What this covers

Andy Matuschak and Dwarkesh Patel discuss why educational technology typically underperforms and what would be required to build tools that genuinely deepen learning. Matuschak argues that most learning tools fail because they treat learning as a straightforward problem to be solved through rapid iteration, when it is actually a difficult design problem rooted in metacognition and human memory. The conversation ranges across educational psychology, the architecture of effective learning environments, why certain media succeed or fail at teaching, and how the design constraints of powerful tools differ from the MVP mindset.

The discussion opens with why people fail to retain material: either they forgot what they once understood, or they never understood it and never noticed the gap. Skillful reading addresses the second failure through active interrogation of confusion. Matuschak argues that memory is underestimated in learning and creativity; having relevant facts in mind is necessary for making novel connections or recognizing patterns. This underpins his view that explicit spaced repetition tools have broader value than typically recognized, particularly for material too rare or disconnected to be learned through natural immersion.

The conversation also examines the tension between engaging media and deep understanding—video teaches a surprising amount but leaves viewers lacking detailed retention—and whether design can resolve this. Scaffolding and outsourced metacognition appear throughout: learning environments that make thinking visible and manageable, from embedded review questions to structured tutoring. Matuschak argues that building such tools requires genuine design ideas working backward from learning goals, not just fast iteration, and discusses how this shapes both tool design and the funding models that enable contemplative research rather than rapid pivoting.

Sharpest takeaway

Matuschak argues that most learning tools fail because effective learning is fundamentally a difficult metacognitive and design problem—skillful reading means continuously interrogating one's confusion, memory is the underrated substrate of understanding and creativity, and powerful tools for thought require genuine design ideas rather than fast MVPs.

  • People fail to retain material either because they forgot it or because they never understood it and never noticed—skillful, active reading addresses the latter.
  • Memory is essential to understanding arguments, decision-making, and creative insight, so explicit memorization tools have broad underappreciated value.
  • Building good tools for thought is a design problem requiring contemplation, not just rapid iteration; this also shapes how independent research can be funded.

The claims · ranked55 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.59

The startup advice to build an MVP fast and iterate is good once you have something worth testing, but adopting it as a default prevents you from sitting in stillness and confusion long enough to have good ideas—because the most powerful designs are expressions of new ideas or primitives (like the nonlinear text editor or Photoshop's layer), which require a theory of change working backward from the goal rather than a mere theory of action of building whatever is straightforwardly possible.

normativecontestednovelty 4/4durability 3/4· Andy Matuschak

“it's difficult to have ideas by means of building an MVP very rapidly. Now, if you have an idea that you think is interesting, it is good to test it rapidly.”

0.58

Following John Dewey, because children lack a fully developed prefrontal cortex and sense of self, letting them do whatever their whim commands in any moment is not freedom but rather chains them to those impulses, making them subject to tides of impulse—an argument that doesn't authorize tyranny but warrants skepticism about the plans of 12-year-olds.

normativecontestednovelty 3/4durability 4/4· Andy Matuschak

“to let them do whatever it is that their whim commands them to do in any given moment is actually not freedom, but rather is chaining them to whatever that impulse is. It makes them the subject of these tides of impulse.”

0.55

Apple manages enormous scope through strong command-and-control combined with extensive delegation: leaders operate in concentric rings of responsibility—a small core they directly manipulate, a wider ring they monitor and influence, and an outer ring fully delegated where they only want to hear about problems—so the head of all software can deeply engage gesture-recognizer heuristics while ignoring 95% of software decisions.

factualestablishednovelty 3/4durability 3/4· Andy Matuschak

“the individual leaders in various areas at all levels of hierarchy have an enormous amount of latitude. That's the only way that any of this can work.”

0.55

There are two distinct reasons people fail to retain material: they understood it at some point but forgot it, or they never understood it in the first place and never noticed that they never understood it—the second failure is insidious because the reader's eyes skid across paragraphs without comprehension.

factualestablishednovelty 3/4durability 3/4· Dwarkesh Patel

“One is they got it at some point, but they forgot it. And the other is they never understood it in the first place and they just never noticed that they never understood it.”

0.55

Being a skillful reader is a learnable skill akin to rock climbing: difficult material that looks like a solid wall to inexperienced readers presents subtle notches—footholds for asking and answering questions—to experienced readers, and when reading is aligned with genuine interest, this skill produces more serious, forward, understanding-oriented reading.

causalestablishednovelty 3/4durability 3/4· Andy Matuschak

“To an experienced rock climber, those subtle notches are like a ladder and they can get right in there and start making some progress and seeing what's up with this wall. But if you're an inexperienced rock climber, it just looks like a solid wall.”

0.55

An enduring result in education psychology is that when you are doing a lot of cognition, metacognition becomes difficult—the harder and less familiar the material, the harder it is to plan, self-regulate, choose the best next action, and evaluate your own understanding—which is why effective learning environments outsource metacognition through scaffolds like syllabi and embedded questions.

causalestablishednovelty 3/4durability 3/4· Andy Matuschak

“when you're doing a lot of cognition, metacognition is difficult. What I mean by that is when you're thinking really hard about the stuff on the page, it's very difficult for you to plan, to regulate yourself, to figure out what the best next action to do is”

0.55

Education has made enormous progress over the 20th century, but almost entirely by raising the floor: high-school graduation went from 6% in 1900, illiteracy was largely eliminated, and the bottom quartile's test scores rose dramatically while the 75th percentile moved slowly and the top barely at all—because the system was built for the masses, the ceiling (von Neumann-level greatness) is very hard to raise and depends on aristocratic tutors and heritable propensities outside mass schooling.

factualestablishednovelty 3/4durability 3/4· Andy Matuschak

“in 1900, 6% of teenagers graduated high school in the US.”

0.55

Apple coordinates a $2 trillion company's vast tradeoffs not by one person integrating everything but through strong compartmentalization: each person owns a domain bounded by others' domains (e.g., a thermals team sets the thermal budget as a hard constraint), with iterative push-and-pull when desired features prove infeasible—as when Hey Siri's battery drain forced creating a dedicated low-power co-processor.

factualestablishednovelty 3/4durability 3/4· Andy Matuschak

“It's very compartmentalized. None of what you just said was relevant to me. It was all pre-specified. At Apple, you have a little domain that's your own, and the boundaries of that domain are determined by everybody else's little domain.”

0.52

Per Anderson's ACT-R theory, the mind turns individual facts into higher-level generalizable patterns through knowledge compilation, but this requires repeated exposure and use across varied contexts; simply reading a book won't reliably compile knowledge, though reading combined with demanding conversations that draw on the material can naturally produce that compilation.

causalestablishednovelty 3/4durability 3/4· Andy Matuschak

“in order to effectively compile things that you've learned into schemas that will match feature scenarios effectively, you need to be exposed repeatedly to those things, you need to use them”

0.51

Much of daily cognition is memory-bound in underappreciated ways: understanding a multi-step argument exceeding working-memory capacity depends on abstracting steps via things you already know; and creative insight—noticing a surprising connection or contradiction—is only possible if the other element is already in memory, since having to look it up means the connection won't seem salient.

causalcontestednovelty 3/4durability 4/4· Andy Matuschak

“your ability to understand an argument that has many steps in it, more steps than you can keep in your working memory, depends on your ability to think of some of those steps in terms of some stuff that you already know, so that you can kind of reduce it or abstract it.”

0.50

Embedding review questions every ~1500 words in a text (as in Quantum Country) helps memory, but has a crucial metacognitive benefit: the question reveals that the reader failed to even notice an apparently important word or concept, providing feedback that prompts rereading and changes reading behavior for future sections toward slower, more attentive engagement.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“it is an opportunity for you to notice that you did not in fact absorb what was in that thing. Not that you don't remember but that there's a word in the question that is apparently important that you simply didn't even notice.”

0.50

Explicit spaced-repetition practice is best understood as a way to bootstrap yourself into a position where you can then learn naturalistically through immersion and social reinforcement—like a beginner who can't yet immerse in a Swahili community or play church piano every Sunday until they have a foundation—but some things, like rare medical diagnoses or many creative insights, occur too infrequently to be naturally reinforced and require an out-of-band mechanism.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“explicit practice is a way to bootstrap yourself.”

0.50

Educational games largely fail to teach explicit subjects because games are ultimately an aesthetic form whose purpose is to produce an experience that feels a particular way; works like The Witness teach complex mechanics about their own environment without words, but no one has succeeded in making a game that teaches explicit external subjects, because that aim conflicts with the game's aesthetic purpose.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“The issue seems to be that games are ultimately in aesthetic form. The purpose of the game is to have an experience that feels a particular way. And so they're sort of serving a different purpose than Grant's videos or a text.”

0.50

Video has taken off enormously as a learning medium—people will watch an hour of Grant Sanderson explaining math who would never open a graduate textbook—but the trade-off is that they leave much more engaged yet without much detailed understanding, raising the open question of whether a version exists that produces both engagement and deep understanding.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“People will watch Grant Sanderson spend an hour going through some explanation of an esoteric math problem, people who would never crack a Springer graduate textbook in mathematics or something like that. The issue is that they will not walk away from that interaction with much understandingbutthey're much more engaged.”

0.50

Being a good teacher is largely about empathy, effective communication, and care rather than postgraduate-level subject mastery; you don't need extraordinary ability to teach a 15-year-old, though inquiry-oriented, open-ended classes do require enough mathematical familiarity to improvise—so the historical loss of intelligent women from teaching to other careers may not have lowered teaching quality if effective teaching is anti-correlated with elite research ability.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“Being a good teacher is often about empathy and effective communication and care. It's very personal. It's very intimate. You need to understand the subject but to teach a 15 year old or something, you actually don't need to understand it at a postgraduate level necessarily.”

0.50

Empirical research shows tutoring is surprisingly effective even when the tutor knows little—just having a warm body present has a large effect—but Andy suspects the most important component of elite tutoring like Aristotle teaching Alexander is communicating a worldview and being inspiring, which wouldn't show up on tests, and inspiring tutors correlate with being expensive though not necessarily with subject expertise.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“it's interesting, how far you get in tutoring efficacy when the tutor doesn't necessarily know anything. Just having another warm body there actually contributes a very large effect.”

0.50

Recasting crowdfunding from 'subscribe to support my work' (a tip jar) to 'become a member' with concrete, even modest member benefits yields very different and better results, because membership means something in people's minds—Andy succeeded more after this reframe and believes Gwern's tip-jar framing, not lack of traffic, explains his low Patreon revenue despite a larger, more impactful following.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“There's a difference between a tip jar and a membership in people's minds. Becoming a member means something. If you could offer something small, that feels membership-ish, you might get very different results.”

0.49

Programming feels like a video game because of its sense of instantaneous direct contact with the environment—you manipulate a conceptual world right under your hands and constantly get feedback (red squiggly lines, pressing Command-R)—producing an addictive sense of forward progress, captured by the doom-laden experience of staying up until 2am because each near-fix feels so good.

causalestablishednovelty 2/4durability 3/4· Andy Matuschak

“what makes programming feel like a video game is this sense of instantaneousness, this sense of direct contact with the environment. You're learning about a conceptual world, but that world is right underneath your hands”

0.49

Manipulable physical/dynamic media like Cuisenaire rods let children explore number relationships unprompted because handling them is pleasurable, and patterns naturally suggest themselves (e.g., an 8-rod divides evenly while a 7-rod has no equal pair)—the open design question is whether equivalent manipulable media can be created for advanced topics in math or debates in history.

factualestablishednovelty 2/4durability 3/4· Andy Matuschak

“Kids will play with these things unprompted because they're fun. It's just a pleasure to handle them... it's very natural to suggest, "Ah, can you make a pattern like this?Why can't you seem to make patterns like that? Why is that?"”

0.49

Studying in the presence of another engaged person reduces the willpower required and is more energizing; Andy could study quantum mechanics for three hours partly because he was effectively performing for Dwarkesh, and similarly a student engaged in a real activity with a peer needs less willpower and learns from peers if structured well.

causalestablishednovelty 2/4durability 3/4· Andy Matuschak

“if I were alone and studying that book for three hours, and I weren't effectively trying to perform for you Dwarkesh, it wouldn't have been nearly as energizing for me. And I definitely would have taken breaks.”

0.48

Experts need less scaffolding in a familiar domain not just because cognitive demand is lower but because rich existing knowledge lets them make better plans from the start—planning requires modeling a path from start to destination, which is feasible when the steps are only a few hops apart but impossible when start and goal are far and the intermediate connections invisible.

causalestablishednovelty 3/4durability 3/4· Andy Matuschak

“if your destination and your starting point are very far away, then you can't necessarily see all the things in between or how to draw those lines. But if those things are only a couple hops away, you can maybe infer pretty accurately.”

0.47

Andy systematically avoids marketing his work as self-protection because audience and marketing exert a corrosive, default-distorting pressure on honest inquiry—tempting you to investigate what people would like rather than what you actually find interesting, to do the boring shallow version, and to portray results in the best possible light while downplaying limitations, all of which is death for discovery.

causalcontestednovelty 3/4durability 4/4· Andy Matuschak

“I am really worried about the corrosive influence of audience and marketing on honest inquiry. It is very easy to distort my work. It's almost a default to try to make it be something that people would be more likely to like rather than the thing that I actually want to investigate”

0.47

Hypertext is a powerful navigational aid for texts amenable to dropping the reader into the middle—encyclopedias, dictionaries, compound references—because those entries stand on their own, but most concepts are embedded in holistic, narrative wholes that resist excerpting; this is why hypertext novels and choose-your-own-adventure books fail, since a hyperlink's destination must work as the endpoint of all references, forcing a lowest-common-denominator sameness that destroys coherent arc and renders the hypertext useless.

causalspeaker onlynovelty 4/4durability 4/4· Andy Matuschak

“the page that is the destination of a hyperlink, it has to work as the endpoint of all of its reference. And so it can't establish any kind of coherent or consistent arc, unless there's a kind of sameness to all of the reference. And the more that there's sameness to the reference, the less useful hypertext is.”

0.47

Learning new material is easier when you already have a rich network of related knowledge for it to slot into; Andy could absorb Apple's complex system internals quickly because his deep prior understanding of Apple's platforms and computer science gave new facts somewhere to connect.

causalestablishednovelty 2/4durability 4/4· Andy Matuschak

“Learning stuff is easier when you have other stuff to connect to. It's a nice principle.”

0.45

There is a flow pocket where the demandingness you bring is matched to your ability and the text is engaging rather than overwhelming; reading dutifully as an undemanding reader produces an adversarial relationship requiring willpower to power through, whereas opening one's curiosity and genuinely engaging the book requires less gumption.

causalestablishednovelty 2/4durability 3/4· Andy Matuschak

“I will find that I need to bring more gumption to bear to power through and make myself sit there and keep flipping the pages than I need if I actually just open my curiosity and attention and really start engaging the book.”

0.45

The first and most important rule of skillful, active reading is to constantly ask questions of the text and try to answer them; an undemanding reader asks no questions and gets no answers, leading their eyes to skid across paragraphs without comprehension despite remembering nothing being a misdiagnosis of the real problem.

normativeestablishednovelty 2/4durability 3/4· Andy Matuschak

“an undemanding reader asks no questions and gets no answers”

0.44

For the median or below-median student, the education system is mostly about making students do things they don't want to do and achieving goals that aren't theirs, rather than helping them achieve their own goals—which is why Andy got out of that space and instead focused on inquiry learning that connects to authentically interesting questions.

normativecontestednovelty 2/4durability 3/4· Andy Matuschak

“For the median student, the education system mostly wants to make the student do things they don't want to do. It's not about helping them achieve their goals more easily or more effectively. For the most part, it's about achieving goals that aren't theirs.”

0.43

The hardest, most draining part of research is not cognitive difficulty but sitting and not knowing—deciding what to do next on an open project taxes what William James called the 'energies of men' (willpower/gumption), which is why studying a textbook for three hours felt like a vacation: it is far more cognitively demanding yet far less demanding on that life-force resource because someone else specifies what to do.

causalspeaker onlynovelty 4/4durability 3/4· Andy Matuschak

“sitting and staring at a page and deciding what you should do next on a research project is incredibly draining on that resource. The sitting and not knowing is the hardest thing that I do in my work.”

0.43

Forgetting may serve attention via spreading activation: memory is encoded in weighted connections that route 'packets' from a stimulus, and predictive utility theory holds that the probability of retrieving an item corresponds to how useful the brain predicts it will be—so retaining everything would create a cacophony, and repeated retrieval in varied situations raises predicted utility across a broader distribution.

causalcontestednovelty 3/4durability 3/4· Andy Matuschak

“predictive utility theory. And it suggests that the probability of retrieval of a particular item in a given situation actually does correspond with basically a model of to what extent the brain predicts it will be useful.”

0.42

Contrary to a resource-management intuition, practicing extensive spaced-repetition memorization does not cause you to forget normal human things; experience and the experimental literature suggest there are probably upper bounds on memory but we are nowhere near them—memory champions and multilingual people have memorized orders of magnitude more without loss.

factualestablishednovelty 2/4durability 3/4· Andy Matuschak

“I've been doing this memory system stuff for years and I just know more stuff now. This is aligned with the experimental literature, which seems to suggest that, there's probably upper bounds but we're not close to them.”

0.41

The limited adoption of spaced repetition is mostly efficient rather than a market failure: it is widely used precisely where it works without rare tacit knowledge (medical students, language learners), but for subjects like quantum physics it is correctly under-adopted because it merely speeds learning rather than making it a fait accompli and requires a large amount of unusual, largely tacit knowledge to apply well.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“In places where spaced repetition, as it stands without substantial novel cultural knowledge that's difficult to transmit and isolate, is valuable, we see a lot of space repetition usage.”

0.41

To bootstrap into a domain you must outsource the question of what is necessary to know to an author or syllabus, but you should treat that answer as tentative and revise it iteratively, skipping around according to interest rather than following it completionist-style on faith—because rigid completionism leads to boredom, demotivation, and quitting.

normativecontestednovelty 2/4durability 3/4· Andy Matuschak

“I think a very common mistake that people make is to feel that they need to do the thing the right way and that is exhaustive and completionist.”

0.40

With abundant online learning resources, the value of conscientiousness as a trait has dramatically increased because if you can motivate yourself, the world is out there for you to absorb.

factualcontestednovelty 2/4durability 2/4· Dwarkesh Patel

“it seems that the value of conscientiousness as a trait has dramatically increased. If you can motivate yourself to learn these things, the world is out there for you to absorb.”

0.40

Over 90% of people in the education space focus on the bottom quartile rather than the median, motivated by equity and by the larger marginal impact on disadvantaged students; Andy understands and endorses this but has chosen to focus elsewhere—on supercharging higher-performing learners—partly because much innovation may come from disaffected 20th-percentile students whom this focus leaves un-helped.

factualcontestednovelty 2/4durability 2/4· Andy Matuschak

“Basically everyone in the educational space, this isn't totally true, but like 90+% of people in the educational space are focused on the bottom quartile, not even the median.”

0.40

School serves many purposes beyond instruction—social, societal, behavioral, and the pragmatic purpose of babysitting—and these can be unbundled; pandemic-era pod models like Schoolhouse showed that splitting one teacher's time across five or six kids can be tractable and cost substantially less than private school, with older grades possibly handled by coaches who connect students to domain specialists.

factualcontestednovelty 2/4durability 2/4· Andy Matuschak

“The school, of course, has many purposes other than instructional, right? It has a social purpose, it has a societal purpose, it has a behavioral purpose, and it also has like a pragmatic purpose of basically babysitting. Those things can be unbundled.”

0.39

Crowdfunding only works for research that is broadly applicable to many people with disposable income and that already has promising, communicable results with regular reportable progress; it is very difficult to use in the early days of a project before a direction is chosen and expressible, which is why a marine geology project wouldn't attract a crowd like Andy's learning research does.

factualspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“this crowdfunding project only even slightly works, because it's understandable and interesting to others. It's already in a place where there's some results that look promising.”

0.39

Andy's durable evergreen notes let knowledge accumulate over time—partial notes expand as he reads relevant papers and conducts interviews, giving him a place to look when designing the next system—but Byrne Hobart and Matt Levine's daily re-explanation of concepts argues for ephemerality, since recreating an explanation anew each day lets it become inflected by current thinking; Andy now does a mix and lacks a clear model of when each is better.

factualspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“This is an argument against the kind of note-taking that I do. It's an argument for ephemerality, for recreating the thing every day, because it will change and it will become inflected by what you're thinking about and your experiences.”

0.39

Unlike programming's search-with-heuristics feel, design work often provides no sense of getting closer—the core action is to duplicate and try hundreds of variations, learning only that parts of the combinatorial search space don't work, so weeks can pass narrowing the space without feeling like progress until you stumble onto something that works.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“you don't have this feeling of getting closer often. Often there will be like weeks that go by without feeling like you're getting closer, because what you're doing is just kind of like narrowing the search space.”

0.39

When learning something you genuinely care about with good tools, misery usually comes not from the material but from resisting the feeling of doing poorly—fear of judgment or insufficient time—which is an emotional problem to be healed rather than a practical learning problem; remaining rote tasks like organic chemistry names can be made cheap with memory systems and pleasant by embedding them in interesting Fermi problems.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“it could be miserable that I think is pretty common is that you have some idea about, you're not going fast enough, or you're failing, or you're struggling, and the misery comes from resisting that.”

0.39

Wikipedia works as well as it does precisely because encyclopedia entries were already forced to stand on their own before hypertext existed; yet Wikipedia is rarely the best introduction or explanation of a topic—it functions as a good jumping-off point and reference that helps you learn the right thing to ask about.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“the reason why Wikipedia works as well as it does is that encyclopedia entries are already forced to stand on their own. And that was true before hypertext existed.”

0.36

Prolific infovores like Tyler Cowen who lack an explicit note-taking system actually do take notes in the form of their writing obligations—daily blogging, weekly columns, and books—written in a first-draft, think-out-loud mode honed over decades that lets them produce a decent take the first time, which functions as the note.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“Tyler's blog posts often have this a real first draft mentality to them. He's just thinking out loud and he's got decades of practice thinking out loud and like writing down a decent take the first time.”

0.36

Skilled designers obsessively rabbit-hole into highly constrained, finite, ephemeral problems like designing a poster because the search space is small enough to find a decent solution rapidly, after which there's an enticing feeling of turning the crank to polish it where all options are okay—a controllable 'snack' that contrasts with open-ended design where you simply don't know how to proceed.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“a poster is really constrained, it's finite, it's ephemeral. You can start it and yeah, there's a search space, but you can find a decent part of the search space pretty rapidly.”

0.35

Building Orbit—a highly general, self-serve, deployed platform generalizing Quantum Country—was the wrong move because the underlying idea wasn't strong enough and a general infrastructure with accounts wasn't the right way to test it; Andy would have learned more by building specific one-offs, working with one author and one different book to form a specific theory, since his aim is to build the idea, not yet a scalable product.

factualspeaker onlynovelty 3/4durability 2/4· Andy Matuschak

“Building this highly general version of it wasn't the right way to test it. I would have been better building more one-offs rather than a self-serve thing that anyone could use.”

0.35

In crowdfunded subscription research, churn (about 2% monthly for Andy) is surprisingly insensitive to anything the creator does—experiments haven't meaningfully changed it—so growth depends almost entirely on getting more people into the top of the funnel through marketing, with only conversion rate being somewhat malleable.

factualspeaker onlynovelty 3/4durability 2/4· Andy Matuschak

“One thing I've learned is that the churn rate is surprisingly insensitive to anything I do. I experiment with a variety of things and it hasn't meaningfully changed the churn rate. What does change things is getting more people into the top of the funnel, in other words, marketing.”

0.34

Andy now favorably endorses the old-fashioned view that you should memorize a subject's taxonomy because modern memory techniques make the price trivial—roughly 15 minutes spread over a few weeks to learn it reliably—whereas his high-school memorization was inefficient, uncertain, and emotionally difficult because he lacked techniques and couldn't be confident he'd learned anything.

causalspeaker onlynovelty 2/4durability 3/4· Andy Matuschak

“for me, it's just trivial. Like, yeah, sure, whatever, throw it into the deck. It'll consume a total of 15 minutes over the next few weeks and then I'll know it. It just doesn't cost anything.”

0.34

A core tension of crowdfunding public work is that one of the main ways to fund it—putting essays behind a Patreon paywall—makes that work less public and less discoverable, so people unfamiliar enough to subscribe never learn the content exists; design solutions exist (visually adjacent public/private work, previews) but Andy has invested zero effort into them.

factualspeaker onlynovelty 2/4durability 3/4· Andy Matuschak

“It's a shame that one of the ways to fund public work is to make some of that work less public.”

0.34

There's no general answer to how much effort a personal website deserves; many engineers do themselves a disservice building vast technical infrastructure when they just want to post markdown files, and the better approach is to first identify the ideal form of what you want to put into the world—once you understand that shape, making it is often not effortful, as Andy's celebrated working-notes site took only a weekend.

normativespeaker onlynovelty 2/4durability 3/4· Andy Matuschak

“Many people, especially engineers, do themselves a disservice by fretting over their websites unnecessarily, building vast technical infrastructure when really what they want is a place to post markdown files.”

0.33

Spaced repetition is accidentally embedded throughout knowledge work—researchers re-explaining a field's history in successive papers' background sections, mentors repeatedly advising students, doctors on rounds covertly retrieving differentials before presentations—so the distinctive value of formal practice is reaching material you wouldn't otherwise repeat because it's too early or not firmly tethered to your life.

factualspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“For a researcher, when you need to write papers regularly and you're writing those background sections, you're repeatedly explaining the history of a particular line of research and citing the appropriate sources. That is a kind of spaced repetition.”

0.33

What's needed is a mass medium that is active and self-regulating—where the choice of activity is built into the structure of the material, as in Nand to Tetris where you actively build a computer—rather than requiring the learner to regulate everything themselves; the non-mass version of this active learning already exists as apprenticeship, but apprenticeship cannot be printed and distributed in a billion copies like a book.

normativespeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“The regulation, the choice of what activity to do, is in the course, is in the structure of the material.”

0.33

Some disciplines like programming are legible enough that books and courses (e.g., MIT OpenCourseWare) can lift learners up to build real things, but other domains—design, musical composition, architecture—are not, requiring studio classes with extensive, highly contextual, contingent hands-on feedback that we haven't figured out how to communicate at scale.

factualspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“we have some disciplines like programming where there are a million books on courses about how to learn to program... This is not true in all domains. In particular, design... Nope. It's normally done in studio classes. Lots and lots of hands-on feedback.”

0.33

Streaming is exciting as a medium for transmitting tacit knowledge—the unwritten pattern recognition engineers develop from handling thousands of weird situations—because watching someone like George Hotz work lets viewers absorb some of that knowledge as apprentices do; but it lacks feedback and contains much meaningless chaff, and the bigger gap is the absence of streamers in domains like design, musical composition, and architecture that depend on highly contingent studio feedback.

factualspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“By watching George do it, I think that people do absorb stuff. They can absorb some of that knowledge. That's part of how apprentices absorb that knowledge. There's a few things that are missing. You're not getting feedback.”

0.33

Tools for thought nerd-snipe Silicon Valley because they promise technologists empowerment and feel actionable in a misleading way: building an interesting tool for thought is almost entirely a design problem, yet 95%+ of interested people are engineers whose design ideas are usually weak, and the meta nature of 'work about the work' scratches the same itch as endlessly tinkering with dotfiles or a near-empty blog.

causalspeaker onlynovelty 3/4durability 3/4· Andy Matuschak

“The problem with this is that building an interesting tool for thought is basically entirely a design problem. Their design ideas are usually not very good”

0.31

Andy was much better at his Apple work than at his current research because Apple work was mostly well-understood engineering with others deciding what should be done, demanding heavy memory but lower metacognitive load, whereas research requires deciding what should be done and what intervention to make—synthesizing unfamiliar literature—which is the hardest and most draining part.

factualspeaker onlynovelty 2/4durability 2/4· Andy Matuschak

“Metacognitive load on me was lighter because others were figuring out what we should be doing.”

0.31

Using memory systems to thoroughly learn detailed facts about the education market early at Khan Academy let Andy participate in high-level strategy conversations, evaluating proposed directions on the fly with specific figures about budget spending and market growth in a way that colleagues without that recall could not.

factualspeaker onlynovelty 2/4durability 2/4· Andy Matuschak

“I could say things like, "Well the total budget spending for instructional materials is this and that market is growing by this percent per year and 10% of students in the US are in this place"and so on and so forth. Basically I could evaluate ideas on the fly in a way that others couldn't.”

0.25

In neural networks you can sometimes improve performance by pruning some of the weights.

factualestablishednovelty 1/4durability 1/4· Dwarkesh Patel

“in neural networks where sometimes you can improve performance by pruning some of the weights.”