Long Form Explainers
In-depth, interactive explainers on AI and the math and systems knowledge pre-requisite to understand it. Each one builds a topic from the ground up, defines every term the first time it appears, and gives you widgets to poke at instead of walls of text.
All explainers
4 explainers · more comingHow Neural Networks Learn, from the ground up
The one idea underneath all of modern AI: a machine that improves itself from examples. Starting from a single straight line through some house prices, we build up — measuring how wrong a guess is, working out which way to improve, stacking simple functions into a network, and the cycle that learns from examples — until the very same method, pointed at text, becomes a large language model. Assumes no machine-learning background. The natural prerequisite to the LLM Pre-training explainer.
LLM Pre-training, from the ground up
How a large language model is actually built from scratch — starting from the simple game of "predict the next word" and following it all the way up: how the model learns, the hardware tricks that make enormous training runs possible, and the design ideas behind every landmark model from the original 2017 Transformer to the 2026 frontier. Begins from the basics.
LLM Post-training, from the ground up
A raw language model knows a lot but does not yet behave like a helpful assistant. This is the story of how we fix that — teaching it to follow instructions, to learn from people choosing which of two answers is better, and finally to reason its way through hard problems. The whole modern recipe, built up one idea at a time, with no prior background assumed.
LLM Inference
What actually happens when you send a prompt and the model answers — from the instant your words are broken into pieces the model can read, through the computation that picks each next word, to the clever shortcuts real services use to answer millions of people quickly and cheaply.