Step 01 / 04

Stacked layers of knobs

Map

You know a model has knobs (parameters). In many modern systems, those knobs are organized into layers stacked on top of each other. That stack is a neural network.

Think of an assembly line: each station changes the product a little. Here each layer changes the numbers a little, then hands them to the next layer.

An LLM is (mostly) a very deep version of this idea.

1

Input

what comes in

passes down ↓
2

Layer 1

mix with knobs

passes down ↓
3

Layer 2

mix again

passes down ↓
4

Output

guess / scores

Knobs live inside the layers. Depth = more stations on the line.

A neural net is a stack