Stop 01 · 4 lessons · 13 min
What's actually happening when a model answers you: text in, numbers, attention, one token at a time out.
Not started
Learn · the road
AI, one stop at a time.
A road through how AI works and how to build with it: 12 stops, 48 lessons of a few minutes each, in plain words with an analogy and something to try. Start anywhere; the road remembers where you are.
0%
Stop 1 · Start here
Tokens: how text becomes numbers · 3 min
- 12 stops
- 48 short lessons
- 2 h for the whole road
- 112 terms explained
- Part 1 · FoundationsWhat a model is, how it learns, and how to talk to it.You are here
- Stop 02 · 4 lessons · 12 min
From a pile of text to an assistant: pretraining, fine-tuning, and learning from human preferences.
Not started - Stop 03 · 4 lessons · 11 min
Getting reliable results out of a model is a skill with rules: context, structure, examples and settings.
Not started - Part 2 · Building with modelsGiving models knowledge, tools, tests and a budget.Stop 04 · 4 lessons · 10 min
Models don't know your data. Retrieval finds the right pieces and puts them in the prompt at the moment they're needed.
Not started - Stop 05 · 4 lessons · 12 min
Models that act: calling tools, taking many steps, and the new ways things go wrong when they do.
Not started - Stop 06 · 4 lessons · 12 min
How you know a model or a prompt is good, keeps being good, and is better than the last one.
Not started - Stop 07 · 4 lessons · 11 min
The economics and engineering of running models: what you pay for, what makes it slow, and how to make it cheaper.
Not started - Part 3 · Beyond textImages and audio, keeping it safe, and the open-weight world.Stop 08 · 4 lessons · 10 min
Models that see, draw, listen and speak, and how they differ from the text models underneath.
Not started - Stop 09 · 4 lessons · 11 min
The ways models go wrong, from confident nonsense to hijacked agents, and the defences that actually work.
Not started - Stop 10 · 4 lessons · 9 min
Weights you can download, licences that matter, and the ecosystem that moves faster than any single lab.
Not started - Part 4 · Shipping and staying currentTurning all of it into products, and reading what comes next.Stop 11 · 4 lessons · 11 min
What changes when the feature is probabilistic: interface patterns, trust, pricing and data.
Not started - Stop 12 · 4 lessons · 9 min
How to read a paper, a benchmark and a launch post without being fooled.
Not started - End of the road · 112 terms
Every term from every lesson, each explained in a sentence, with a way back to the lesson it comes from. Handy beside the day's news.
48 lessons still ahead