Studying With AI: Feynman, Mock Exams, Error Hunting

⚠️ Version note: the interface details change; the study techniques don't. Last checked: 24 July 2026. If a menu has moved, the "If it moved" notes point at what to look for.

The wrong way to study with AI, first

Almost everyone uses AI for studying like this: "summarise this chapter", "explain photosynthesis", "give me notes on the French Revolution."

Every one of those makes the AI do the work — and LN-01 is unambiguous that having material handed to you, however clearly, is close to useless for actually remembering it. A perfect summary you read is a perfect summary you'll forget, exactly like a re-read textbook. You've automated the study method that doesn't work.

The whole value of AI for studying is the opposite move: make it test you, not tell you. An AI is an infinitely patient examiner that generates unlimited questions, marks your answers, never makes you feel stupid, and is available at 2am. Used that way it's the best implementation of active recall that's ever existed. Used the lazy way, it actively makes your studying worse.

This lesson builds the good version — and ends with a saved bot you can reuse for every exam.

What you'll have at the end

• Three techniques that use AI for retrieval, not delivery
• A saved "exam bot" — a Project that quizzes you the right way every time
• The one instruction that stops it just handing you answers

Prerequisite: AI-02 (you'll build a Project). The mechanism why this works is LN-01 — this lesson assumes you've met active recall and spaced repetition.

Technique 1 — The mock exam that won't give you the answer (15 min)

The core move. You upload your material and have the AI examine you on it — one question at a time, withholding answers until you've tried.

The instruction that makes or breaks it:

I'm studying [topic]. I've attached my notes / the syllabus.

Act as an examiner. Ask me one question at a time, increasing in
difficulty. Wait for my answer before saying anything.

When I answer:
- If I'm right, confirm briefly and ask the next question.
- If I'm wrong or partial, DO NOT give me the answer. Tell me which
  concept I've missed and ask me to try again.
- Only after two failed attempts, explain it — then ask a fresh
  question testing the same idea.

Never show me the answer before I've attempted it. Keep going until I
say stop. At the end, list the concepts I was weakest on.

The load-bearing line is "do not give me the answer". Without it the AI is helpful in exactly the wrong way — you ask, it tells, you feel informed, you retain nothing. With it, you're forced to retrieve, which is the entire mechanism from LN-01.

Two refinements worth adding:

• "Ask me to explain my reasoning, not just state the answer." A right answer with wrong reasoning is a gap waiting to fail you on a differently-worded question.
• "At the end, tell me which topics to review first." That weak-topics list feeds straight into your spaced repetition schedule (LN-01).

✅ Check: you've done a session where the AI made you attempt every answer before explaining, and it produced a list of your weak spots.