Learn AI

Track progress · learn offline

Open

Glossary

What is a neural network?

The structure behind most modern AI: layers of simple connected units that pass numbers forward, turning an input into an output. What it “knows” lives in the connection strengths, learned from data.

Loosely inspired by the brain: lots of tiny units (“neurons”) arranged in layers. Each unit takes numbers in, weighs them, and passes a number on. Information flows from the input, through one or more hidden layers, to the output.

Input · your data
Hidden layer
Hidden layer
Output · prediction

Each layer transforms the numbers a little; stack enough of them and the network can capture very complex patterns.

The network learns by nudging its connection weights — the parameters — during training until its outputs match the examples. The transformer, the architecture behind modern language models, is one especially powerful kind of neural network.

Related terms

Appears in these lessons

Look inside

Open the black box

How AI Really Works takes the network apart layer by layer — tokens, attention and all — so it stops feeling like magic.