A VISUAL LABORATORY FOR EXPLANATION
The Explanation Lab
20 ways to make hard things clear. Each one shown by doing it.
Each lesson is a short clip plus a page that uses the method it teaches. Pick any one. They take 2 to 5 minutes each.
Why this exists
On October 1, 2026, Andrej Karpathy posted that we'll spend a lot more time trying to understand what language models produce. He suggested a ladder of better output formats, each one "even better" than the last:
- Writing. Ask for output in ASD-STE100, a controlled language from aerospace manuals.Simplified Technical EnglishPlain languageFeynman-styleELI5
- Diagrams. Instead of writing, ask for a diagram. It can be much easier to take in.GestaltVisual variablesGraphical perceptionTufte
- Web pages. Ask for the answer "in HTML": a page you can read, explore and interact with.Progressive disclosureExternal memoryISO 9241-112ISO/IEC/IEEE 26514
- Explainer videos. Custom explainer videos on any topic, made on request. The format he is most bullish on.3Blue1Brown-styleMayer's principlesSemantic animationCognitive load
Every rung works better when the model knows how to explain. That knowledge already exists, in standards, research and the habits of great explainers. This lab collects 20 of them. Each lesson ends with a prompt you can hand to your own agent.
The 20 lessons
Prefer one long watch?
All 20 are also one video, 11 minutes long. Watch the full video. It goes fast. That's why the lessons exist.