Neuro 13
0 / 37 chapters finished
Part 5Your turn9 min read

Chapter 37: Where to go next

The best videos, courses and books for after this one.

This book ends here. What you can do with it doesn't. Below is a short list of the best places to go next, sorted by what you feel like doing: watching, building, reading, or filling in the math and the Python. It leaves out far more than it includes. Everything on it is free to use, and every link was opened and checked in November 2026.

Each entry says how hard it is, in one of three ways. Gentle means you could start today. A stretch means it expects some Python or some patience, and repays both. For later means it was written for university students. It is good to know those exist, and no reason to feel bad if you bounce off one now.

All of these links leave this book for other people's websites, and several are YouTube videos. If you aren't sure you're allowed, ask first.

If you want to watch

If you want to build

  • Teachable Machine, from Google. Train a classifier on your own pictures, sounds or poses, in the browser, with no code at all. Gentle. It uses your camera or microphone, and the site says you can choose to keep everything on your own device. It continues chapter 27.
  • TensorFlow Playground. A famous neural network toy that runs in the browser: dots, layers and a play button. Gentle. It continues chapter 33. Its hardest dataset is a spiral, like the one in chapter 17. See how few neurons you can solve it with.
  • CNN Explainer and Transformer Explainer. Two interactive pages. The first shows the numbers inside a small image network as it looks at a photo. The second shows a real small GPT-2 model, with 124 million weights, choosing its next word, and it has the temperature and top-k knobs you already know. Its ready-made examples work straight away. To try a sentence of your own it downloads the whole model, about 600 megabytes, so wait until you are on Wi-Fi. Gentle to poke at, a stretch to follow completely. They continue chapter 28 and chapter 22.
  • "Neural Networks: Zero to Hero", by Andrej Karpathy. Eight long videos, most of them about two hours, in which he types out every line of a neural network starting from nothing. A stretch: he asks for solid Python and a little calculus. The first video builds the same kind of engine you built in chapter 32. The second builds a name inventor from the very list of names that chapter 21 borrowed. The seventh, "Let's build GPT", builds a small transformer that writes, which is chapter 22 and chapter 23 made real. If you do one hard thing after this book, do this.
  • Practical Deep Learning for Coders, from fast.ai. A course of nine lessons, about 90 minutes each, that has you training a real image classifier almost at once and explains the insides afterwards. A stretch: it asks for about a year of coding experience. It continues chapter 32.

If you want to read

  • "The Illustrated Transformer", by Jay Alammar (2018). One long page of drawings that walks through a whole transformer, piece by piece. A stretch. It continues chapter 22.
  • "What Is ChatGPT Doing … and Why Does It Work?", by Stephen Wolfram (2023). A very long essay whose first heading is "It's Just Adding One Word at a Time". Gentle at the start, a stretch by the end. It continues chapter 23.
  • Distill. A journal of carefully illustrated explanations, published from 2016 to 2021. Nothing new has appeared since, but the old articles still have some of the best pictures of what goes on inside an image network. Start with "Feature Visualization". A stretch. It continues chapter 28 and the "Looking inside" section of chapter 35.
  • Neural Networks and Deep Learning, by Michael Nielsen (2015). A free online book that builds a digit reader in Python and explains backpropagation with great care. A stretch: it has real math in it. It was finished before transformers existed. It continues chapter 27 and chapter 14. One warning: the site is old and has no https, so your browser may refuse to open it or ask you first. It is a book to read, with nothing to type in.
  • Understanding Deep Learning, by Simon Prince (2023). A modern textbook with a free PDF and 68 Python notebooks to go with it. For later. It is a university book, and the best one to grow into.
  • Deep Learning, by Ian Goodfellow, Yoshua Bengio and Aaron Courville (2016). The classic textbook, free to read online. For later. It came out a year before the transformer, so read it for the foundations and not for the news.

If you want the math

You got through this book on arithmetic and graphs. To go further you will want four things, in this order:

  1. Algebra and graphs. Variables, functions, and the slope of a line. Everything else stands on this. It is the language of chapter 1 and chapter 6.
  2. Probability and a little statistics. Chances that add up to 1, averages, and logarithms. That is softmax in chapter 10 and the loss in chapter 21.
  3. Vectors and matrices. Lists and grids of numbers, dot products, and multiplying a grid by a list. The subject is called linear algebra, and it is every layer in chapter 18 and every match score in chapter 22.
  4. Calculus. Mainly derivatives and the chain rule: the slope of a curve, and how slopes multiply along a chain. That is chapter 13 and chapter 14.

You do not have to finish that list before you build anything. Most people learn the math when a project makes them need it, and it sticks better that way.

If you want to learn Python

You have been reading and changing Python all through this book. Writing your own from a blank page is a different skill, and it is the one that opens up everything in the "build" list.

Then go back to this book's own scripts. They are all in one zip file, and Build your own explains how to run them on your own computer.

Five habits for learning on your own

  1. Build small things. A tiny network that you finished teaches you more than a big one you gave up on.
  2. Change one thing at a time. If you change three settings and the result gets better, you have learned nothing about any of them. That was the lesson of chapter 33.
  3. Read other people's code. Start with the scripts in this book. Type them out instead of pasting them, and you'll notice every line.
  4. Explain it to someone. A friend, a parent, a rubber duck. The place where your explanation gets vague is the part you don't understand yet.
  5. Expect to be confused. Being confused doesn't mean you are bad at this. It is what learning something hard feels like from the inside.

You don't need permission, a degree or a powerful computer to keep going. You need a small project, the habit of changing one thing at a time, and the patience to stay confused for a while.

That's the book

You started with one neuron and three numbers. You now know how the machines that read, write, see and draw are put together, and where the people who build them have run out of answers.

If you haven't picked a project from Build your own yet, pick one. It is the best next step on this whole page. And when every chapter is ticked off, your certificate is waiting. Thank you for reading. Go and build something.