How to learn fast and remember what you learned

Fast learning is the method that leaves the least to re-learn, and you can only see which method that is at a retest a week later.

By Ojasvi YadavPublished 6 September 2026

Fast learning is measured a week later

Learning speed is what you can still do at the retest, so measure it a week after the session rather than at the end of it. Roediger and Karpicke (2006) gave students prose passages, then had one group take recall tests and the other restudy the same text. The final test came after 5 minutes, 2 days, or 1 week. At 5 minutes the restudy group scored higher. At 2 days and at 1 week the tested group retained substantially more. One session produced opposite rankings depending on when the researchers looked.

Pick the interval you care about, then judge methods on it. If you are learning a field you intend to work in, the interval is months. Any method that wins at 5 minutes and loses at 1 week has booked you a second pass through the same material, and that second pass never appears in the hours you logged today.

The fastest method is therefore the one with the least re-learning in it: an order that never skips a prerequisite, one lesson written for what you already know, a test on the same day, an explanation on the next day, and spaced reviews after that.

Why rereading feels fast and is slow

Rereading feels fast because fluent text is easy to process, and you read that ease as knowing. Roediger and Karpicke (2006) recorded that restudying raised students' confidence while it lowered their retention at 2 days and 1 week. Confidence and retention moved in opposite directions.

Karpicke and Blunt (2011) measured that gap on a one-week test. Students who practiced retrieval scored 0.67 against 0.45 for students who built concept maps, about 50 percent better. 101 of the 120 students did better after retrieval practice. Before the test, 75 percent had predicted that concept mapping would be equal or better. Most students chose the weaker method and expected it to win.

Dunlosky and colleagues (2013) reviewed ten common study techniques and rated rereading, highlighting, underlining, and summarization low utility. Practice testing and distributed practice were the only two rated high. That ranking is the starting point for the best way to learn anything technical.

Stop treating fluency as a measurement. Close the page and write what it said. Whatever you cannot produce from memory is the part you have not learned, and today is the cheapest day to find that out.

The fastest order is the smallest step

The fastest order puts each new concept one step away from something you already hold, because a missing prerequisite converts reading time into re-reading time. When a lesson uses a term you do not have, you stop, look it up, absorb a partial answer, and come back with the thread broken. The paragraph then needs a second pass, and often a third. None of that work teaches the new concept, and it can cost more than the concept itself.

Find your starting point before the first lesson rather than after the first failure. A few placement questions do it. Ask three broad questions that cover the spine of the field, then up to six targeted questions only where the broad answers left a real gap. Write the answers down. A recorded starting point stops the next lesson from teaching what you already hold, and stops it from assuming what you do not.

Then order the topics so nothing appears before its prerequisites, and take them in that order even when a later topic looks more interesting. Two learners can read the same number of pages in a week and finish with different amounts, because one of them spent half the week rebuilding background the other already had.

A lesson written for your prior knowledge beats a chapter written for everyone

A lesson written for one reader drops the paragraphs you do not need and adds the step you are missing, and the measured result is more learning in less time. Kestin and colleagues (2025) ran a randomized trial with 194 students in a Harvard introductory physics course in fall 2023. One group worked with an AI tutor designed around pedagogical best practices, the other took an active-learning class on identical material. Median learning gains in the AI group were more than double, the students spent less time, and they reported more engagement and motivation.

Written for you means three concrete things. It starts from your recorded prior knowledge, so it defines the terms you lack and skips the ones you hold. It carries one worked example through the whole lesson, so the example itself becomes something you know. It uses the field's own terminology from the first page, because you will read that field's papers and error messages later.

A general chapter cannot do any of this. It is written for the median reader of a large group, and on most pages you are not that reader. The conditions that make an AI-written lesson worth reading are set out in the guide on how to learn with AI.

Test the same day, then explain it back after sleep

Two tests on two days measure two different things, so run both. The same-day quiz measures whether the lesson landed at all. Recognition is still available and the wording is still fresh, so a failure here tells you the lesson was wrong for you and needs rewriting. The next-day explanation measures whether the concept survived a night. You write it in your own words with the lesson closed, and a grader marks it against the lesson.

Sleep is why the second test earns its day of delay. Diekelmann and Born (2010) describe the consolidation of declarative memory during sleep, with memories reactivated and stabilized in slow-wave sleep. An explanation you give ten minutes after reading is partly a readback of short-term memory. The same explanation the next morning reads what was kept.

Treat a failed explanation as an instruction to teach the concept again. Rereading restores your fluency and not your ability to state the mechanism. Have the concept taught again from a different angle, then tested again the next day. The graded explain-back is the Feynman technique with AI, run on a fixed schedule.

Let the review schedule do the remembering

Once a concept passes both tests, hand the remembering to a spaced schedule and stop rereading it. Cepeda and colleagues (2006) synthesized the distributed practice literature and found that spaced study beats massed study across retention intervals, and that the best gap between sessions grows with the retention interval you want.

Murre and Dros (2015) replicated Ebbinghaus' forgetting curve and located most forgetting in the first hours and the first day. The same-day quiz and the next-day explanation already sit inside that window. The card schedule covers everything after it.

Keep the cards small. Two to four per concept is enough: the definition, the trap you fell into, and one mechanism or number. Add one card for every question you asked while learning, because each question marks a place where your model was incomplete. Ten minutes a day clears the queue at a normal rate of new concepts.

Retention then has its own measurement. The intervals and lapses in your review history show which concepts are holding and which are decaying, which is a different question from how confident you feel. Spaced repetition with AI covers how to write those cards from a lesson you just passed.

Where AI makes learning faster, and where it makes it slower

AI speeds up two steps, writing the lesson and grading your answer, and it slows you down the moment it supplies an answer you were about to produce yourself. Bastani and colleagues (2025) ran a field experiment with nearly 1,000 high-school math students in Turkey in the 2023 to 2024 school year, with two interfaces on the same material. A ChatGPT-like interface raised practice scores 48 percent and lowered scores on the later unassisted exam by 17 percent against students who had no AI. A version prompted to give teacher-designed hints instead of answers raised practice scores 127 percent and left exam scores statistically indistinguishable from the control group.

Students in the first group copied answers and did not perceive the harm, which is the confidence trap from rereading, delivered faster.

The rule is narrow. Use AI to write the lesson for your prior knowledge, to generate the questions, and to grade what you produced. Keep it out of any step you are in the middle of. During practice, ask for a hint or a mark, and finish the step yourself.

The Learning Harness is my implementation of this routine. It asks what you already know, orders the topics so nothing precedes its prerequisites, writes each lesson for one learner, quizzes you the same day, grades your explanation the next day, and writes the cards once a concept passes both. The method sets out those rules in the order a learner meets them.

Sources

1. Roediger, H. L., & Karpicke, J. D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3), 249–255. https://doi.org/10.1111/j.1467-9280.2006.01693.x
2. Karpicke, J. D., & Blunt, J. R. (2011). Retrieval Practice Produces More Learning than Elaborative Studying with Concept Mapping. Science, 331(6018), 772–775. https://doi.org/10.1126/science.1199327
3. Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., & Willingham, D. T. (2013). Improving Students' Learning With Effective Learning Techniques. Psychological Science in the Public Interest, 14(1), 4–58. https://doi.org/10.1177/1529100612453266
4. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning. Proceedings of the National Academy of Sciences. https://doi.org/10.1073/pnas.2422633122
5. Diekelmann, S., & Born, J. (2010). The memory function of sleep. Nature Reviews Neuroscience, 11, 114–126. https://doi.org/10.1038/nrn2762
6. Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed practice in verbal recall tasks: A review and quantitative synthesis. Psychological Bulletin, 132(3), 354–380. https://doi.org/10.1037/0033-2909.132.3.354
7. Kestin, G., Miller, K., Klales, A., Milbourne, T., & Ponti, G. (2025). AI tutoring outperforms in-class active learning. Scientific Reports, 15, 17458. https://doi.org/10.1038/s41598-025-97652-6
8. Murre, J. M. J., & Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLoS ONE, 10(7), e0120644. https://doi.org/10.1371/journal.pone.0120644