The big picture
Build a useful mental model without starting with the mathematics. Follow the path from text to a response, then learn where that picture stops.
By the endExplain why a fluent response is a prediction, and why a chatbot is more than its underlying model.
Start path: Tokens: the pieces a model reads →5 lessons · 30 min total
Mark lessons as read to save your place on this browser.
- 16 MIN READING · Not read yet
Tokens: the pieces a model reads
Separate words on the screen from the units the model sees.
Read the lesson → - 26 MIN READING · Not read yet
Next-token prediction: a small objective with broad demands
Understand the prediction task before opening the network.
Read the lesson → - 36 MIN READING · Not read yet
Autoregression: one choice becomes the next input
See how one prediction becomes a whole response.
Read the lesson → - 46 MIN READING · Not read yet
Base model versus assistant
Add the product layer around the base model.
Read the lesson → - 56 MIN READING · Not read yet
Hallucinations: fluent text without support
Keep fluency and factual support separate.
Read the lesson →