Studying with AI: the four moves that work
Lesson 2 said what not to do. This lesson is the positive version: four concrete uses of an AI assistant that make studying genuinely better rather than merely faster, with the prompts.
Each one shares a design principle. The assistant is arranged so that you are the one doing the retrieving. That is the entire trick, and once you see it you can invent your own.
Move one: explain it at three levels
When you meet something you do not understand, ask for the explanation three times at increasing depth.
"Explain urban heat islands to me as if I have never studied geography. Then explain it again for a second-year urban planning student. Then explain it the way a paper in the field would put it, and tell me which technical terms I now need to know."
Why three: the first gives you a hook to attach things to, the second is the level you are actually working at, and the third is a vocabulary list disguised as an explanation. That third level is what lets you read the literature, and asking for it explicitly saves you the several weeks it otherwise takes to absorb the jargon by osmosis.
Then — this is not optional — close it and write level two from memory.
Move two: ask for questions, not answers
This is the highest-value move in the course, and almost nobody uses it.
"I am studying urban heat islands. Ask me five questions of increasing difficulty, one at a time. Wait for my answer before asking the next. After each, tell me what was missing and do not simply give me the full answer."
One at a time and wait both matter — without them you get five questions and five answers, which is a document, not a session. What you are building is a tutor that quizzes you, which is retrieval practice, which is the mechanism from lesson 2.
The variant Ananya used most: "Ask me questions about my own notes below, and tell me which parts of my understanding are shaky." Feeding your notes in makes the questions land exactly where your gaps are, which no textbook question set can do.
Move three: the Feynman check
Explain the idea to the assistant as if teaching it, in your own words, badly. Then ask it to find the holes.
"I am going to explain urban heat islands to you as if you are a classmate who missed the lecture. Afterwards, tell me what I got wrong, what I left out, and where my explanation would confuse someone. Be blunt."
This works because writing the explanation forces retrieval, and because it surfaces the specific thing you cannot yet articulate — which is reliably the thing you do not actually understand. It is the single fastest way to convert a vague sense of knowing into either real knowledge or a specific question.
Be blunt is load-bearing. Assistants default to encouraging, and encouraging feedback on a wrong explanation is worse than none.
Move four: spaced self-testing
Retrieval works better spread out than crammed. The assistant makes this practical in a way flashcards never quite did.
"From the notes below, write twelve short-answer questions. Number them. Do not include the answers."
Keep the twelve. Answer them from memory tomorrow, then in three days, then in a week — from a blank page, each time, before checking. When you check, ask the assistant to mark your answers against your own notes and tell you which are weak.
That last step is the one that makes this stick, because self-marking is where students are most generous with themselves. An external marker with your notes in hand is neither generous nor tired.
What Ananya's week looks like
Ninety minutes on a paper: read the abstract and conclusion herself, write two sentences on the argument from memory, ask for the hard paragraphs to be explained, do a Feynman check on the main claim, and generate five questions for later. One tool, four moves, and the assistant never once told her what a paper argued before she had formed her own view.
The mid-term revision runs on move four alone: a growing bank of questions from her own notes, answered from a blank page on a schedule.
The anti-pattern to name
If your session consists of asking questions and reading answers, you are not studying. You are browsing. It will feel like studying, for the reasons lesson 2 gave.
The check is simple: at the end of a session, how many times did you write something from your own head? If the answer is zero, the session did nothing, regardless of how long it was or how much you understood while it was happening.
Do this today: run move two for twenty minutes on whatever you are studying now. Five questions, one at a time, answered from memory before checking. Notice how different it feels from reading — that difference is the work.