Lessons

Each lesson builds on the one before. Sign in to save your progress; or just try one.

1. 1. Values flow through a graphConnect nodes to compute with numbers, then with whole arrays at once.3 steps · ~5 minTry it
2. 2. Fit a lineYour first model: predict y from x with a slope and an offset, measure the error, and let training fix it.4 steps · ~15 minTry it
3. 3. From a line to a neural networkA single layer can only draw straight boundaries. Add a hidden layer and a ReLU to bend them.3 steps · ~20 minTry it
4. 4. Classify handwritten digitsTurn 784 pixels into 10 scores, one per digit, and train it on real handwriting.2 steps · ~20 minTry it
5. 5. Attention: predicting the next characterTurn characters into vectors, predict what comes next, then let each position look back at earlier ones, the core idea of transformers.4 steps · ~25 minTry it
6. 6. Diffusion: making new data from noiseThe idea behind image generators, on 2-D points: blur data with noise, learn to undo one step, then turn pure noise into new data.3 steps · ~25 minTry it