How ChatGPT-style Models Work
ChatGPT and similar assistants feel like magic, but the core trick is simple. They break text into small pieces, then predict what piece should come next, over and over. This lesson shows how that one idea produces whole answers.
New words
- LLM
- Large language model: an AI trained on huge amounts of text.
- Token
- A small piece of text, often a word or part of a word.
- Prediction
- A guess about what comes next.
- Temperature
- A setting that makes the model play safe or take risks.
- Hallucination
- A confident answer that is false or made up.
- Feedback
- People rating answers so the model learns which ones are good.
What you will learn
- You already use a tiny version
- Step 1: text is cut into tokens
- Try it: split a sentence into tokens
- Step 2: predict the next token, then repeat
- Try it: be the model
- How did it learn to do this?
- Why it can sound sure and still be wrong
- Where you see it every day, and where to go deeper
- Explain it to a friend
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.