Neuro 13
0 / 37 chapters finished

How a neural network actually works

A hands-on book for curious 13-year-olds. Start with one tiny neuron. Finish having trained networks that read handwriting, invent names, sort the news and turn noise into pictures. Everything runs right here in your browser, with nothing to install.

  • 37 chapters
  • 77 things to play with
  • 22 real Python programs

What you'll build

Every one of these is a real model that learns in front of you, from random numbers, on your own device.

The whole book

37 chapters in five parts. Go in order: each one builds on the one before.

Part 1

One neuron

Meet the tiny machine everything else is built from, teach it a rule, then find the puzzle it can't solve.

  1. 1Tiny math mapThe handful of symbols this whole book is written in.
  2. 2The big pictureWhat a neural network is, before any detail.
  3. 3One tiny neuronMultiply, add, squish. Drag the sliders and watch it decide.
  4. 4Training the neuronShow it examples, nudge the weights, repeat.
  5. 5What training actually changesA trained model is its numbers and nothing else.
  6. 6Tables, dots, and linesWhy a neuron's decision is always a straight line.
  7. 7When one neuron isn't enoughFour dots that no single neuron can separate, and the research slump they came to stand for.
  8. 8Hidden neurons fix itAdd a middle layer and the impossible becomes easy.
  9. 9Checkpoint: one neuronWhat to remember, a mixed quiz, and your first build-it project.
Part 2

How networks learn

More answers, more layers, real data, and the difference between learning a rule and memorizing the homework.

  1. 10Picking from a listFrom yes/no to choosing one of many.
  2. 11What does that "squish" actually do?Sigmoid, ReLU, and why the bend matters.
  3. 12Meal guesserA network that has to combine clues.
  4. 13How learning actually worksRolling downhill on the loss landscape.
  5. 14Backpropagation: passing the blame backwardsThe one algorithm that trains every network, stepped through with real numbers.
  6. 15Training on a real dataset150 real flowers and the four habits of honest training.
  7. 16Learning vs. memorizingWatch a network ace the homework and fail the test.
  8. 17Stacking hidden layersEight neurons, one layer or two: the spiral showdown.
  9. 18Scaling this upWhat changes between 9 weights and 9 billion.
  10. 19Checkpoint: learningWhat to remember, a mixed quiz, and a project with your own data.
Part 3

Machines that read and write

Turn text into numbers, train a model that invents words, and see what a chatbot is really doing.

  1. 20Tokens: turning text into numbersBuild a tokenizer's dictionary one merge at a time.
  2. 21A language model you can trainTeach a network to invent names, one letter at a time.
  3. 22Attention: a fuzzy lookupWatch a network learn where to look.
  4. 23How AI chatbots actually writeNext-token prediction, sampling, and the training that makes it polite.
  5. 24A model that decides instead of writesTrain a decision model on 10,000 real headlines.
  6. 25When models go wrongMade-up facts, skewed data, and how to judge a model honestly.
  7. 26Checkpoint: languageWhat to remember, a mixed quiz, and a classifier for your own messages.
Part 4

Seeing, creating, playing

Networks that read handwriting, turn noise into pictures, and learn a game by trial and error.

  1. 27Read my handwritingTrain a real digit reader, then draw for it.
  2. 28How networks seeSlide a tiny filter over a picture and watch edges light up.
  3. 29Pictures from noiseThe trick behind AI image generators, small enough to watch.
  4. 30Learning by trial and errorNo right answers given: just rewards. Watch a creature learn a maze.
  5. 31Checkpoint: seeing and creatingWhat to remember, a mixed quiz, and a project you can show off.
Part 5

Your turn

The tools professionals use, a lab to experiment in, a map of the frontier, and projects of your own.

  1. 32What the professionals useBuild your own mini-PyTorch in 60 lines, then read the real thing.
  2. 33The playgroundPick the data, the layers and the knobs. Break it. Fix it.
  3. 34The story so farEighty-three years on one line, and you have built a small version of almost all of it.
  4. 35The frontierModels that think first, tools and agents, looking inside, and what nobody knows yet.
  5. 36Build your ownThree capstone projects, from an afternoon to a weekend.
  6. 37Where to go nextThe best videos, courses and books for after this one.