# 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))

## Related terms

- [Model parameters](https://tidbit.legios.com.ar/model-parameters/): The billions of numbers a model learns in training; '70B' means 70 billion of them.
- [Training and inference](https://tidbit.legios.com.ar/training-and-inference/): Training builds the model once; inference is using it, millions of times a day.
- [Claude](https://tidbit.legios.com.ar/claude/): Anthropic's family of AI models, and the assistant app built on them.
- [Transformer](https://tidbit.legios.com.ar/transformer/): The neural network design behind modern language models, built around 'attention'.