Credits & onward

The shoulders this stands on

Every chapter ends with the papers it stood on, each one linked to somewhere you can read it without paying. This page is for the other kind of debt: the work that decided what this book would be, and the code it runs on.

• Christopher Olah, Neural Networks, Manifolds, and Topology — the essay behind Bending Space, and the reason this book exists.
• TensorFlow Playground — the original proof that you can feel a neural network through a browser tab.
• Gradient Lab — the optimization laboratory whose landscapes the Prologue descends.
• jax-js — JAX, reborn in TypeScript on WebGPU; the engine under every live plate here.
• cm-chessboard by Stefan Haack (MIT) — every board in the Rook chapter, wearing the Staunty pieces from lichess (CC BY-NC-SA 4.0), with the black set darkened here so it reads on this book's board.
• Fashion-MNIST (Zalando Research, MIT) — the ten classes of clothing the last two chapters learn to draw, a drop-in replacement for MNIST that is still small enough to train on in a browser tab and various enough that a silhouette can be judged by eye.
• Kenneth Li et al., Emergent World Representations (Othello-GPT) and Adam Karvonen’s chess-GPT work — the lineage Rook belongs to: small game-playing transformers that were never shown a board and turn out to carry one anyway.

Onward

Start over from the beginning, or open the hood: everything on this site — every model, chapter, and mistake — lives in its repository. Train something.