The principle
Learning happens when *you* retrieve, not when a model generates. So a good study prompt puts you in the position of producing the answer and the model in the position of marking it. Every prompt below inverts the usual direction.
Prompts that make you work
- "Ask me one question at a time about X. Don't tell me the answer — tell me what I missed after I answer." This is retrieval practice with a marker attached, and it is the single most valuable prompt on this page.
- "Here is my explanation of X. What did I get wrong, and what did I leave out?" Feynman, with an examiner.
- "I think the answer is Y because Z. Attack my reasoning." Forces you to commit before you're corrected — the generation effect.
- "Give me a problem like this one but harder, and don't show me the solution." Unlimited practice, no answer key to peek at.
- "Keep asking me 'why' until I can't answer." Finds the edge of your understanding in about four minutes.
- "What's the most common misconception about X, and do I have it?" Then answer honestly.
Prompts that feel productive and aren't
- "Summarise this chapter." You'll read it once, feel informed, and retain nothing.
- "Make me flashcards about X." Cards from the model's general knowledge, not your syllabus, unverifiable, possibly wrong — and spaced repetition will make any error permanent.
- "Explain X simply." Fine once. But if this is your whole study method, you are reading, and reading is not studying.
- "Write me a study plan." You'll get a beautiful plan you won't follow. Plans aren't the problem.