Test yourself instead of rereading: how memory researchers beat forgetting

The Short Version

Rereading feels like studying, but the experts here say the thing that makes knowledge last is pulling it back out of your own head and finding out where you were wrong. Andrew Huberman, reviewing more than a century of research, calls self-testing the single most powerful and underused study tool, especially when it is done soon after you first meet the material [1]. Memory researcher Charan Ranganath explains why: the errors a test exposes are what let the brain keep the good connections and drop the bad ones [2]. UCLA's Alan Castel adds that an attempt that fails, followed by the right answer, still beats easy repetition [3]. And math educator Grant Sanderson supplies the warning that ties it together: a lecture can make perfect sense while you watch it and be gone a week later, because how satisfying something feels does not track how much of it you keep [4].

Table of contents

What They Said

Andrew Huberman: studying is defense against forgetting

Huberman's solo episode on studying starts from a reframe. Most new information is naturally discarded, so learning is mostly about offsetting forgetting, and good studying is the set of practices that protect what you want to keep [1]. He asks listeners to stop thinking of studying as putting information in and to think of it as defending against the loss that follows any new exposure, whether the material is cognitive, musical, mathematical or motor [1]. He also says the best learning practices are not intuitive, and that when you look at the research on preferred learning styles, such as being a visual or auditory learner, the supposed differences largely melt away [1]. What matters is whether you engage the parts of your memory systems that prevent forgetting [1].

The center of his case is the research on testing. He describes a 1917 study in which grade school children either reread biographies several times or read them once and then repeatedly tested themselves by recalling details in their heads; the self-testers vastly outperformed the rereaders in both the amount and the accuracy of what they recalled [1]. More recent studies compared people who studied a passage four times, people who studied it three times and took one test, and people who studied it once and took three tests, with a final test days, weeks or up to a year later [1]. Performance on the final test was essentially proportional to the number of tests taken beforehand, even when practice scores were only 40 or 50 percent, provided the learners saw the correct answers afterward [1].

The finding he calls most striking is about confidence. The students who repeatedly restudied felt the most confident and predicted they would do best, and the students tested three times felt the least confident; the results came out the other way round [1]. Huberman's explanation is that repeated exposure breeds familiarity, a sense of recognizing the material, which he separates from true recall and mastery [1].

He gives his own example. Overwhelmed by a demanding undergraduate neuroanatomy lab course, he began lying on his bed with his eyes closed, mentally flying through the nervous system from the ear, the eye and the top of the brain [1]. Whenever he hit a gap and did not know what came next, he looked it up and went back to the mental journey [1]. The key to testing, he says, is finding exactly what you know and what you do not [1]. He practices this on the audience too, stopping the episode for pop quizzes and pointing out that noticing where you got something wrong is itself valuable [1].

On timing, he reports experiments comparing a test right after first exposure, evenly spaced tests, and tests only shortly before the final exam. The group that tested itself very soon after first exposure, the same day or the next, did best; the group that waited a long time and then crammed two tests back to back did worst; spaced testing fell in between [1]. By his account a single self-test soon after learning cuts forgetting roughly in half compared with no test, across musical, mathematical, language and motor learning, regardless of how well you score [1].

On format, he says the best self-tests are open-ended, short-answer or long-answer questions that require recall, not recognition [1]. Standard multiple choice mostly taps familiarity, with the exception of well-designed trick questions of the kind found on the GRE, LSAT or MCAT [1]. He frames the whole approach as testing for learning, not for evaluation [1].

Charan Ranganath: stress test the memory

Ranganath, a UC Davis psychologist and neuroscientist, starts from the same fact as Huberman: we forget most of what we experience. Hermann Ebbinghaus lost about 60% of the nonsense material he memorized within a day, and nobody has overturned that basic picture [2]. So his advice is that "you don't want to remember more, you want to remember better" [2].

He describes the testing effect in plain terms. If you are learning words in Spanish, or anything else, "that act of testing yourself helps you retain it better over time than if you just studied it" [2]. He admits this looks odd from the point of view of some traditional learning theories: why would giving yourself an error do more than giving yourself perfect input that copies what you are trying to learn [2]? His answer, drawn from his lab's computational modeling, is that the benefit comes from the error itself [2]. In the conversation's words, "The error is where the learning happens" [2].

From this he draws a rule he says he returns to again and again: "If you are feeling like you're coasting, then you're actually not learning" [2]. And: "you should always be stress testing the memory system" [2].

Ranganath also describes the confidence problem from the learner's side. Even people who complain that their memory is terrible are, he says, overconfident in the moment: "they're overconfident about what they'll retain later on" [2]. A self-test punctures that. The reaction is "oh my god, I thought I knew that, but I don't," and he grants that "it can be demoralizing until you get around that" and see that this is how you learn best [2]. His comparison is an actor preparing for a film: you do not just sit reading the script, you act it out, and you botch lines from time to time [2].

He pairs testing with spacing. Host Lex Fridman describes using the spaced repetition app Anki every day, and Ranganath explains that spreading out study sessions improves long-term retention [2]. Cramming has a narrow use. He recalls studying in the hallway right before a test, and says "actually not spacing can really help you if you need it very quickly" [2]. The catch follows at once: "But the problem is that you tend to forget it later on" [2].

Alan Castel: struggle first, then check

Castel, who studies memory and cognitive aging at UCLA, says "a lot of good learning happens through making mistakes" [3]. His demonstration is the Apple logo. People have seen it countless times, yet cannot say which side the bite is on or whether there is a stem or a leaf [3]. The best way to learn it is to try drawing it from memory, struggle, and then look again; even if you never redraw it correctly, the errorful attempt followed by corrective feedback produces better learning [3].

He points to UCLA colleagues who study desirable difficulties and notes that teachers often try to make learning easier, when the aim should be different: "You want it to be not a struggle, but a challenge" [3]. In one of his studies, people who had to get up and find the nearest fire extinguisher learned its location far better than people who were simply told where it was [3]. "I think that's the difference between seeing something and noticing it," he says [3]. He would like airline safety briefings to ask passengers to find the exits themselves [3].

Castel's own history supplies two cautions about easy-feeling methods. As a child he practiced a single line for a school production of The Wizard of Oz over and over and still swapped the words on opening night [3]. Later he memorized much of the periodic table with mnemonics and found in college that he did not really understand chemistry [3]. He calls mnemonic tricks a workaround that operates at a shallow level of processing [3].

He also uses self-testing in a modern form, asking ChatGPT to generate self-tests, because testing yourself and finding out what you got wrong is one of the best ways to learn [3].

Grant Sanderson: the clear explanation that does not stick

Sanderson, the creator of 3Blue1Brown, comes at the same problem as someone who makes explanations for a living. He describes the "Feynman effect": while you watch or read Richard Feynman's lectures everything makes perfect sense, yet a week later you cannot recall what gave you the insight [4]. He worries that his own videos risk being "intellectual candy" in the same way [4].

His remedy is activity. Unless learners do something active, such as reinventing the idea, working problems, or using spaced repetition to lock in the terminology, the material will not stick [4]. He warns educators that how satisfying content feels does not correlate with long-term learning [4]. He admires Feynman's habit of working out what problem a paper was trying to solve and attempting it himself before reading what others did, and says the things he has learned best are the ones with some element of rediscovery [4]. Approaching a few core building blocks on your own, he says, fills your head with counterexamples and intuitive examples you would never get from just reading a proof [4].

Where They Agree, and Where They Don't

All four agree on the central point: exposure is not learning, and the feeling of understanding is a poor gauge of what will be there later. Huberman has the restudy group predicting the best scores and getting them wrong [1]. Ranganath says people are overconfident in the moment about what they will retain [2]. Castel says our sense of how well we know something, which he calls metacognition, often misleads us, and that seeing something many times does not guarantee you will remember it [3]. He draws the same lesson from eyewitness memory: "high confidence, doesn't necessarily mean high accuracy" [3]. Sanderson says the same of his own medium [4].

They also agree that effort is a signal to welcome. Huberman calls effort the cornerstone of learning and says the courses he remembers most fondly are the ones he struggled with most [1]. Ranganath says coasting means you are not learning [2]. Castel wants a challenge built in [3]. Sanderson accepts that rediscovery slows him down [4].

The differences are of emphasis, not direct contradiction.

• How fast to test, and how to schedule it. Huberman stresses a first self-test the same day or the next, and in the experiments he describes, early testing beat evenly spaced testing [1]. Ranganath stresses spacing study sessions out for long-term retention [2], and Sanderson mentions spaced repetition as a way to lock in terminology [4]. These are compatible, since one concerns the first retrieval and the other the schedule afterward, but the sources do not settle how to combine them.
• Cramming. In Huberman's account the group that waited and crammed its tests did worst [1]. Ranganath grants cramming a real but limited use when you need the information very soon, at the cost of forgetting it later [2].
• Memory techniques. Ranganath speaks well of the memory palace, saying it works because it gives arbitrary information a structure to hang on, and that memory athletes practiced strategies until they became automatic [2]. Castel is cooler: mnemonics take effort, can produce predictable errors and work at a shallow level, as his chemistry experience showed him [3]. The gap reflects their goals. Ranganath is talking about holding arbitrary information; Castel is talking about understanding.
• How deep to go. Sanderson's rediscovery is a heavier form of retrieval than a recall quiz, and he says plainly that it slows you down and is partly a fault, one reason he is not an active researcher [4]. He reserves it for a few core building blocks [4].

Why It Works

Errors mark what to fix. Ranganath's lab model treats a test as a way of exposing which connections are good and which are bad, so the brain can keep the parts of the cell assembly that serve the memory and lose the ones that do not; he compares it conceptually to backpropagation in AI [2]. Without a stress test, the memory is never fully exposed to its own errors [2]. Huberman puts it similarly: testing forces errors, and errors cue the nervous system to lock in what is right and correct what is wrong [1].

Familiarity is not recall. Huberman distinguishes the sense of recognizing material from being able to produce it, and points to a review in the journal Hippocampus on the neuroscience of familiarity versus recollection [1]. Rereading builds the first, which is why it breeds confidence without the performance to match [1].

Strain signals the brain to change. Huberman says the sense of strain during focused study reflects the release of neuromodulators such as epinephrine that cue circuits to change later, and that if learning felt effortless the nervous system would have no reason to change [1]. In the Castel conversation he adds that plasticity is triggered by the gap between current and desired performance, and that the frustration that comes with that gap is part of what allows change [3]. Ranganath says more generally that when expectations break, the prediction errors are where learning happens [2].

Recalling in new contexts strips away the incidental. Ranganath's model suggests that recalling something in different contexts removes context-specific details and strengthens the core content, which is why well-rehearsed memories feel like facts and not vivid scenes [2].

The rewiring happens later. Huberman notes that the actual changes in connections occur during deep sleep and sleep-like states, and that material learned on a given day is mostly consolidated during that night's sleep [1].

On the evidence itself: the 1917 study and the study-versus-test experiments are reported by Huberman, without the detail needed to check them here [1]. He also flags that the survey of close to 700 medical students is self-report data with correlation-versus-causation limits [1]. Ranganath says of his lab's work that he does not want to make it sound as if they have figured out everything [2].

Where It Applies, and Where It Doesn't

• It reaches well beyond exam facts. Huberman says the benefit of an early self-test shows up across musical, mathematical, language and motor learning [1]. Castel's fire extinguisher and emergency exit examples extend it to everyday knowledge that matters in a crisis [3].
• It depends on feedback. The low practice scores in Huberman's studies still helped because learners saw the correct answers afterward [1]. Castel's logo exercise is an attempt followed by corrective feedback [3]. A test without checking is not what these sources describe.
• Recognition formats are weaker. Standard multiple choice mostly taps familiarity [1].
• If you only need it for the next hour, cramming works. Ranganath is candid about that, and equally candid that the material fades [2].
• Not everything deserves the effort. Huberman says forgetting is not a flaw and that people who remember every detail struggle in work and relationships [1]. Castel calls forgetting healthy: you need your hotel room number this week, not next month [3]. Sanderson limits rediscovery to a few core ideas [4].
• Testing sits on top of the basics. Huberman treats focus and alertness as the first step of learning and sleep as the second, and warns that late caffeine and all-nighters undermine the learning they are meant to serve [1]. Ranganath says attention decides what gets in, and that checking laptops during talks leaves him with no memory of the day [2].
• Expect it to feel worse. Tested students felt least confident [1], and Ranganath calls the experience demoralizing at first [2]. Huberman notes that students dislike pop quizzes and punish teachers for them in evaluations, and suggests announcing quizzes in advance [1].

If trouble with attention or sleep is getting in the way of studying, that is worth raising with a doctor or other qualified professional.

What To Do With It

1. Recall before you reach for your phone. When a class or study session ends, take a few seconds or minutes to go back over what you just learned from memory [1].
2. Test yourself the same day or the next. In the experiments Huberman describes, the earliest self-testers did best, and he reports that one early test roughly halves forgetting [1].
3. Use open-ended questions. Write or say short and long answers from memory, and do not rely on recognizing the right option in a list [1].
4. Always check the answer. Look up every gap, as Huberman did in his mental tour of the nervous system, then return to the recall [1]. Low scores still pay off when the correct answer follows [1][3].
5. Attempt before you look. Draw it, solve it or state it first, as in Castel's logo exercise [3], and for a few core ideas try Sanderson's approach of working out the problem yourself before reading the solution [4].
6. Read discomfort as progress. If you feel you are coasting, Ranganath says you are not learning [2], and the confidence that rereading gives is the feeling the research shows to be misleading [1].
7. Space the later sessions. Spread reviews out for long-term retention, and keep cramming for cases where you only need the information very soon [2].
8. Teach it. The top medical students in the survey Huberman cites taught the material to classmates, and that was linked to their own strong performance [1].
9. Generate your own quizzes. Castel asks ChatGPT to produce self-tests so he can find out what he got wrong [3].
10. After a lecture or video that felt clear, do something with it. Work problems or reconstruct the argument yourself, because a satisfying explanation is not evidence that it stuck [4].

Sources

1. Optimal Protocols for Studying & Learning · Huberman Lab
2. Charan Ranganath on How Memory Works: Imagination, Deja Vu and False Memories · Lex Fridman Podcast
3. Dr. Alan Castel on Improving Memory and Aging Well: Mistakes, Curiosity, Balance and the ABCs of Successful Aging · Huberman Lab
4. Grant Sanderson (3Blue1Brown) on Feynman, Exponential Growth, Teaching Online, GPT-3 and Manim · Lex Fridman Podcast