a visual course

How AI
Pays Attention

(Transformers & Attention)

The one idea behind every modern AI model — attention — visualized from scratch, in about half an hour.

A free, visual course on attention and the Transformer — the mechanism behind every modern LLM. Why words need context, the query–key–value operation, self-attention, multi-head attention, and the full architecture.

in the real world

Attention is the 'T' in GPT — and it also powers translation, vision transformers, AlphaFold's protein folding and code assistants. It's the engine of the whole AI boom.

Start with a single word →

Part 1 — The problem

1

Words need context

A word's meaning depends on the others around it — so every word needs a way to look at the rest.

Part 2 — The mechanism

2

Query, key, value

Each word asks a question, every word answers, and the best matches get blended into a new meaning.

3

Everyone looks at everyone

Run that operation for every word at once and you get an attention matrix — the heart of a transformer.

4

Many heads, many patterns

Several attention heads run in parallel, each learning to track a different kind of relationship.

Part 3 — The architecture

5

The Transformer

Stack attention with a few extras — positions, feed-forward layers — and you have the engine behind every LLM.

Part 4 — Out in the world

6

Where attention shows up

LLMs, translation, vision, protein folding, code — the transformer took over far more than text.