a visual course

How ChatGPT
Actually Works

(Large Language Models)

The one trick behind every chatbot: predicting the next word — in under half an hour.

A free, visual course on how large language models like ChatGPT actually work — from predicting the next word, to tokens, probabilities, temperature, the context window, and why they sometimes make things up. No math required.

in the real world

It's the engine behind chatbots, coding copilots, and the writing and search tools you already use — and understanding it is how you get more out of them and catch them when they bluff.

Start with the next word →

Part 1 — The core trick

1

It's just autocomplete

A chatbot does one thing: guess the next word, add it, and repeat.

2

Tokens, not words

Models read text in chunks called tokens — sometimes a word, sometimes a piece of one.

Part 2 — Inside the prediction

3

A score for every word

At each step the model ranks the entire vocabulary by how likely it is to come next.

4

Careful or creative

A single dial — temperature — decides whether it plays safe or takes risks.

5

How it remembers

The context window is everything the model can see at once — and its hard limit.

Part 3 — Quirks & limits

6

Why it makes things up

It predicts what sounds right, not what's true — so it can be confidently wrong.

Part 4 — Out in the world

7

Where this shows up

Chatbots, coding copilots, customer support, search over your docs, and tool-using agents.