# Tidbit · every entry > A free bilingual dictionary of technology: AI, cloud, AWS, networks, security, data, hardware and tech companies. Every term in plain words, with an analogy, a diagram and official sources. ## 5G > The fifth generation of mobile networks: faster data, less delay, more devices at once. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/5g/ - In Spanish: [5G](https://tidbit.legios.com.ar/es/5g/) **5G** is the fifth generation of the technology that connects phones and other devices to mobile networks, after 2G (calls and text messages), 3G (the first usable mobile internet) and 4G (video and apps everywhere). It is a set of standards written by an industry body called 3GPP and adopted around the world, so phones and networks from different makers work together. It improves three things. **Speed**: more data per second, so big downloads finish much sooner. **Latency**: a shorter delay between asking and getting an answer, which helps video calls, games and remote control of machines. And **capacity**: many more devices connected in the same area, which matters in stadiums, crowded city centers and factories full of sensors. Part of this comes from using new radio frequencies, including very high ones that carry a lot of data but travel short distances and are easily blocked by walls. What you actually get varies a lot. Many carriers first launched 5G on top of their existing 4G equipment and lower frequencies, which works well but feels only somewhat faster; the most dramatic speeds appear only near dense groups of antennas. For most people, 5G mostly means a phone that keeps working smoothly in crowded places. Its bigger long-term uses are less visible: private networks in factories and ports, and connecting huge numbers of devices at once. ### An analogy Widening a highway. Each car does not go dramatically faster, but many more cars fit, there are fewer jams, and a few special express lanes are very fast over short stretches. ### Keep in mind - The fifth generation of mobile network standards, after 4G. - Improves speed, delay (latency) and how many devices fit in one area. - Its highest frequencies carry lots of data but only over short distances. - Real-world speed depends heavily on where you are and how the network was built. ### Go deeper - [5G on Wikipedia](https://en.wikipedia.org/wiki/5G) - [5G system overview (3GPP)](https://www.3gpp.org/technologies/5g-system-overview) - [5G, fifth generation of mobile technologies (ITU)](https://www.itu.int/en/mediacentre/backgrounders/Pages/5G-fifth-generation-of-mobile-technologies.aspx) --- ## Adobe > The company behind Photoshop, Illustrator, Premiere Pro, Acrobat and the PDF format. - Level: Basic - Topics: Tech companies - URL: https://tidbit.legios.com.ar/adobe/ - In Spanish: [Adobe](https://tidbit.legios.com.ar/es/adobe/) **Adobe** is an American software company founded in 1982 by John Warnock and Charles Geschke, two researchers from Xerox's lab in Palo Alto. Its first product, PostScript, was a language that described a page so that any printer could print it exactly. That idea of a page that looks the same everywhere led, in 1993, to the **PDF** format and the Acrobat software to read and create it. PDF later became an open ISO standard. Adobe also makes the tools most creative professionals use: **Photoshop** for editing photos, **Illustrator** for drawings and logos, **InDesign** for magazines and books, **Premiere Pro** for video and **Lightroom** for organizing and developing photos. In recent years it added generative AI to them, through its own Firefly models, so you can fill or extend an image by describing what you want. For decades Adobe sold its programs in boxes, for a single large payment. In 2013 it switched to **Creative Cloud**, a monthly or yearly subscription, and today most of its money comes from subscriptions: Creative Cloud for creative work, Document Cloud for Acrobat and PDFs, and Experience Cloud, software that helps large companies run their websites, marketing campaigns and analytics. ### An analogy A gym membership for creative tools. Instead of buying each machine for your home, you pay every month and get access to the whole room, with new equipment added over time. ### Keep in mind - Founded in 1982 by John Warnock and Charles Geschke, both from Xerox PARC. - Created PostScript and the PDF format, which is now an open ISO standard. - Makes Photoshop, Illustrator, InDesign, Premiere Pro, Lightroom and Acrobat. - Since 2013 it sells its software mainly as subscriptions. ### Go deeper - [Adobe Inc. on Wikipedia](https://en.wikipedia.org/wiki/Adobe_Inc.) - [Adobe Creative Cloud on Wikipedia](https://en.wikipedia.org/wiki/Adobe_Creative_Cloud) - [PDF on Wikipedia](https://en.wikipedia.org/wiki/PDF) --- ## AGI > A hypothetical AI able to do most mental tasks as well as a person. Not here yet. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/agi/ - In Spanish: [AGI](https://tidbit.legios.com.ar/es/agi/) **AGI**, short for **artificial general intelligence**, is the idea of an AI system that could learn and do most intellectual tasks a human can, at a human level or better, instead of being good at one narrow thing. A chess program that cannot read a recipe is 'narrow' AI. An AGI could, in principle, switch from diagnosing a disease to planning a trip to doing original research, and learn new skills along the way the way people do. There is no agreed definition, and that is the first thing to know. Some define it by tasks ('matches an expert at most jobs done at a computer'), some by economics ('could do most work people are paid for'), some by abilities like learning quickly from little data and adapting to the unexpected. Depending on the definition, a given system might count or not, which is why claims that AGI is near, or already here, are hard to evaluate. Today's large language models are much more general than older AI: the same model can write, translate, code and explain. But they still make mistakes no competent person would, struggle to learn on the job, and depend on human direction. Experts disagree widely on whether and when AGI will arrive, with guesses ranging from a few years to many decades to never. Several AI companies say building it is their goal, and its possible benefits and risks are a serious topic in research and policy. When someone mentions AGI, a good first question is: what exactly do you mean by it? ### An analogy Today's AI is like a set of very good specialist tools: a great saw, a great drill, a great level. AGI would be a craftsperson who can pick up any tool, learn a new trade and run the whole job. Nobody agrees on how close we are to that craftsperson. ### Keep in mind - AGI means AI that is broadly capable, not good at just one task. - There is no single agreed definition or test. - Current models are more general than before but still fall short. - Expert predictions on timing differ enormously. ### Go deeper - [Artificial general intelligence on Wikipedia](https://en.wikipedia.org/wiki/Artificial_general_intelligence) - [What is AGI? (IBM)](https://www.ibm.com/think/topics/artificial-general-intelligence) --- ## Agile and Scrum > Building software in short cycles, showing results often and adjusting as you learn. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/agile-and-scrum/ - In Spanish: [Agile y Scrum](https://tidbit.legios.com.ar/es/agile-and-scrum/) **Agile** is a family of ways to organize software work around short cycles and constant feedback, instead of one big plan carried out over a year. It was named in 2001 by a group of developers who wrote the **Agile Manifesto**, which values working software, collaboration with the customer and responding to change over heavy documentation and rigid contracts. The older style it reacted against is often called **waterfall**: specify everything, then build everything, then test everything, then deliver. **Scrum** is the most widely used way of putting Agile into practice. Work is done in **sprints**, fixed periods of up to a month, very often two weeks. All pending work lives in a prioritized list called the **product backlog**. At the start of a sprint the team picks what it can finish; every day it holds a short meeting, the **daily scrum** or stand-up; at the end it shows what it built in a **sprint review** and discusses how to work better in a **retrospective**. Scrum defines three roles: the **product owner**, who decides what matters most; the **Scrum Master**, who helps the team follow the process and removes obstacles; and the developers, who do the work. **Kanban** is a lighter alternative without sprints: tasks move across a board from "to do" to "done", with a limit on how many can be in progress at once. In practice many teams mix both, and "agile" sometimes ends up meaning just "lots of meetings", which misses the point. ### An analogy Like driving with a GPS that recalculates, instead of printing the whole route before leaving. You still know where you are going, but every few kilometers you check where you actually are and adjust. ### Keep in mind - Agile is a set of values; Scrum is one concrete way to apply them. - Scrum works in sprints of up to a month, often two weeks. - Each sprint should end with something working that people can see and use. - Kanban drops sprints and limits how much work is in progress. ### Go deeper - [Manifesto for Agile Software Development](https://agilemanifesto.org/) - [The Scrum Guide](https://scrumguides.org/scrum-guide.html) - [Scrum on Wikipedia](https://en.wikipedia.org/wiki/Scrum_(software_development)) --- ## AI agent > An LLM that can take actions, check the results and keep going until the job is done. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/ai-agent/ - In Spanish: [Agente de IA](https://tidbit.legios.com.ar/es/ai-agent/) An **AI agent** is an LLM wrapped in a loop with tools. A plain chatbot answers once and stops. An agent receives a goal, decides what to do, uses a tool (search the web, read a file, run code, call an API, send an email), looks at the result, and decides the next step. It repeats until the goal is met or it gets stuck. The tools are what make it an agent. Each tool is a function the developer exposes with a description; the model chooses when to call it and with what arguments. A coding agent like Claude Code, for example, can read your project, edit files, run the tests, see them fail and fix the code, without a person typing each command. This is powerful and also where the risks live. An agent that can act can also act wrongly: delete the wrong file, buy the wrong ticket, loop forever. That is why serious agents ask for confirmation before irreversible steps, run in sandboxes with limited permissions, and keep a log of what they did. Why it matters now: in 2023 the question was 'what can the model say?'; by 2026 it is 'what can the model do?'. Agents are how AI moves from answering questions to completing tasks, which is where most of the economic value, and most of the open problems, are. ### An analogy A chatbot is a consultant who writes you a memo. An agent is an assistant with your calendar, your inbox and a company card, who gets the trip booked and tells you when it is done. ### Keep in mind - Agent = model + tools + a loop: think, act, observe, repeat. - Tools are ordinary functions or APIs the model can choose to call. - Agents act on the world, so permissions, confirmations and logs matter. - Open standards like MCP (Model Context Protocol) let agents connect to tools and data in a uniform way. ### Go deeper - [Anthropic: building effective agents](https://www.anthropic.com/research/building-effective-agents) - [Intelligent agent on Wikipedia](https://en.wikipedia.org/wiki/Intelligent_agent) --- ## AI bias > When an AI system treats some groups of people unfairly, usually learned from its data. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/ai-bias/ - In Spanish: [Sesgo en la IA](https://tidbit.legios.com.ar/es/ai-bias/) **AI bias** is when an AI system produces results that are systematically unfair to certain groups of people: by gender, skin color, age, accent, disability, neighborhood. Usually nobody programmed it on purpose. A model learns from examples, and if the examples reflect an unequal past, the model learns that inequality as if it were a rule. Real cases made this concrete. In 2018 Reuters reported that Amazon had dropped an experimental hiring tool after finding it downgraded résumés that included the word 'women's', because it had learned from a decade of mostly male hires. The same year, the Gender Shades study found that commercial face-analysis systems made far more errors on darker-skinned women than on lighter-skinned men, because their training photos were skewed. Language models can do it too, for instance by assuming a doctor is a man and a nurse is a woman when translating. Bias can enter at many points: data that underrepresents some people, labels that carry human prejudice, a goal that measures the wrong thing, or a system used on a population it was never tested on. It matters most where AI touches decisions about people: hiring, loans, insurance, healthcare, policing. Reducing it takes deliberate work: more representative data, testing results separately for each group, independent audits, and keeping a human able to question and override the system. Laws such as the EU AI Act now require this kind of checking for high-risk uses. ### An analogy Like a new employee trained only by reading the company's old decisions. If those decisions favored one kind of person, the new employee will copy the pattern, sincerely believing it is just how the job is done. ### Keep in mind - Usually learned from data that reflects past unfairness. - It can hide in data, labels, goals or where the system is used. - It matters most in hiring, credit, health and policing. - Testing results per group and human oversight help reduce it. ### Go deeper - [Algorithmic bias on Wikipedia](https://en.wikipedia.org/wiki/Algorithmic_bias) - [What is AI bias? (IBM)](https://www.ibm.com/think/topics/ai-bias) - [Identifying and managing bias in AI (NIST)](https://www.nist.gov/publications/towards-standard-identifying-and-managing-bias-artificial-intelligence) --- ## Algorithm > A precise, step-by-step method for solving a problem or doing a task. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/algorithm/ - In Spanish: [Algoritmo](https://tidbit.legios.com.ar/es/algorithm/) An **algorithm** is a finite list of clear steps that, followed exactly, solves a problem. Long division as you learned it at school is an algorithm. So is "to find a word in a paper dictionary, open it in the middle, check whether the word comes before or after, and repeat with that half". Computers need algorithms because they cannot improvise: every program is algorithms written in a programming language. The same problem can often be solved by several algorithms, and they are not equally good. Checking a dictionary page by page from the start works, but halving it each time (called **binary search**) finds any word in a few dozen steps even in a huge book. Choosing a smarter algorithm usually matters far more than buying a faster computer. In everyday news the word is used more loosely, as in "the algorithm decided what you see". That usually means a ranking or recommendation system: software that scores posts, videos or ads and puts the highest scores first. Those systems are built from many algorithms, often combined with machine learning models trained on what people clicked before. ### An analogy A recipe, but one written so precisely that someone who has never cooked, and never thinks for themselves, would still end up with the same cake every time. ### Keep in mind - A finite sequence of exact steps that solves a problem. - Different algorithms for the same task can differ enormously in speed. - Classic examples: sorting a list, searching, finding the shortest route. - "The algorithm" of a social app is a ranking system built from many of them. ### Go deeper - [Algorithms course (Khan Academy)](https://www.khanacademy.org/computing/computer-science/algorithms) - [Algorithm on Wikipedia](https://en.wikipedia.org/wiki/Algorithm) --- ## Alibaba > China's giant of online marketplaces, which also runs a major cloud. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/alibaba/ - In Spanish: [Alibaba](https://tidbit.legios.com.ar/es/alibaba/) **Alibaba Group** was founded in 1999 in Hangzhou, China, by Jack Ma and a group of partners. It started with **Alibaba.com**, a website where small Chinese manufacturers could find buyers abroad, and it is still where many businesses around the world go to order products in bulk directly from factories. It then built marketplaces for consumers. **Taobao**, launched in 2003, lets anyone open a shop and sell to shoppers in China; **Tmall** is the version for established brands; **AliExpress** sells to buyers in other countries. Alibaba usually does not own the goods: it makes money from the fees and, above all, the advertising sellers pay to stand out. It also popularized the 11 November "Singles' Day" sales festival, and runs logistics through its Cainiao network. Like Amazon with AWS, Alibaba turned the computing infrastructure it needed for its own shopping peaks into a business. **Alibaba Cloud**, also called Aliyun, started in 2009 and is the largest cloud provider in China, renting servers, storage and databases, and it develops the Qwen family of AI models, many released as open weights. The payments app Alipay, often associated with it, belongs to Ant Group, a separate company in which Alibaba holds a stake. ### An analogy A huge market where the building owner does not sell anything: it rents stalls, charges for the signs that make a stall visible, and also runs the power plant that keeps the lights on for the whole neighborhood. ### Keep in mind - Founded in 1999 in Hangzhou; started as a site connecting Chinese factories with buyers abroad. - Runs Taobao, Tmall and AliExpress, earning mostly from seller fees and ads. - Alibaba Cloud is China's largest cloud provider and makes the Qwen AI models. - Alipay belongs to Ant Group, a separate company linked to Alibaba. ### Go deeper - [About Alibaba Group (official site)](https://www.alibabagroup.com/en-US/about-alibaba) - [Alibaba Group on Wikipedia](https://en.wikipedia.org/wiki/Alibaba_Group) - [About Alibaba Cloud](https://www.alibabacloud.com/about) --- ## Amazon > An online store with its own delivery network, and the company that runs AWS. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/amazon/ - In Spanish: [Amazon](https://tidbit.legios.com.ar/es/amazon/) **Amazon** was founded by Jeff Bezos in 1994, in Seattle, as an online bookstore. It grew into a store that sells almost everything, and most of what it sells today comes from independent **sellers** who list their products on Amazon's marketplace and pay fees for the sale, and often for Amazon to store and ship the goods. Behind the website sits one of the largest logistics networks in the world: warehouses, sorting centers, planes and delivery vans. Several businesses grew around the store. **Prime** is a yearly or monthly membership with fast shipping, Prime Video and other perks, which keeps customers coming back. Amazon also sells ads inside its own store, so brands pay to appear first in search results, and makes devices such as the Kindle and the Echo speakers with the **Alexa** assistant. The part that matters most in tech is **AWS** (Amazon Web Services). Amazon built huge computing infrastructure for its own store and, in 2006, started renting it to others: S3 for storage and EC2 for servers were among the first services. AWS became the largest cloud provider and, although it is a smaller share of Amazon's sales than the store, it brings in a large share of the company's profit. Many of the apps and websites you use every day run on it, and it is also a major investor in Anthropic. ### An analogy A shopping mall that also owns the trucks, the warehouses and the power plant. Shops rent space and pay for delivery; and since the power plant had capacity to spare, the mall started selling electricity to the whole city too. ### Keep in mind - Founded in 1994 by Jeff Bezos in Seattle, as an online bookstore. - Most items in its store come from third-party sellers who pay Amazon fees. - Prime, advertising and devices like Alexa grew around the store. - AWS, launched in 2006, is the largest cloud provider and a big part of Amazon's profit. ### Go deeper - [About Amazon](https://www.aboutamazon.com/about-us) - [What is AWS?](https://aws.amazon.com/what-is-aws/) - [Amazon on Wikipedia](https://en.wikipedia.org/wiki/Amazon_(company)) --- ## Amazon API Gateway > The front door of your API: it receives requests and passes them to your code. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-api-gateway/ - In Spanish: [Amazon API Gateway](https://tidbit.legios.com.ar/es/amazon-api-gateway/) **Amazon API Gateway** gives your API a public address on the internet and takes care of everything that happens before your code runs. You define routes like `GET /orders` or `POST /users`, tell it where each one goes (usually an AWS Lambda function, sometimes a server or another AWS service), and it handles the rest. 'The rest' is a lot: serving over HTTPS, checking API keys or login tokens (it works directly with Amazon Cognito), limiting how many requests per second each client can make, caching responses, logging every call to Amazon CloudWatch, and even keeping WebSocket connections open and turning their messages into events for your code. Example: a mobile app calls `https://api.example.com/orders`. API Gateway receives the request, verifies the user's token, rejects it if that client already exceeded its quota, and otherwise invokes a Lambda function with the request body. The function returns JSON, and the gateway sends it back to the phone. No server was ever provisioned. Together with Lambda and Amazon DynamoDB it is the classic recipe for a serverless backend: you pay per request, and it scales from zero to millions of calls without touching anything. ### An analogy The receptionist of an office building. Visitors do not walk straight to a desk: the receptionist checks their ID, limits how many go up at once, and points each one to the right door. ### Keep in mind - You define routes and point each one at Lambda, a server or an AWS service. - Authentication, rate limiting, caching and logging happen before your code runs. - Two flavors: HTTP APIs (simpler and cheaper) and REST APIs (more features). - Billed per million requests; nothing to run or pay for while idle. ### Go deeper - [What is Amazon API Gateway? (official docs)](https://docs.aws.amazon.com/apigateway/latest/developerguide/welcome.html) - [Amazon API Gateway product page](https://aws.amazon.com/api-gateway/) --- ## Amazon Athena > Run SQL directly on files sitting in S3, with no database to set up. - Level: Intermediate - Topics: AWS, Data - URL: https://tidbit.legios.com.ar/amazon-athena/ - In Spanish: [Amazon Athena](https://tidbit.legios.com.ar/es/amazon-athena/) **Amazon Athena** is a query service: you point it at files stored in Amazon S3 (CSV, JSON, Parquet, logs), describe their columns once, and run standard SQL against them. There is no server to start and no data to load: Athena reads the files when you run the query, and you pay per amount of data scanned (roughly 5 dollars per terabyte). It exists because a lot of valuable data lives as files, not inside a database: application logs, exports from other systems, clickstreams, sensor readings. Loading all of that into a data warehouse is slow and costly; with Athena you query it where it is. It is the query engine of the data lake pattern: S3 as the storage, the AWS Glue Data Catalog as the list of tables and columns, Athena as the SQL layer. Example: a company keeps a year of web server logs in S3, in one folder per day. An analyst runs `SELECT status, COUNT(*) FROM logs WHERE day = '2026-10-01' GROUP BY status` and gets the answer in seconds, scanning only that day's folder. Converting the logs to Parquet, a compressed column-oriented format, makes the same query scan far less data and cost far less. Athena is built on Trino (formerly Presto) and can also run Apache Spark. It shines for ad hoc analysis and scheduled reports; for a team hitting the same tables all day and expecting sub-second answers, a warehouse like Amazon Redshift is the better fit. ### An analogy A librarian who answers questions by reading the boxes in the archive directly, instead of first copying every document into a filing system. Slower per question than a filing system, but it works on day one and you pay only for what gets read. ### Keep in mind - Serverless SQL over files in S3: no cluster to run, no loading step. - Billed per data scanned, so columnar formats like Parquet and folder partitions cut costs a lot. - Table definitions live in the Glue Data Catalog and are shared with other AWS tools. - Great for ad hoc analysis and logs; a warehouse suits heavy, constant workloads. ### Go deeper - [What is Amazon Athena? (official docs)](https://docs.aws.amazon.com/athena/latest/ug/what-is.html) - [Amazon Athena product page](https://aws.amazon.com/athena/) --- ## Amazon Bedrock > A single AWS service to call AI models from Anthropic, Meta, Amazon and others. - Level: Basic - Topics: AWS, AI - URL: https://tidbit.legios.com.ar/amazon-bedrock/ - In Spanish: [Amazon Bedrock](https://tidbit.legios.com.ar/es/amazon-bedrock/) **Bedrock** is the AWS service for using generative AI models through an API. Instead of running a model yourself, you send a prompt and get the answer back, paying per token. The catalog includes Claude from Anthropic, Llama from Meta, Mistral, Amazon's own Nova models and more, all behind the same API, so trying a different model is a one-line change. What AWS adds on top of the models: your data stays in your account and your Region (it is not used to train anyone's models), access is controlled with IAM, and there are building blocks for the common patterns: **Knowledge Bases** for RAG (letting a model answer from your own documents), **Agents** that can call tools and APIs, **Guardrails** that filter topics and personal data, and fine-tuning for some models. Typical example: a bank already on AWS builds an internal assistant that answers questions from its policy manuals. The documents live in Amazon S3, Bedrock indexes them into a Knowledge Base, and Claude writes the answers. Nothing leaves the company's AWS account, which is usually the reason Bedrock gets picked over calling a model provider directly. ### An analogy A restaurant with one menu and several kitchens. You order from the same waiter, and whether the dish comes from the Anthropic kitchen or the Meta kitchen, the bill, the plates and the health inspection are the restaurant's job. ### Keep in mind - Pay per token; no servers or GPUs to manage. - Several model providers behind one API, including Claude. - Your prompts and data are not used to train the models. - Built-in pieces for RAG, agents and content guardrails. ### Go deeper - [What is Amazon Bedrock? (official docs)](https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html) - [Amazon Bedrock](https://aws.amazon.com/bedrock/) --- ## Amazon CloudFront > AWS's CDN: copies of your content kept close to users all over the world. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-cloudfront/ - In Spanish: [Amazon CloudFront](https://tidbit.legios.com.ar/es/amazon-cloudfront/) **CloudFront** is AWS's CDN (content delivery network). You point it at an **origin** (an Amazon S3 bucket, a load balancer, any web server) and it serves your files from hundreds of **edge locations**: small data centers in cities on every continent. A visitor in Buenos Aires gets your images from an edge nearby instead of from a server in Virginia, so the page loads noticeably faster. The first visitor near an edge triggers a fetch from the origin; the edge keeps a copy (a **cache**) for as long as you configure and serves everyone else from it. That also protects the origin: a million visitors turn into a handful of requests to your server. CloudFront handles HTTPS with a free certificate, blocks common attacks in front of your site, and charges per gigabyte served. A common setup is a static website: HTML, CSS and images in an S3 bucket, CloudFront in front, a domain from Amazon Route 53. No server at all, and it holds up fine when a post goes viral. ### An analogy A bakery chain that bakes once at headquarters and keeps fresh loaves on shelves in every neighborhood. You buy around the corner instead of driving to the factory, and the factory only bakes again when a shelf runs empty. ### Keep in mind - The origin is where the content lives; the edges keep temporary copies. - Faster for users and cheaper for the origin: most requests never reach it. - Includes HTTPS with a free certificate and basic protection against attacks. - You decide how long each file stays cached (its TTL). ### Go deeper - [What is Amazon CloudFront? (official docs)](https://docs.aws.amazon.com/AmazonCloudFront/latest/DeveloperGuide/Introduction.html) - [Amazon CloudFront](https://aws.amazon.com/cloudfront/) --- ## Amazon CloudWatch > The dashboard and alarm system that watches everything running in your AWS account. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-cloudwatch/ - In Spanish: [Amazon CloudWatch](https://tidbit.legios.com.ar/es/amazon-cloudwatch/) **CloudWatch** is the AWS service for monitoring. Almost every AWS service reports **metrics** to it on its own: CPU use of an Amazon EC2 instance, number of calls to an AWS Lambda function, free space in an Amazon RDS database, messages waiting in a queue. You see them as charts going back months, without installing anything. Second piece: **logs**. Applications and services send their text output to CloudWatch Logs, where you can search it ('every line with `error` in the last hour') instead of logging into machines. Third piece: **alarms**. You write a rule such as 'CPU above 80% for five minutes' or 'more than ten errors a minute', and CloudWatch sends an email through Amazon SNS, pages the on-call person, or triggers an action like adding servers. Why it matters: in the cloud you cannot walk over to a machine and look at it. CloudWatch is how you find out something broke, ideally before the users do. A team sets an alarm on a Lambda function's error count; at 2 a.m. it fires, the on-call engineer gets a text, and the fix goes in before the morning rush. ### An analogy The dashboard of a car. Speed, fuel and engine temperature in one place, with a red light that comes on before the engine actually overheats. ### Keep in mind - Metrics are numbers over time, logs are text, alarms act on either. - Most AWS services send basic metrics for free; detailed metrics and logs cost per gigabyte. - Alarms can notify people or trigger automatic actions like scaling. - Dashboards put metrics from many services on one screen. ### Go deeper - [What is Amazon CloudWatch? (official docs)](https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitoring/WhatIsCloudWatch.html) - [Amazon CloudWatch](https://aws.amazon.com/cloudwatch/) --- ## Amazon Cognito > Sign-up, login and passwords for your app, handled by AWS instead of you. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/amazon-cognito/ - In Spanish: [Amazon Cognito](https://tidbit.legios.com.ar/es/amazon-cognito/) **Amazon Cognito** manages the users of an application: sign-up, login, password reset, email or phone verification, two-factor authentication, and 'log in with Google' style federation. Your app does not store passwords; Cognito does, and hands your app a signed token that says who the user is. It has two parts. A **user pool** is the directory of users plus the login machinery: it issues JSON Web Tokens (JWT) that your backend, or Amazon API Gateway directly, can verify. An **identity pool** goes one step further: it exchanges that token for temporary AWS credentials, so a mobile app can upload a photo straight to Amazon S3 under IAM rules, with no server in between. Example: the app of a gym. Members sign up with their email, confirm with a code, and log in with a password plus an SMS code. The app sends the Cognito token to the API; API Gateway checks the signature and the expiry, and the AWS Lambda function behind it reads the member id from the token. Nobody wrote login code, and the passwords live in a store that AWS hardens. Why it matters: login is where many security mistakes happen (passwords stored badly, reset flows that leak). Handing it to a managed service is one of the cheapest wins in security. Cognito has a free tier for a first batch of monthly active users, then charges per user. ### An analogy The membership desk of a club. It issues cards, checks faces, handles lost cards and renewals. Every other door in the club only has to look at the card. ### Keep in mind - User pools authenticate people and issue JWT tokens; identity pools turn tokens into temporary AWS credentials. - Supports MFA, social login (Google, Apple, Facebook) and corporate SAML or OIDC through OAuth and SSO standards. - API Gateway can verify Cognito tokens with zero backend code. - Your app never touches a password. ### Go deeper - [What is Amazon Cognito? (official docs)](https://docs.aws.amazon.com/cognito/latest/developerguide/what-is-amazon-cognito.html) - [Amazon Cognito product page](https://aws.amazon.com/cognito/) --- ## Amazon DynamoDB > A key-value database that answers in milliseconds no matter how big the table gets. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-dynamodb/ - In Spanish: [Amazon DynamoDB](https://tidbit.legios.com.ar/es/amazon-dynamodb/) **DynamoDB** is AWS's NoSQL database. Instead of tables with rows and columns joined by SQL, it stores **items** (JSON-like documents) and finds them by a **key** you choose, such as a user id or an order number. Fetching an item by key takes a few milliseconds whether the table holds a thousand items or a billion. There is no server to size: you pay per read, per write and per gigabyte stored, and AWS spreads the table over as many machines as it needs. It is fully managed, replicated across three Availability Zones, and can keep live copies in other Regions. The catch is that it only does fast what it was designed for: getting items by key. A question like 'all orders over 100 dollars last March' needs either a careful table design or a different tool, such as Amazon RDS or Amazon Athena. Typical uses: shopping carts, user sessions and profiles, game leaderboards, the state behind a mobile app. Amazon's own retail site leans on it for its busiest shopping days, which is roughly the scale it was built for. ### An analogy A coat check. You hand over a coat, get a numbered ticket, and later get the coat back instantly by showing the ticket. Asking the attendant for 'all the blue coats' is not what the system is for. ### Keep in mind - You access items by a key you choose; design the key around the questions you will ask. - Serverless pricing: per read, per write and per gigabyte, with no instance to manage. - Reads take single-digit milliseconds at any table size. - Great for carts, sessions and profiles; a poor fit for ad hoc reporting. ### Go deeper - [What is Amazon DynamoDB? (official docs)](https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/Introduction.html) - [Amazon DynamoDB](https://aws.amazon.com/dynamodb/) --- ## Amazon EBS > The virtual hard drive of an EC2 instance, kept alive even when the instance stops. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-ebs/ - In Spanish: [Amazon EBS](https://tidbit.legios.com.ar/es/amazon-ebs/) **EBS** (Elastic Block Store) provides **volumes**: virtual disks that you attach to an Amazon EC2 instance and that the operating system sees as a normal drive. The instance's root disk is an EBS volume; you can add more, like plugging in extra hard drives, and each one can be resized while in use. The key property is that a volume is independent from the instance. Stop or terminate the machine and the volume can stay, be attached to a different instance in the same Availability Zone, or be captured as a **snapshot**: a backup stored in Amazon S3 that you can turn into a new volume anywhere in the Region, or copy to another one. AWS replicates every volume inside its zone, so one failed disk does not lose data. Volume types trade price for speed: `gp3` (general purpose SSD) is the default for almost everything, `io2` offers guaranteed high performance for demanding databases, `st1` and `sc1` are cheap spinning disks for large sequential files. You pay per gigabyte provisioned per month, used or not, plus the snapshots. EBS is block storage attached to one machine; for files shared by many or served to the web, Amazon S3 is the tool. ### An analogy An external hard drive. You plug it into one laptop, unplug it, plug it into another, and keep a copy in a safe (the snapshot). It is still a drive: one computer at a time, and it has to be in the same room (the zone). ### Keep in mind - A volume lives in one Availability Zone and attaches to one instance at a time. - Snapshots are incremental backups stored in S3; copy them across Regions for disaster recovery. - `gp3` is the default; pick `io2` for heavy databases, `st1` or `sc1` for cheap bulk storage. - You pay for the gigabytes you provision, not for the ones you fill. ### Go deeper - [What is Amazon EBS? (official docs)](https://docs.aws.amazon.com/ebs/latest/userguide/what-is-ebs.html) - [Amazon EBS](https://aws.amazon.com/ebs/) --- ## Amazon EC2 > Virtual computers you rent by the hour, in any size, and switch off when done. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-ec2/ - In Spanish: [Amazon EC2](https://tidbit.legios.com.ar/es/amazon-ec2/) **EC2** (Elastic Compute Cloud) is the AWS service for renting virtual machines, called **instances**. You choose an operating system (Linux or Windows), a size (how many CPU cores, how much RAM), press launch, and about a minute later you have a computer you can log into and run anything on: a website, a database, a game server, a script that runs once a night. You pay by the second while the instance is running and nothing for the machine while it is stopped. Sizes go from tiny (one shared core, half a gigabyte of RAM, a fraction of a cent per hour) to huge machines with hundreds of cores or several GPUs for training AI models. Instance types have names like `t3.micro` or `m7g.large`: the letters say what the family is good for, the suffix says how big it is. An instance lives in one Availability Zone of a Region, its disk is usually an Amazon EBS volume, and its permissions come from an IAM role. EC2 is the most classic piece of AWS: when something cannot run as a container or as an AWS Lambda function, it ends up on an EC2 instance. ### An analogy Renting a car instead of buying one. You pick the size you need today (a compact for errands, a truck for moving house), pay for the hours you drive, and return it. Tomorrow you can take a different one. ### Keep in mind - An instance is a virtual machine; you choose the OS, the CPU and the RAM. - Billing is per second while running; a stopped instance only pays for its disk. - Families: `t` and `m` for general work, `c` for compute, `r` for memory, `p` and `g` for GPUs. - Spot instances are the same machines at a steep discount, with the risk of being reclaimed. ### Go deeper - [What is Amazon EC2? (official docs)](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/concepts.html) - [Amazon EC2](https://aws.amazon.com/ec2/) --- ## Amazon ECR > A private warehouse on AWS for your container images, ready to be deployed. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-ecr/ - In Spanish: [Amazon ECR](https://tidbit.legios.com.ar/es/amazon-ecr/) **ECR** (Elastic Container Registry) is where you store container images on AWS. An image is the packaged version of an app: the code, the libraries and the settings it needs, frozen in one file that any machine can run. Before you can run it on AWS, the image has to live somewhere AWS can fetch it from, and that place is a **container registry**. The flow is short. A developer or a build pipeline creates the image with Docker, tags it with a version like `web:1.4.2`, and **pushes** it to an ECR repository. Then Amazon ECS, Amazon EKS or AWS Lambda **pulls** that exact version and starts it. Access is controlled with IAM, so only the people and services you allow can push or pull. ECR also does some housekeeping: it can scan images for known vulnerabilities in the software they contain, and lifecycle rules can delete old versions automatically so the repository does not grow forever. There is a public side too, for sharing images with anyone, but most teams use ECR as the private, nearby copy of what they deploy. ### An analogy The loading dock of a factory. Finished, sealed boxes (images) wait on labeled shelves, and the delivery trucks (ECS, EKS, Lambda) pick up exactly the box with the label they were told to take. ### Keep in mind - Stores and versions container images inside your AWS account. - You push images to it; ECS, EKS and Lambda pull them to run. - Access goes through IAM; images are private by default. - Can scan images for known vulnerabilities and expire old versions. ### Go deeper - [What is Amazon ECR? (official docs)](https://docs.aws.amazon.com/AmazonECR/latest/userguide/what-is-ecr.html) - [Amazon ECR](https://aws.amazon.com/ecr/) --- ## Amazon ECS and Fargate > AWS runs your containers; with Fargate you never see the servers at all. - Level: Intermediate - Topics: AWS, Cloud - URL: https://tidbit.legios.com.ar/amazon-ecs-and-fargate/ - In Spanish: [Amazon ECS y Fargate](https://tidbit.legios.com.ar/es/amazon-ecs-and-fargate/) **Amazon ECS** (Elastic container Service) is the AWS service that runs containers for you. You write a **task definition**: which container image to use, how much CPU and memory it needs, which ports it opens, which environment variables it gets. Then you create a **service**: 'keep three copies of this task running behind this load balancer', and ECS makes sure three are always alive, replacing any that die. Where do those containers physically run? Two options. With the EC2 launch type you keep a fleet of Amazon EC2 instances and ECS places tasks on them; you patch and size the machines. With **Fargate** there is no machine to manage: you say 'this task needs 1 vCPU and 2 GB', AWS finds the capacity, starts the container and bills you per second while it runs. Example: a Python API packaged as a Docker image and pushed to Amazon ECR, the AWS image registry. An ECS service on Fargate runs two copies in private subnets of a VPC with a load balancer in front; autoscaling adds copies when CPU stays above 70%. Shipping a new version means pushing a new image tag and letting ECS roll it out copy by copy. ECS is the simpler, AWS-native way to run containers. Amazon EKS is the choice when a team already works with Kubernetes and wants to keep that toolset. ### An analogy A food court. ECS is the manager who guarantees your stall is always open and staffed. With Fargate you also stop renting floor space: you say how big a stall you need and it appears, and you pay only while it is open. ### Keep in mind - A task definition is the recipe; a service keeps N copies of it running and healthy. - Fargate removes servers entirely: you pay per vCPU and GB for the seconds a task runs. - Images usually live in Amazon ECR, the AWS container registry. - ECS plugs into IAM, load balancers and CloudWatch without extra setup. ### Go deeper - [What is Amazon ECS? (official docs)](https://docs.aws.amazon.com/AmazonECS/latest/developerguide/Welcome.html) - [AWS Fargate product page](https://aws.amazon.com/fargate/) --- ## Amazon EKS > Kubernetes with the hard part, the control plane, run by AWS. - Level: Advanced - Topics: AWS, Cloud - URL: https://tidbit.legios.com.ar/amazon-eks/ - In Spanish: [Amazon EKS](https://tidbit.legios.com.ar/es/amazon-eks/) **Amazon EKS** (Elastic Kubernetes Service) is managed Kubernetes on AWS. Kubernetes has two halves: the **control plane** (the brain: the API server, the scheduler and the etcd database that stores the state of the cluster) and the **worker nodes** (the machines where your containers actually run). EKS runs the control plane for you, spread across several Availability Zones, patched and backed up, and exposes the standard Kubernetes API so your usual tools (`kubectl`, Helm, GitOps pipelines) work unchanged. For the nodes you choose: managed node groups of Amazon EC2 instances that EKS creates and upgrades, Fargate so pods run with no node to manage at all, or the Karpenter autoscaler that launches the right-sized instance on demand. EKS also wires Kubernetes into AWS: a pod can carry an IAM role, a Kubernetes Service of type LoadBalancer becomes a real AWS Load balancer, and a persistent volume becomes an Amazon EBS disk. Why pick it over Amazon ECS and Fargate? Portability and ecosystem. A team that already runs Kubernetes elsewhere, or depends on its tools (operators, service meshes, Argo CD), brings its manifests to EKS almost as they are. The price is complexity: Kubernetes has many moving parts, and version upgrades still need planning on your side. Example: a company with 40 microservices runs them on one EKS cluster with three node groups (general purpose, memory-heavy, and GPU for model inference), one Helm chart per service, and Argo CD deploying every merge to main. ### An analogy Renting a theater with the technical crew included. AWS runs the lighting booth and the sound desk (the control plane); you bring the actors (your containers) and decide how big a stage you need (the nodes). ### Keep in mind - AWS runs and patches the control plane; you choose how nodes run: EC2, Fargate or Karpenter. - It is upstream Kubernetes: standard API, standard tooling, portable manifests. - Deep AWS integration: IAM roles for pods, load balancers, EBS and EFS volumes. - More flexible than ECS, and noticeably more to learn and operate. ### Go deeper - [What is Amazon EKS? (official docs)](https://docs.aws.amazon.com/eks/latest/userguide/what-is-eks.html) - [Amazon EKS product page](https://aws.amazon.com/eks/) --- ## Amazon ElastiCache > A managed in-memory cache that keeps hot data close so apps answer instantly. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-elasticache/ - In Spanish: [Amazon ElastiCache](https://tidbit.legios.com.ar/es/amazon-elasticache/) **ElastiCache** is the AWS service that runs an in-memory data store for you, using the open source engines Valkey, Redis OSS or Memcached. Because the data lives in RAM instead of on disk, reading a value takes well under a millisecond. AWS handles the servers, patches, replicas and failover. Its classic use is as a **cache**: a copy of answers that are expensive to compute or slow to fetch. Say the home page of a shop shows the 20 best-selling products, and building that list takes a heavy database query. The app runs the query once, stores the result in ElastiCache for a minute, and the next thousand visitors get it straight from memory. The database breathes, and pages load faster. It is also used for things that need speed more than permanence: user sessions, rate limits, game leaderboards, short-lived counters. The trade-off is that memory is more expensive than disk and a cache can be emptied, so the real, durable copy of the data should still live in a database like Amazon RDS or Amazon DynamoDB. ### An analogy The sticky note on your monitor with the phone numbers you dial every day. Looking there is instant; the full phone book is still in the drawer for everything else. ### Keep in mind - Managed Valkey, Redis OSS or Memcached, run by AWS. - Data in RAM means reads in microseconds, not milliseconds. - Usually sits in front of a database to absorb repeated reads. - A cache is a copy: the source of truth stays in the database. ### Go deeper - [What is Amazon ElastiCache? (official docs)](https://docs.aws.amazon.com/AmazonElastiCache/latest/dg/WhatIs.html) - [Caching overview (AWS)](https://aws.amazon.com/caching/) - [Cache (computing) on Wikipedia](https://en.wikipedia.org/wiki/Cache_(computing)) --- ## Amazon EventBridge > A switchboard that routes "something happened" messages to whoever cares. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-eventbridge/ - In Spanish: [Amazon EventBridge](https://tidbit.legios.com.ar/es/amazon-eventbridge/) Many systems are easier to build if parts announce what happened instead of calling each other directly. An order was placed, a file was uploaded, a server stopped. **EventBridge** is the AWS service that carries those announcements, called **events**: small JSON messages that go into an **event bus**. On the bus you set **rules**. Each rule describes a pattern, like "events from the orders app where the total is above 1000", and one or more targets to send matching events to: an AWS Lambda function, an Amazon SQS queue, an AWS Step Functions workflow, another account. The app that emits the event does not know or care who is listening. Adding a new reaction, say a fraud check, means adding a rule, not changing the orders app. AWS services already publish their own events to a default bus (an EC2 machine changed state, a backup finished), and many SaaS tools can send theirs too. EventBridge also includes a **scheduler** for running something at a set time or on a repeating schedule, which replaced the old trick of cron jobs on a server. It used to be called CloudWatch Events, and some older documentation still uses that name. ### An analogy The announcements board of a building. Residents pin notices ("water cut on Tuesday") without knowing who will read them, and each neighbor has asked the concierge to alert them only about the kind of notice they care about. ### Keep in mind - Events are small JSON messages describing something that happened. - Rules match events by pattern and route them to targets like Lambda or SQS. - Senders and receivers stay decoupled; new listeners need no change to the sender. - Includes a scheduler for timed and recurring tasks. ### Go deeper - [What is Amazon EventBridge? (official docs)](https://docs.aws.amazon.com/eventbridge/latest/userguide/eb-what-is.html) - [Amazon EventBridge](https://aws.amazon.com/eventbridge/) --- ## Amazon Lightsail > The simple corner of AWS: a ready-made server for a flat monthly price. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-lightsail/ - In Spanish: [Amazon Lightsail](https://tidbit.legios.com.ar/es/amazon-lightsail/) **Lightsail** is AWS's answer for people who want one server and nothing more complicated. You pick a plan, pick what it should come with (plain Linux or Windows, or a ready-made setup like WordPress), and a minute later you have a running machine with a fixed IP address and a browser-based terminal. It is the AWS version of the classic "virtual private server" sold by hosting companies. The main difference with Amazon EC2 is packaging. With EC2 you choose and pay separately for the machine, the disk, the network traffic and the public address, and you configure a VPC and security rules. With Lightsail all of that comes in one bundle at a predictable monthly price, with a much smaller set of options and a simpler console. It suits a personal blog, a small business website, a test environment or a side project. Lightsail also offers simple managed databases, load balancers and container hosting in the same style. When a project grows beyond it, you can take a snapshot of the server and move it to EC2, where the full range of AWS options is available. ### An analogy A furnished studio with utilities included in the rent, instead of an apartment where you set up electricity, gas and internet one by one. Less flexible, but one bill and you move in today. ### Keep in mind - Virtual servers in fixed-price bundles: machine, disk, transfer and IP together. - One-click setups for common apps like WordPress. - Far fewer options than EC2, on purpose. - Snapshots can be exported to EC2 when a project outgrows it. ### Go deeper - [What is Amazon Lightsail? (official docs)](https://docs.aws.amazon.com/lightsail/latest/userguide/what-is-amazon-lightsail.html) - [Amazon Lightsail](https://aws.amazon.com/lightsail/) --- ## Amazon OpenSearch Service > Managed search engine on AWS, for site search boxes and digging through logs. - Level: Intermediate - Topics: AWS, Data - URL: https://tidbit.legios.com.ar/amazon-opensearch-service/ - In Spanish: [Amazon OpenSearch Service](https://tidbit.legios.com.ar/es/amazon-opensearch-service/) **OpenSearch** is an open source search and analytics engine, the technology behind the search box of many websites. **Amazon OpenSearch Service** runs it for you on AWS: you get a cluster that stays patched and backed up, and you send it documents to make searchable. It grew out of Elasticsearch, and the service was called Amazon Elasticsearch Service until 2021. Its core is an **inverted index**: for every word, a list of the documents where it appears, like the index at the back of a book. That is why it can search millions of product descriptions for "red running shoes" in milliseconds, tolerate typos, and rank the results by relevance, things a regular database does poorly. The other big use is logs. Apps and servers produce endless lines of text, and teams send them to OpenSearch to search for an error message, count failures per minute, or build charts in **OpenSearch Dashboards**, its visual tool. More recently it is also used for vector search, finding documents by meaning rather than exact words, which is a common piece of RAG systems. A serverless option scales capacity without managing a cluster. ### An analogy The index at the back of a giant book. Instead of reading every page to find "volcano", you look it up and jump straight to pages 12, 87 and 341, the most relevant first. ### Keep in mind - Managed OpenSearch, the open source descendant of Elasticsearch. - Uses an inverted index for fast, typo-tolerant, ranked full-text search. - Widely used to search and chart application logs. - Also supports vector search for finding text by meaning. ### Go deeper - [What is Amazon OpenSearch Service? (official docs)](https://docs.aws.amazon.com/opensearch-service/latest/developerguide/what-is.html) - [What is OpenSearch? (AWS)](https://aws.amazon.com/what-is/opensearch/) - [OpenSearch on Wikipedia](https://en.wikipedia.org/wiki/OpenSearch_(software)) --- ## Amazon RDS > A ready-made relational database that AWS keeps patched, backed up and running. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-rds/ - In Spanish: [Amazon RDS](https://tidbit.legios.com.ar/es/amazon-rds/) **RDS** (Relational database Service) gives you a database such as PostgreSQL, MySQL, MariaDB, SQL Server or Oracle without installing or maintaining it yourself. You choose the engine and the size, and AWS hands you an address and a password. Your app connects with SQL exactly as it would to a database on your own server. What you stop doing: installing updates, scheduling backups, replacing a failed disk, getting up at night when it crashes. RDS takes automatic backups, applies patches in a maintenance window you pick, and with **Multi-AZ** keeps a synchronized standby copy in another Availability Zone that takes over in a minute or two if the main one fails. **Aurora** is AWS's own engine, compatible with PostgreSQL and MySQL, built for more speed and resilience. A typical example: an online shop keeps its customers, orders and products in a PostgreSQL database on RDS, and the developers never log into the database server, because there is none to log into. The trade-off is money: you pay for the instance every hour it runs, busy or idle, and it costs more than running the same engine yourself on Amazon EC2. ### An analogy A rented office with cleaning, security and maintenance included. You bring your furniture (the data) and get to work; the landlord fixes the leaks and keeps a spare key in case you lose yours. ### Keep in mind - Managed engines: PostgreSQL, MySQL, MariaDB, SQL Server, Oracle and Aurora. - Automatic backups let you restore to any point within the last 35 days. - Multi-AZ keeps a standby copy that takes over if the main one fails. - You still design the tables and write the queries; RDS only runs the server. ### Go deeper - [What is Amazon RDS? (official docs)](https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/Welcome.html) - [Amazon RDS](https://aws.amazon.com/rds/) --- ## Amazon Redshift > AWS's data warehouse: SQL over huge tables, built for analysis rather than apps. - Level: Intermediate - Topics: AWS, Data - URL: https://tidbit.legios.com.ar/amazon-redshift/ - In Spanish: [Amazon Redshift](https://tidbit.legios.com.ar/es/amazon-redshift/) **Redshift** is the data warehouse of AWS: a database designed to answer analytical questions over very large amounts of data, like "revenue by country and month for the last five years". You load data from your apps, Amazon S3 or other sources, and analysts query it with SQL, usually through a dashboard tool. Two design choices make it fast at that job. It stores data **by column** instead of by row, so a query that only needs `country` and `amount` reads just those two columns, highly compressed. And it splits each table across several machines (**nodes**) that work in parallel: a leader node plans the query, the compute nodes each scan their slice, and the leader combines the answers. You can run it as a provisioned **cluster** with a fixed number of nodes, or as **Redshift Serverless**, where AWS adds capacity while queries run and you pay for the compute actually used. It can also query files that stay in S3 without loading them. What it is not good for is the stream of small reads and writes behind an app's checkout page; that is the job of a regular database like Amazon RDS or Amazon DynamoDB. ### An analogy A team of archivists, each holding one part of a huge archive, sorted by topic instead of by date. Ask a question and every archivist searches their part at the same time while a coordinator puts the pieces together. ### Keep in mind - A columnar data warehouse queried with SQL (its dialect descends from PostgreSQL). - Spreads data across nodes that scan in parallel. - Comes as provisioned clusters or as Redshift Serverless. - Built for analytics over big tables, not for an app's many small transactions. ### Go deeper - [What is Amazon Redshift? (official docs)](https://docs.aws.amazon.com/redshift/latest/mgmt/welcome.html) - [Amazon Redshift](https://aws.amazon.com/redshift/) - [Amazon Redshift on Wikipedia](https://en.wikipedia.org/wiki/Amazon_Redshift) --- ## Amazon Route 53 > AWS's DNS service: it turns your domain name into the address of your servers. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-route-53/ - In Spanish: [Amazon Route 53](https://tidbit.legios.com.ar/es/amazon-route-53/) **Route 53** is the AWS service for DNS, the system that translates a name like `example.com` into an IP address. You register a domain or bring one, create a **hosted zone** for it, and add **records**: 'the website is at this CloudFront distribution', 'mail goes to this server', 'the API is at this load balancer'. When someone types your domain, Route 53 answers in a few milliseconds from servers spread around the world. Beyond plain name-to-address answers, it can route traffic cleverly: send users to the closest Region (latency routing), split them 90/10 between two versions of a site (weighted routing), or check that a server is healthy and stop sending people to it when it is not (**health checks** with failover). The name is a nod to port 53, the port DNS uses. A typical example: a startup buys `miapp.com` through Route 53, points `www` at Amazon CloudFront, `api` at a load balancer in front of Amazon EC2, and sets a failover so that if the main Region goes down, traffic moves to a backup. It costs roughly half a dollar per hosted zone per month plus a small fee per million queries. ### An analogy A phone book that updates itself. You look up a name, it gives you the number, and if that line is down it quietly gives you the backup number instead. ### Keep in mind - A hosted zone holds the DNS records for one domain. - Routing policies: simple, weighted, latency, geolocation, failover. - Health checks let it stop pointing users at a server that is down. - You can register domains here too, but any registrar works with it. ### Go deeper - [What is Amazon Route 53? (official docs)](https://docs.aws.amazon.com/Route53/latest/DeveloperGuide/Welcome.html) - [Amazon Route 53](https://aws.amazon.com/route53/) --- ## Amazon S3 > A giant online drawer for files, rented by the gigabyte. - Level: Basic - Topics: AWS, Cloud - URL: https://tidbit.legios.com.ar/amazon-s3/ - In Spanish: [Amazon S3](https://tidbit.legios.com.ar/es/amazon-s3/) **S3** (Simple Storage Service) is the AWS service for storing files of any size: photos, backups, videos, logs, website pages. You create a **bucket** (a named container) and put **objects** (files) inside it, each with a key that works like a file path. There is no disk to manage and no limit to worry about. AWS copies every object to several machines so it survives hardware failures, and you pay only for the space you use and the data you move out. S3 is the quiet foundation of a lot of the internet: static websites are served from it, data warehouses read from it, and machine learning datasets live in it. ### An analogy A self-storage facility with infinite units. You get a unit (bucket), you drop boxes in it (objects), each box has a label (key), and the facility is responsible for keeping the building standing. ### Keep in mind - Bucket names are unique across all of AWS, worldwide. - Objects are private by default; you choose who can read them. - Cheaper storage classes exist for files you rarely open. ### Go deeper - [What is Amazon S3? (official docs)](https://docs.aws.amazon.com/AmazonS3/latest/userguide/Welcome.html) - [S3 pricing](https://aws.amazon.com/s3/pricing/) --- ## Amazon SageMaker > The AWS workshop for building, training and serving machine learning models. - Level: Intermediate - Topics: AWS, AI - URL: https://tidbit.legios.com.ar/amazon-sagemaker/ - In Spanish: [Amazon SageMaker](https://tidbit.legios.com.ar/es/amazon-sagemaker/) **Amazon SageMaker** is the AWS platform for the whole machine learning lifecycle: preparing data, writing and running notebooks, training a model on GPU machines, and deploying it as an endpoint your app can call. Each step is a managed piece. You say 'train this script on four GPU instances', SageMaker launches them, runs the job, saves the model to Amazon S3 and shuts the machines down. The difference with Amazon Bedrock is who makes the model. Bedrock gives you ready-made foundation models (Claude, Llama, Amazon's own) through an API. SageMaker is for when you train or fine-tune your own, whether a small fraud detector on tabular data or a custom vision model. SageMaker JumpStart sits in between: a catalog of open models you can deploy or fine-tune in a few clicks. Example: a retailer wants to predict which customers will stop buying. A data scientist explores the data in a SageMaker notebook, launches a training job with XGBoost on past purchases, compares a few experiments, and deploys the best model to a real-time endpoint. Each night a batch job scores every customer and writes the result to a table that marketing reads. SageMaker has grown into a large family (Studio, Pipelines, Feature Store, Model Monitor, Ground Truth for labeling). Nobody uses all of it; most teams start with notebooks, training jobs and endpoints. ### An analogy A fully equipped professional kitchen rented by the hour. The ovens (GPUs), the prep stations and the pass are already there; you bring the recipe and the ingredients (your data), and you stop paying when the dish is out. ### Keep in mind - Covers data prep, notebooks, training, deployment and monitoring in one managed service. - Training jobs spin up GPU instances only while they run; endpoints serve predictions over HTTPS. - Bedrock is for using existing foundation models; SageMaker is for training or fine-tuning your own. - JumpStart offers open models ready to deploy or fine-tune. ### Go deeper - [What is Amazon SageMaker? (official docs)](https://docs.aws.amazon.com/sagemaker/latest/dg/whatis.html) - [Amazon SageMaker product page](https://aws.amazon.com/sagemaker/) --- ## Amazon SNS > One message in, many copies out: a megaphone for systems. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-sns/ - In Spanish: [Amazon SNS](https://tidbit.legios.com.ar/es/amazon-sns/) **Amazon SNS** (Simple Notification Service) is a publish/subscribe service. You create a **topic**, several subscribers sign up to it, and every message published to the topic is delivered to all of them at once. A subscriber can be an AWS Lambda function, an Amazon SQS queue, an HTTPS endpoint, an email address or a phone number (SMS). The difference with a queue: Amazon SQS hands each message to one consumer; a topic copies each message to every subscriber. The two are often combined: SNS fans an event out to several SQS queues, each owned by a different team, and each team processes it at its own pace. Example: when an order is paid, the shop publishes 'order 4521 paid' to a topic. The warehouse queue receives it, the invoicing function receives it, the analytics pipeline receives it, and the customer gets an SMS. The checkout code published once and has no idea who is listening. SNS is also how Amazon CloudWatch alarms reach people: the alarm publishes to a topic, and the topic emails or texts whoever is on call. ### An analogy A WhatsApp broadcast list. You write the message once; every contact on the list gets their own copy and does whatever they want with it. ### Keep in mind - Publish once, deliver to every subscriber at the same time (fan-out). - Subscribers can be code (Lambda, SQS, HTTPS) or people (email, SMS, push notifications). - SNS does not keep messages for later: if nobody is subscribed when you publish, the message is gone. - Filters let each subscriber receive only the messages it cares about. ### Go deeper - [What is Amazon SNS? (official docs)](https://docs.aws.amazon.com/sns/latest/dg/welcome.html) - [Amazon SNS product page](https://aws.amazon.com/sns/) --- ## Amazon SQS > A waiting line for messages between programs, so none of them gets lost. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/amazon-sqs/ - In Spanish: [Amazon SQS](https://tidbit.legios.com.ar/es/amazon-sqs/) **Amazon SQS** (Simple Queue Service) is a managed message queue. One program (the producer) drops messages into the queue; another (the consumer) picks them up when it is ready. The two never talk directly and do not even need to be running at the same time. It solves a common problem: one part of a system is faster than another. An online shop takes 5,000 orders in a minute during a sale, but the service that prints shipping labels handles 50 per minute. With a queue in the middle, orders wait in line instead of failing, and the label printer works through them at its own pace. SQS keeps each message until a consumer deletes it. If the consumer crashes halfway, the message becomes visible again after a timeout and another consumer retries it. Messages that fail too many times are moved to a **dead-letter queue** for a human to inspect. It pairs naturally with AWS Lambda: a function can be triggered for every batch of messages that arrives. There are two kinds: standard queues (very high throughput, order mostly preserved, a message may rarely be delivered twice) and FIFO queues (strict order, no duplicates, lower throughput). ### An analogy The number dispenser at a deli counter. Customers take a ticket and wait; the staff calls the next number when free. Nobody gets lost, even when the line is long. ### Keep in mind - Producer and consumer are decoupled: neither needs to know about the other or be online at the same time. - A message stays in the queue until it is explicitly deleted, so a crashed consumer does not lose it. - Standard queues favor throughput; FIFO queues guarantee order and no duplicates. - Messages can wait in a queue for up to 14 days (4 by default). ### Go deeper - [What is Amazon SQS? (official docs)](https://docs.aws.amazon.com/AWSSimpleQueueService/latest/SQSDeveloperGuide/welcome.html) - [Amazon SQS product page](https://aws.amazon.com/sqs/) --- ## Amazon VPC > Your own private slice of the AWS network, with walls you draw yourself. - Level: Intermediate - Topics: AWS, Networks - URL: https://tidbit.legios.com.ar/amazon-vpc/ - In Spanish: [Amazon VPC](https://tidbit.legios.com.ar/es/amazon-vpc/) **Amazon VPC** (Virtual Private Cloud) is the service that gives you an isolated network inside AWS. Everything you launch, an Amazon EC2 server, an Amazon RDS database, a container, lives inside a VPC, with its own range of private IP addresses that nobody else on AWS can reach. Inside the VPC you cut **subnets**: smaller blocks of addresses, each tied to one Availability Zone. A public subnet has a route to the internet; a private one does not. **Security groups** act as a firewall around each machine, saying which ports can be reached and from where. Route tables decide where traffic goes next. A typical setup: a load balancer in a public subnet, application servers in a private subnet, and the database in another private subnet that only those servers can talk to. Even if someone learns the database's address, there is no path to it from outside. Every AWS account comes with a default VPC so you can start without thinking about networking, but production systems almost always design their own. ### An analogy An empty office building you rent. You decide which floors have a street entrance, which are staff-only, and which doors need a key card. ### Keep in mind - A VPC lives in one region and spans its Availability Zones. - Public subnets can reach the internet; private ones cannot unless you add a NAT gateway. - Security groups are per-machine firewalls: closed by default, you open only what you need. - The default VPC is fine for experiments, not for production. ### Go deeper - [What is Amazon VPC? (official docs)](https://docs.aws.amazon.com/vpc/latest/userguide/what-is-amazon-vpc.html) - [Amazon VPC product page](https://aws.amazon.com/vpc/) --- ## AMD > The chip designer that competes with Intel in CPUs and with NVIDIA in GPUs. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/amd/ - In Spanish: [AMD](https://tidbit.legios.com.ar/es/amd/) **AMD** (Advanced Micro Devices) designs both main kinds of processors. Its CPUs are sold as Ryzen for PCs and EPYC for servers; its GPUs as Radeon for gaming and Instinct for AI data centers. It also designs the custom chips inside the PlayStation and Xbox consoles. For most of its history it was the smaller alternative to Intel in x86 processors, the instruction set both companies have the rights to make. AMD is **fabless**: it designs chips but owns no factories. It sold its manufacturing arm in 2009 (it became GlobalFoundries), and today its most advanced chips are made by TSMC. That turned out to be a strength: when TSMC's manufacturing pulled ahead of Intel's, AMD's processors benefited directly, and the company went from near-bankruptcy in the mid-2010s to a major player in servers and PCs. In AI, AMD is the main challenger to NVIDIA. Its Instinct GPUs compete for the same data center budgets, and its software stack, **ROCm**, is its answer to NVIDIA's CUDA. Closing that software gap matters as much as the chips themselves, because most AI code was written with NVIDIA in mind. ### An analogy An architecture studio that draws excellent buildings and hires the best construction company in the world to build them, instead of keeping its own cranes. ### Keep in mind - Designs CPUs (Ryzen, EPYC) and GPUs (Radeon, Instinct), plus console chips. - Fabless: TSMC manufactures its most advanced chips. - Competes with Intel in x86 processors and with NVIDIA in AI hardware. - Its ROCm software is its alternative to NVIDIA's CUDA. ### Go deeper - [AMD on Wikipedia](https://en.wikipedia.org/wiki/AMD) - [About AMD (official site)](https://www.amd.com/en/corporate.html) --- ## Anthropic > The AI company behind Claude, built around making models safe and useful. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/anthropic/ - In Spanish: [Anthropic](https://tidbit.legios.com.ar/es/anthropic/) **Anthropic** is an AI company based in San Francisco, founded in 2021 by a group of researchers who left OpenAI, among them the siblings Dario and Daniela Amodei. It builds **Claude**, a family of large language models, and safety is its stated reason to exist: the founders' bet is that powerful AI is coming either way, and that it is better to have safety-focused labs at the frontier than to leave it to others. What it sells is access to Claude. People and teams use the Claude app (web, desktop, mobile); developers call the models through an API, paying per token, and can also reach them inside Amazon Bedrock and Google Cloud. Claude Code, a tool that lets Claude work inside a software project (reading files, editing code, running tests), became one of its best-known products. The models come in sizes with different price and speed, named Haiku, Sonnet and Opus. Anthropic is also a research lab. It published **Constitutional AI**, a method for training a model to follow a written set of principles; it is a leading group in **interpretability**, the effort to understand what happens inside a neural network; and it maintains a Responsible Scaling Policy that defines which safety measures each level of capability requires. In late 2024 it released the Model Context Protocol (MCP), an open standard for connecting AI models to tools and data that other companies adopted. Why it matters: it is one of a handful of labs training frontier models, Claude is widely used for coding and for enterprise work, and its research and policy positions shape how governments and other labs think about AI risk. Amazon and Google are among its major investors. It is organized as a public benefit corporation, which lets it weigh its mission alongside profit. ### An analogy A pharmaceutical lab that makes the drug and also runs the trials on its side effects. Anthropic sells Claude and spends much of its effort studying what could go wrong with models like it. ### Keep in mind - Founded in 2021 by former OpenAI researchers; based in San Francisco. - Products: the Claude models, the Claude app, the API and Claude Code. - Known for safety research: Constitutional AI, interpretability, the Responsible Scaling Policy. - Backed by Amazon and Google; Claude is also sold through their clouds. ### Go deeper - [anthropic.com](https://www.anthropic.com) - [Claude developer docs](https://docs.anthropic.com) - [Anthropic on Wikipedia](https://en.wikipedia.org/wiki/Anthropic) --- ## Antivirus > A program that watches your files and apps and blocks the ones known to be harmful. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/antivirus/ - In Spanish: [Antivirus](https://tidbit.legios.com.ar/es/antivirus/) An **antivirus** is software that checks what arrives on and runs on your computer or phone, looking for malware: viruses, ransomware, spyware and the like. It scans files when you download or open them, and when it finds something dangerous it blocks it or moves it to **quarantine**, a locked area where it cannot run, so you can delete it or restore it if it was a false alarm. It recognizes threats in two main ways. The classic one is **signatures**: a constantly updated list of fingerprints of known malware, like a wanted-posters board. The newer one is **behavior**: watching what programs do, and raising the alarm when, for example, an unknown app suddenly starts encrypting hundreds of your documents. Signatures catch known threats reliably; behavior analysis catches some new ones, at the cost of occasional false alarms. Today most computers already have one built in. Windows comes with Microsoft Defender turned on, and macOS, iOS and Android check apps in their own ways. In companies the same idea grows into **EDR** tools that also record what happened on each machine so a security team can investigate. An antivirus is a seatbelt, not a guarantee. It cannot stop you from typing your password into a fake site, and brand new malware may slip past it. It works best together with habits: keep the system and apps updated, install software only from official sources, and keep backups. ### An analogy A security guard at a building entrance with a board of wanted photos. He stops anyone who matches, and also watches for people acting strangely in the corridors, but a well disguised newcomer can still get through. ### Keep in mind - Scans files and running programs and quarantines what looks harmful. - Detects known malware by signature and some new malware by behavior. - Windows, macOS and phones already include protection; keep it on and updated. - It does not protect against phishing or bad habits; updates and backups matter as much. ### Go deeper - [Understanding anti-virus software (CISA)](https://www.cisa.gov/news-events/news/understanding-anti-virus-software) - [Antivirus software on Wikipedia](https://en.wikipedia.org/wiki/Antivirus_software) --- ## Apache Spark > An open source engine that splits work on huge datasets across many machines. - Level: Advanced - Topics: Data - URL: https://tidbit.legios.com.ar/apache-spark/ - In Spanish: [Apache Spark](https://tidbit.legios.com.ar/es/apache-spark/) **Apache Spark** is an open source engine for processing data that is too big, or too slow, for one computer. You describe what you want (read these files, filter, join, group, write the result) and Spark splits the data into **partitions**, sends them to many machines in a cluster, runs the steps in parallel and gathers the answer. You can write it in SQL (**Spark SQL**), Python (**PySpark**), Scala, Java or R. It started at UC Berkeley around 2009 and became an Apache project, designed to improve on Hadoop MapReduce, which wrote intermediate results to disk between every step. Spark keeps as much as it can in memory and plans the whole job before running it, which makes multi-step work much faster. A central **driver** builds that plan; **executors** on the worker machines do the actual work, and if one fails, Spark recomputes only the lost pieces. Today Spark is a workhorse of data engineering: nightly ETL jobs over a data lake, cleaning billions of events, preparing training data for machine learning, and processing streams in small batches. You rarely install it by hand; it runs as a managed service in Databricks, Amazon EMR, AWS Glue or Google Cloud Dataproc. For data that fits on one machine, simpler tools are often faster and cheaper. ### An analogy Counting all the votes of a country. One person would take months; instead, each polling station counts its own box at the same time and phones the total to a central office that adds them up. Spark is the system that organizes the stations and the phone calls. ### Keep in mind - Splits data into partitions and processes them in parallel on a cluster. - Keeps intermediate data in memory, which made it much faster than MapReduce. - Used from SQL, Python (PySpark), Scala, Java or R. - Usually run as a managed service, such as Databricks, EMR or Glue. ### Go deeper - [Apache Spark (official site)](https://spark.apache.org/) - [What is Apache Spark? (AWS)](https://aws.amazon.com/what-is/apache-spark/) - [Apache Spark on Wikipedia](https://en.wikipedia.org/wiki/Apache_Spark) --- ## API > The menu of things one program lets another program ask for. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/api/ - In Spanish: [API](https://tidbit.legios.com.ar/es/api/) An **API** (Application Programming Interface) is a set of rules for how two programs talk. One side publishes what you can ask for and in what format; the other side sends requests and gets answers. You never see the kitchen, only the menu and the dishes. On the web, most APIs work over HTTP: your app sends a request to a URL like `/weather?city=Rosario` and receives data back, usually as JSON. Maps in a food delivery app, the login with Google button, a chatbot inside a website: all APIs. APIs are how companies sell software to other software. AWS, Stripe, and Anthropic are, at their core, collections of APIs with a bill attached. ### An analogy A restaurant counter. You do not walk into the kitchen; you order from the menu, in the restaurant's language, and your plate comes out the window. ### Keep in mind - An API defines what you can ask, how to ask it, and what you get back. - Web APIs usually use HTTP and return JSON. - An API key is the credential that says who is asking (and who pays). ### Go deeper - [What is an API? (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Extensions/Client-side_APIs/Introduction) - [Example: the Claude API reference](https://docs.anthropic.com/en/api/messages) --- ## API key > A secret string that identifies your app when it calls someone else's API. - Level: Intermediate - Topics: Security, Software - URL: https://tidbit.legios.com.ar/api-key/ - In Spanish: [Clave de API](https://tidbit.legios.com.ar/es/api-key/) An **API key** is a long, random-looking string, something like `sk_live_4f9a…`, that a service gives you so your program can use its API. Every request your code sends includes the key, usually in a header, and the service uses it to know who is calling: which account to bill, which limits apply and what that caller is allowed to do. Because whoever has the key can act as you, it works like a password. If one leaks, a stranger can run up your bill on a paid AI or maps API, read your data or send messages in your name. The most common leak is a key written directly in the code and pushed to a public GitHub repository; automated bots scan for exactly that within minutes. Good habits follow from that. Keep keys out of the code, in environment variables or a secrets manager such as AWS Secrets Manager. Never put a secret key in a website's or mobile app's code, since anyone can read it there; call the API from your server instead. Give each key only the permissions it needs, use separate keys for testing and production, and **rotate** them: create a new one, switch over, and revoke the old. Keys identify a program; for acting on behalf of a person, services use OAuth tokens instead. ### An analogy Like the badge a contractor gets at an office building. It opens the doors they need and logs every entry under their company's name, so if they lose it, you deactivate it and print a new one. ### Keep in mind - Identifies the calling app and its account, often for billing and limits. - Treat it like a password: anyone holding it can use it. - Store it in environment variables or a secrets manager, never in code. - If it leaks, revoke it and create a new one right away. ### Go deeper - [What is an API key? (AWS)](https://aws.amazon.com/what-is/api-key/) - [Manage API keys (Google Cloud docs)](https://cloud.google.com/docs/authentication/api-keys) - [Secrets Management Cheat Sheet (OWASP)](https://cheatsheetseries.owasp.org/cheatsheets/Secrets_Management_Cheat_Sheet.html) --- ## Apple > The company that sells the iPhone and the Mac, plus the services that run on them. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/apple/ - In Spanish: [Apple](https://tidbit.legios.com.ar/es/apple/) **Apple** was founded in 1976 by Steve Jobs and Steve Wozniak in California, and its first business was personal computers. Today it is best known for the **iPhone**, launched in 2007, which is still the product that brings in the most money. Around it sit the Mac, the iPad, the Apple Watch and AirPods, all designed in Cupertino and assembled by partner factories, mostly in Asia. What makes Apple different is that it builds the hardware and the software together: it writes the operating systems (iOS, macOS) for its own devices only, and designs its own chips. Since 2020 the Mac uses **Apple silicon**, chips based on the Arm design that Apple engineers and TSMC manufactures, replacing the Intel processors it used before. Controlling every layer lets it tune battery life, speed and privacy features across the whole device. The second business is **services**: the App Store, where it takes a commission on many app sales and subscriptions, plus iCloud storage, Apple Music, Apple TV+, Apple Pay and AppleCare. Services grow with every device already in people's hands, and the ecosystem (photos, messages and passwords syncing between your devices) makes switching to another brand inconvenient, which is part of the strategy. ### An analogy A furniture brand that also owns the house it fits in. Each piece is designed to match the others and the rooms, so everything works smoothly together, and moving one piece to another house is more trouble than it seems. ### Keep in mind - Founded in 1976 by Steve Jobs and Steve Wozniak; based in Cupertino, California. - The iPhone is its largest product; Mac, iPad, Watch and AirPods complete the line. - It designs its own chips (Apple silicon), which TSMC manufactures. - Services such as the App Store and iCloud are its second big source of money. ### Go deeper - [Apple Newsroom](https://www.apple.com/newsroom/) - [Apple Inc. on Wikipedia](https://en.wikipedia.org/wiki/Apple_Inc.) --- ## Arm > The British company that designs the processor blueprints inside almost every phone. - Level: Intermediate - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/arm/ - In Spanish: [Arm](https://tidbit.legios.com.ar/es/arm/) **Arm** (Arm Holdings) designs processor architectures and licenses them; it does not manufacture or sell chips. Its product is intellectual property: the **ARM architecture**, the instruction set a processor understands, and ready-made CPU core designs. Companies like Apple, Qualcomm, Samsung and MediaTek pay Arm for a license, build their own chips with it, and have them manufactured by a foundry such as TSMC. ARM processors were built from the start to use little power, which is why nearly every smartphone and tablet runs on one. That efficiency has since moved up: Apple's M-series Mac chips, Amazon's Graviton servers on AWS and many other data center chips are ARM-based. This is the 'ARM vs x86' contest: x86, from Intel and AMD, ruled PCs and servers for decades, and ARM keeps winning ground there. Arm earns money in two ways: a license fee to use its designs, and a small **royalty** for every chip shipped with them. Each royalty is tiny, but tens of billions of ARM chips ship every year, in phones, cars, TVs and appliances. Founded in Cambridge, England, in 1990, Arm is majority-owned by the Japanese group SoftBank. ### An analogy A company that writes the rules of a language and sells dictionaries. It never writes the books itself, but almost every author in a huge market writes in its language and pays a little for each copy sold. ### Keep in mind - Licenses processor designs; never manufactures chips. - Its customers (Apple, Qualcomm, Samsung) build ARM chips and a foundry makes them. - ARM rules phones for its low power use and is growing in laptops and servers. - Earns license fees plus a royalty per chip shipped. ### Go deeper - [Arm, company overview (official site)](https://www.arm.com/company) - [Arm Holdings on Wikipedia](https://en.wikipedia.org/wiki/Arm_Holdings) - [ARM architecture family on Wikipedia](https://en.wikipedia.org/wiki/ARM_architecture_family) --- ## Artificial intelligence > Getting computers to do things that, until recently, only people could do. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/artificial-intelligence/ - In Spanish: [Inteligencia artificial](https://tidbit.legios.com.ar/es/artificial-intelligence/) **Artificial intelligence** (AI) is the broad name for building computer systems that do tasks we associate with human thinking: understanding language, recognizing faces, translating, playing chess, writing a summary, spotting a tumor in a scan. The term was coined in 1956, at a summer workshop at Dartmouth College, and the field has gone through several waves of excitement and disappointment since. Early AI was mostly hand-written rules: if the patient has a fever and a rash, consider measles. That works for narrow problems but breaks in the messy real world. Almost all of today's AI is **machine learning** instead: rather than writing the rules, engineers show the system millions of examples and let it find the patterns itself. Deep learning, which uses large neural networks, is the branch behind the recent leaps. When people say 'AI' today they usually mean one specific kind: large language models like Claude, ChatGPT or Gemini, which can hold a conversation and write. But AI also sorts your spam, suggests the next song, unlocks your phone with your face and routes your delivery. It is a set of statistical tools, very powerful in some tasks and surprisingly fragile in others, not a single thinking machine. ### An analogy AI is a word like 'medicine': it names a whole field, not one pill. Machine learning, chatbots and face recognition are specific treatments inside it, each good for some things and useless for others. ### Keep in mind - AI is the umbrella; machine learning and deep learning are branches inside it. - Most modern AI learns patterns from data instead of following hand-written rules. - Today's systems are good at specific tasks, not general-purpose minds. - It can be confidently wrong, so its output still needs checking. ### Go deeper - [Artificial intelligence on Wikipedia](https://en.wikipedia.org/wiki/Artificial_intelligence) - [What is artificial intelligence? (IBM)](https://www.ibm.com/think/topics/artificial-intelligence) --- ## ASML > The Dutch company that builds the machines every advanced chip factory depends on. - Level: Intermediate - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/asml/ - In Spanish: [ASML](https://tidbit.legios.com.ar/es/asml/) **ASML** is a Dutch company that builds **lithography** machines: the equipment that prints circuit patterns onto silicon wafers. Making a chip involves many steps, but lithography decides how small the transistors can be. The machine shines light through a mask holding the circuit's pattern, shrinks it with optics and projects it onto the wafer, layer after layer. Its most advanced machines use **EUV** (extreme ultraviolet) light, with a wavelength far shorter than visible light, which is what makes today's smallest transistors possible. ASML is the only company in the world that makes EUV machines. Each one is the size of a bus, costs roughly a couple of hundred million dollars, ships in many containers and takes months to install. TSMC, Samsung and Intel all depend on them for their leading-edge chips. That monopoly turns ASML into a geopolitical piece. The machines rely on parts from a network of European, American and Japanese suppliers, and the Dutch government, coordinating with the United States, restricts which machines can be sold to China. Whoever cannot buy EUV machines cannot make the most advanced chips, whatever their designs. ### An analogy The only maker of the printing presses fine enough to print the smallest letters in the world. Authors and printers compete fiercely, but all of them buy their presses from the same workshop. ### Keep in mind - Builds lithography machines, which print circuit patterns onto silicon wafers. - The only company that makes EUV machines, needed for the most advanced chips. - TSMC, Samsung and Intel are its key customers. - Export rules on its machines shape who can make cutting-edge chips. ### Go deeper - [ASML, about the company (official site)](https://www.asml.com/en/company) - [ASML Holding on Wikipedia](https://en.wikipedia.org/wiki/ASML_Holding) - [Extreme ultraviolet lithography on Wikipedia](https://en.wikipedia.org/wiki/Extreme_ultraviolet_lithography) --- ## Autoscaling > Adding and removing servers automatically as demand goes up and down. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/autoscaling/ - In Spanish: [Autoescalado](https://tidbit.legios.com.ar/es/autoscaling/) **Autoscaling** is a cloud feature that changes the number of servers running an application on its own, based on how busy it is. When traffic rises, it adds copies (**scaling out**); when traffic falls, it removes them (**scaling in**). A load balancer in front spreads requests across whatever copies exist at the moment. You set the rules: keep average CPU around 50%, never run fewer than two servers or more than twenty, add capacity every weekday at 8 a.m. before the rush. The system watches metrics like CPU, memory, request count or queue length and acts when they cross the line. On AWS this is done with **Auto Scaling groups** for EC2; Kubernetes does the same for containers, and serverless platforms like AWS Lambda scale by themselves without any of this setup. The point is to pay for what you use without falling over at the peak. A ticket site can run a few servers at night and dozens when a big concert goes on sale. It only works well if the app is designed for it: any copy must be able to handle any request, with no important data kept only on one server, and new copies must start quickly. Scaling also takes time, so a sudden spike can still hit before new servers are ready. ### An analogy A supermarket that opens more checkouts when the lines get long and closes them when the store empties. Customers never wait too long, and the store does not pay cashiers to stand at empty tills. ### Keep in mind - Adds servers when load rises and removes them when it falls. - Driven by rules on metrics like CPU, requests or queue length, plus minimums and maximums. - Works with a load balancer and apps where any copy can serve any request. - New capacity takes time to start, so sudden spikes still need headroom. ### Go deeper - [What is Amazon EC2 Auto Scaling? (AWS docs)](https://docs.aws.amazon.com/autoscaling/ec2/userguide/what-is-amazon-ec2-auto-scaling.html) - [Autoscaling workloads (Kubernetes docs)](https://kubernetes.io/docs/concepts/workloads/autoscaling/) - [Autoscaling on Wikipedia](https://en.wikipedia.org/wiki/Autoscaling) --- ## AWS account and billing > What an AWS account is, how you get charged, and how to avoid a surprise bill. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-account-and-billing/ - In Spanish: [Cuenta y facturación de AWS](https://tidbit.legios.com.ar/es/aws-account-and-billing/) An **AWS account** is the container for everything you create in AWS: it has a twelve-digit id, an email, a **root user** that can do anything, and one monthly bill. Companies usually have many accounts (one per team or per environment, like test and production) grouped under AWS Organizations, so a mistake in one does not touch the others and each one's cost is visible on its own. Billing is **pay as you go**: no upfront payment, no contract. Each service is metered in its own unit: Amazon EC2 by the second an instance runs, Amazon S3 by gigabyte stored and gigabyte sent out, AWS Lambda by request and millisecond. At the end of the month everything is added up and charged to your card. Prices vary by Region, and sending data out of AWS to the internet is the line people most often forget to budget. New accounts get a **Free Tier**: a starter allowance to try many services at no cost, within limits, plus small always-free quotas on services like Lambda and Amazon DynamoDB (the exact rules change, so check the current ones). The important habit: set a **budget** with an email alert on day one, because a forgotten instance or a public bucket serving a viral file keeps metering while you sleep. For steady workloads, Savings Plans and Reserved Instances cut the price in exchange for a one- or three-year commitment. ### An analogy A utility bill, electricity rather than rent. Nobody asks for a deposit, every outlet has its own meter, and the bill at the end of the month depends on what you left switched on. ### Keep in mind - One account means one bill, one root user and one set of resources. - Everything is metered: seconds, gigabytes, requests. Nothing is flat-rate. - Set a budget alert the day you open the account. - Steady workloads get discounts through Savings Plans or Reserved Instances. ### Go deeper - [What is AWS Billing? (official docs)](https://docs.aws.amazon.com/awsaccountbilling/latest/aboutv2/billing-what-is.html) - [AWS Free Tier](https://aws.amazon.com/free/) - [AWS pricing](https://aws.amazon.com/pricing/) --- ## AWS Amplify > A shortcut for building and publishing web and mobile apps on AWS. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-amplify/ - In Spanish: [AWS Amplify](https://tidbit.legios.com.ar/es/aws-amplify/) **AWS Amplify** is a set of tools for people who want to build an app, not learn fifty AWS services first. It covers the two halves of a typical web or mobile app: putting the frontend online, and giving it a backend with sign-in, a database and file uploads. The hosting part is what most people meet first. You connect a Git repository, and every time you push a change Amplify builds the site and publishes it at a URL with HTTPS, served through a global CDN. Each branch can get its own preview address, so a teammate can try a change before it goes live. It works with common frameworks like React, Next.js or Vue. For the backend, Amplify sets up real AWS services behind the scenes: Amazon Cognito for user accounts, a database on Amazon DynamoDB, file storage on Amazon S3. You describe what you need in a few lines of code, and its client libraries let the app sign users in or save a record without writing that plumbing yourself. The trade-off is the usual one for shortcuts: it is fast to start, and very custom architectures may eventually outgrow it. ### An analogy A furnished apartment instead of an empty plot of land. You move in and start living right away; the wiring and plumbing are already there, even if you cannot easily move the walls. ### Keep in mind - Hosts web apps straight from a Git repository, rebuilding on every push. - Sets up sign-in, data and storage using Cognito, DynamoDB and S3. - Gives each branch its own preview URL. - Aimed at frontend and full-stack developers who want speed over fine control. ### Go deeper - [What is AWS Amplify Hosting? (official docs)](https://docs.aws.amazon.com/amplify/latest/userguide/welcome.html) - [AWS Amplify documentation](https://docs.amplify.aws/) - [AWS Amplify](https://aws.amazon.com/amplify/) --- ## AWS Backup > One place to schedule, keep and restore backups of your AWS data. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-backup/ - In Spanish: [AWS Backup](https://tidbit.legios.com.ar/es/aws-backup/) Each AWS storage service has its own way of taking snapshots: Amazon EBS disks, Amazon RDS databases, Amazon DynamoDB tables, file systems, S3 buckets. **AWS Backup** puts them all under one roof, so instead of remembering a dozen different settings, you write one set of rules and see every backup in a single console. The rules live in a **backup plan**: how often to copy (every night, every hour), how long to keep each copy (35 days, 7 years), and where to store it. You then say which resources the plan covers, often simply "everything tagged `backup=daily`", so new databases get protected the moment someone tags them. Copies go into a **backup vault**, and can also be copied automatically to another Region or another account, in case a whole Region has trouble or an account is compromised. That last point matters because of ransomware. An attacker who gains admin access may try to delete the backups before encrypting everything else. **Vault Lock** makes a vault write-once: until the retention period ends, nobody, not even the account's administrators, can delete or shorten those copies. A backup is only real once you have restored it, so AWS Backup also lets you run restore tests on a schedule. ### An analogy A building's caretaker who, every night, photocopies the important papers of every office, files the copies in a fireproof safe across town, and has a rule that nothing in that safe can be shredded before its date. ### Keep in mind - Centralizes backups of EBS, RDS, DynamoDB, file systems, S3 and more. - A backup plan sets frequency, retention and which resources, often by tag. - Copies can go to another Region or account for disaster recovery. - Vault Lock makes backups undeletable until they expire, a defense against ransomware. ### Go deeper - [What is AWS Backup? (official docs)](https://docs.aws.amazon.com/aws-backup/latest/devguide/whatisbackup.html) - [AWS Backup](https://aws.amazon.com/backup/) --- ## AWS CDK > Describe your AWS infrastructure in a real programming language instead of YAML. - Level: Advanced - Topics: AWS, Cloud - URL: https://tidbit.legios.com.ar/aws-cdk/ - In Spanish: [AWS CDK](https://tidbit.legios.com.ar/es/aws-cdk/) **AWS CDK** (Cloud Development Kit) is an infrastructure as code tool where you describe servers, buckets, queues and permissions in TypeScript, Python, Java, C# or Go. Under the hood it does not talk to AWS directly: it **synthesizes** your program into an AWS CloudFormation template, and CloudFormation creates or updates the resources. The two key commands are `cdk synth`, which produces the template, and `cdk deploy`, which applies it. Writing a real program brings the things programmers are used to: loops, functions, autocompletion, type checks, tests, and packages to share code. A few lines like `new s3.Bucket(this, 'Uploads', { versioned: true })` can expand into dozens of lines of template. Ten similar environments become one loop instead of ten copied files. The building blocks are **constructs**, which come in levels. Low-level ones map one-to-one to CloudFormation resources. Higher-level ones bundle sensible defaults (a bucket that is encrypted and blocks public access unless you say otherwise), and the highest-level ones assemble whole patterns, like a load-balanced container service. One helper worth knowing: `bucket.grantRead(fn)` writes the IAM policy for you, scoped to exactly that bucket. Each account and Region needs a one-time `cdk bootstrap` before the first deploy. ### An analogy Building with prefabricated modules instead of drawing every brick. You say "a kitchen here" and the module arrives with plumbing and wiring already done; the city still inspects the final blueprint, which in this case is CloudFormation. ### Keep in mind - Infrastructure in TypeScript, Python, Java, C# or Go. - Synthesizes to CloudFormation templates; CloudFormation does the actual deploy. - Constructs package resources with safe defaults and reusable patterns. - Helpers like `grantRead` generate least-privilege IAM permissions. ### Go deeper - [What is the AWS CDK? (official docs)](https://docs.aws.amazon.com/cdk/v2/guide/home.html) - [AWS Cloud Development Kit](https://aws.amazon.com/cdk/) --- ## AWS CloudFormation > Describe your AWS infrastructure in a text file; AWS builds it exactly like that. - Level: Intermediate - Topics: AWS, Cloud - URL: https://tidbit.legios.com.ar/aws-cloudformation/ - In Spanish: [AWS CloudFormation](https://tidbit.legios.com.ar/es/aws-cloudformation/) **AWS CloudFormation** is the AWS service for infrastructure as code. Instead of clicking through the console to create a VPC, two servers, a database and the permissions between them, you write a **template** (a YAML or JSON file) that lists those resources and their settings. You hand the template to CloudFormation and it creates everything, in the right order, as a **stack**. The value shows up the second time. Need the same environment for testing? Deploy the same template under another name. Need a bigger database? Edit the file, update the stack, and CloudFormation works out the difference and applies only that. Did the change break? It rolls back to the previous state. The template lives in Git, so every change to the infrastructure has an author, a review and a history. Example: a 150-line template defines an AWS Lambda function, its IAM role, an Amazon API Gateway in front and an Amazon DynamoDB table. One command (`aws cloudformation deploy`) creates all four in a couple of minutes; deleting the stack deletes all four, with nothing forgotten and still billing. Raw templates get verbose, so many teams generate them: the AWS CDK lets you describe the same resources in TypeScript or Python code that compiles to CloudFormation, and AWS SAM is a shorthand for serverless apps. Terraform is the popular alternative that works across clouds. ### An analogy A LEGO instruction booklet. Anyone with the booklet builds the same model, and when you want a second one you do not design it again, you follow the booklet. ### Keep in mind - A template (YAML or JSON) describes resources; a stack is the deployed result. - Updates are diffs: CloudFormation changes only what differs, and rolls back on failure. - Deleting a stack removes every resource it created. - CDK and SAM generate CloudFormation templates from code; Terraform is the multi-cloud alternative. ### Go deeper - [What is AWS CloudFormation? (official docs)](https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/Welcome.html) - [AWS CloudFormation product page](https://aws.amazon.com/cloudformation/) --- ## AWS CloudTrail > The logbook of an AWS account: who did what, when, and from where. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/aws-cloudtrail/ - In Spanish: [AWS CloudTrail](https://tidbit.legios.com.ar/es/aws-cloudtrail/) Everything that happens in AWS, whether someone clicks in the console, a script runs a command or a service acts on its own, is an API call. **CloudTrail** records those calls. Each **event** says who made it (which user or role), what they did (`DeleteBucket`, `RunInstances`), on which resource, when, from which IP address, and whether it succeeded. That record answers the questions that come up after something goes wrong. Who deleted this database? When was this firewall rule opened to the internet? Did anyone use the leaked access key, and for what? Security teams also feed CloudTrail into alerting, so that a sensitive action, like disabling logging or creating a new admin user, sets off an alarm right away. It is on by default: the console shows the recent history of management actions (roughly the last 90 days) at no cost. To keep events longer, or to record finer-grained activity like every file read in Amazon S3, you create a **trail** that delivers the logs to an S3 bucket, or you use CloudTrail Lake to store and query them with SQL. Many companies send the trails of all their accounts to one locked-down account, so an intruder cannot erase their own tracks. ### An analogy The visitor log and the security cameras of a building, combined. Every door opened is written down with a name and a time, so when something goes missing you can rewind and see who was there. ### Keep in mind - Records API calls: who, what, which resource, when, from where, and the result. - The recent history of management events is available by default. - A trail stores events long-term in Amazon S3; CloudTrail Lake lets you query them with SQL. - Essential for audits, incident investigation and compliance. ### Go deeper - [What is AWS CloudTrail? (official docs)](https://docs.aws.amazon.com/awscloudtrail/latest/userguide/cloudtrail-user-guide.html) - [AWS CloudTrail](https://aws.amazon.com/cloudtrail/) --- ## AWS Glue > AWS's service for finding, describing and cleaning up data, with no servers to run. - Level: Intermediate - Topics: AWS, Data - URL: https://tidbit.legios.com.ar/aws-glue/ - In Spanish: [AWS Glue](https://tidbit.legios.com.ar/es/aws-glue/) **AWS Glue** is the AWS service for ETL: taking data from where it lands, cleaning and reshaping it, and putting it where people can analyze it. A typical case: every night an app drops thousands of JSON files into Amazon S3, and a **Glue job** reads them, removes duplicates, fixes dates, joins them with a customer table and writes the result as Parquet files ready for queries. Its second half is the **Glue Data Catalog**, a register of every dataset: where the files live, what columns they have and of what type. A **crawler** can fill it for you by opening the files and guessing their structure. The catalog is what lets Amazon Athena or Amazon Redshift query a folder of files in S3 as if it were a table. Glue is serverless: you write the transformation (in Python or SQL on top of Apache Spark, or by dragging boxes in a visual editor), say when it should run, and AWS spins up the machines for the duration of the job. You pay for the processing time used, not for idle clusters. That makes it the glue, as the name says, of many data lakes built on AWS. ### An analogy A librarian and a bookbinder in one. The librarian walks the shelves writing down what each book contains; the bookbinder takes loose, messy pages and turns them into clean volumes that readers can actually use. ### Keep in mind - Runs ETL jobs on Apache Spark without you managing a cluster. - The Data Catalog describes datasets so Athena and Redshift can query them as tables. - Crawlers inspect files and infer their columns and types automatically. - Billed by processing time while jobs run; nothing runs between jobs. ### Go deeper - [What is AWS Glue? (official docs)](https://docs.aws.amazon.com/glue/latest/dg/what-is-glue.html) - [AWS Glue](https://aws.amazon.com/glue/) - [What is ETL? (AWS)](https://aws.amazon.com/what-is/etl/) --- ## AWS KMS > AWS's vault for encryption keys: you use them, but they never leave. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/aws-kms/ - In Spanish: [AWS KMS](https://tidbit.legios.com.ar/es/aws-kms/) Encryption is only as good as the protection of its keys. Store the key next to the data and a thief takes both. **KMS** (Key Management Service) solves that on AWS: it creates and keeps encryption keys inside dedicated, tamper-resistant hardware, and you never get to see or download the key itself. You send data to KMS and ask it to encrypt or decrypt, and IAM decides who is allowed to ask. For large files it uses a trick called **envelope encryption**. KMS generates a fresh **data key**, which encrypts the file locally; then the KMS key encrypts that data key, and the encrypted data key is stored next to the file. To read the file, a service asks KMS to unlock the small data key. The big file never travels to KMS, and the master key never leaves it. Most AWS services plug into KMS with a checkbox: Amazon S3 buckets, Amazon EBS disks, Amazon RDS databases and many more encrypt their data with a KMS key. Every use of a key is logged in AWS CloudTrail, so you can see who decrypted what. And because the key is the only way back to the data, deleting a KMS key is deliberately slow: AWS makes you wait a period of days first, in case it was a mistake. ### An analogy A bank vault with a teller window. You never get the master key; you hand the teller your small lockbox key, they lock or unlock it behind the glass, and every visit is written in the bank's ledger. ### Keep in mind - Keys are created and used inside KMS; their raw material never leaves. - Envelope encryption: a data key encrypts the data, a KMS key encrypts the data key. - Built into S3, EBS, RDS and most other AWS services. - IAM controls who can use each key, and CloudTrail logs every use. ### Go deeper - [AWS Key Management Service (official docs)](https://docs.aws.amazon.com/kms/latest/developerguide/overview.html) - [AWS KMS](https://aws.amazon.com/kms/) - [Key management on Wikipedia](https://en.wikipedia.org/wiki/Key_management) --- ## AWS Lambda > Run a piece of code only when something happens, with no server to manage. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-lambda/ - In Spanish: [AWS Lambda](https://tidbit.legios.com.ar/es/aws-lambda/) **Lambda** lets you upload a function (a chunk of code in Python, JavaScript, Java and other languages) and tell AWS when to run it: when a file lands in Amazon S3, when an HTTP request arrives through Amazon API Gateway, when a message shows up in an Amazon SQS queue, or every day at 3 a.m. AWS finds a machine, runs your code, and throws the machine away. You never see a server. You pay per request and per millisecond of execution. If nobody calls the function, the bill is zero. If a million people call it at once, AWS runs thousands of copies in parallel without you changing anything. The limits: one run lasts at most 15 minutes, and the first call after a quiet period is a bit slower (a **cold start**) while AWS prepares a fresh copy. This is what people mean by serverless: there are still servers, but someone else owns them. A typical example: a photo app uploads an image to S3, which triggers a Lambda function that makes a thumbnail and stores it back. No machine sits idle waiting for the next photo. ### An analogy A plumber on call instead of one on your payroll. You pay only for the visits, you do not care where they park the van, and if ten pipes burst at once the company sends ten plumbers. ### Keep in mind - A function runs in response to an event: an upload, a request, a schedule. - Billing is per request and per millisecond; idle time costs nothing. - One execution lasts up to 15 minutes; longer jobs belong somewhere else. - Scales on its own, from zero to thousands of parallel copies. ### Go deeper - [What is AWS Lambda? (official docs)](https://docs.aws.amazon.com/lambda/latest/dg/welcome.html) - [AWS Lambda](https://aws.amazon.com/lambda/) --- ## AWS Organizations > One place to group, govern and pay for many AWS accounts at once. - Level: Intermediate - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-organizations/ - In Spanish: [AWS Organizations](https://tidbit.legios.com.ar/es/aws-organizations/) Companies rarely use a single AWS account. They keep one for production, one for testing, one per team or per client, because an account is the strongest wall AWS offers: what happens in one does not touch the others. **AWS Organizations** is the service that ties all those accounts together under one roof. You create an organization from a **management account** and invite or create **member accounts**. You can arrange them in a tree of **organizational units** (OUs), like folders: `Production`, `Sandbox`, `Finance`. All of them land on one bill (**consolidated billing**), so the company pays once and sees the cost of each account separately. The most powerful piece is the **service control policy** (SCP): a rule attached to an OU or account that sets the maximum of what anyone inside can do. For example, "no one in Sandbox can turn off CloudTrail" or "only these Regions are allowed". It does not grant permissions; it caps them, even for the account's own administrators. That is how a security team sets guardrails across hundreds of accounts without logging into each one. ### An analogy A company with many branches. Each branch runs its own shop, but they share one accounting department, and head office can issue rules ("no branch opens after midnight") that no branch manager can override. ### Keep in mind - Groups many AWS accounts under one management account. - Accounts are arranged in organizational units, like folders. - Consolidated billing puts every account on a single bill. - Service control policies cap what accounts can do; they never grant access. ### Go deeper - [What is AWS Organizations? (official docs)](https://docs.aws.amazon.com/organizations/latest/userguide/orgs_introduction.html) - [AWS Organizations](https://aws.amazon.com/organizations/) --- ## AWS Secrets Manager > A locked vault for passwords and API keys, so they never live in your code. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/aws-secrets-manager/ - In Spanish: [AWS Secrets Manager](https://tidbit.legios.com.ar/es/aws-secrets-manager/) **AWS Secrets Manager** stores the sensitive values an application needs: database passwords, API keys for third-party services, private keys, tokens. You save a secret under a name such as `prod/payments/stripe-key`, and your code asks for it by name at runtime through the AWS API, with IAM deciding who is allowed to read it. The value is encrypted at rest and never printed in logs. The problem it attacks is secrets in the wrong place: inside source code, in a config file committed to Git, in an environment variable visible to anyone who can log in to the server. Once a password reaches a repository it is effectively public, even after you delete it. Secrets Manager also **rotates** secrets: for an Amazon RDS database it can change the password every 30 days on its own and update the stored value, so a leaked credential has a short life. Example: an AWS Lambda function needs to connect to a database. Its IAM role allows `secretsmanager:GetSecretValue` on one specific secret. On startup the function fetches the secret, connects, and caches it for a few minutes. When the password rotates, the next fetch gets the new one. No developer ever sees the production password. AWS also offers Systems Manager Parameter Store, a simpler and cheaper place for configuration values (and for secrets that do not need rotation). Secrets Manager is the choice when rotation and auditing matter. ### An analogy The key safe at a hotel reception. Staff who need the master key sign it out and return it; nobody carries it around in a pocket or writes the combination on a sticky note. ### Keep in mind - Secrets are encrypted, read by name through the API, and access is controlled by IAM. - Automatic rotation for RDS, Redshift and DocumentDB; custom rotation through a Lambda function for anything else. - Every read is recorded in CloudTrail, so you know who accessed what and when. - Priced per secret per month plus API calls; Parameter Store is the cheaper option for plain configuration. ### Go deeper - [What is AWS Secrets Manager? (official docs)](https://docs.aws.amazon.com/secretsmanager/latest/userguide/intro.html) - [AWS Secrets Manager product page](https://aws.amazon.com/secrets-manager/) --- ## AWS Shield > AWS's protection against DDoS attacks, on by default for everyone. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/aws-shield/ - In Spanish: [AWS Shield](https://tidbit.legios.com.ar/es/aws-shield/) A **DDoS** attack tries to knock a site offline by burying it in traffic from thousands of hijacked machines at once. **AWS Shield** is the AWS service that absorbs and filters those floods before they reach your servers. It works at the edges of the AWS network, where there is far more capacity than any single website has. It comes in two tiers. **Shield Standard** is free and automatic for every AWS customer: it protects against the most common network-level attacks, like floods of junk packets or half-open connections, without you doing anything. **Shield Advanced** is a paid subscription for organizations that are likely targets. It adds detection tuned to your own traffic, detailed reports during an attack, access to an AWS response team that can step in around the clock, and protection against the surprise bill when an attack makes your services scale up. Shield and AWS WAF are usually used together. Shield handles brute-force volume at the network level; WAF handles smarter attacks that look like real web requests, such as one bot hammering the login page. Putting the site behind Amazon CloudFront helps both, since traffic then meets AWS's large edge network first. ### An analogy A sea wall in front of a harbor town. The ordinary wall is there for everyone and stops everyday waves; towns that expect storms pay for a taller wall plus a crew on call and insurance for the damage. ### Keep in mind - Defends against DDoS floods before they reach your resources. - Shield Standard is free and on for every AWS account. - Shield Advanced adds tailored detection, a response team and cost protection. - Pairs with AWS WAF for attacks that look like normal web requests. ### Go deeper - [How AWS Shield works (official docs)](https://docs.aws.amazon.com/waf/latest/developerguide/shield-chapter.html) - [AWS Shield](https://aws.amazon.com/shield/) --- ## AWS Step Functions > A flowchart that runs: it chains steps, retries failures and remembers where it was. - Level: Advanced - Topics: AWS - URL: https://tidbit.legios.com.ar/aws-step-functions/ - In Spanish: [AWS Step Functions](https://tidbit.legios.com.ar/es/aws-step-functions/) **AWS Step Functions** orchestrates processes that have many steps. You describe a **state machine**: step A, then B; if B fails retry it three times; then C and D in parallel; wait 24 hours; then E. Each step calls something, most often an AWS Lambda function, but also Amazon ECS tasks, Amazon SQS, Amazon DynamoDB and most AWS APIs. Step Functions runs the flow, tracks exactly where every execution is, and shows it as a live diagram. The problem it solves: once a process has several steps, someone has to answer 'what if step 3 fails after step 2 already charged the card?'. Writing that logic inside one big function is fragile, and a crashed process forgets everything. A state machine keeps its state outside your code, so an execution can pause for days and pick up again, and a failure in one step triggers the retry or the compensation you defined. Example: a loan application. Validate the form (Lambda), run a credit check (an external API with a 2-minute timeout and retries), then in parallel calculate the offer and generate the PDF. Then wait for the applicant to sign, up to 7 days, paused at no cost. Then open the account. If any step fails for good, a notification goes out through Amazon SNS. Definitions are written in a JSON dialect called Amazon States Language, or drawn in the visual Workflow Studio. Standard workflows can run for up to a year; Express workflows are for short, high-volume flows that finish in under five minutes. ### An analogy A production-line foreman with a clipboard. The foreman builds nothing, but knows which station every product is at, sends it back when a station fails, and never loses track even when the shift changes. ### Keep in mind - A state machine defines sequence, choices, parallel branches, waits and retries. - State lives in the service, not in your code: an execution can pause for days and survive crashes. - Every execution is visible as a diagram with the input and output of each step. - Standard workflows for long processes, Express for fast, high-volume ones. ### Go deeper - [What is AWS Step Functions? (official docs)](https://docs.aws.amazon.com/step-functions/latest/dg/welcome.html) - [AWS Step Functions product page](https://aws.amazon.com/step-functions/) --- ## AWS WAF > A filter in front of your website that reads each request and blocks the bad ones. - Level: Intermediate - Topics: AWS, Security - URL: https://tidbit.legios.com.ar/aws-waf/ - In Spanish: [AWS WAF](https://tidbit.legios.com.ar/es/aws-waf/) A regular firewall looks at addresses and ports: who is connecting, and to which door. A **web application firewall** (WAF) looks inside the web request itself: the URL, the headers, the form fields. **AWS WAF** is that filter on AWS. You put it in front of Amazon CloudFront, a load balancer or Amazon API Gateway, and every request is checked against your rules before it reaches your app. The rules live in a **web ACL** and can be as simple or as sharp as you need. Block requests that try **SQL injection** (sneaking database commands into a search box) or cross-site scripting. Limit any single IP address to a certain number of requests per few minutes, which stops crude scrapers and password-guessing bots. Allow only some countries, or block known bad addresses. Each rule says allow, block, or just count, so you can test a rule before enforcing it. You do not have to write everything yourself: AWS and security vendors publish **managed rule groups** that cover common attacks and are kept up to date. WAF does not fix bugs in your code, but it buys time, catching the predictable attacks at the door while the real fix is on its way. ### An analogy A bouncer who does not just check the guest list but also looks inside the bags. Ordinary visitors walk in; anyone carrying something suspicious, or trying to enter fifty times in a minute, stays outside. ### Keep in mind - Inspects web requests (layer 7), not just addresses and ports. - Sits in front of CloudFront, load balancers and API Gateway. - Rules can block injection attacks, cap requests per IP, or filter by country. - Managed rule groups give ready-made, maintained protection. ### Go deeper - [What is AWS WAF? (official docs)](https://docs.aws.amazon.com/waf/latest/developerguide/waf-chapter.html) - [AWS WAF](https://aws.amazon.com/waf/) - [Web application firewall on Wikipedia](https://en.wikipedia.org/wiki/Web_application_firewall) --- ## Backup > A separate copy of your data, kept so you can get it back when the original is lost. - Level: Basic - Topics: Data, Security - URL: https://tidbit.legios.com.ar/backup/ - In Spanish: [Copia de seguridad](https://tidbit.legios.com.ar/es/backup/) A **backup** is a copy of data stored separately from the original, so it can be restored if something goes wrong: a laptop is stolen, a disk dies, someone deletes the wrong folder, a fire hits the office, or ransomware encrypts every file. It sounds obvious, but many people only discover their backups were missing, incomplete or impossible to restore on the day they need them. A widely used guideline is the **3-2-1 rule**: keep at least **3** copies of your data (the original and two backups), on **2** different kinds of storage (say, the laptop and an external drive or a cloud service), with **1** copy off-site, in another building or another region. Each part protects against a different failure: one copy can break, one kind of media can fail, one place can burn. Many teams now add that one copy should be offline or **immutable**, impossible to change or delete for a while, because modern ransomware hunts for backups it can reach. A sync service is not a backup by itself: if you delete or corrupt a file, the change is copied everywhere within seconds. Real backups keep older versions, so you can go back to how things were last Tuesday. And a backup only counts if it can be restored, so testing a restore every so often is part of the job. ### An analogy Keeping photocopies of your passport: one in your suitcase, one in your email and one with a relative at home. Lose your bag on a trip and you still have proof of who you are. ### Keep in mind - A separate copy that lets you restore data after loss, damage or deletion. - 3-2-1: three copies, two kinds of storage, one copy off-site. - Keep at least one copy offline or immutable as protection against ransomware. - Sync is not backup; keep old versions and test that restores work. ### Go deeper - [Data backup options (CISA, PDF)](https://www.cisa.gov/sites/default/files/publications/data_backup_options.pdf) - [The 3-2-1 backup strategy (Backblaze)](https://www.backblaze.com/blog/the-3-2-1-backup-strategy/) - [Backup on Wikipedia](https://en.wikipedia.org/wiki/Backup) --- ## Bandwidth and latency > How much data fits per second, and how long each trip takes. Two different things. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/bandwidth-and-latency/ - In Spanish: [Ancho de banda y latencia](https://tidbit.legios.com.ar/es/bandwidth-and-latency/) **Bandwidth** is how much data a connection can move per second, measured in megabits per second (Mbps) or gigabits (Gbps). **Latency** is how long a single piece of data takes to go from you to the other end and back, measured in milliseconds (ms) and often called **ping**. A connection can have lots of one and little of the other, and they feel very different. Downloading a movie or streaming in 4K needs bandwidth: the pipe has to be wide. A video call, an online game or just clicking around a site needs low latency: each action waits for a round trip, and 200 ms of lag is noticeable even on a gigabit line. Latency is mostly physics and distance: light in a fiber takes roughly 100 ms to cross the Atlantic and come back, and every router on the way adds a bit more. That is why a CDN puts copies of content near you. When the internet feels slow, ask which one it is. A page that takes a moment to start loading but then arrives all at once is a latency problem (far server, satellite link, overloaded Wi-Fi). A video that keeps buffering is a bandwidth problem (too many devices, a small plan, a weak Wi-Fi link). Internet plans advertise bandwidth; they almost never mention latency. ### An analogy A highway. Bandwidth is the number of lanes: how many cars pass per minute. Latency is the length of the trip: even on a ten-lane highway, going to the next city takes an hour. ### Keep in mind - Bandwidth: Mbps, how much per second. Latency: ms, how long each trip takes. - Streaming and downloads need bandwidth; calls, games and clicks need low latency. - Distance sets a floor on latency; nothing beats the speed of light, so CDNs move content closer. - More bandwidth does not fix lag. ### Go deeper - [What is latency? (Cloudflare)](https://www.cloudflare.com/learning/performance/glossary/what-is-latency/) - [What is bandwidth? (Cloudflare)](https://www.cloudflare.com/learning/network-layer/what-is-bandwidth/) - [Latency (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Latency) --- ## BGP > The protocol networks use to tell each other which addresses they can reach. - Level: Advanced - Topics: Networks - URL: https://tidbit.legios.com.ar/bgp/ - In Spanish: [BGP](https://tidbit.legios.com.ar/es/bgp/) The internet is not one network but tens of thousands of independent ones: internet providers, universities, cloud companies, big websites. Each is an **autonomous system** (AS) with its own number. **BGP**, the Border Gateway Protocol, is how they tell each other which IP addresses they can deliver to. It is the reason a packet that leaves your home in Buenos Aires knows how to reach a server in Frankfurt. The mechanism is a chain of announcements. A network says to its neighbors 'I am the home of these addresses'. Each neighbor passes the news on, adding itself to the path: 'you can reach them through me'. Every router at the edge ends up with several possible routes to each destination and picks one according to its own rules, often preferring shorter paths or cheaper business agreements. BGP is not about finding the fastest path; it is about policy, about who agrees to carry whose traffic. BGP was designed when the networks on the internet largely trusted each other, and by default a router believes what its neighbors announce. That makes mistakes contagious. If a network wrongly announces addresses that are not its own, by accident or on purpose (a **BGP hijack**), traffic can be pulled toward it. In 2008 a provider in Pakistan meant to block YouTube locally and briefly took it offline for much of the world; in 2021 Facebook withdrew its own routes during maintenance and disappeared from the internet for hours. The fixes are gradual. **RPKI** lets an address owner publish a signed statement of which networks may announce its addresses, so others can reject impostors, and providers increasingly filter what their customers are allowed to announce. For most people BGP is invisible, until it breaks and a huge site is suddenly unreachable. ### An analogy A chain of travel agents passing along notes: 'I can get you to Lisbon'. Each one adds its own name and forwards the note. It works because they believe each other, and that is also why one agent writing a wrong note can send everyone's trips the wrong way. ### Keep in mind - Connects the tens of thousands of independent networks that make up the internet. - Networks announce the addresses they reach; neighbors pass the routes along. - Route choice follows business policy, not just distance or speed. - Built on trust, so bad announcements spread; RPKI and filtering reduce the risk. ### Go deeper - [Border Gateway Protocol on Wikipedia](https://en.wikipedia.org/wiki/Border_Gateway_Protocol) - [RFC 4271, the BGP-4 specification](https://www.rfc-editor.org/rfc/rfc4271) --- ## Big data > Data so large, fast or messy that one ordinary computer cannot handle it. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/big-data/ - In Spanish: [Big data](https://tidbit.legios.com.ar/es/big-data/) **Big data** describes collections of data too big or too fast for the usual tools on a single computer. The classic definition uses three Vs: **volume** (huge amounts, far beyond what fits on one machine), **velocity** (arriving constantly, like every click on a website or every reading from thousands of sensors) and **variety** (many shapes at once: tables, text, images, logs). The term became popular in the late 2000s, when companies like Google and Yahoo needed to process the whole web. Their answer was to split the data into pieces and process them in parallel across many cheap machines, an idea that led to open source tools like Hadoop and later Spark. Today the same idea powers cloud data warehouses, data lakes and streaming systems, so a company can query years of records without owning a single server. Examples are everywhere: a streaming service recommending shows from what millions of people watched, a bank flagging an unusual card payment in real time, a city timing traffic lights from road sensors. The term itself has faded a little, because handling large data has become normal. What has not changed is the lesson behind it: more data only helps if it is good data and someone asks it a useful question. ### An analogy Counting the coins in your pocket is easy. Counting every coin that passes through a country's shops in a day needs thousands of counters working at once, each on its own pile, and someone adding up their totals. ### Keep in mind - The three Vs: volume, velocity and variety. - Handled by splitting work across many machines in parallel. - Hadoop and Spark started it; cloud warehouses and lakes made it common. - Big is not the same as useful: quality and good questions still matter. ### Go deeper - [What is big data? (AWS)](https://aws.amazon.com/what-is/big-data/) - [Big data on Wikipedia](https://en.wikipedia.org/wiki/Big_data) --- ## Biometrics > Proving who you are with your body: a fingerprint, your face, your voice. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/biometrics/ - In Spanish: [Biometría](https://tidbit.legios.com.ar/es/biometrics/) **Biometrics** means recognizing a person by something they physically are or do: a fingerprint, the shape of a face, the pattern of an iris, a voice, even the way someone types. It is what happens when you unlock a phone with your thumb or your face, cross an airport gate that checks your passport photo, or clock in at work by touching a reader. The device does not keep a photo of your finger. When you enroll, it measures distinctive features (where the ridges of a fingerprint split and end, the distances between points of a face) and stores them as a compact **template**. Each time you unlock, it measures again and compares. Bodies and cameras are never exactly the same twice, so the answer is a matter of probability: the system is tuned to balance letting the wrong person in against rejecting the right one. Its big advantage is convenience: you cannot forget your face, and nobody can read it over your shoulder. Its big weakness is that you cannot change it. A leaked password can be replaced in a minute; a leaked fingerprint template cannot. That is why well-designed phones and laptops keep the template in a protected chip on the device itself and never send it to a server, and why biometrics usually unlock a key or a passkey stored locally rather than acting as a password sent over the internet. Biometrics also raise questions beyond security. Faces can be captured from a distance without consent, so facial recognition in public spaces, workplaces or schools is regulated in many countries and debated in others. For your own devices, the sensible view is that biometrics are a convenient lock that works best together with a PIN or password as backup. ### An analogy A doorman who knows your face. It is faster than showing an ID card, but if someone ever made a perfect mask of you, you could not get a new face the way you get a new card. ### Keep in mind - Your device stores a mathematical template of the feature, not a photo. - Matching is probabilistic: tuned between false accepts and false rejects. - You cannot change your fingerprint, so templates should stay on the device. - It is usually one factor, backed up by a PIN or password. ### Go deeper - [Biometrics (NIST)](https://www.nist.gov/biometrics) - [Biometrics on Wikipedia](https://en.wikipedia.org/wiki/Biometrics) --- ## BIOS and UEFI > The first program a PC runs: it checks the hardware and starts the operating system. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/bios-and-uefi/ - In Spanish: [BIOS y UEFI](https://tidbit.legios.com.ar/es/bios-and-uefi/) When you press the power button, the processor knows nothing about your disk, your keyboard or Windows. It starts by running a small program stored in a chip on the motherboard. On older PCs that program was the **BIOS** (Basic Input/Output System); on almost every computer made since the early 2010s it is **UEFI** (Unified Extensible Firmware Interface), although many people still call it "the BIOS" out of habit. Its job is short and essential. It checks that the CPU, the RAM and the other parts respond, sets them up, finds a disk with an operating system and hands control to the **boot loader**, the small program that starts the system proper. It also keeps settings like the boot order, the clock and whether virtualization is turned on. You reach its setup screen by pressing a key such as `F2` or `Del` right after power-on. UEFI replaced the BIOS because the old design had hard limits: it could not boot from disks larger than roughly 2 TB and offered little protection. UEFI handles big disks, starts faster, has a setup screen you can use with a mouse, and adds **Secure Boot**, which only lets boot loaders signed with trusted keys start. That makes it much harder for malware to hide underneath the operating system. Because it is firmware, manufacturers occasionally release updates for it, usually to fix bugs or security holes. ### An analogy The stage manager who arrives before everyone else, turns on the lights, checks the microphones and then hands the stage to the main act. The audience barely notices, but without it the show never starts. ### Keep in mind - It runs from a chip on the motherboard every time the computer turns on. - UEFI is the modern replacement for the BIOS; many people still call both "BIOS". - Secure Boot only lets signed, trusted boot loaders start. - Updates fix bugs and security holes, but a failed one can leave the machine unable to start. ### Go deeper - [BIOS on Wikipedia](https://en.wikipedia.org/wiki/BIOS) - [UEFI on Wikipedia](https://en.wikipedia.org/wiki/UEFI) - [Booting on Wikipedia](https://en.wikipedia.org/wiki/Booting) --- ## Blockchain > A shared record book, copied on many computers, where old pages can't be quietly changed. - Level: Basic - Topics: Security, Software - URL: https://tidbit.legios.com.ar/blockchain/ - In Spanish: [Blockchain](https://tidbit.legios.com.ar/es/blockchain/) A **blockchain** is a way of keeping a record, a **ledger**, that many parties share without having to trust a single owner. Entries, such as "Ana sent 5 units to Ben", are grouped into **blocks**. Each block also stores the **hash** of the previous block: a short fingerprint calculated from its contents. That chains the blocks together, because if someone edits an old block, its fingerprint changes and no longer matches the one stored in the next block, so the tampering is obvious. The second idea is that the ledger is not kept in one place. Thousands of computers, called **nodes**, each keep a full copy and follow the same rules to agree on which block comes next. That agreement is called **consensus**. In Bitcoin, the first blockchain, described in 2008 by someone using the name Satoshi Nakamoto, nodes compete to solve a costly math puzzle (**proof of work**); newer systems such as Ethereum let nodes that lock up funds as a deposit take turns adding blocks (**proof of stake**). Rewriting history would require outvoting or outspending most of the network. Blockchains are best known for cryptocurrencies, but the structure can record anything: ownership of digital items, steps in a supply chain, or small programs called **smart contracts** that run automatically when conditions are met. The trade-offs are real: they are slower and more expensive than an ordinary database, mistakes are hard to undo, and proof of work uses a lot of electricity. When one trusted organization keeps the records, a normal database is usually the better tool. ### An analogy A notebook that a whole neighborhood keeps in copies, where each new page starts by quoting the last line of the previous page. Tearing out or rewriting a page would not match everyone else's copy, so nobody gets away with it. ### Keep in mind - Each block stores the hash of the previous one, so edits break the chain. - Many computers keep copies and agree on new blocks through consensus rules. - Proof of work and proof of stake are the two main ways to reach consensus. - Useful when parties don't trust a single keeper; otherwise a database is simpler. ### Go deeper - [Blockchain Technology Overview (NIST)](https://csrc.nist.gov/pubs/ir/8202/final) - [Bitcoin: the original 2008 paper](https://bitcoin.org/bitcoin.pdf) - [Blockchain on Wikipedia](https://en.wikipedia.org/wiki/Blockchain) --- ## Bluetooth > Short-range wireless for connecting nearby gadgets: headphones, watches, keyboards. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/bluetooth/ - In Spanish: [Bluetooth](https://tidbit.legios.com.ar/es/bluetooth/) **Bluetooth** is a wireless standard for connecting devices that are close to each other, typically within about ten meters. It is what links your phone to wireless headphones, a smartwatch, a car's audio system, a keyboard or a mouse, with no cables and no need for Wi-Fi or an internet connection. Devices from different brands work together because they follow the same specification, managed by an industry group called the Bluetooth SIG. Before two devices can talk, they **pair**: you confirm on both that they may connect, they exchange keys, and from then on they recognize each other and reconnect automatically. Bluetooth shares radio frequencies with many Wi-Fi networks and hops quickly between many narrow channels to dodge interference. Over the years it has become faster and more efficient, and newer versions add better audio and longer range. A low-power variant, **Bluetooth Low Energy** (BLE), is made for devices that send small amounts of data now and then and must run for months on a tiny battery: fitness trackers, item finders, medical sensors, smart locks. Phones also use it to discover devices nearby. One practical note: range and quality depend a lot on walls, other radio noise and even your own body, which is why audio can stutter when the phone is in a back pocket. ### An analogy A quiet conversation between two people standing close together. It needs no phone network or loudspeaker, but it only works within a few steps, and the two have to be introduced first. ### Keep in mind - Connects nearby devices wirelessly, usually within about ten meters. - Devices pair once and then reconnect by themselves. - Bluetooth Low Energy lets small sensors run for months on a battery. - Walls and bodies weaken the signal, which can cause dropouts. ### Go deeper - [Bluetooth technology overview (Bluetooth SIG)](https://www.bluetooth.com/learn-about-bluetooth/tech-overview/) - [Bluetooth on Wikipedia](https://en.wikipedia.org/wiki/Bluetooth) --- ## Broadcom > An American chip designer behind data center networking, Wi-Fi and custom AI chips. - Level: Intermediate - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/broadcom/ - In Spanish: [Broadcom](https://tidbit.legios.com.ar/es/broadcom/) **Broadcom** designs chips that most people never see but use every day. Today's company was formed in 2016, when Avago Technologies, a chipmaker whose roots go back to Hewlett-Packard's semiconductor division, bought the original Broadcom Corporation and took its name. Like NVIDIA or Apple, it is **fabless**: it designs the chips and pays foundries such as TSMC to manufacture them. Its chips cover several areas. Its **switch chips** move data between the thousands of servers in a data center, so they are central to the networks that connect GPUs for AI training. It makes Wi-Fi and Bluetooth chips for phones, chips for broadband modems and set-top boxes, and controllers for storage. It also builds **custom AI accelerators** (ASICs) together with large tech companies: the customer brings its ideas, Broadcom turns them into a chip that can be manufactured. Google's TPU is the best-known example. Broadcom has grown mostly through acquisitions, and in recent years it bought large software companies too: CA Technologies, the enterprise security business of Symantec and, in 2023, VMware, the leading maker of virtualization software. Its usual approach is to focus on the products and big customers that bring the most profit, which sometimes means price and licensing changes that customers notice. ### An analogy A firm that makes the plumbing inside buildings. You never see the pipes, valves and pumps behind the walls, but nothing works without them, and the architects of the biggest towers ask it to design custom pieces. ### Keep in mind - Fabless chip designer; foundries such as TSMC manufacture its designs. - Its switch chips are key to data center and AI networks. - Co-designs custom AI chips (ASICs), such as Google's TPU, with big tech companies. - Grew through acquisitions, including VMware in 2023. ### Go deeper - [About Broadcom (official site)](https://www.broadcom.com/company/about-us) - [Broadcom on Wikipedia](https://en.wikipedia.org/wiki/Broadcom) --- ## Brute-force attack > Breaking into an account by trying passwords, one after another, until one works. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/brute-force-attack/ - In Spanish: [Ataque de fuerza bruta](https://tidbit.legios.com.ar/es/brute-force-attack/) A **brute-force attack** is the least clever way to get into an account, and still one of the most common: a program tries passwords over and over until one is accepted. It does not exploit a flaw in the software. It relies on the password being short, common, or already known. Attackers rarely try every possible combination, because that takes too long. A **dictionary attack** starts with the passwords people actually use, like `123456`, names, sports teams and keyboard patterns, plus common tweaks such as adding a year. **Credential stuffing** goes one step further: it takes email and password pairs leaked from one site and tries them on others, betting that people reuse passwords. When a company's password database is stolen, attackers can also guess at full speed on their own machines, offline, without any login page to slow them down. The math favors long passwords. Each extra character multiplies the number of possibilities, so a long passphrase of four random words is far harder to guess than a short password full of symbols. That is why a password manager, which creates a long, different password for every site, defeats most of these attacks, and why two-factor authentication helps even when the password is guessed. Services defend themselves too: they limit how many attempts each account or address can make, add delays or a CAPTCHA after failures, alert you about strange logins, and store passwords with deliberately slow hashing so a stolen database is expensive to crack. ### An analogy A thief standing at a combination lock with patience and a list of the most popular codes. A four-digit bike lock falls quickly; a long combination, or a lock that jams after five wrong tries, makes the whole approach useless. ### Keep in mind - It guesses passwords instead of exploiting a bug, so weak and reused passwords are the target. - Credential stuffing reuses passwords leaked from other sites. - Length beats complexity: each extra character multiplies the work. - Unique passwords from a password manager plus 2FA stop almost all of it. ### Go deeper - [Brute force attack (OWASP)](https://owasp.org/www-community/attacks/Brute_force_attack) - [Use strong passwords (CISA)](https://www.cisa.gov/secure-our-world/use-strong-passwords) - [Brute-force attack on Wikipedia](https://en.wikipedia.org/wiki/Brute-force_attack) --- ## Bug > A mistake in software that makes it behave differently from what was intended. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/bug/ - In Spanish: [Bug](https://tidbit.legios.com.ar/es/bug/) A **bug** is a flaw in a program that makes it do something its authors did not intend: a button that does nothing, a total that adds up wrong, an app that crashes when you rotate the phone. Computers follow instructions exactly, so a bug is almost always a human mistake in those instructions, or a situation nobody planned for, like a name with an apostrophe or a date in a leap year. Engineers used the word for machine faults long before computers. The famous story is from 1947: operators of the Harvard Mark II computer, a team that included Grace Hopper, found a real moth stuck in a relay and taped it into their logbook as the "first actual case of bug being found". The joke works precisely because the word was already in use. Some bugs are cosmetic, others are serious. A bug that lets an attacker read data they should not see is a security vulnerability. Bugs are normal in any large program; what matters is catching them early through testing and code review, and fixing them with **debugging**. Teams usually track them in a bug tracker, with steps to reproduce, how severe they are and who is working on them. ### An analogy A typo in a recipe that says "3 cups of salt" instead of "3 teaspoons". The cook follows it faithfully, and the result is ruined, not because the cook failed but because the instructions were wrong. ### Keep in mind - A mistake or unforeseen case in a program's instructions. - Ranges from cosmetic glitches to crashes and security holes. - The famous 1947 moth was a joke on a word engineers already used. - Found through testing and reports, fixed through debugging. ### Go deeper - [Software bug on Wikipedia](https://en.wikipedia.org/wiki/Software_bug) - [Grace Hopper on Wikipedia](https://en.wikipedia.org/wiki/Grace_Hopper) --- ## Bug bounty > A company pays outside researchers who find and privately report its security holes. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/bug-bounty/ - In Spanish: [Bug bounty](https://tidbit.legios.com.ar/es/bug-bounty/) A **bug bounty** is a standing offer from a company: if you find a security flaw in our products and tell us privately, following our rules, we will pay you. The amount depends on how serious the problem is, from a small thank-you for a minor issue to large sums for a flaw that could expose user accounts. Big tech companies, banks, governments and many startups run them, often through platforms that handle the reports and payments. Every program publishes a policy that is the heart of the deal. It lists which sites and apps are **in scope**, what is off limits (attacking other users, overloading servers, touching real customer data beyond what is needed to prove the point), and how to report. Inside those rules, researchers have permission to test; outside them, the same activity can be illegal. Reports go privately to the company, which confirms the flaw, fixes it, pays, and often lets the researcher publish a write-up afterwards. That last step is called **coordinated disclosure** (or responsible disclosure): the problem is made public only once users are protected. Even companies without a paid program usually publish a **vulnerability disclosure policy** that says how to report a flaw and promises not to take legal action against people acting in good faith. Bug bounties complement penetration testing rather than replace it. A pentest is a small team for a few weeks; a bounty is many people with different skills looking all year. The catch is that the company must be ready to triage a steady stream of reports, many of them duplicates or not real problems. ### An analogy A reward poster in reverse: instead of paying to catch the burglar, the shop pays anyone who points out an unlocked window, as long as they tell the owner and do not climb in. ### Keep in mind - Researchers get paid for privately reporting real security flaws. - The program policy defines scope and rules; testing outside it is not authorized. - Coordinated disclosure: fix first, publish after. - It complements pentests: many testers, all year, paid per finding. ### Go deeper - [Vulnerability Disclosure Cheat Sheet (OWASP)](https://cheatsheetseries.owasp.org/cheatsheets/Vulnerability_Disclosure_Cheat_Sheet.html) - [Vulnerability Disclosure Policy Template (CISA)](https://www.cisa.gov/vulnerability-disclosure-policy-template) - [Bug bounty program on Wikipedia](https://en.wikipedia.org/wiki/Bug_bounty_program) --- ## Business intelligence > Turning a company's data into reports and dashboards people use to decide. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/business-intelligence/ - In Spanish: [Business intelligence](https://tidbit.legios.com.ar/es/business-intelligence/) **Business intelligence**, or **BI**, is the set of tools and practices that turn a company's raw data into reports, charts and **dashboards** that people can read without writing code. A sales manager opens a page on Monday and sees revenue by region, this month against last, and which products are falling behind. That page is BI. Behind the dashboard there is usually a pipeline. Data from the systems a company runs on, like sales, billing, the CRM and the website, is copied by ETL processes into a data warehouse, cleaned and organized into tables with agreed definitions. BI tools such as Power BI, Tableau, Looker or Amazon QuickSight connect to that warehouse, run SQL behind the scenes and draw the results as charts that update on their own. The hard part is rarely the charts. It is agreeing on what the numbers mean: does "active customer" mean bought this month or logged in this year? Good BI teams define these **KPIs**, key performance indicators, once and reuse them everywhere, so two departments do not arrive at a meeting with different totals. BI mostly answers "what happened and where?"; predicting what will happen next is usually the job of data science. ### An analogy Like the dashboard of a car. You do not need to understand the engine to see your speed, fuel and warning lights at a glance, but someone had to wire the sensors correctly for those gauges to tell the truth. ### Keep in mind - Reports and dashboards that let non-technical people explore data. - Usually built on top of a data warehouse fed by ETL. - Shared definitions of each KPI matter more than pretty charts. - Describes what happened; prediction is closer to data science. ### Go deeper - [What is business intelligence? (AWS)](https://aws.amazon.com/what-is/business-intelligence/) - [Business intelligence on Wikipedia](https://en.wikipedia.org/wiki/Business_intelligence) --- ## ByteDance > The Chinese tech company behind TikTok and its recommendation engine. - Level: Basic - Topics: Tech companies - URL: https://tidbit.legios.com.ar/bytedance/ - In Spanish: [ByteDance](https://tidbit.legios.com.ar/es/bytedance/) **ByteDance** is a Chinese technology company founded in 2012 in Beijing by Zhang Yiming. Its first hit was Toutiao, a news app that, instead of having editors choose the stories, learned from each reader what to show them. That idea, a feed run entirely by a recommendation algorithm, became the core of everything it built later. Its best-known products are short-video apps: **Douyin** in China, launched in 2016, and **TikTok** for the rest of the world, which in 2018 absorbed the lip-sync app Musical.ly. You do not need to follow anyone: the "For You" feed watches what you finish, rewatch, skip and like, and within minutes it adjusts what comes next. ByteDance also makes CapCut, a video editor, and Lark, a workplace chat and documents tool. It makes money mostly from **advertising** placed between videos, and also from in-app shopping, where creators and stores sell products, and from virtual gifts viewers buy for creators during live streams. Because its apps are used by many people outside China while the company is Chinese, governments in several countries have raised questions about where user data is stored and who can access it, and ByteDance has set up separate data arrangements in some regions in response. ### An analogy A TV channel with one programmer per viewer, who watches your face as you watch and changes the next show the moment you look bored. ### Keep in mind - Founded in 2012 in Beijing; owns TikTok and its Chinese twin, Douyin. - Its strength is the recommendation algorithm that builds each person's feed. - Makes money mainly from ads, plus in-app shopping and live-stream gifts. - Its foreign reach has drawn government scrutiny over data and ownership. ### Go deeper - [ByteDance (official site)](https://www.bytedance.com/en/) - [ByteDance on Wikipedia](https://en.wikipedia.org/wiki/ByteDance) - [TikTok on Wikipedia](https://en.wikipedia.org/wiki/TikTok) --- ## CAPTCHA > A small test a website uses to tell a person from an automated program. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/captcha/ - In Spanish: [CAPTCHA](https://tidbit.legios.com.ar/es/captcha/) A **CAPTCHA** is the little challenge that appears before you sign up, log in or buy tickets: type the wavy letters, click every picture with a traffic light, or just tick a box that says "I'm not a robot". The name is an acronym for "Completely Automated Public Turing test to tell Computers and Humans Apart". The idea is a task that is easy for people and hard for software. It exists because bots are cheap. A single program can try thousands of stolen passwords per minute, create fake accounts by the hundred, post spam or buy a whole concert in seconds to resell it. A CAPTCHA does not stop a determined attacker, but it makes each attempt slower and more expensive, which is often enough. The tests have changed a lot. Distorted text became easy for image-recognition software, so many sites moved to "invisible" checks such as Google's **reCAPTCHA** or Cloudflare's Turnstile: they look at signals like how the mouse moves, the browser and the network's reputation, and only show a puzzle when something looks off. The trade-offs are real: puzzles can be a wall for people with visual or motor disabilities, and the invisible kind works by observing your behavior. ### An analogy A bouncer who asks a quick question at the door. Anyone can answer it, but it slows down the crowd of fake guests trying to rush in at once. ### Keep in mind - Its job is telling people apart from bots, not checking who you are. - It protects sign-ups, logins, forms and ticket sales from automated abuse. - Modern versions are often invisible and only show a puzzle when something looks suspicious. - Puzzles can block people with disabilities, so good sites offer accessible alternatives. ### Go deeper - [CAPTCHA on Wikipedia](https://en.wikipedia.org/wiki/CAPTCHA) - [What is a CAPTCHA? (Kaspersky)](https://www.kaspersky.com/resource-center/definitions/what-is-a-captcha) - [Inaccessibility of CAPTCHA (W3C)](https://www.w3.org/TR/turingtest/) --- ## CDN > Copies of a website kept on servers around the world, close to whoever asks. - Level: Intermediate - Topics: Networks, Cloud - URL: https://tidbit.legios.com.ar/cdn/ - In Spanish: [CDN](https://tidbit.legios.com.ar/es/cdn/) A **CDN** (content delivery network) is a set of servers spread across many cities that keep copies of a website's files: images, videos, scripts, whole pages. When you open the site, your request goes to the nearest of those **edge** servers instead of travelling to the company's own server, the **origin**, which may be on another continent. Shorter distance means lower latency, and the page appears faster. The edge works as a **cache**. If it already has the file you asked for, that is a **cache hit** and it answers on the spot. If not, a **cache miss**, it fetches the file from the origin once, keeps a copy, and serves it to everyone nearby who asks next. The site decides how long each copy may live: a logo can stay for a month, a news front page maybe for a minute. Almost every big site uses one. A video service in Buenos Aires does not stream every episode from Virginia; it streams from a server a few kilometers away. The origin also gets far fewer requests, so it needs less capacity, and because a CDN has huge total bandwidth spread across the world, it can absorb traffic floods like a DDoS attack that would knock a single server over. Amazon CloudFront, Cloudflare, Akamai and Fastly are well-known CDNs. ### An analogy A bakery chain. The recipe and the main kitchen (the origin) are in one place, but every neighborhood has a branch with fresh bread on the shelf (the edge), so you walk two blocks instead of crossing the city. ### Keep in mind - Edge servers near users keep copies of files from the origin server. - A cache hit is answered at the edge; a cache miss goes to the origin once. - The site controls how long each file may stay cached. - Besides speed, a CDN shields the origin from load and traffic floods. ### Go deeper - [What is a CDN? (AWS)](https://aws.amazon.com/what-is/cdn/) - [CDN (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/CDN) - [Content delivery network on Wikipedia](https://en.wikipedia.org/wiki/Content_delivery_network) --- ## Chatbot > A program you talk to by typing or speaking, as if it were a person. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/chatbot/ - In Spanish: [Chatbot](https://tidbit.legios.com.ar/es/chatbot/) A **chatbot** is a program designed to hold a conversation. You write or say something, and it answers in words. The idea is old: in 1966 Joseph Weizenbaum at MIT built ELIZA, which imitated a therapist by turning your sentences back into questions. Some users became emotionally attached to it, even knowing it was a simple trick. For decades, most chatbots were **scripted**. The bank or airline bot on a website follows a decision tree written by people: it looks for keywords and offers fixed menus ('Type 1 for a refund'). That is predictable and cheap, but it breaks the moment you ask something nobody anticipated. Modern chatbots like ChatGPT, Claude or Gemini are built on large language models, so they generate each reply on the spot instead of choosing from a script. They understand messy questions and answer naturally, and companies can connect them to their own documents and systems. The trade-off is control: a generated answer can be wrong. In 2024 a Canadian tribunal ordered Air Canada to honor a refund policy its website chatbot had invented, which made clear that a company is responsible for what its bot says. ### An analogy A scripted chatbot is a phone menu that reads you options. A language-model chatbot is a new employee who can talk about anything, but who sometimes makes things up when they do not know. ### Keep in mind - Old chatbots follow scripts; modern ones generate replies with an LLM. - ELIZA, from 1966, is usually called the first chatbot. - Generated replies sound natural but can be wrong. - Companies are responsible for what their chatbots tell customers. ### Go deeper - [Chatbot on Wikipedia](https://en.wikipedia.org/wiki/Chatbot) - [ELIZA on Wikipedia](https://en.wikipedia.org/wiki/ELIZA) - [What is a chatbot? (IBM)](https://www.ibm.com/think/topics/chatbots) --- ## ChatGPT > OpenAI's chat assistant, the app that brought AI chatbots to the general public. - Level: Basic - Topics: AI, Tech companies - URL: https://tidbit.legios.com.ar/chatgpt/ - In Spanish: [ChatGPT](https://tidbit.legios.com.ar/es/chatgpt/) **ChatGPT** is a chat assistant made by OpenAI. You type or say something in plain language (a question, a request, a half-formed idea) and it answers in conversation: it explains, drafts, summarizes, translates, writes code, and keeps the thread so you can ask follow-ups. It launched as a free research preview on 30 November 2022 and spread faster than almost any consumer product before it. For many people, 'ChatGPT' became the name of the whole idea. It is important to separate the app from the model. ChatGPT is the product you use, on the web, on phones and on computers. The thinking happens in OpenAI's **GPT** models, large language models that OpenAI keeps replacing with newer versions; GPT stands for 'generative pre-trained transformer'. Over time the app gained more abilities around the model: searching the web, reading uploaded files and photos, generating images, talking by voice and remembering preferences across chats. There is a free version and paid plans for individuals and companies. Developers who want GPT inside their own products do not use ChatGPT itself; they use OpenAI's API. Like every chatbot built on a language model, ChatGPT can sound sure of things that are false, so anything important it tells you (a dose, a law, a quote) is worth checking against a real source. ### An analogy ChatGPT is the shop window and GPT is the workshop behind it. You talk to the person at the counter, but the work is done in the back, and the workshop's tools get upgraded every so often while the counter looks the same. ### Keep in mind - Made by OpenAI and launched on 30 November 2022. - ChatGPT is the app; GPT models are the engines inside it. - Can search the web, read files and images, and talk by voice. - Free to start, with paid plans; developers use the API instead. ### Go deeper - [ChatGPT on Wikipedia](https://en.wikipedia.org/wiki/ChatGPT) - [OpenAI's models (official docs)](https://platform.openai.com/docs/models) --- ## Chip > A sliver of silicon carrying billions of tiny switches that compute and remember. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/semiconductor-chip/ - In Spanish: [Chip](https://tidbit.legios.com.ar/es/semiconductor-chip/) A **chip** (or **integrated circuit**) is a small square of silicon, often smaller than a fingernail, with an entire electronic circuit built into its surface: billions of transistors, the tiny switches that do all digital work, plus the microscopic wiring between them. The CPU and GPU are chips, so is the RAM, the flash memory in an SSD, and the controllers inside a microwave or a car. Silicon is a **semiconductor**: a material that can be made to conduct electricity or block it, depending on how it is treated and on the voltage applied. That on/off control is exactly what a switch needs. Chips are made on round, polished **wafers** of pure silicon. In a factory called a **fab**, the circuit is projected onto the wafer with light, layer after layer, with features measured in nanometers. The wafer is then cut into hundreds of individual chips, each one tested and sealed in a plastic or ceramic package with metal contacts. Two kinds of companies are involved, and the news often mixes them up. Some design chips (NVIDIA, Apple, AMD), and some manufacture them (TSMC, Samsung, Intel, which also designs its own). Building a leading-edge fab costs many billions of dollars and needs machines only a few suppliers can make, which is why chips have become a matter of economics and geopolitics, not just engineering. ### An analogy A city printed on a stamp. Streets, buildings and traffic lights, all drawn so small that a whole metropolis fits on a fingernail, and all printed at once, many cities per sheet. ### Keep in mind - A whole circuit of transistors and wiring built into one piece of silicon. - Silicon is a semiconductor: it can be made to conduct or to block current. - Made by the hundreds on a wafer, then cut, tested and packaged. - Designing chips and manufacturing them are often done by different companies. ### Go deeper - [Integrated circuit on Wikipedia](https://en.wikipedia.org/wiki/Integrated_circuit) - [Semiconductor device fabrication on Wikipedia](https://en.wikipedia.org/wiki/Semiconductor_device_fabrication) - [Semiconductor on Wikipedia](https://en.wikipedia.org/wiki/Semiconductor) --- ## CI/CD > Automation that builds, tests and ships every code change, so releases become routine. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/ci-cd/ - In Spanish: [CI/CD](https://tidbit.legios.com.ar/es/ci-cd/) **CI/CD** names two linked practices. **Continuous integration** (CI) means developers merge their changes into the shared code often, at least daily, and every merge automatically triggers a build and a run of the automated tests. If something breaks, the team learns within minutes, while the change is still fresh, instead of weeks later. The CD half has two flavors. **Continuous delivery** keeps the code always ready to release: every change that passes the checks is packaged and can go to production with the press of a button. **Continuous deployment** goes one step further and releases every passing change automatically, with no button. Teams pick based on how much risk and human approval their product needs. The automation runs as a **pipeline**: a list of steps written in a file inside the repository, such as "install, build, run tests, scan for vulnerabilities, build the container image, deploy to staging, deploy to production". Services like GitHub Actions, GitLab CI and Jenkins run these pipelines on every push. The payoff is that shipping software stops being a scary monthly event and becomes a small, frequent, boring one, which is exactly what you want. ### An analogy An assembly line with quality checks at every station. Each part is inspected as soon as it is added, and a defective one stops the belt before it reaches the customer, not after a full truck has left. ### Keep in mind - CI: merge often and test every change automatically. - Continuous delivery: always ready to release. Continuous deployment: releases automatically. - The pipeline is a file in the repository, so it is versioned like code. - Small, frequent releases are easier to check and to undo than big ones. ### Go deeper - [What is continuous integration? (AWS)](https://aws.amazon.com/devops/continuous-integration/) - [What is CI/CD? (GitLab)](https://about.gitlab.com/topics/ci-cd/) - [CI/CD on Wikipedia](https://en.wikipedia.org/wiki/CI/CD) --- ## Cisco > The company that makes much of the network gear connecting offices to the internet. - Level: Basic - Topics: Tech companies, Networks - URL: https://tidbit.legios.com.ar/cisco/ - In Spanish: [Cisco](https://tidbit.legios.com.ar/es/cisco/) **Cisco Systems** was founded in 1984 by Leonard Bosack and Sandy Lerner, who worked on Stanford University's computer networks, and is based in San Jose, California. Its first products were routers, the boxes that pass data between different networks, and it grew with the internet itself in the 1990s. Today much of the equipment inside corporate offices, universities, internet providers and data centers carries its name. Its main business is **networking hardware**: routers, **switches** (which connect the computers, phones and printers inside a building), Wi-Fi access points and firewalls. Around that it sells software and services: tools to manage thousands of devices from the cloud (its Meraki line), security products, the **Webex** video-conferencing service and, since buying Splunk, software for collecting and searching logs and security events. It makes money from selling equipment and, more and more, from software licenses, subscriptions and support contracts. Cisco has also shaped the profession: its certifications, such as the **CCNA**, are a common first step for network engineers, and its Networking Academy teaches networking in schools and universities around the world. ### An analogy A company that builds the roads, junctions and traffic lights for data. You rarely see its name, but most office traffic passes through its intersections on the way out to the internet. ### Keep in mind - Founded in 1984 by Stanford network engineers; grew with the internet. - Makes routers, switches, Wi-Fi access points and firewalls for businesses and providers. - Also sells security software, Webex and, through Splunk, log analytics. - Its CCNA certification is a classic entry point for network engineers. ### Go deeper - [Cisco company overview (Cisco newsroom)](https://newsroom.cisco.com/c/r/newsroom/en/us/company.html) - [Cisco on Wikipedia](https://en.wikipedia.org/wiki/Cisco) - [Cisco Networking Academy](https://www.netacad.com/) --- ## Claude > Anthropic's family of AI models, and the assistant app built on them. - Level: Basic - Topics: AI, Tech companies - URL: https://tidbit.legios.com.ar/claude/ - In Spanish: [Claude](https://tidbit.legios.com.ar/es/claude/) **Claude** is the name of the large language models made by Anthropic, and of the assistant people use to talk to them. Like other chat assistants, it answers questions, writes and edits text, summarizes long documents, analyzes data, reads images and PDFs, and writes code. The first version came out in March 2023, and Anthropic releases new generations regularly, numbered like Claude 3 or Claude 4. The models come in sizes with different trade-offs. **Haiku** is the small, fast and cheap one; **Sonnet** sits in the middle; **Opus** is the largest and most capable, and also the slowest and most expensive. There are several ways to reach them: the Claude app on the web, desktop and phone; an API that developers pay for per token; cloud platforms such as Amazon Bedrock and Google Cloud; and **Claude Code**, a tool that lets Claude work inside a software project, reading files, changing code and running commands. What sets it apart is mostly how it is trained. Anthropic uses a method it calls **Constitutional AI**, where the model learns to follow a written set of principles, and it publishes research on safety and on understanding what happens inside the model. In practice Claude is widely used for writing, for analysis of long documents and for programming. Like any language model, it can be wrong while sounding sure, so it is a collaborator to check, not an oracle. ### An analogy Claude is like a firm of assistants with the same training but different seniority: a quick junior for simple errands, a reliable mid-level for most work, and a senior partner for the hardest problems, who takes longer and bills more. ### Keep in mind - Made by Anthropic; the first version came out in March 2023. - Three sizes: Haiku (fast), Sonnet (balanced), Opus (most capable). - Available as an app, an API, through clouds and as Claude Code. - Trained with Constitutional AI, guided by written principles. ### Go deeper - [Claude on Wikipedia](https://en.wikipedia.org/wiki/Claude_(language_model)) - [Intro to Claude (official docs)](https://platform.claude.com/docs/en/intro) - [Meet Claude (Anthropic)](https://www.anthropic.com/claude) --- ## Clock speed > How many times per second a processor ticks, measured in gigahertz (GHz). - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/clock-speed/ - In Spanish: [Velocidad de reloj](https://tidbit.legios.com.ar/es/clock-speed/) Inside every CPU there is a **clock**: a signal that pulses at a perfectly regular rhythm, and every part of the chip moves one step forward on each pulse. **Clock speed** is how fast that rhythm is. One hertz is one tick per second, so a CPU at 3 **GHz** (gigahertz) ticks three billion times every second. Typical desktop and laptop processors run at roughly 2 to 5 GHz. A faster clock means more steps per second, but it is not the whole story. How much work a core gets done on each tick depends on its design, so a newer CPU at 3 GHz can easily beat an older one at 4 GHz. The number of cores matters too. Comparing GHz only makes sense between chips of the same family and generation. Clock speeds climbed fast until the mid-2000s and then mostly stopped, at a few gigahertz. Each extra tick costs power, and power turns into heat; past a point the chip simply gets too hot to cool. The industry turned to more cores and smarter designs instead. Modern CPUs also change speed constantly: they slow down when idle to save battery, and briefly speed up (**boost** or **turbo**) when you need a burst, as long as they stay cool enough. ### An analogy The drummer on a galley ship. Every beat, the rowers pull once. A faster drummer means more strokes per minute, but how far the boat moves also depends on how strong each rower is and how many there are. ### Keep in mind - Clock speed is the number of ticks per second; 1 GHz is a billion ticks. - Work per tick depends on the chip's design, so GHz alone does not rank CPUs. - Speeds plateaued at a few GHz because heat grows with every extra tick. - CPUs slow down when idle and boost briefly under load. ### Go deeper - [Clock rate on Wikipedia](https://en.wikipedia.org/wiki/Clock_rate) - [Hertz on Wikipedia](https://en.wikipedia.org/wiki/Hertz) --- ## Cloud computing > Renting computers over the internet instead of buying your own. - Level: Basic - Topics: Cloud - URL: https://tidbit.legios.com.ar/cloud-computing/ - In Spanish: [Computación en la nube](https://tidbit.legios.com.ar/es/cloud-computing/) **Cloud computing** means using someone else's computers over the internet, paid by the hour or by use, instead of buying and running your own. The "cloud" is not in the sky: it is huge buildings full of servers (a data center) owned by companies like AWS, Microsoft and Google, which rent out slices of them. What you rent ranges from a bare virtual machine (Amazon EC2), to file storage (Amazon S3), to a managed database, to a function that runs for a few milliseconds (AWS Lambda). You click a button or call an API, and the machine exists a minute later; delete it and the bill stops. Why it matters: a startup can begin with one small server and grow to thousands without ever buying hardware. Netflix, your bank's app and most AI chatbots run in the cloud. The trade-off is that you depend on the provider, and the bill grows with use, so a mistake left running over a weekend can get expensive. ### An analogy Electricity from the grid. You do not build a power plant in your backyard; you plug in, use what you need, and pay the meter at the end of the month. ### Keep in mind - The three biggest providers are AWS, Microsoft Azure and Google Cloud. - You pay for what you use: per hour, per gigabyte or per request. - "Public cloud" means shared data centers; "on-premises" means servers you own. - Elasticity, adding or removing capacity in minutes, is the whole point. ### Go deeper - [What is cloud computing? (AWS)](https://aws.amazon.com/what-is-cloud-computing/) - [Cloud computing (Wikipedia)](https://en.wikipedia.org/wiki/Cloud_computing) --- ## Cloudflare > A global network that sits in front of websites to make them faster and safer. - Level: Basic - Topics: Tech companies, Networks - URL: https://tidbit.legios.com.ar/cloudflare/ - In Spanish: [Cloudflare](https://tidbit.legios.com.ar/es/cloudflare/) **Cloudflare** is an American company founded in 2009 by Matthew Prince, Michelle Zatlyn and Lee Holloway. It runs a network of data centers in cities all over the world, and websites put it **in front of** their own servers. When you visit a site that uses Cloudflare, your request first reaches the Cloudflare data center closest to you, not the site's real server, which is called the **origin**. From that position it can do several jobs at once. It works as a CDN, keeping copies of images and pages near visitors so they load faster. It absorbs DDoS attacks, where floods of fake traffic try to knock a site offline, and filters malicious requests with a firewall. It also handles the site's HTTPS certificates and its DNS. Beyond websites, it offers **Cloudflare Workers**, which run small programs in its data centers, close to users, and the public DNS service `1.1.1.1`. Its business model is freemium: a basic plan is free, which is why millions of small sites use it, and companies pay monthly for more features or sign larger contracts. The flip side of being in front of so much of the web is that when Cloudflare has an outage, many unrelated sites and apps fail at the same moment. ### An analogy A front desk with a security guard for a building. Visitors check in at the lobby, the guard turns away troublemakers, and common questions get answered right there without bothering the people upstairs. ### Keep in mind - Founded in 2009; runs data centers in cities around the world. - Sits between visitors and a site's origin server, acting as CDN, shield and DNS. - Free basic plan; companies pay for advanced features and contracts. - Because so many sites rely on it, its outages are felt across the web. ### Go deeper - [About Cloudflare (official site)](https://www.cloudflare.com/about-overview/) - [Cloudflare developer docs](https://developers.cloudflare.com/) - [Cloudflare on Wikipedia](https://en.wikipedia.org/wiki/Cloudflare) --- ## CMS > Software that lets people publish and edit a website's content without writing code. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/cms/ - In Spanish: [CMS](https://tidbit.legios.com.ar/es/cms/) A **CMS** (content management system) is software for creating, editing and publishing the content of a website through forms, without touching HTML. An editor logs in, writes a title, pastes some text, uploads a photo and clicks "Publish"; the CMS stores it in a database and fills in the site's design template to produce the finished page. The point is a clean split of jobs. Developers and designers build the templates and features once; writers, marketers and shop owners then add pages, news and products every day on their own. A CMS usually adds users and roles (who can write, who can approve), drafts, scheduled posts and a history of changes. **WordPress** is the best-known example, open source and used by a very large share of the web; Drupal and Joomla are other long-standing ones, and services like Wix or Squarespace bundle a CMS with hosting. A newer style, the **headless CMS**, only stores and serves the content through an API, and leaves the look to separate apps, so the same article can appear on a website, a phone app and a smart screen. ### An analogy A newspaper's layout system. Reporters type their stories into fixed boxes for headline, body and photo; the page designers already decided how a front page looks, so each story lands in the right place automatically. ### Keep in mind - Lets non-programmers create and edit web content through forms. - Separates content (stored in a database) from design (templates). - Adds roles, drafts, scheduling and change history. - A headless CMS serves content through an API to any app or site. ### Go deeper - [CMS (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/CMS) - [Content management system on Wikipedia](https://en.wikipedia.org/wiki/Content_management_system) - [About WordPress (wordpress.org)](https://wordpress.org/about/) --- ## Code review > Having another person read a code change before it becomes part of the product. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/code-review/ - In Spanish: [Revisión de código](https://tidbit.legios.com.ar/es/code-review/) **Code review** is the practice of having at least one other developer read and approve a change before it is merged into the main code. Today it almost always happens through a **pull request** (GitLab calls it a **merge request**): the author pushes their change to a separate branch, opens a page that shows exactly which lines were added and removed, and explains why. Reviewers read the diff and leave comments on specific lines: "this breaks if the list is empty", "there is already a function for this", "can we name it something clearer?". The author answers or updates the change, and the conversation repeats until a reviewer **approves** it. Meanwhile automated checks run on the same pull request, from tests to formatting rules, so humans can focus on what machines cannot judge. It catches bugs, but research and experience suggest its bigger value is spreading knowledge: more than one person understands each part of the system, newcomers learn the codebase's conventions, and decisions are written down next to the code. It works best with small changes, which are reviewed faster and more carefully than huge ones, and with a kind, specific tone, since the code is being reviewed, not the person. AI tools now often add a first pass of automated review comments. ### An analogy Like having a colleague proofread an important email before you send it to a client. They catch typos, but more often they say "this part could be misunderstood", and next time you write it better yourself. ### Keep in mind - A pull request shows the exact change and collects the discussion about it. - Usually at least one approval is required before merging. - Shares knowledge as much as it catches bugs. - Small changes get better reviews than big ones. ### Go deeper - [About pull request reviews (GitHub docs)](https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/reviewing-changes-in-pull-requests/about-pull-request-reviews) - [Google's code review guidelines](https://google.github.io/eng-practices/review/) - [Code review on Wikipedia](https://en.wikipedia.org/wiki/Code_review) --- ## Command line > Controlling a computer by typing text commands instead of clicking icons. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/command-line/ - In Spanish: [Línea de comandos](https://tidbit.legios.com.ar/es/command-line/) The **command line** is a way to use a computer by typing instructions and reading text replies, with no buttons or icons. You open a **terminal** app (Terminal on a Mac, Windows Terminal or PowerShell on Windows) and type something like `ls` to list the files in a folder, `cd photos` to move into a folder, or `cp report.pdf backup/` to copy a file. The program that reads and runs what you type is called the **shell**; Bash and Zsh are common ones. It looks old-fashioned, but it is still central to software work. Commands are exact and repeatable: you can save a sequence of them in a script and run it a thousand times, or chain small tools together so the output of one becomes the input of the next. Many developer tools, like Git or package managers, are used mainly or only from the command line, and most servers have no screen at all, so people manage them by typing commands over a remote connection. The flip side is that it trusts you. There is usually no "Are you sure?" and no trash can: a mistyped delete command can erase files for good. That is why careful users read a command before pressing Enter, especially one copied from the internet. ### An analogy Ordering at a counter by saying exactly what you want, instead of pointing at pictures on a menu. It takes learning the words, but once you know them it is faster and you can ask for things that are not on the pictures. ### Keep in mind - You type commands in a terminal; a shell such as Bash or Zsh runs them. - Exact and repeatable: easy to script and to automate. - The standard way to manage servers and many developer tools. - Powerful and unforgiving: deleted files usually do not go to a trash can. ### Go deeper - [Command line crash course (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Getting_started/Environment_setup/Command_line) - [The Linux command line for beginners (Ubuntu)](https://ubuntu.com/tutorials/command-line-for-beginners) - [Command-line interface on Wikipedia](https://en.wikipedia.org/wiki/Command-line_interface) --- ## Compiler > A program that translates source code into instructions a computer can run directly. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/compiler/ - In Spanish: [Compilador](https://tidbit.legios.com.ar/es/compiler/) A **compiler** is a program that reads source code written in a language like C, Rust, Go or Swift and translates the whole thing, ahead of time, into **machine code**: the low-level instructions a specific kind of CPU executes. The result is a file you can run on its own, like the app you double-click, without the original source code or the compiler. Along the way the compiler checks the code. If you call something that does not exist, pass text where a number is required or forget a bracket, it refuses to finish and points at the line. It also optimizes: it rearranges and simplifies the instructions so the program runs faster than a literal translation would. That is a big reason compiled programs tend to be fast. The alternative is an **interpreter**, which reads the code and runs it on the spot, as Python and JavaScript traditionally do. In practice the line is blurry. Java and C# compile to an intermediate form that a virtual machine runs, and browsers compile JavaScript to machine code while the page is running (**just-in-time compilation**). Because machine code is specific to a processor family, the same program has to be compiled separately for, say, Intel and Arm chips. ### An analogy A translator who receives a whole book, translates it carefully, checks it for mistakes and hands you a finished edition. An interpreter is more like a live interpreter, translating sentence by sentence as the speaker talks. ### Keep in mind - Translates source code into machine code before the program runs. - Catches many mistakes early and refuses to build broken code. - Optimizes the result, so compiled programs are often fast. - Machine code is tied to a CPU family, so builds differ for Intel and Arm. ### Go deeper - [Compile (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Compile) - [Compiler on Wikipedia](https://en.wikipedia.org/wiki/Compiler) - [Interpreter (computing) on Wikipedia](https://en.wikipedia.org/wiki/Interpreter_(computing)) --- ## Computer vision > The part of AI that lets computers find and understand what is in images and video. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/computer-vision/ - In Spanish: [Visión por computadora](https://tidbit.legios.com.ar/es/computer-vision/) **Computer vision** is the field of AI that gets computers to make sense of images and video. To a computer a photo is just a grid of numbers, one per pixel; computer vision turns that grid into useful statements like 'there is a pedestrian 10 meters ahead', 'this X-ray shows a fracture' or 'this receipt says 4,250'. It covers a family of tasks. **Classification** says what an image shows ('cat'). **Object detection** finds each thing and draws a box around it, with a confidence score. **Segmentation** outlines objects pixel by pixel. Others read text in images (**OCR**), recognize faces, estimate body poses or track objects across video frames. Modern systems learn all this with deep learning, trained on millions of labeled images; a turning point was 2012, when a deep network won the ImageNet image-recognition contest by a wide margin. You use it constantly: unlocking a phone with your face, searching your photos for 'beach', depositing a check from the bank app, scanning a QR code, supermarket checkouts that recognize products. It also guides factory inspection, crop monitoring, medical imaging and cars that brake on their own. The weak spots are the usual ones for machine learning: it can fail on situations unlike its training images, and facial recognition in particular raises serious questions about errors, bias and surveillance. ### An analogy A trainee radiologist who has looked at a million labeled scans. They spot patterns fast and tirelessly, but show them something truly unusual and they may confidently call it the closest thing they have seen before. ### Keep in mind - Turns pixels into labels, boxes, outlines or extracted text. - Main tasks: classification, detection, segmentation and OCR. - Learned from huge sets of labeled images, using deep learning. - Face recognition raises concerns about bias and surveillance. ### Go deeper - [Computer vision on Wikipedia](https://en.wikipedia.org/wiki/Computer_vision) - [What is computer vision? (IBM)](https://www.ibm.com/think/topics/computer-vision) --- ## Container > A sealed box with a program and everything it needs to run anywhere. - Level: Basic - Topics: Cloud, Software - URL: https://tidbit.legios.com.ar/container/ - In Spanish: [Contenedor](https://tidbit.legios.com.ar/es/container/) A **container** packages a program together with everything it needs to run: the libraries, the settings, the exact version of the language runtime. That package, called an **image**, runs the same on your laptop, on a colleague's machine and on a server in the cloud. **Docker** is the tool that made containers popular, and a `Dockerfile` is the recipe that builds an image. Before containers, "it works on my machine" was a daily problem: the server had a different Python version, a missing library, a setting nobody remembered. A container ends that argument, because the environment travels with the code. It also starts in seconds and is far lighter than a full virtual machine, since containers share the operating system of the machine that hosts them. Today most web services ship as containers. One server can run dozens of them, isolated from each other, and tools like Kubernetes or Amazon ECS and Fargate decide where each one runs and restart it if it dies. ### An analogy A shipping container. Whatever is inside, every crane, truck and ship knows how to handle the box, so the cargo moves from port to port without being repacked. ### Keep in mind - An image is the frozen package; a container is a running copy of it. - Registries like Docker Hub or Amazon ECR are where images are stored and shared. - Containers share the host's operating system, so they start fast and use little memory. - Lighter than a virtual machine, with somewhat weaker isolation. ### Go deeper - [What is Docker? (official docs)](https://docs.docker.com/get-started/docker-overview/) - [What is containerization? (AWS)](https://aws.amazon.com/what-is/containerization/) - [Containerization (Wikipedia)](https://en.wikipedia.org/wiki/Containerization_(computing)) --- ## Context window > How much text an AI model can take into account at once, measured in tokens. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/context-window/ - In Spanish: [Ventana de contexto](https://tidbit.legios.com.ar/es/context-window/) The **context window** is the amount of text a language model can look at in one go. Everything the model uses to produce an answer has to fit inside it: the hidden instructions from the app, your conversation so far, any documents you pasted or uploaded, and the reply it is writing. It is measured in tokens, the chunks of text models work with; in English a token is roughly three quarters of a word. A model has no memory of its own between messages. When you chat, the app sends the whole conversation back to the model each turn, and that is how it 'remembers' what you said. Once the conversation or the documents get longer than the window, something has to go: older messages are dropped or summarized, and the model simply cannot see them anymore. That is why a very long chat can start to forget details from the beginning. Context windows have grown a lot, from a few thousand tokens in early chatbots to hundreds of thousands, and in some models over a million, which is enough for several books. Bigger is not free, though: long inputs cost more, take longer, and models can pay less attention to details buried in the middle of a huge text. For large collections of documents, it is often better to retrieve only the relevant pieces (RAG) than to paste everything. ### An analogy Like a desk with limited space. You can spread out only so many pages at once; to bring in a new one, an old one has to go back in the drawer, and whatever is in the drawer you cannot read right now. ### Keep in mind - Everything the model sees at once: instructions, chat, files and reply. - Measured in tokens; a token is roughly three quarters of an English word. - Beyond the limit, older content is dropped or summarized. - Bigger windows cost more, and the middle of very long inputs gets less attention. ### Go deeper - [Context windows (Claude docs)](https://platform.claude.com/docs/en/build-with-claude/context-windows) - [What is a context window? (IBM)](https://www.ibm.com/think/topics/context-window) --- ## Cookies > Small notes a website asks your browser to keep, so it can recognize you later. - Level: Basic - Topics: Security, Software - URL: https://tidbit.legios.com.ar/cookies/ - In Spanish: [Cookies](https://tidbit.legios.com.ar/es/cookies/) A **cookie** is a small piece of text that a website asks your web browser to store, like `session=8f3a…`. Every time you go back to that site, the browser sends the cookie along with each request. That is how a site remembers that you are logged in, what is in your cart or which language you picked, even though each page load is otherwise a separate conversation. Cookies belong to the site that set them. A **first-party cookie** comes from the site you are visiting. A **third-party cookie** comes from another company whose content is embedded in the page, like an ad network or a social media button. If the same ad network sits on thousands of sites, its cookie lets it recognize you on all of them and build a profile of your browsing. That is what people usually mean by tracking cookies. Because of this, privacy laws such as the EU's GDPR and ePrivacy rules require sites to ask before setting non-essential cookies, which is where the cookie banners come from. Browsers have also tightened the rules: Safari and Firefox block third-party cookies by default, and deleting your cookies logs you out of sites and makes them forget you. ### An analogy The wristband you get at a festival. Show it at the gate and they let you back in without paying again. A tracking cookie is like a sponsor who stamped your hand and has staff at every festival in the country checking for the stamp. ### Keep in mind - Small text the browser stores and sends back to the same site. - They keep you logged in and remember carts and preferences. - Third-party cookies can follow you across many sites; that is tracking. - Clearing them logs you out and resets what sites remember about you. ### Go deeper - [Using HTTP cookies (MDN)](https://developer.mozilla.org/en-US/docs/Web/HTTP/Guides/Cookies) - [HTTP cookie on Wikipedia](https://en.wikipedia.org/wiki/HTTP_cookie) --- ## Cores and threads > Why a CPU is really several processors, and why it can juggle even more tasks. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/cpu-cores-and-threads/ - In Spanish: [Núcleos e hilos](https://tidbit.legios.com.ar/es/cpu-cores-and-threads/) A **core** is a complete processor: it can fetch instructions and execute them on its own. For decades a CPU had one. Since the mid-2000s, chipmakers have put several cores on the same chip instead of making a single core ever faster, so a laptop processor might have 8 or more and a server processor many dozens. Each core can run a different program, or a different piece of the same program, at the same time. A **thread** is a sequence of instructions waiting to be run. Software creates threads, the operating system hands them to cores, and switches between them so quickly that dozens of apps seem to run at once. Many CPUs also offer **simultaneous multithreading** (Intel calls it **Hyper-Threading**): each core presents itself as two, so it can work on a second thread while the first is waiting for data. That is where specs like "8 cores, 16 threads" come from. The two threads share one core's machinery, so it is a useful boost, not double the speed. More cores help only when the work can be split. Rendering video, compiling code or running many users' requests on a server split well, and get much faster with more cores. A single long chain of steps, where each step needs the previous result, cannot be split, and runs at the speed of one core. That is why both the number of cores and the speed of each one matter. ### An analogy Cores are cooks in a kitchen; threads are the orders on the rail. Hyper-Threading is a cook who keeps two pans going, stirring one while the other heats. Ten cooks will not make one soufflé rise faster. ### Keep in mind - A core is an independent processor; modern CPUs have several to many. - A thread is a stream of instructions the operating system assigns to a core. - Simultaneous multithreading lets one core handle two threads, sharing its parts. - More cores speed up work that can be split, not single-chain tasks. ### Go deeper - [Multi-core processor on Wikipedia](https://en.wikipedia.org/wiki/Multi-core_processor) - [Simultaneous multithreading on Wikipedia](https://en.wikipedia.org/wiki/Simultaneous_multithreading) - [Thread (computing) on Wikipedia](https://en.wikipedia.org/wiki/Thread_(computing)) --- ## CPU > The chip that runs your programs, one instruction after another, very fast. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/cpu/ - In Spanish: [CPU](https://tidbit.legios.com.ar/es/cpu/) The **CPU** (Central Processing Unit) is the chip that executes programs. Everything a computer does, from opening a file to running a browser, is a stream of tiny instructions (add these two numbers, copy this to memory, jump to that line), and the CPU runs through them billions of times per second. It is the general-purpose brain: not specialized in anything, able to do everything. A modern CPU has several **cores**, each one an independent processor, so it can run several programs (or several parts of one) at once; a laptop has around 8, a server 64 or more. Its speed is quoted in gigahertz (clock ticks per second), but the design of each core matters as much as the clock. Intel and AMD make most PC and server CPUs; Apple designs its own (the M series) on the ARM architecture, as do phone makers. The CPU is good at varied, branching work with lots of decisions, which is most software. It is bad at doing the same simple calculation on millions of numbers at once, which is why graphics and AI moved to the GPU. Even then the CPU is still there, running the operating system and feeding the GPU its work. When a computer feels slow today, the CPU is rarely the reason; more often it is waiting on RAM, the SSD or the network. But the CPU still decides how fast a single task finishes, and its power use decides how long a laptop battery lasts. ### An analogy A chef who can cook any recipe, one step at a time, very fast. More cores means more chefs in the kitchen; a GPU is a separate room with a thousand helpers who only chop. ### Keep in mind - Runs all general-purpose software, instruction by instruction. - Cores work in parallel; clock speed (GHz) is how fast each core ticks. - Two instruction families dominate: x86 (Intel, AMD) and ARM (Apple, phones, more and more servers). - Great at varied logic, weak at massive repeated math; that is the GPU's job. ### Go deeper - [Central processing unit on Wikipedia](https://en.wikipedia.org/wiki/Central_processing_unit) - [Multi-core processor on Wikipedia](https://en.wikipedia.org/wiki/Multi-core_processor) --- ## Cross-site scripting > Sneaking code into a website so it runs in other visitors' browsers. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/cross-site-scripting/ - In Spanish: [Cross-site scripting](https://tidbit.legios.com.ar/es/cross-site-scripting/) Websites constantly show text that someone else wrote: comments, product reviews, usernames, the words you searched for. **Cross-site scripting**, almost always called **XSS**, happens when a site shows that text without making sure it stays text. If an attacker writes a comment that contains a bit of JavaScript disguised inside HTML, and the site pastes it into the page as is, every visitor's web browser reads it as part of the page and runs it. That code runs with the site's own identity. It can read what the page shows, act as the logged-in visitor (post, change settings, send messages), show a fake login box, or grab session cookies so the attacker can take over the account. The visitor did nothing wrong: they opened a trusted site. The flaw is in the site, which let an outsider's code into its pages. There are a few versions. **Stored** XSS lives in the site's database, like the comment above, and hits everyone who views it. **Reflected** XSS bounces off a link: the malicious bit is in the URL and the page echoes it back to whoever clicks. **DOM-based** XSS happens entirely inside the browser, in the site's own JavaScript. The defenses are well known. Developers **escape** or encode anything that came from outside before putting it in a page, so `<` shows up as the character instead of starting a tag; modern frameworks do this automatically. A **Content Security Policy** tells the browser which scripts the site actually expects, and marking session cookies as `HttpOnly` keeps scripts from reading them. OWASP's guidance covers all of this in detail. ### An analogy A community bulletin board where the person in charge reads every note aloud to the room. If someone pins up a note saying 'everyone, hand your house keys to the person by the door', a careless reader announces it as if it were official. A good reader says 'this note says…' and never treats it as an order. ### Keep in mind - XSS puts an attacker's script into a trusted page, run by other visitors. - The script acts with the visitor's session on that site. - Stored, reflected and DOM-based are the three main variants. - Escaping output, Content Security Policy and HttpOnly cookies are the main defenses. ### Go deeper - [Cross Site Scripting (OWASP)](https://owasp.org/www-community/attacks/xss/) - [Cross-site scripting (MDN)](https://developer.mozilla.org/en-US/docs/Web/Security/Attacks/XSS) - [Cross-site scripting on Wikipedia](https://en.wikipedia.org/wiki/Cross-site_scripting) --- ## CSS > The language that decides how a web page looks: colors, fonts, spacing, layout. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/css/ - In Spanish: [CSS](https://tidbit.legios.com.ar/es/css/) **CSS** (Cascading Style Sheets) is the language that styles web pages. HTML says "this is a heading, this is a list"; CSS says "headings are dark blue and large, the list sits in a column on the right, and on a phone everything stacks". A CSS file is a list of **rules**, each one picking some elements and setting properties, like `h1 { color: navy; }`. Keeping looks separate from content is the big idea. The same HTML can look completely different with another stylesheet, and one stylesheet can restyle thousands of pages at once. It also lets a single site adapt to a big monitor, a phone, a printer or a dark mode setting without rewriting the content. The "cascading" part is how CSS settles conflicts. When several rules apply to the same element, the more specific one wins, and among equals the later one wins. That makes it powerful and, for beginners, sometimes puzzling: a color that "will not change" is usually being overridden by another rule somewhere else. ### An analogy HTML is the furniture in a house; CSS is the interior designer's instructions: paint this wall green, put the sofa against the window, use smaller chairs in the small room. ### Keep in mind - Styles HTML: colors, fonts, spacing, layout and animations. - A rule picks elements with a selector and sets properties on them. - When rules conflict, the more specific one wins (the cascade). - Lets one page adapt to phones, big screens, print and dark mode. ### Go deeper - [CSS: Cascading Style Sheets (MDN)](https://developer.mozilla.org/en-US/docs/Web/CSS) - [CSS on Wikipedia](https://en.wikipedia.org/wiki/CSS) --- ## CSV > A plain text file that holds a table: one row per line, commas between values. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/csv/ - In Spanish: [CSV](https://tidbit.legios.com.ar/es/csv/) A **CSV** file (comma-separated values) is the simplest way to write a table down. Each line is a row, and inside a line the values are separated by commas; the first line usually holds the column names. Because it is plain text, any program can read it: a spreadsheet, a database, a script, even a text editor. A small one looks like this: `name,city,age` on the first line, then `Ana,Rosario,34` and `Luis,Salta,29`. When a value contains a comma itself, it goes between double quotes (`"Buenos Aires, AR"`). There is no single official version, so details vary: many spreadsheets set up for Spanish use semicolons, because there the comma is the decimal separator, and accents come out garbled when the file was not saved as UTF-8. CSV is the universal export button: banks, CRMs, open-data portals and analytics tools all offer it, and it is often the first step when moving data from one system to another. Its weakness is that it has no types: the file does not say whether `03/04` is a date, a fraction or text, so every program has to guess. For large analytics datasets, formats like Parquet are much smaller and faster to read. ### An analogy A shopping list written one item per line, with the quantity after a comma. Anyone can read it and anyone can write it, but nothing stops someone from writing 'a few' where a number should go. ### Keep in mind - One row per line, values separated by commas, column names on the first line. - Plain text: every spreadsheet, database and programming language can open it. - It has no types, so dates and numbers are open to interpretation. - Separators and encodings vary; if accents look broken, check that it is UTF-8. ### Go deeper - [Comma-separated values on Wikipedia](https://en.wikipedia.org/wiki/Comma-separated_values) - [RFC 4180, the common CSV format](https://www.rfc-editor.org/rfc/rfc4180) --- ## Cybersecurity > Protecting computers, networks and data from theft, damage and disruption. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/cybersecurity/ - In Spanish: [Ciberseguridad](https://tidbit.legios.com.ar/es/cybersecurity/) **Cybersecurity** is everything people and organizations do to keep their computers, networks and data safe from attack and misuse. It covers a lot of ground: passwords and two-factor authentication, encryption, firewalls, software updates, backups, training people to spot phishing, and the teams who watch for intruders. A classic way to think about it is three goals. **Confidentiality**: only the right people can see the information (your medical record, a company's plans). **Integrity**: nobody can change it without permission (your bank balance, a hospital's dosage). **Availability**: it is there when needed (the online banking site, the hospital's systems during an emergency). Most attacks break at least one: a data leak breaks confidentiality, ransomware breaks availability, a forged transfer breaks integrity. Security is not a product you install once. It is technology plus people plus habits, and attackers usually go for the weakest of the three, which is often a person tricked by a convincing message. That is why frameworks like the NIST Cybersecurity Framework organize the work as a cycle: understand what you have, protect it, detect problems, respond and recover. For an individual, a few habits cover most of the risk: updates on, a password manager, 2FA on important accounts and a backup. ### An analogy Keeping a house safe is not just a good lock: it is also closing windows, knowing who has a key, having an alarm, and insurance for when something still goes wrong. ### Keep in mind - The three classic goals: confidentiality, integrity and availability. - It mixes technology, people and processes; attackers aim for the weakest link. - Work runs as a cycle: identify, protect, detect, respond, recover. - Updates, a password manager, 2FA and backups cover most personal risk. ### Go deeper - [Computer security on Wikipedia](https://en.wikipedia.org/wiki/Computer_security) - [NIST Cybersecurity Framework](https://www.nist.gov/cyberframework) - [Cybersecurity best practices (CISA)](https://www.cisa.gov/topics/cybersecurity-best-practices) --- ## Dark web > Websites that are reachable only with special software that hides who is talking to whom. - Level: Basic - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/dark-web/ - In Spanish: [Dark web](https://tidbit.legios.com.ar/es/dark-web/) The **dark web** is a small part of the internet made of sites you cannot open with a normal browser or find on a search engine. They live on overlay networks, the best known being **Tor**, and have addresses ending in `.onion`. Visiting them requires software like the Tor Browser, which is designed so that neither the site nor anyone watching the network can easily tell who is visiting. It helps to separate three layers. The **surface web** is everything a search engine can find. The **deep web** is everything behind a login or not indexed: your email inbox, your bank account, a company's intranet. It is by far the largest part, and it is perfectly ordinary. The dark web is a thin slice on top that is deliberately hidden and anonymous. The anonymity comes from **onion routing**. Your traffic is wrapped in several layers of encryption and bounced through three volunteer-run relays around the world. Each relay removes one layer and only learns the previous and next stop, so no single point sees both who you are and where you are going. The idea came out of research at the US Naval Research Laboratory in the 1990s, and today the nonprofit Tor Project maintains the software. People use it for very different things. Journalists receive tips from sources there, people in countries with censorship use it to read the open web, and several major news outlets run `.onion` versions of their sites. It also hosts illegal markets, forums where stolen data is sold, and many scams. Using Tor is legal in most countries, but the usual laws still apply to what you do with it. For most people, the dark web matters mainly as the place where leaked passwords end up, which is why changing them after a breach matters. ### An analogy Sending a letter inside three sealed envelopes through three couriers. Each courier opens only the outer envelope to see where to take it next, so none of them knows both who sent it and who finally receives it. ### Keep in mind - Sites reachable only through networks like Tor, with addresses ending in `.onion`. - Not the same as the deep web, which is just everything behind a login. - Onion routing passes traffic through three relays, each seeing only part of the path. - Used both for privacy and free speech and for crime; using Tor itself is legal in most countries. ### Go deeper - [Dark web on Wikipedia](https://en.wikipedia.org/wiki/Dark_web) - [History of Tor (The Tor Project)](https://www.torproject.org/about/history/) --- ## Data breach > When private data an organization holds ends up with people who should not have it. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/data-breach/ - In Spanish: [Filtración de datos](https://tidbit.legios.com.ar/es/data-breach/) A **data breach** happens when information an organization is supposed to protect, such as customer emails, passwords, addresses, card numbers or medical records, is accessed or copied by someone without permission. It can be the work of attackers, but also of an employee who looks where they should not, or of a simple mistake that leaves a file open to the internet. The causes repeat. Stolen or guessed passwords for an employee account, often thanks to phishing. A known vulnerability that was never patched. A database or cloud storage bucket configured as public by accident. A laptop lost without encryption. Many breaches are discovered weeks or months later, sometimes when the data shows up for sale. For the people affected, the risk is what comes next. Leaked email and password pairs get tried on other sites, so reused passwords spread the damage. Personal details make phishing messages more convincing, and identity documents can be used for fraud. Many laws require companies to notify authorities and affected users; under Europe's GDPR, for example, the authority must usually be told within 72 hours. When you get a breach notice, act on it calmly. Change the password for that service and anywhere you reused it, turn on two-factor authentication, and watch for messages that mention the breach, because scammers follow the news too. Services like Have I Been Pwned let you check whether your email appeared in known breaches. ### An analogy A clinic's filing cabinet left unlocked in a hallway overnight. Nobody may have taken anything yet, but you cannot know who looked inside, and every patient now has to be a little more careful. ### Keep in mind - Any unauthorized access to protected data counts, malicious or accidental. - Common causes: stolen credentials, unpatched systems, public storage by mistake. - Reused passwords turn one breach into many; use a different password per site. - Laws often require the company to notify authorities and the people affected. ### Go deeper - [Data breach on Wikipedia](https://en.wikipedia.org/wiki/Data_breach) - [Have I Been Pwned: check your email in known breaches](https://haveibeenpwned.com/) --- ## Data catalog > A searchable inventory of an organization's datasets, with what each one means. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/data-catalog/ - In Spanish: [Catálogo de datos](https://tidbit.legios.com.ar/es/data-catalog/) A **data catalog** is an inventory of the data an organization has: every table, file and dashboard, with a description of what it contains, who owns it, when it was last updated and how it connects to the rest. It does not hold the data itself; it holds metadata about it and points to where the data lives. The everyday use is search. An analyst who needs "monthly revenue by country" types it into the catalog and finds the right table in the data warehouse, sees that the `amount` column is in dollars and excludes refunds, and knows who to ask if something looks off. Many catalogs also show **lineage**: which jobs produced a table and which reports depend on it. Catalogs exist because data grows faster than anyone's memory. In a data lake with thousands of folders, a file that nobody can find or understand is as good as gone. Tools such as AWS Glue Data Catalog, Google's Dataplex, Unity Catalog or the open source DataHub fill this role, and the catalog is usually where data governance rules become visible to everyone. ### An analogy A library catalog: you do not walk every shelf looking for a book, you search the catalog, read the card with the author, the topic and the shelf number, and then go straight to it. ### Keep in mind - Lists datasets with descriptions, owners, columns and freshness. - Stores metadata and pointers, not the data itself. - Lineage shows where a table comes from and what depends on it. - Keeps a data lake from turning into a swamp. ### Go deeper - [What is a data catalog? (AWS)](https://aws.amazon.com/what-is/data-catalog/) - [Data catalog on Wikipedia](https://en.wikipedia.org/wiki/Data_catalog) - [What is a data catalog? (IBM)](https://www.ibm.com/think/topics/data-catalog) --- ## Data center > A building full of servers where the internet, the cloud and AI actually run. - Level: Basic - Topics: Hardware, Cloud - URL: https://tidbit.legios.com.ar/data-center/ - In Spanish: [Centro de datos](https://tidbit.legios.com.ar/es/data-center/) A **data center** is a building designed to hold thousands of computers (servers) running around the clock. When you watch a video, send a message, pay with your phone or ask an AI a question, the work happens in one of these buildings, not on your device. The cloud is not a cloud; it is data centers, owned by companies like AWS, Google, Microsoft and Meta, or rented out rack by rack by operators like Equinix. Inside, servers sit in rows of metal racks connected by fast networks. Around them is the real engineering: a huge electricity supply with backup generators and batteries, so nothing ever goes dark; cooling that removes the heat all those machines produce; physical security; and redundant fiber links to the internet. A big data center draws as much power as a small city, which is why they are built near cheap electricity and, increasingly, next to their own power plants. AI changed the scale. Training and running LLMs needs tens of thousands of GPUs packed together, drawing far more power per rack than ordinary servers and often needing liquid cooling. The 'AI data center' buildout, now measured in gigawatts, is why electricity, water and land became part of the AI conversation, and where NVIDIA's chips end up. For the cloud, data centers are grouped into regions (a city or area) with several separate buildings each, so that a fire or a power failure in one does not take down the others. When a cloud provider says a service lives in `us-east-1`, it means a cluster of data centers in northern Virginia. ### An analogy A factory for computing. Electricity goes in, racks of machines do the work, heat comes out, and the product (every web page, stream and AI answer) ships out over fiber. ### Keep in mind - The physical place where cloud services, websites and AI models run. - Power and cooling matter as much as the servers; big sites draw hundreds of megawatts. - Built with redundancy: duplicated power, network and buildings so failures stay invisible. - AI data centers pack GPUs densely and are driving a global race for electricity. ### Go deeper - [Data center on Wikipedia](https://en.wikipedia.org/wiki/Data_center) - [AWS global infrastructure](https://aws.amazon.com/about-aws/global-infrastructure/) --- ## Data cleaning > Fixing errors, gaps and duplicates in data before anyone trusts it. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/data-cleaning/ - In Spanish: [Limpieza de datos](https://tidbit.legios.com.ar/es/data-cleaning/) **Data cleaning** is the work of finding and fixing problems in a dataset so it can be used safely. Real data is messy: the same customer typed twice with slightly different names, dates written as `03/04/2025` in one system and `2025-04-03` in another, an age of 250, a blank email, prices with a comma in one country and a dot in another. Typical steps are removing duplicates, standardizing formats (dates, currencies, country names), handling missing values (fill them, flag them or drop the row), catching values that cannot be true, and checking that references match, for example that every order belongs to a customer who exists. Each decision is a judgment call, so good teams write the rules down and run them as code instead of fixing cells by hand. It is famously the least glamorous and most time-consuming part of working with data, and also one of the most important. A dashboard or a machine learning model is only as good as the data it is fed; "garbage in, garbage out" is the oldest saying in the field. In a pipeline, cleaning usually happens in the transform step of ETL, and the checks keep running every time new data arrives. ### An analogy Washing and sorting vegetables before cooking: you throw out the bruised ones, remove the dirt and cut everything the same size. Skip it and even a great recipe tastes off. ### Keep in mind - Fixes duplicates, inconsistent formats, missing values and impossible numbers. - Every rule is a decision; write it down and automate it. - Usually part of the transform step in a data pipeline. - Garbage in, garbage out: reports and AI models inherit every error. ### Go deeper - [Data cleansing on Wikipedia](https://en.wikipedia.org/wiki/Data_cleansing) - [What is data cleansing? (AWS)](https://aws.amazon.com/what-is/data-cleansing/) - [What is data cleaning? (IBM)](https://www.ibm.com/think/topics/data-cleaning) --- ## Data governance > The rules and roles that decide who owns data, who can use it and how it is kept right. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/data-governance/ - In Spanish: [Gobierno de datos](https://tidbit.legios.com.ar/es/data-governance/) **Data governance** is the set of decisions, roles and rules an organization uses to manage its data on purpose instead of by accident. It answers questions like: who is responsible for the customer table, what does "active customer" mean, who is allowed to see salaries, how long do we keep old records, and what happens when two reports disagree. In practice it is part paperwork and part tooling. There are **data owners** who answer for a dataset, **data stewards** who look after its quality day to day, written definitions, and access policies. Then there are tools that enforce them: permissions in the database or data warehouse, a data catalog that documents what exists, automatic quality checks and records of where each number came from (its **lineage**). It matters because data is shared by many teams and regulated by laws such as Europe's GDPR. Without governance, the same metric gets calculated three different ways, sensitive data ends up in a spreadsheet nobody controls, and nobody can answer a regulator's question. Done badly, it turns into bureaucracy that slows everyone down; done well, it is mostly invisible and lets people trust the numbers. ### An analogy The rules of a shared kitchen in an office: who restocks the fridge, what gets labeled, what is off limits and when leftovers are thrown out. Without them, the kitchen still works for a week, and then nobody wants to touch the milk. ### Keep in mind - Defines ownership, meanings, access and retention for an organization's data. - Mixes people (owners, stewards) with tools (permissions, catalogs, quality checks). - Helps meet privacy and record-keeping laws such as GDPR. - The goal is trusted data, not more bureaucracy. ### Go deeper - [What is data governance? (AWS)](https://aws.amazon.com/what-is/data-governance/) - [Data governance on Wikipedia](https://en.wikipedia.org/wiki/Data_governance) --- ## Data lake > Cheap storage where all your data is kept as raw files, to be sorted out later. - Level: Intermediate - Topics: Data, Cloud - URL: https://tidbit.legios.com.ar/data-lake/ - In Spanish: [Data lake](https://tidbit.legios.com.ar/es/data-lake/) A **data lake** is a big, cheap store where an organization keeps all its data as files, in whatever shape it arrives: app logs, CSV exports, JSON events, images, sensor readings. Nothing has to be modeled before it lands. On AWS it is usually Amazon S3 with a folder layout and some rules; the structure is applied later, when someone reads the data. That is the key difference with a data warehouse. A warehouse asks you to clean and model data before loading it (**schema on write**); a lake accepts everything and lets each reader decide how to interpret it (**schema on read**). Query engines like Amazon Athena or Spark read the files where they are, and a **catalog** records what each folder contains so people can find it. Lakes are popular because storage is cheap and nothing gets thrown away: tomorrow's question, or tomorrow's machine learning model, may need data nobody thought was useful today. The risk is the **data swamp**: years of undocumented, duplicated or half-broken files that nobody trusts. Many teams now combine both ideas in a **lakehouse**: files in the lake, stored in formats like Parquet, with a table layer on top that adds warehouse-style guarantees. ### An analogy A huge storage unit where everything goes in its original box. It is cheap and nothing is lost, but if nobody labels the boxes, finding last year's tax papers becomes an expedition. ### Keep in mind - Stores raw files of any type; structure is applied when the data is read. - Usually built on object storage like Amazon S3, which is cheap and practically unlimited. - Without a catalog and some discipline, it turns into a data swamp. - Lakes and warehouses complement each other; a lakehouse mixes both. ### Go deeper - [What is a data lake? (AWS)](https://aws.amazon.com/what-is/data-lake/) - [Data lake on Wikipedia](https://en.wikipedia.org/wiki/Data_lake) --- ## Data mesh > An organizational approach where each business team owns and publishes its own data. - Level: Advanced - Topics: Data - URL: https://tidbit.legios.com.ar/data-mesh/ - In Spanish: [Data mesh](https://tidbit.legios.com.ar/es/data-mesh/) A **data mesh** is a way of organizing who does what with data in a large company, more than a piece of technology. In the classic setup, every department sends its data to one central data team, which builds the data warehouse or data lake for everyone. As the company grows, that team becomes a bottleneck: it is far from the business, it does not know what each field really means, and every request waits in its queue. The idea, described by Zhamak Dehghani around 2019, rests on four principles. **Domain ownership**: the team that runs a part of the business (sales, shipping, payments) also owns its data. **Data as a product**: that team publishes datasets other teams can rely on, with documentation, quality guarantees and someone accountable. **Self-serve platform**: a central platform team provides the shared tools so each domain does not reinvent storage, pipelines and catalogs. **Federated governance**: common rules on security, privacy and formats, agreed together and enforced automatically. Data mesh makes the most sense for big organizations with many teams and mature engineering; for a small company, one good central team is usually simpler. It is often confused with a product you can buy, but no tool gives you a mesh. Tools like a data catalog, dbt or a shared warehouse can support it; the hard part is changing responsibilities and incentives. ### An analogy A food market instead of one big restaurant kitchen. Each stall cooks its own specialty and is responsible for it, but they all follow the same health rules and share the tables, the water and the electricity. ### Keep in mind - Each business domain owns, documents and serves its own data. - Datasets are treated as products, with users, quality guarantees and an owner. - A central platform provides shared tools; governance rules are common. - An organizational model for large companies, not a product you install. ### Go deeper - [Data mesh principles (Zhamak Dehghani, martinfowler.com)](https://martinfowler.com/articles/data-mesh-principles.html) - [What is a data mesh? (AWS)](https://aws.amazon.com/what-is/data-mesh/) - [Data mesh on Wikipedia](https://en.wikipedia.org/wiki/Data_mesh) --- ## Data science > Using statistics, code and business sense to answer questions with data. - Level: Basic - Topics: Data, AI - URL: https://tidbit.legios.com.ar/data-science/ - In Spanish: [Ciencia de datos](https://tidbit.legios.com.ar/es/data-science/) **Data science** is the work of getting useful answers out of data. It mixes three skills: **statistics**, to know what the numbers really say and how sure you can be; **programming**, usually in Python, R and SQL, to handle data too big for a spreadsheet; and **domain knowledge**, to ask questions that matter for a business, a hospital or a lab. A typical project starts with a question, like "which customers are about to cancel?". The **data scientist** gathers data from databases and files, then cleans it: fixing typos, filling or removing gaps, merging sources that disagree. That step famously takes most of the time. Then comes exploring with charts, building a **model** that finds patterns or makes predictions, and finally explaining the result to people who will act on it, often in a notebook, a dashboard or a short presentation. It overlaps with neighbors. **Machine learning** is one of its main tools, used when the goal is prediction; **business intelligence** focuses more on reporting what already happened; **data engineering** builds the pipelines that bring the data in. The role grew in the 2010s as companies collected more data than they knew what to do with, and it remains central now that AI models are trained and evaluated with the same methods. ### An analogy Like being a detective with spreadsheets. Most of the job is gathering and checking evidence, the dramatic deduction takes an afternoon, and then you still have to convince the jury. ### Keep in mind - Combines statistics, programming and knowledge of the field. - Cleaning and preparing data takes most of the time. - Ends with communication: an answer nobody understands changes nothing. - Machine learning is a core tool, but not the whole job. ### Go deeper - [What is data science? (AWS)](https://aws.amazon.com/what-is/data-science/) - [Data science on Wikipedia](https://en.wikipedia.org/wiki/Data_science) --- ## Data streaming > Processing data as a continuous flow of events, seconds after they happen. - Level: Advanced - Topics: Data, Cloud - URL: https://tidbit.legios.com.ar/data-streaming/ - In Spanish: [Streaming de datos](https://tidbit.legios.com.ar/es/data-streaming/) **Data streaming** means handling data as a continuous flow of small events instead of in big periodic batches. Each card payment, click, GPS position or sensor reading is sent the moment it happens, and programs downstream react to it within seconds. A nightly ETL answers 'what happened yesterday?'; a stream answers 'what is happening right now?'. At the center there is usually a **log** of events, such as Apache Kafka or Amazon Kinesis. Producers append events to the end, the log keeps them in order for days, and any number of consumers read at their own pace, each remembering where it left off. Because events are kept, a new consumer can start from the beginning, and one that crashes picks up where it stopped without losing anything. Streams power things that cannot wait: a bank blocking a suspicious card payment before it is approved, a delivery app moving the courier on your map, a factory raising an alert when a machine overheats. The cost is complexity. Events arrive late or out of order, the same event can be delivered twice, and a question like 'sales in the last five minutes' needs **windows** over a flow that never ends. When an answer every hour is good enough, a batch pipeline is simpler and cheaper. ### An analogy Reading the morning paper versus following a live radio broadcast. The paper (batch) is tidy but a day late; the radio (stream) tells you as it happens, and you can tune in at any moment. ### Keep in mind - Events are processed one by one, seconds after they happen, instead of in nightly batches. - A durable log such as Kafka or Kinesis keeps events in order and lets many consumers read them. - Late, duplicated and out-of-order events are normal and must be handled. - Use it when minutes matter; otherwise a batch ETL is simpler. ### Go deeper - [What is streaming data? (AWS)](https://aws.amazon.com/what-is/streaming-data/) - [Introduction to Apache Kafka](https://kafka.apache.org/intro) - [Stream processing on Wikipedia](https://en.wikipedia.org/wiki/Stream_processing) --- ## Data warehouse > A database built for analysis: years of history, from many sources, in one place. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/data-warehouse/ - In Spanish: [Data warehouse](https://tidbit.legios.com.ar/es/data-warehouse/) A **data warehouse** is a database designed to answer business questions, not to run an app. The app's own database is busy recording each sale as it happens; the warehouse receives copies of that data, plus data from the CRM, the payments provider and the ad platforms, and keeps years of history in one place, cleaned and organized so the pieces fit together. The difference shows in the questions. An app asks 'what is in Ana's cart right now?', touching one row. An analyst asks 'how did sales per region change month by month over the last three years?', reading hundreds of millions of rows. Warehouses store data **column by column** and spread the work across many machines, so that second kind of question takes seconds. Amazon Redshift, Snowflake and Google BigQuery are well-known ones, and you talk to all of them in SQL. Data usually arrives through ETL pipelines that run every night or every few minutes, and the tables are modeled around the business: **facts** (each sale) surrounded by **dimensions** (customer, product, date). Dashboards, monthly reports and forecasts are built on top. The warehouse becomes the single source of truth: when two teams disagree about a number, it is where they go to settle it. ### An analogy A company's central archive. Each department keeps its working papers at its own desk (the app databases); every night copies are filed in the archive, labeled the same way, so anyone can pull five years of records and compare them side by side. ### Keep in mind - Built for analysis over a long history, not for the app's moment-to-moment work. - Stores data by columns and spreads each query over many machines. - Fed by ETL pipelines from many sources, and queried with SQL. - Amazon Redshift, Snowflake and Google BigQuery are common examples. ### Go deeper - [What is a data warehouse? (AWS)](https://aws.amazon.com/what-is/data-warehouse/) - [Data warehouse on Wikipedia](https://en.wikipedia.org/wiki/Data_warehouse) --- ## Database > An organized store of data that programs can search and update safely. - Level: Basic - Topics: Software, Data - URL: https://tidbit.legios.com.ar/database/ - In Spanish: [Base de datos](https://tidbit.legios.com.ar/es/database/) A **database** is a program whose whole job is to store data and answer questions about it quickly and safely: every user of an app, every order, every message. Unlike a spreadsheet, it is built to be used by many programs at once, to hold millions of rows, and to never lose or half-save a change, even if the power goes out in the middle of a write. The most common kind is **relational**: data lives in tables with rows and columns (a `customers` table, an `orders` table), tables point to each other by ids, and you ask questions in SQL. PostgreSQL, MySQL and SQLite are relational and open source; Amazon RDS runs them for you in the cloud. **NoSQL** databases drop the strict tables for flexible documents (MongoDB), key-value pairs (Redis, Amazon DynamoDB) or graphs, trading some guarantees for speed or scale. Almost every app you use is a thin layer over a database. Open Instagram and a query fetches your feed; pay someone and a **transaction** moves money between two accounts as a single all-or-nothing step. Backups, indexes (so a row can be found without scanning the whole table) and permissions are the daily work of keeping one healthy. ### An analogy A library with a catalog. Books (data) sit on numbered shelves (tables), the catalog (index) tells you where to look without walking every aisle, and the librarian (the database engine) makes sure two people do not rewrite the same card at the same time. ### Keep in mind - Relational databases store tables and speak SQL; NoSQL ones store documents, key-value pairs or graphs. - A transaction is a group of changes that either all happen or none do. - An index makes searches fast at the cost of some space and slightly slower writes. - The database is usually the hardest part of a system to scale and the most important to back up. ### Go deeper - [What is a database? (AWS)](https://aws.amazon.com/what-is/database/) - [Database (Wikipedia)](https://en.wikipedia.org/wiki/Database) --- ## Database index > A sorted shortcut that lets a database find rows without reading the whole table. - Level: Intermediate - Topics: Data, Software - URL: https://tidbit.legios.com.ar/database-index/ - In Spanish: [Índice de base de datos](https://tidbit.legios.com.ar/es/database-index/) A **database index** is an extra structure, kept next to a table, that lets the database jump straight to the rows you ask for. Without one, a query like 'find the customer with email ana@example.com' forces a **full scan**: reading every row until it shows up. With an index on the email column, the database looks the value up in a sorted structure and goes directly to the right row. Most indexes are a **B-tree**: a sorted, branching structure where each step narrows the search, so finding one value among a hundred million takes a handful of jumps instead of a hundred million reads. You create one with a line of SQL, `CREATE INDEX ON customers (email);`, and from then on the database keeps it up to date by itself and decides when to use it. Indexes are not free. Each one takes disk space, and every insert or update has to update the indexes too, so writes get slower. The craft is choosing them from the queries the app really runs: the columns used to filter, join and sort. When a page that used to load instantly starts taking seconds as the data grows, a missing index is the first suspect. ### An analogy The index at the back of a book. To find every page that mentions 'photosynthesis' you check the alphabetical list instead of reading the whole book; the price is a few extra pages that must be redone in every new edition. ### Keep in mind - An index trades disk space and slower writes for much faster reads. - Without an index, the database has to scan every row to answer a query. - Index the columns you filter, join and sort by, not every column. - Most indexes are B-trees, which keep values sorted so a lookup takes a few steps. ### Go deeper - [Indexes, introduction (PostgreSQL docs)](https://www.postgresql.org/docs/current/indexes-intro.html) - [Database index on Wikipedia](https://en.wikipedia.org/wiki/Database_index) --- ## Database schema > The blueprint of a database: which tables exist, their columns and how they connect. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/database-schema/ - In Spanish: [Esquema de base de datos](https://tidbit.legios.com.ar/es/database-schema/) A **database schema** is the description of how a database is organized: which tables it has, which columns each table has, what type of value goes in each column (a number, a date, some text), which columns are required, and how tables relate to each other. It describes the shape of the data, not the data itself. Take an online shop. The schema might say there is a `customers` table with an `id`, a `name` and an `email` that cannot be empty, and an `orders` table with an `id`, a `customer_id`, a `total` with two decimals and a `created_at` date. With that in place, the database refuses an order with a total of "banana" or with no customer. In SQL, the schema is written with commands like `CREATE TABLE` and changed with `ALTER TABLE`. Schemas change as products change, and that is delicate: renaming a column can break every report and app that reads it. Teams handle this with **migrations**, small versioned scripts that change the schema step by step and can be reviewed like code. The word also has a narrower meaning in some databases, such as PostgreSQL, where a schema is a named folder that groups tables. ### An analogy The blueprint of a building: it does not tell you who lives there, but it says how many rooms there are, how big each one is and which doors connect them. You cannot put a fridge where the plan says there is a staircase. ### Keep in mind - Defines tables, columns, data types, required fields and relationships. - Lets the database reject data with the wrong shape. - Written and changed in SQL, ideally through versioned migrations. - A warehouse checks the schema on write; a data lake applies it on read. ### Go deeper - [Database schema on Wikipedia](https://en.wikipedia.org/wiki/Database_schema) - [What is a database schema? (IBM)](https://www.ibm.com/think/topics/database-schema) --- ## dbt > A tool for writing data transformations as SQL files, with tests and docs, like software. - Level: Advanced - Topics: Data - URL: https://tidbit.legios.com.ar/dbt/ - In Spanish: [dbt](https://tidbit.legios.com.ar/es/dbt/) **dbt** (originally "data build tool") is an open source tool for the transform step of a data pipeline, the T in ELT. The data has already been loaded into a data warehouse such as Snowflake, BigQuery, Redshift or Databricks; dbt helps you turn those raw tables into clean, trusted ones. Each transformation is a **model**: a plain `.sql` file with a `SELECT` statement, saved in a git repository. Models can refer to each other with `{{ ref('orders') }}`, so dbt knows the order in which to build them and draws the whole chain as a graph. Running `dbt run` sends each model's SQL to the warehouse, which does the heavy work; dbt itself moves no data. `dbt test` checks rules like "this column is never empty" or "every order has a customer", and `dbt docs` generates a website describing every table and column, with lineage. Its impact was cultural as much as technical: it brought software habits like version control, code review, testing and reusable pieces to analysts who already knew SQL. That gave rise to the role of **analytics engineer**. dbt Labs, the company behind it, also sells a hosted platform, but the core tool is open source and runs from the command line. ### An analogy A recipe book for a restaurant kitchen where each recipe says which other preparations it needs. The chef (the warehouse) does the cooking; dbt reads the book, works out the order, and checks every plate before it goes out. ### Keep in mind - Each model is a SELECT in a .sql file; dbt builds them inside the warehouse. - ref() links models into a dependency graph with automatic build order. - Built-in tests and generated documentation with lineage. - Brought git, code review and testing to SQL analytics work. ### Go deeper - [What is dbt? (official docs)](https://docs.getdbt.com/docs/introduction) - [dbt Core on GitHub](https://github.com/dbt-labs/dbt-core) - [About dbt models (official docs)](https://docs.getdbt.com/docs/build/models) --- ## DDoS > Knocking a site offline by flooding it with more traffic than it can handle. - Level: Intermediate - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/ddos/ - In Spanish: [DDoS](https://tidbit.legios.com.ar/es/ddos/) A **DDoS attack** (distributed denial of service) does not break into anything: it buries a website or service under so many requests that real users cannot get through. Every server, link and router has a limit; push enough traffic at it and it slows to a crawl or falls over. 'Distributed' means the flood comes from thousands or millions of machines at once, so blocking a single IP address does nothing. Those machines are usually a **botnet**: computers, routers, security cameras and other devices infected with malware and controlled remotely, mostly without their owners noticing. The attacker, or someone who rents the botnet to them by the hour, tells them all to send traffic at the same target. Some attacks fill the network pipe with raw volume; others are subtler and send requests that look normal but are expensive for the server to answer, like repeated searches or logins. The defense is mostly size and filtering. Services like AWS Shield or Cloudflare sit in front of the site, spread across data centers all over the world, so the flood is absorbed by a network far bigger than any attack, and **scrubbing** systems separate the bot traffic from the people. A CDN helps on its own, because cached pages are served from the edge without touching the origin. Attacks are often used for extortion, protest or as a distraction while something else happens. ### An analogy A shop where a crowd of hired people walks in all at once and stands in line asking for nothing. The shop is not robbed, but no real customer can get to the counter until a doorman starts telling them apart. ### Keep in mind - A DDoS overwhelms a service with traffic; it steals nothing, it makes the service unusable. - The traffic comes from a botnet of infected devices, so blocking one address is useless. - Protection comes from huge networks that absorb and filter the flood before it reaches you. - Some attacks use raw volume, others cheap-to-send, expensive-to-answer requests. ### Go deeper - [What is a DDoS attack? (AWS Shield)](https://aws.amazon.com/shield/ddos-attack-protection/) - [Denial-of-service attack on Wikipedia](https://en.wikipedia.org/wiki/Denial-of-service_attack) --- ## Debugging > The detective work of finding why a program misbehaves, and fixing it. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/debugging/ - In Spanish: [Debugging](https://tidbit.legios.com.ar/es/debugging/) **Debugging** is the process of tracking down a bug and fixing it. The hard part is rarely the fix itself; it is finding the cause. A wrong number on the screen might come from a typo, from bad data, from two parts of the program disagreeing about a unit, or from something that only happens on Tuesdays in one time zone. Most debugging follows the same loop. First, **reproduce** the problem reliably: if you cannot make it happen on purpose, you cannot tell whether you fixed it. Then **narrow it down**, ruling out half the possibilities at a time until the faulty line is found. Then fix it and test again, ideally adding an automated test so the same bug cannot quietly return. Programmers have tools for this. The simplest is printing values to see what the program is actually doing. A **debugger** goes further: it can pause the program at a chosen line (a **breakpoint**), show every variable at that moment and run the code one step at a time. Logs, error reports from users and even explaining the problem out loud to a colleague, or to a rubber duck, are part of the toolkit. ### An analogy A mechanic chasing a strange noise in a car. They first make the noise happen again, then check one part after another, eliminating suspects, until the loose bolt turns up. ### Keep in mind - Finding the cause is usually harder than writing the fix. - Reproduce, narrow down, fix, then test that it stays fixed. - A debugger pauses code at breakpoints and shows its values. - Explaining the problem out loud often reveals the answer. ### Go deeper - [Debugging on Wikipedia](https://en.wikipedia.org/wiki/Debugging) - [Debug JavaScript with Chrome DevTools (Chrome for Developers)](https://developer.chrome.com/docs/devtools/javascript) --- ## Deep learning > Machine learning with many-layered neural networks, behind most recent AI leaps. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/deep-learning/ - In Spanish: [Deep learning](https://tidbit.legios.com.ar/es/deep-learning/) **Deep learning** is machine learning done with neural networks that have many layers stacked one after another. 'Deep' refers to that depth, not to any profound understanding. Each layer takes the output of the previous one and turns it into something a bit more abstract, so the network builds its own understanding of the data step by step. In an image model, the first layers end up reacting to simple things like edges and color patches, the middle layers to textures and shapes like an eye or a wheel, and the last layers to whole concepts like 'cat' or 'bicycle'. Nobody programs those stages; they emerge during training. That is the big change from older machine learning, where people had to decide by hand which features of the data mattered. The ideas are decades old, but deep learning only took off around 2012, when a network called AlexNet won an image recognition contest by a wide margin. Three things had come together: huge labeled datasets, GPUs fast enough to train large networks, and better training tricks. Since then it has powered speech recognition, translation, image generation and the large language models behind chatbots. The cost: it needs a lot of data and computing power, and it is hard to explain why a deep network gave a particular answer. ### An analogy Like an assembly line of interpreters: the first only notices lines, the next turns lines into shapes, the next turns shapes into objects. No single worker understands the whole picture, but the last one can say 'that is a cat'. ### Keep in mind - 'Deep' means many layers, each building on the one before. - It learns which features matter on its own, without hand-picked rules. - It took off around 2012 thanks to big datasets and GPUs. - Powerful but data-hungry, expensive to train and hard to explain. ### Go deeper - [Deep learning on Wikipedia](https://en.wikipedia.org/wiki/Deep_learning) - [What is deep learning? (IBM)](https://www.ibm.com/think/topics/deep-learning) --- ## Deepfake > A fake video, photo or voice made with AI to look and sound like a real person. - Level: Basic - Topics: AI, Security - URL: https://tidbit.legios.com.ar/deepfake/ - In Spanish: [Deepfake](https://tidbit.legios.com.ar/es/deepfake/) A **deepfake** is media created or altered with AI so that a real person appears to say or do something they never did. It can be a face swapped into a video, a cloned voice reading any text, or a photo that never happened. The word joins 'deep learning' and 'fake', and comes from a Reddit user who posted face-swapped videos in 2017. Making one used to take skill and hours of footage. Today generative AI tools can clone a voice from a short recording and animate a face from a few photos, and the results keep getting harder to tell apart from the real thing. There are honest uses: dubbing films into other languages, restoring voices for people who lost them, visual effects, satire that is clearly labeled. The harm comes from deception. Scammers call parents with a cloned voice of their child asking for money, or pose as a company's boss on a video call to order an urgent transfer. Deepfakes are used for non-consensual sexual images, which many countries now treat as a crime, and to spread political misinformation. Detection tools exist but are not reliable. The best defense is a habit: if a message asks for money, passwords or urgency, check it through another channel you already trust, such as calling back on a known number or agreeing on a family code word. ### An analogy A deepfake is a forged signature for faces and voices. A forger used to need years of practice; now the forging machine is available to anyone, so you check the document through another route instead of trusting how it looks. ### Keep in mind - AI-made video, audio or images that imitate a real person. - A voice can be cloned from a short recording. - Common in scams: fake calls from relatives or bosses asking for money. - Verify urgent requests through a second, trusted channel. ### Go deeper - [Deepfake on Wikipedia](https://en.wikipedia.org/wiki/Deepfake) - [What is a deepfake? (IBM)](https://www.ibm.com/think/topics/deepfake) - [Increasing threat of deepfake identities (US DHS, PDF)](https://www.dhs.gov/sites/default/files/publications/increasing_threats_of_deepfake_identities_0.pdf) --- ## DeepSeek > A Chinese AI lab whose open-weight models showed that top AI could be built for less. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/deepseek/ - In Spanish: [DeepSeek](https://tidbit.legios.com.ar/es/deepseek/) **DeepSeek** is an AI company based in Hangzhou, China, founded in 2023 by Liang Wenfeng and funded by High-Flyer, the quantitative investment fund he co-founded. It builds large language models, offers them in a free chat app and sells access to them through an API at low prices. It drew worldwide attention in January 2025 with **DeepSeek-R1**, a reasoning model that thinks step by step before answering, built on top of its earlier model DeepSeek-V3. R1 performed close to the best models from American labs on many tests, but DeepSeek published its **weights** under a permissive license, so anyone could download it and run it, together with technical reports explaining how it was trained. The reports also described tricks to train efficiently, such as a mixture of experts design and using reinforcement learning to teach reasoning, and a training cost far lower than people expected, although that figure covered only the final training run. Its arrival mattered for several reasons. It suggested that strong models did not require the biggest budgets, it came from China despite US limits on exports of the most advanced AI chips, and it gave researchers everywhere an open model to study and adapt. It also raised concerns: the app stores data on servers in China, some governments banned it on official devices, and its hosted service avoids or follows official lines on topics that are sensitive in China. ### An analogy A small kitchen that cooks a dish as good as the famous restaurants for a fraction of the cost, and then publishes the full recipe for anyone to try at home. ### Keep in mind - Chinese AI company founded in 2023 in Hangzhou, backed by the fund High-Flyer. - DeepSeek-R1 (2025) rivaled top models and was released with open weights. - Its technical reports showed how to train strong models more efficiently. - Its app has raised privacy and censorship concerns in several countries. ### Go deeper - [DeepSeek on GitHub (models and reports)](https://github.com/deepseek-ai) - [DeepSeek on Wikipedia](https://en.wikipedia.org/wiki/DeepSeek) - [DeepSeek official site](https://www.deepseek.com/) --- ## DevOps > Breaking the wall between people who write software and people who run it. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/devops/ - In Spanish: [DevOps](https://tidbit.legios.com.ar/es/devops/) **DevOps** is a way of working that joins **development** (writing the software) and **operations** (running it in production) into one continuous effort. The name and the movement took off around 2009. Before, a common setup was that developers finished a feature and "threw it over the wall" to an operations team that had to install it, keep it alive and get woken up at 3 a.m. when it broke. Each side blamed the other. DevOps says the same team should care about the whole life of the software, captured in the phrase "you build it, you run it". In practice that comes with habits and tools: CI/CD pipelines that test and deploy every change, infrastructure as code so servers are created from files instead of by hand, monitoring and alerts so problems are seen before customers report them, and blameless reviews after incidents to fix causes rather than find culprits. It is mainly a culture and set of practices, but the word is also used as a job title: a **DevOps engineer** usually builds and maintains the pipelines, cloud infrastructure and monitoring the rest of the team relies on. Variations add other concerns to the loop, like **DevSecOps**, which builds security checks into every step instead of leaving them for the end. ### An analogy Like a restaurant where the cooks also see the dining room. If a dish keeps coming back cold, they find out that night and fix it, instead of reading a complaint from a separate department a month later. ### Keep in mind - Development and operations share responsibility for software in production. - Built on automation: CI/CD, infrastructure as code, monitoring. - Aims for small, frequent, safe changes and fast recovery from failures. - It is a culture first; the job title came later. ### Go deeper - [What is DevOps? (AWS)](https://aws.amazon.com/devops/what-is-devops/) - [DevOps on Wikipedia](https://en.wikipedia.org/wiki/DevOps) --- ## DHCP > The service that gives each device an IP address automatically when it joins a network. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/dhcp/ - In Spanish: [DHCP](https://tidbit.legios.com.ar/es/dhcp/) **DHCP** (Dynamic Host Configuration Protocol) is why you can walk into a café, pick the Wi-Fi and be online a second later without typing any numbers. When a device joins a network, it shouts 'is there a DHCP server here?'. The server, usually built into the router at home or a dedicated machine at an office, answers with an IP address that is free, and the device accepts it. The answer contains more than the address. It also says which addresses are local (the subnet mask), which router to send everything else to (the default gateway) and which DNS servers to use for looking up names. Before DHCP, someone had to type all of that by hand into every computer, and two machines given the same address would knock each other offline. The address is a **lease**, not a gift: it is lent for a while, from minutes to days, and the device quietly renews it while it stays. When it leaves, the address goes back to the pool for the next visitor. Printers and servers that need to be found at the same place every day get a **reservation**: the server always hands that device, recognized by its MAC address, the same IP address. When something goes wrong, DHCP is a common suspect. If your computer shows an address starting with `169.254`, it asked and nobody answered, so it gave itself a placeholder that cannot reach the internet. Restarting the router or turning Wi-Fi off and on usually triggers a fresh request. ### An analogy The front desk of a hotel. You arrive, they hand you a room key, tell you where breakfast is and when checkout is, and when you leave the room goes to the next guest. ### Keep in mind - Hands out an IP address, the gateway and the DNS servers automatically. - Addresses are leased for a time and renewed while the device stays. - At home, the router is almost always the DHCP server. - An address starting with `169.254` means the device got no answer. ### Go deeper - [DHCP on Wikipedia](https://en.wikipedia.org/wiki/Dynamic_Host_Configuration_Protocol) - [Dynamic Host Configuration Protocol (Microsoft Learn)](https://learn.microsoft.com/en-us/windows-server/networking/technologies/dhcp/dhcp-top) --- ## Diffusion model > An image generator that starts from random noise and cleans it up, step by step. - Level: Advanced - Topics: AI - URL: https://tidbit.legios.com.ar/diffusion-model/ - In Spanish: [Modelo de difusión](https://tidbit.legios.com.ar/es/diffusion-model/) A **diffusion model** is the kind of generative AI behind most image generators, such as Stable Diffusion, and many video generators. It creates a picture by starting from pure random noise, like TV static, and removing a little of that noise at each step until a clear image emerges. A text description, the prompt, steers each step, so the noise turns into 'a mountain at sunrise' rather than anything else. The training is the clever part. Take millions of images with captions and gradually ruin each one by adding noise, step by step, until nothing is left but static. Then train a neural network to do the reverse: given a noisy image and its caption, predict what noise was added so it can be subtracted. A model that is good at undoing one small step of damage can, by repeating it dozens of times, build a brand-new image from static. The modern form of this approach was laid out in a 2020 paper on denoising diffusion, and it quickly overtook earlier techniques in image quality. Running directly on millions of pixels is slow, so most systems use **latent diffusion**: they first compress images into a much smaller internal representation, do the denoising there, and expand the result into a full image at the end. The same method works for video, audio and even protein design. Because the output depends on the random noise it starts from, the same prompt gives a different picture each time. And because these models learn from huge collections of images from the web, they raise ongoing questions about artists' rights, deepfakes and the stereotypes they reproduce. ### An analogy Like a sculptor who sees a figure in a rough block of marble and chips away a little at a time. Here the block is random static, the chisel is the model, and the prompt tells the sculptor what figure to look for. ### Keep in mind - Generates images by removing noise in many small steps. - Trained by adding noise to real images and learning to undo it. - Latent diffusion works in a compressed space to be faster. - Same prompt, different starting noise, different image. ### Go deeper - [Diffusion model on Wikipedia](https://en.wikipedia.org/wiki/Diffusion_model) - [What are diffusion models? (IBM)](https://www.ibm.com/think/topics/diffusion-models) - [Denoising Diffusion Probabilistic Models (paper, arXiv)](https://arxiv.org/abs/2006.11239) --- ## Disaster recovery > The plan for getting systems and data back after a major failure. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/disaster-recovery/ - In Spanish: [Recuperación ante desastres](https://tidbit.legios.com.ar/es/disaster-recovery/) **Disaster recovery** (DR) is how an organization gets its systems running again after something big goes wrong: a whole data center or cloud region goes down, a ransomware attack encrypts the servers, a bad deploy corrupts the database, or a flood reaches the building. Backups are part of it, but DR is the whole plan: what gets restored first, where it runs, who decides, and how long it all takes. Two numbers shape every plan. The **RPO** (recovery point objective) is how much data you can afford to lose, measured in time: an RPO of one hour means losing at most the last hour of changes. The **RTO** (recovery time objective) is how long you can afford to be down. A blog might accept a day for both; a payment system may need seconds. Tighter numbers cost more, because they require more copies, more automation and spare capacity sitting ready. On the cloud, common strategies go from cheap and slow to expensive and fast: **backup and restore** in another region, a **pilot light** with only the core pieces running, a **warm standby** with a smaller copy of everything, and **active-active**, where two regions serve traffic at the same time. Whatever the choice, the plan only works if it is practiced; teams run regular drills where they actually fail over and time how long it takes. ### An analogy A fire drill for your systems. Having extinguishers is the backup; knowing the exits, who calls whom and where everyone meets is the disaster recovery plan, and you only know it works if you have practiced it. ### Keep in mind - Covers restoring systems and data after a big failure, not just keeping copies. - RPO: how much data you can lose. RTO: how long you can be down. - Strategies range from backup and restore to active-active across regions. - Untested plans fail; run drills regularly. ### Go deeper - [What is disaster recovery? (AWS)](https://aws.amazon.com/what-is/disaster-recovery/) - [Disaster recovery of workloads on AWS (whitepaper)](https://docs.aws.amazon.com/whitepapers/latest/disaster-recovery-workloads-on-aws/disaster-recovery-workloads-on-aws.html) - [Disaster recovery on Wikipedia](https://en.wikipedia.org/wiki/Disaster_recovery) --- ## Display resolution > How many dots, or pixels, make up a screen's image: more dots, finer detail. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/display-resolution/ - In Spanish: [Resolución de pantalla](https://tidbit.legios.com.ar/es/display-resolution/) Every screen is a grid of tiny colored dots called pixels. The **display resolution** is how many there are, written as width by height. Full HD, also called 1080p, is 1920 × 1080 pixels, about two million dots. **4K** (on TVs, strictly "4K UHD") is 3840 × 2160: twice as many across and twice as many down, so four times as many pixels in total. More pixels can show finer detail, but only if they are small enough for your eyes to tell apart. What really matters is **pixel density**, measured in pixels per inch (PPI), together with how far away you sit. A phone held close to your face needs a very high density to look sharp; a big TV across the room looks fine at a much lower one. That is why a 4K phone and a 4K TV have the same number of pixels and feel completely different. Resolution also has costs. Every extra pixel has to be computed, sent and stored: a 4K video needs much more bandwidth and storage than an HD one, and a game in 4K asks far more of the GPU. And picture quality depends on other things too, such as contrast, brightness, color accuracy and how many times per second the screen refreshes, so a bigger number on the box does not automatically mean a better image. ### An analogy A mosaic made of tiles. With many small tiles you can draw eyelashes; with a few big ones you only get the outline of a face. From across the room, though, both mosaics may look the same. ### Keep in mind - Resolution is the number of pixels, written as width × height. - Full HD is 1920 × 1080; 4K is 3840 × 2160, four times as many pixels. - Sharpness depends on pixel density and viewing distance, not on resolution alone. - More pixels need more bandwidth, storage and graphics power. ### Go deeper - [Display resolution on Wikipedia](https://en.wikipedia.org/wiki/Display_resolution) - [4K resolution on Wikipedia](https://en.wikipedia.org/wiki/4K_resolution) - [Pixel on Wikipedia](https://en.wikipedia.org/wiki/Pixel) --- ## DNS > The internet's phone book: it turns names like wikipedia.org into IP addresses. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/dns/ - In Spanish: [DNS](https://tidbit.legios.com.ar/es/dns/) **DNS** (Domain Name System) is the service that translates the names people type, like `wikipedia.org`, into the IP address computers actually use to find each other. Every time you open a website, send an email or tap an app, a DNS lookup happens first, usually in a few milliseconds and without you noticing. The lookup is a chain of questions. Your device asks a **resolver** (normally run by your internet provider, or a public one like `1.1.1.1` or `8.8.8.8`). If the resolver does not have the answer cached, it asks the root servers, then the servers for `.org`, then the servers responsible for `wikipedia.org`, which reply with the IP address. The answer is cached along the way, so the next person asking gets it instantly. DNS matters because names are stable and addresses are not. A company can move its site to another server or a CDN, change the DNS record, and nobody has to update anything. It is also a point of control: a DNS outage makes sites look down even when they are fine, and swapping resolvers is how some networks block or filter sites. Services like Amazon Route 53 sell DNS hosting. ### An analogy Your contacts app. You tap a name, not a number; if someone changes their number, you update one entry and keep tapping the same name. ### Keep in mind - DNS turns a domain name into an IP address before any connection is made. - Resolvers cache answers, so most lookups are answered close to you. - A DNS record change takes time to propagate because of those caches. - If DNS is down, the site may be fine but nobody can find it. ### Go deeper - [What is DNS? (Cloudflare)](https://www.cloudflare.com/learning/dns/what-is-dns/) - [DNS (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/DNS) - [Domain Name System (Wikipedia)](https://en.wikipedia.org/wiki/Domain_Name_System) --- ## Docker > The tool that made containers easy: package an app once, run it anywhere. - Level: Intermediate - Topics: Cloud, Software - URL: https://tidbit.legios.com.ar/docker/ - In Spanish: [Docker](https://tidbit.legios.com.ar/es/docker/) **Docker** is a set of tools for building and running containers, made by the company of the same name. Released in 2013, it took an old Linux idea and made it usable by any developer, which is why many people still say "a Docker" when they mean a container. The workflow has three steps. You write a **Dockerfile**, a short text recipe: start from this base system, copy my code, install these libraries, run this command. `docker build` turns the recipe into an **image**, a sealed package with everything the app needs. `docker run` starts a container from that image, and you can start as many identical copies as you want. Images are shared through a **registry**, a library of images; **Docker Hub** is the best-known public one, and clouds offer private ones like Amazon ECR. This is what makes Docker useful: the same image runs on a laptop, a test server and production, so "it works on my machine" stops being an excuse. The image format is now an open standard, so images built with Docker also run on Kubernetes and other tools that never touch Docker itself. ### An analogy A Dockerfile is a recipe, the image is the frozen meal packed from it, and each container is one portion heated up. Any microwave heats it the same way. ### Keep in mind - Dockerfile is the recipe, image is the package, container is the running copy. - Registries such as Docker Hub store and share images. - The image format is an open standard, so images run outside Docker too. - Docker is the tool; the container is the general idea. ### Go deeper - [What is Docker? (official docs)](https://docs.docker.com/get-started/docker-overview/) - [Docker on Wikipedia](https://en.wikipedia.org/wiki/Docker_(software)) --- ## Domain name > The readable name of a website, like wikipedia.org, that stands in for a numeric address. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/domain-name/ - In Spanish: [Nombre de dominio](https://tidbit.legios.com.ar/es/domain-name/) A **domain name** is the human-friendly name of a place on the internet, like `wikipedia.org` or `bbc.co.uk`. Computers find each other by IP address, a string of numbers that is hard to remember and can change. The domain name is the stable label people type, and DNS is the system that translates it into the current IP address. Domain names are read from right to left. The last part is the **top-level domain** (TLD): generic ones like `.com` and `.org`, or country ones like `.ar`, `.es` and `.mx`. To its left is the name an organization registers, and anything further left is a subdomain it creates freely, like `es.wikipedia.org` or `mail.example.com`. Nobody owns a domain forever: you rent it, usually by the year, through a **registrar**, a company accredited to sell names under each TLD. The whole naming system is coordinated by ICANN, and each TLD has a registry that keeps its official list. If a renewal is forgotten, the name can lapse and someone else can register it, which is why companies guard their domains closely. ### An analogy A contact name in your phone. You tap "Mom" instead of remembering her number, and if she changes numbers you only update the contact, not your habit. ### Keep in mind - A readable name that DNS translates into an IP address. - Read right to left: top-level domain, registered name, subdomains. - You rent a domain through a registrar, usually renewing yearly. - ICANN coordinates the system; each TLD has its own registry. ### Go deeper - [What is a domain name? (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Howto/Web_mechanics/What_is_a_domain_name) - [Domain name on Wikipedia](https://en.wikipedia.org/wiki/Domain_name) - [Root zone database of all TLDs (IANA)](https://www.iana.org/domains/root/db) --- ## Edge computing > Processing data close to where it is created, instead of in a distant data center. - Level: Intermediate - Topics: Cloud, Networks - URL: https://tidbit.legios.com.ar/edge-computing/ - In Spanish: [Edge computing](https://tidbit.legios.com.ar/es/edge-computing/) **Edge computing** means running computation near the people or devices that need it, at the "edge" of the network, rather than sending everything to a big data center hundreds or thousands of kilometers away. The edge can be a small server in a factory or a store, equipment in a mobile network tower, a city-level point of presence, or the device itself: a camera, a car or a phone. The main reasons are physics and volume. Every trip to a distant data center adds latency, and some decisions cannot wait: a factory robot that must stop when a person walks in front of it, or a car reacting to the road. Other times the problem is sheer data: a set of security cameras produces far too much video to upload, so it is analyzed on site and only the alerts go to the cloud. Keeping data local can also help with privacy rules and keep working when the internet connection drops. Edge and cloud work together; the edge does not replace the cloud. A common split is: react locally, then send summaries to the cloud for storage, analytics and training models. Cloud providers offer edge options, from functions that run in CDN locations close to users, to hardware they install in your own building. The trade-off is many small, scattered machines that are harder to update, secure and monitor than a few big ones. ### An analogy A first-aid kit in every office versus one big hospital across town. Small cuts are handled on the spot in seconds; only the serious cases make the long trip to the hospital. ### Keep in mind - Computes near the data source: in a store, a factory, a tower or on the device. - Cuts latency and the amount of data sent over the network. - Works with the cloud: react locally, send summaries centrally. - Many scattered machines are harder to update, secure and monitor. ### Go deeper - [What is edge computing? (AWS)](https://aws.amazon.com/what-is/edge-computing/) - [Edge computing on Wikipedia](https://en.wikipedia.org/wiki/Edge_computing) --- ## Email protocols > The shared rules, SMTP, IMAP and POP3, that carry an email from one inbox to another. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/email-protocols/ - In Spanish: [Protocolos de correo](https://tidbit.legios.com.ar/es/email-protocols/) Email works between any two providers because they all follow the same open rules. Sending and receiving are separate jobs with separate protocols. **SMTP** (Simple Mail Transfer Protocol) moves a message forward, from your app to your provider's **mail server** and from there to the recipient's. **IMAP** and **POP3** are how the recipient's app later picks it up. The trip goes like this. You hit send, and your app hands the message over SMTP to your provider. Your provider looks at the part after the `@`, say `b.com`, and asks DNS for that domain's **MX record**, which names the server that accepts mail for it. It connects to that server, again with SMTP, and delivers. The message now sits in the recipient's mailbox, on their provider's server, until they open their app. Usually this takes seconds; if the other server is down, yours keeps retrying for a while before bouncing it back to you. Reading is where IMAP and POP3 differ. **IMAP** leaves the mail on the server and keeps every device in sync: read it on the phone and it shows as read on the laptop, delete it anywhere and it is gone everywhere. **POP3** is older and simpler: it downloads messages to one device, often removing them from the server. Today IMAP, or a provider's own app, is the normal choice. SMTP comes from the early 1980s and originally trusted whatever 'From' address a sender claimed, which is why spam and phishing are so easy to send. Extra checks added later, called **SPF**, **DKIM** and **DMARC**, let a receiving server verify that a message really comes from the domain it claims. Connections between servers are usually encrypted with TLS, but that is not end-to-end encryption: each provider can read the mail it stores. ### An analogy SMTP is the postal service carrying letters between post offices; IMAP is a PO box you can check from any branch, and POP3 is taking the letters home and emptying the box. ### Keep in mind - SMTP sends and relays; IMAP and POP3 are for reading. - DNS MX records tell servers where to deliver mail for a domain. - IMAP keeps all devices in sync; POP3 downloads to one. - SPF, DKIM and DMARC help prove a message really comes from its domain. ### Go deeper - [Simple Mail Transfer Protocol on Wikipedia](https://en.wikipedia.org/wiki/Simple_Mail_Transfer_Protocol) - [Internet Message Access Protocol on Wikipedia](https://en.wikipedia.org/wiki/Internet_Message_Access_Protocol) - [POP, IMAP and SMTP settings for Outlook.com (Microsoft Support)](https://support.microsoft.com/en-us/outlook/pop-imap-and-smtp-settings-for-outlook-com) --- ## Embeddings > Turning words, images or documents into lists of numbers that capture meaning. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/embeddings/ - In Spanish: [Embeddings](https://tidbit.legios.com.ar/es/embeddings/) An **embedding** is a list of numbers (a vector) that represents a piece of content: a word, a sentence, a whole document, an image. A neural network produces it so that things with similar meaning get similar numbers. 'Dog' and 'puppy' end up close together; 'dog' and 'invoice' end up far apart. A typical embedding has a few hundred to a few thousand numbers. Computers cannot compare meanings, but they can compare numbers very fast. Once everything is an embedding, 'find documents like this one' becomes 'find the nearest points', a geometry problem. That powers semantic search (results that match the idea, not the exact words), recommendations, duplicate detection and clustering. Inside an LLM, every token becomes an embedding first; that is how text enters the math. Outside, embeddings are the backbone of RAG: documents are embedded once and stored in a **vector database**, the question is embedded at query time, and the closest chunks are retrieved. Example: a help desk that gets 'my screen went black' should find the article titled 'Display not turning on' even though no word matches. With embeddings, the two sentences sit close together, so the search works. ### An analogy A library map where every book has coordinates and books on similar topics are shelved near each other. To find related books you do not read them; you look at what is nearby. ### Keep in mind - An embedding is a vector; distance between vectors approximates distance in meaning. - The same model must embed both the documents and the query, or the numbers are not comparable. - Vector databases are built to find nearest neighbors among millions of embeddings. - Embeddings exist for text, images, audio and code, sometimes in one shared space. ### Go deeper - [3Blue1Brown: But what is a GPT? (embeddings explained visually)](https://www.3blue1brown.com/lessons/gpt) - [Word embedding on Wikipedia](https://en.wikipedia.org/wiki/Word_embedding) - [OpenAI: embeddings guide](https://platform.openai.com/docs/guides/embeddings) --- ## Emoji > Little pictures that are really text characters, drawn differently by each phone maker. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/emoji/ - In Spanish: [Emoji](https://tidbit.legios.com.ar/es/emoji/) An **emoji** is a small picture you can type, like 😀, 🍕 or ❤️. The word is Japanese: "e" means picture and "moji" means character, so it has nothing to do with "emotion". Emoji appeared on Japanese mobile phones in the late 1990s, and in 2010 they were added to **Unicode**, the standard that gives every character in every writing system its own number. That is the key idea: to a computer an emoji is text, not an image. 😀 is the character `U+1F600`, officially named "grinning face", in the same way that the letter A is `U+0041`. When you send it, your phone sends that number, a few bytes, and the receiving phone looks it up in its own emoji font and draws its own picture. Some emoji are combinations: a skin tone is a modifier added after the base character, and many family or profession emoji are several characters glued together with an invisible joiner. Because Unicode only sets the number, the name and a rough description, every company draws its own version. Apple, Google, Samsung and Microsoft each have their own designs, so the same emoji can look happier, sadder or simply different on each phone, and occasionally that changes its meaning. New emoji are proposed to the Unicode Consortium and a batch is approved most years; if your phone is not updated, a new emoji may show up as an empty box or as its separate pieces. ### An analogy Like a numbered list of songs shared by many bands. Everyone agrees which song is number 42, but each band plays it in its own style. ### Keep in mind - An emoji is a Unicode character, sent as a number, not as an image. - Each company draws its own design, so emoji look different across phones. - Skin tones and many family emoji are combinations of several characters. - Unicode approves new emoji; old phones may show them as boxes. ### Go deeper - [About emoji (Unicode Consortium)](https://home.unicode.org/emoji/about-emoji/) - [Unicode emoji resources](https://www.unicode.org/emoji/) - [Emoji on Wikipedia](https://en.wikipedia.org/wiki/Emoji) --- ## Encryption > Scrambling data so that only someone with the key can read it. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/encryption/ - In Spanish: [Cifrado](https://tidbit.legios.com.ar/es/encryption/) **Encryption** turns readable data (a message, a photo, a card number) into scrambled bytes that look like noise, using a **key**. Without the key the scrambled version is useless; with it, the original comes back exactly. The recipe for scrambling is called a cipher, and the good ones, like **AES**, are public and studied by everyone. The only secret is the key. It exists for one situation: someone else can see your bytes and you do not want them to understand them. The storage in your phone is encrypted, so a thief who pulls out the chip gets gibberish. Your messages travel encrypted, so the owner of the café Wi-Fi sees traffic but not words. HTTPS is exactly this, applied to web pages. There are two families. **Symmetric** encryption uses the same key to lock and unlock; it is fast, so it protects the bulk of the data. Public-key cryptography uses a pair of keys and solves a different problem: agreeing on a secret with someone you have never met. Real systems combine both. Encryption also hides content, not context: who you talk to, when, and how much is usually still visible. Practical takeaway: turn on disk encryption on your laptop (FileVault on a Mac, BitLocker or Device encryption on Windows). Phones with a lock screen already have it. And when an app says end-to-end encryption, it means not even the company can read what you send. ### An analogy Shuffling a deck of cards following a precise sequence written on a note. Anyone can look at the shuffled deck, but only someone with the note can put it back in order. ### Keep in mind - The algorithm is public; the key is the only secret. - Symmetric encryption uses one key; public-key encryption uses a pair. - It hides what you send, not usually who you send it to. - Turn on disk encryption on your laptop; most phones already have it. ### Go deeper - [What is encryption? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/ssl/what-is-encryption/) - [Encryption on Wikipedia](https://en.wikipedia.org/wiki/Encryption) --- ## End-to-end encryption > Only the sender and the recipient can read a message; the service in between cannot. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/end-to-end-encryption/ - In Spanish: [Cifrado de extremo a extremo](https://tidbit.legios.com.ar/es/end-to-end-encryption/) **End-to-end encryption** (E2EE) means a message is encrypted on your phone and decrypted only on the recipient's phone. The company that carries it (WhatsApp, Signal, Apple for iMessage) stores and forwards scrambled bytes it cannot read, because the keys exist only on the two devices. That is the difference with ordinary HTTPS, which protects the trip to the server but lets the server read everything once it arrives. It works with public-key cryptography: each device publishes a public key, and your phone uses the recipient's public key to lock the message so that only their private key opens it. The Signal protocol, used by Signal, WhatsApp and others, adds a trick: keys rotate with every message, so even a key stolen later cannot unlock old conversations. The limits matter. E2EE hides content, not metadata: the service still knows who talks to whom, when and how often. It does nothing if the phone itself is compromised, if someone screenshots the chat, or if your cloud backup of the chat is not encrypted the same way (check that setting; it is often the weak spot). And when a government asks for a 'backdoor' for investigators, the technical answer is that a door for one party is a door for anyone who finds it. Practically: your chat app probably already has it on; the setting worth checking is encrypted backups. For email, E2EE is rare and awkward, so treat email as a postcard. ### An analogy HTTPS is an armored truck between post offices: safe on the road, opened at each office. End-to-end encryption is a letter in a box only the recipient's key opens; the post office just carries the box. ### Keep in mind - Encrypted on one device, decrypted on the other; the service never has the key. - HTTPS protects the road to the server; E2EE also protects against the server. - It hides content, not who talks to whom or when. - Unencrypted cloud backups are the usual gap; check that setting. ### Go deeper - [What is end-to-end encryption? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/privacy/what-is-end-to-end-encryption/) - [End-to-end encryption on Wikipedia](https://en.wikipedia.org/wiki/End-to-end_encryption) --- ## Environment variables > Named settings a program reads from outside its code, like where its database lives. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/environment-variables/ - In Spanish: [Variables de entorno](https://tidbit.legios.com.ar/es/environment-variables/) **Environment variables** are name-and-value pairs that the operating system hands to every program when it starts, such as `DATABASE_URL=postgres://…` or `LOG_LEVEL=debug`. The program reads them instead of having those values written into its code. They are an old Unix idea; one you have probably used without noticing is `PATH`, which tells the computer where to look for commands. Their main job today is to let the **same code** run in different places. On a developer's laptop `DATABASE_URL` points to a small local database; in production it points to the real one. The code does not change, only the environment around it. This split between code and configuration is one of the principles of the well-known **Twelve-Factor App** guide for building cloud software. They are also the usual way to give programs **secrets**, like an API key or a database password, without putting them in the repository. During development many projects keep these in a file named `.env` that a library loads at startup; that file must be listed in `.gitignore` so it never reaches Git. In the cloud, services like AWS Lambda, containers and CI/CD pipelines all let you set environment variables, and for sensitive values they can be filled from a secrets manager. ### An analogy Like a recipe that says "bake at the temperature your oven needs" instead of a fixed number. The same recipe works in every kitchen; each kitchen supplies its own setting. ### Keep in mind - Configuration lives outside the code, so one build runs anywhere. - Each environment, like local, staging or production, sets its own values. - The common way to pass secrets without committing them. - Keep `.env` files out of Git. ### Go deeper - [Config, from The Twelve-Factor App](https://12factor.net/config) - [Environment variable on Wikipedia](https://en.wikipedia.org/wiki/Environment_variable) --- ## ETL > Copy data out of its sources, clean it up, load it where it can be analyzed. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/etl/ - In Spanish: [ETL](https://tidbit.legios.com.ar/es/etl/) **ETL** stands for extract, transform, load: the three steps of moving data from the systems where it is born to the place where it is analyzed. **Extract** reads it from the sources: the app's database, a CRM, a payments provider's API, spreadsheets. **Transform** cleans and reshapes it. **Load** writes the result into a data warehouse or a data lake. The transform step is where most of the work hides. Dates come in three formats, the same customer appears as 'ANA PÉREZ' in one system and 'Ana Perez' in another, prices are in different currencies, some rows are duplicated and others are missing a field. The pipeline fixes all of that the same way every time, so the numbers in the dashboard can be trusted and compared month to month. Pipelines run on a schedule (every night, every hour) or continuously, and a broken one shows up as a dashboard that quietly stops updating. Many modern teams flip the order to **ELT**: load the raw data first, then transform it inside the warehouse with SQL, which makes sense now that warehouses are powerful and storage is cheap. Tools like AWS Glue, Apache Airflow, dbt and Fivetran exist to build, schedule and monitor these pipelines. ### An analogy A restaurant's prep kitchen. Ingredients arrive from different suppliers (extract), get washed, peeled and cut the same way every morning (transform), and go into labeled containers on the line (load), so the cooks never deal with a muddy potato. ### Keep in mind - Extract from the sources, transform (clean, join, standardize), load into the destination. - Most of the effort, and most of the bugs, live in the transform step. - ELT loads the raw data first and transforms it inside the warehouse. - Pipelines run on a schedule or continuously, and need monitoring for when they break. ### Go deeper - [What is ETL? (AWS)](https://aws.amazon.com/what-is/etl/) - [Extract, transform, load on Wikipedia](https://en.wikipedia.org/wiki/Extract,_transform,_load) --- ## Fiber optic > Thin strands of glass that carry data as pulses of light, fast and over long distances. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/fiber-optic/ - In Spanish: [Fibra óptica](https://tidbit.legios.com.ar/es/fiber-optic/) A **fiber-optic** cable carries data as light instead of electricity. Inside is a strand of very pure glass, about as thin as a human hair, that guides light along its length by reflecting it off its own walls. A laser at one end switches on and off extremely fast, and a sensor at the other end reads those flashes back as ones and zeros. Light in glass has big advantages over electrical signals in copper. It loses very little strength with distance, so a signal can travel many kilometers before it needs a boost. Electrical interference from motors or storms does not disturb it. And one fiber can carry many separate signals at once, each on a different color of light. That is why the backbone of the internet, including the cables on the ocean floor that connect continents, is made of fiber. At home, **fiber to the home** (FTTH) means the fiber reaches your building or your door, instead of stopping at a cabinet down the street and finishing the trip over old copper phone or TV cable. You notice it in steadier speeds, faster uploads and lower latency. At your end, a small box called an ONT turns the light back into electrical signals; it is often combined with the modem and the router in a single device. ### An analogy Signaling with a flashlight down a long tube lined with mirrors. The light cannot escape, so the message arrives clear even very far away, and you can send several colors at once without them mixing. ### Keep in mind - Carries data as pulses of light through thin strands of glass. - Travels far with little loss and ignores electrical interference. - The internet's backbone, including undersea cables, runs on fiber. - FTTH brings the fiber all the way to your home for faster, steadier connections. ### Go deeper - [Optical fiber on Wikipedia](https://en.wikipedia.org/wiki/Optical_fiber) - [Fiber-optic communication on Wikipedia](https://en.wikipedia.org/wiki/Fiber-optic_communication) - [Fiber to the x on Wikipedia](https://en.wikipedia.org/wiki/Fiber_to_the_x) --- ## Fine-tuning > Extra training that adapts an existing AI model to a specific task or style. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/fine-tuning/ - In Spanish: [Fine-tuning](https://tidbit.legios.com.ar/es/fine-tuning/) **Fine-tuning** means taking a model that has already been trained on a huge amount of general data and training it a little more on a small, specific set of examples. Training a large model from scratch costs enormous amounts of data and computing; fine-tuning reuses all that general knowledge and only nudges the model's internal numbers (its parameters) toward what you need. Typical uses: making a model always answer in a particular format, write in a company's tone, classify support tickets into your own categories, or handle the jargon of a field like law or radiology. You prepare hundreds or thousands of example pairs ('given this input, the ideal output is this'), run the training, and get a new version of the model. Techniques such as **LoRA** make it cheaper by training only a small add-on instead of the whole model. Fine-tuning is also how chat assistants are made in the first place: a base model that only continues text is fine-tuned on conversations so it learns to follow instructions and reply helpfully. For everyday business problems, though, it is often not the first tool to reach for. If the goal is to give the model up-to-date facts or your own documents, retrieval (RAG) is usually easier and easier to update; if the goal is behavior, a good prompt sometimes gets you most of the way. Fine-tuning shines when you have many good examples of a task you repeat a lot. ### An analogy Like a doctor who finished general medical school and then does a short residency in dermatology. They do not relearn biology; they sharpen what they already know for one specialty. ### Keep in mind - Starts from a pretrained model instead of training from zero. - Needs a good set of examples of the exact task you want. - Best for style, format and repeated specialized tasks. - For fresh facts or private documents, RAG is often better. ### Go deeper - [Fine-tuning (deep learning) on Wikipedia](https://en.wikipedia.org/wiki/Fine-tuning_(deep_learning)) - [What is fine-tuning? (IBM)](https://www.ibm.com/think/topics/fine-tuning) - [Fine-tuning guide (OpenAI)](https://platform.openai.com/docs/guides/fine-tuning) --- ## FinOps > Understanding and controlling cloud spending, as a job shared by engineers and finance. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/finops/ - In Spanish: [FinOps](https://tidbit.legios.com.ar/es/finops/) **FinOps** (from finance and operations) is the practice of managing cloud costs as a shared responsibility. In a traditional data center, buying servers was a big, slow decision approved by finance. In the cloud, any engineer can start resources in seconds and the bill arrives at the end of the month, already spent. FinOps brings engineering, finance and business teams together so that spending decisions are made with the numbers in view. The FinOps Foundation, part of the Linux Foundation, describes it as a loop of three phases. **Inform**: make costs visible and assign them to teams and products, usually by **tagging** every resource with its owner. **Optimize**: cut waste, like idle servers left running, oversized machines (**rightsizing**), forgotten disks and snapshots, and commit to steady usage with discounts such as reserved instances or savings plans. **Operate**: set budgets and alerts, review costs regularly and make cost one more thing teams track, next to speed and reliability. The goal is not to spend as little as possible but to get the most value per dollar: spending more on a product that grows can be the right call. Clouds provide tools for it, such as AWS Cost Explorer and Budgets, Azure Cost Management and Google Cloud billing reports. The most common surprises are data leaving the cloud, logs and storage that grow quietly, and test environments nobody turned off. ### An analogy A shared household budget for the electricity bill. Everyone can turn on lights and heaters, so the bill only stays reasonable if everyone can see what uses the most and agrees on turning things off when they leave the room. ### Keep in mind - Makes cloud spending a shared responsibility of engineering, finance and business. - Three phases: inform (visibility), optimize (cut waste), operate (budgets and habits). - Common levers: tagging, rightsizing, turning off idle resources, commitment discounts. - Aims for value per dollar, not the lowest possible bill. ### Go deeper - [What is FinOps? (FinOps Foundation)](https://www.finops.org/introduction/what-is-finops/) - [What is FinOps? (Microsoft Learn)](https://learn.microsoft.com/en-us/cloud-computing/finops/overview) - [Cloud cost management (AWS)](https://aws.amazon.com/aws-cost-management/) --- ## Firewall > A gatekeeper that decides which network traffic gets in or out. - Level: Basic - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/firewall/ - In Spanish: [Firewall](https://tidbit.legios.com.ar/es/firewall/) A **firewall** sits between a network (or a single computer) and the outside world and checks every connection against a list of rules: where it comes from, where it goes, which port it uses, which direction. Whatever matches an allow rule passes; the rest is dropped. The usual default is simple: anything that starts from inside can go out, nothing that starts from outside can come in unless a rule says so. You already have several. The router at home blocks unsolicited connections to your laptop and TV. Windows and macOS ship with a firewall turned on, so a random machine on the café Wi-Fi cannot open a connection to your file sharing. In the cloud, a web server might allow exactly two ports, `443` for HTTPS and `22` for the admin, and nothing else. Specialized ones, like a **WAF** (web application firewall), read the content of web requests and block known attack patterns. What it does not do is just as important. A firewall does not stop you from downloading Malware, does not detect phishing, and cannot see inside traffic that is encrypted. It is a border control, not a bodyguard: one layer among several. Practical takeaway: leave the firewall on your laptop enabled, and do not open ports on your home router ('port forwarding') unless you know exactly which device and why. Every open port is a door that you are now responsible for. ### An analogy A building receptionist with a list. Residents walk out whenever they want; visitors only get in if someone registered them, and only to the floor they were registered for. ### Keep in mind - A firewall allows or drops connections based on rules: address, port, direction. - The default that matters: outbound is free, inbound is blocked unless allowed. - Your router, your laptop and every cloud server already have one. - It does not stop malware you download or phishing you click; it is one layer. ### Go deeper - [What is a firewall? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/security/what-is-a-firewall/) - [Firewall on Wikipedia](https://en.wikipedia.org/wiki/Firewall_(computing)) --- ## Firmware > Software built into a device to run its hardware, from routers to washing machines. - Level: Basic - Topics: Hardware, Software - URL: https://tidbit.legios.com.ar/firmware/ - In Spanish: [Firmware](https://tidbit.legios.com.ar/es/firmware/) **Firmware** is the software that lives inside a device and makes its hardware work. It is stored on a memory chip in the device itself, not on a disk you can browse, and it starts running the moment the device gets power. Your router, your TV, your printer, a smartwatch, a car's dashboard and even a computer's keyboard and SSD each have their own firmware. It sits between the hardware and everything else. In a simple gadget, like a thermostat or a microwave, the firmware is the whole program: it reads the buttons, drives the display and switches the heater on and off. In a computer or a phone, the firmware does the low-level work (waking up the chips, talking to the disk, managing the battery) and then hands over to the operating system. The BIOS and UEFI of a PC are firmware. The name says where it sits: firmer than software you install, but not fixed forever like the hardware. Manufacturers release **firmware updates** to fix bugs, add features or close security holes, which is why a router or a phone sometimes asks to update and restart. Old devices that no longer get updates are a real risk, because their known vulnerabilities never get fixed. And an update should never be interrupted: if the power goes out halfway, the device can be left unable to start. ### An analogy The metal drum inside a music box: it comes with the device and decides what it does every time you wind it. Unlike the drum, firmware can be swapped for a newer version with an update. ### Keep in mind - Stored on a chip inside the device and runs as soon as it powers on. - Almost every electronic device has it, from routers to keyboards. - Updates fix bugs and security holes; devices that stop getting them stay exposed. - Never cut the power in the middle of a firmware update. ### Go deeper - [Firmware on Wikipedia](https://en.wikipedia.org/wiki/Firmware) - [Firmware (NIST glossary)](https://csrc.nist.gov/glossary/term/firmware) --- ## Framework > A ready-made skeleton for an app: it sets the structure and calls your code. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/framework/ - In Spanish: [Framework](https://tidbit.legios.com.ar/es/framework/) A **framework** is a large body of ready-made code that gives an application its basic structure, so programmers only write the parts that are unique to their product. A web framework, for example, already knows how to receive requests, read forms, talk to a database and send pages back; you fill in what your pages show and what your rules are. The key difference with a software library is who is in charge. You call a library when you need it, like grabbing a tool from a drawer. A framework calls you: it runs the main loop and invokes your code at the right moments, such as "when someone opens this page, run this function". Engineers call this **inversion of control**. Well-known examples include Angular and Vue for building user interfaces in the browser, Django and Ruby on Rails for web backends, and Spring for business software in Java. Choosing one is a big decision, because it shapes how the whole project is organized and is hard to swap later. In exchange, teams get a lot done fast and follow conventions that new members already know. ### An analogy The frame of a prefabricated house: walls, roof and plumbing paths are already in place. You choose the paint, the kitchen and the furniture, but you cannot easily move a load-bearing wall. ### Keep in mind - Provides the structure of an app; you fill in the specific parts. - It calls your code, unlike a library, which you call. - Examples: Angular, Vue, Django, Ruby on Rails, Spring. - Saves a lot of work, but is hard to replace once chosen. ### Go deeper - [Introduction to client-side frameworks (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/Frameworks_libraries/Introduction) - [Software framework on Wikipedia](https://en.wikipedia.org/wiki/Software_framework) --- ## Frontend and backend > The part of an app you see and touch, and the part that works behind it. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/frontend-and-backend/ - In Spanish: [Frontend y backend](https://tidbit.legios.com.ar/es/frontend-and-backend/) Almost every app is split in two. The **frontend** is what runs on your device: the screens, buttons and forms of a website or mobile app. The **backend** runs on a server somewhere else: it checks your password, reads and writes the database, charges the card and sends the data the frontend displays. The two talk through an API, usually over HTTPS. Frontends are built with HTML, CSS and JavaScript in the browser (often with a framework like React), or with Swift and Kotlin on phones. Backends are written in Python, Java, Go, Node.js and many others, and run on a cloud machine, in a container or as a serverless function. When you tap "buy", the frontend sends a request; the backend validates it, updates stock and orders, and replies "done"; the frontend shows the confirmation. The split matters because each side has different problems. The frontend worries about screen sizes, speed on old phones and accessibility; the backend about security, data integrity and handling thousands of users at once. Anything that must be trusted (prices, permissions) lives in the backend, because anything in the frontend can be read and modified by the user. A "full-stack" developer works on both. ### An analogy A restaurant. The dining room, the menu and the waiters are the frontend; the kitchen, the pantry and the cash register are the backend. You only ever see the dining room, but the kitchen decides what can actually be served. ### Keep in mind - Frontend: runs on the user's device, built with HTML, CSS, JavaScript or native mobile code. - Backend: runs on servers, owns the data and the business rules. - They communicate through APIs, and the backend never trusts the frontend blindly. - "Full stack" means working on both sides. ### Go deeper - [How the web works (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Getting_started/Web_standards/How_the_web_works) - [Frontend and backend (Wikipedia)](https://en.wikipedia.org/wiki/Frontend_and_backend) --- ## FTP and SFTP > Two ways to move files to and from a server; only one of them keeps them private. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/ftp-and-sftp/ - In Spanish: [FTP y SFTP](https://tidbit.legios.com.ar/es/ftp-and-sftp/) **FTP** (File Transfer Protocol) is one of the oldest services on the internet, older than the web itself. You connect to an **FTP server** with a username and password, see its folders, and upload or download files. For decades it was how people published websites to their web hosting, how printers dropped scans into shared folders and how companies swapped files with banks and suppliers. Its big flaw is that it was designed before anyone worried about eavesdroppers. FTP sends everything as plain text, including the password, so anyone on the path, such as someone on the same café Wi-Fi, can read it. It also opens separate connections for commands (port 21) and for each file, which makes it awkward to get through firewalls and NAT. **SFTP** is the modern replacement, and despite the name it is not FTP with security added: it is a different protocol that runs inside SSH, over a single encrypted connection, usually on port 22. Logins can use SSH keys instead of passwords, and the files travel encrypted. There is also **FTPS**, which is the old FTP wrapped in TLS, the same encryption HTTPS uses; it works, but keeps FTP's multi-connection complications. In practice: if a provider or a partner asks you to send files 'by FTP', ask for SFTP. Many businesses still exchange daily files this way, and cloud services can offer an SFTP endpoint that stores everything straight into object storage like Amazon S3. For sharing with people, a link from a cloud drive is usually simpler. ### An analogy FTP is mailing documents on postcards: they arrive, but every mail carrier on the way can read them. SFTP is the same documents in a locked courier bag that only the recipient can open. ### Keep in mind - FTP sends passwords and files as plain text. - SFTP runs inside SSH: one encrypted connection, usually on port 22. - FTPS is FTP wrapped in TLS; similar goal, different protocol. - If someone asks for FTP, ask for SFTP instead. ### Go deeper - [File Transfer Protocol on Wikipedia](https://en.wikipedia.org/wiki/File_Transfer_Protocol) - [SSH File Transfer Protocol on Wikipedia](https://en.wikipedia.org/wiki/SSH_File_Transfer_Protocol) - [RFC 959, the FTP standard](https://www.rfc-editor.org/rfc/rfc959) --- ## Gemini > Google's family of AI models, and the assistant app that carries the same name. - Level: Basic - Topics: AI, Tech companies - URL: https://tidbit.legios.com.ar/gemini/ - In Spanish: [Gemini](https://tidbit.legios.com.ar/es/gemini/) **Gemini** is the name Google uses for two related things: a family of large language models built by Google DeepMind, its AI lab, and the chat assistant people use to talk to them. Google announced the models in December 2023. A few months later, in February 2024, it renamed its earlier chatbot, Bard, to Gemini, so the app and the models now share one name. The models were designed from the start to be **multimodal**: the same model can take in text, photos, audio, video and code, rather than bolting separate systems together. They come in sizes for different jobs, with names like Pro for the most capable, Flash for fast and cheap, and Nano, small enough to run on a phone. Google also publishes **Gemma**, a smaller sibling whose weights anyone can download. What makes Gemini distinct is where it lives. Besides the Gemini app on the web and phones, it is built into Google's own products: Android phones, Gmail, Docs and the rest of Google Workspace, Chrome and Search. Developers reach it through the Gemini API and through Google Cloud. Like any language model it can be wrong while sounding confident, so answers that matter deserve a second look. ### An analogy Gemini is less a single shop than a franchise: the same kitchen supplies the stand-alone restaurant (the app) and the food counters inside Google's other buildings, from Gmail to your phone. ### Keep in mind - Made by Google DeepMind; the models were announced in December 2023. - The Bard chatbot was renamed Gemini in February 2024. - Multimodal by design: text, images, audio and video in one model. - Built into Android and Google Workspace, and available as an API. ### Go deeper - [Google Gemini on Wikipedia](https://en.wikipedia.org/wiki/Google_Gemini) - [Gemini models (Google DeepMind)](https://deepmind.google/models/gemini/) --- ## Generative AI > AI that creates new text, images, audio or code instead of just sorting things. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/generative-ai/ - In Spanish: [IA generativa](https://tidbit.legios.com.ar/es/generative-ai/) **Generative AI** is the kind of AI that produces new content: a paragraph, an email, a picture, a song, a voice, a video, a piece of code. Older AI was mostly about deciding things, like whether a photo shows a dog or whether a payment looks like fraud. Generative AI goes the other way: you describe what you want, usually in plain words, and it makes something that did not exist before. It works by learning the patterns of an enormous amount of existing material. A text model like the ones behind ChatGPT, Claude or Gemini has read so much writing that it can predict, word after word, what a plausible answer looks like. An image model has seen so many pictures with captions that it can turn 'a cat in a raincoat, watercolor' into a new image. What comes out is new, but built from patterns in what went in. It became a mainstream topic in late 2022, when ChatGPT and image generators reached the general public. It is useful for drafting, summarizing, translating, brainstorming and coding. It also brings real problems: it can state false things with confidence, imitate real people's voices and faces, repeat the biases of its training data, and raise open questions about the copyright of the material it learned from. ### An analogy Like a musician who has listened to thousands of songs and can improvise a new one in any style you ask for. The tune is new, but you can hear where the influences come from, and sometimes it hits a wrong note with total confidence. ### Keep in mind - Creates new content from a description, called a prompt. - Learns from huge amounts of existing text, images or sound. - Output sounds or looks convincing even when it is wrong. - Raises open questions about copyright, deepfakes and bias. ### Go deeper - [Generative artificial intelligence on Wikipedia](https://en.wikipedia.org/wiki/Generative_artificial_intelligence) - [What is generative AI? (AWS)](https://aws.amazon.com/what-is/generative-ai/) --- ## Git > A time machine for code: every change saved, by whom and why. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/git/ - In Spanish: [Git](https://tidbit.legios.com.ar/es/git/) **Git** is a program that records the history of a folder of files. Each time you save a set of changes you make a **commit**: a snapshot with a message ("fix login bug"), an author and a date. You can go back to any snapshot, compare two of them, or see who changed a given line and when. Linus Torvalds wrote it in 2005 to manage the Linux kernel, and today it is used for almost all software. Its second trick is **branches**. A branch is a parallel line of history: you open one to try a new feature, commit freely without touching the main version, and when it works you **merge** it back. Several people can work on different branches of the same project at once, and Git combines the results, flagging the rare lines where two people changed the same thing (a "conflict"). Git itself works on your machine. **GitHub**, GitLab and Bitbucket are websites that host a copy of the repository so a team can share it: `git push` sends your commits there and `git pull` brings down everyone else's. A **pull request** on GitHub is a proposal to merge a branch, with a review conversation attached. Most open source software lives this way. ### An analogy Track changes in a document, but for a whole folder, with the option of trying an edit in a parallel copy and folding it back in once it is ready. ### Keep in mind - A commit is a snapshot plus a message; the history is a chain of them. - Branches let several people work in parallel; merging joins the work. - Git is the tool; GitHub is one place to host the repository. - `clone`, `commit`, `push`, `pull` and `merge` are the five commands you will hear most. ### Go deeper - [Pro Git book (official)](https://git-scm.com/book/en/v2) - [Git documentation](https://git-scm.com/doc) - [Git (Wikipedia)](https://en.wikipedia.org/wiki/Git) --- ## GitHub > A website where teams store their code with Git and work on it together. - Level: Basic - Topics: Software, Tech companies - URL: https://tidbit.legios.com.ar/github/ - In Spanish: [GitHub](https://tidbit.legios.com.ar/es/github/) **GitHub** is an online service for hosting code. Each project lives in a **repository**: a folder of files plus its full history, tracked with Git. Developers keep a copy on their laptop, make changes, and **push** them to GitHub so the rest of the team can **pull** them. It started in 2008 and has been owned by Microsoft since 2018. Git is the tool; GitHub is a place built around it. On top of storing code, it adds what a team needs to collaborate: **pull requests** to propose a change and discuss it before it is merged, **issues** to track bugs and ideas, **GitHub Actions** to run tests and deployments automatically, and pages to document the project. It matters because it became the default meeting point for software. Most well-known open source projects live there, companies use private repositories for their own code, and a developer's public GitHub profile often works as a portfolio. Alternatives like GitLab and Bitbucket do similar things, and since everything is plain Git, a project can move between them. ### An analogy Git is like keeping a detailed diary of every edit to a document. GitHub is the shared office where everyone's diaries meet, with a meeting room for proposing changes and a bulletin board for problems. ### Keep in mind - Git tracks changes; GitHub hosts Git repositories online and adds collaboration tools. - A pull request is a proposed change that others review before it joins the main code. - Repositories can be public, open to anyone, or private to a team. - Owned by Microsoft since 2018; GitLab and Bitbucket are common alternatives. ### Go deeper - [About GitHub and Git (official docs)](https://docs.github.com/en/get-started/start-your-journey/about-github-and-git) - [GitHub on Wikipedia](https://en.wikipedia.org/wiki/GitHub) --- ## GitHub Copilot > An AI assistant inside the code editor that suggests and writes code. - Level: Basic - Topics: AI, Software - URL: https://tidbit.legios.com.ar/github-copilot/ - In Spanish: [GitHub Copilot](https://tidbit.legios.com.ar/es/github-copilot/) **GitHub Copilot** is an AI coding assistant made by GitHub, launched in 2021. It runs inside the editors developers already use, such as Visual Studio Code, and is powered by large language models that were trained on huge amounts of code and text. It works in a few ways. As you type, it suggests the next lines in grey, and you press Tab to accept them. In a chat panel you can ask things like "what does this function do?" or "write a test for this". In its more autonomous modes it can take a task, edit several files, run commands and open a pull request on GitHub for a person to review. It matters because it changed how a lot of everyday programming is done: less typing of repetitive code, more reading and checking. That last part is key. Copilot can be confidently wrong, invent functions that do not exist, or suggest code with security problems, so its output needs the same review as a colleague's. Similar tools include Cursor, Claude Code and others; they differ in how much they do on their own. ### An analogy Like the autocomplete on your phone keyboard, but for code and much more ambitious: sometimes it finishes your word, sometimes it drafts the whole paragraph. You still decide what gets sent. ### Keep in mind - Suggests code as you type and answers questions in a chat inside the editor. - Built on large language models; it predicts likely code, it does not understand your business. - Its suggestions can be wrong or insecure, so they need review. - It is a paid service with plans for individuals, companies and a limited free tier. ### Go deeper - [What is GitHub Copilot? (official docs)](https://docs.github.com/en/copilot/get-started/what-is-github-copilot) - [GitHub Copilot on Wikipedia](https://en.wikipedia.org/wiki/GitHub_Copilot) --- ## Google > The search company that gives most products away and makes its money on ads. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/google/ - In Spanish: [Google](https://tidbit.legios.com.ar/es/google/) **Google** started in 1998 as a search engine built by two Stanford students, Larry Page and Sergey Brin. Their idea was to rank web pages by how many other pages linked to them, which gave much better results than the search engines of the time. Since 2015 Google is part of **Alphabet**, a parent company that also holds other bets such as Waymo, the self-driving car business. Most people never pay Google directly, and yet it is one of the largest companies in the world. The trick is **advertising**: Search, YouTube, Maps, Gmail, Chrome and Android are free, billions of people use them every day, and businesses pay to show ads next to what those people search for or watch. Because a search says exactly what someone wants right now ('plumber near me'), those ads are unusually valuable, and ads still bring in most of the company's revenue. On top of that it sells things directly. **Google Cloud** rents computing, storage and data tools to companies, competing with AWS and Microsoft Azure. Google DeepMind, its AI lab, builds the **Gemini** models, and Google designs its own AI chips, called TPUs. It also sells Pixel phones and other devices, and takes a share of the sales in the Play Store, the app store of Android. ### An analogy A free TV channel. The shows (search, video, maps) cost you nothing because the channel sells the commercial breaks to advertisers, and the more people watch, the more each break is worth. ### Keep in mind - Founded in 1998 by Larry Page and Sergey Brin; part of Alphabet since 2015. - Most of its revenue comes from ads shown in Search, YouTube and partner sites. - Android, Chrome, Gmail and Maps are free and keep people inside its ecosystem. - It also sells Google Cloud, Gemini models and Pixel devices. ### Go deeper - [About Google](https://about.google/) - [Google products](https://about.google/products/) - [Google on Wikipedia](https://en.wikipedia.org/wiki/Google) --- ## Google Cloud > Google's cloud business: computing, storage, data and AI services rented by the hour. - Level: Basic - Topics: Cloud, Tech companies - URL: https://tidbit.legios.com.ar/google-cloud/ - In Spanish: [Google Cloud](https://tidbit.legios.com.ar/es/google-cloud/) **Google Cloud**, often called **GCP** (Google Cloud Platform), is the part of Google that rents computing infrastructure and services to other companies, running in the same kind of global data centers that power Google Search and YouTube. It competes with AWS and Microsoft Azure, and is usually counted as the third-largest of the big public clouds. It offers the usual building blocks: virtual machines (Compute Engine), object storage (Cloud Storage), databases, networking and serverless options like Cloud Run. Its best-known strengths come from technology Google built for itself. **BigQuery** is a serverless data warehouse that can scan huge tables with plain SQL. **GKE** runs Kubernetes, which started at Google before being released as open source. **Vertex AI** and Google's own AI chips, called **TPUs**, are used to train and serve machine learning models, including Google's Gemini models. Like other clouds, it is organized into regions and zones around the world, and you pay for what you use, with discounts for steady or committed usage. In Google's financial reports, the Google Cloud segment also includes Google Workspace (Gmail, Docs and Drive for businesses), so the name covers both renting infrastructure and selling office software to companies. ### An analogy Renting space and tools in the same workshop where a famous manufacturer builds its own products. You get access to the machines they designed for themselves, and you pay by the hour for what you use. ### Keep in mind - Google's public cloud, also called GCP, competing with AWS and Azure. - Known for BigQuery, Kubernetes (GKE), and AI services with its own TPU chips. - Pay-as-you-go pricing across regions and zones worldwide. - As a business segment, it also includes Google Workspace. ### Go deeper - [Google Cloud overview (official docs)](https://cloud.google.com/docs/overview) - [Google Cloud Platform on Wikipedia](https://en.wikipedia.org/wiki/Google_Cloud_Platform) --- ## GPU > The chip that does thousands of simple calculations at once. - Level: Basic - Topics: Hardware, AI - URL: https://tidbit.legios.com.ar/gpu/ - In Spanish: [GPU](https://tidbit.legios.com.ar/es/gpu/) A **GPU** (Graphics Processing Unit) was invented to draw video game frames: millions of pixels that all need the same small calculation at the same time. Instead of a few powerful cores like a CPU, a GPU has thousands of small ones that work in parallel. It turns out that training and running AI models is the same kind of problem. A neural network is mostly huge multiplications of number grids (matrices), and a GPU can do those much faster than a CPU. That is why NVIDIA became one of the most valuable companies in the world, why data centers are full of GPUs, and why a modern AI model is measured in how many GPUs it took to train. ### An analogy A CPU is a brilliant chef who can cook any dish. A GPU is a thousand line cooks who each chop one onion. For a banquet of onions, the line cooks win. ### Keep in mind - CPU: few fast cores, good at varied tasks. GPU: many simple cores, good at repeated math. - AI training and inference run mostly on GPUs or similar chips (TPUs). - GPU memory (VRAM) decides how big a model fits on one card. ### Go deeper - [GPU on Wikipedia](https://en.wikipedia.org/wiki/Graphics_processing_unit) - [NVIDIA: what is accelerated computing](https://blogs.nvidia.com/blog/what-is-accelerated-computing/) --- ## GraphQL > An API language where the app asks for exactly the fields it needs, in one request. - Level: Advanced - Topics: Software - URL: https://tidbit.legios.com.ar/graphql/ - In Spanish: [GraphQL](https://tidbit.legios.com.ar/es/graphql/) **GraphQL** is a way of building APIs in which the client writes a **query** describing the exact data it wants, and the server answers with JSON in that same shape. It was created at Facebook in 2012 to feed its mobile apps, released publicly in 2015, and is now maintained by the GraphQL Foundation, part of the Linux Foundation. Compare it with a REST API. To show a profile screen, a REST app might call `/users/7`, then `/users/7/orders`, then `/orders/91/items`, and receive many fields it never shows. With GraphQL it sends one request to a single endpoint, usually `/graphql`, saying "user 7: name, and the last two orders with their totals", and gets just that. This avoids **over-fetching** (too much data) and **under-fetching** (too many round trips), which matters on slow mobile connections. Behind the endpoint sits a **schema**: a typed description of every object and field the API offers, which tools use to autocomplete and validate queries before they run. Each field is filled in by a small function called a **resolver**, which can read from a database, a REST service or anything else. The flexibility has costs: caching is harder than with plain URLs, and a careless query can ask for so much nested data that it overloads the server, so production APIs limit query depth and cost. ### An analogy REST is a set menu: each dish comes as the kitchen decided. GraphQL is ordering à la carte from a printed list of every ingredient: you get exactly what you asked for, but the kitchen has to watch out for absurd orders. ### Keep in mind - One endpoint; the client's query decides the shape of the answer. - A typed schema lists everything that can be asked for. - Avoids fetching too much or making many round trips. - Harder to cache and needs limits against very expensive queries. ### Go deeper - [Introduction to GraphQL (official site)](https://graphql.org/learn/) - [What is GraphQL? (AWS)](https://aws.amazon.com/graphql/) - [GraphQL on Wikipedia](https://en.wikipedia.org/wiki/GraphQL) --- ## Hacker > Someone who takes systems apart to see how they work, for better or for worse. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/hacker/ - In Spanish: [Hacker](https://tidbit.legios.com.ar/es/hacker/) The word **hacker** has two lives. It was born in the late 1950s and 1960s among students at MIT, where a "hack" was a clever, playful solution to a technical problem. In that original sense, a hacker is a curious tinkerer who loves understanding how things work and bending them to do new things. Much of today's free and open source software came out of that culture. In the 1980s the press started using "hacker" for people who broke into computers, and that meaning stuck. Inside the field, people tell them apart by the color of the hat, borrowed from old westerns. A **white hat** or **ethical hacker** looks for weaknesses with permission and reports them so they get fixed. A **black hat** breaks in without permission, to steal, extort or cause damage. A **grey hat** is in between: no permission, but no harm intended, which is still often illegal. White hats are a real profession. Companies hire **penetration testers** to attack their own systems before criminals do, and many run **bug bounty** programs that pay outsiders who report a vulnerability responsibly: privately, to the owner, giving time to fix it before telling anyone. So when you read "hacker", look at the context: it can be a compliment, a job title or an accusation. ### An analogy A locksmith. One studies locks to make better ones and opens doors when the owner asks; another uses the same skill to break into houses at night. The knowledge is the same; permission and intent make the difference. ### Keep in mind - Originally a hacker was a curious tinkerer; the meaning of "attacker" came later, through the press. - White hats work with permission, black hats without it, grey hats in between. - Penetration testing and bug bounties are legal, paid ways of hacking. - Responsible disclosure means telling the owner privately and giving time to fix the flaw. ### Go deeper - [Hacker on Wikipedia](https://en.wikipedia.org/wiki/Hacker) - [White hat (computer security) on Wikipedia](https://en.wikipedia.org/wiki/White_hat_(computer_security)) --- ## Hallucination > When an AI model states something false with total confidence. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/hallucination/ - In Spanish: [Alucinación](https://tidbit.legios.com.ar/es/hallucination/) A **hallucination** is an answer from an AI model that sounds right and is wrong: a court case that never existed, a quote nobody said, a library function that does not exist, a wrong date delivered in a perfect sentence. The model is not lying; it has no idea it is wrong. It produced the most plausible-sounding text, and plausible is not the same as true. It happens because of how an LLM works. It writes by predicting likely next tokens, based on patterns in its training data. When the fact was well represented in that data, the likely continuation is usually correct. When it was not (an obscure topic, a private detail, something after its training cutoff), the likely continuation is still fluent, just invented. There is no built-in step that checks against reality. Hallucinations are the main reason you cannot use a chatbot as an encyclopedia. Lawyers have been sanctioned for filing briefs with made-up citations. The fixes are partial: giving the model the real documents (RAG), letting it search and cite its sources, asking it to say 'I don't know', and newer models trained to decline more often when unsure. None of them bring the rate to zero. The practical rule: use models for drafting, summarizing, explaining and brainstorming, and verify any specific fact, number, name or citation before you rely on it. ### An analogy A confident student in an oral exam who, instead of saying 'I don't remember', improvises an answer in the same assured tone as the ones they actually know. From the tone alone you cannot tell which is which. ### Keep in mind - A hallucination is fluent, confident and false; the model cannot tell the difference. - Most common with obscure facts, exact citations, numbers and recent events. - Grounding the model in real documents (RAG) and letting it cite sources lowers the rate, but does not eliminate it. - Verify names, numbers and sources before you act on them. ### Go deeper - [Hallucination (artificial intelligence) on Wikipedia](https://en.wikipedia.org/wiki/Hallucination_(artificial_intelligence)) - [Anthropic: reduce hallucinations](https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) --- ## Hard disk drive > Storage that keeps data on spinning magnetic platters: slower than an SSD, but cheap. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/hard-disk-drive/ - In Spanish: [Disco rígido](https://tidbit.legios.com.ar/es/hard-disk-drive/) A **hard disk drive** (HDD) stores data magnetically on one or more round metal platters stacked inside a sealed box. The platters spin at several thousand revolutions per minute, and an arm sweeps a tiny **read/write head** across them, floating a hair's breadth above the surface. Writing flips the magnetic direction of microscopic spots; reading senses them. The data stays there with the power off. Because it has moving parts, a hard disk is slow to reach data: before reading, the arm has to swing to the right track and wait for the right spot to spin under it, which takes several milliseconds. An SSD has no moving parts and answers in a fraction of a millisecond, which is why SSDs took over laptops and phones, and why a computer with a hard disk feels sluggish when booting or opening apps. Hard disks are also noisier and more fragile if dropped while running. Their advantage is cost. Per terabyte, a hard disk is still much cheaper than an SSD, and capacities are large. So they are far from gone: data centers fill racks with them for backups, archives and the bulk of cloud storage, and home users put them in external drives and network storage boxes, anywhere lots of space matters more than speed. ### An analogy A record player with a very fast needle. To play a song, the arm has to travel to the right groove and wait for it to come around; a pile of records is cheap, but jumping between songs takes time. ### Keep in mind - Data lives on spinning magnetic platters read by a moving head. - Several milliseconds to reach data, far slower than an SSD. - Much cheaper per terabyte, so it remains the choice for bulk storage. - Keeps data without power, but is sensitive to shocks while spinning. ### Go deeper - [Hard disk drive on Wikipedia](https://en.wikipedia.org/wiki/Hard_disk_drive) - [Solid-state drive on Wikipedia](https://en.wikipedia.org/wiki/Solid-state_drive) --- ## Hashing > A one-way fingerprint of data: easy to compute, not meant to be reversed. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/hashing/ - In Spanish: [Hashing](https://tidbit.legios.com.ar/es/hashing/) A **hash function** takes any input (a word, a file, a whole disk) and produces a short, fixed-size string called a **hash**: `SHA-256` always gives 64 hexadecimal characters. The same input always gives the same hash. Change one letter and the hash is completely different. And there is no way to run it backwards: from the hash you cannot rebuild the input, and finding two inputs with the same hash is, for a good function, impractical. That makes hashes useful as fingerprints. A download page shows the file's hash so you can check your copy is intact. Git names every commit by a hash of its content. Blockchains chain blocks by hashing the previous one. And websites store a hash of your password instead of the password: when you log in, they hash what you typed and compare. If the database leaks, the attacker gets fingerprints, not passwords. The catch with passwords is that attackers can hash billions of guesses per second and compare. So passwords are hashed with slow, purpose-built functions (`bcrypt`, `scrypt`, `Argon2`) and a random **salt** per user, so that the same password gives different hashes for different people and precomputed tables are useless. A fast hash like `MD5` for passwords is a classic mistake that keeps appearing in breach reports. Hashing is not encryption. Encryption is reversible with the key; a hash has no key and no way back. If a site can email you your old password, it was not hashing it. ### An analogy A fingerprint. It identifies a person reliably and fits on a small card, but you cannot rebuild the person from the print. ### Keep in mind - Same input, same hash; a tiny change gives a completely different hash. - It is one way: the hash does not reveal the input. - Passwords should be stored as slow, salted hashes (Argon2, bcrypt), never as plain text. - Hashing verifies and identifies; encryption hides. They are different tools. ### Go deeper - [Cryptographic hash function on Wikipedia](https://en.wikipedia.org/wiki/Cryptographic_hash_function) - [OWASP Password Storage Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Password_Storage_Cheat_Sheet.html) --- ## High availability > Designing a system so it keeps working when one of its parts fails. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/high-availability/ - In Spanish: [Alta disponibilidad](https://tidbit.legios.com.ar/es/high-availability/) **High availability** (HA) is the goal of keeping a service up nearly all the time, even when individual parts break. Servers crash, disks fail, networks drop and whole data centers lose power; a highly available system is built so that none of those events, on its own, takes the service down. It is usually measured as a percentage of uptime, like 99.9% or 99.99%. The core idea is **redundancy**: no **single point of failure**. Instead of one server, run several behind a load balancer. Instead of one data center, spread them across two or three **availability zones**, separate buildings in the same cloud region. Keep a standby copy of the database that takes over automatically if the main one fails, a switch called **failover**. Health checks constantly test each piece and route traffic away from anything that stops answering. High availability is about everyday failures and keeps the service running through them, often without anyone noticing. Disaster recovery is the plan for the rare big event that knocks out a whole region. Every extra "nine" of uptime costs more in duplicated machines and engineering effort, so teams pick a target that matches how much an outage really hurts, and write it down in an SLA. ### An analogy A plane with two engines. It can fly on one if the other fails, so a single engine problem is an incident, not a crash. You pay for the second engine on every flight so that you never need to think about it. ### Keep in mind - Keeps a service up when individual servers, disks or data centers fail. - Built on redundancy: no single point of failure, spread across availability zones. - Health checks and automatic failover move traffic away from broken parts. - Each extra nine of uptime costs more; match the target to the real impact. ### Go deeper - [Reliability pillar, AWS Well-Architected](https://docs.aws.amazon.com/wellarchitected/latest/reliability-pillar/welcome.html) - [High availability on Wikipedia](https://en.wikipedia.org/wiki/High_availability) --- ## HTML > The text format that describes what is on a web page: headings, paragraphs, links. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/html/ - In Spanish: [HTML](https://tidbit.legios.com.ar/es/html/) **HTML** (HyperText Markup Language) is the format every web page is written in. It is plain text with **tags**, short labels in angle brackets, that mark what each piece is: `h1` for a main heading, `p` for a paragraph, `a` for a link, `img` for an image. In the file a tag is written between angle brackets, and most come in pairs that open and close around the content. The browser reads those tags and draws the page. HTML is not a programming language: it does not make decisions or calculate anything. It describes structure and meaning. How the page looks (colors, fonts, layout) is the job of CSS, and what it does when you click is the job of JavaScript. The three work together on almost every site. Good HTML matters more than it seems. Screen readers used by blind people rely on the tags to know what is a heading, a button or a menu. Search engines and AI tools read them to understand a page. You can see the HTML of any site yourself: in most desktop browsers, right-click a page and choose "View page source". ### An analogy The labels an editor writes in the margin of a manuscript: "title here", "new paragraph", "photo goes here". They do not change the words; they tell the printer what each part is. ### Keep in mind - Plain text with tags like `h1`, `p` and `a` that mark each part. - Describes structure; CSS handles looks and JavaScript handles behavior. - The standard is maintained in the open by the WHATWG, with browser makers involved. - Accessibility and search depend on using the right tags. ### Go deeper - [HTML: HyperText Markup Language (MDN)](https://developer.mozilla.org/en-US/docs/Web/HTML) - [The HTML Living Standard (WHATWG)](https://html.spec.whatwg.org/) - [HTML on Wikipedia](https://en.wikipedia.org/wiki/HTML) --- ## HTTP > The request-and-response language browsers and websites speak. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/http/ - In Spanish: [HTTP](https://tidbit.legios.com.ar/es/http/) **HTTP** (HyperText Transfer Protocol) is the set of rules a browser and a web server use to talk. The browser sends a **request** (give me this page) and the server sends back a **response** (the page, or an error). Every web page, image and API call you use runs on this simple back-and-forth. A request has a method and a path. `GET /menu` means fetch the menu; `POST /order` means here is some data, create an order. The response starts with a status code: `200` is OK, `404` is not found, `500` is the server broke. Both carry **headers**, small notes such as the type of content, the language, or a cookie that identifies you. HTTP is stateless: each request stands alone, and the server does not remember the previous one, which is why cookies and sessions exist. Plain HTTP is readable by anyone in between, so today almost everything uses HTTPS, which is the same conversation inside an encrypted tunnel. Newer versions (HTTP/2 and HTTP/3) make the exchange faster without changing what you ask for. ### An analogy Ordering at a counter with a numbered ticket. You say what you want in a fixed form, you get a plate or a we are out of that, and the counter does not remember you between orders. ### Keep in mind - A request has a method (`GET`, `POST`...) and a path; a response has a status code (`200`, `404`, `500`). - HTTP is stateless: cookies are how a site remembers who you are between requests. - HTTPS is HTTP with encryption; the conversation is the same, only protected. - Most APIs on the web are just HTTP requests that return JSON instead of pages. ### Go deeper - [An overview of HTTP (MDN)](https://developer.mozilla.org/en-US/docs/Web/HTTP/Guides/Overview) - [What is HTTP? (Cloudflare)](https://www.cloudflare.com/learning/ddos/glossary/hypertext-transfer-protocol-http/) - [HTTP (Wikipedia)](https://en.wikipedia.org/wiki/HTTP) --- ## HTTPS > The padlock in your browser: web traffic nobody in between can read. - Level: Basic - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/https/ - In Spanish: [HTTPS](https://tidbit.legios.com.ar/es/https/) **HTTPS** is plain HTTP (the language browsers and websites speak) wrapped in encryption, using a protocol called **TLS**. When you open a site with the padlock, your browser and the server first agree on a secret key that only they know, then every request and response is scrambled with it. Anyone sitting in the middle (the café Wi-Fi, your internet provider, a hacker on the same network) sees that you are talking to a site, but not what you send: no passwords, no card numbers, no page content. HTTPS also proves identity. The server shows a **certificate** signed by an authority your browser trusts, so you know you are talking to your bank and not to a copy of it. Since Let's Encrypt made certificates free in 2015, almost the whole web moved to HTTPS. ### An analogy Sending a letter in a locked box where only you and the recipient have the key. The mail carrier still delivers it, but can only see the address, never the letter. ### Keep in mind - HTTPS = HTTP + TLS encryption. The padlock means the connection is encrypted. - It hides what you send, not which site you visit. - Certificates prove the site is who it says it is. ### Go deeper - [How HTTPS works (comic)](https://howhttps.works/) - [Let's Encrypt: how it works](https://letsencrypt.org/how-it-works/) --- ## Huawei > A Chinese company that makes mobile network equipment, phones and its own chips. - Level: Basic - Topics: Tech companies, Networks - URL: https://tidbit.legios.com.ar/huawei/ - In Spanish: [Huawei](https://tidbit.legios.com.ar/es/huawei/) **Huawei** was founded in 1987 in Shenzhen, China, by Ren Zhengfei. It started by reselling telephone switches and grew into one of the world's largest makers of **telecom equipment**: the antennas, base stations, fiber gear and core systems that mobile operators install so your phone can connect. When a carrier builds a 4G or 5G network, Huawei is one of a handful of companies, along with Ericsson and Nokia, that can supply the whole thing. It is also known for its consumer devices: smartphones, tablets, laptops, watches and earbuds. Its chip design unit, **HiSilicon**, designs processors such as the Kirin chips in its phones. Huawei also sells cloud services and equipment for companies, and its shares are not traded on any stock market: it is a private company that says it is owned by its employees. Huawei has been at the center of a long dispute between the United States and China. Several governments, citing national security concerns that Huawei rejects, have limited or banned its equipment in their 5G networks. Since 2019, US export restrictions cut Huawei off from American technology, including Google's apps on its new Android phones and chips made with US tools. In response it built its own operating system, **HarmonyOS**, and works with Chinese factories to make its chips. ### An analogy A company that builds both the roads and the cars. It sells the highways to the countries that run them, and the vehicles to the people who drive on them. ### Keep in mind - Founded in 1987 in Shenzhen by Ren Zhengfei; privately held. - One of the main suppliers of 4G and 5G network equipment, with Ericsson and Nokia. - Makes phones and other devices, and designs chips through HiSilicon. - US export restrictions since 2019 led it to build HarmonyOS and rely on Chinese suppliers. ### Go deeper - [Huawei corporate information (official site)](https://www.huawei.com/en/corporate-information) - [Huawei on Wikipedia](https://en.wikipedia.org/wiki/Huawei) --- ## Hugging Face > The website where people share AI models and datasets, and the tools to use them. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/hugging-face/ - In Spanish: [Hugging Face](https://tidbit.legios.com.ar/es/hugging-face/) **Hugging Face** is a company founded in 2016 in New York by three French entrepreneurs, Clément Delangue, Julien Chaumond and Thomas Wolf. It started as a chatbot app for teenagers, named after the hugging face emoji, and changed course when the open-source code it wrote for its own AI models became popular with researchers. Its best-known piece is the **Hub**, a website that works like a GitHub for AI. Anyone can upload a **model**, a **dataset** or a small demo app (called a **Space**), and anyone can download them. Most open-weight models, from large companies and from individuals, are published there. Each model has a page that explains what it does, its license and how to use it, and in many cases you can try it right in the browser. It also maintains popular open-source libraries, above all `transformers`, which lets a programmer load and run thousands of different models with a few lines of Python. Most of the Hub is free; Hugging Face makes money from paid plans for individuals and companies, private and enterprise features, and computing services that run models for customers without them having to manage servers. ### An analogy A big public library for AI, where anyone can donate a book, borrow one for free, or sit in the reading room and try it out, and the library charges only for private rooms and extra services. ### Keep in mind - Founded in 2016 in New York by three French entrepreneurs; named after the 🤗 emoji. - The Hub hosts models, datasets and demo apps called Spaces. - Its open-source library `transformers` is a standard way to run models in Python. - Free for most uses; earns money from paid plans, enterprise features and hosted computing. ### Go deeper - [Hugging Face Hub documentation](https://huggingface.co/docs/hub/index) - [Hugging Face on Wikipedia](https://en.wikipedia.org/wiki/Hugging_Face) --- ## Hybrid and multi-cloud > Running systems across your own servers and the cloud, or across several cloud providers. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/hybrid-and-multi-cloud/ - In Spanish: [Nube híbrida y multicloud](https://tidbit.legios.com.ar/es/hybrid-and-multi-cloud/) A **hybrid cloud** mixes a company's own servers, in its offices or a rented data center (called **on-premises** or on-prem), with a public cloud like AWS, Azure or Google Cloud, connected so they work as one system. A bank might keep its core accounts on its own hardware for regulatory reasons while running its mobile app and analytics in the cloud. A factory might process sensor data locally and send summaries to the cloud. **Multi-cloud** means using more than one public cloud provider. Sometimes it is deliberate: one provider's AI services, another's data warehouse, or a requirement not to depend on a single vendor. Very often it is just history: one team chose AWS, an acquired company runs on Azure, and marketing uses tools hosted on Google Cloud. Hybrid and multi-cloud are different ideas, and many companies are both. The benefits are flexibility, keeping data where rules require it, and less dependence on one vendor. The costs are real: each provider has its own services, security model, billing and skills, and moving data between them takes time and can carry fees for data leaving a provider. Tools like Kubernetes and infrastructure as code help by describing systems in a way that works in more than one place, but they do not make the differences disappear. ### An analogy Hybrid is keeping your own car while also using taxis for some trips. Multi-cloud is having accounts with two taxi companies. Both give you options, and both mean more apps, more bills and more rules to remember. ### Keep in mind - Hybrid: your own servers (on-premises) plus a public cloud, working together. - Multi-cloud: two or more public cloud providers at the same time. - Reasons include regulation, latency, best-of-breed services and avoiding lock-in. - Costs: more complexity, more skills to keep, and fees to move data out. ### Go deeper - [What is hybrid cloud? (AWS)](https://aws.amazon.com/what-is/hybrid-cloud/) - [What is multicloud? (Google Cloud)](https://cloud.google.com/learn/what-is-multicloud) - [Multicloud on Wikipedia](https://en.wikipedia.org/wiki/Multicloud) --- ## Hypervisor > The software layer that splits one physical computer into many virtual machines. - Level: Advanced - Topics: Cloud - URL: https://tidbit.legios.com.ar/hypervisor/ - In Spanish: [Hipervisor](https://tidbit.legios.com.ar/es/hypervisor/) A **hypervisor** is the software that creates and runs virtual machines. It sits between the physical hardware and the VMs, hands each one a share of CPU time, memory, storage and network, and keeps them isolated: a VM thinks it has a whole computer to itself, and it cannot read another VM's memory. This whole technique is called **virtualization**. There are two broad kinds. A **type 1** or bare-metal hypervisor runs directly on the hardware, with no ordinary operating system underneath; this is what data centers use, with examples like VMware ESXi, Microsoft Hyper-V, Xen and KVM, which is built into Linux. A **type 2** hypervisor runs as an app on your normal operating system, like VirtualBox or Parallels on a laptop; it is convenient for testing but adds a layer of overhead. Modern CPUs include special instructions that let hypervisors run VMs at close to native speed. Cloud providers build their own on top of these: AWS uses the **Nitro** system, which moves much of the virtualization work onto dedicated hardware cards, and **Firecracker**, a tiny hypervisor that starts lightweight VMs in a fraction of a second for AWS Lambda and Fargate. Because the hypervisor is the wall between customers, its security is critical. ### An analogy The manager of a shared office building who assigns each tenant their own floor, meters their electricity and makes sure nobody's key opens someone else's door. The tenants only see their own floor; the manager sees the whole building. ### Keep in mind - Creates virtual machines and shares out CPU, memory, disk and network among them. - Type 1 runs on bare hardware (ESXi, Hyper-V, KVM); type 2 runs on top of an OS (VirtualBox). - CPU support for virtualization makes VMs run at near-native speed. - The hypervisor is the isolation boundary between cloud customers. ### Go deeper - [What is a hypervisor? (AWS)](https://aws.amazon.com/what-is/hypervisor/) - [Hypervisor on Wikipedia](https://en.wikipedia.org/wiki/Hypervisor) --- ## IAM > The rulebook that says who can do what in your AWS account. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/iam/ - In Spanish: [IAM](https://tidbit.legios.com.ar/es/iam/) **IAM** (Identity and Access Management) is the AWS service that controls permissions. Every action in AWS (reading a file in Amazon S3, starting an Amazon EC2 instance, deleting a database) is checked against IAM first: who is asking, and are they allowed? By default the answer is no; you grant permissions explicitly. It has three main pieces. **Users** are people or programs with their own credentials. **Roles** are sets of permissions that something can take on temporarily: an AWS Lambda function, an EC2 instance, a person from another account. **Policies** are JSON documents that list what is allowed, for example 'allow `s3:GetObject` on the bucket `invoices` and nothing else'. You attach policies to users, groups or roles. Why it matters: most cloud security incidents are not clever hacks but leaked keys with too many permissions. The habit IAM encourages is **least privilege**: give each person and each program exactly what it needs. A reporting script should be able to read one bucket, not wipe the account. ### An analogy A building's badge system. Each badge (a user or a role) opens only the doors listed on it (the policies). The cleaning crew's badge opens the offices at night but not the server room, and a lost badge gets cancelled without changing the locks. ### Keep in mind - Everything is denied until a policy explicitly allows it. - Prefer roles over long-lived access keys; roles hand out temporary credentials. - Protect the root user with MFA and almost never use it. - IAM is free and comes with every AWS account. ### Go deeper - [What is IAM? (official docs)](https://docs.aws.amazon.com/IAM/latest/UserGuide/introduction.html) - [AWS IAM](https://aws.amazon.com/iam/) --- ## IBM > A century-old computer company that sells software, consulting and mainframes. - Level: Basic - Topics: Tech companies - URL: https://tidbit.legios.com.ar/ibm/ - In Spanish: [IBM](https://tidbit.legios.com.ar/es/ibm/) **IBM** (International Business Machines) is an American company that began in 1911 making machines to tabulate, record and count, and took its current name in 1924. For most of the 20th century it was the name in computing: its punched-card systems ran censuses and payrolls, and its **mainframes**, large, very reliable computers starting with the System/360 in 1964, became the core of banks, airlines and governments. In 1981 the IBM PC set the standard for the personal computer. IBM no longer sells PCs (it sold that business to Lenovo in 2005). Today it sells mainly to large organizations, in three ways. **Software**, much of it from Red Hat, which IBM bought in 2019, to run applications across a company's own data centers and several clouds, an idea it calls **hybrid cloud**. **Consulting**, where its staff help companies modernize systems and adopt AI. And **infrastructure**: modern mainframes, still used for card payments and bank accounts, and powerful servers. It earns money from software licenses and subscriptions, consulting fees and hardware sales with long support contracts. IBM is also known for research: its labs developed the hard disk drive, the relational database idea behind SQL and the DRAM memory chip, built Deep Blue, which beat the world chess champion in 1997, and Watson, which won a TV quiz show in 2011. Today one of its long bets is quantum computing. ### An analogy An old, large engineering firm that once built the city's power plants. It no longer makes the light bulbs people buy at the store, but it still runs and modernizes the plants, and sells the know-how to do it. ### Keep in mind - Founded in 1911; named IBM since 1924. - Its mainframes still process a large part of the world's banking and card transactions. - Makes money from software (including Red Hat), consulting and hardware for large organizations. - Its labs invented the hard disk drive and the relational database idea, and work on quantum computing. ### Go deeper - [About IBM (official site)](https://www.ibm.com/about) - [IBM on Wikipedia](https://en.wikipedia.org/wiki/IBM) --- ## IDE > The app where programmers write, run and fix code, with all the tools in one window. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/ide/ - In Spanish: [IDE](https://tidbit.legios.com.ar/es/ide/) An **IDE** (integrated development environment) is the main workplace of a programmer: an app that combines a code editor with the tools needed around it. In one window you see the project's files, edit the source code, run the program, read its errors, use a debugger and type commands in a built-in terminal. Visual Studio Code, IntelliJ IDEA, Xcode and Android Studio are well-known examples. What makes an IDE more than a text editor is that it understands the code. It colors the different parts so they are easy to read, suggests the name you are typing, underlines mistakes before you even run the program, and lets you jump to where something is defined or rename it everywhere at once. Many IDEs now also include AI assistants that suggest whole lines or explain code. The line between a light code editor and a full IDE is fuzzy. Some, like Xcode for Apple platforms, come with everything for one ecosystem. Others, like Visual Studio Code, start small and grow with extensions for each language. The choice is personal; what matters is that the tools are close at hand, because programmers spend most of their day inside them. ### An analogy A well-equipped workbench instead of a single screwdriver: the vise, the drill, the measuring tape and the light are all within reach, set up for the job you do every day. ### Keep in mind - Combines code editor, run button, debugger and terminal in one app. - Understands code: autocomplete, error underlines, jump to definition. - Examples: Visual Studio Code, IntelliJ IDEA, Xcode, Android Studio. - Extended with plugins for languages, tools and AI assistants. ### Go deeper - [What is an IDE? (AWS)](https://aws.amazon.com/what-is/ide/) - [IDE (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/IDE) - [Integrated development environment on Wikipedia](https://en.wikipedia.org/wiki/Integrated_development_environment) --- ## Identity theft > Someone uses your personal details to pass as you, usually to borrow or buy in your name. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/identity-theft/ - In Spanish: [Robo de identidad](https://tidbit.legios.com.ar/es/identity-theft/) **Identity theft** is when someone gets enough of your personal information to pretend to be you: your full name, ID or social security number, date of birth, address, phone number, sometimes a photo of your ID card or a selfie. With that, they can open a credit card or a loan in your name, sign up for a phone line, take over your existing accounts, claim benefits, or give your name when they are stopped by the police. The details usually come from many places at once. A data breach at a shop leaks emails and addresses; a phishing message tricks you into typing your bank login; a social media profile shows your birthday and your pet's name; a stolen wallet or a letter taken from a mailbox adds the rest. Criminals combine those pieces, and often buy and sell them in bulk. Victims often find out late: a call from a collection agency, a loan they never asked for on a credit report, or a bank account that suddenly does not let them in. Recovering takes paperwork: reporting it to the police and the institutions involved, closing fraudulent accounts, and in many countries asking credit bureaus to freeze or flag your file so nothing new can be opened. Prevention is mostly habits. Share ID numbers only when there is a real need, use unique passwords and two-factor authentication on email and banking, be suspicious of unexpected messages asking you to confirm details, check your bank statements and your credit report from time to time, and shred documents before throwing them away. ### An analogy Like someone collecting the pieces of your house key scattered around town, one from your mailbox, one from your social profile, one from a store that lost its records, until they can cut a full copy and walk in as if they lived there. ### Keep in mind - It is using your personal data to act, borrow or buy as you. - The data is pieced together from breaches, phishing, social media and stolen papers. - Act fast: report it, close fraudulent accounts and freeze your credit where possible. - Unique passwords, 2FA and sharing less data are the best prevention. ### Go deeper - [IdentityTheft.gov (US Federal Trade Commission)](https://www.identitytheft.gov/) - [Identity theft on Wikipedia](https://en.wikipedia.org/wiki/Identity_theft) --- ## Incognito mode > A browser window that forgets your visit on this device. It does not make you invisible. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/incognito-mode/ - In Spanish: [Modo incógnito](https://tidbit.legios.com.ar/es/incognito-mode/) **Incognito mode** (Chrome's name; Safari, Firefox and others call it **private browsing**) opens a browser window that keeps nothing once you close it: no history, no cookies, no saved form entries, no logins. The next person who uses the same computer will not see which sites you visited, and sites will not recognize you from an earlier session in that window. That is all it does, and it is useful: checking your email on a friend's laptop, buying a present on the shared family computer, or seeing how a website looks to someone who is not logged in. Every window opens clean and closes clean. What it does not do is hide you from anyone outside your device. The websites you visit still see your IP address and what you do there; if you log in, they know exactly who you are. Your employer or school can still see traffic on their network, and your internet provider can still see which sites you connect to. Files you download and bookmarks you save stay on the computer. Malware on the device is not stopped either. If you want to hide your traffic from the local network or your provider, that is a different tool, such as a VPN, which moves the trust to the VPN company instead. For privacy from trackers, browser settings that block third-party cookies help more than a private window does. ### An analogy Like a hotel room that is cleaned the moment you check out: the next guest finds no trace of you. But the front desk, the cameras in the lobby and the taxi that brought you all saw you arrive. ### Keep in mind - It forgets history, cookies and logins when the window closes. - It protects you from other people who use the same device. - Websites, network owners and your internet provider still see your visits. - Downloads and bookmarks stay; it is not a VPN or an antivirus. ### Go deeper - [Browse in Incognito mode (Chrome Help)](https://support.google.com/chrome/answer/95464) - [Private browsing on Wikipedia](https://en.wikipedia.org/wiki/Private_browsing) --- ## Infrastructure as code > Describing servers, networks and databases in files instead of clicking. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/infrastructure-as-code/ - In Spanish: [Infraestructura como código](https://tidbit.legios.com.ar/es/infrastructure-as-code/) **Infrastructure as code** (**IaC**) means writing down the cloud resources you need in a text file and letting a tool create them. Instead of clicking through the AWS console to make a virtual machine, a bucket and a database, you write "one Amazon EC2 instance of this size, one Amazon S3 bucket with this name, one Amazon RDS database", run a command, and the tool makes the cloud match the file. The big win is repeatability. The file lives in Git next to the application code, so every change is reviewed, has an author and can be undone. Need an identical staging environment? Run the same file again. Someone deleted a server by accident? Re-apply and it comes back. Clicking gives you none of that, and nobody remembers what was clicked six months ago. The common tools are **Terraform** (works with every cloud, with its own language called HCL), AWS CloudFormation (AWS only, in YAML or JSON), Pulumi and the AWS CDK (both use regular programming languages). All of them compare what the file says with what exists and show a plan ("will create 2, change 1, destroy 0") before touching anything. ### An analogy A written recipe instead of a chef's memory. With the recipe on paper, any kitchen can cook the same dish tonight, next week or in another city, and a typo in it is visible before dinner, not during. ### Keep in mind - Infrastructure files are versioned, reviewed and reused like application code. - Tools show a plan of changes before applying them. - Terraform is multi-cloud; CloudFormation and CDK are AWS-native. - "Drift" is when someone changes things by hand and the file no longer matches reality. ### Go deeper - [What is infrastructure as code? (AWS)](https://aws.amazon.com/what-is/iac/) - [What is Terraform? (official docs)](https://developer.hashicorp.com/terraform/intro) - [Infrastructure as code (Wikipedia)](https://en.wikipedia.org/wiki/Infrastructure_as_code) --- ## Intel > The American company behind most PC and server processors for decades. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/intel/ - In Spanish: [Intel](https://tidbit.legios.com.ar/es/intel/) **Intel** makes CPUs, the main processor of a computer. Its chips, sold under names like Core for PCs and Xeon for servers, have powered most of the world's desktops, laptops and data centers since the 1990s. It created **x86**, the instruction set that Windows PCs and most servers still use, and in 1971 it sold one of the first microprocessors. What sets it apart is that it both designs and manufactures its chips, in its own factories (fabs) in the United States, Ireland and Israel, among other places. Most rivals only design: AMD, Apple and NVIDIA send their designs to TSMC. Owning the factories was Intel's great advantage for decades; when it fell behind on the most advanced manufacturing processes in the late 2010s, that advantage turned into a burden. Today Intel is trying to win back its lead and to become a **foundry** too, making chips designed by other companies the way TSMC does, with backing from the US government, which wants advanced chipmaking on its own soil. Meanwhile it faces competition on several fronts: AMD in PCs and servers, ARM-based chips in laptops and the cloud, and NVIDIA in AI hardware. ### An analogy A bakery that writes its own recipes and runs its own ovens. For years that made it the best bakery in town; when a rival oven shop (TSMC) got better ovens, the bakers who only write recipes could suddenly bake finer bread. ### Keep in mind - Created the x86 architecture used by Windows PCs and most servers. - Designs and manufactures its own chips, unlike most competitors. - Fell behind TSMC in manufacturing and is now also offering to make chips for others. - Competes with AMD in CPUs, with ARM-based chips and with NVIDIA in AI. ### Go deeper - [Intel on Wikipedia](https://en.wikipedia.org/wiki/Intel) - [x86 on Wikipedia](https://en.wikipedia.org/wiki/X86) --- ## Internet > A worldwide network of networks that lets any connected device reach any other. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/internet/ - In Spanish: [Internet](https://tidbit.legios.com.ar/es/internet/) The **Internet** is a global network made of many smaller networks: those of homes, universities, companies, mobile carriers and internet providers, all connected to each other. No single company owns it or runs it. What holds it together is a set of shared rules, above all the **Internet Protocol**, which gives every connected device an IP address and lets data find its way from any point to any other. Data does not travel as one continuous stream. It is cut into small packets, each labeled with where it comes from and where it is going. Routers along the way pass each packet one step closer to its destination, often across several countries and through undersea fiber-optic cables, and the receiving device puts the pieces back together. If one path fails, packets can take another, which makes the whole network very hard to break. The Internet and the web are not the same thing. The **web**, the pages you open in a browser, is one service running on top of the Internet; email, video calls, online games, app updates and messaging are others. The Internet grew out of research networks in the United States in the late 1960s and 1970s, opened to the public in the 1990s, and today is coordinated by non-profit bodies that write the standards and hand out addresses and domain names. ### An analogy A worldwide postal system run by thousands of local post offices that agreed on how to write addresses. Nobody owns the whole system, yet a letter sent from anywhere can reach anywhere, handed from one office to the next. ### Keep in mind - A network of networks that nobody owns as a whole. - Data travels in packets, passed router to router until it reaches an IP address. - The web is one service on the Internet, alongside email, video calls and more. - It works because everyone follows the same open standards. ### Go deeper - [Internet on Wikipedia](https://en.wikipedia.org/wiki/Internet) - [A brief history of the Internet (Internet Society)](https://www.internetsociety.org/internet/history-internet/brief-history-internet/) --- ## Internet of things > Everyday objects with sensors and a network connection: thermostats, meters, doorbells. - Level: Basic - Topics: Hardware, Networks - URL: https://tidbit.legios.com.ar/internet-of-things/ - In Spanish: [Internet de las cosas](https://tidbit.legios.com.ar/es/internet-of-things/) The **Internet of things** (**IoT**) is the name for ordinary objects that carry a small computer, a few sensors and a network connection: a thermostat you control from your phone, a smart electricity meter, a video doorbell, a fitness band, a soil-moisture sensor in a field, a tracker inside a shipping container. Each one measures or does something and reports over the internet. Most IoT devices are simple. A cheap chip reads a sensor and sends a few numbers every so often, over Wi-Fi, Bluetooth or a mobile network, to a service in the cloud. There the data is stored, shown in an app and used to trigger actions: turn on the heating, send an alert when a freezer warms up, schedule maintenance before a machine breaks. The value is rarely in one device; it is in thousands of them reporting together. The weak spot is security. Many devices ship with default passwords, rarely get firmware updates and stay plugged in for years, and attackers have taken over large numbers of cameras and routers to build botnets. Good practice is boring but it works: change the default password, keep the firmware updated, and put these devices on a separate network when you can. ### An analogy Giving objects a voice and a phone line. One plant pot that can say "I'm dry" is not very clever, but a farm full of them can tell you exactly where to water. ### Keep in mind - Ordinary objects with a sensor, a small chip and a network connection. - They usually send small readings to a cloud service and an app. - The value comes from many devices reporting together. - Default passwords and missing updates make them easy targets. ### Go deeper - [Internet of things on Wikipedia](https://en.wikipedia.org/wiki/Internet_of_things) - [What is the Internet of Things? (IBM)](https://www.ibm.com/think/topics/internet-of-things) - [What is IoT? (AWS)](https://aws.amazon.com/what-is/iot/) --- ## IP address > The number that identifies a device on a network, like a postal address. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/ip-address/ - In Spanish: [Dirección IP](https://tidbit.legios.com.ar/es/ip-address/) An **IP address** is a number that identifies a device on a network, so that data can find its way to it. Your phone has one on the café Wi-Fi, the server hosting a website has one, your router has one. The classic form (IPv4) looks like `192.168.1.10`: four numbers from 0 to 255 separated by dots. Every packet of data travelling the internet carries a destination IP address and a sender IP address, exactly like an envelope with two addresses. Routers along the way read the destination and pass the packet closer, hop by hop. When you open a website, your browser first asks DNS to turn the name into an IP address, then sends its request there. There are two kinds. A **public IP address** is visible to the whole internet and is usually assigned to your home connection by your provider. A **private IP address** (like the `192.168.x.x` range) only works inside your local network; many devices share one public address through NAT. Because the world ran out of IPv4 numbers, the newer **IPv6** uses much longer addresses such as `2001:db8::1`, with room for every device on the planet. ### An analogy A postal address. The mail carrier does not know who you are, but with the street and number the letter arrives, and the sender's address on the back lets you reply. ### Keep in mind - IPv4 addresses look like `192.168.1.10`; IPv6 addresses are longer, like `2001:db8::1`. - Public IPs are reachable from the internet; private IPs only work inside your local network. - Your IP usually reveals your rough location and your internet provider, not your name. - DNS exists so people type names, not IP addresses. ### Go deeper - [What is my IP address? (Cloudflare)](https://www.cloudflare.com/learning/dns/glossary/what-is-my-ip-address/) - [IP address (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/IP_Address) - [IP address (Wikipedia)](https://en.wikipedia.org/wiki/IP_address) --- ## IPv6 > The newer internet addressing system, with practically unlimited addresses. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/ipv6/ - In Spanish: [IPv6](https://tidbit.legios.com.ar/es/ipv6/) Every device on the internet needs an IP address so data can find it. The original system, **IPv4**, uses addresses of 32 bits, written like `192.168.1.10`, which allows roughly 4.3 billion of them. That seemed plenty in the early 1980s; with phones, laptops, servers and connected gadgets everywhere, the free supply ran out in the 2010s. **IPv6** is its replacement, standardized in the late 1990s. An IPv6 address has 128 bits and is written as groups of hexadecimal digits separated by colons, like `2001:db8::8a2e:370:7334`. That gives roughly 340 undecillion addresses, a number with 39 digits, enough for every device to have its own public address with plenty to spare. It also simplifies things: devices can set up their own address automatically, and there is less need for NAT, the trick that lets a whole home share a single IPv4 address. The two versions cannot talk to each other directly, so the switch has been slow. For years most networks have run both side by side, which is called **dual stack**, and a large and growing share of traffic, especially on mobile networks, already travels over IPv6 without anyone noticing. You rarely need to do anything about it: your router, your phone and the sites you visit agree on which one to use by themselves. ### An analogy A city that ran out of phone numbers and switched to much longer ones. For years old and new numbers work side by side, but the new plan has so many that nobody will ever run out again. ### Keep in mind - IPv4 has roughly 4.3 billion addresses, and the free supply has run out. - IPv6 uses 128-bit addresses, enough for practically every device imaginable. - Its addresses look like `2001:db8::1`: hexadecimal groups separated by colons. - Most networks run both versions at once while the transition continues. ### Go deeper - [IPv6 on Wikipedia](https://en.wikipedia.org/wiki/IPv6) - [RFC 8200, the IPv6 specification (IETF)](https://www.rfc-editor.org/rfc/rfc8200) - [IPv6 (Google)](https://www.google.com/intl/en/ipv6/) --- ## ISP > The company you pay to connect your home or phone to the internet. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/isp/ - In Spanish: [ISP](https://tidbit.legios.com.ar/es/isp/) An **ISP** (internet service provider) is the company that connects you to the internet: the cable or phone company that brings a line to your home, a fiber provider, a mobile carrier or a satellite service. You pay it a monthly fee, and in exchange it links your home network, through your modem, to its own network, and from there to the rest of the internet. Behind that simple service there is a lot of infrastructure. The ISP runs the cables or antennas in your area, gives your connection an IP address, often provides a DNS service that turns names like `example.com` into addresses, and connects its network to others, either at shared meeting points called **internet exchange points** or by paying larger carriers. Smaller ISPs buy their access from bigger ones, which connect with each other, and together they make up the internet. Because all your traffic passes through it, your ISP can see a lot: which sites you connect to and when, even when the content itself is protected by HTTPS. What it may do with that information depends on the laws of each country. It also explains some everyday experiences: a slow connection at peak evening hours is often the ISP's network being busy, and an outage across your whole neighborhood is usually its problem, not your router's. A VPN or encrypted DNS hides part of your activity from the ISP, although the VPN provider then sees it instead. ### An analogy The company that builds and maintains the road from your house to the highway. You pay a toll to use it, it decides how many lanes there are, and it can see every car leave your driveway, even if it cannot see inside them. ### Keep in mind - The company that connects your home or phone to the internet. - It can reach you by cable, phone line, fiber, mobile network or satellite. - Assigns your IP address and often provides your DNS. - Sees which sites you connect to, even when the content is encrypted. ### Go deeper - [Internet service provider on Wikipedia](https://en.wikipedia.org/wiki/Internet_service_provider) - [Internet exchange point on Wikipedia](https://en.wikipedia.org/wiki/Internet_exchange_point) --- ## JavaScript > The programming language that runs inside web pages and makes them interactive. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/javascript/ - In Spanish: [JavaScript](https://tidbit.legios.com.ar/es/javascript/) **JavaScript** is the programming language built into every web browser. HTML says what is on a page and CSS says how it looks; JavaScript says what it does. When a menu opens, a form warns you that the email is missing, or a map lets you drag it around without reloading, JavaScript is doing that work on your own device. It was created in 1995 at Netscape, in a famously short time, and its official standard is called **ECMAScript**. Despite the name, it has nothing to do with Java; the name was a marketing choice of the era. Because every browser has to run the same JavaScript, the standard changes carefully, and old code keeps working. Today JavaScript also runs outside the browser, on servers and in desktop and phone apps, through runtimes like Node.js. That means one language can power both the frontend and the backend of a product, which is a big reason it is among the most used languages in the world. Many teams write **TypeScript**, a stricter version that adds type checks and is translated into plain JavaScript before it runs. ### An analogy If a web page were a theater play, HTML would be the script, CSS the costumes and set, and JavaScript the actors reacting to the audience. ### Keep in mind - Every web browser runs JavaScript; no installation needed. - Not related to Java despite the name. - Its standard is ECMAScript; TypeScript is a typed variant that compiles to it. - Also runs on servers through runtimes like Node.js. ### Go deeper - [What is JavaScript? (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/Scripting/What_is_JavaScript) - [JavaScript reference and guides (MDN)](https://developer.mozilla.org/en-US/docs/Web/JavaScript) - [JavaScript on Wikipedia](https://en.wikipedia.org/wiki/JavaScript) --- ## JSON > The plain-text format programs use to hand each other structured data. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/json/ - In Spanish: [JSON](https://tidbit.legios.com.ar/es/json/) **JSON** (JavaScript Object Notation, said "jason") is a text format for writing structured data: names with values, lists, and nested groups of them. A user might be `{"name": "Ana", "age": 31, "languages": ["es", "en"]}`. Humans can read it, every programming language can parse it, and it takes one line of code to turn it into an object you can work with. It was born inside JavaScript in the early 2000s and spread because it is far simpler than XML, the format it replaced. Today nearly every web API answers in JSON, many configuration files are JSON (`package.json`, `tsconfig.json`), and databases like Amazon DynamoDB or MongoDB store documents that are essentially JSON. It has only six kinds of value: string, number, true/false, null, array (`[...]`) and object (`{...}`). No dates, no comments, no trailing commas. That strictness is why it is so portable, and also why a single missing quote breaks the whole file. When size or speed matter more than readability, binary formats like Protocol Buffers or Parquet take over. ### An analogy A filled-in form with labeled boxes. Any clerk in any office can read "Name: Ana, Age: 31" because the labels travel with the values, and a box can contain another whole form. ### Keep in mind - Keys are strings in double quotes; values can be text, numbers, booleans, null, lists or objects. - It is the default response format of web APIs. - Every major language has a built-in JSON parser. - Strict syntax: one stray comma makes the file invalid. ### Go deeper - [Introducing JSON (json.org)](https://www.json.org/json-en.html) - [Working with JSON (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Core/Scripting/JSON) - [JSON (Wikipedia)](https://en.wikipedia.org/wiki/JSON) --- ## Keylogger > Software or a gadget that secretly records everything you type. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/keylogger/ - In Spanish: [Keylogger](https://tidbit.legios.com.ar/es/keylogger/) A **keylogger** records every key you press and sends the log to someone else. Passwords, card numbers, private messages, searches: if you type it, it ends up in the list. Most keyloggers are a kind of malware (often grouped under spyware) that sneaks onto a computer or phone through a phishing attachment, a pirated program or a fake app. A few are physical: a small plug between the keyboard and the computer, or a skimmer stuck on an ATM keypad. Not all of them are criminal. Some parental-control and employee-monitoring products record keystrokes openly, and the law in many countries limits when that is allowed. What makes a keylogger an attack is that it runs without the person's knowledge, to steal what they type. Protection is mostly good habits. Keep the operating system and the browser updated, install apps only from official stores, and be suspicious of attachments. A password manager helps because it fills passwords in without you typing them. And two-factor authentication is the real safety net: even if a keylogger captures your password, the attacker still needs the code on your phone or your security key. On shared or public computers, avoid logging into anything important. ### An analogy Someone standing behind you writing down every word as you type, except they are invisible and never get tired. ### Keep in mind - Records keystrokes and sends them to whoever installed it. - Usually arrives as malware; some are small devices plugged into the keyboard. - A password manager reduces how much you type; 2FA makes stolen passwords far less useful. - Be careful logging into important accounts on shared or public computers. ### Go deeper - [Keystroke logging on Wikipedia](https://en.wikipedia.org/wiki/Keystroke_logging) - [What is a keylogger? (Kaspersky)](https://www.kaspersky.com/resource-center/definitions/keylogger) --- ## Kubernetes > The system that runs and babysits containers across many machines. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/kubernetes/ - In Spanish: [Kubernetes](https://tidbit.legios.com.ar/es/kubernetes/) **Kubernetes** (often shortened to **K8s**) is open source software that runs containers across a group of machines, called a **cluster**. You tell it what you want: "run three copies of this image, give each one 1 GB of memory, expose them on port 80". Kubernetes finds room on the machines, starts the containers, and if one dies or a whole machine fails, it starts a replacement somewhere else without anyone being paged. That desired state is written in YAML files. A **Pod** is the smallest unit (one container, or a few that must run together); a **Deployment** keeps N copies of a Pod alive and rolls out new versions gradually; a **Service** gives them a stable address so traffic can find them. Change the YAML, apply it, and Kubernetes moves the cluster from the old state to the new one. Google built it from its internal system (Borg) and open-sourced it in 2014; today it is the default way to run large fleets of containers, and every cloud sells a managed version (Amazon EKS, Google GKE, Azure AKS). The cost is complexity: for one small app it is usually overkill, and simpler options like Amazon ECS and Fargate or a serverless function do the job. ### An analogy A conductor and an orchestra. The score says what each section should play; the conductor keeps everyone in time, brings in a substitute when a musician gets sick, and the audience only hears the music. ### Keep in mind - You declare the desired state; Kubernetes keeps reality matching it. - Pods, Deployments and Services are the three words to learn first. - It scales by adding machines to the cluster and copies of your Pods. - Managed offerings (EKS, GKE, AKS) run the control plane for you. ### Go deeper - [Kubernetes overview (official docs)](https://kubernetes.io/docs/concepts/overview/) - [What is Kubernetes? (AWS)](https://aws.amazon.com/what-is/kubernetes/) - [Kubernetes (Wikipedia)](https://en.wikipedia.org/wiki/Kubernetes) --- ## Linux > A free, open source operating system that runs most of the internet's servers. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/linux/ - In Spanish: [Linux](https://tidbit.legios.com.ar/es/linux/) **Linux** is a free and open source operating system. Strictly, the name refers to its kernel, the core that talks to the hardware, which Linus Torvalds started as a hobby project in 1991 and which thousands of people and companies now maintain together. In everyday use, "Linux" means the whole system: the kernel plus the tools, desktop and apps packaged around it. Those packages are called **distributions** ("distros"): Ubuntu, Debian, Fedora and Red Hat Enterprise Linux are well-known ones. They all share the same kernel but choose different software, update schedules and support. Anyone can download most of them for free, read the source code and change it. You probably use Linux every day without noticing. Most web servers and cloud machines run it, Android is built on the Linux kernel, and it powers routers, smart TVs, cars and virtually all of the world's fastest supercomputers. On desktops and laptops it is less common, but developers often like it because the same system runs on their computer and on their servers. ### An analogy A public engine design that anyone can build and improve. Different carmakers (the distributions) put it in their own bodywork, with their own seats and dashboard, but under the hood it is the same engine. ### Keep in mind - Started by Linus Torvalds in 1991; free and open source. - "Linux" is the kernel; distributions like Ubuntu or Debian package a full system around it. - Runs most servers and cloud machines, and Android is built on it. - Usually controlled from the command line on servers. ### Go deeper - [The Linux Kernel Archives (kernel.org)](https://www.kernel.org/) - [Linux on Wikipedia](https://en.wikipedia.org/wiki/Linux) --- ## LLM > A neural network trained on huge amounts of text to predict the next word. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/llm/ - In Spanish: [LLM](https://tidbit.legios.com.ar/es/llm/) An **LLM** (Large Language Model) is a neural network trained on an enormous amount of text: books, websites, code, conversations. Its only job during training is to guess the next token in a sequence. Do that well enough across trillions of words and the model ends up absorbing grammar, facts, writing styles and a surprising amount of reasoning. When you type a question into Claude or ChatGPT, the model does not look anything up. It reads your prompt, turns it into tokens, and produces the answer one token at a time, each time picking a likely continuation. The text you see is that stream of predictions, glued together. 'Large' refers to the number of weights (parameters), billions of them, and to the scale of data and GPU time used to train it. Bigger models usually generalize better, but they are also slower and more expensive to run, which is why companies offer several sizes of the same model. LLMs matter because they turned language into an interface: you describe what you want in plain words and get a draft, a summary, a translation or working code. They also make things up with confidence (hallucination), so they work best with a person checking the output. ### An analogy The autocomplete on your phone, but trained on most of the written internet and so good at guessing the next word that it can hold a conversation. ### Keep in mind - An LLM generates text one token at a time, predicting the most likely next one. - Its knowledge is frozen in its weights at training time; it does not browse unless given tools. - Claude, GPT, Gemini and Llama are all LLMs; they differ in size, data and training choices. - Fluent is not the same as correct: check the facts. ### Go deeper - [3Blue1Brown: But what is a GPT?](https://www.3blue1brown.com/lessons/gpt) - [Large language model on Wikipedia](https://en.wikipedia.org/wiki/Large_language_model) --- ## Load balancer > The traffic cop that spreads incoming requests across many servers. - Level: Intermediate - Topics: Networks, Cloud - URL: https://tidbit.legios.com.ar/load-balancer/ - In Spanish: [Balanceador de carga](https://tidbit.legios.com.ar/es/load-balancer/) A **load balancer** sits in front of a group of servers and distributes incoming requests among them. Users all connect to one address (the balancer's); behind it, the same website or API runs on two, ten or a thousand identical machines, and each request is sent to one of them. The user never knows which. It solves two problems at once. **Capacity**: one server can only handle so many requests per second, so you add more and split the load. **Availability**: the balancer checks each server's health every few seconds and stops sending traffic to one that fails, so a crashed machine means a slightly busier fleet instead of an outage. It also lets you update servers one at a time with no downtime. Balancers choose servers with simple rules (round robin, fewest connections, by region) and many also terminate HTTPS, so the TLS certificate lives in one place. In the cloud this is a managed service: AWS has the Application Load Balancer and the Network Load Balancer, and platforms like Kubernetes build one in. Combined with autoscaling (add servers when busy, remove them when quiet), a load balancer is what lets a site survive being on the news. ### An analogy The person at the bank entrance who sends each customer to the next free window. Ten windows, one queue; if a teller goes on break, nobody is sent there. ### Keep in mind - One public address in front, many identical servers behind. - Health checks pull a broken server out of rotation automatically. - It is what makes zero-downtime deploys and autoscaling possible. - Often also terminates HTTPS, so certificates live in one place. ### Go deeper - [What is load balancing? (Cloudflare)](https://www.cloudflare.com/learning/performance/what-is-load-balancing/) - [Load balancing, computing (Wikipedia)](https://en.wikipedia.org/wiki/Load_balancing_(computing)) --- ## Low-code and no-code > Building apps and automations by dragging blocks instead of writing code. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/low-code-and-no-code/ - In Spanish: [Low-code y no-code](https://tidbit.legios.com.ar/es/low-code-and-no-code/) **No-code** tools let people build software without programming: you drag fields onto a form, connect boxes in a flow, or set rules in menus, and the tool turns that into a working app. **Low-code** tools work the same way but let you drop into real code when the blocks run out, so they are aimed at developers who want to go faster, and at technical people who are not full-time developers. Familiar examples include Airtable and Notion for databases with a friendly face, Zapier and Make for automations ("when a form is submitted, add a row and send an email"), Webflow and Bubble for websites and web apps, and Microsoft Power Apps or Salesforce's builders inside large companies. A typical win: an operations team builds its own approval tracker in a week instead of waiting months in the IT queue. The limits show up as things grow. You can only do what the platform allows; complex logic gets awkward; performance and costs can climb with usage; and moving away later is hard because your app only exists inside that vendor's tool, which is called **lock-in**. Companies also worry about **shadow IT**: important processes built by individuals that nobody else knows about, secures or maintains. Many teams use these tools for prototypes and internal tools, and switch to code when something becomes core to the business. ### An analogy Like building with LEGO versus carving wood. LEGO is fast and anyone can do it, but you are limited to the bricks in the box. Low-code is LEGO that also lets you carve a custom piece when you need one. ### Keep in mind - No-code: visual building only. Low-code: visual building plus some code. - Great for prototypes, internal tools and automations. - You are limited to what the platform supports, and leaving it is hard. - Apps built outside IT still need owners, security and backups. ### Go deeper - [What is low-code? (AWS)](https://aws.amazon.com/what-is/low-code/) - [Low-code development platform on Wikipedia](https://en.wikipedia.org/wiki/Low-code_development_platform) - [No-code development platform on Wikipedia](https://en.wikipedia.org/wiki/No-code_development_platform) --- ## MAC address > The built-in name of a network card, used to deliver data inside a local network. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/mac-address/ - In Spanish: [Dirección MAC](https://tidbit.legios.com.ar/es/mac-address/) A **MAC address** is an identifier that comes with every network interface: the Wi-Fi chip in your phone, the Ethernet port in your laptop, the radio in a smart bulb. It is 48 bits long and usually written as six pairs of hexadecimal digits, like `3c:22:fb:9a:10:4e`. The first half says which manufacturer made the hardware, from blocks handed out by the IEEE; the second half is that maker's serial for the card. It has nothing to do with Apple's Mac computers: MAC stands for Media Access Control. Its job is local delivery. Inside one network, say your home Wi-Fi, devices find each other by MAC address: the router knows that `3c:22:fb:9a:10:4e` is your laptop and sends its data there. To get from an IP address to the right MAC address, devices simply ask the network out loud, 'who has 192.168.1.20?', and the owner answers (that exchange is called **ARP**). A MAC address never leaves your local network. When your data goes out to the internet, each router along the way rewraps it with its own MAC address for the next hop, while the IP address stays the same from end to end. That is why the IP address works like a postal address and the MAC address like the name of the person who opens the door. Because it used to be fixed, a MAC address could be used to recognize the same phone in every café and shopping mall. Modern phones and laptops now use a **private** or **random MAC address**, a different one for each Wi-Fi network. It is also why 'MAC filtering' on a router is weak protection: any device can be told to announce a different address. ### An analogy The IP address is the street address on an envelope; the MAC address is the name of the person inside the building. The mail carrier uses the street, and the doorman uses the name. ### Keep in mind - 48 bits, written as six pairs like `3c:22:fb:9a:10:4e`. - The first half identifies the manufacturer of the network hardware. - Only used inside a local network; routers swap it at every hop. - Phones now use a random MAC address per Wi-Fi network for privacy. ### Go deeper - [MAC address on Wikipedia](https://en.wikipedia.org/wiki/MAC_address) - [Address Resolution Protocol (ARP) on Wikipedia](https://en.wikipedia.org/wiki/Address_Resolution_Protocol) --- ## Machine learning > Teaching computers by showing them examples instead of writing every rule. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/machine-learning/ - In Spanish: [Machine learning](https://tidbit.legios.com.ar/es/machine-learning/) **Machine learning** (ML) is the part of AI where a program learns from examples instead of following rules a person wrote. You give it lots of data where the answer is known, it adjusts itself until it gets those answers mostly right, and then it applies what it learned to new cases it has never seen. The term goes back to 1959, when Arthur Samuel used it for a program that learned to play checkers. Take a spam filter. Writing rules by hand ('block anything that says FREE MONEY') is endless, because spammers change words faster than you can. With machine learning you collect thousands of emails people already marked as spam or not, and the system learns which combinations of words, senders and links tend to go with spam. When a new email arrives, it gives a probability, not a certainty. The same recipe is everywhere: predicting which card payment is fraud, which customers might cancel, how long a delivery will take, which movie you will like. Learning from examples has a catch, though: the model can only be as good as its data. If the examples are few, outdated or skewed, the predictions will be too. Deep learning and today's large language models are machine learning at a much larger scale. ### An analogy Like learning to tell ripe avocados by squeezing hundreds of them with someone who knows, instead of reading a rulebook. After a while you just know, even if you cannot fully explain how. ### Keep in mind - Learns patterns from examples where the right answer is known. - Outputs are predictions with a probability, not guarantees. - Bad, scarce or biased data produces bad predictions. - Two phases: training builds the model, inference uses it. ### Go deeper - [Machine learning on Wikipedia](https://en.wikipedia.org/wiki/Machine_learning) - [Intro to machine learning (Google)](https://developers.google.com/machine-learning/intro-to-ml) --- ## Malware and ransomware > Software built to harm you; ransomware locks your files and charges to unlock them. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/malware-and-ransomware/ - In Spanish: [Malware y ransomware](https://tidbit.legios.com.ar/es/malware-and-ransomware/) **Malware** is the umbrella word for software written to do damage: **viruses** that copy themselves, **trojans** that hide inside a program you wanted (a cracked game, a fake PDF reader), **spyware** that records what you type, and bots that quietly turn your computer into a soldier for someone else's attack. It arrives through phishing attachments, downloads from shady sites, infected USB sticks, or by exploiting a Vulnerability in software you never updated. **Ransomware** is the kind that makes the news. It uses encryption against you: it scrambles your documents, photos and databases with a key only the attacker has, then leaves a note asking for payment, usually in cryptocurrency, to get the key. Modern gangs also copy the data first and threaten to publish it, so a backup alone does not end the blackmail. Why it matters: in 2017 WannaCry spread through unpatched Windows machines and shut down hospitals in the UK; in 2021 Colonial Pipeline paused fuel delivery on the US east coast after an attack. Hospitals, city governments, schools and small businesses are the typical victims, not because they are special, but because they run old software and have no second copy of their data. Practical takeaway: keep the operating system and browser on automatic updates, keep a backup that is not permanently connected to the computer (a cloud backup with version history counts), and do not install programs from outside the official stores or the maker's site. Paying is a last resort: it funds the next attack and the key does not always arrive. ### An analogy A burglar who does not steal your photo albums. Instead, they change the lock on the cabinet that holds them and offer to sell you the new key. ### Keep in mind - Malware is any software made to harm; ransomware is the kind that locks your files for money. - It usually enters through an attachment, a dubious download or software that was never updated. - Updates and an offline or versioned backup defuse most of the damage. - Paying the ransom does not guarantee the key and funds the next attack. ### Go deeper - [What is ransomware? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/security/ransomware/what-is-ransomware/) - [Malware on Wikipedia](https://en.wikipedia.org/wiki/Malware) - [Ransomware on Wikipedia](https://en.wikipedia.org/wiki/Ransomware) --- ## Mercado Libre > Latin America's largest online marketplace, with its own payments and delivery network. - Level: Basic - Topics: Tech companies - URL: https://tidbit.legios.com.ar/mercado-libre/ - In Spanish: [Mercado Libre](https://tidbit.legios.com.ar/es/mercado-libre/) **Mercado Libre** was founded in 1999 in Buenos Aires by Marcos Galperin and a few partners, inspired by the online auction sites of the time. It grew into the biggest e-commerce company in Latin America, operating in many countries of the region, with Brazil, Mexico and Argentina as its largest markets. Like Amazon's marketplace, most products are listed by independent **sellers**, from people clearing out their garage to large brands, who pay a commission on each sale. To make online shopping work in a region where many people had no credit card and trusted strangers little, it built the missing pieces itself. **Mercado Pago** started as a way to pay safely inside the marketplace and became a fintech of its own: a digital wallet, QR payments in shops, card readers for small merchants, transfers, savings and loans. **Mercado Envíos** is its logistics network, with warehouses, sorting centers, delivery fleets and cargo planes, so a package can arrive the next day or even the same day. It makes money from sale commissions, shipping fees, payment processing fees, interest on loans and ads that sellers buy to appear first. Behind all of it is a large technology operation: Mercado Libre builds most of its own software and is one of the biggest employers of software engineers in Latin America. ### An analogy A huge regional fair that, to make sure deals actually happen, also opened its own bank at the entrance and its own delivery service at the exit. ### Keep in mind - Founded in 1999 in Buenos Aires; the largest e-commerce company in Latin America. - Most items are sold by independent sellers who pay a commission per sale. - Mercado Pago is its fintech: wallet, QR payments, card readers and loans. - Mercado Envíos, its own logistics network, handles storage and delivery. ### Go deeper - [Mercado Libre investor relations (official site)](https://investor.mercadolibre.com/) - [Mercado Libre on Wikipedia](https://en.wikipedia.org/wiki/Mercado_Libre) - [Mercado Pago on Wikipedia](https://en.wikipedia.org/wiki/Mercado_Pago) --- ## Mesh Wi-Fi > Several Wi-Fi units spread around a home that work together as one network. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/mesh-wifi/ - In Spanish: [Wi-Fi mesh](https://tidbit.legios.com.ar/es/mesh-wifi/) A **mesh Wi-Fi** system covers a home with several small units, called **nodes**, instead of a single router. One node connects to your modem; the others go around the house, and they talk to each other to pass traffic along. To your phone and laptop it all looks like one Wi-Fi network with one name and one password, and devices switch to the nearest node as you walk from room to room. It solves a common problem. A single router sits in one spot, and its signal weakens with distance and with every wall, floor and appliance in the way, so far rooms get slow Wi-Fi or none. Older **range extenders** helped by repeating the signal, but they often created a second network and cut the speed roughly in half. A mesh system coordinates its nodes: it picks the best path for each device and, in better models, uses a separate radio channel or a cable between nodes so their link back to the router does not compete with your devices. Mesh is not magic. Each hop between nodes adds some delay and can lose speed, so placement matters: nodes should be close enough to hear each other well, ideally halfway between the router and the dead zone, not inside it. Connecting the nodes with an Ethernet cable, when you can, gives the best results. In a small apartment one good router is usually enough; mesh pays off in larger homes, on several floors or behind thick walls. ### An analogy A relay team instead of one runner. Rather than one person trying to carry the message to every room, several stand at different points and pass it along, so it arrives everywhere without anyone running out of breath. ### Keep in mind - Several nodes work together as one Wi-Fi network with a single name. - Devices move to the nearest node as you walk around. - Better than old extenders, which often split the network and roughly halved speed. - Placement matters; wiring the nodes together gives the best speed. ### Go deeper - [Wireless mesh network on Wikipedia](https://en.wikipedia.org/wiki/Wireless_mesh_network) - [Wi-Fi EasyMesh (Wi-Fi Alliance)](https://www.wi-fi.org/discover-wi-fi/wi-fi-easymesh) --- ## Meta > The company behind Facebook, Instagram and WhatsApp, paid for by advertising. - Level: Basic - Topics: Tech companies - URL: https://tidbit.legios.com.ar/meta/ - In Spanish: [Meta](https://tidbit.legios.com.ar/es/meta/) **Meta** is the company that started as **Facebook**, created by Mark Zuckerberg in 2004 while he was a student at Harvard. It bought Instagram in 2012 and WhatsApp in 2014, and in 2021 renamed itself Meta Platforms to signal a bet beyond social networks. It is based in Menlo Park, California, and its apps are used by billions of people. The apps are free, and nearly all of Meta's revenue comes from **advertising**. Businesses, from a corner bakery to global brands, pay to show ads in the feeds and stories of Facebook and Instagram, and Meta's systems choose who sees each ad based on what people do in its apps. That ability to reach a specific audience cheaply is what advertisers pay for. WhatsApp makes money mostly by charging businesses to message their customers. Two other efforts are important for tech. Meta's AI lab builds the **Llama** models and publishes their weights, so anyone can download them and run them on their own computers, which made them a common base for open models. And **Reality Labs** makes virtual reality headsets (Quest) and smart glasses; it is a long-term bet that has cost a lot of money so far. ### An analogy A huge town square that charges no entry. People come to meet their friends, and the square makes its money renting the walls and the noticeboards to shops that want those people to see their posters. ### Keep in mind - Started as Facebook in 2004; renamed Meta in 2021; owns Instagram and WhatsApp. - Its apps are free; nearly all of its revenue comes from targeted ads. - It publishes the weights of its Llama AI models. - Reality Labs builds VR headsets and smart glasses as a long-term bet. ### Go deeper - [About Meta](https://about.meta.com/) - [Llama](https://www.llama.com/) - [Meta Platforms on Wikipedia](https://en.wikipedia.org/wiki/Meta_Platforms) --- ## Metadata > Data that describes other data: who made it, when, where and what it contains. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/metadata/ - In Spanish: [Metadatos](https://tidbit.legios.com.ar/es/metadata/) **Metadata** is information about a piece of data rather than the data itself. A photo on your phone is the image, but it also carries the date it was taken, the place, the camera model and the file size. An email has a body, and also a sender, a recipient, a subject line and a timestamp. All of that is metadata. In companies, metadata is what makes data usable. A table in a database is just rows and columns until something says what each column means, who owns the table, when it was last updated and where the numbers came from. Search engines, music apps and libraries all work the same way: they find things by their metadata, not by reading every file from scratch. Metadata can also be sensitive on its own. Even without reading a single message, knowing who called whom, when and from where says a lot about a person. That is why privacy rules often cover metadata too, and why some apps strip the location from photos before you share them. ### An analogy The label on a jar of homemade jam: it is not the jam, but it tells you the flavor, who made it and when, so you know what you are holding without opening it. ### Keep in mind - Describes data: author, date, size, format, origin, meaning. - It is how search, catalogs and file systems find things quickly. - In a company, good metadata is what lets people trust and reuse data. - Metadata alone can reveal private details, so it deserves protection too. ### Go deeper - [Metadata on Wikipedia](https://en.wikipedia.org/wiki/Metadata) - [What is metadata? (IBM)](https://www.ibm.com/think/topics/metadata) --- ## Metaverse > The idea of shared 3D virtual worlds where people meet, play and work as avatars. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/metaverse/ - In Spanish: [Metaverso](https://tidbit.legios.com.ar/es/metaverse/) The word **metaverse** comes from Snow Crash, a 1992 science fiction novel by Neal Stephenson, where people put on goggles and enter a virtual city as **avatars**. Today it names a broad idea: online 3D spaces that are always on, where many people are present at the same time, can talk, play, buy things and build, and maybe carry their identity and belongings from one world to another. The term became a buzzword in 2021, when Facebook renamed itself Meta and said building the metaverse was its next big goal. Other companies followed with announcements, and for a while almost any 3D or virtual reality product was called a metaverse. Interest cooled in the following years, as attention moved to generative AI. What actually exists is a set of separate pieces. Online games and platforms such as Roblox, Minecraft, Fortnite and VRChat already host huge shared worlds and concerts. **Virtual reality** headsets like Meta Quest and **mixed reality** devices like Apple Vision Pro let you feel inside a 3D space, and some companies use them for training or meetings. What does not exist is a single, connected metaverse: each world has its own accounts, rules and owner, and you cannot walk from one to another as you can between websites. ### An analogy Like the word "cyberspace" in the 1990s: a vivid name for a future that was still being built, used to describe many different things that only partly match the dream. ### Keep in mind - The word comes from Neal Stephenson's 1992 novel Snow Crash. - It means shared, always-on 3D worlds where people meet as avatars. - Became a buzzword in 2021, when Facebook renamed itself Meta. - Today there are many separate virtual worlds, not one connected metaverse. ### Go deeper - [Metaverse on Wikipedia](https://en.wikipedia.org/wiki/Metaverse) - [Virtual world on Wikipedia](https://en.wikipedia.org/wiki/Virtual_world) --- ## Microservices > Building one application as many small services that each do one job. - Level: Advanced - Topics: Software, Cloud - URL: https://tidbit.legios.com.ar/microservices/ - In Spanish: [Microservicios](https://tidbit.legios.com.ar/es/microservices/) **Microservices** is a way of structuring software in which an application is split into many small, independent services. An online shop might have one service for users, one for the catalog, one for orders and one for payments. Each runs on its own, usually in its own container, often owns its own database, and talks to the others only through APIs or messages in a queue. The alternative is a **monolith**: one program that contains everything and is deployed as a single piece. Monoliths are simpler to build, test and understand, which is why most products start as one. Trouble appears when the code and the team get large: a small change requires redeploying the whole thing, one slow part drags down the rest, and dozens of people step on each other's toes. Microservices answer that by letting each team own a service, change it and deploy it on its own schedule, and scale only the parts under load. The price is complexity. A function call inside one program becomes a network call that can be slow or fail; a single database transaction becomes coordination across several services; and finding a bug means following a request through many logs. Done without need, it creates a "distributed monolith" with the problems of both. Tools like Kubernetes, load balancers and tracing exist largely to manage this. ### An analogy A monolith is a single restaurant kitchen where every cook shares one stove. Microservices is a food court: each stall has its own kitchen, menu and staff, and can renovate without closing the others, but now someone has to coordinate orders that need food from three stalls. ### Keep in mind - Each service does one job, can be deployed alone, and usually owns its own data. - Services talk over the network through APIs or message queues. - Gains: independent teams, deploys and scaling. Costs: network failures and operational complexity. - Many teams start with a monolith and split it only when it hurts. ### Go deeper - [Microservices, by Martin Fowler and James Lewis](https://martinfowler.com/articles/microservices.html) - [What are microservices? (AWS)](https://aws.amazon.com/microservices/) - [Microservices on Wikipedia](https://en.wikipedia.org/wiki/Microservices) --- ## Microsoft > The company of Windows and Office, now just as much a cloud company with Azure. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/microsoft/ - In Spanish: [Microsoft](https://tidbit.legios.com.ar/es/microsoft/) **Microsoft** was founded in 1975 by Bill Gates and Paul Allen, and grew by selling software for personal computers: first MS-DOS, then **Windows**, which became the operating system of most office and home PCs, and **Office** (Word, Excel, PowerPoint, Outlook). It is based in Redmond, near Seattle. Its biggest shift was from selling software in a box to renting it by subscription. Office became **Microsoft 365**, paid monthly per person, and companies also pay for Teams, Outlook email and the tools that manage employees' accounts and devices. Most of its customers are businesses and governments, which is why Microsoft matters even to people who never think about it: their employer probably runs on it. The other half is **Azure**, its cloud, one of the largest next to AWS: companies rent servers, databases and AI services in Microsoft's data centers. Microsoft has been a major investor in and partner of OpenAI since 2019; OpenAI's models run on Azure and power Microsoft's **Copilot** assistants inside Windows, Office and GitHub. It also owns GitHub, where most open source code lives, LinkedIn, and the Xbox gaming business. ### An analogy The company that supplies an office building with its desks, filing cabinets and phones, and rents them by the month instead of selling them. Lately it also rents out the building's back rooms full of computers to anyone who needs them. ### Keep in mind - Founded in 1975 by Bill Gates and Paul Allen; based in Redmond, Washington. - Windows and Microsoft 365 (Office) are sold mostly to businesses, by subscription. - Azure is one of the largest clouds and runs OpenAI's models. - It also owns GitHub, LinkedIn and Xbox. ### Go deeper - [Azure documentation](https://learn.microsoft.com/en-us/azure/) - [About GitHub](https://github.com/about) - [Microsoft on Wikipedia](https://en.wikipedia.org/wiki/Microsoft) --- ## Mistral AI > A French company that builds AI language models, many of them with open weights. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/mistral-ai/ - In Spanish: [Mistral AI](https://tidbit.legios.com.ar/es/mistral-ai/) **Mistral AI** was founded in Paris in 2023 by Arthur Mensch, who came from Google DeepMind, and Guillaume Lample and Timothée Lacroix, who came from Meta's AI research lab. It builds large language models, the kind of AI behind chat assistants, and quickly became one of the best-known AI companies in Europe. What made it stand out early was releasing models with **open weights**. Mistral 7B and Mixtral, among its first models, could be downloaded by anyone and run on their own computers or servers, under a permissive license. Mixtral popularized the **mixture of experts** design, where only part of the model is used for each word, which makes it cheaper to run. Mistral also offers more capable models that are available only through its paid service. It makes money by selling access to its models through an API, by charging companies that want to run them on their own infrastructure, often for privacy or legal reasons, and through **Le Chat**, its chat assistant for people and businesses. Being European is part of its pitch: some governments and companies prefer a provider subject to European law. ### An analogy A bakery that sells bread at the counter but also publishes some of its recipes. You can buy a loaf ready to eat, or take a recipe home and bake it in your own oven. ### Keep in mind - Founded in Paris in 2023 by former DeepMind and Meta researchers. - Releases many of its models with open weights that anyone can download. - Its Mixtral models helped popularize the mixture of experts design. - Earns money from its API, enterprise deployments and the Le Chat assistant. ### Go deeper - [About Mistral AI (official site)](https://mistral.ai/about) - [Mistral AI documentation](https://docs.mistral.ai/) - [Mistral AI on Wikipedia](https://en.wikipedia.org/wiki/Mistral_AI) --- ## Model Context Protocol > An open standard for plugging tools and data into AI assistants the same way every time. - Level: Intermediate - Topics: AI, Software - URL: https://tidbit.legios.com.ar/model-context-protocol/ - In Spanish: [Model Context Protocol](https://tidbit.legios.com.ar/es/model-context-protocol/) The **Model Context Protocol** (MCP) is an open standard that defines how an AI application talks to outside tools and data. Anthropic published it in November 2024, and it was soon adopted across the industry, including by OpenAI, Google and Microsoft. Before MCP, every AI app needed its own custom integration for every service it wanted to use; with MCP, a tool built once works with any app that speaks the protocol. It has two sides. An **MCP server** is a small program that wraps some system (your files, a calendar, a company database, GitHub, a design tool) and describes what it offers: **tools** the AI can call to do things, **resources** it can read, and ready-made **prompts**. An **MCP client** lives inside the AI app, such as Claude, ChatGPT or a code editor, connects to servers, and shows the model what is available. When you ask 'what is on my calendar tomorrow?', the model decides to call the calendar server's tool, gets the answer, and uses it in its reply. Servers can run on your own computer or remotely over the internet. That makes MCP a building block for AI agents, which need to act, not just talk. It also deserves care. A server can do whatever its permissions allow, so a careless or malicious one could read private data or take actions you did not intend, and text coming back from a tool can carry hidden instructions meant to trick the model (prompt injection). Install servers only from sources you trust, give them the minimum access they need, and keep an eye on what an assistant asks to do. ### An analogy Like the standard electrical outlet: appliance makers do not need a deal with each house, they just use the standard plug. MCP is that outlet for AI apps, and each MCP server is an appliance you can plug in. ### Keep in mind - An open standard released by Anthropic in November 2024. - Servers expose tools, data and prompts; clients sit inside AI apps. - Build an integration once and it works with many AI apps. - A server acts with your permissions, so install only trusted ones. ### Go deeper - [What is the Model Context Protocol? (official site)](https://modelcontextprotocol.io/docs/getting-started/intro) - [Introducing the Model Context Protocol (Anthropic)](https://www.anthropic.com/news/model-context-protocol) - [Model Context Protocol on Wikipedia](https://en.wikipedia.org/wiki/Model_Context_Protocol) --- ## Model parameters > The billions of numbers a model learns in training; '70B' means 70 billion of them. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/model-parameters/ - In Spanish: [Parámetros de un modelo](https://tidbit.legios.com.ar/es/model-parameters/) A neural network is, underneath, a huge collection of numbers. During training each of those numbers is adjusted again and again until the model gets good at its task. Those learned numbers are the **parameters**, also called **weights**. They are the model: when people 'download a model', they are downloading a file full of them. When you see a name like 'Llama 70B' or hear 'a 7-billion-parameter model', the number is how many parameters it has: B means billion. In 2020 GPT-3 had 175 billion, which seemed enormous at the time. More parameters give a model more room to store patterns, so bigger models tend to be more capable, but the quality and amount of training data matter just as much, and a well-trained small model can beat a sloppy big one. Many companies no longer publish parameter counts for their top models. The count also tells you what hardware you need. Each parameter is stored with a certain precision: at 16 bits it takes 2 bytes, so a 70B model needs roughly 140 GB of memory just to load, far more than a laptop has. **Quantization** stores each number with fewer bits, for example 4, shrinking the same model to roughly 35 GB at a small cost in quality. Some models, called **mixture of experts**, only use part of their parameters for each word, so they run faster than their total size suggests. ### An analogy Parameters are like the knobs on a gigantic mixing desk. Training turns each knob a tiny bit until the sound is right. A '70B' desk has 70 billion knobs, which is why it needs a very big room. ### Keep in mind - Parameters, or weights, are the numbers a model learns in training. - '70B' means roughly 70 billion parameters. - Memory needed is about parameters times bytes per parameter. - Bigger is often more capable, but data and training matter as much. ### Go deeper - [Large language model on Wikipedia](https://en.wikipedia.org/wiki/Large_language_model) - [Machine learning glossary: parameter (Google)](https://developers.google.com/machine-learning/glossary#parameter) - [Mixture of experts on Wikipedia](https://en.wikipedia.org/wiki/Mixture_of_experts) --- ## Modem > The box that connects your home to your internet provider's network. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/modem/ - In Spanish: [Módem](https://tidbit.legios.com.ar/es/modem/) A **modem** is the device that connects your home to the network of your internet provider. The name comes from "modulator-demodulator": it turns digital data into a signal that can travel over the line coming into your home, and turns the incoming signal back into data. Which kind you have depends on that line: a DSL modem for a telephone line, a cable modem for TV cable, a box called an ONT for fiber optic, or a cellular modem for 4G or 5G. The modem and the router do different jobs. The modem talks to the outside world over that one line. The router builds your home network: it gives each device a local address, runs the Wi-Fi and directs traffic between your devices and the modem. In many homes both live in a single box supplied by the provider, which is why people mix up the names, but inside they are still two separate functions. Knowing the difference helps when something breaks. If no device in the house can reach the internet, the problem is often the modem or the provider's line, and the modem's lights usually show whether it has a signal. If one device works and another does not, or the Wi-Fi is weak in some rooms, the router or the Wi-Fi coverage is the more likely culprit. The classic fix, restarting the modem first and then the router, follows the order in which they set up the connection. ### An analogy The modem is the front door that connects the house to the street; the router is the hallway that leads each visitor to the right room. Sometimes both are built into the same entrance, but they are still different parts of the house. ### Keep in mind - A modem connects your home to the provider's line: phone, cable, fiber or mobile. - The router builds the home network and the Wi-Fi; the modem talks to the outside. - Many providers supply one box that does both jobs. - If nothing in the house connects, check the modem and the line first. ### Go deeper - [Modem on Wikipedia](https://en.wikipedia.org/wiki/Modem) - [Cable modem on Wikipedia](https://en.wikipedia.org/wiki/Cable_modem) --- ## Moore's law > The observation that chips fit about twice as many transistors every two years. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/moores-law/ - In Spanish: [Ley de Moore](https://tidbit.legios.com.ar/es/moores-law/) In 1965, Gordon Moore, who later co-founded Intel, noticed that the number of transistors that could be put on a chip was doubling at a steady pace. In 1975 he revised the pace to roughly every two years. That prediction, known as **Moore's law**, held up remarkably well for about half a century. It is not a law of physics. It is an observation about engineering and economics, and for decades the industry treated it as a target, planning each new generation of factories to hit the next doubling. The results compound quickly: doubling every two years means roughly a thousand times more transistors every twenty years. That is why a phone today outperforms the supercomputers of the early 1990s, and why computers kept getting cheaper and more capable at the same time. The pace has slowed. Transistors are now so small that each further step is harder, slower and far more expensive, and making them faster stopped being free around the mid-2000s because chips got too hot. The industry keeps progressing in other ways: more cores per chip, specialized chips like the GPU, stacking chips in 3D, and packaging several smaller chips together. So people argue about whether Moore's law is dead, but its effect, cheaper computing every year, still shapes the industry. ### An analogy Folding a sheet of paper in half again and again. Each fold only doubles it, which feels modest, but after a few dozen folds the stack would reach far past the clouds. ### Keep in mind - Transistors per chip doubled roughly every two years for decades. - An observation and an industry target, not a law of nature. - Compounding made computers vastly cheaper and more powerful. - It has slowed; progress now also comes from specialization and packaging. ### Go deeper - [Moore's law on Wikipedia](https://en.wikipedia.org/wiki/Moore%27s_law) - [Transistor on Wikipedia](https://en.wikipedia.org/wiki/Transistor) --- ## Motherboard > The main circuit board that every other part of a computer plugs into. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/motherboard/ - In Spanish: [Placa madre](https://tidbit.legios.com.ar/es/motherboard/) The **motherboard** is the large flat board inside a computer that connects everything else. The CPU sits in a socket on it, the RAM sticks clip into slots next to it, the SSD screws into its own slot, the graphics card plugs into a long connector, and the USB, audio and network ports on the back of a desktop are part of it. Thin copper lines printed on the board, called **traces**, carry the signals between all of them. Besides wiring, the board carries a few chips of its own: a **chipset** that manages traffic between the CPU and the slower parts, the circuits that turn the power supply's current into the precise voltages the CPU needs, and a small piece of firmware (**BIOS** or **UEFI**) that wakes the hardware up and starts the operating system when you press the power button. The motherboard decides what a computer can become. Its socket fixes which CPUs fit, its slots fix how much RAM and how many drives you can add, and its ports fix what you can plug in. In a laptop or a phone the same job is done by a smaller board where most parts are soldered directly on, which saves space but means you usually cannot upgrade them later. ### An analogy The street grid and plumbing of a city. Buildings (CPU, memory, storage) do the actual work, but the grid decides where they can stand, how they are connected and how big the city can grow. ### Keep in mind - Connects CPU, RAM, storage, graphics card and ports. - Its socket and slots decide which parts are compatible. - Holds the firmware (BIOS/UEFI) that starts the computer. - In laptops and phones, most parts are soldered on and cannot be swapped. ### Go deeper - [Motherboard on Wikipedia](https://en.wikipedia.org/wiki/Motherboard) - [Integrated circuit on Wikipedia](https://en.wikipedia.org/wiki/Integrated_circuit) --- ## Multimodal AI > AI that works with more than one kind of data at once: text, images, sound, video. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/multimodal-ai/ - In Spanish: [IA multimodal](https://tidbit.legios.com.ar/es/multimodal-ai/) **Multimodal AI** is AI that works with more than one kind of information, or 'mode', at the same time: text, images, audio, video. A multimodal model can look at a photo and answer a question about it, listen to your voice and reply out loud, read a chart in a PDF, or take a sketch and turn it into a working web page. Earlier systems handled each mode separately and chained them together: one program turned speech into text, a language model answered, another program read the answer aloud. Details got lost at each step, like tone of voice or what was in the picture. Modern multimodal models are trained on text, images and sound together, so they connect the word 'dog' with pictures of dogs and the sound of barking inside one model. Most current assistants, including ChatGPT, Claude and Gemini, accept images as well as text, and several also handle audio and video. That opens practical uses. Photograph a strange error message, a medication label or a broken part and ask what it means. Blind and low-vision people use apps that describe their surroundings through the phone camera. Companies process scanned forms, invoices and screenshots without retyping them. The usual caveat applies with extra force: models can misread an image, miss a small detail or describe something that is not there, so important readings still need a human look. ### An analogy Like the difference between texting a friend and having them next to you: in person you can point at things, show them a photo and hear each other's tone, and they understand much more of what you mean. ### Keep in mind - Handles several kinds of input or output in one model. - Common today: image plus text; many also do voice and video. - Trained on mixed data, so it links words with sights and sounds. - It can still misread images, so check what matters. ### Go deeper - [What is multimodal AI? (IBM)](https://www.ibm.com/think/topics/multimodal-ai) - [Multimodal learning on Wikipedia](https://en.wikipedia.org/wiki/Multimodal_learning) --- ## NAT > How a whole home or office shares one public IP address on the internet. - Level: Advanced - Topics: Networks - URL: https://tidbit.legios.com.ar/nat/ - In Spanish: [NAT](https://tidbit.legios.com.ar/es/nat/) **NAT** (network address translation) is what lets many devices share a single public IP address. At home, your phone, laptop and TV each get a **private** address like `192.168.0.12` from the router. Those addresses mean nothing on the internet; they are reused in millions of homes. Your internet provider gives the router just one **public** address, and the router translates between the two worlds. When your laptop sends a packet to a website, the router rewrites the sender from `192.168.0.12, port 51000` to its own public address and a port it picks, say `200.45.10.7, port 40001`, and writes that pairing in a table. When the answer comes back to port 40001, the router looks it up, rewrites the destination back to your laptop, and passes it in. Each conversation gets its own entry, so the phone and the laptop never get each other's replies. NAT exists because IPv4 has only about 4.3 billion addresses, far fewer than the devices online, and it stretched them for decades. A side effect is that devices behind NAT cannot be reached from outside unless they started the conversation, which works as a basic shield but complicates video calls, online games and running your own server (hence port forwarding). Providers now often add a second layer, carrier-grade NAT, and IPv6, with enough addresses for every device, removes the need for it. In the cloud, a NAT gateway lets private servers reach the internet without being exposed to it. ### An analogy An office with one public phone number and a receptionist. Employees call out from their extensions; the receptionist notes who called whom, and when the callback arrives at the main number, transfers it to the right extension. ### Keep in mind - Private addresses (like 192.168.x.x or 10.x.x.x) are reused everywhere and are not reachable from the internet. - The router rewrites addresses and ports, and keeps a table to route the answers back. - Devices behind NAT accept incoming traffic only for conversations they started. - It exists because IPv4 ran short of addresses; IPv6 makes it unnecessary. ### Go deeper - [Network address translation on Wikipedia](https://en.wikipedia.org/wiki/Network_address_translation) - [RFC 1918, the private address ranges](https://www.rfc-editor.org/rfc/rfc1918) --- ## Natural language processing > The field of AI that works with human language: reading, writing, translating. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/natural-language-processing/ - In Spanish: [Procesamiento del lenguaje natural](https://tidbit.legios.com.ar/es/natural-language-processing/) **Natural language processing** (NLP) is the branch of AI that deals with human language, the 'natural' kind we speak and write, as opposed to programming languages. Its goal is to get computers to do useful things with text and speech: translate it, summarize it, answer questions about it, sort it, or turn it into a spoken reply. Language is hard for machines because meaning depends on context. 'The movie was not bad' is mildly positive, even though it contains the word 'bad'. 'Bank' can be a river or a place for money. For decades, NLP relied on grammar rules written by linguists, then on statistics over large collections of text. Since the late 2010s it has been dominated by neural networks, especially the **transformer**, and today's large language models are its most visible product. Classic NLP tasks are still everywhere, often hidden: spam filters, autocomplete on your phone, machine translation, voice assistants that turn speech into text, search engines that understand what you meant, systems that read thousands of customer reviews and tell a company which ones are angry. Many of these tasks that used to need a separate specialized model can now be done by one general language model with the right prompt. ### An analogy NLP is to computers what learning a second language is to people: first memorizing grammar rules, then picking up patterns from lots of reading, and finally getting a feel for sarcasm and context. ### Keep in mind - Covers translation, summarizing, classifying, search and speech. - Hard because the same words change meaning with context. - Moved from hand-written rules to statistics to neural networks. - Large language models are the current, most general form of NLP. ### Go deeper - [Natural language processing on Wikipedia](https://en.wikipedia.org/wiki/Natural_language_processing) - [What is NLP? (IBM)](https://www.ibm.com/think/topics/natural-language-processing) --- ## Netflix > A streaming service, and the engineering it built to deliver video to the whole world. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/netflix/ - In Spanish: [Netflix](https://tidbit.legios.com.ar/es/netflix/) **Netflix** was founded in 1997 in California by Reed Hastings and Marc Randolph, renting DVDs by mail. In 2007 it started streaming video over the internet, and later began producing its own series and films. It makes money mainly from monthly **subscriptions**, and also from cheaper plans that show ads. To the viewer it is an app; behind the app is some of the best-known engineering on the internet. After a database failure stopped DVD shipments for days in 2008, Netflix moved its systems from its own data centers to AWS, finishing in 2016. It was one of the first big companies to bet everything on the cloud, splitting its software into many small services that can fail without taking down the whole site. To make sure they really could, its engineers wrote **Chaos Monkey**, a tool that switches off servers at random during work hours. That practice of breaking things on purpose to find weaknesses is now called **chaos engineering**. The video itself does not come from the cloud. Netflix built its own CDN, **Open Connect**: servers full of films that it gives to internet providers to install inside their networks, close to viewers, so a popular episode travels a short distance. Each title is also encoded in many qualities, and the app switches between them as your connection speeds up or slows down. Netflix has published much of this work as open source, and its recommendation system, which decides the rows you see on the home screen, is a classic example of machine learning in a product. ### An analogy A newspaper that, instead of printing everything in one city and shipping it by truck, puts small printing presses in every neighborhood's kiosk, so each morning the paper is printed right where it is read. ### Keep in mind - Started in 1997 renting DVDs by mail; streaming since 2007; earns mostly from subscriptions. - An early large company to move everything to the cloud, running on AWS. - Popularized chaos engineering with Chaos Monkey. - Its own CDN, Open Connect, places video servers inside internet providers. ### Go deeper - [About Netflix (official site)](https://about.netflix.com/en) - [Netflix Open Connect](https://openconnect.netflix.com/en/) - [Chaos engineering on Wikipedia](https://en.wikipedia.org/wiki/Chaos_engineering) --- ## Neural network > A pile of tiny math units that learns patterns from examples. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/neural-network/ - In Spanish: [Red neuronal](https://tidbit.legios.com.ar/es/neural-network/) A **neural network** is a program made of layers of simple units called neurons. Each neuron takes some numbers, multiplies them by its own **weights**, adds them up, and passes the result to the next layer. Alone, a neuron is trivial. Millions of them, stacked, can recognize a face or write a paragraph. Nobody programs the weights by hand. During **training**, the network sees examples (a photo and the word 'cat'), makes a guess, measures how wrong it was, and nudges every weight a tiny bit in the direction that would have made the guess better. Repeat billions of times. Large language models like Claude are neural networks of a specific shape (transformers) trained on text. Their 'knowledge' is nothing more than a very large set of adjusted weights. ### An analogy Tuning a radio with millions of knobs. You turn them a little, listen if the music got clearer, and keep the turns that helped. After enough rounds, it plays the station. ### Keep in mind - Neurons are simple; the power comes from stacking millions of them. - Training means adjusting weights to reduce errors on examples. - A trained model is just a file full of numbers (the weights). ### Go deeper - [3Blue1Brown: But what is a neural network?](https://www.3blue1brown.com/lessons/neural-networks) - [Neural network on Wikipedia](https://en.wikipedia.org/wiki/Neural_network_(machine_learning)) --- ## NFC > Very short-range wireless that works with a tap: contactless cards, phones, transit. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/nfc/ - In Spanish: [NFC](https://tidbit.legios.com.ar/es/nfc/) **NFC** (near-field communication) is a wireless technology that only works over a few centimeters, roughly the distance of a tap. It is what happens when you pay by holding a card or a phone against a terminal, go through a subway gate with a transit card, or open a door with a badge. Two devices come close, exchange a small amount of data, and that is it. One side can be a **tag** with no battery at all: a tiny chip and an antenna inside a card, a sticker or a key fob. The reader's own radio field powers the tag for an instant, just long enough for it to answer. Phones can act as a reader, as a tag, or both, which lets them pay, read a label on a museum display or start pairing with headphones just by touching them. The very short range is a feature, not a limitation. Because you have to bring the device almost into contact, it is hard to trigger a payment by accident, and someone across the room cannot easily listen in. Cards and phone wallets add their own protections on top: phones usually ask you to unlock them or confirm with your face or fingerprint, and each payment uses a one-time code instead of your real card number. NFC is slow and carries little data, so it is used to start things, not to move files. ### An analogy A handshake instead of a phone call. It only happens when you are close enough to touch, it is over in a second, and it is enough to say "it's me, go ahead". ### Keep in mind - Works only within a few centimeters, about the distance of a tap. - Tags in cards and stickers need no battery; the reader powers them. - Used for contactless payments, transit cards, badges and quick pairing. - The tiny range makes accidental or remote triggering hard. ### Go deeper - [NFC technology (NFC Forum)](https://nfc-forum.org/learn/nfc-technology/) - [Near-field communication on Wikipedia](https://en.wikipedia.org/wiki/Near-field_communication) --- ## NoSQL > Databases that store data in shapes other than tables with fixed columns. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/nosql/ - In Spanish: [NoSQL](https://tidbit.legios.com.ar/es/nosql/) **NoSQL** is an umbrella name for databases that do not use the classic relational model of tables, rows and fixed columns queried with SQL. The name is usually read as "not only SQL". They became popular in the late 2000s, when big web companies needed to store huge amounts of data spread over many servers, with flexible shapes and very fast reads and writes. There are four main families. **Document databases**, like MongoDB, store each record as a JSON-like document that can have its own fields. **Key-value stores**, like Redis or Amazon DynamoDB, work like a giant dictionary: give a key, get the value back, extremely fast. **Wide-column databases**, like Cassandra, store huge tables where each row can have different columns. **Graph databases**, like Neo4j, store things and the connections between them, ideal for social networks or fraud detection. The trade-off is that you design around how the data will be read. A relational database lets you ask almost any question later by joining tables; many NoSQL databases are fast and scale out easily precisely because they limit joins and sometimes accept that different copies may be briefly out of sync. Neither is better in general. Many companies use both: a relational database for orders and payments, and a NoSQL one for sessions, caches, catalogs or activity feeds. ### An analogy A relational database is a filing cabinet with identical printed forms. NoSQL is a set of other containers: folders where each document can look different, a key rack, a huge ledger, a corkboard with strings between photos. Each fits some jobs much better than the cabinet. ### Keep in mind - "Not only SQL": databases without the fixed-table model. - Four families: document, key-value, wide-column and graph. - Usually scale across many servers easily, often with fewer joins and looser consistency. - Often used alongside a relational database, not instead of it. ### Go deeper - [What is NoSQL? (AWS)](https://aws.amazon.com/nosql/) - [NoSQL on Wikipedia](https://en.wikipedia.org/wiki/NoSQL) --- ## NPU > A small AI engine inside phones and laptops that runs models fast on little battery. - Level: Intermediate - Topics: Hardware, AI - URL: https://tidbit.legios.com.ar/npu/ - In Spanish: [NPU](https://tidbit.legios.com.ar/es/npu/) An **NPU** (neural processing unit) is a part of a chip dedicated to running AI models. Modern phones and many recent laptops have one inside their main chip, next to the CPU and the GPU. It is what blurs the background in a video call, finds faces in your photo library, transcribes a voice note or runs a small language model without sending anything over the internet. Like a TPU, it is specialized in the multiplications that neural networks are made of, using low-precision numbers. The difference is the goal. A TPU lives in a data center and trains or serves giant models; an NPU lives in your device and runs small models for inference using as little energy as possible. The same task on the CPU would be slower and drain the battery faster. NPUs matter because of **on-device AI**: processing data on the device itself is quick, works offline and keeps private things like photos and voice off someone else's server. Manufacturers often advertise NPU speed in TOPS (trillions of operations per second), but that number says little on its own; what counts is whether the apps you use actually take advantage of it. Brand names vary too: Apple, for example, calls its NPU the Neural Engine. ### An analogy A pocket translator versus phoning an interpreter. The interpreter far away can handle anything, but the little device in your pocket answers instantly, works without signal and keeps the conversation to yourself. ### Keep in mind - A part of a phone or laptop chip made for running AI models. - Runs small models locally, quickly and on little battery. - Keeps data on the device, so some AI features work offline and more privately. - TOPS figures in ads are a rough guide, not a measure of real usefulness. ### Go deeper - [Neural processing unit on Wikipedia](https://en.wikipedia.org/wiki/Neural_processing_unit) - [What is a neural processing unit? (IBM)](https://www.ibm.com/think/topics/neural-processing-unit) --- ## NVIDIA > The company that sells the shovels of the AI gold rush. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/nvidia/ - In Spanish: [NVIDIA](https://tidbit.legios.com.ar/es/nvidia/) **NVIDIA** designs GPUs. For twenty years that meant graphics cards for gamers. Around 2012 researchers discovered those same chips were ideal for training neural networks, and NVIDIA leaned in hard: it built **CUDA**, the software layer that lets programmers use its GPUs for any kind of math, and it became the default for AI. Today its data center chips (H100, B200 and successors) are what OpenAI, Anthropic, Google and Meta use to train and run models. Demand is so high that access to NVIDIA GPUs is a geopolitical topic and a line item in national budgets. NVIDIA does not manufacture the chips itself: it designs them and TSMC in Taiwan fabricates them. Its moat is less the silicon than the software ecosystem everyone already wrote on top of CUDA. ### An analogy During a gold rush, the surest money is in selling shovels. NVIDIA sells the shovels (GPUs) and the only handle that fits them (CUDA). ### Keep in mind - Started with gaming graphics; now most revenue is AI data centers. - CUDA is the software lock-in that keeps developers on NVIDIA. - Fabless: designs chips, TSMC manufactures them. ### Go deeper - [NVIDIA on Wikipedia](https://en.wikipedia.org/wiki/Nvidia) - [NVIDIA data center platforms](https://www.nvidia.com/en-us/data-center/) --- ## OAuth and SSO > Log in once with an account you trust, and let it vouch for you in other apps. - Level: Intermediate - Topics: Security, Software - URL: https://tidbit.legios.com.ar/oauth-and-sso/ - In Spanish: [OAuth y SSO](https://tidbit.legios.com.ar/es/oauth-and-sso/) When an app offers **Sign in with Google**, it is using **OAuth**: a standard that lets one service act on your behalf in another without ever seeing your password. The app sends you to Google, you sign in there, and Google asks whether you agree to share, say, your name and email. If you say yes, Google hands the app a **token**: a temporary pass that works only for what you approved and can be revoked at any time. Those permissions are called **scopes**, and they are what the consent screen lists: 'read your calendar', 'see your email address'. Strictly, OAuth is about permission to access things; the part that tells the app who you are is **OpenID Connect**, a layer built on top of it. In practice both travel together, and people say OAuth for the whole flow. **SSO** (single sign-on) applies the same idea inside a company. Employees log in once to an identity provider, such as Okta, Microsoft Entra ID or Google Workspace, and from there get into the mail, the chat, the CRM and dozens of other tools without a separate password for each. Security improves because there is one place to enforce two-factor authentication, and when someone leaves the company, disabling one account closes every door at once. ### An analogy A wristband at a festival. You show your ID once at the entrance (the identity provider) and get a wristband (the token); each stage checks the wristband, not your ID, and the wristband only opens the areas your ticket includes. ### Keep in mind - With OAuth the app gets a limited, revocable token, never your password. - Scopes say exactly what the app may do; the consent screen lists them. - OpenID Connect adds identity (who you are) on top of OAuth. - SSO gives one login for many apps, and one switch to turn access off. ### Go deeper - [OAuth 2.0 (oauth.net)](https://oauth.net/2/) - [How OpenID Connect works (OpenID Foundation)](https://openid.net/developers/how-connect-works/) - [What is SSO? (AWS)](https://aws.amazon.com/what-is/sso/) --- ## Object storage > Files stored as whole named objects, reached over the web, almost without limit. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/object-storage/ - In Spanish: [Almacenamiento de objetos](https://tidbit.legios.com.ar/es/object-storage/) **Object storage** keeps data as **objects**: a whole file (a photo, a video, a backup, a CSV export) plus some metadata about it, saved under a unique name called a **key**, inside a container usually called a **bucket**. There are no real folders and no disks to manage; you save an object with a `PUT` request over HTTP and get it back with a `GET`. Amazon S3 made this model popular, and Google Cloud Storage, Azure Blob Storage and many others follow it. It is different from the two older kinds of storage. **Block storage**, like a laptop's disk or an EBS volume, is split into small blocks a computer formats and edits in place, which is fast for databases. **File storage** is the shared drive with folders that many machines can open at once. Object storage gives up editing in place (you replace the whole object) and in return scales to billions of objects, spreads copies across several data centers for durability and is cheap per gigabyte. That combination makes it the default home for things that are written once and read many times: website images and videos, app uploads, backups, logs and the files of a data lake. Most providers offer storage classes, from fast to archive, where colder data costs less to keep but more, or longer, to get back. ### An analogy A coat check. You hand over the whole coat and get a numbered ticket; to get it back you show the ticket. You cannot sew a button on while it is hanging there, but the coat check can hold an enormous number of coats and never loses one. ### Keep in mind - Data is stored as whole objects with a key and metadata, inside buckets. - Read and written over HTTP with simple requests like PUT and GET. - Objects are replaced, not edited in place, unlike block or file storage. - Cheap, durable and nearly unlimited: the home of backups, media and data lakes. ### Go deeper - [What is object storage? (AWS)](https://aws.amazon.com/what-is/object-storage/) - [What is object storage? (Google Cloud)](https://cloud.google.com/learn/what-is-object-storage) - [Object storage on Wikipedia](https://en.wikipedia.org/wiki/Object_storage) --- ## OLAP and OLTP > Two kinds of database work: many small transactions, or a few huge analytical questions. - Level: Advanced - Topics: Data - URL: https://tidbit.legios.com.ar/olap-and-oltp/ - In Spanish: [OLAP y OLTP](https://tidbit.legios.com.ar/es/olap-and-oltp/) **OLTP** (online transaction processing) is the work behind an app while people use it: create this order, charge this card, update this address. Each operation touches a handful of rows, there are thousands of them per second, and every one must be correct and safe even if two happen at once. Databases like PostgreSQL, MySQL, Amazon RDS or DynamoDB are built for this, usually storing data row by row with indexes to find a single record quickly. **OLAP** (online analytical processing) is the work of asking big questions about all that history: revenue by region and month for the last five years, or which products are bought together. A single query reads millions or billions of rows but only a few columns, and nobody minds waiting a few seconds. Data warehouses such as Snowflake, BigQuery or Amazon Redshift are built for this, storing data column by column (like Parquet files) so a query reads only what it needs and compresses well. The two are kept apart on purpose. Running a heavy report on the production OLTP database can slow down checkout for real customers, and a columnar warehouse is clumsy at updating one row at a time. So data is copied from OLTP systems into an OLAP one through ETL or streaming. Some newer systems, called **HTAP**, try to do both, but the split remains the default design in most companies. ### An analogy A supermarket checkout versus the head office. The checkout handles one customer at a time, fast and without mistakes; the office looks at a year of receipts to decide what to stock. You would not want the office doing its yearly analysis at the till. ### Keep in mind - OLTP: many small reads and writes, row storage, correctness under load. - OLAP: few large read queries over history, columnar storage, aggregations. - Data flows from OLTP systems into an OLAP warehouse through ETL or streaming. - Keeping them apart protects the live app from heavy reports. ### Go deeper - [OLAP vs OLTP (AWS)](https://aws.amazon.com/compare/the-difference-between-olap-and-oltp/) - [Online analytical processing on Wikipedia](https://en.wikipedia.org/wiki/Online_analytical_processing) - [Online transaction processing on Wikipedia](https://en.wikipedia.org/wiki/Online_transaction_processing) --- ## Open source > Software whose code anyone can read, use, change and share. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/open-source/ - In Spanish: [Código abierto](https://tidbit.legios.com.ar/es/open-source/) **Open source** software is software whose source code, the human-readable text that programmers write, is published for anyone to read, use, modify and redistribute. The rules are set by a license. Some (MIT, Apache) let you do almost anything, including selling it inside a closed product; others (GPL) require that if you distribute a modified version, you publish your changes too. It is not a niche. Linux runs most servers and every Android phone, Git tracks most of the world's code, Python and JavaScript are open source, and so are Kubernetes, PostgreSQL and the browser engine behind Chrome. Most of it lives on GitHub, where anyone can report a bug or propose a fix in a pull request, and a small group of maintainers decides what gets in. Companies fund it because shared infrastructure is cheaper than everyone building their own, and because an open project attracts users and talent. The business models are paid support, hosted versions (you can run PostgreSQL yourself for free, or pay AWS for Amazon RDS) and "open core", where the base is free and the extras are paid. In AI, "open-weight" models like Llama publish the trained model but not always the data or code that produced it. ### An analogy A recipe printed in the cookbook rather than kept in the chef's safe. Anyone can cook it, improve it and pass it on; the license says whether you must share your improvements. ### Keep in mind - "Free" here means the freedom to read and change the code; it is usually also free of charge, but that is a separate thing. - The license decides what you may do: MIT and Apache are permissive, GPL requires sharing changes. - It is maintained by volunteers and by paid engineers at large companies alike. - Open source does not mean no owner: maintainers decide what gets merged. ### Go deeper - [The Open Source Definition (OSI)](https://opensource.org/osd) - [Open Source Guides (GitHub)](https://opensource.guide/) - [Open-source software (Wikipedia)](https://en.wikipedia.org/wiki/Open-source_software) --- ## Open standard > A public rulebook anyone can follow, so products from different makers work together. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/open-standard/ - In Spanish: [Estándar abierto](https://tidbit.legios.com.ar/es/open-standard/) A **standard** is a written specification of how something must work: the shape of a plug, the format of a file, the messages two computers exchange. A standard is **open** when anyone can read it and build products that follow it without asking one company for permission. Typically it is written and maintained through a public process by a standards body, such as the W3C and WHATWG for the web, the IETF for internet protocols, the IEEE for Wi-Fi or ISO for many industrial formats, and it can be implemented royalty-free or under fair, published terms. Familiar examples are everywhere. **HTML** is open, so any company can build a web browser and any person can publish a web page that all of them can show. **USB** is published by an industry group, which is why one cable fits laptops, phones and printers from different brands. **PDF** started as Adobe's format, but Adobe published it and in 2008 it became the ISO 32000 standard, so countless programs can create and read PDFs. TCP/IP, HTTP and Unicode are open standards too. "Open" is a spectrum, not a yes or no. Some standards are free to read and free to implement; others require buying the document or paying patent licenses, and people disagree on whether those count. An open standard is also different from open source: the standard is the rulebook, while open source code is one particular implementation of it. And a **de facto standard**, a format everyone uses because one product became dominant, may still be controlled by a single company. The payoff of open standards is interoperability, competition and files that can still be read decades later. ### An analogy Like the rules of football. They are published for everyone, so any club can form a team, any company can make the balls and any stadium can host a match, and everything fits together without asking one owner for permission. ### Keep in mind - Anyone can read it and build products that follow it. - Usually maintained by a standards body through a public process. - HTML, USB, PDF, TCP/IP and Unicode are well-known examples. - Not the same as open source, and not the same as a de facto standard. ### Go deeper - [Open standard on Wikipedia](https://en.wikipedia.org/wiki/Open_standard) - [About W3C standards](https://www.w3.org/standards/about/) - [About USB-IF](https://www.usb.org/about) --- ## Open-weights models > AI models whose trained numbers anyone can download and run on their own machines. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/open-weights-models/ - In Spanish: [Modelos de pesos abiertos](https://tidbit.legios.com.ar/es/open-weights-models/) An **open-weights model** is an AI model whose weights, the billions of numbers learned in training, are published for anyone to download. With them you can run the model on your own computer or servers, inspect it and adapt it, instead of only renting access through someone else's API. Meta's Llama, Google's Gemma, Mistral's models and those from Chinese labs such as Alibaba (Qwen) and DeepSeek are well-known examples, and OpenAI released open-weights models too. 'Open weights' is not quite the same as 'open source'. Open-source software gives you the full recipe; most open-weights releases give you the finished dish but not the training data or all the code used to make it. Licenses also vary: some allow almost any use, others add conditions, for example for very large companies. In 2024 the Open Source Initiative published a definition of open-source AI that asks for much more than weights, and few popular models meet it. Why people use them: privacy, because data never leaves your machines; control over cost and versions, since nobody can change or retire the model under you; and freedom to fine-tune. Tools like Ollama and sites like Hugging Face make downloading and running them fairly easy, as long as you have enough memory. The trade-offs: you run and secure it yourself, the best closed models are often still more capable, and once weights are released they cannot be recalled, which means safety restrictions can be removed by whoever downloads them. ### An analogy A closed model is a restaurant: you order, you eat, you never see the kitchen. An open-weights model is a frozen dish you take home: you can heat it, add spices or serve it to friends, but the original recipe and ingredient list usually stay with the chef. ### Keep in mind - The trained weights are published and can be downloaded. - Runs on your own hardware, so data can stay private. - Usually not fully open source: training data is rarely included. - Licenses differ, and released weights cannot be taken back. ### Go deeper - [Open-source artificial intelligence on Wikipedia](https://en.wikipedia.org/wiki/Open-source_artificial_intelligence) - [The Open Source AI Definition (OSI)](https://opensource.org/ai/open-source-ai-definition) - [Models on Hugging Face](https://huggingface.co/models) --- ## OpenAI > The company that made ChatGPT and turned AI into a household topic. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/openai/ - In Spanish: [OpenAI](https://tidbit.legios.com.ar/es/openai/) **OpenAI** is the company that made AI something everyone talks about. It was founded in 2015 in San Francisco as a nonprofit research lab, with Sam Altman among its founders and now its CEO, and it builds the **GPT** family of large language models. On 30 November 2022 it released **ChatGPT**, a chat interface on top of GPT, and it became one of the fastest-adopted consumer products in history. For many people 'ChatGPT' is still the name of the whole category. What it sells: ChatGPT subscriptions for individuals and companies; an API where developers pay per token to use GPT models, image generation, speech-to-text (Whisper) and video generation (Sora) inside their own products; and enterprise deals. Microsoft is its largest partner and investor: OpenAI's models run on Microsoft's Azure cloud and power Copilot across Windows and Office. Its structure is unusual and has changed several times. OpenAI started as a nonprofit, created a for-profit subsidiary in 2019 to raise the money that training frontier models requires, and reorganized again in 2025 into a public benefit corporation still controlled by the nonprofit. The tension between its original mission ('ensure that AGI benefits all of humanity') and the commercial race has been a recurring source of news, including the brief removal and return of Altman as CEO in November 2023. Why it matters: it set the pace. GPT-3 showed that scaling up models produced surprising abilities, ChatGPT showed the public cared, and since then every big tech company has been racing to match it. OpenAI is also one of the largest buyers of NVIDIA GPUs and is building its own data centers through the Stargate project. ### An analogy The company that turned a lab curiosity into a product everyone could try, the way the first iPhone did for smartphones. Others make comparable phones now, but that launch is the one people remember. ### Keep in mind - Founded in 2015 as a nonprofit; today a for-profit structure under nonprofit control. - ChatGPT (2022) made LLMs mainstream; the GPT models power it and the API. - Microsoft is its main partner; its models run on Azure and inside Copilot. - Also makes image, speech and video models (DALL-E, Whisper, Sora). ### Go deeper - [openai.com](https://openai.com) - [OpenAI developer platform](https://platform.openai.com/docs) - [OpenAI on Wikipedia](https://en.wikipedia.org/wiki/OpenAI) --- ## Operating system > The base software that runs a device and lets every app share its hardware. - Level: Basic - Topics: Software, Hardware - URL: https://tidbit.legios.com.ar/operating-system/ - In Spanish: [Sistema operativo](https://tidbit.legios.com.ar/es/operating-system/) An **operating system** (OS) is the software that starts when you turn on a device and stays in charge until you turn it off. Windows, macOS, Linux, Android and iOS are operating systems. Its job is to manage the hardware, the CPU, the memory (RAM), the storage, the screen, the network, and to share it fairly among all the apps you run. Without an OS, every app would need to know how to talk to your exact disk, keyboard and Wi-Fi chip, and two apps could overwrite each other's memory. The OS hides those details behind a common set of services: "save this file", "draw this window", "send this over the network". It also keeps apps apart from each other, so a crashing app does not take the whole machine down, and asks your permission before one uses the camera or your location. The core of an operating system is called the **kernel**: the part that talks directly to the hardware and decides which program runs on the CPU at each moment. Around it come the things you actually see, like the desktop, the file browser, settings and built-in apps. That is why an app made for one OS usually does not run on another without changes. ### An analogy The manager of a busy shared kitchen. The cooks (apps) never fight over the stove directly: they ask the manager, who hands out burners, keeps each cook's ingredients separate and makes sure nobody burns the place down. ### Keep in mind - Examples: Windows, macOS, Linux, Android, iOS. - Shares the CPU, memory, storage and devices among all running apps. - Its core, the kernel, talks directly to the hardware. - Apps are written for a specific OS, which is why some only exist for one. ### Go deeper - [Operating system on Wikipedia](https://en.wikipedia.org/wiki/Operating_system) - [Kernel (operating system) on Wikipedia](https://en.wikipedia.org/wiki/Kernel_(operating_system)) --- ## Oracle > The company behind one of the most used business databases, now also a cloud provider. - Level: Basic - Topics: Tech companies, Data - URL: https://tidbit.legios.com.ar/oracle/ - In Spanish: [Oracle](https://tidbit.legios.com.ar/es/oracle/) **Oracle** is an American software company founded in 1977 by Larry Ellison, Bob Miner and Ed Oates. Its first and most famous product is **Oracle Database**, one of the first commercial relational databases using SQL. For decades it has been where banks, airlines, telecoms and governments keep their most important records: accounts, bookings, invoices, payroll. Around the database it built a large set of **business applications**: software for accounting, purchasing, human resources and supply chains, the kind of system often called ERP. It grew partly by buying other companies, such as PeopleSoft, NetSuite and Sun Microsystems, which brought it the Java programming language and the MySQL database. It also owns Cerner, a maker of electronic health record systems used by hospitals. Its traditional business model is **licenses plus support**: customers pay for the right to use the software and then a yearly fee for updates and help, which brings in steady income for many years. More recently it sells the same things as a cloud service, through **Oracle Cloud Infrastructure**, which rents servers and databases from Oracle's own data centers, and it has become a large supplier of computing capacity for training AI models. ### An analogy The company that makes the safe where a bank keeps its ledgers, and that also sells the accounting books, the clerks' desks and, lately, rents rooms in its own guarded building to put the safe in. ### Keep in mind - Founded in 1977; Oracle Database is one of the most widely used business databases. - Sells business applications (ERP, HR, finance) and owns Java and MySQL. - Makes money from licenses, yearly support fees and cloud subscriptions. - Oracle Cloud Infrastructure rents servers and databases, including capacity for AI. ### Go deeper - [About Oracle (official site)](https://www.oracle.com/corporate/) - [Oracle Corporation on Wikipedia](https://en.wikipedia.org/wiki/Oracle_Corporation) - [What is a database? (Oracle)](https://www.oracle.com/database/what-is-database/) --- ## Packet > The small chunk every piece of internet data is cut into before it travels. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/packet/ - In Spanish: [Paquete](https://tidbit.legios.com.ar/es/packet/) Nothing travels the internet in one piece. A photo, a web page, a second of a video call: everything is chopped into **packets**, small blocks of roughly 1,500 bytes at most, each sent on its own. A packet has a **header** (sender and destination IP address, ports, a sequence number, a few flags) and a **payload**, the actual slice of data. The receiving side puts the slices back in order. Cutting data up is what makes the internet work at all. Packets from thousands of conversations share the same cables, interleaved; a router handles each one independently, reading only the header; if one gets lost or corrupted, only that small piece is resent. Two packets of the same file can even take different routes and still arrive, because each one carries the full destination address. This design, called **packet switching**, is the opposite of an old phone call, where a line was reserved end to end. You meet packets when things go wrong. **Packet loss** is the share of packets that never arrive; a few percent is enough to make a video call stutter or a game lag, because TCP waits and resends. Tools like `ping` and `traceroute` send small test packets to measure latency and see which routers are on the way. Firewalls, VPNs and NAT all work by looking at, rewriting or wrapping packets. ### An analogy Moving a library by mail. You cannot ship the building, so you box the books, number every box, and the boxes may take different trucks. At the other end you shelve them back in order and ask for any box that went missing. ### Keep in mind - A packet = header (addresses, ports, sequence number) + payload (the data). - Packets from many conversations share the same cables; routers forward each one on its own. - Packet loss, not low bandwidth, is what usually makes calls stutter. - `ping` and `traceroute` are test packets that measure the path. ### Go deeper - [What is a packet? (Cloudflare)](https://www.cloudflare.com/learning/network-layer/what-is-a-packet/) - [Packet (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Packet) - [Network packet (Wikipedia)](https://en.wikipedia.org/wiki/Network_packet) --- ## Parquet > A file format that stores tables by column, compressed, so analytics reads far less. - Level: Advanced - Topics: Data - URL: https://tidbit.legios.com.ar/parquet/ - In Spanish: [Parquet](https://tidbit.legios.com.ar/es/parquet/) **Parquet** (Apache Parquet) is an open file format for tables, designed for analytics. A CSV writes a table row by row; Parquet writes it **column by column**: all the dates together, then all the amounts, then all the countries. It is binary rather than plain text, so you cannot open it in a text editor, but almost every data tool reads it: Spark, Amazon Athena, pandas, DuckDB and the major data warehouses. Storing by column pays off twice. First, a query that needs two columns out of fifty reads only those two and skips the rest of the file. Second, values in a column look alike (the same few countries repeated millions of times), so they compress very well, often to a fraction of the CSV size. Each file also records the minimum and maximum of every column for each block of rows, so a query for March can skip blocks that only contain January. Parquet also keeps the **schema** inside the file: column names and types, so a date stays a date and nobody has to guess. That is why it became the default format of the data lake: engines like Athena charge by data scanned, and converting raw logs from CSV or JSON to Parquet can make the same query both faster and much cheaper. The trade-off is that it is made for writing once and reading many times; changing a single row means rewriting a file. ### An analogy A filing cabinet with one drawer per kind of information instead of one folder per person. To add up everyone's salary you open only the salary drawer, and since it holds nothing but numbers, it packs tightly. ### Keep in mind - Columnar: a query reads only the columns it needs. - Similar values sit together, so files compress far better than CSV. - The schema travels inside the file, with real types. - Built for write-once, read-many analytics, not for editing single rows. ### Go deeper - [Parquet overview (Apache Parquet docs)](https://parquet.apache.org/docs/overview/) - [Apache Parquet on Wikipedia](https://en.wikipedia.org/wiki/Apache_Parquet) --- ## Passkeys > A way to sign in without a password, using your phone's or computer's lock instead. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/passkeys/ - In Spanish: [Passkeys](https://tidbit.legios.com.ar/es/passkeys/) A **passkey** replaces the password for a site or app. Instead of typing something you memorized, you unlock your phone or computer the way you normally do, with your fingerprint, your face or your device PIN, and you are in. Google, Apple, Microsoft, banks, shops and many other services support them. Underneath it uses public-key cryptography. When you create a passkey, your device makes a pair of keys: a **private key** that stays with you, and a **public key** that is sent to the site. To sign in, the site sends a random challenge, your device signs it with the private key after you unlock it, and the site checks the signature with the public key. Your fingerprint never leaves your device; it only unlocks the key. The big win is that there is nothing worth stealing on the site's side and nothing you can be tricked into typing. If the site's database leaks, the public keys are useless to an attacker. And each passkey is tied to the real web address it was created for, so a look-alike phishing site simply cannot ask for it. That is why passkeys are considered phishing-resistant, which ordinary passwords and SMS codes are not. Passkeys can live in the password manager of your phone or computer, such as Apple's or Google's, or in a third-party one, which syncs them across your devices; they can also be stored on a physical security key. If you lose your phone, you recover them from that account like any other synced data. The technology is an open standard, built on FIDO2 and WebAuthn, created by the FIDO Alliance and the W3C. ### An analogy Instead of whispering a password to the doorman, who could overhear it or be an impostor, you show a seal only your ring can make. The doorman has a copy of the pattern to check it, but cannot make the seal himself. ### Keep in mind - You sign in by unlocking your device; there is no password to remember or type. - The site only stores a public key, useless to anyone who steals it. - Each passkey works only on the real site, so phishing pages cannot capture it. - Synced by your device's password manager, or kept on a physical security key. ### Go deeper - [Sign in with a passkey instead of a password (Google)](https://support.google.com/accounts/answer/13548313?hl=en) - [About the security of passkeys (Apple)](https://support.apple.com/en-us/102195) - [Passkey on Wikipedia](https://en.wikipedia.org/wiki/Passkey_(authentication)) --- ## Password manager > An encrypted vault that remembers a different strong password for every site. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/password-manager/ - In Spanish: [Gestor de contraseñas](https://tidbit.legios.com.ar/es/password-manager/) A **password manager** is an app that creates, stores and types your passwords for you. You remember one long **master password**; the app remembers the hundreds of others, each one different, random and long (`v9!Qm#2kLp...`), and fills them in when you log in. The vault is protected with encryption, so in a well-designed manager the company that runs it cannot read your passwords either. The problem it solves is reuse. Sites get breached, and when one leaks its list of emails and passwords, attackers try those same pairs on every other service, automatically. If your email password is also your bank password, one leak opens both. With a manager, each site has its own key, so a leak stays contained to that site. It also helps against phishing: the manager fills a password only on the exact site it was saved for. If a fake page imitates your bank on a look-alike address, the autofill does nothing, and that silence is a warning. Popular options are the built-in ones from Apple, Google and browsers, and dedicated apps like Bitwarden (open source) or 1Password. Practical takeaway: start with the one already on your phone or browser. Make the master password a long passphrase of four or five random words, turn on two-factor authentication for the manager itself, and let it replace your reused passwords one site at a time. ### An analogy A keyring inside a safe. You memorize one combination; inside are a hundred different keys, one per door, and you never have to carry them in your head. ### Keep in mind - Reused passwords are the real risk: one leak opens every account that shares it. - You memorize one long master passphrase; everything else is generated and random. - Autofill refuses to work on a look-alike site, which blocks a lot of phishing. - The built-in managers from Apple, Google and browsers are a fine place to start. ### Go deeper - [Password manager on Wikipedia](https://en.wikipedia.org/wiki/Password_manager) - [OWASP Authentication Cheat Sheet](https://cheatsheetseries.owasp.org/cheatsheets/Authentication_Cheat_Sheet.html) --- ## Penetration testing > Hiring someone to break into your systems, with permission, before a criminal does. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/penetration-testing/ - In Spanish: [Pruebas de penetración](https://tidbit.legios.com.ar/es/penetration-testing/) **Penetration testing**, or **pentesting**, is an authorized, simulated attack on an organization's systems. A company hires security specialists, agrees in writing on what they may touch and when, and lets them try to get in the way a real attacker would: through the website, the office Wi-Fi, a forgotten server, or a convincing phone call to the help desk. At the end they deliver a report: what they reached, how, how serious it is, and how to fix it. The point is to find the gaps that checklists miss. An automated scanner can list software with a known vulnerability; a human tester notices that a minor bug on the login page, combined with a reused password and an overly broad permission, adds up to full access to customer data. That chain of small problems is usually how real breaches happen. What separates a pentest from an attack is consent and rules. The **scope** says which systems are in and which are off limits, the **rules of engagement** say what is not allowed (no deleting data, no testing during the payroll run), and the client knows it is happening. Tests come in flavors: **black box** (the tester starts knowing nothing, like an outsider), **white box** (they get the source code and diagrams) and gray box in between. A **red team** exercise is a longer, stealthier version that also tests whether the defenders notice. Pentests are often required by regulators, card payment rules or large customers, typically once a year and after big changes. They are a snapshot, not a guarantee: they show what one team found in a limited time. The value is in fixing what they found and retesting. ### An analogy Paying a locksmith to try to get into your house while you watch, then hand you a list: the back window latch is weak, the spare key under the plant is obvious, the side gate does not lock. ### Keep in mind - A pentest is a simulated attack with written permission and a defined scope. - Humans find chains of small weaknesses that automated scanners miss. - The deliverable is a report with findings, severity and fixes. - It is a snapshot in time; retest after fixing and after big changes. ### Go deeper - [Penetration testing (NIST glossary)](https://csrc.nist.gov/glossary/term/penetration_testing) - [OWASP Web Security Testing Guide](https://owasp.org/www-project-web-security-testing-guide/) - [Penetration test on Wikipedia](https://en.wikipedia.org/wiki/Penetration_test) --- ## Perplexity > An AI answer engine that searches the web and replies with a summary and its sources. - Level: Basic - Topics: Tech companies, AI - URL: https://tidbit.legios.com.ar/perplexity/ - In Spanish: [Perplexity](https://tidbit.legios.com.ar/es/perplexity/) **Perplexity** is an American company founded in 2022 in San Francisco by Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski. Its product, also called Perplexity, calls itself an **answer engine**: instead of returning a list of blue links like a classic search engine, it returns a written answer to your question. Under the hood it combines a search engine with a large language model. When you ask something, Perplexity searches the web, picks the most relevant pages, reads them and asks the model to write an answer based only on what it found, with **numbered citations** pointing to each source so you can check them. This technique is known as RAG (retrieval-augmented generation). It can use its own models or models from other companies, such as OpenAI and Anthropic, and lets paying users choose. It makes money mainly through subscriptions: a free version with limits and paid plans for individuals and companies, plus an API for developers. Like other AI search tools, it raises a fair question about the open web: some news publishers have objected to how it uses their articles, since readers may get the answer without ever visiting the page that produced it. ### An analogy A research assistant who, when you ask a question, goes to the library, skims the relevant books and comes back with a one-page summary that has footnotes on every claim. ### Keep in mind - Founded in 2022 in San Francisco; its product is also called Perplexity. - Searches the web, then has an LLM write an answer with numbered sources. - Uses its own models and models from other AI companies. - Makes money from subscriptions and an API for developers. ### Go deeper - [Perplexity Help Center (official)](https://www.perplexity.ai/help-center/en) - [Perplexity AI on Wikipedia](https://en.wikipedia.org/wiki/Perplexity_AI) --- ## Personal data and GDPR > Information about you, and the laws that say who can use it and how. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/personal-data-and-gdpr/ - In Spanish: [Datos personales y RGPD](https://tidbit.legios.com.ar/es/personal-data-and-gdpr/) **Personal data** is any information that identifies a person or can be linked to one: a name, an email, a phone number, an ID number, a home address, but also a photo, a location history, an IP address or the record of what you bought. Some kinds are treated as especially sensitive, like health, religion, political views or biometric data such as fingerprints. **Data protection laws** set the rules for organizations that collect and use that information. The best known is the European Union's **GDPR** (General Data Protection Regulation), in force since 2018. It says a company needs a valid reason to use your data (your consent, a contract, a legal duty), may use it only for the purpose it told you, should keep no more than it needs and must protect it. It gives you rights: to see what they hold about you, to correct it, to have it deleted, and to take it elsewhere. Companies must report serious data breaches quickly, and fines can be very large. It applies to anyone handling data of people in the EU, wherever the company is based. Many other places have their own laws built on similar ideas. **Argentina** has had a personal data protection law since 2000, with a public agency that oversees it and receives complaints; Brazil has the LGPD, California the CCPA, and most of Latin America has something comparable. That is why sites ask about cookies, why apps show privacy policies and why you can ask a company what it knows about you. For engineers it means designing systems that collect less, encrypt more and can actually delete a person on request. ### An analogy Your data is like a key to your house that you lend to a company. The law says they must tell you why they want it, use it only for that, keep it safe, and give it back when you ask. ### Keep in mind - Personal data is anything that identifies you, directly or combined with other data. - GDPR is the EU's law, in force since 2018, and it applies to anyone handling EU residents' data. - Core rights: see, correct, delete and take your data elsewhere. - Argentina and many other countries have their own data protection laws and authorities. ### Go deeper - [Data protection in the EU (European Commission)](https://commission.europa.eu/law/law-topic/data-protection_en) - [General Data Protection Regulation on Wikipedia](https://en.wikipedia.org/wiki/General_Data_Protection_Regulation) - [Personal data in Argentina (AAIP, in Spanish)](https://www.argentina.gob.ar/aaip/datospersonales) --- ## Phishing > A fake message that tricks you into handing over a password or a payment. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/phishing/ - In Spanish: [Phishing](https://tidbit.legios.com.ar/es/phishing/) **Phishing** is a message (email, SMS, WhatsApp, a phone call) that pretends to come from someone you trust, such as your bank, a courier, the tax office or your boss, and pushes you to act fast: click a link, log in, pay a small fee, open an attachment. The link goes to a copy of the real site that captures what you type. The attachment installs Malware. Nothing is hacked; you are convinced to open the door yourself. The classic example: 'Your package is on hold, pay $2.50 in customs fees' with a link to `correo-envios-ar.com`, a page that looks exactly like the courier. Targeted versions (**spear phishing**) use your name, your company and a believable story, like an invoice from a real supplier with a changed bank account. Most large breaches in the news started with one of these. The tells are consistent: urgency, a request for credentials or money, and an address that is almost right. A real company rarely needs you to confirm your password through a link, and the sender name in an email is just text anyone can write; the domain after the `@` is what counts. Practical takeaway: do not click, go. Open the bank or courier by typing its address or from its app and check there. A Password manager helps because it refuses to fill your password on a look-alike domain, and two-factor authentication with a passkey cannot be phished at all. If you already clicked, change that password and tell the service; it happens to careful people too. ### An analogy Someone in a convincing uniform knocking at your door, saying they are from the gas company and need to check the meter inside. They never pick the lock; they get you to open it. ### Keep in mind - Phishing targets people, not software: it convinces you to hand over the keys. - Urgency plus a request for a password or a payment is the pattern to recognize. - Check the domain after the @ and in the link, not the display name. - Do not click: open the site yourself, from your bookmarks or the app. ### Go deeper - [What is a phishing attack? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/access-management/phishing-attack/) - [Phishing on Wikipedia](https://en.wikipedia.org/wiki/Phishing) --- ## Ping > A tiny "are you there?" message that tests a connection and times the reply. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/ping/ - In Spanish: [Ping](https://tidbit.legios.com.ar/es/ping/) **Ping** is the simplest test of a network connection. Your computer sends a tiny message to another machine, basically "are you there?", and waits for the echo. If a reply comes back, the other side is reachable, and the time it took, shown in milliseconds, tells you how long a round trip takes. The name comes from the sound of sonar, which works the same way. Technically, ping sends an **echo request** using ICMP, a helper protocol of the internet meant for control and error messages, and the other machine answers with an **echo reply**. You can try it yourself: open a terminal and type `ping example.com`. Each line shows one reply and its time; if some replies never arrive, packets are being lost somewhere along the way. In everyday speech, "ping" also means that delay itself. Gamers complain about "high ping" when their moves reach the game server late, and speed tests show it next to download speed. A good ping depends on distance and on the network: a nearby server answers in a few milliseconds, while one on another continent takes much longer, because even light in a fiber needs time to travel. One caution: some servers and firewalls are set to ignore pings, so no answer does not always mean the machine is down. ### An analogy Shouting "hello?" into a cave and listening for the echo. If it comes back, you know something is there, and how long it takes tells you how far away the wall is. ### Keep in mind - Sends a small echo request and waits for the reply. - The round-trip time, in milliseconds, measures the connection's delay. - Missing replies point to packets being lost along the way. - No answer may just mean the server is set to ignore pings. ### Go deeper - [Ping (networking utility) on Wikipedia](https://en.wikipedia.org/wiki/Ping_(networking_utility)) - [Internet Control Message Protocol on Wikipedia](https://en.wikipedia.org/wiki/Internet_Control_Message_Protocol) --- ## Port > The numbered door on a computer that says which program a connection is for. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/port/ - In Spanish: [Puerto](https://tidbit.legios.com.ar/es/port/) An IP address gets a packet to the right computer; a **port** gets it to the right program on that computer. It is a number from 0 to 65535 that both TCP and UDP attach to every packet, next to the address. A server listens on a port, and anything arriving there is handed to the program behind it: a web server, a database, a mail server. Many ports are conventions everyone follows. `80` is HTTP, `443` is HTTPS, `22` is SSH (remote terminals), `53` is DNS, `25` is email between servers, `5432` is PostgreSQL. When you type `https://example.com`, the browser silently uses port 443; a URL like `http://localhost:3000` is how a developer reaches an app running on port 3000 on their own machine. Your side of the connection also gets a port, a random high number, so the reply knows which browser tab it belongs to. Ports are also where security starts. A firewall is, at its simplest, a list of which ports are open to whom: a web server exposes 443 and nothing else. An open port is a door that answers; scanning a server's ports is the first thing an attacker (or a security audit) does. Closing everything you do not use is the cheapest hardening there is. ### An analogy An office building with one street address and many numbered doors. The mail arrives at the building (the IP address) and the number on the envelope (the port) says which office gets it. ### Keep in mind - IP address = which machine; port = which program on that machine. - Well-known ports: 80 HTTP, 443 HTTPS, 22 SSH, 53 DNS. - A firewall decides which ports are reachable from outside. - The browser hides the port when it is the standard one for the scheme. ### Go deeper - [What is a computer port? (Cloudflare)](https://www.cloudflare.com/learning/network-layer/what-is-a-computer-port/) - [Port (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Port) - [Port, computer networking (Wikipedia)](https://en.wikipedia.org/wiki/Port_(computer_networking)) --- ## Primary and foreign keys > A primary key names each row uniquely; a foreign key points to a row in another table. - Level: Intermediate - Topics: Data - URL: https://tidbit.legios.com.ar/primary-and-foreign-keys/ - In Spanish: [Claves primarias y foráneas](https://tidbit.legios.com.ar/es/primary-and-foreign-keys/) In a relational database, a **primary key** is the column (or set of columns) that identifies each row uniquely. In a `customers` table it is usually an `id`: customer 42 is always that customer and no one else, even if two people share the same name. The database refuses a second row with the same primary key and does not allow it to be empty. A **foreign key** is a column in one table that holds the primary key of a row in another table. An `orders` table has a `customer_id` column, and each value in it points to a row in `customers`. That is how a database links data without copying it: the order does not repeat the customer's name and address, it just says "customer 42", and a SQL `JOIN` brings the two together when you need them. Foreign keys also protect consistency. If they are declared in the database schema, the database will not accept an order for customer 999 when there is no customer 999, and it can stop you from deleting a customer who still has orders. Many analytics systems skip these checks for speed, which is one reason data cleaning so often finds orphan rows. ### An analogy A national ID number is a primary key: one per person, never repeated. When a hospital file writes your ID number instead of copying your whole life story, that number is a foreign key pointing back to you. ### Keep in mind - Primary key: unique and never empty, one per row. - Foreign key: a column that refers to a primary key in another table. - Lets tables connect without duplicating data; joins follow these links. - Declared keys stop orphan records, like an order for a customer who does not exist. ### Go deeper - [Primary key on Wikipedia](https://en.wikipedia.org/wiki/Primary_key) - [Foreign key on Wikipedia](https://en.wikipedia.org/wiki/Foreign_key) - [Constraints, including keys (PostgreSQL docs)](https://www.postgresql.org/docs/current/ddl-constraints.html) --- ## Process node > The name of a generation of chip manufacturing, like "3 nm". A label, not a size. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/process-node/ - In Spanish: [Nodo de proceso](https://tidbit.legios.com.ar/es/process-node/) A **process node** is the name a chipmaker gives to one generation of its manufacturing technology: "7 nm", "5 nm", "3 nm" and so on. Each new node packs transistors more tightly, so a chip of the same size can hold more of them, run faster or use less power, usually some mix of the three. Companies like TSMC, Samsung and Intel compete hard to reach the next node first, and phone and computer makers advertise which node their chips are built on. Here is the catch: the number no longer measures anything on the chip. Decades ago it roughly matched the length of a part of the transistor called the gate. From the late 1990s on, names and physical sizes drifted apart, and today a "3 nm" chip has no feature that is 3 nanometers long; its key dimensions are several times larger. Transistors did keep shrinking and improving, largely by changing shape and going three-dimensional, but the names simply kept the old rhythm, each one roughly 0.7 times the previous. So read the node as a brand for a generation: useful for comparing one company's chips over time, much less so across companies. Intel's former "10 nm" was roughly as dense as TSMC's "7 nm", which is one reason Intel renamed its later nodes. For a real comparison, engineers look at **transistor density** (how many transistors fit in a square millimeter) and at measured speed and power. The node names are a shorthand for the story of Moore's law, not a ruler. ### An analogy Like clothing sizes. A "medium" from one brand is not the same as a "medium" from another, and neither is a measurement you could check with a tape. The label tells you where the item sits in the range, not its actual dimensions. ### Keep in mind - A node is a generation of chip manufacturing, named like "5 nm" or "3 nm". - Each new node fits more transistors in the same space, for more speed or less power. - The number is a label: no part of a "3 nm" chip is 3 nm long. - Compare chips by transistor density and measured results, not by node names. ### Go deeper - [3 nm process on Wikipedia](https://en.wikipedia.org/wiki/3_nm_process) - [Semiconductor device fabrication on Wikipedia](https://en.wikipedia.org/wiki/Semiconductor_device_fabrication) - [A better way to measure progress in semiconductors (IEEE Spectrum)](https://spectrum.ieee.org/a-better-way-to-measure-progress-in-semiconductors) --- ## Programming language > A precise, written language people use to give instructions to a computer. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/programming-language/ - In Spanish: [Lenguaje de programación](https://tidbit.legios.com.ar/es/programming-language/) A **programming language** is a set of words, symbols and rules for writing instructions that a computer can follow. Computers only understand very simple operations expressed as numbers; a language like Python or JavaScript lets a person write `if total > 100: apply_discount()` instead, and a tool translates that into something the machine can run. Unlike English or Spanish, a programming language leaves no room for interpretation. Every line has exactly one meaning, and a missing comma can stop a program from running. That strictness is the point: the computer does exactly what it is told, so the instructions have to be exact. There are hundreds of languages because they make different trade-offs. Some are built for speed and control over the hardware (C, Rust), some for the web (JavaScript), some for being easy to read and quick to write (Python), and some for one narrow job, like querying a database with SQL. Most working programmers know a few and pick the one that suits the task. ### An analogy A recipe written for a cook who follows it literally and has no common sense. "A pinch of salt" will not do; you must say how many grams, in which bowl, at which step. ### Keep in mind - It lets humans write instructions in readable text instead of raw numbers. - Every statement has one exact meaning; small typos cause errors. - A compiler or an interpreter turns the text into something the machine runs. - Different languages suit different jobs: web pages, data, apps, hardware. ### Go deeper - [Programming language on Wikipedia](https://en.wikipedia.org/wiki/Programming_language) - [History of programming languages on Wikipedia](https://en.wikipedia.org/wiki/History_of_programming_languages) --- ## Prompt > Everything you send to an AI model: the question, the instructions, the context. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/prompt/ - In Spanish: [Prompt](https://tidbit.legios.com.ar/es/prompt/) A **prompt** is everything you send to an LLM before it answers: your question, any instructions ('answer in three bullet points'), examples of what you want, and material to work with (an email to summarize, a table to read). The model has no memory of you beyond what is in the prompt, so for that request the prompt is its whole world. Small changes in the prompt change the answer a lot. 'Write a summary' and 'Write a two-sentence summary for a busy doctor, no jargon' produce very different text. Saying who it is for, what format you want, how long, and what to avoid gets better results than hoping the model guesses. This craft is called **prompt engineering**, though most of it is just clear writing. Products built on LLMs add a hidden part, the **system prompt**: instructions from the developer that sit before yours and set the tone, the rules and the persona. When a chatbot answers in a particular style or declines something, the system prompt is often the reason. Why it matters: with the same model, the quality of what you get depends mostly on what you ask. Learning to write a good prompt is the cheapest way to get more out of any AI tool. ### An analogy Briefing a very capable new colleague who has read everything but knows nothing about you or today's task. The clearer the brief, the better the work. ### Keep in mind - The prompt is all the model knows about your request: say what you want, for whom, in what format. - Showing one or two examples often beats describing the format. - System prompts are hidden instructions from the developer that shape every answer. - Prompt and answer are both measured in tokens, which is what you pay for. ### Go deeper - [Anthropic: prompt engineering overview](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview) - [Prompt engineering on Wikipedia](https://en.wikipedia.org/wiki/Prompt_engineering) --- ## Prompt engineering > Writing instructions for an AI model so it gives you what you actually need. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/prompt-engineering/ - In Spanish: [Ingeniería de prompts](https://tidbit.legios.com.ar/es/prompt-engineering/) **Prompt engineering** is the practice of writing the instructions you give an AI model (the prompt) so that it does the job well. The same model can give a bland, generic answer or a sharp, useful one depending on how it was asked. 'Engineering' sounds grand; most of it is clear writing plus some trial and error. The techniques that work are a lot like briefing a smart new colleague who knows nothing about your situation. Say what the goal is and who the result is for. Give the context it cannot guess: the audience, the constraints, the document to work from. Show one or two examples of what good looks like. Say what form you want back: a table, three bullet points, under 100 words, in Spanish. For tricky problems, asking the model to work through the steps before answering often helps. It matters at two levels. For everyday use, it is the difference between 'write a post about our product' and a request that gets something you can actually publish. For companies building products on top of models, the prompt is part of the software: a carefully tested set of instructions, often hidden from the user as a **system prompt**, decides how the assistant behaves. As models get better at understanding plain requests, fewer tricks are needed, but clear instructions and good context still make the biggest difference. ### An analogy Like ordering at a tailor instead of a clothing rack: 'a suit' gets you something generic, but your measurements, the occasion and a photo of a cut you like get you something that fits. ### Keep in mind - Clear goal, relevant context and an example beat clever wording. - Saying the format you want back saves a lot of rewriting. - In products, the system prompt sets how an assistant behaves. - Good prompts reduce mistakes but do not make the model infallible. ### Go deeper - [Prompt engineering overview (Claude docs)](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview) - [Prompt engineering guide (OpenAI)](https://platform.openai.com/docs/guides/prompt-engineering) - [Prompt engineering on Wikipedia](https://en.wikipedia.org/wiki/Prompt_engineering) --- ## Proxy > A middleman computer that makes requests on someone else's behalf. - Level: Intermediate - Topics: Networks, Security - URL: https://tidbit.legios.com.ar/proxy/ - In Spanish: [Proxy](https://tidbit.legios.com.ar/es/proxy/) A **proxy** is a server that sits between two parties and passes messages along. Instead of your browser talking to a website directly, it asks the proxy, the proxy asks the website, and the answer comes back the same way. To the website, the visitor is the proxy: it sees the proxy's IP address, not yours. The classic kind, a **forward proxy**, works for the people inside a network. Companies and schools use one so all browsing goes through a single door, where they can block certain sites, scan downloads for malware and keep logs. People use public proxies to appear to be in another country or to hide their address from the sites they visit. A **reverse proxy** works for the servers instead. It stands in front of a website and receives every visitor, then decides which server behind it should answer. That is where TLS certificates are often handled, where repeated content is cached, and where abusive traffic gets filtered. Load balancers and CDNs are reverse proxies with a specific job; tools like NGINX are commonly used to build one. A proxy is not a VPN. A proxy is usually set up for one app, like the browser, and by itself does not encrypt anything; a VPN tunnels all the traffic of your device through an encrypted connection. Either way, the operator of the middleman can see where you go, and with plain HTTP also what you send, so a free proxy run by strangers is a bad place to log in to anything. ### An analogy Asking a friend to buy something for you at a shop. The shop only sees your friend, and your friend knows exactly what you bought. ### Keep in mind - The other side sees the proxy's IP address, not yours. - Forward proxies serve users; reverse proxies stand in front of servers. - Load balancers and CDNs are specialized reverse proxies. - A proxy is not a VPN, and the operator can see your traffic. ### Go deeper - [Proxy server on Wikipedia](https://en.wikipedia.org/wiki/Proxy_server) - [Reverse proxy on Wikipedia](https://en.wikipedia.org/wiki/Reverse_proxy) - [NGINX reverse proxy (official docs)](https://docs.nginx.com/nginx/admin-guide/web-server/reverse-proxy/) --- ## Public-key cryptography > Two linked keys: one you share with everyone to lock, one you keep to unlock. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/public-key-cryptography/ - In Spanish: [Criptografía de clave pública](https://tidbit.legios.com.ar/es/public-key-cryptography/) **Public-key cryptography** uses a mathematically linked **key pair**. What the **public key** locks, only the **private key** opens, and the public key cannot be used to work out the private one. So you can publish your public key anywhere, and anyone can send you something only you can read. The older kind, symmetric encryption, uses one shared key for both directions, which leaves a hard problem: how do two strangers agree on a secret over a network everyone can listen to? That is the problem it solves, and it is why the internet works. When your browser opens an HTTPS site, it uses the server's public key (delivered inside its TLS certificate) to agree on a fresh symmetric key, and the rest of the session runs on that faster key. The same idea protects your `ssh` login to a server (your public key sits on the server, your private key never leaves your laptop), the messages in end-to-end encryption apps, cryptocurrency wallets and passkeys. The pair also works backwards, as a **digital signature**: something locked with your private key can be opened by anyone with your public key, which proves it came from you and was not changed. Software updates, Git commits and certificates are signed this way. Common algorithms are RSA and elliptic-curve ones like Ed25519; the latter are smaller and faster and are now the default. The one rule that matters in daily life: the private key is the whole secret. Whoever has the file (or the seed phrase of a crypto wallet) is you, as far as the math is concerned. ### An analogy A mailbox with a slot. Anyone can drop a letter in (the public key), but only the owner's key opens the door (the private key). A signature is the reverse: a wax seal only you can press, which anyone can recognize. ### Keep in mind - A key pair: public to lock, private to unlock, and the public one does not reveal the private one. - It lets two strangers agree on a secret over a network anyone can listen to. - Reversed, it makes digital signatures: proof of who sent something and that it was not altered. - The private key is the whole secret; whoever holds it is you. ### Go deeper - [How does public key encryption work? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/ssl/how-does-public-key-encryption-work/) - [Public-key cryptography on Wikipedia](https://en.wikipedia.org/wiki/Public-key_cryptography) --- ## Python > A popular programming language known for being easy to read and quick to write. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/python/ - In Spanish: [Python](https://tidbit.legios.com.ar/es/python/) **Python** is a general-purpose programming language created by Guido van Rossum and first released in 1991. Its main idea is that code is read far more often than it is written, so it should look clean: a Python program uses plain words like `for`, `in` and `if`, and indentation (the spaces at the start of a line) to show which lines belong together. It is used almost everywhere: data analysis, AI and machine learning, automating boring tasks with small scripts, web servers, scientific research and teaching. Much of its strength comes from its huge collection of free libraries, ready-made code you can install with one command, for things like reading spreadsheets, drawing charts or training neural networks. Python is usually run by an **interpreter**, a program that reads your code and executes it line by line, instead of being compiled ahead of time. That makes it fast to try things out but slower than languages like C for heavy number crunching; in practice the heavy parts of popular Python libraries are written in faster languages underneath. ### An analogy A well-written instruction manual with big headings and clear indentation: you can tell at a glance which steps belong to which section, even if you did not write it. ### Keep in mind - Created by Guido van Rossum; first released in 1991. The name comes from Monty Python, not the snake. - Indentation is part of the language: it defines which lines go together. - Dominant in data science and AI thanks to its libraries. - Free and open source, run by the Python Software Foundation. ### Go deeper - [About Python (python.org)](https://www.python.org/about/) - [The official Python tutorial](https://docs.python.org/3/tutorial/index.html) - [Python on Wikipedia](https://en.wikipedia.org/wiki/Python_(programming_language)) --- ## QR code > A square pattern of dots that a phone camera reads, usually to open a link. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/qr-code/ - In Spanish: [Código QR](https://tidbit.legios.com.ar/es/qr-code/) A **QR code** ("quick response") is a two-dimensional barcode: a square grid of black and white modules that stores text, most often a web address. It was invented in 1994 by Masahiro Hara at Denso Wave, a Japanese company of the Toyota group, to track car parts in factories. Denso Wave holds the patent but chose not to enforce it, so anyone can create and read QR codes for free, which helped them spread everywhere. The three big squares in the corners tell the camera where the code is and which way it is turned, so it can be read at any angle. The data is stored with **error correction**: extra information that lets the code be read even when part of it is dirty, torn or covered, up to roughly a third of it at the highest setting. That is why some codes can have a logo in the middle. Today any phone camera can read them, and they are used for restaurant menus, payments, boarding passes, Wi-Fi passwords and tickets. The risk is that a QR code is unreadable to people: you cannot tell where it leads just by looking. Scammers print fake codes and stick them over real ones on parking meters, posters or restaurant tables, sending you to a copy of a payment or login page to steal your card or password. This is sometimes called **quishing**, a mix of QR and phishing. Before opening a link from a code, check the address your phone shows, be suspicious of stickers on top of other codes, and pay through the official app or website when you can. ### An analogy A sealed envelope with an address written in a language only phones can read. Convenient, but if someone swaps the envelope, you won't notice until you've already gone to the wrong place. ### Keep in mind - Invented in 1994 at Denso Wave to track car parts; free to use. - Error correction lets it work even when part of it is damaged. - Most codes just hold a link; the phone opens it for you. - Fake codes stuck over real ones are a common scam: check the link first. ### Go deeper - [History of QR Code (Denso Wave)](https://www.qrcode.com/en/history/) - [QR code on Wikipedia](https://en.wikipedia.org/wiki/QR_code) --- ## Qualcomm > Designs the chips, and owns key patents, that let phones talk to cell towers. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/qualcomm/ - In Spanish: [Qualcomm](https://tidbit.legios.com.ar/es/qualcomm/) **Qualcomm** is an American company founded in 1985 in San Diego by Irwin Jacobs and partners. It became known for CDMA, a way of sharing radio frequencies among many phones at once, which ended up at the heart of 3G networks. Since then it has been one of the main inventors behind each generation of mobile technology, 4G and 5G included. It makes money in two ways. The first is **chips**. Its **Snapdragon** line puts a whole phone's brain on a single chip: the CPU (based on Arm designs), the GPU for graphics, an AI accelerator, the camera processor and the **modem**, the part that talks to cell towers. Many Android phones run on Snapdragon, and Qualcomm also sells modems on their own, plus chips for cars, laptops, headsets and connected devices. Like NVIDIA or Apple, it is **fabless**: it designs the chips and companies such as TSMC and Samsung manufacture them. The second is **licensing**. Qualcomm owns a large portfolio of patents that are essential to cellular standards, so almost anyone who makes a phone that connects to 3G, 4G or 5G pays it a royalty, even if they use another company's chips. Licensing is a smaller share of its sales than chips, but a very profitable one, and it has led to long legal disputes with phone makers and regulators over how much it can charge. ### An analogy A company that both builds car engines and owns the patent on the wheel. Some carmakers buy its engines; nearly all of them pay it something for the wheels. ### Keep in mind - Founded in 1985 in San Diego; a key inventor of 3G, 4G and 5G technology. - Its Snapdragon chips power many Android phones, with CPU, GPU and modem in one piece. - Fabless: it designs chips and others, like TSMC, manufacture them. - Earns royalties from patents essential to cellular standards. ### Go deeper - [About Qualcomm (official site)](https://www.qualcomm.com/company/about) - [Qualcomm on Wikipedia](https://en.wikipedia.org/wiki/Qualcomm) --- ## Quantum computer > A machine that uses quantum physics for a few specific problems. Not a faster PC. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/quantum-computer/ - In Spanish: [Computadora cuántica](https://tidbit.legios.com.ar/es/quantum-computer/) A **quantum computer** stores information in **qubits** instead of ordinary bits. A normal bit is either 0 or 1. A qubit can be in a **superposition**, a blend of both, and several qubits can be **entangled**, so their states are linked. A quantum algorithm arranges those states so that, when the qubits are finally measured, wrong answers tend to cancel each other out and the right one is likely to appear. It does not "try every answer at once", as is often said; it relies on a very particular kind of interference. That trick only helps with certain problems. The best known are simulating molecules and materials, which are quantum systems themselves, and factoring very large numbers, which would break much of today's public-key cryptography. For everyday tasks like email, spreadsheets, games or running a website, a quantum computer offers no advantage and would do far worse than a laptop. It is a specialized tool, not a faster replacement for normal computers. Today's machines are real but small and fragile. Qubits lose their state with the slightest heat, vibration or stray signal, so many designs are cooled to a fraction of a degree above absolute zero, and they still make frequent errors. Building enough reliable, error-corrected qubits to run useful algorithms at scale is the central unsolved problem, and nobody can honestly say when it will be solved. Security experts are not waiting: NIST has already published **post-quantum cryptography** standards, new algorithms designed to resist future quantum attacks. ### An analogy Less a faster car than a submarine. For most trips it is useless and expensive, but for a few places nothing else can reach, it is the only vehicle that works. ### Keep in mind - Uses qubits, which can be in superposition and entangled with each other. - Helps with specific problems like simulating molecules or factoring, not everyday tasks. - Current machines are small, error-prone and need extreme isolation, often extreme cold. - Post-quantum cryptography is already being adopted to protect today's data. ### Go deeper - [Quantum computing on Wikipedia](https://en.wikipedia.org/wiki/Quantum_computing) - [What is quantum computing? (IBM)](https://www.ibm.com/think/topics/quantum-computing) - [Post-quantum cryptography (NIST)](https://csrc.nist.gov/projects/post-quantum-cryptography) --- ## RAG > Let the model look things up in your documents before it answers. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/rag/ - In Spanish: [RAG](https://tidbit.legios.com.ar/es/rag/) **RAG** (Retrieval-Augmented Generation) is a way to make an LLM answer from documents it never saw during training: your company's manuals, last week's contracts, a product catalog. Instead of retraining the model, you fetch the relevant pages at question time and paste them into the prompt. The model then writes its answer with those pages in front of it. It works in two steps. First, **retrieval**: the documents are split into chunks and each chunk is turned into embeddings, numbers that capture its meaning, stored in a database. When a question arrives, the system finds the chunks closest in meaning to it. Second, **generation**: the question plus those chunks go to the LLM, which answers and can say where it found each thing. Why it matters: an LLM's knowledge is frozen at training time, and it knows nothing private. RAG fixes both without the cost of training, keeps the data where it already lives, and makes hallucinations rarer because the model is grounded in real text. It is the most common architecture behind 'chat with your documents' products and internal company assistants. Example: a support bot for an airline. The question 'can I bring my cat?' retrieves the pet policy page, and the model answers from that page instead of guessing what airlines usually do. ### An analogy An open-book exam. The student (the model) is smart but did not memorize your files; you hand them the three relevant pages and they answer from those. ### Keep in mind - RAG adds knowledge at question time, through the prompt, not by retraining. - Retrieval usually relies on embeddings and a vector search. - Answers can cite their sources, which makes them easier to check. - Quality depends on retrieval: if the wrong chunks come back, the answer is wrong too. ### Go deeper - [Retrieval-augmented generation on Wikipedia](https://en.wikipedia.org/wiki/Retrieval-augmented_generation) - [AWS: what is RAG?](https://aws.amazon.com/what-is/retrieval-augmented-generation/) --- ## RAM > The computer's short-term memory: fast, small and wiped when you turn it off. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/ram/ - In Spanish: [RAM](https://tidbit.legios.com.ar/es/ram/) **RAM** (Random Access Memory) is where a computer keeps whatever it is working on right now: the open programs, the document you are editing, the browser tabs. The CPU can read and write any part of it in nanoseconds, far faster than an SSD. The catch is that it is volatile: cut the power and everything in it is gone, which is why you save files to disk. For most people the amount matters more than the speed. If the programs you have open need more RAM than you have, the operating system starts shuffling parts of them to the SSD (swapping) and everything crawls. That is the classic 'my laptop is slow with 40 tabs open' problem. Phones have roughly 6 to 12 GB, laptops 8 to 32, servers hundreds. RAM comes as chips soldered onto the board or on removable sticks. Samsung, SK Hynix and Micron make most of the memory chips in the world. GPUs have their own, faster kind (VRAM, and HBM in AI chips), and for AI the question 'does the model fit in memory?' is usually about that. Example: an LLM with 7 billion parameters needs roughly 14 GB just to hold its weights. On a laptop with 16 GB of RAM it barely fits; on one with 8 GB it does not run at all. Memory, not speed, is the first wall you hit. ### An analogy Your desk versus your filing cabinet. The desk (RAM) holds the papers you are working on, within reach; the cabinet (SSD) holds everything else. A bigger desk lets you work on more things at once, and when you leave, the desk is cleared. ### Keep in mind - Fast and volatile: holds running programs, forgets everything when powered off. - Too little RAM forces swapping to disk, the usual cause of a sluggish computer. - Measured in gigabytes; more RAM lets you keep more open at once. - AI models must fit in memory (GPU VRAM or system RAM) to run at all. ### Go deeper - [Random-access memory on Wikipedia](https://en.wikipedia.org/wiki/Random-access_memory) - [Memory hierarchy on Wikipedia](https://en.wikipedia.org/wiki/Memory_hierarchy) --- ## Raspberry Pi > A cheap, credit-card-sized computer made for learning and tinkering. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/raspberry-pi/ - In Spanish: [Raspberry Pi](https://tidbit.legios.com.ar/es/raspberry-pi/) A **Raspberry Pi** is a small, inexpensive computer built on a single board about the size of a credit card. It has a processor, memory, USB ports, a video output and a row of pins for connecting electronics. Add a power supply, a microSD card with an operating system (usually a version of Linux), a keyboard and a screen, and it works as a basic desktop computer. It was created in the UK by the Raspberry Pi Foundation, an educational charity, and first went on sale in 2012 to help children learn programming and how computers work inside. It quickly became a favorite of hobbyists, schools and engineers too. People use it as a home media center, a retro game console, a weather station, a small home server, the brain of a robot or the controller of a sensor in a factory. What makes it special is that row of **GPIO pins** (general-purpose input/output). From a short Python program you can switch on a light, read a temperature sensor or drive a motor, which turns code into something you can touch. It will not replace a laptop for heavy work, but it is a cheap and safe place to experiment: if you break the setup, you rewrite the memory card and start again. ### An analogy A box of LEGO for computing. It is a real computer, but small and cheap enough that you can wire it to things, take it apart and get it wrong without worrying. ### Keep in mind - A full computer on one small board, usually running Linux from a microSD card. - Made by the Raspberry Pi Foundation, a UK charity, to teach computing. - Its GPIO pins let programs control lights, sensors and motors. - Common uses: learning, home servers, media centers and IoT projects. ### Go deeper - [Raspberry Pi Foundation (official)](https://www.raspberrypi.org/) - [Raspberry Pi on Wikipedia](https://en.wikipedia.org/wiki/Raspberry_Pi) --- ## Refactoring > Reorganizing code to make it clearer, without changing what it does. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/refactoring/ - In Spanish: [Refactoring](https://tidbit.legios.com.ar/es/refactoring/) **Refactoring** means changing the internal structure of code without changing its behavior. From the outside nothing is different: same screens, same results, same bugs, even. Inside, the code becomes easier to read and to change. Typical moves are giving a variable a clearer name, splitting a 300-line function into small ones with meaningful names, removing duplicated code, or moving a piece to where it logically belongs. The idea was popularized by Martin Fowler's book Refactoring, first published in 1999, which catalogued dozens of these small transformations. The key rule is to go in **small steps** and keep the program working after each one. That is only safe with good automated tests: you run them after every step, and if they still pass, you know you did not break anything. Teams refactor because code that is easy to understand is cheap to change, and software is changed constantly. It is the main way to pay down technical debt. A common habit is to clean up the area you are about to work on before adding a new feature there, sometimes summed up as "leave the campground cleaner than you found it". What refactoring is not: rewriting everything from scratch, or sneaking new features in at the same time. ### An analogy Like reorganizing a kitchen: same pots, same recipes, same meals come out. But now the spices are labeled and next to the stove, so cooking tomorrow is faster and fewer things get spilled. ### Keep in mind - Changes how code is organized, never what it does. - Done in small steps, running tests after each one. - Makes later changes faster and safer; it pays down technical debt. - Mixing refactoring with new features in one change makes both harder to review. ### Go deeper - [Refactoring (Martin Fowler's site)](https://refactoring.com/) - [Code refactoring on Wikipedia](https://en.wikipedia.org/wiki/Code_refactoring) --- ## Regex > A compact pattern language for finding and checking pieces of text. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/regex/ - In Spanish: [Regex](https://tidbit.legios.com.ar/es/regex/) A **regex** (regular expression) is a short pattern that describes a shape of text, so a program can find every piece of text with that shape. Instead of searching for one exact word, you describe what you want: "four digits, a dash, two digits". In regex that is written `\d{4}-\d{2}`, where `\d` means any digit and `{4}` means exactly four of the previous thing. Regexes are everywhere text is processed. A signup form uses one to check that an email looks like an email. A spreadsheet or code editor uses them for powerful find-and-replace, like turning every date from `2026-10-17` into `17/10/2026` in one go. Programmers use them to pull order numbers out of emails, clean up messy data or scan logs for errors. Almost every programming language and many text tools support them. Their power comes with a reputation for being hard to read. A pattern like `^[\w.+-]+@[\w-]+\.[\w.]+$` is dense, and small mistakes match too much or too little. Good practice is to keep patterns simple, test them against real examples with tools like regex101, and not use them for jobs they handle badly, such as fully parsing HTML. ### An analogy Searching a phone book not for a name but for a description: "every surname that starts with Mac, followed by any letters, ending in -son". The regex is that description written in a tiny, strict shorthand. ### Keep in mind - Describes a pattern of text, not one fixed word. - Used to search, validate (like emails) and find-and-replace. - Symbols like `\d` (digit) and `+` (one or more) build the pattern. - Powerful but easy to get subtly wrong; test patterns on real data. ### Go deeper - [Regular expressions guide (MDN)](https://developer.mozilla.org/en-US/docs/Web/JavaScript/Guide/Regular_expressions) - [regex101, an online tester that explains each part](https://regex101.com/) - [Regular expression on Wikipedia](https://en.wikipedia.org/wiki/Regular_expression) --- ## Regions and Availability Zones > Where AWS physically lives: cities worldwide, each split into separate data centers. - Level: Basic - Topics: AWS - URL: https://tidbit.legios.com.ar/regions-and-availability-zones/ - In Spanish: [Regiones y zonas de disponibilidad](https://tidbit.legios.com.ar/es/regions-and-availability-zones/) A **Region** is a geographic area where AWS has built its infrastructure: Northern Virginia, São Paulo, Frankfurt, Tokyo and a few dozen more. Each Region is independent from the others. When you create something in AWS (a bucket in Amazon S3, a server in Amazon EC2) you pick the Region it lives in, and it stays there unless you copy it somewhere else. Inside each Region there are several **Availability Zones** (AZs): groups of one or more data centers with their own power, cooling and network, far enough apart that a fire or a flood in one does not take down the others, but close enough to talk to each other in a couple of milliseconds. A Region has a code like `us-east-1` and each zone adds a letter: `us-east-1a`, `us-east-1b`. Why it matters: picking a Region near your users makes your app feel faster, and sometimes the law requires it (some data cannot leave the country). Spreading an app across two or more AZs is the standard way to survive a data center outage without anybody noticing. A bank in Buenos Aires might run its app in `sa-east-1` (São Paulo) across three zones, with a backup copy in Virginia. ### An analogy A supermarket chain. A Region is a city where the chain operates; each Availability Zone is a separate store in that city with its own stock and its own generator. If one store loses power, customers still shop at the other two. ### Keep in mind - Regions are isolated from each other; nothing is shared or copied automatically. - An Availability Zone is one or more data centers; every Region has at least three. - Running in two AZs is the usual minimum for an app that must stay up. - Prices and the list of available services differ a little from Region to Region. ### Go deeper - [Regions and Availability Zones (official docs)](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using-regions-availability-zones.html) - [AWS global infrastructure](https://aws.amazon.com/about-aws/global-infrastructure/regions_az/) --- ## Reinforcement learning > Learning by trial and error, guided by rewards for good outcomes. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/reinforcement-learning/ - In Spanish: [Aprendizaje por refuerzo](https://tidbit.legios.com.ar/es/reinforcement-learning/) **Reinforcement learning** (RL) is a way of training AI where nobody shows the system the right answers. Instead, an **agent** acts in some environment (a game, a simulation, a conversation), gets a **reward** when things go well and a penalty when they go badly, and gradually learns which actions lead to more reward. It is trial and error, repeated millions of times. Its most famous success came in 2016, when DeepMind's AlphaGo beat Lee Sedol, one of the world's best players of Go, a board game long thought too subtle for computers. AlphaGo improved partly by playing against itself over and over. RL is also used to train robots to walk or grasp, to control cooling in data centers, and in many game-playing systems. For today's chatbots, RL matters in two ways. **RLHF** (reinforcement learning from human feedback) is a step where people compare pairs of answers, a separate model learns their preferences, and the chatbot is then trained to produce answers that preference model scores well. It is a big part of why assistants sound helpful and polite. More recently, RL on problems with checkable answers, like math and code, has been used to train models that reason step by step. The classic risk is **reward hacking**: the agent finds a loophole that earns points without doing what you actually wanted, so the reward has to be designed with care. ### An analogy Like training a dog with treats. You never explain the rule; you reward the behavior you want, and the dog figures out what earns the treat. If you reward the wrong thing by accident, it will learn that instead. ### Keep in mind - Learns from rewards and penalties, not from labeled answers. - AlphaGo's 2016 win at Go is the landmark example. - RLHF uses human preferences to shape how chatbots respond. - A badly designed reward gets gamed: reward hacking. ### Go deeper - [Reinforcement learning on Wikipedia](https://en.wikipedia.org/wiki/Reinforcement_learning) - [Illustrating RLHF (Hugging Face)](https://huggingface.co/blog/rlhf) - [What is reinforcement learning? (IBM)](https://www.ibm.com/think/topics/reinforcement-learning) --- ## REST API > The most common style of web API: things have URLs and you act on them with HTTP verbs. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/rest-api/ - In Spanish: [API REST](https://tidbit.legios.com.ar/es/rest-api/) A **REST API** is an API designed around a simple idea: everything the system knows about is a **resource** with its own URL, and you use the standard verbs of HTTP to act on it. `GET /orders/42` reads order 42, `POST /orders` creates a new one, `PUT /orders/42` replaces it and `DELETE /orders/42` removes it. The answer usually comes back as JSON. The style was described by Roy Fielding in his 2000 doctoral thesis, and the name stands for **REpresentational State Transfer**. One of its rules is that each request is **stateless**: it carries everything the server needs to answer it, so the server does not have to remember previous requests. That makes REST APIs easy to spread over many servers and to cache. Every answer also brings a **status code**, a three-digit number that says how it went: 200 means OK, 201 created, 404 not found, 401 you are not logged in, 500 the server failed. Because it reuses what the web already has, REST became the default way apps, websites and companies talk to each other. When people say "the Stripe API" or "the GitHub API", they usually mean a REST API. ### An analogy Like a well-organized library where every book has a fixed shelf address and there are only a few allowed actions: look at it, add one, replace one, remove one. Anyone who knows the rules can find their way around. ### Keep in mind - Each thing has a URL; the HTTP verb says what to do with it. - GET reads, POST creates, PUT or PATCH updates, DELETE removes. - Requests are stateless: each one stands on its own. - Status codes like 200, 404 and 500 tell you how it went. ### Go deeper - [What is a RESTful API? (AWS)](https://aws.amazon.com/what-is/restful-api/) - [REST (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/REST) - [REST on Wikipedia](https://en.wikipedia.org/wiki/REST) --- ## Router > The box that connects your devices to each other and to the internet. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/router/ - In Spanish: [Router](https://tidbit.legios.com.ar/es/router/) A **router** is a device that passes data between networks. The one at home connects your local network (phones, laptop, TV) to your internet provider's network, and from there to the rest of the world. Big routers at providers and data centers do the same job between huge networks, millions of times per second. Its core trick is choosing the next hop. Every packet carries a destination IP address; the router looks it up in its **routing table** and forwards the packet out the right door. No single router knows the whole path to a server across the ocean; each one only knows which neighbour is closer, and the packet gets there by a chain of such decisions. The home router actually bundles several things in one box: a router, a Wi-Fi access point, a small switch for the Ethernet sockets, a DHCP server that hands out private IP addresses, NAT so all devices share one public address, and a basic firewall. That is why turning it off and on again fixes so much: you are restarting half a dozen services at once. ### An analogy A roundabout with signposts. Each car arriving reads the sign for its destination and takes the right exit; the roundabout does not need to know the whole trip, only the next road. ### Keep in mind - A router forwards packets between networks using a routing table. - Each router only decides the next hop; the internet is a chain of those decisions. - Your home router also does Wi-Fi, NAT, DHCP and a basic firewall. - A modem talks to the provider's line; the router shares that connection. Often it is the same box. ### Go deeper - [What is a router? (Cloudflare)](https://www.cloudflare.com/learning/network-layer/what-is-a-router/) - [Router (Wikipedia)](https://en.wikipedia.org/wiki/Router_(computing)) --- ## SaaS, PaaS and IaaS > Three ways to rent computing, depending on how much of the work you hand over. - Level: Basic - Topics: Cloud - URL: https://tidbit.legios.com.ar/saas-paas-iaas/ - In Spanish: [SaaS, PaaS e IaaS](https://tidbit.legios.com.ar/es/saas-paas-iaas/) Cloud computing is usually sold in three layers, depending on how much the provider takes care of. With **IaaS** (infrastructure as a service) you rent the raw pieces: virtual machines, disks and networks, like Amazon EC2. The provider runs the data center and the hardware; you install and look after the operating system, the software and everything on top. With **PaaS** (platform as a service) you hand over more. You bring your code and the platform runs it: it provides the operating system, the runtime and often the database, and handles patching and scaling. Heroku, Google App Engine and AWS Elastic Beanstalk are classic examples. With **SaaS** (software as a service) you rent a finished application and just use it in the browser: Gmail, Slack, Salesforce, Microsoft 365. The trade-off is always the same: the more the provider manages, the less you can customize and the less work you do yourself. Many services sit between the labels; serverless functions, for example, are often described as a step beyond PaaS. Security responsibility is split the same way: in IaaS you patch your own servers, in SaaS you mostly manage who has access. ### An analogy Pizza. IaaS is renting a kitchen: you buy ingredients and cook. PaaS is a take-and-bake pizza: it comes ready, you just use your oven. SaaS is eating at the pizzeria: you only choose and enjoy. ### Keep in mind - IaaS: rent virtual machines, storage and networks; you manage the rest. - PaaS: bring your code; the platform runs and scales it. - SaaS: use a finished app, usually in the browser, on a subscription. - More managed means less work and less control, including on security. ### Go deeper - [Types of cloud computing (AWS)](https://aws.amazon.com/types-of-cloud-computing/) - [The NIST definition of cloud computing](https://csrc.nist.gov/pubs/sp/800/145/final) - [Software as a service on Wikipedia](https://en.wikipedia.org/wiki/Software_as_a_service) --- ## Salesforce > The company that sells, as an online subscription, the software to manage customers. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/salesforce/ - In Spanish: [Salesforce](https://tidbit.legios.com.ar/es/salesforce/) **Salesforce** is an American company founded in 1999 in San Francisco by Marc Benioff and partners. It sells **CRM** (customer relationship management) software: the system where a company keeps everything about its customers, such as who they are, what they bought, every call and email, which deals are in progress and which complaints are open. Sales, support and marketing teams all work from the same record. What made it famous was how it sold it. Business software used to come on discs, installed on the company's own servers after long projects. Salesforce ran everything on its own servers and customers just logged in through the browser, paying a monthly fee per user. It was one of the pioneers of **SaaS** (software as a service), the model most business software follows today. Over time it added products, often by buying companies: tools for customer service, marketing and online stores, **Slack** for team chat, **Tableau** for charts and dashboards, and **MuleSoft** for connecting systems together. Companies also build their own apps on top of its platform, and a whole profession of administrators and consultants works on it. It makes money mainly from those subscriptions, priced per user and per product, and lately from AI assistants that work inside the CRM. ### An analogy A shared address book and diary for a whole company, kept in the cloud, where every salesperson, support agent and marketer writes down what happened with each customer, and the company pays a monthly fee per person who uses it. ### Keep in mind - Founded in 1999; its core product is CRM, software to manage customer relationships. - A SaaS pioneer: software in the browser, paid by subscription per user. - Owns Slack, Tableau and MuleSoft, among other products. - Many companies build custom apps on its platform. ### Go deeper - [About Salesforce (official site)](https://www.salesforce.com/company/about-us/) - [What is CRM? (Salesforce)](https://www.salesforce.com/crm/what-is-crm/) - [Salesforce on Wikipedia](https://en.wikipedia.org/wiki/Salesforce) --- ## Samsung > Korean giant that sells phones and TVs, and makes much of the world's memory chips. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/samsung/ - In Spanish: [Samsung](https://tidbit.legios.com.ar/es/samsung/) **Samsung Electronics**, the best-known part of the South Korean Samsung group, is two businesses in one. The one you see is consumer electronics: Galaxy phones, TVs, fridges, washing machines. The one you mostly do not see is semiconductors, and it is just as important: Samsung is one of the world's largest makers of **memory chips**, the RAM and flash storage inside phones, laptops, SSDs and servers. Memory is a commodity business with dramatic cycles: when demand outruns supply, prices and profits soar; when factories overproduce, prices crash. Only a few companies make it at scale (Samsung, SK Hynix and Micron are the big three), so Samsung's results are a common barometer for the whole electronics industry. The AI boom made one kind of memory especially valuable: **HBM** (high-bandwidth memory), the stacked chips that sit next to GPUs in AI accelerators. Samsung also runs a **foundry**, manufacturing chips designed by other companies the way TSMC does, and it is one of the very few able to work at the leading edge. It also designs its own phone processors, the Exynos line. That makes it unusual: a company that competes with Apple in phones while also being a supplier to Apple and many others. ### An analogy A big department store that also owns the warehouses where everyone else's shelves come from. You know it for what it sells at the front; much of its weight is in the back. ### Keep in mind - Sells phones, TVs and appliances, and is one of the largest makers of memory chips. - Memory (RAM, flash, HBM for AI) is a cyclical business with few big players. - Its foundry manufactures chips for other companies, competing with TSMC. - It is both a competitor and a supplier to many of its customers. ### Go deeper - [Samsung Electronics on Wikipedia](https://en.wikipedia.org/wiki/Samsung_Electronics) - [Samsung Semiconductor (official site)](https://semiconductor.samsung.com/) --- ## Sandboxing > Running a program in a closed box so it cannot harm anything outside it. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/sandboxing/ - In Spanish: [Sandboxing](https://tidbit.legios.com.ar/es/sandboxing/) **Sandboxing** means running a piece of software inside a restricted space, the **sandbox**, where it can only touch what it has been explicitly given. It cannot read your other files, talk to other programs or change the system unless a narrow, controlled door allows it. If the program is buggy or malicious, the damage stays inside the box. You use sandboxes all day without noticing. Your web browser runs each website in its own sandboxed process, so a malicious page that exploits a bug still cannot read your documents. Phone apps live in sandboxes too, which is why an app has to ask for permission before it sees your photos, contacts or location. Containers and virtual machines are sandboxes at a bigger scale, isolating whole applications on a shared server. Security teams also use sandboxes on purpose: suspicious email attachments are opened in a disposable environment to watch what they do before anyone opens them for real. And it matters more now with AI: when an assistant writes and runs code, it is usually run in a sandbox so a mistake cannot delete or leak anything important. Sandboxes are not perfect, though. Attackers look for "sandbox escapes", bugs that break through the walls, so the walls also need updates. ### An analogy A sandbox in a playground: kids can dig, build and wreck whatever they want inside it, and none of the sand ends up in the living room. ### Keep in mind - A sandbox limits what a program can see and touch to what it was explicitly given. - Browsers, phone apps, containers and virtual machines all rely on it. - Used to safely examine suspicious files and to run AI-generated code. - Bugs that break out of a sandbox are serious, so it still needs updates. ### Go deeper - [Sandbox (computer security) on Wikipedia](https://en.wikipedia.org/wiki/Sandbox_(computer_security)) - [The Chromium sandbox (design document)](https://www.chromium.org/developers/design-documents/sandbox/) --- ## SAP > The German company whose software runs the back office of many large businesses. - Level: Basic - Topics: Tech companies, Data - URL: https://tidbit.legios.com.ar/sap/ - In Spanish: [SAP](https://tidbit.legios.com.ar/es/sap/) **SAP** is a German software company founded in 1972 by five former IBM engineers and based in Walldorf. It is best known for **ERP** (enterprise resource planning) software: one system where a company records its accounting, purchases, inventory, sales, production and payroll. When a factory sells a truckload of goods, the same record updates the stock, the invoice and the books, instead of each department keeping its own spreadsheet. Its flagship product today is **S/4HANA**, which runs on **HANA**, SAP's own database that keeps data in memory so reports can be computed on the fly. Around it, SAP sells software for specific jobs: SuccessFactors for human resources, Ariba for purchasing, Concur for travel expenses, plus tools to analyze the data that all these systems collect. SAP makes money from software licenses, yearly maintenance fees and, more and more, cloud subscriptions where SAP runs the system for the customer. Installing SAP is rarely quick: big companies spend months or years adapting it to their processes, usually with consulting firms that specialize in it. Once a company runs on SAP, its whole operation depends on it, which is why changing systems is a big decision. ### An analogy The nervous system of a company. Every department is an organ doing its own job, but they all report to the same spine, so a sale in one place is instantly felt in the warehouse and in the accounts. ### Keep in mind - Founded in 1972 by five former IBM engineers; based in Walldorf, Germany. - Best known for ERP software that ties finance, stock, sales and payroll together. - S/4HANA is its main ERP; HANA is its in-memory database. - Earns money from licenses, maintenance and cloud subscriptions. ### Go deeper - [SAP on Wikipedia](https://en.wikipedia.org/wiki/SAP) - [Enterprise resource planning on Wikipedia](https://en.wikipedia.org/wiki/Enterprise_resource_planning) - [SAP HANA on Wikipedia](https://en.wikipedia.org/wiki/SAP_HANA) --- ## Satellite internet > Internet access beamed from space to a dish, where cables and cell towers don't reach. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/satellite-internet/ - In Spanish: [Internet satelital](https://tidbit.legios.com.ar/es/satellite-internet/) **Satellite internet** connects a home, a ship or a plane to the internet without any cable running to it. A small dish or flat antenna points at the sky and exchanges radio signals with a satellite. The satellite relays them to a **ground station**, a site with big antennas that is plugged into the regular internet. Everything you send and receive makes that trip up and down. For decades the satellites sat in **geostationary orbit**, roughly 36,000 km up, where one satellite seems fixed in the sky and covers a third of the planet. The catch is distance: the signal takes roughly half a second to go up, down, and back again, so video calls feel awkward and online games are nearly unplayable, even when downloads are reasonably fast. Newer systems use thousands of smaller satellites in **low Earth orbit**, a few hundred to about two thousand kilometers up. Being closer, they cut the delay, the latency, to something similar to a regular home connection. Because each one crosses the sky in minutes, the dish keeps handing your connection from one satellite to the next. SpaceX's Starlink is the best-known example; others are being built. It is a lifeline for rural areas, islands, boats, flights and places where a disaster knocked out the ground network. The trade-offs: the dish needs a clear view of the sky, heavy rain or snow can weaken the signal, and capacity is shared among everyone under the same patch of sky. Large constellations also raise concerns about space debris and about streaks across astronomers' images. ### An analogy Shouting to a friend across a valley by bouncing your voice off a mountain. A distant mountain carries it far but the echo comes back late; a closer hill answers quickly, but you need many hills to cover the whole valley. ### Keep in mind - A dish talks to a satellite, which relays to a ground station on the internet. - High geostationary satellites add roughly half a second of delay. - Low-orbit constellations use thousands of satellites to cut that delay. - Needs a clear view of the sky; heavy weather can slow it down. ### Go deeper - [Satellite internet access on Wikipedia](https://en.wikipedia.org/wiki/Satellite_Internet_access) - [Low Earth orbit on Wikipedia](https://en.wikipedia.org/wiki/Low_Earth_orbit) - [Geostationary orbit on Wikipedia](https://en.wikipedia.org/wiki/Geostationary_orbit) --- ## SDK > A kit of code, tools and docs a company provides so you can build for its platform. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/sdk/ - In Spanish: [SDK](https://tidbit.legios.com.ar/es/sdk/) An **SDK** (software development kit) is a bundle that a platform gives to developers so they can build software for it. It usually contains libraries you call from your own code, documentation, sample projects and some tools, such as an emulator to test a phone app or a command-line program to set things up. Apple and Google publish SDKs for building iPhone and Android apps; AWS, Stripe and many others publish SDKs for using their services. Many SDKs are a friendly layer on top of an API. Instead of building raw HTTP requests, handling authentication, retries and error codes by hand, you call something like `payments.create(amount)` in your own programming language and the SDK does the rest. The same service often offers one SDK per language: Python, JavaScript, Java, Go and so on. There is a trade-off. An SDK saves time and avoids common mistakes, but it adds a dependency to your project and ties your code to that vendor's way of doing things. SDKs embedded in phone apps, for example for ads or analytics, can also collect data about users, which is why app stores now ask developers to disclose what their SDKs gather. ### An analogy The starter kit that comes with a new espresso machine: the right filters, a tamper, a cleaning brush and a booklet. You could find your own tools, but the kit is made to fit exactly that machine. ### Keep in mind - A bundle of libraries, docs, samples and tools for one platform. - Often wraps an API so you can call it in your own language. - Usually offered in several languages for the same service. - Saves time, but adds a dependency and ties you to the vendor. ### Go deeper - [What is an SDK? (AWS)](https://aws.amazon.com/what-is/sdk/) - [SDK (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/SDK) - [Software development kit on Wikipedia](https://en.wikipedia.org/wiki/Software_development_kit) --- ## Security operations center > The team that watches an organization's systems around the clock for attacks. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/security-operations-center/ - In Spanish: [Centro de operaciones de seguridad](https://tidbit.legios.com.ar/es/security-operations-center/) A **security operations center**, or **SOC**, is the team (and often the room) in charge of watching an organization's systems for attacks and reacting when one happens. Think of it as the security guards of the digital side: they monitor alerts day and night, decide which ones are real, and contain the damage when something gets through. The day-to-day work is a loop. Tools such as a SIEM and endpoint protection on every laptop send alerts. **Tier 1** analysts triage them: most are false alarms, some are a user who clicked a phishing link. Real cases go to more experienced analysts who investigate, isolate an infected computer, reset stolen passwords or block an attacker's address. Afterwards the team writes down what happened and adjusts its rules so the same trick is caught earlier next time. Some SOCs also do **threat hunting**: actively looking for intruders no alert has spotted. Running a SOC 24/7 takes several shifts of skilled people, which is expensive. That is why many mid-size companies hire one as a service from a specialized provider (often called **MDR**, managed detection and response) instead of building their own. Whatever the setup, the point is the same: attacks happen at night and on holidays too, and the sooner someone notices, the smaller the damage. ### An analogy A fire station for computers. Most calls are false alarms or a burnt toast, but the team is always on duty, knows the building, and when there is real smoke it arrives in minutes, not days. ### Keep in mind - A SOC monitors, detects, responds to and learns from security incidents. - Analysts are organized in tiers: triage first, deep investigation after. - Its main tools are a SIEM, endpoint protection and well-rehearsed playbooks. - Many organizations rent a SOC as a managed service instead of staffing one. ### Go deeper - [Security operations center on Wikipedia](https://en.wikipedia.org/wiki/Security_operations_center) - [What is a security operations center? (IBM)](https://www.ibm.com/think/topics/security-operations-center) --- ## Semantic versioning > A rule for version numbers like 2.4.1, so each part tells you how risky an update is. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/semantic-versioning/ - In Spanish: [Versionado semántico](https://tidbit.legios.com.ar/es/semantic-versioning/) **Semantic versioning**, or **SemVer**, is a convention for numbering software releases as three numbers: `MAJOR.MINOR.PATCH`, for example `2.4.1`. Each number has a precise meaning, so anyone using the software can tell what kind of change a new version brings just by looking at which number went up. The rules are written at semver.org, first published by Tom Preston-Werner, one of GitHub's founders. A **patch** bump, `2.4.1` to `2.4.2`, means bug fixes only: nothing new, nothing removed. A **minor** bump, to `2.5.0`, adds new features but keeps everything that worked before working the same way. A **major** bump, to `3.0.0`, means there are **breaking changes**: something was removed or behaves differently, and code that used the old version may need adjusting. Versions starting with `0.`, like `0.7.3`, signal early development, where anything may change. This matters because modern software is built from hundreds of libraries, and keeping them updated by hand would be impossible. Package managers like npm or pip read these numbers to update safely: a rule like `^2.4.1` means "accept any 2.x version from 2.4.1 up, but not 3.0". It only works if authors follow the rules honestly, and in practice a minor update occasionally breaks something anyway, which is why projects also lock exact versions and run their tests after updating. ### An analogy Like the labels on a building's renovation notices. "Painting the hallway" (patch), "adding a new elevator" (minor), "we are moving the entrance to the other street" (major): you instantly know which ones require you to change your routine. ### Keep in mind - Three numbers: MAJOR.MINOR.PATCH. - Patch fixes bugs, minor adds compatible features, major can break things. - Versions below 1.0.0 promise no stability. - Package managers rely on it to update dependencies automatically. ### Go deeper - [Semantic Versioning 2.0.0 (official spec)](https://semver.org/) - [Software versioning on Wikipedia](https://en.wikipedia.org/wiki/Software_versioning) --- ## Server > A computer whose job is to answer requests from other computers, all day, every day. - Level: Basic - Topics: Hardware, Cloud - URL: https://tidbit.legios.com.ar/server/ - In Spanish: [Servidor](https://tidbit.legios.com.ar/es/server/) A **server** is a computer that provides something to other computers over a network: web pages, email, files, a database, the data behind an app. Your phone or laptop is the **client** that asks; the server is the one that answers. When you open a website, your browser sends a request across the internet and a server somewhere sends the page back. "Server" describes a role more than a kind of machine. Any computer can act as one, even a Raspberry Pi at home. But machines built for the job look different from a PC: flat boxes stacked in racks inside a data center, with no screen or keyboard, lots of memory, two power supplies in case one fails and disks that can be swapped while they run. They are managed remotely and designed to stay on for years. The word also names the software. A **web server** is a program that answers web requests, and one physical machine can run many server programs at once or be split into several virtual machines. When people say an app runs "in the cloud", it still runs on servers; they just belong to a provider like AWS, Google Cloud or Azure and are rented by the hour. Even serverless uses servers: you simply do not manage them yourself. ### An analogy A waiter in a restaurant. Customers, the clients, ask for things and the waiter brings them. One good waiter can look after many tables at once, and when the place gets busy, the restaurant hires more waiters. ### Keep in mind - A server answers requests from other computers, called clients. - It is a role: the same word names the machine and the program doing the job. - Data center servers sit in racks, with no screen, and run nonstop. - The cloud, and even serverless, still run on servers owned by someone else. ### Go deeper - [Server (computing) on Wikipedia](https://en.wikipedia.org/wiki/Server_(computing)) - [What is a server? (IBM)](https://www.ibm.com/think/topics/server) --- ## Serverless > Code that runs only when called; you never see or manage the server. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/serverless/ - In Spanish: [Serverless](https://tidbit.legios.com.ar/es/serverless/) **Serverless** is a way of running code in the cloud where you upload a function and the provider runs it only when something triggers it: an HTTP request, a new file landing in Amazon S3, a timer. There are still servers, of course, but you never see them, size them or patch them. AWS Lambda is the best-known example, and the word also describes services like Amazon DynamoDB that scale by themselves with no machines to manage. Billing is per execution. If your function runs 10 times a day for 200 milliseconds each, you pay for 2 seconds of computing. If a viral post sends a million requests, the provider starts thousands of copies in parallel and bills for those. Zero traffic costs zero, or close to it. It suits spiky or small workloads: generating thumbnails, receiving webhooks, nightly reports, glue between services. It fits worse for long jobs, since a function usually has a time limit (15 minutes on Lambda), and the first call after a quiet period can be slow while the provider spins it up (a "cold start"). ### An analogy A taxi instead of owning a car. With a server you pay for the car even while it sits parked; with serverless you pay per trip, and someone else worries about the maintenance. ### Keep in mind - You pay per invocation and per millisecond, not per hour of a running machine. - It scales from zero to thousands of copies with no configuration. - Functions are short-lived and stateless: save anything that matters in a database or in storage. - Cold starts and vendor lock-in are the usual complaints. ### Go deeper - [Serverless on AWS](https://aws.amazon.com/serverless/) - [What is AWS Lambda? (official docs)](https://docs.aws.amazon.com/lambda/latest/dg/welcome.html) - [Serverless computing (Wikipedia)](https://en.wikipedia.org/wiki/Serverless_computing) --- ## SIEM > The system that gathers security logs from everywhere and raises the alarm. - Level: Advanced - Topics: Security - URL: https://tidbit.legios.com.ar/siem/ - In Spanish: [SIEM](https://tidbit.legios.com.ar/es/siem/) A **SIEM** (security information and event management) is software that collects logs and events from across an organization, stores them in one place and looks for signs of an attack. Servers, laptops, the firewall, the VPN, cloud accounts, email and identity systems all write records of what happens: who logged in, from where, which file was opened, which connection was blocked. On its own, each record says little. Put together, they can tell a story. The SIEM first **normalizes** the data, so a login looks the same whether it came from Windows, a Linux server or a cloud service. Then it **correlates**: it applies detection rules that combine events across sources and time. "Fifty failed logins followed by a success from a new country" or "an admin account created at 3 a.m. and then used to export a database" are typical rules. When one matches, it raises an **alert** for an analyst, usually in a security operations center. It is also the organization's memory. During an investigation, analysts search months of logs to reconstruct what an attacker did, and auditors ask for proof that access is monitored, which is why regulations often push companies toward one. The hard part is tuning: a SIEM with too many noisy rules buries real attacks under thousands of false alarms, and one with too few misses them. Modern products add behavior analytics, threat intelligence feeds and automated responses, and many now run as a cloud service. ### An analogy The control room of a building's security, where every camera, door sensor and badge reader sends its feed. One guard watching many screens, with software that rings a bell when the same badge opens doors on two different floors at once. ### Keep in mind - Collects and centralizes logs from servers, devices, network gear and cloud services. - Correlation rules combine events across sources to spot attacks no single log reveals. - Keeps history for investigations and compliance audits. - Without careful tuning it drowns analysts in false alarms. ### Go deeper - [Security information and event management on Wikipedia](https://en.wikipedia.org/wiki/Security_information_and_event_management) - [What is SIEM? (IBM)](https://www.ibm.com/think/topics/siem) --- ## SLA > A written promise about how reliable a service will be, and what happens if it is not. - Level: Basic - Topics: Cloud - URL: https://tidbit.legios.com.ar/sla/ - In Spanish: [SLA](https://tidbit.legios.com.ar/es/sla/) An **SLA** (service level agreement) is the part of a contract where a provider commits to a level of service: usually how much of the time it will be up, sometimes also how fast it will respond or how quickly support will answer. It also says what you get if the promise is broken, which in cloud services is almost always **service credits**, a discount on next month's bill, not a refund for the business you lost. The famous number is uptime, written as a percentage. It sounds tiny but each extra "nine" matters a lot. **99%** allows roughly 3.65 days of downtime a year. **99.9%** (three nines) allows roughly 8.8 hours a year, or about 43 minutes a month. **99.99%** (four nines) allows roughly 53 minutes a year, about 4 minutes a month. Going from three to four nines usually means running in several availability zones or regions, with automatic failover. Engineers use two related terms. An **SLI** (service level indicator) is what you actually measure, like the share of requests that succeed. An **SLO** (service level objective) is the internal target for it, usually stricter than the public SLA so there is a safety margin. Read the fine print: an SLA defines what counts as "down", often excludes planned maintenance, and covers only that one service, not the whole system you built on top of it. ### An analogy A pizza place that promises delivery in 30 minutes or the next pizza is free. The promise tells you what to expect, but a free pizza does not save the dinner party that went hungry. ### Keep in mind - A contractual promise of service level, usually uptime, with credits if missed. - 99.9% allows roughly 8.8 hours of downtime a year; 99.99% roughly 53 minutes. - SLI is what you measure, SLO is the internal target, SLA is the external promise. - Check the definitions and exclusions; credits rarely cover real losses. ### Go deeper - [Service-level agreement on Wikipedia](https://en.wikipedia.org/wiki/Service-level_agreement) - [Service level objectives (Google SRE book)](https://sre.google/sre-book/service-level-objectives/) - [AWS service level agreements](https://aws.amazon.com/legal/service-level-agreements/) --- ## Social engineering > Manipulating people, instead of machines, into giving away access or information. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/social-engineering/ - In Spanish: [Ingeniería social](https://tidbit.legios.com.ar/es/social-engineering/) **Social engineering** is the art of getting what you want from a system by working on the people around it. Instead of breaking encryption or finding a bug, the attacker convinces someone to share a password, read out a verification code, approve a payment or hold a door open. Phishing is its most common form, but it also happens by phone, in person, through social networks and in messaging apps. The tricks lean on ordinary human instincts. **Authority**: 'This is your bank's fraud department' or 'the CEO needs this transfer today'. **Urgency**: 'your account will be closed in one hour'. **Helpfulness**: a new 'colleague' who forgot their badge. **Curiosity or fear**: a USB stick labeled 'salaries' left in the parking lot. A carefully built cover story is called **pretexting**; a fraudulent phone call is **vishing** (voice phishing). It works on smart, careful people because it does not attack intelligence; it attacks routine and good manners. The attacker often does homework first, using public information from LinkedIn or a company's website, so the story includes real names and details. Voice cloning and convincing fake messages make the stories even more believable. The defense is mostly process. Verify requests through a separate channel you already know, such as calling back the official number or asking the person face to face. No real support team needs your password or the code a service just texted you. In companies, clear rules help: payments to new bank accounts need a second approval, and nobody gets punished for slowing down to check. ### An analogy A con artist in a hotel lobby who never touches a lock. With a confident smile and a believable story about a lost key card, he gets the receptionist to hand him a new one. ### Keep in mind - It targets people's trust, habits and good manners, not the technology. - Authority, urgency, helpfulness and fear are the usual levers. - Verify any unusual request through a channel you choose, not the one offered. - Never share passwords or one-time codes, whoever asks. ### Go deeper - [Avoiding social engineering and phishing attacks (CISA)](https://www.cisa.gov/news-events/news/avoiding-social-engineering-and-phishing-attacks) - [Social engineering on Wikipedia](https://en.wikipedia.org/wiki/Social_engineering_(security)) --- ## Software library > Ready-made code for a common task that programmers plug into their own programs. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/software-library/ - In Spanish: [Biblioteca de software](https://tidbit.legios.com.ar/es/software-library/) A **software library** is a package of code written to solve one kind of problem well, so other programmers do not have to solve it again. There are libraries for handling dates and time zones, drawing charts, resizing images, reading spreadsheets, encrypting data or talking to a payment provider. You add the library to your project and call its functions when you need them. Modern software is mostly assembled this way. A typical app has a small layer of its own code on top of dozens or hundreds of libraries, called its **dependencies**. Each language has a **package manager** that downloads them with one command, like `pip install` for Python or `npm install` for JavaScript, and many of those libraries are open source. The convenience has a cost. Every dependency is code someone else wrote and maintains, so a bug or a security vulnerability in a popular library can affect thousands of apps at once, and an abandoned one can leave a project stuck. That is why teams keep a list of exact versions, update them regularly and use tools that warn about known problems. ### An analogy Buying ready-made parts for a piece of furniture: hinges, drawer rails, screws. You still design and build the cabinet, but you do not forge your own hinges. ### Keep in mind - Reusable code for a specific task, called from your own program. - Installed with a package manager such as pip or npm. - Most apps rely on many libraries, known as dependencies. - A flaw in a popular library can affect many apps at once. ### Go deeper - [Library (computing) on Wikipedia](https://en.wikipedia.org/wiki/Library_(computing)) - [Python Package Index (PyPI), a public library catalog](https://pypi.org/) --- ## Software testing > Checking that software does what it should, mostly with code that tests other code. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/software-testing/ - In Spanish: [Testing de software](https://tidbit.legios.com.ar/es/software-testing/) **Software testing** is everything done to find out whether a program works before its users do. Some of it is manual: a person clicks through the app looking for problems. Most of it, in modern teams, is **automated tests**: small programs that call the real code with known inputs and check the result. A test might say: "`total([10, 5])` must return `15`". If someone later breaks that, the test fails and tells them. Tests come in layers. **Unit tests** check one small piece, like a single function, in milliseconds. **Integration tests** check that pieces work together, for example the code and a real database. **End-to-end tests** drive the whole app like a user would, opening a browser and clicking buttons; they are the most realistic but also the slowest and most fragile. The usual advice, called the **test pyramid**, is many unit tests, fewer integration tests and a handful of end-to-end ones. The main value is confidence to change things. A good **test suite** runs automatically on every change in the CI pipeline, catching **regressions**, old features broken by new code, within minutes. Tests are not proof of correctness, though: they only check the cases someone thought of. As the computer scientist Edsger Dijkstra put it, testing can show the presence of bugs, never their absence. ### An analogy Like a pilot's pre-flight checklist. It does not guarantee a perfect flight, but it reliably catches the known ways things go wrong, every single time, without depending on anyone's memory. ### Keep in mind - Automated tests are code that runs your code and checks the results. - Unit, integration and end-to-end tests trade speed for realism. - They run on every change, so broken features are caught early. - Passing tests mean the checked cases work, not that there are no bugs. ### Go deeper - [The Practical Test Pyramid (martinfowler.com)](https://martinfowler.com/articles/practical-test-pyramid.html) - [Software testing on Wikipedia](https://en.wikipedia.org/wiki/Software_testing) --- ## Source code > The human-readable text that programmers write and that a program is built from. - Level: Basic - Topics: Software - URL: https://tidbit.legios.com.ar/source-code/ - In Spanish: [Código fuente](https://tidbit.legios.com.ar/es/source-code/) **Source code** is the set of text files where programmers write a program, in a programming language like Python, Java or JavaScript. It is meant to be read by people: it has names that make sense, comments that explain decisions, and a structure split into files and folders. All the source code of a product together is often called its **codebase**. What you install on a phone or a laptop is usually not the source code but the result of building it: a compiler or other tools turn the text into a form the computer runs quickly and that is very hard for humans to read. That is why having the source code matters. With it you can understand exactly what a program does, fix it, change it or rebuild it; without it you mostly cannot. Companies usually treat their source code as a valuable secret. **Open source** projects do the opposite: they publish it so anyone can read, reuse and improve it, under a license that sets the rules. Either way, source code today almost always lives in a version control system like Git, which keeps the full history of who changed what and when. ### An analogy The architect's plans for a building. Living in the house does not require them, but if you want to know where the pipes run, or add a floor, you need the plans. ### Keep in mind - Readable text, written by people in a programming language. - It is built (compiled or packaged) into the program that users actually run. - Having the source lets you inspect, fix and change a program. - Usually kept in Git; published openly in open source projects. ### Go deeper - [Source code on Wikipedia](https://en.wikipedia.org/wiki/Source_code) - [Compile (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Compile) --- ## Spam > Unwanted messages sent in bulk: ads, scams and junk you never asked for. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/spam/ - In Spanish: [Spam](https://tidbit.legios.com.ar/es/spam/) **Spam** is unsolicited messages sent to huge numbers of people at once. It started with email and now arrives everywhere: text messages, messaging apps, social media comments, phone calls, fake reviews. Sending one more message costs the sender almost nothing, so even if only a tiny fraction of people respond, it pays. That is why there is so much of it; by most estimates a large share of all email sent is spam. Some spam is just annoying advertising. A lot of it is a vehicle for something worse: phishing links that try to steal passwords, attachments that install Malware, fake prizes, investment scams, or messages that only check whether your address or number is active so it can be sold on. **Spammers** get addresses from data breaches, from scraping websites, from buying lists, or simply by guessing common names at popular domains. Email providers fight it with **spam filters** that score every incoming message: who sent it, whether the sending server is who it claims to be (checks called SPF, DKIM and DMARC), what it contains and how other users reacted to similar messages. Most spam never reaches your inbox at all. Marking a message as spam, rather than just deleting it, helps train those filters for everyone. Good habits: do not reply or click to 'unsubscribe' in a message that looks shady, since that confirms your address is real (unsubscribe links from companies you actually signed up with are usually fine). Do not open unexpected attachments. Think twice before publishing your email address or phone number openly, and consider a separate address for shop sign-ups. ### An analogy The paper flyers stuffed into every mailbox on the street. Printing one more costs almost nothing, so they keep coming even if nearly everyone throws them away, and once in a while one of them is a scam dressed as a bill. ### Keep in mind - Spam is unsolicited bulk messaging, by email, SMS, apps or calls. - It is cheap to send, so even a tiny response rate makes it profitable. - Much of it carries phishing links, scams or malware. - Mark it as spam instead of replying; filters learn from that. ### Go deeper - [Email spam on Wikipedia](https://en.wikipedia.org/wiki/Email_spam) - [Report spam in Gmail (Gmail Help)](https://support.google.com/mail/answer/1366858) --- ## Spotify > A Swedish streaming service for music and podcasts, free with ads or paid monthly. - Level: Basic - Topics: Tech companies, Cloud - URL: https://tidbit.legios.com.ar/spotify/ - In Spanish: [Spotify](https://tidbit.legios.com.ar/es/spotify/) **Spotify** was founded in 2006 in Stockholm by Daniel Ek and Martin Lorentzon, and launched in 2008, at a time when much of the music online was pirated. Its bet was that listening instantly, without downloading files, could be more convenient than piracy. You open the app, pick a song and it starts playing in a second or two, from a catalog of many millions of tracks, plus podcasts and audiobooks. Behind that second there is a lot of engineering. Each song is stored on servers in the cloud, compressed at several quality levels. The app asks for the first few seconds, starts playing and keeps fetching the rest while you listen, with a small buffer so a weak connection does not cut the music. Songs you play often are kept on your phone, and Premium users can download them, encrypted, to listen offline. In 2016 Spotify started moving its infrastructure to Google Cloud. The other big piece is **recommendations**. Spotify looks at what millions of people play, skip and save, and finds patterns: people who love these ten songs also tend to love this other one. It combines that with an analysis of the audio itself and of what is written about artists online, and turns it into personalized playlists such as Discover Weekly. It makes money in two ways: ads for free listeners and **Premium** subscriptions, and it pays a large share of that money to record labels, publishers and artists for the rights to the music. ### An analogy A radio station that plays only for you. It has almost every record ever made, and its DJ learns your taste by watching what you skip. ### Keep in mind - Founded in 2006 in Stockholm by Daniel Ek and Martin Lorentzon; launched in 2008. - Streams music in small pieces with a buffer, instead of downloading whole files. - Recommendations come from comparing the listening habits of many people. - Earns money from ads and Premium subscriptions, and pays rights holders a large share. ### Go deeper - [About Spotify (official newsroom)](https://newsroom.spotify.com/company-info/) - [Spotify Research](https://research.atspotify.com/) - [Spotify on Wikipedia](https://en.wikipedia.org/wiki/Spotify) --- ## Spreadsheet vs database > Why a company that runs on Excel eventually needs a real database. - Level: Basic - Topics: Data - URL: https://tidbit.legios.com.ar/spreadsheet-vs-database/ - In Spanish: [Planilla vs base de datos](https://tidbit.legios.com.ar/es/spreadsheet-vs-database/) A **spreadsheet**, like Excel or Google Sheets, is a grid where you can type anything in any cell and add formulas. That freedom is its strength: anyone can start a customer list or a budget in minutes. A **database** is stricter. You define the structure first (this column is a date, this one an email, every order must belong to an existing customer), and the system enforces it for every record, forever. Most businesses start on spreadsheets, and the trouble arrives with growth. Copies multiply by email, and nobody knows whether `clients_final_v3.xlsx` or `clients_final_REAL.xlsx` is current. Two people edit at once and one overwrites the other. Someone types "N/A" in a price column and a total silently goes wrong. Files get slow, and Excel itself stops at roughly a million rows. And there is no reliable record of who changed what, or a way to let one person see only their own region. A database solves those exact problems: one shared source of truth, many people and apps reading and writing safely at the same time, rules that reject bad data, relationships between tables so customer details live in one place, permissions and backups. The move does not have to be dramatic. Many teams go first to tools like Airtable, then to a proper database with an app or a BI tool on top, and keep spreadsheets for what they do best: quick personal analysis. ### An analogy A spreadsheet is a notebook: perfect for one person, flexible, fast. A database is the reception desk of a busy clinic: forms with required fields, one shared appointment book, and rules that stop two patients being booked in the same slot. ### Keep in mind - Spreadsheets are flexible; databases enforce structure and rules. - Warning signs: many file versions, overwritten edits, broken totals, slow files. - Databases handle many simultaneous users, permissions and history. - Spreadsheets stay useful for quick analysis on top of the database. ### Go deeper - [What is a database? (AWS)](https://aws.amazon.com/what-is/database/) - [Spreadsheet on Wikipedia](https://en.wikipedia.org/wiki/Spreadsheet) - [Database on Wikipedia](https://en.wikipedia.org/wiki/Database) --- ## SQL > The language you use to ask a database questions. - Level: Basic - Topics: Data, Software - URL: https://tidbit.legios.com.ar/sql/ - In Spanish: [SQL](https://tidbit.legios.com.ar/es/sql/) **SQL** (Structured Query Language, often pronounced 'sequel') is the language for talking to relational databases. You do not tell the database how to find the data, step by step; you describe what you want, and the database works out the fastest way to get it. That is why a few lines of SQL can answer a question over millions of rows. A query reads almost like an English sentence: `SELECT name, city FROM customers WHERE country = 'AR' ORDER BY name;` asks for the name and city of every customer in Argentina, alphabetically. Other commands add rows (`INSERT`), change them (`UPDATE`) or remove them (`DELETE`). `JOIN` combines tables that point to each other, like orders with the customers who placed them, and `GROUP BY` summarizes, like total sales per month. SQL dates from the 1970s and is still everywhere: PostgreSQL, MySQL, SQLite (the database inside most phones) and big analytics tools like Amazon Athena or a data warehouse. Each engine adds its own extras, called a dialect, but the core is the same, so learning it once pays off for decades. It is also one of the most useful skills outside engineering: analysts, product managers and marketers use it to answer their own questions. ### An analogy Ordering at a restaurant. You say what you want ('the steak, medium, no fries'), not how to cook it; the kitchen (the database engine) decides the steps. ### Keep in mind - SQL is declarative: you describe the result and the database decides how to get it. - `SELECT` reads data; `INSERT`, `UPDATE` and `DELETE` change it. - `JOIN` combines related tables and `GROUP BY` summarizes them. - Every engine has its own dialect, but the basics work almost everywhere. ### Go deeper - [What is SQL? (AWS)](https://aws.amazon.com/what-is/sql/) - [The SQL language, PostgreSQL tutorial](https://www.postgresql.org/docs/current/tutorial-sql.html) - [SQL on Wikipedia](https://en.wikipedia.org/wiki/SQL) --- ## SQL injection > Typing text into a form that the database mistakes for its own instructions. - Level: Intermediate - Topics: Security, Data - URL: https://tidbit.legios.com.ar/sql-injection/ - In Spanish: [Inyección SQL](https://tidbit.legios.com.ar/es/sql-injection/) Behind most login forms and search boxes there is a database that speaks SQL. When you search a shop for 'red shoes', the website builds a command along the lines of `find products whose name contains: red shoes` and sends it to the database. **SQL injection** happens when the site builds that command by simply gluing your text into it. Then a visitor can type something that is not a search term at all but a fragment of SQL, and the database cannot tell where the site's instructions end and the visitor's begin. Depending on what the attacker types, the changed command might return every customer's record instead of a few products, skip the password check on a login page, or alter data. It is one of the oldest web vulnerabilities, it has been behind many large data breaches, and OWASP still lists injection among the most important risks for web applications. The fix is well known and not complicated: never build a command by pasting text together. Developers use **parameterized queries** (also called prepared statements), where the command is written first with a fixed placeholder and the user's input is passed separately, always treated as a value and never as code. Modern frameworks and libraries do this by default. On top of that, the database account the website uses should only be allowed to do what the site needs, so a mistake cannot become a disaster. As a user you cannot do much about it, beyond using unique passwords so a leak from one site does not open your other accounts. As someone who hires or manages a team building software, it is worth asking whether queries are parameterized and whether the code has had a security review or a penetration test. ### An analogy A bank teller who reads aloud whatever is written on the slip: 'Pay to: Ana Pérez'. If someone writes 'Ana Pérez, and also empty the vault', a careless teller reads it as one instruction. The fix is a form with a box for the name that is only ever treated as a name. ### Keep in mind - It happens when user input is pasted straight into a database command. - The database then runs part of the input as instructions. - Parameterized queries keep input as data and prevent it. - Limit the database permissions of the app to contain any mistake. ### Go deeper - [SQL Injection (OWASP)](https://owasp.org/www-community/attacks/SQL_Injection) - [SQL Injection Prevention Cheat Sheet (OWASP)](https://cheatsheetseries.owasp.org/cheatsheets/SQL_Injection_Prevention_Cheat_Sheet.html) - [SQL injection on Wikipedia](https://en.wikipedia.org/wiki/SQL_injection) --- ## SSD > Storage with no moving parts: where your files live when the power is off. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/ssd/ - In Spanish: [SSD](https://tidbit.legios.com.ar/es/ssd/) An **SSD** (Solid-State Drive) is the storage in almost every modern laptop, phone and server: the place where files, programs and the operating system stay saved when the power is off. It stores data in flash memory chips, the same technology as a USB stick, and it has no moving parts. The older alternative, the **hard disk drive** (HDD), spins magnetic platters and reads them with a moving arm. The difference you feel is speed. An SSD finds any file in a fraction of a millisecond; a hard disk has to physically move its arm, which takes several milliseconds. That is why a computer with an SSD boots in seconds and opens apps instantly, and why replacing an old laptop's hard disk with an SSD was for years the single best upgrade. SSDs are also silent, lighter and survive drops. Hard disks have not disappeared: per gigabyte they are still much cheaper, so data centers use them for backups and the bulk of cloud storage like Amazon S3, with SSDs in front for the hot, frequently accessed part. Flash cells also wear out after many writes, though modern drives manage this well enough that a normal user never notices. SSD versus RAM, since people confuse them: RAM is the fast working memory that empties when you shut down; the SSD is the persistent storage that keeps your files. More RAM makes multitasking smoother; a bigger SSD lets you keep more files. ### An analogy A hard disk is a record player: the needle has to travel to the right groove. An SSD is a book with tabs: you open straight to any page, with nothing to move. ### Keep in mind - Persistent storage made of flash chips; no moving parts, very fast access. - Replaced hard disks in laptops and phones; hard disks remain for cheap bulk storage. - Connects through SATA (older, slower) or NVMe over PCIe (newer, much faster). - Not the same as RAM: the SSD keeps data when powered off, RAM does not. ### Go deeper - [Solid-state drive on Wikipedia](https://en.wikipedia.org/wiki/Solid-state_drive) - [Flash memory on Wikipedia](https://en.wikipedia.org/wiki/Flash_memory) --- ## SSH > An encrypted way to log in to another computer and type commands on it. - Level: Intermediate - Topics: Networks, Security - URL: https://tidbit.legios.com.ar/ssh/ - In Spanish: [SSH](https://tidbit.legios.com.ar/es/ssh/) **SSH** (Secure Shell) lets you open the command line of a computer somewhere else, usually a server in a data center or in the cloud, and work on it as if you were sitting in front of it. You type `ssh ana@server.example.com`, and from then on every command runs over there. Everything that travels between the two machines, including what you type and what comes back, is encrypted. It replaced older tools like Telnet that sent everything, passwords included, as plain text anyone on the path could read. SSH normally listens on port 22, and the same encrypted channel also carries file copies (`scp` and SFTP), connections to Git hosts like GitHub, and **tunnels** that forward one program's traffic through the server. You can log in with a password, but the usual way is a **key pair**, an application of public-key cryptography. You create two matching keys on your laptop: the private key stays there, ideally protected with a passphrase, and the public key is copied to the server. When you connect, the server checks that you hold the private key without the key ever being sent. Losing a public key is harmless; leaking a private key is like losing the master key. The server also proves who it is. The first time you connect, SSH shows the server's **fingerprint** and asks whether to trust it; afterwards, if that fingerprint ever changes, you get a loud warning, because someone could be impersonating the machine. Common hardening steps are turning off password logins, disallowing direct login as the administrator account, and removing keys of people who left. ### An analogy A sealed, private phone line straight into a room in another building, where the door opens only for the key you carry, and the room has to show you its badge before you speak. ### Keep in mind - Gives you a remote command line over an encrypted connection, usually on port 22. - Key pairs beat passwords: the public key goes on the server, the private key never leaves you. - A changed server fingerprint warning means stop and check. - Also carries file transfers (SFTP, `scp`) and Git connections. ### Go deeper - [Secure Shell on Wikipedia](https://en.wikipedia.org/wiki/Secure_Shell) - [About SSH (GitHub Docs)](https://docs.github.com/en/authentication/connecting-to-github-with-ssh/about-ssh) - [OpenSSH (official site)](https://www.openssh.com/) --- ## Stripe > A company that lets websites and apps accept online payments with a few lines of code. - Level: Basic - Topics: Tech companies, Software - URL: https://tidbit.legios.com.ar/stripe/ - In Spanish: [Stripe](https://tidbit.legios.com.ar/es/stripe/) **Stripe** was founded in 2010 by Patrick and John Collison, two brothers from Ireland, and is based in San Francisco and Dublin. It sells the plumbing for **online payments**: when you pay with a card in an app or on a website, there is a good chance Stripe is the one moving the money behind the scenes. Accepting cards used to mean contracts with banks, long paperwork and strict security rules for storing card numbers. Stripe turned it into an API: a developer adds a few lines of code, the customer's card details go straight from the payment form to Stripe, and the shop receives a **token**, a stand-in that represents the card without revealing it. Stripe then talks to the card networks and the customer's bank, checks for fraud, and later deposits the money in the shop's account. Around that core it built more products: recurring billing for subscriptions, tools for marketplaces that split each payment among many sellers, invoices, tax calculation and support for many local payment methods. It makes money by charging a fee on each transaction, usually a small percentage plus a fixed amount, so it earns more as its customers sell more. ### An analogy A cashier you can hire for your online shop. You never touch the customer's card: the cashier takes it, checks it with the bank, keeps the details locked away and hands you the money minus a small cut. ### Keep in mind - Founded in 2010 by brothers Patrick and John Collison, from Ireland. - Turns accepting card payments into a few calls to its API. - Card data goes to Stripe; the shop only keeps a token that stands for it. - Earns a fee on each payment it processes. ### Go deeper - [About Stripe (official site)](https://stripe.com/about) - [Stripe payments documentation](https://docs.stripe.com/payments) - [Stripe on Wikipedia](https://en.wikipedia.org/wiki/Stripe,_Inc.) --- ## Subnet > A slice of a network's IP addresses, grouped so its devices can talk directly. - Level: Advanced - Topics: Networks - URL: https://tidbit.legios.com.ar/subnet/ - In Spanish: [Subred](https://tidbit.legios.com.ar/es/subnet/) A **subnet** is a block of consecutive IP addresses that belong together, like all the addresses on one office floor or in one part of a cloud network. Devices in the same subnet reach each other directly. To reach anything outside it, they hand their traffic to a router, known as the **default gateway**, which forwards it on. Subnets are usually written in **CIDR** notation: `10.0.1.0/24`. The number after the slash says how many of the 32 bits of an IPv4 address are fixed for the whole subnet; the rest are free for devices. A `/24` fixes the first three numbers, so `10.0.1.0` to `10.0.1.255` is roughly 250 usable addresses. A `/16` fixes only the first two and holds about 65,000. The bigger the number after the slash, the smaller the subnet. The older way to say the same thing is a **subnet mask**: `/24` is `255.255.255.0`. Splitting a network has practical reasons. Each subnet can get its own rules, so a firewall can let the web servers in one subnet talk to the database in another, while the guest Wi-Fi in a third reaches only the internet. Smaller subnets also keep local chatter, like devices asking 'who has this address?', from flooding every machine in the building. In the cloud this is everyday work. When you create an Amazon VPC you pick a range like `10.0.0.0/16` and carve it into subnets, typically one public and one private per Availability Zone. Public subnets hold what must be reachable from the internet, such as a load balancer; private ones hold databases and internal services. Ranges cannot overlap if networks will ever be connected, so it pays to plan them early. ### An analogy A building's address range split into floors. People on the same floor walk over to each other's desks; to reach another floor, they go through the reception desk, which knows where everyone is. ### Keep in mind - Devices in the same subnet talk directly; anything else goes through the router. - `/24` means the first 24 bits are fixed: roughly 250 usable addresses. - A bigger number after the slash means a smaller subnet. - In the cloud, public subnets face the internet and private ones stay inside. ### Go deeper - [Subnet on Wikipedia](https://en.wikipedia.org/wiki/Subnet) - [Subnets for your VPC (AWS docs)](https://docs.aws.amazon.com/vpc/latest/userguide/configure-subnets.html) --- ## System on a chip > One chip that holds a computer's main parts: CPU, GPU, memory control and more. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/system-on-a-chip/ - In Spanish: [Sistema en un chip](https://tidbit.legios.com.ar/es/system-on-a-chip/) A **system on a chip** (often shortened to SoC) puts most of a computer's main parts on a single piece of silicon. Instead of a separate processor, graphics card and controller chips spread over a motherboard, one chip holds the CPU cores, a GPU, often an NPU for AI tasks, the controllers for memory, screen, camera and USB, and in phones the cellular modem. Every smartphone runs on one, and so do tablets, smartwatches, many laptops and the Raspberry Pi. Packing everything together has big advantages. Signals travel millimeters instead of centimeters, so the parts talk to each other faster and use much less energy, which means longer battery life and less heat. The device can also be thinner and cheaper to build. Apple silicon Macs, most Android phones and many small devices are built around one, usually based on the Arm architecture. The trade-off is flexibility. In a desktop PC you can swap the graphics card or add more RAM; in a system on a chip the parts are fixed when the chip is designed, and the memory is often soldered right next to it or inside the same package. Designing one is also very expensive, so only large companies design their own from scratch, while many others combine licensed building blocks. ### An analogy A studio apartment instead of a house. Kitchen, bedroom and living room share one compact space, so everything is close and cheap to heat, but you cannot knock down a wall to add a room. ### Keep in mind - Puts the CPU, the GPU and many controllers on one piece of silicon. - Uses less power and space, which is why every smartphone has one. - Its parts cannot be upgraded separately, unlike in a desktop PC. - Most are based on the Arm architecture. ### Go deeper - [System on a chip on Wikipedia](https://en.wikipedia.org/wiki/System_on_a_chip) - [Apple silicon, an example of SoCs, on Wikipedia](https://en.wikipedia.org/wiki/Apple_silicon) --- ## TCP and UDP > Two ways to send packets: TCP checks everything arrives, UDP just sends and moves on. - Level: Intermediate - Topics: Networks - URL: https://tidbit.legios.com.ar/tcp-and-udp/ - In Spanish: [TCP y UDP](https://tidbit.legios.com.ar/es/tcp-and-udp/) IP gets packets to the right machine, but it promises nothing: packets can get lost, duplicated or arrive out of order. **TCP** (Transmission Control Protocol) and **UDP** (User Datagram Protocol) are the two protocols that sit on top of IP and decide how much to care. Every connection on the internet uses one or the other, plus a port number to reach the right program. **TCP** is reliable. It opens a connection with a three-step handshake, numbers every packet, waits for acknowledgements, resends what went missing and delivers everything in order. It also slows down when the network is congested. Web pages (HTTP, HTTPS), email, file transfers and almost every API use TCP, because a web page with a missing chunk is useless. **UDP** is the opposite: no handshake, no acknowledgements, no retries. Each packet goes out on its own, and if it is lost, it is lost. That sounds worse, but it is exactly right when a late packet is as useless as a missing one: a video call, a live stream, an online game, a DNS lookup. Better to skip a frame than to freeze while waiting for it. HTTP/3, the newest version of the web's protocol, runs on top of UDP (through QUIC) and adds its own reliability, to avoid some of TCP's delays. ### An analogy TCP is a phone conversation with confirmation: did you get that? yes, go on; repeat if not. UDP is shouting across a stadium: fast, no confirmation, and if a word gets lost you do not stop the game to repeat it. ### Keep in mind - TCP: connection, ordering, retransmission. Reliable but with some delay. - UDP: no connection, no retries. Fast and simple; losing a packet is acceptable. - Web, email and APIs use TCP; calls, streaming, games and DNS use UDP. - HTTP/3 runs over UDP (QUIC) and brings its own reliability. ### Go deeper - [What is TCP/IP? (Cloudflare)](https://www.cloudflare.com/learning/ddos/glossary/tcp-ip/) - [What is UDP? (Cloudflare)](https://www.cloudflare.com/learning/ddos/glossary/user-datagram-protocol-udp/) - [RFC 9293: Transmission Control Protocol](https://datatracker.ietf.org/doc/html/rfc9293) --- ## Technical debt > The future cost of shortcuts taken in software today. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/technical-debt/ - In Spanish: [Deuda técnica](https://tidbit.legios.com.ar/es/technical-debt/) **Technical debt** is the extra work a team will have to do later because of a quicker, messier choice made now. The programmer Ward Cunningham coined the metaphor in 1992 to explain it to non-technical managers: shipping imperfect code is like borrowing money. You get something sooner, but you pay **interest**: every future change in that area takes longer and breaks more easily, until you pay back the **principal** by cleaning it up. Debt is not automatically bad. A startup testing whether anyone wants its product may reasonably cut corners, the same way a business takes a loan to grow. The trouble is debt nobody tracks. It piles up from many sources: rushed deadlines, copy-pasted code, missing tests, outdated libraries, an early design that no longer fits what the product became, or simply people who left without documenting anything. You notice it when simple requests start taking weeks, when each fix causes two new bugs, or when only one person dares to touch a certain module. Teams manage it by naming it in their backlog, reserving part of each cycle to pay it down, and refactoring as they go. Ignored long enough, it can make a system so costly to change that a full rewrite looks tempting, which is its own risky bet. ### An analogy Like not washing the dishes after dinner. One night it saves time. After a week, cooking anything means first scrubbing a mountain of pans, and every meal takes longer. ### Keep in mind - Shortcuts now cost extra effort on every future change. - Taken on deliberately and tracked, it can be a reasonable trade-off. - Signs: slow changes, fragile code, areas nobody wants to touch. - Paid down by refactoring, tests and upgrades, little by little. ### Go deeper - [Technical Debt, by Martin Fowler](https://martinfowler.com/bliki/TechnicalDebt.html) - [Technical debt on Wikipedia](https://en.wikipedia.org/wiki/Technical_debt) --- ## Terraform > A popular tool for describing cloud infrastructure in files and creating it for you. - Level: Intermediate - Topics: Cloud - URL: https://tidbit.legios.com.ar/terraform/ - In Spanish: [Terraform](https://tidbit.legios.com.ar/es/terraform/) **Terraform** is a tool for infrastructure as code, created by HashiCorp (now part of IBM) and released in 2014. You write files ending in `.tf` that say what you want, for example "a server of this size, a bucket with this name, a database in this network", and Terraform talks to the cloud to make it real. The everyday loop has two commands. `terraform plan` compares your files with what exists and shows a list of changes it would make: create this, modify that, delete this other thing. `terraform apply` carries them out. To know what already exists, Terraform keeps a **state file**, a record of every resource it manages; teams store it in a shared place so two people do not change the same thing at once. Its big draw is that one language works for many platforms. Plugins called **providers** connect it to AWS, Azure, Google Cloud, and hundreds of other services, from DNS to monitoring tools. In 2023 HashiCorp changed Terraform's license to one that is no longer open source, and part of the community forked it into **OpenTofu**, which works almost the same way. ### An analogy Like an architect's blueprint plus a builder who reads it. You change the drawing, the builder tells you exactly which walls will go up or come down, and only after you agree does the work start. ### Keep in mind - You describe the desired result; Terraform works out the steps. - `plan` shows the changes before `apply` makes them. - The state file is Terraform's memory of what it built; losing it causes real trouble. - OpenTofu is an open source fork that works almost the same way. ### Go deeper - [What is Terraform? (official docs)](https://developer.hashicorp.com/terraform/intro) - [Terraform on Wikipedia](https://en.wikipedia.org/wiki/Terraform_(software)) --- ## Tesla > An electric car maker that treats the car as a computer on wheels. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/tesla/ - In Spanish: [Tesla](https://tidbit.legios.com.ar/es/tesla/) **Tesla** is an American company founded in 2003 by Martin Eberhard and Marc Tarpenning; Elon Musk joined as its main early investor and later became CEO. It makes electric cars, such as the Model S, Model 3 and Model Y, plus large batteries for homes and power grids. It earns money mostly by selling cars, and also from energy storage, charging and software features sold as upgrades or subscriptions. The heart of an electric car is the **battery**. Tesla's packs are built from thousands of small lithium-ion cells, similar in shape to oversized AA batteries, wired together and lying flat in the floor of the car. Software constantly manages their temperature and charge, which matters for range and for how long the battery lasts. Tesla builds its own large battery and car factories, and its fast-charging network, the **Superchargers**, uses a plug design that other carmakers in North America have since adopted. Tesla also changed how cars get better over time. Most of the car is controlled by its central computer and software, so Tesla sends **over-the-air updates**, like a phone's, that can add features or fix problems without a visit to the workshop. Its driver-assistance systems, sold as Autopilot and "Full Self-Driving (Supervised)", use cameras and neural networks to steer, brake and change lanes, but they require an attentive driver ready to take over; regulators in several countries have examined how they are used and marketed. ### An analogy Like a smartphone that happens to have wheels: the hardware is built once, and the software keeps changing what it can do long after you take it home. ### Keep in mind - Founded in 2003; makes electric cars and large batteries for homes and grids. - Battery packs are made of thousands of small cells managed by software. - Over-the-air updates add features and fixes without a workshop visit. - Its driver assistance relies on cameras and needs an attentive driver. ### Go deeper - [Tesla, Inc. on Wikipedia](https://en.wikipedia.org/wiki/Tesla,_Inc.) - [Tesla Autopilot on Wikipedia](https://en.wikipedia.org/wiki/Tesla_Autopilot) - [Over-the-air update on Wikipedia](https://en.wikipedia.org/wiki/Over-the-air_update) --- ## TLS certificate > The ID card a website shows your browser to prove it is who it claims to be. - Level: Intermediate - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/tls-certificate/ - In Spanish: [Certificado TLS](https://tidbit.legios.com.ar/es/tls-certificate/) A **TLS certificate** is a small file that says 'the public key for `example.com` is this one', signed by a **certificate authority** (CA) that your browser already trusts. When you open a site over HTTPS, the server sends its certificate. The browser checks that the signature is from a known CA, that the name matches the site you asked for and that it has not expired. If all three pass, you get the padlock and the browser uses the public key from the certificate to start encryption with that server, and nobody else. The CA's job is to check, before signing, that whoever asked for the certificate actually controls the domain. Since 2015 Let's Encrypt does this automatically and for free: the server proves control by publishing a token the CA asked for, gets a certificate valid for 90 days, and renews it on its own. That is why nearly every site has HTTPS today and why 'SSL certificate', the old name, is still what many people call it. What a certificate proves is narrow: that you are really talking to `mybank.com`. It says nothing about whether `mybank-secure-login.com` is honest; a phishing site can have a perfectly valid certificate for its own look-alike name. And an 'expired certificate' warning usually means an administrator forgot to renew, not an attack, although the safe move is still not to type anything there. Click the padlock in your browser and you can read the certificate yourself: who issued it, for which names, and until when. It is the same mechanism, in reverse, that lets a company sign its software so your computer knows an update really came from them. ### An analogy A passport. An authority the border guard trusts issued it, it binds a name to a face (a domain to a public key) and it expires. A valid passport proves identity, not good intentions. ### Keep in mind - A certificate binds a domain name to a public key, signed by a trusted authority. - The browser checks the signature, the name and the expiry date before showing the padlock. - Let's Encrypt made certificates free and automatic; most sites renew every 90 days. - It proves you reached the real domain, not that the domain is trustworthy. ### Go deeper - [What is an SSL certificate? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/ssl/what-is-an-ssl-certificate/) - [Let's Encrypt: how it works](https://letsencrypt.org/how-it-works/) - [Public key certificate on Wikipedia](https://en.wikipedia.org/wiki/Public_key_certificate) --- ## Token > The small chunk of text a language model actually reads. - Level: Basic - Topics: AI - URL: https://tidbit.legios.com.ar/token/ - In Spanish: [Token](https://tidbit.legios.com.ar/es/token/) A language model like Claude or ChatGPT does not read letters or whole words. It reads **tokens**: pieces of text that are usually a short word, part of a long word, or a punctuation mark. The word `cat` is one token; `unbelievable` might be three. Before the model sees your message, a program called a tokenizer splits it into tokens and turns each one into a number. The model works with those numbers, predicts the next token, and the output is turned back into text. Tokens matter in practice because that is what you pay for and what fits in the model's memory. A page of English is roughly 400 tokens; Spanish and code usually need a bit more. ### An analogy Think of Lego bricks. You do not build with raw plastic, and you do not buy a whole castle: you work with standard pieces that snap together. Tokens are the bricks of text. ### Keep in mind - One token is about 3 or 4 English characters on average. - Prices and limits for AI models are measured in tokens, not words. - The same text can be more tokens in another language or in code. ### Go deeper - [Anthropic tokenizer and token counting](https://docs.anthropic.com/en/docs/build-with-claude/token-counting) - [OpenAI tokenizer playground](https://platform.openai.com/tokenizer) --- ## TPU > Google's chip designed specifically for the math of neural networks. - Level: Intermediate - Topics: Hardware, AI - URL: https://tidbit.legios.com.ar/tpu/ - In Spanish: [TPU](https://tidbit.legios.com.ar/es/tpu/) A **TPU** (Tensor Processing Unit) is a chip Google designed for one job: the arithmetic behind neural networks. Training and running an AI model comes down to multiplying huge tables of numbers, called **tensors** or matrices, over and over. A CPU can do it, slowly; a GPU does it well because it has thousands of small cores; a TPU goes a step further and is built almost entirely around that one operation. Its heart is a **matrix unit**: a grid of multipliers where numbers flow from cell to cell, each one multiplying, adding and passing the result to its neighbor, so data is reused instead of being fetched from memory again and again. TPUs also work with lower-precision numbers, which are accurate enough for neural networks and much cheaper to compute. Google has run them in its data centers since 2015 for search, translation and its Gemini models, and rents them to other companies through Google Cloud. You cannot buy a TPU for your PC. They come in large groups called **pods**, linked by very fast connections, and are available only as a cloud service. They are a clear example of a wider trend: big tech companies designing their own AI chips instead of relying only on NVIDIA's GPUs, to lower costs and control their supply. Each generation is faster, but the core idea stays the same. ### An analogy A GPU is a big kitchen with many cooks who can make almost any dish. A TPU is a factory line built for a single recipe: it cannot do much else, but for that recipe each plate comes out fast and cheap. ### Keep in mind - Designed by Google for the matrix math of neural networks. - Used both to train models and to run them (inference). - Available only through Google Cloud, in large groups called pods. - Part of a trend of tech companies making their own AI chips. ### Go deeper - [Introduction to Cloud TPU (Google Cloud)](https://cloud.google.com/tpu/docs/intro-to-tpu) - [Tensor Processing Unit on Wikipedia](https://en.wikipedia.org/wiki/Tensor_Processing_Unit) --- ## Training and inference > Training builds the model once; inference is using it, millions of times a day. - Level: Intermediate - Topics: AI - URL: https://tidbit.legios.com.ar/training-and-inference/ - In Spanish: [Entrenamiento e inferencia](https://tidbit.legios.com.ar/es/training-and-inference/) An AI model has two very different lives. **Training** is when it learns: the neural network sees enormous amounts of data, makes predictions, measures its errors and adjusts its weights, over and over, for weeks or months on thousands of GPUs. At the end you have a file of numbers, the trained model. It happens once per model version, and it is the expensive part. **Inference** is everything after: using the trained model to answer. Every time you send a prompt to Claude or ChatGPT, that is one inference. The weights do not change; the model runs the input through the network and produces an output, token by token. Each one is cheap compared with training, but there are billions of them, so inference ends up being most of the computing bill. The split explains a lot of what you read in the news. A model's 'knowledge cutoff' is the end of its training data. **Fine-tuning** is a short extra round of training on specific examples. 'Serving' a model means running inference for users, which is why companies need data centers full of GPUs even after the model is finished. Example: a model that reads chest X-rays is trained on hundreds of thousands of labeled scans over several weeks. Afterward, in a hospital, it looks at one new scan in under a second. That second is inference. ### An analogy Training is medical school: years, expensive, done once. Inference is every patient the doctor sees after graduating: fast, repeated, and what the school was for. ### Keep in mind - Training adjusts the weights; inference uses them without changing them. - Training is a huge one-time cost; inference is a small cost repeated billions of times. - Fine-tuning is a short, targeted extra round of training. - Both run mostly on GPUs, but small models can do inference on a laptop or a phone. ### Go deeper - [3Blue1Brown: gradient descent, how neural networks learn](https://www.3blue1brown.com/lessons/gradient-descent) - [NVIDIA: the difference between training and inference](https://blogs.nvidia.com/blog/difference-deep-learning-training-inference-ai/) - [Machine learning on Wikipedia](https://en.wikipedia.org/wiki/Machine_learning) --- ## Transformer > The neural network design behind modern language models, built around 'attention'. - Level: Advanced - Topics: AI - URL: https://tidbit.legios.com.ar/transformer/ - In Spanish: [Transformer](https://tidbit.legios.com.ar/es/transformer/) The **transformer** is a way of building neural networks, introduced in 2017 by a team at Google in a paper titled 'Attention Is All You Need'. It is the architecture behind almost every large language model today: the T in GPT stands for it, and Claude, Gemini and Llama are transformers too. It is also used for images, audio, video and even protein structures. Its central idea is **self-attention**. Text is first split into tokens, and each token becomes a list of numbers. Then, layer after layer, every token looks at every other token in the context and decides how much each one matters for understanding it. In 'the cat sat because it was tired', attention lets 'it' connect strongly to 'cat'. Many attention 'heads' work in parallel, each picking up different relations: grammar, who did what, which earlier sentence a word refers back to. A language model stacks dozens of these layers and ends by producing probabilities for the next token. Before transformers, language models mostly read text one word at a time, in order (recurrent networks), which was slow to train and forgot things over long passages. Attention looks at the whole sequence at once, so training can be split across thousands of GPUs running in parallel, and that is what made today's giant models practical. The cost is that comparing every token with every other grows with the square of the text length, one reason context windows are limited and long inputs are expensive. ### An analogy Like a meeting where, before speaking, each person glances at everyone else and decides whose earlier comments matter most to what they are about to say. Everyone does this at the same time, round after round, until the group's understanding is sharp. ### Keep in mind - Introduced in 2017 in the paper 'Attention Is All You Need'. - Self-attention lets each token weigh every other token. - Processes whole sequences in parallel, which made huge models trainable. - Attention cost grows with the square of the input length. ### Go deeper - [Transformer (deep learning architecture) on Wikipedia](https://en.wikipedia.org/wiki/Transformer_(deep_learning_architecture)) - [The Illustrated Transformer (Jay Alammar)](https://jalammar.github.io/illustrated-transformer/) - [Attention Is All You Need (original paper, arXiv)](https://arxiv.org/abs/1706.03762) --- ## Transistor > An electrically controlled switch, the basic building block of every chip. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/transistor/ - In Spanish: [Transistor](https://tidbit.legios.com.ar/es/transistor/) A **transistor** is a switch with no moving parts. It has three connections: current wants to flow between two of them (the **source** and the **drain**), and a small voltage on the third (the **gate**) decides whether it can. Gate on, current flows; gate off, it stops. On means 1, off means 0, and that is where the ones and zeros of computing physically live. One switch does not do much. But wire a few together and you get **logic gates**, little circuits that answer questions like "are both inputs on?" (AND) or "is at least one on?" (OR). Combine gates and you can add numbers, compare them, and store a bit of memory. Stack that up by the billion and you have a CPU. The kind used in almost every modern chip is the **MOSFET**, built directly into silicon. The first transistor was built at Bell Labs in 1947, a fist-sized device that replaced fragile vacuum tubes. Since then they have shrunk so much that a modern phone chip holds tens of billions of them, so small that more than a thousand would fit side by side across a human hair. Smaller transistors switch faster, use less energy and fit more per chip, which is the story Moore's law describes. ### An analogy A water tap controlled by a tiny lever. Pressure is always waiting in the pipe; flick the lever and water flows, release it and the flow stops. A computer is billions of these taps opening and closing in patterns, billions of times per second. ### Keep in mind - A switch controlled by voltage: gate on lets current through, gate off blocks it. - On and off are the 1s and 0s of digital computing. - Combined into logic gates, they add, compare and remember. - Invented in 1947; modern chips hold billions of them. ### Go deeper - [Transistor on Wikipedia](https://en.wikipedia.org/wiki/Transistor) - [MOSFET on Wikipedia](https://en.wikipedia.org/wiki/MOSFET) --- ## TSMC > The factory that makes the world's most advanced chips for everyone else. - Level: Intermediate - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/tsmc/ - In Spanish: [TSMC](https://tidbit.legios.com.ar/es/tsmc/) **TSMC** (Taiwan Semiconductor Manufacturing Company) manufactures chips for other companies and sells none under its own name. Apple designs the chip in your iPhone, NVIDIA designs the GPU in an AI data center, AMD designs its CPUs, and all of them send those designs to TSMC, which fabricates them in its factories (**fabs**) in Taiwan. A company that only manufactures is called a **foundry**, and TSMC invented the model in 1987 under its founder Morris Chang. What makes it matter is that it is the best at it. Making a modern chip means printing billions of transistors a few nanometers wide onto silicon wafers, in a process with hundreds of steps and machines (from the Dutch company ASML) that cost hundreds of millions each. Only a handful of companies can do it at the leading edge (Samsung and Intel are the others), and TSMC has led for years: nearly every cutting-edge AI and smartphone chip is made there. That concentration is a global risk. Most of the world's most advanced chips come from a few fabs on one island that China claims as its territory. Governments noticed: the United States, Japan and Germany have subsidized TSMC fabs on their own soil (Arizona, Kumamoto, Dresden), and 'chip sovereignty' became policy. Export controls on which chips may be sold to China run partly through what TSMC is allowed to make for whom. Why you hear about it: when NVIDIA cannot ship enough GPUs, when Apple announces a new iPhone chip, or when a '2-nanometer' process is in the news, TSMC is the factory behind it. Its capacity, more than any single design, sets the ceiling on how much AI hardware exists in a given year. ### An analogy The world's finest printing press, owned by one company. Authors (Apple, NVIDIA, AMD) write the books; TSMC is the only press that can print them at that resolution, so everyone lines up at its door. ### Keep in mind - A pure foundry: manufactures chips designed by others, never its own brand. - Makes the most advanced chips for Apple, NVIDIA, AMD, Qualcomm and many more. - Based in Taiwan, which makes it a geopolitical concern; new fabs in the US, Japan and Germany. - Its capacity is a bottleneck for the whole AI industry. ### Go deeper - [tsmc.com](https://www.tsmc.com) - [TSMC on Wikipedia](https://en.wikipedia.org/wiki/TSMC) - [Semiconductor fabrication plant on Wikipedia](https://en.wikipedia.org/wiki/Semiconductor_fabrication_plant) --- ## Two-factor authentication > A second proof that you are you, so a stolen password alone is not enough. - Level: Basic - Topics: Security - URL: https://tidbit.legios.com.ar/two-factor-authentication/ - In Spanish: [Autenticación de dos factores](https://tidbit.legios.com.ar/es/two-factor-authentication/) **Two-factor authentication** (2FA, or MFA when there can be more than two) means logging in with two different kinds of proof: something you **know** (a password) plus something you **have** (your phone, a hardware key) or something you **are** (a fingerprint, your face). A password can be guessed, leaked or phished from across the world. Your phone cannot, at least not as easily. In practice: you type your password, and the service asks for a six-digit code from an authenticator app, a tap on a notification, or a touch on a small USB key. Codes sent by SMS also count, but they are the weakest option, because attackers can take over a phone number by convincing the carrier to move it to a new SIM. The strongest and most convenient option today is a **passkey** or hardware key: it only works on the real site, so a phishing page cannot ask you for it. For the rest, an authenticator app (Google Authenticator, Microsoft Authenticator, or the one inside your password manager) is a solid default. Practical takeaway: turn it on for your email account first, because the email is what resets every other password. Then your bank, your main social network and your work accounts. Save the backup codes somewhere safe, so that losing your phone does not mean losing the account. ### An analogy A debit card at the ATM: you need the card (something you have) and the PIN (something you know). A thief who shoulder-surfed your PIN still cannot withdraw a peso without the card. ### Keep in mind - Two different kinds of proof: know, have, are. A second password is not a second factor. - Authenticator apps and hardware keys beat SMS codes; SMS still beats nothing. - Passkeys and hardware keys resist phishing because they only answer the real site. - Enable it on your email first: it is the account that resets all the others. ### Go deeper - [What is two-factor authentication? (Cloudflare Learning Center)](https://www.cloudflare.com/learning/access-management/what-is-two-factor-authentication/) - [Multi-factor authentication on Wikipedia](https://en.wikipedia.org/wiki/Multi-factor_authentication) --- ## Unicode > The universal numbering of every letter and symbol, and why accents sometimes break. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/unicode/ - In Spanish: [Unicode](https://tidbit.legios.com.ar/es/unicode/) Computers store text as numbers, so every character needs a number. **Unicode** is the worldwide standard that assigns one: `A` is U+0041, `ñ` is U+00F1, the Greek `Ω` is U+03A9, and the emoji 😀 is U+1F600. It covers virtually every writing system in use, plus many historic ones and a lot of symbols, well over a hundred thousand characters in all. Before Unicode, each region used its own incompatible table, and a file written in one often came out garbled in another. Unicode says which number each character gets; an **encoding** says how to write those numbers as bytes in a file or a network message. The dominant one is **UTF-8**. It stores plain English letters in one byte each, exactly like the old ASCII standard, and uses two to four bytes for everything else. That backward compatibility is a big reason it won: today the vast majority of web pages are UTF-8. When accents break, this is usually why. If a program writes "año" in UTF-8 and another reads those bytes assuming an older encoding such as Latin-1, each half of the two-byte `ñ` is shown as a separate character and you get "año". That garbage has a name, **mojibake**. The fix is almost never to edit the text by hand, but to make both sides agree that the text is UTF-8. ### An analogy A global phone book that gives every character on Earth its own number. UTF-8 is the way of writing those numbers on paper: short numbers take one box, long ones take several. If the reader assumes every number fits in one box, a two-box number turns into two wrong ones. ### Keep in mind - Unicode gives every character, from A to ñ to emoji, a unique number. - UTF-8 is the most common way to store those numbers as bytes. - Plain English text in UTF-8 is identical to old ASCII. - "ñ" instead of "ñ" means the text was read with the wrong encoding. ### Go deeper - [What is Unicode? (Unicode Consortium)](https://www.unicode.org/standard/WhatIsUnicode.html) - [UTF-8 (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/UTF-8) - [The Absolute Minimum Every Software Developer Must Know About Unicode (Joel Spolsky)](https://www.joelonsoftware.com/2003/10/08/the-absolute-minimum-every-software-developer-absolutely-positively-must-know-about-unicode-and-character-sets-no-excuses/) --- ## URL > The full address of something on the web: protocol, site and path. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/url/ - In Spanish: [URL](https://tidbit.legios.com.ar/es/url/) A **URL** (Uniform Resource Locator) is the address of one specific thing on the web: a page, an image, a video, an API endpoint. It is what you type in the address bar or share in a chat. `https://es.wikipedia.org/wiki/Mate?lang=es#historia` is one URL, and every piece of it means something. Read it left to right. `https://` is the **scheme**: how to talk (HTTPS, encrypted). `es.wikipedia.org` is the **host**, the domain name DNS turns into an IP address. `/wiki/Mate` is the **path**, which page on that site. `?lang=es` is the **query**, extra options for the server. `#historia` is the **fragment**, a spot inside the page that only the browser uses. Sometimes a port appears after the host, like `:8080`, when the server is not on the standard one. URLs matter because they are the web's links: a search result, a QR code, a share button, all just pass a URL around. For you, the host is the part to check: a phishing page can copy a bank's look, but `banco-seguro.ejemplo.com` is not `banco.com`, and the browser shows the real host in the bar. ### An analogy A full postal address on an envelope: country, city, street, floor and apartment. Each part narrows it down until there is exactly one door. ### Keep in mind - Scheme, host, path, query, fragment: `https://` + `site.com` + `/page` + `?a=1` + `#section`. - The host is the part that says who you are really talking to. - Spaces and special characters get encoded, which is why you see `%20` in links. ### Go deeper - [What is a URL? (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Howto/Web_mechanics/What_is_a_URL) - [URL (Wikipedia)](https://en.wikipedia.org/wiki/URL) --- ## USB > The standard plug for connecting and charging almost everything. - Level: Basic - Topics: Hardware - URL: https://tidbit.legios.com.ar/usb/ - In Spanish: [USB](https://tidbit.legios.com.ar/es/usb/) **USB** (Universal Serial Bus) is the standard that lets a computer talk to keyboards, mice, printers, phones, cameras and drives through one kind of port, and lets a cable carry power at the same time. Before it arrived in the mid-1990s, each kind of device had its own connector. USB replaced most of them, and also became the default way to charge small electronics. Two things get confused because both are called USB. One is the **connector**, the physical shape: the flat rectangular **USB-A**, the small micro-USB of older phones, and the oval, reversible **USB-C** that most new devices use. The other is the **version**, which sets the speed and features: USB 2.0, USB 3.x, USB4. A USB-C port can be anything from slow to very fast, depending on the version behind it and on the cable. USB-C also carries much more power than older USB, enough to charge most laptops, using a negotiation called **USB Power Delivery**: the charger and the device agree on how much power to send before sending it. It can carry video to a monitor as well. That is why one cable can now replace several, and why the European Union made USB-C the required charging port for phones and many other devices sold there. ### An analogy A universal power outlet that also works as a phone line. One shape of plug, and through it the device both gets its energy and talks to the computer. ### Keep in mind - One standard for connecting devices and carrying power. - Connector shape (USB-A, USB-C) and version (USB 2.0, 3.x, USB4) are different things. - USB-C is reversible and can carry data, video and enough power for a laptop. - Two USB-C ports or cables that look the same can differ a lot in speed. ### Go deeper - [USB on Wikipedia](https://en.wikipedia.org/wiki/USB) - [USB-C on Wikipedia](https://en.wikipedia.org/wiki/USB-C) - [About USB-IF, the group behind the standard](https://www.usb.org/about) --- ## Vector database > A database that finds things by similar meaning instead of exact matching words. - Level: Intermediate - Topics: AI, Data - URL: https://tidbit.legios.com.ar/vector-database/ - In Spanish: [Base de datos vectorial](https://tidbit.legios.com.ar/es/vector-database/) A **vector database** stores **embeddings**: lists of numbers that an AI model produces to represent the meaning of a piece of text, an image or a sound. Things with similar meaning get similar numbers, so they end up close together, like points on a map. 'Puppy' lands near 'dog' and 'kitten', far from 'invoice'. A vector database is built to answer one question fast: which stored items are closest to this one? That is different from a regular database search. Searching a help center for 'cannot log in' with plain keywords misses an article titled 'Resetting your password'. Searching by vectors finds it, because the meanings are close even though the words are different. To do this over millions of items without comparing against every single one, vector databases use special indexes (names like HNSW come up) that find nearly the closest matches very quickly, trading a tiny bit of precision for a lot of speed. Vector databases became popular with large language models because they are the usual memory behind **RAG**: your documents are split into chunks, each chunk is embedded and stored, and when someone asks a question, the closest chunks are fetched and handed to the model as context. They also power semantic search, 'similar products' recommendations and duplicate detection. Some are dedicated products; others are features added to existing systems, like the `pgvector` extension for PostgreSQL. ### An analogy A library where books are shelved by what they are about rather than by title. Walk to the shelf for the book you have in mind, and the neighbors on that shelf are the ones most similar to it, whatever their titles say. ### Keep in mind - Stores embeddings: numbers that represent meaning. - Finds the nearest items to a query, not exact word matches. - Uses approximate search to stay fast with millions of items. - The usual storage behind RAG and semantic search. ### Go deeper - [What is a vector database? (AWS)](https://aws.amazon.com/what-is/vector-databases/) - [Vector database on Wikipedia](https://en.wikipedia.org/wiki/Vector_database) - [What is a vector database? (IBM)](https://www.ibm.com/think/topics/vector-database) --- ## Video streaming > Watching a video while it arrives in small pieces, instead of downloading it all first. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/video-streaming/ - In Spanish: [Streaming de video](https://tidbit.legios.com.ar/es/video-streaming/) **Video streaming** is how Netflix, YouTube or a live match reach your screen: the video starts playing after a moment, while the rest is still on its way. Behind the scenes, the video has been **compressed** with a codec such as H.264 or AV1, which shrinks it by storing only what changes from one frame to the next, and then cut into short **segments** of a few seconds each. Each video is also prepared in several versions, from low resolution to 4K, each with a different **bitrate**: how many bits per second it needs. A small text file, the **manifest**, lists all the versions and segments. Your player reads it and requests the segments one after another over ordinary HTTP, usually from a CDN server near you, keeping a few seconds or minutes stored ahead in a **buffer**. The clever part is **adaptive bitrate streaming**. The player constantly measures how fast segments are arriving. If your Wi-Fi slows down, it asks for the next segments in a lower quality, so the picture gets blurrier for a while but keeps moving; when the connection improves, it climbs back up. The spinning wheel of **buffering** appears only when the buffer runs empty faster than it can be refilled. HLS and MPEG-DASH are the two common formats for this, and live streams work the same way, with segments created on the fly a few seconds behind real time. ### An analogy A conveyor belt of sushi plates. You start eating as soon as the first plate arrives, the kitchen keeps a few plates ahead of you, and if it gets busy it sends simpler dishes so you never sit in front of an empty belt. ### Keep in mind - The video is compressed and cut into segments of a few seconds. - Each video exists at several bitrates; a manifest lists them. - The player switches quality segment by segment to match your connection. - Buffering happens when the buffer empties faster than it refills. ### Go deeper - [Adaptive bitrate streaming on Wikipedia](https://en.wikipedia.org/wiki/Adaptive_bitrate_streaming) - [Streaming audio and video (MDN)](https://developer.mozilla.org/en-US/docs/Web/Media/Guides/Streaming) - [Streaming media on Wikipedia](https://en.wikipedia.org/wiki/Streaming_media) --- ## Virtual machine > A complete computer made of software, running inside a real one. - Level: Basic - Topics: Cloud, Hardware - URL: https://tidbit.legios.com.ar/virtual-machine/ - In Spanish: [Máquina virtual](https://tidbit.legios.com.ar/es/virtual-machine/) A **virtual machine** (VM) is a computer that exists as software. It has what looks like its own processor, memory, disk and network card, runs its own operating system and its own programs, and as far as that software can tell, it is a real machine. Underneath, it is sharing a physical computer with other virtual machines. A layer called a **hypervisor** makes this possible: it splits the real CPU, RAM and disks into slices and gives each VM its own, keeping them isolated so one cannot see or crash another. That lets you run Windows inside a Mac, test software on several systems at once, or keep an old program alive on an operating system that no longer runs on modern hardware. VMs are the foundation of cloud computing. When you launch an Amazon EC2 instance or a Google Compute Engine machine, you get a virtual machine on a server in someone's data center, ready in a minute and billed while it runs. Containers are a lighter alternative: they share the host's operating system instead of each bringing their own, so they start faster but are less isolated. ### An analogy Apartments in a building. Everyone shares the same foundation, pipes and wiring, but each apartment has its own door, its own key and its own furniture, and what happens in one does not spill into the next. ### Keep in mind - Software that behaves like a full computer, with its own operating system. - A hypervisor splits one physical machine into many isolated VMs. - Cloud servers like EC2 instances are virtual machines. - Containers are lighter because they share the host's operating system. ### Go deeper - [What is a virtual machine? (AWS)](https://aws.amazon.com/what-is/virtual-machine/) - [Virtual machine on Wikipedia](https://en.wikipedia.org/wiki/Virtual_machine) --- ## VLAN > A way to split one physical network into several separate ones, using software rules. - Level: Advanced - Topics: Networks - URL: https://tidbit.legios.com.ar/vlan/ - In Spanish: [VLAN](https://tidbit.legios.com.ar/es/vlan/) A **VLAN** (virtual LAN) lets one set of cables and switches behave like several independent networks. An office might plug everyone into the same switches, yet put staff laptops in one VLAN, guest Wi-Fi in another, and security cameras in a third. Devices in one VLAN see each other as if they shared a private network, and do not see the others at all. It works by labeling traffic. Each switch port is assigned to a VLAN, and when frames travel between switches over a shared link (a **trunk**), each one carries a small tag with its VLAN number, defined by the standard **802.1Q**. The tag has room for roughly 4,000 different VLANs. A switch never delivers a frame to a port of a different VLAN, so broadcasts and casual snooping stay inside their group. To go from one VLAN to another, traffic has to pass through a router or a firewall, and that is the point: it becomes a checkpoint where you can write rules like 'cameras may talk to the recording server and nothing else'. This is called **network segmentation**. It limits the damage when one device is compromised, keeps noisy traffic contained, and lets teams reorganize who belongs where without rewiring the building. VLANs are a tool for separation, not a security guarantee on their own. A misconfigured trunk can let traffic leak between groups, so sensitive segments also get firewall rules and monitoring. In the cloud the same idea shows up as virtual networks and subnets, configured entirely in software. ### An analogy One office floor divided with movable partitions into separate departments. The building and the hallways are shared, but each department has its own door, and to go from one to another you have to pass the reception desk. ### Keep in mind - One physical network, several logical ones, separated by switch configuration. - Frames between switches carry an 802.1Q tag with the VLAN number. - Traffic between VLANs must cross a router or firewall, where rules apply. - Useful for segmentation, but needs careful configuration to be a real barrier. ### Go deeper - [VLAN on Wikipedia](https://en.wikipedia.org/wiki/VLAN) - [IEEE 802.1Q on Wikipedia](https://en.wikipedia.org/wiki/IEEE_802.1Q) --- ## VPN > An encrypted tunnel that sends all your traffic through another network first. - Level: Intermediate - Topics: Networks, Security - URL: https://tidbit.legios.com.ar/vpn/ - In Spanish: [VPN](https://tidbit.legios.com.ar/es/vpn/) A **VPN** (Virtual Private Network) wraps all the traffic from your device inside an encrypted tunnel to a VPN server, which then sends it on to the internet on your behalf. Everything in between (the café Wi-Fi, your provider, the network you are on) sees only scrambled packets going to one address. The websites you visit see the VPN server's IP address, not yours. There are two very different uses. Companies use VPNs so that a laptop at home can reach the office network as if it were plugged in there: internal tools, file servers, databases that are not exposed to the internet. Consumers use VPN services to hide their traffic from the local network, to appear to be in another country, or to get around blocks. Same technology (protocols like WireGuard, OpenVPN or IPsec), different purpose. A VPN moves trust, it does not remove it. The VPN provider can see everything your internet provider used to see, so a VPN is only as trustworthy as the company running it. It does not make you anonymous (the sites you log into still know who you are), and it does not replace HTTPS, which already protects the content of each site. It does protect you on untrusted networks and hides which sites you visit from whoever runs that network. Many companies are replacing the office VPN with zero trust access, where each app checks identity directly. ### An analogy Mailing all your letters inside one sealed bag to a trusted friend in another city, who opens the bag and posts them from there. Your local post office sees only bags addressed to your friend; the recipients see your friend's return address. ### Keep in mind - All traffic goes encrypted to the VPN server, which forwards it to the internet. - Sites see the VPN's IP address; your local network sees only an encrypted tunnel. - The VPN provider can see what your ISP could: you are choosing whom to trust. - Office VPN = reach the internal network; consumer VPN = privacy on untrusted networks, or location. ### Go deeper - [What is a VPN? (Cloudflare)](https://www.cloudflare.com/learning/access-management/what-is-a-vpn/) - [Virtual private network (Wikipedia)](https://en.wikipedia.org/wiki/Virtual_private_network) --- ## VRAM > The memory built into a graphics card, where the GPU keeps what it is working on. - Level: Intermediate - Topics: Hardware - URL: https://tidbit.legios.com.ar/vram/ - In Spanish: [VRAM](https://tidbit.legios.com.ar/es/vram/) **VRAM** (video RAM) is the memory that belongs to a GPU. It sits on the graphics card or right beside the GPU chip, separate from the computer's main RAM, and holds whatever the GPU is working on: the textures and frames of a game, the video being edited, the weights of an AI model. Today it is usually a fast type called **GDDR** or, in data-center GPUs, **HBM** (High Bandwidth Memory), made of stacked chips placed right next to the processor. Two numbers matter. **Capacity**, in gigabytes, decides what fits: if a game's textures or a model's weights do not fit in VRAM, the GPU has to fetch data from much slower memory elsewhere, and performance collapses or the job refuses to run. **Bandwidth**, in gigabytes per second, decides how fast the GPU can be fed. A GPU has thousands of cores, and they sit idle if data cannot reach them quickly enough. VRAM is why conversations about AI talk so much about memory. A large language model's weights must fit in VRAM to run quickly, so the amount of memory often matters more than raw speed when choosing hardware, and big models are split across several GPUs. That is also why high-end AI chips carry large amounts of expensive HBM. Many laptops and all phones have no separate VRAM: the GPU shares the main memory with the CPU, which is cheaper and simpler but usually slower. ### An analogy A chef's countertop next to the stove. The pantry holds far more, but if the ingredients for tonight's dishes do not fit on the counter, the chef keeps walking back and forth and dinner slows to a crawl. ### Keep in mind - VRAM is the GPU's own memory, separate from the main RAM. - Capacity decides what fits; bandwidth decides how fast the GPU is fed. - AI models must fit in VRAM to run quickly, so memory size often limits what you can run. - Gaming cards use GDDR; data-center AI chips use stacked HBM. ### Go deeper - [Video random-access memory on Wikipedia](https://en.wikipedia.org/wiki/Video_random-access_memory) - [High Bandwidth Memory on Wikipedia](https://en.wikipedia.org/wiki/High_Bandwidth_Memory) - [GDDR SDRAM on Wikipedia](https://en.wikipedia.org/wiki/GDDR_SDRAM) --- ## Vulnerability and CVE > A flaw that lets an attacker misuse software, and the public id that names it. - Level: Intermediate - Topics: Security - URL: https://tidbit.legios.com.ar/vulnerability-and-cve/ - In Spanish: [Vulnerabilidad y CVE](https://tidbit.legios.com.ar/es/vulnerability-and-cve/) A **vulnerability** is a mistake in software or its configuration that someone can exploit to do what they should not: read other people's data, take over a server, crash a service. It is usually an ordinary bug with bad consequences, like a login check that can be skipped or a form that runs whatever text you type into it. Every large program has some; the question is who finds them first. When one is found in software many people use, it gets a **CVE** id (Common Vulnerabilities and Exposures), like `CVE-2021-44228`, the famous Log4Shell. The id is a shared name, so a vendor, a security team and a news article can be sure they are talking about the same flaw. Each CVE usually comes with a **CVSS** score from 0 to 10 that estimates how severe it is: how easy it is to exploit, whether it needs a password, and how much damage it allows. The fix is a **security patch**, and the dangerous window is the time between the flaw becoming public and the patch being installed, because attackers read the same announcements and scan the internet for systems that did not update. Worse is a **zero-day**: a flaw attackers already exploit before the vendor knows about it, so there are zero days to prepare. That is why 'install updates' is still the most useful security advice there is. ### An analogy A recall for a car model with faulty brakes. The defect gets an official number so every dealer knows which one it is, it is rated by how dangerous it is, and owners are safe only after they take the car in for the fix. ### Keep in mind - A vulnerability is a flaw that can be exploited, usually a bug with security consequences. - A CVE id gives each public vulnerability one shared name. - CVSS scores severity from 0 to 10; 9 and up is critical. - A zero-day is exploited before a patch exists; updating quickly closes the rest. ### Go deeper - [CVE program overview (cve.org)](https://www.cve.org/About/Overview) - [Common Vulnerability Scoring System (FIRST)](https://www.first.org/cvss/) - [Vulnerability (computer security) on Wikipedia](https://en.wikipedia.org/wiki/Vulnerability_(computer_security)) --- ## Web browser > The app that fetches web pages from the internet and draws them on your screen. - Level: Basic - Topics: Software, Networks - URL: https://tidbit.legios.com.ar/web-browser/ - In Spanish: [Navegador web](https://tidbit.legios.com.ar/es/web-browser/) A **web browser** is the app you use to visit websites: Chrome, Safari, Firefox and Edge are browsers. When you type an address or click a link, the browser looks up which server holds that site (using DNS), asks it for the page over HTTP or HTTPS, and receives a bundle of files back: HTML for the content, CSS for the looks, JavaScript for the behavior, plus images and fonts. Then comes the hard part. The browser's **rendering engine** turns those files into the page you see, working out the position of every word and box, and runs the JavaScript that makes the page interactive. A modern site can involve hundreds of files, and the browser fetches and assembles them in a fraction of a second. Browsers are also a security wall. Each site runs in its own sandbox, so a page cannot read your files or peek at what another tab is doing. The browser shows the padlock for HTTPS, stores passwords and cookies, warns about known phishing sites and asks before a page uses your camera or location. Under the hood only a few engines exist, so many browsers with different names share the same core. ### An analogy A translator and builder in one. It phones the store (the server), receives a box of flat-pack parts and instructions, and assembles the furniture in your living room in an instant. ### Keep in mind - Fetches pages from servers using DNS and HTTP(S), then draws them. - Its rendering engine turns HTML and CSS into pixels and runs JavaScript. - Each site is isolated, so pages cannot read your files or other tabs. - Many browsers share the same few engines underneath. ### Go deeper - [How the web works (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Getting_started/Web_standards/How_the_web_works) - [Browser (MDN glossary)](https://developer.mozilla.org/en-US/docs/Glossary/Browser) - [Web browser on Wikipedia](https://en.wikipedia.org/wiki/Web_browser) --- ## Web hosting > Renting space on an always-on computer so your website is reachable by anyone. - Level: Basic - Topics: Cloud - URL: https://tidbit.legios.com.ar/web-hosting/ - In Spanish: [Hosting web](https://tidbit.legios.com.ar/es/web-hosting/) **Web hosting** is the service of keeping a website's files and programs on a server that is connected to the internet day and night. You could run a site from your own laptop, but it would vanish whenever the laptop slept or the Wi-Fi dropped. A hosting provider rents you space on machines in a data center with stable power, fast connections and people watching them. Hosting comes in several sizes. **Shared hosting** puts many small sites on one server, which is cheap and simple. A virtual private server (VPS) gives you your own slice of a machine to configure. Cloud platforms let you rent exactly the computing you use and grow when traffic spikes. **Static hosting** serves plain HTML, CSS and JavaScript files, often through a CDN, and is ideal for blogs, portfolios and documentation. Hosting is separate from the domain name. The domain is the address people type; the hosting is the place where the site actually lives. You connect them by pointing the domain's DNS records at the host. Many companies sell both together, but you can keep them apart and move your site to another host without changing its address. ### An analogy Renting a shop on a busy street instead of selling from your living room. The landlord keeps the lights on and the doors open; you just stock the shelves. ### Keep in mind - Keeps a website on servers that are always on and online. - Options range from cheap shared hosting to VPS and full cloud platforms. - Domain and hosting are different things, connected through DNS. - Static sites can be hosted very cheaply, often through a CDN. ### Go deeper - [Publishing your website (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Getting_started/Your_first_website/Publishing_your_website) - [What is a web server? (MDN)](https://developer.mozilla.org/en-US/docs/Learn_web_development/Howto/Web_mechanics/What_is_a_web_server) - [Web hosting service on Wikipedia](https://en.wikipedia.org/wiki/Web_hosting_service) --- ## Webhook > A way for one service to call your app the moment something happens. - Level: Intermediate - Topics: Software - URL: https://tidbit.legios.com.ar/webhook/ - In Spanish: [Webhook](https://tidbit.legios.com.ar/es/webhook/) A **webhook** is a message one system sends to another automatically when an event happens. You give a service a URL of yours, and when the event occurs (a payment succeeds, someone pushes code to GitHub, a form is submitted) the service makes an HTTP `POST` request to that URL with the details, usually as JSON. It is an API call in the opposite direction. With a normal API your app asks: "any new payments?" To find out quickly it would have to ask every few seconds, which is called **polling** and wastes effort when the answer is almost always "no". With a webhook your app says once: "call me here when it happens", and then just listens. Since anyone on the internet could send a request to that URL, well-designed webhooks are **signed**: the sender attaches a code computed with a shared secret, and your app checks it before trusting the message. Senders also retry when your app does not answer, so the same event may arrive twice, and good receivers are prepared to ignore duplicates. Tools like Zapier are built largely on webhooks, connecting one service's events to another's actions. ### An analogy Polling is calling the restaurant every five minutes to ask if your table is ready. A webhook is leaving your number so they text you when it is. ### Keep in mind - The service calls you when an event happens; you do not have to ask. - It is just an HTTP POST to a URL you registered. - Check the signature, because anyone can send requests to a public URL. - Events can arrive twice or late, so handle duplicates. ### Go deeper - [About webhooks (GitHub docs)](https://docs.github.com/en/webhooks/about-webhooks) - [Webhook on Wikipedia](https://en.wikipedia.org/wiki/Webhook) --- ## WebSocket > A connection between browser and server that stays open, so both can talk anytime. - Level: Advanced - Topics: Networks, Software - URL: https://tidbit.legios.com.ar/websocket/ - In Spanish: [WebSocket](https://tidbit.legios.com.ar/es/websocket/) Plain HTTP works like a letter exchange: the browser asks, the server answers, and the conversation is over. The server cannot speak first. A **WebSocket** is a connection that stays open after it is created, so either side can send a message at any moment, in both directions, without asking again. It starts as an ordinary HTTP request with a special header that says, in effect, 'let us switch to WebSocket' (the **upgrade handshake**). If the server agrees, the same TCP connection is kept and from then on carries small messages, called frames, instead of full requests and responses. In the browser it is a few lines of code: `new WebSocket("wss://chat.example.com")`, then a function that runs every time a message arrives. The `wss://` prefix means it is encrypted, like HTTPS. It is what makes real-time web apps feel instant: chat apps, live sports scores, shared documents where you see other people typing, trading screens, multiplayer browser games. The old alternative was **polling**, asking the server 'anything new?' every few seconds, which is slower and wastes requests when the answer is no. The cost is that the server has to keep one open connection per user, which takes memory and care when scaling, and connections drop on bad networks, so apps must reconnect. ### An analogy HTTP is sending text messages and waiting for each reply; a WebSocket is a phone call that stays connected, where either person can speak the moment they have something to say. ### Keep in mind - One connection stays open and carries messages both ways. - It begins as an HTTP request that is upgraded; `wss://` is the encrypted version. - It replaces polling for chats, live data, collaboration and games. - Servers must hold one connection per user and clients must handle reconnecting. ### Go deeper - [The WebSocket API (MDN)](https://developer.mozilla.org/en-US/docs/Web/API/WebSockets_API) - [RFC 6455, the WebSocket protocol](https://www.rfc-editor.org/rfc/rfc6455) - [WebSocket on Wikipedia](https://en.wikipedia.org/wiki/WebSocket) --- ## Wi-Fi > Networking over radio waves: the internet without the cable, within a few rooms. - Level: Basic - Topics: Networks - URL: https://tidbit.legios.com.ar/wi-fi/ - In Spanish: [Wi-Fi](https://tidbit.legios.com.ar/es/wi-fi/) **Wi-Fi** is a way of connecting devices to a local network using radio waves instead of cables. An **access point** (usually built into your router) broadcasts a network name, your phone picks it and, with the password, they start exchanging data over the air. Wi-Fi only covers the last few meters: from your device to the router. From the router onwards, the internet is mostly cables and fiber. It is a family of standards called IEEE 802.11, which is why newer versions have friendlier names: Wi-Fi 5, Wi-Fi 6, Wi-Fi 7. Each one is faster and handles crowded rooms better. Two radio bands are common: **2.4 GHz** reaches further and goes through walls better but is slower and crowded (microwaves and Bluetooth live there too), and **5 GHz** (and now 6 GHz) is faster but shorter range. Because it is radio, anyone in range can hear it. That is why the password matters: with WPA2 or WPA3 the traffic between you and the router is encrypted. An open café network has no such protection, so what keeps your bank login private there is HTTPS, not the Wi-Fi. Slow Wi-Fi is also the most common reason a slow internet is not actually the internet: walls, distance and many neighbours sharing the same channel all cost speed. ### An analogy A conversation across a room. Close and quiet, you understand everything; far away, through walls, with thirty people talking at once, you both have to repeat yourselves and it all slows down. ### Keep in mind - Wi-Fi connects your device to the router over radio; the rest of the trip is cables. - 2.4 GHz reaches further; 5 GHz and 6 GHz are faster but shorter range. - WPA2/WPA3 encrypt the air link; on open networks only HTTPS protects you. - Walls, distance and crowded channels are the usual cause of slow internet at home. ### Go deeper - [Wi-Fi (Wikipedia)](https://en.wikipedia.org/wiki/Wi-Fi) - [IEEE 802.11, the standards behind Wi-Fi (Wikipedia)](https://en.wikipedia.org/wiki/IEEE_802.11) --- ## Wi-Fi security > How a Wi-Fi password keeps neighbors out and scrambles what travels through the air. - Level: Intermediate - Topics: Security, Networks - URL: https://tidbit.legios.com.ar/wifi-security/ - In Spanish: [Seguridad Wi-Fi](https://tidbit.legios.com.ar/es/wifi-security/) Wi-Fi is radio, so anyone nearby can pick up the signal. **Wi-Fi security** is the set of standards that decides two things: who is allowed to join the network, and how the data flying between your device and the router is encrypted so that listeners only hear noise. The Wi-Fi password is the visible part of it. The standards came in generations. **WEP** (1999) was the first, and its encryption was broken within a few years; today it offers almost no protection. **WPA** was a quick patch in 2003, and **WPA2** (2004) brought strong AES encryption and became the norm for over a decade. **WPA3** (2018) improves the weak spot of WPA2 at home: with WPA2, someone who records you connecting can try millions of password guesses on their own computer later, while WPA3 makes each guess require a live attempt against the router. WPA3 also protects past traffic even if the password leaks later. For a home network, the practical rules are short. Use WPA3, or WPA2 with AES if some device cannot handle WPA3; never WEP or an open network. Pick a long passphrase rather than a short clever word. Change the router's admin password (it is not the same as the Wi-Fi one), keep its firmware updated and turn off **WPS**, the push-button pairing feature with a history of flaws. A separate guest network keeps visitors and cheap smart gadgets away from your laptops. ### An analogy A conversation in a crowded square. The password decides who can join the circle, and the encryption is a private language the members speak, so passers-by hear sounds but no words. ### Keep in mind - Wi-Fi security controls who joins the network and encrypts the radio link. - WEP is broken; WPA2 with AES is acceptable; WPA3 is the current standard. - A long passphrase matters more than a clever short password. - It only protects the hop to the router; on public networks, HTTPS or a VPN does the rest. ### Go deeper - [Wi-Fi Protected Access on Wikipedia](https://en.wikipedia.org/wiki/Wi-Fi_Protected_Access) - [Wi-Fi security (Wi-Fi Alliance)](https://www.wi-fi.org/discover-wi-fi/security) - [Wired Equivalent Privacy on Wikipedia](https://en.wikipedia.org/wiki/Wired_Equivalent_Privacy) --- ## Xiaomi > A Chinese maker of phones and smart home gadgets, and more recently electric cars. - Level: Basic - Topics: Tech companies, Hardware - URL: https://tidbit.legios.com.ar/xiaomi/ - In Spanish: [Xiaomi](https://tidbit.legios.com.ar/es/xiaomi/) **Xiaomi** was founded in 2010 in Beijing by Lei Jun and a group of partners. It started with software: MIUI, a customized version of Android that it updated every week based on users' comments. A year later it sold its first phone, with high-end parts at a low price, sold mostly online to save on stores and middlemen. Phones, including its cheaper **Redmi** line, are still its biggest business. Around the phone it built an **ecosystem** of connected devices: TVs, fitness bands, robot vacuums, air purifiers, electric scooters, rice cookers, lamps. Many are made by smaller companies that Xiaomi invests in and helps with design and distribution, all controlled from the same app. In 2023 it replaced MIUI with **HyperOS**, a system meant to run across phones, home devices and cars alike. In 2024 it started selling its own electric cars in China. Its business model is unusual: Xiaomi has said it keeps the profit margin on its hardware low and earns more from **internet services**, such as ads, app recommendations and paid content inside its software. The idea is to sell many affordable devices, keep people in its ecosystem and make money over time from the services they use. ### An analogy A supermarket that sells its own brand of almost everything at low prices. You come for cheap milk, and soon your whole kitchen is the same brand, all talking to the same app. ### Keep in mind - Founded in 2010 in Beijing by Lei Jun; started with the MIUI software. - Sells phones, including the Redmi line, mostly at low prices. - Built an ecosystem of smart home gadgets, now running on HyperOS, and since 2024 electric cars. - Keeps hardware margins thin and earns money from internet services. ### Go deeper - [About Xiaomi (official site)](https://www.mi.com/global/about/) - [Xiaomi on Wikipedia](https://en.wikipedia.org/wiki/Xiaomi) --- ## XML > A text format that wraps data in named tags, so both people and programs can read it. - Level: Basic - Topics: Data, Software - URL: https://tidbit.legios.com.ar/xml/ - In Spanish: [XML](https://tidbit.legios.com.ar/es/xml/) **XML** (Extensible Markup Language) is a way to write structured data as plain text, using tags in angle brackets. Each piece of information sits between an opening tag and a closing tag with a name you choose: a `title` tag holding "Dune", inside a `book` tag, inside a `library` tag. The closing tag repeats the name with a slash. Because tags can be nested, XML can describe almost any shape of data as a tree. It was standardized by the W3C in 1998 and became the common language for exchanging data between companies and systems. It is still everywhere, often out of sight: Word and Excel files (`.docx`, `.xlsx`) are zipped folders of XML, RSS feeds and sitemaps are XML, SVG images like the one on this page are XML, and many banks, governments and electronic invoice systems exchange XML documents. For new web APIs, JSON has mostly taken its place, because it is shorter and maps directly to the objects programmers use. XML is more verbose, but it brings tools JSON lacks out of the box, such as schemas that strictly validate a document and attributes inside tags. That rigor is why it survives in formal, long-lived document exchanges. ### An analogy Labeled envelopes inside labeled envelopes: a big one that says "book", holding smaller ones that say "title" and "author". Anyone can open them and know exactly what each piece of paper is. ### Keep in mind - Plain text with named opening and closing tags, nested like a tree. - A W3C standard since 1998, still used for documents, feeds, invoices and SVG. - Office files like .docx and .xlsx are XML inside a zip. - JSON replaced it in most web APIs; XML remains where strict validation matters. ### Go deeper - [XML on Wikipedia](https://en.wikipedia.org/wiki/XML) - [XML introduction (MDN)](https://developer.mozilla.org/en-US/docs/Web/XML/Guides/XML_introduction) - [XML at the W3C](https://www.w3.org/XML/) --- ## Zero trust > Security that checks every request, instead of trusting whatever is inside the network. - Level: Advanced - Topics: Security - URL: https://tidbit.legios.com.ar/zero-trust/ - In Spanish: [Zero trust](https://tidbit.legios.com.ar/es/zero-trust/) For decades, company security followed the **castle and moat** model: a strong wall around the office network (the firewall), a gate to get in from outside (the VPN), and once inside, almost everything was reachable and trusted. That breaks down when one phishing email gives an attacker a foothold inside, when people work from home and cafés, and when the apps live in the cloud rather than in the office. **Zero trust** drops the idea that being on the right network means anything. Every request, to every app, is checked on its own: who is asking (a strong login, usually with two-factor authentication), from what device (is it the company laptop, is it up to date?), and whether this person should have access to this specific thing right now. The rule behind it is **least privilege**: each person and each service gets only the access they need, nothing more. In practice it means an identity provider with SSO in front of every app, device checks, short-lived credentials, fine-grained permissions like the ones IAM gives on AWS, and logs of every access. Google popularized it internally after a major breach in 2009, and the US government's NIST published a reference architecture for it. It is not a product you buy but a direction: every step that removes implicit trust makes a stolen password or an infected laptop do less damage. ### An analogy An office building where your badge is checked at every door, not only at the main entrance. Getting into the lobby no longer means you can walk into the server room; each door asks again whether you, today, should go through. ### Keep in mind - Network location grants nothing: being inside is not a reason to be trusted. - Every request is checked for identity, device and permission. - Least privilege: each person and service gets only the access they need. - It limits the damage of stolen passwords and infected devices instead of assuming they will not happen. ### Go deeper - [NIST SP 800-207, Zero Trust Architecture](https://csrc.nist.gov/pubs/sp/800/207/final) - [Zero Trust on AWS](https://aws.amazon.com/security/zero-trust/) - [Zero trust architecture on Wikipedia](https://en.wikipedia.org/wiki/Zero_trust_architecture)