{"course":{"slug":"ai-for-students","title":"AI for Students","url":"https://addaly.com/learn/ai-for-students","level":"beginner","track":"work"},"lesson":{"slug":"make-it-quiz-you","title":"Make It Quiz You","url":"https://addaly.com/learn/ai-for-students/make-it-quiz-you","position":20,"of":73,"minutes":7,"answer":"A model that asks you questions is worth more than one that answers them.","keyIdea":"A model that asks you questions is worth more than one that answers them."},"body":{"format":"markdown","text":"## The single highest-return change\n\nMost students use AI as an answering machine. The version that actually moves exam scores is the one where it asks and you answer.\n\nThe reason is one of the best-supported findings in the study of memory. Roediger and Karpicke (2006) had one group reread a passage four times, and another read it once then try to recall it repeatedly. Tested five minutes later, the rereaders won. Tested a week later, the recall group remembered far more: the gap was large, roughly 60% against 40% recall in one of their experiments. And the rereaders had confidently predicted the opposite.\n\nThis is called the testing effect, and \"testing\" is a bad name for it, because it makes it sound like measurement. It is not measurement. **The act of pulling something out of your head changes how available it is next time.** A quiz is not a check on studying. It is studying.\n\nAI is unusually good at this job, because the limiting factor was always that somebody has to write the questions.\n\n## The prompt that works\n\nThe key details are: one at a time, wait, no hints, and grade against what you actually said.\n\n```text\nHere are my lecture notes on the Krebs cycle.\n\nAsk me one question at a time. Wait for my answer before\nthe next one. Do not give hints unless I ask. After each\nanswer, tell me only whether I'm right and what I left out.\nDon't teach. Ten questions.\n```\n\nWhat you must watch for: after two or three rounds, the model will start volunteering explanations, and your session will quietly turn back into reading. Re-anchor with `Stop explaining. Next question.`\n\nSome variants worth having:\n\n```text\nAsk me the question my exam is most likely to ask that\nmy notes would not prepare me for.\n```\n\n```text\nI'm going to explain photosynthesis to you as if you're 12.\nInterrupt at the first point where what I say stops being\ntrue, and don't correct it — just tell me where you stopped.\n```\n\nThat second one is the Feynman check, and the stopping point is the whole value. It locates the exact sentence where your understanding runs out, which is almost never where you would have guessed.\n\n## Writing your own questions beats having them written\n\nIf you have time for only one thing, do this instead. Read a chapter, close it, and write eight exam questions on it. Then hand your questions to the model and ask which important ideas you failed to ask about.\n\nThis works for two reasons. Generating something yourself makes it stick harder than reading the same thing: the generation effect. And to write a good question you have to know what the chapter's load-bearing ideas were, which is a stiffer test than answering.\n\nThe model's job here is to find your blind spots, not to do the thinking.\n\n## Two honest warnings\n\n**It will get facts wrong, and you cannot audit it.** You are asking a machine to quiz you on material you do not yet know, which is exactly the situation where you cannot catch its errors. So: quiz from *your own* notes and set texts, pasted in, rather than from the model's general knowledge. And when it corrects you on something and you feel surprised, check the textbook before you update. Surprise is the signal.\n\nThis matters more the further you get from big, well-covered subjects. Introductory biology in English: mostly reliable. Your national curriculum's specific treatment of a local legal code, in a language with less material online: much less so.\n\n**A quiz you pass immediately after reading proves almost nothing.** The material is still sitting in working memory. Real retrieval practice happens the next day, or three days later. Same questions, cold. That is when the answer either comes or does not, and that is the answer that predicts the exam.\n\n## What a week looks like\n\n- **Day 0:** read the chapter. Write eight questions from memory.\n- **Day 1:** answer your own questions cold. Get the model to mark you against the source text.\n- **Day 3:** only the ones you missed.\n- **Day 8:** all eight again, plus whatever came up in the lecture since.\n\nFour short sessions, maybe eighty minutes total. It will feel less productive than eighty minutes of rereading with a highlighter. That feeling is the point: it is the difficulty doing the work."},"check":{"question":"You reread a chapter three times and feel confident. A friend reads it once, then answers twenty questions from memory, gets many wrong, and feels shaky. Who is likely to know more a week later, and why?","options":[{"text":"The friend, because attempting recall changes what you can retrieve later; your confidence was fluency with the page rather than access to the content.","correct":true},{"text":"You, because you covered the material three times and repeated exposure is what builds memory, while your friend saw it once.","correct":false,"tempting":"Treats number of exposures as the mechanism, which is what rereading maximises and what the testing-effect studies show is the weaker route."},{"text":"Neither reliably, because the friend's high error rate means they encoded wrong answers, and wrong answers are worse than no answers.","correct":false,"tempting":"A reasonable fear of practising errors, but errors that get corrected do not stick as errors: the attempt still does the work."},{"text":"The friend, but only because twenty questions happened to cover more of the chapter than three readings did.","correct":false,"tempting":"Attributes the gain to coverage rather than to the act of retrieval, which would predict that reading twenty answers works equally well."}]},"prev":{"slug":"getting-unstuck-not-finished","title":"Getting Unstuck, Not Getting Finished","url":"https://addaly.com/learn/ai-for-students/getting-unstuck-not-finished"},"next":{"slug":"building-a-hint-ladder","title":"Building a hint ladder","url":"https://addaly.com/learn/ai-for-students/building-a-hint-ladder"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Make It Quiz You', in AI for Students. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-students/make-it-quiz-you (Accessed: 31 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Make It Quiz You","courseTitle":"AI for Students","module":{"index":3,"slug":"study-that-raises-difficulty","title":"Study that raises the difficulty"},"answer":null,"keyIdea":"A model that asks you questions is worth more than one that answers them.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"the-single-highest-return-change","runs":[{"text":"The single highest-return change"}]},{"t":"p","runs":[{"text":"Most students use AI as an answering machine. The version that actually moves exam scores is the one where it asks and you answer."}]},{"t":"p","runs":[{"text":"The reason is one of the best-supported findings in the study of memory. Roediger and Karpicke (2006) had one group reread a passage four times, and another read it once then try to recall it repeatedly. Tested five minutes later, the rereaders won. Tested a week later, the recall group remembered far more: the gap was large, roughly 60% against 40% recall in one of their experiments. And the rereaders had confidently predicted the opposite."}]},{"t":"p","runs":[{"text":"This is called the testing effect, and “testing” is a bad name for it, because it makes it sound like measurement. It is not measurement. "},{"text":"The act of pulling something out of your head changes how available it is next time.","b":true},{"text":" A quiz is not a check on studying. It is studying."}]},{"t":"p","runs":[{"text":"AI is unusually good at this job, because the limiting factor was always that somebody has to write the questions."}]},{"t":"h2","id":"the-prompt-that-works","runs":[{"text":"The prompt that works"}]},{"t":"p","runs":[{"text":"The key details are: one at a time, wait, no hints, and grade against what you actually said."}]},{"t":"code","lang":"text","label":"Text","text":"Here are my lecture notes on the Krebs cycle.\n\nAsk me one question at a time. Wait for my answer before\nthe next one. Do not give hints unless I ask. After each\nanswer, tell me only whether I'm right and what I left out.\nDon't teach. Ten questions."},{"t":"p","runs":[{"text":"What you must watch for: after two or three rounds, the model will start volunteering explanations, and your session will quietly turn back into reading. Re-anchor with "},{"text":"Stop explaining. Next question.","code":true}]},{"t":"p","runs":[{"text":"Some variants worth having:"}]},{"t":"code","lang":"text","label":"Text","text":"Ask me the question my exam is most likely to ask that\nmy notes would not prepare me for."},{"t":"code","lang":"text","label":"Text","text":"I'm going to explain photosynthesis to you as if you're 12.\nInterrupt at the first point where what I say stops being\ntrue, and don't correct it — just tell me where you stopped."},{"t":"p","runs":[{"text":"That second one is the Feynman check, and the stopping point is the whole value. It locates the exact sentence where your understanding runs out, which is almost never where you would have guessed."}]},{"t":"h2","id":"writing-your-own-questions-beats-having-them-written","runs":[{"text":"Writing your own questions beats having them written"}]},{"t":"p","runs":[{"text":"If you have time for only one thing, do this instead. Read a chapter, close it, and write eight exam questions on it. Then hand your questions to the model and ask which important ideas you failed to ask about."}]},{"t":"p","runs":[{"text":"This works for two reasons. Generating something yourself makes it stick harder than reading the same thing: the generation effect. And to write a good question you have to know what the chapter’s load-bearing ideas were, which is a stiffer test than answering."}]},{"t":"p","runs":[{"text":"The model’s job here is to find your blind spots, not to do the thinking."}]},{"t":"h2","id":"two-honest-warnings","runs":[{"text":"Two honest warnings"}]},{"t":"p","runs":[{"text":"It will get facts wrong, and you cannot audit it.","b":true},{"text":" You are asking a machine to quiz you on material you do not yet know, which is exactly the situation where you cannot catch its errors. So: quiz from "},{"text":"your own","i":true},{"text":" notes and set texts, pasted in, rather than from the model’s general knowledge. And when it corrects you on something and you feel surprised, check the textbook before you update. Surprise is the signal."}]},{"t":"p","runs":[{"text":"This matters more the further you get from big, well-covered subjects. Introductory biology in English: mostly reliable. Your national curriculum’s specific treatment of a local legal code, in a language with less material online: much less so."}]},{"t":"p","runs":[{"text":"A quiz you pass immediately after reading proves almost nothing.","b":true},{"text":" The material is still sitting in working memory. Real retrieval practice happens the next day, or three days later. Same questions, cold. That is when the answer either comes or does not, and that is the answer that predicts the exam."}]},{"t":"h2","id":"what-a-week-looks-like","runs":[{"text":"What a week looks like"}]},{"t":"ul","items":[[{"text":"Day 0:","b":true},{"text":" read the chapter. Write eight questions from memory."}],[{"text":"Day 1:","b":true},{"text":" answer your own questions cold. Get the model to mark you against the source text."}],[{"text":"Day 3:","b":true},{"text":" only the ones you missed."}],[{"text":"Day 8:","b":true},{"text":" all eight again, plus whatever came up in the lecture since."}]]},{"t":"p","runs":[{"text":"Four short sessions, maybe eighty minutes total. It will feel less productive than eighty minutes of rereading with a highlighter. That feeling is the point: it is the difficulty doing the work."}]}],"check":{"question":[{"text":"You reread a chapter three times and feel confident. A friend reads it once, then answers twenty questions from memory, gets many wrong, and feels shaky. Who is likely to know more a week later, and why?"}],"options":[{"text":[{"text":"The friend, because attempting recall changes what you can retrieve later; your confidence was fluency with the page rather than access to the content."}],"correct":true,"tempting":null},{"text":[{"text":"You, because you covered the material three times and repeated exposure is what builds memory, while your friend saw it once."}],"correct":false,"tempting":[{"text":"Treats number of exposures as the mechanism, which is what rereading maximises and what the testing-effect studies show is the weaker route."}]},{"text":[{"text":"Neither reliably, because the friend’s high error rate means they encoded wrong answers, and wrong answers are worse than no answers."}],"correct":false,"tempting":[{"text":"A reasonable fear of practising errors, but errors that get corrected do not stick as errors: the attempt still does the work."}]},{"text":[{"text":"The friend, but only because twenty questions happened to cover more of the chapter than three readings did."}],"correct":false,"tempting":[{"text":"Attributes the gain to coverage rather than to the act of retrieval, which would predict that reading twenty answers works equally well."}]}],"order":[2,3,0,1],"why":null},"prev":{"slug":"getting-unstuck-not-finished","title":"Getting Unstuck, Not Getting Finished","minutes":8},"next":{"slug":"building-a-hint-ladder","title":"Building a hint ladder","minutes":8}}}