1Guess first
How many numbers are inside GPT-3?
Each is a setting the model learned, called a parameter. Drag the marker to your guess, then lock it in.
Make your guess
ChatGPT grew from one small idea, drawn on paper in 1958. You can move it with one finger.
Every lesson starts with a picture you push on. The words come after the picture, and the math comes after the words.
Drag the blue handle until every dot agrees.
8 of 28 dots agree
The same boards the lessons use. Each one takes about half a minute, and nothing needs reading first.
1Guess first
Each is a setting the model learned, called a parameter. Drag the marker to your guess, then lock it in.
Make your guess
2Turn a knob
A model is finishing “The capital of France is”. Turn up the temperature until Paris gets 50% to 70% of the chances, while pizza stays under 5%.
Turn the dials
3Cut it up
It reads words in pieces it has seen many times. Cut “unbelievable” into three: a start, a middle and an end.
Cut it up
4Run it
The dot is a model's error. Pick a step size and press Run. Reach the bottom in 20 steps without jumping past it.
Run it
An idea is yours once you have felt it, guessed it, run it, and seen where it breaks.
Drag a weight and watch the line move. Every picture in a lesson answers your hand.
We ask what you think the curve looks like before we draw it. A wrong guess is the one you remember.
Train a real model right on your device, with no Python, no install and no cloud bill.
Story, Build or Deep. No code and no formulas unless you want them, and they are always one tap away.
Native phrasing, right-to-left layout and Arabic-Indic digits.
Ask about the step you are on. It answers with the same pictures you just used.
One road in three stages. The level test finds where you join it.
Stage 1 · Beginner
From one neuron to a network you train yourself.
No background needed. By the end you can explain how a machine learns from examples, and train a small neural network.
11 chapters · 142 lessons
Stage 2 · Intermediate
Open up the networks that see, remember and write.
You pick up the working tools and look inside the transformer. By the end you know what happens inside a chat model.
17 chapters · 216 lessons
Stage 3 · Expert
Build real systems, then explore the open edge of the field.
You build with models, train your own, and rebuild landmark papers. By the end you can pick an open problem and start on it.
15 chapters · 192 lessons
An eight-minute test finds what you already know and skips it.
One idea per screen. A lesson takes five to twelve minutes.
Quizzes as you go, then a chapter exam. Pass it and the chapter is yours.
You keep reading about AI and still feel one step behind.
You ship features with model APIs and want to know what is inside.
You have the math and want to see where it lands.
You want to read papers and reproduce the result yourself.