YouSaid · the spoken record

Andy Matuschak

lines on the record: 158
first: 2023-07-12
most recent: 2023-07-12
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“A combinatorial search space programming often feels like a search problem too. You have a sense that there's some right way to solve the problem. There might be some set of right ways to solve the problem and you're kind of looking for it. And you have some heuristics that guide you to like, oh, this might be a dynamic programming problem. This might be something that is solved well by separating concerns or something like that. Design often feels less like that. You have those heuristics too. You have those patterns too. Often it just feels like nope, I just need to try like 300 things. And so there's this characteristic where you'll look at a designer's figma file and what you do in Figma is you press command D. This is like the core action of Figma is to duplicate. So you have an artboard, you tried something, and you're like, that didn't work. You press command D. And what you end up with, and you look at design Twitter, and it's just all these screenshots of people's figmas with like...”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Then immediately you run into one more thing all the way through, it's almost going end to end, and you're like, well, I'll just stay up a little bit longer. And you know, before you know it, it's like 2 a.m. And you keep going, cause it feels so good. Like, you feel the sense of forward progress. You're not just staring at a wall. I think for the programming problems where you are at a brick wall, it doesn't feel like this. It feels bad. So can every field be transformed into something where you can feel the sense of forward progress? You can get this rapid feedback cycle. I think that's really hard. I think some fields, like it's not clear to me that they can be transformed in that way. I think we can get a lot closer in most cases than we're at right now. So what's hard about designing the user interface is that often there's this feeling of exploring kind of”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“I think what makes programming feel like This sense of instantaneousness, this sense of direct contact with the environment, you're learning about a kind of a conceptual world, but that world is right underneath your hands. And as you manipulate it, you're constantly getting this feedback, the red squiggly, you're pressing command R regularly, and you're seeing it fail. And that feels great. Like there's this feeling that's very common for programmers and it's laden with doom. The feeling is it's like 9 p.m. and you've been working on a thing all day and it's almost working. It's almost working. And if you just debug this one thing, then your project will be done and you'll be able to go to you. You're like, well, I'll just stay up and I'll debug this one last thing. And then you start debugging it and you get it and you solve it and that feels great.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“The feedback is highly contingent, it's highly contextual. We just haven't figured out how to communicate this. And so it's good to see lots of programmer streamers, but I really want to see the streamers in these other domains.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“So I'm pretty actually pretty excited about streaming videos. I've complained loudly that there aren't more designer streamers. So one of the things that I think is really interesting is that we have some disciplines like programming where there are like a million books on courses about how to learn to program. And they don't give you everything you need. There's this tacit knowledge stuff that you need to develop. You know, like if you work through these courses, if you go through the MIT open courseware for computer science, you'll be able to build some stuff. And you'll be able to leave yourself up. This is not true in all domains. If you're looking at, well, in particular, design, but lots of other domains that are sort of like that, like musical composition, architecture, something like this. Nope. It's normally done in studio classes. Lots and lots of hands-on feedback. Stuff has”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“I'm incredibly excited about streaming, actually, as a medium for this. So, we're kind of gesturing at a particular kind of learning that needs to happen that's often called tacit knowledge. So one of the things that you have to learn to do as an engineer is to learn to deal with 100,000 different weird situations where something is not behaving the right way. And eventually you learn a kind of pattern recognition, you learn ways of dealing with this, and much of this is not described in any book. It's not explicitly taught. You just kind of learn it by doing it over a long period of time. So 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. There's a whole lot of chaff there. Like there's a whole lot of stuff that probably isn't all that meaningful. It's also true for apprentices.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“That, but that can be applied in many, many circumstances. We have the non mass medium version of it already, and it's apprenticeship. If you want to be a good yoga teacher, you go hang out in yoga studios. If you want to be a good surfer, you go to the beach when the other surfers are there and you participate peripherally and you talk to them and you learn about their tactics. They might give you some feedback eventually and you'll start to participate less and less peripherally over time and eventually you'll be part of the community. This is in a mass medium. We can't print a billion copies of it. We can with a book.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“This is a very interesting project that's kind of along these lines. And what characterizes it like games is doing, it's active. So when I was asking all those questions of the book, that was active learning, active reading, NAND to Tetris is naturally active. So this is a course in which you kind of start with basically nothing. You start with memory and you build a virtual computer and you build Tetris. You build a processor and stuff. And so the whole thing's active. The whole time you're making the computer go. This is doing a similar job to the question asking that I was doing, except that you don't have to regulate all of that yourself. The regulation, the choice of what activity to do is in the course, is in the structure of the material. I think there is waiting to be created some kind of mass medium that is like”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Maybe kind of we don't quite have those yet. We have some things that are kind of like that. I don't know if you've seen from NAND to Tetris.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“People learn some things about project management or orbital mechanics from that. Zachtronics has a bunch of games that are sort of about assembly language, roughly speaking. Maybe you can learn some things about that. The issue seems to be that games are ultimately, they're 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. Grant's videos are also serving a different purpose from the text. Like the text you might pick up because you're like, I want to be able to build a robot. So you pick up a textbook on robotics or something. And so is there something that you could pick up that's sort of like a game insofar as it's an active environment that you use in a similar situation to, I want to learn to build a robot.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Extremely complex mechanics, basically, of a system that exists in this world. So you learn a bunch of stuff, and it gets to the point where it feels like you're in conversation with the game's designers. It's like, ah, he's asking me to do this here. No one's asking you, right? There's no text, but you can feel that you are being asked and you perform some interaction in the environment and you feel that you have answered and the game responds in kind. This is very, very interesting. It's like it's a medium of action. Some people have tried to make educational games, games which are explicitly about arithmetic or something. Jonathan Blow's game is not about that. It's the mechanics that you learn, they're about the environment. I don't think anybody has yet really succeeded in doing this about explicit subjects. There are, for instance, things like Kurpal Space program.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“I'm fascinated by this question. Let's see, I can say a few things about it. One is that I would argue that one medium is taken off in an absolutely enormous way, and that's video. People love video. 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 in general not walk away from that.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“And often then I will find that I need to bring more gumption to bear to kind of power through and like, you know, make myself sit there and keep flipping the pages, then I need if I actually just open my curiosity and open my attention and really start engaging with the book.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“But I want to make the case that there is a kind of pocket that you can fall into. Maybe you call it flow, I don't know, where the demandingness that you're bringing to bear is matched to your ability. The book is not overwhelming. You feel like you can make your way through it. And this is actually more engaging. So I occasionally will find myself reading as an undemanding reader and finding my attention kind of slipping because I feel like I'm just not that attached to the text emotionally. I'm kind of reading dutifully. I'm like trying to get through it. That produces sometimes like an adversarial aspect where the text is in my way or it's kind of something to be I don't know accomplished.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, I mean, so this is sort of true, and it's definitely the case that I will occasionally do some of this before bed reading where I think, oh, let me just do a little bit more, and it's like it's basically useless.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Demanding in kind of a direct way than what I actually do day to day. But it's much less demanding on what William James calls the energies of men, which is something like a life force that permits you to act according to your will or something like that. Maybe it's gumption. Maybe it's willpower. Maybe some people call it spoons. I don't know. These aren't all the same thing exactly. But sitting and staring at a page and deciding what you should do next on a research project is incredibly draining on that resource. The not knowing, sitting and not knowing is like the hardest thing that I do in my work. And so there's something, it's like a wonderful vacation to be presented with, oh, great, somebody else is going to tell me what to do. This is great.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“It wouldn't have been nearly as energizing for me, and I definitely would have taken breaks. I still would have been able to go for three hours, I think. And part of the reason for that is that it's simply way less hard than things I normally do. In some sense, learning quantum mechanics should be much harder, and it kind of is cognitively demanding in a lot of ways. It's much more cognitively...”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“So, one of the things that I think is funny about stamina is first off, there's some kind of weird grass is always greener kind of situation where I often feel struck by other people's stamina and feel like I have very little of it. I struggle with energy. I've actually written extensively about all my struggles with energy and ways of managing energy. I spend a lot of time thinking about it and managing my energy levels, structuring my day around it. So I think there is something where one often feels maybe lower stamina than one actually is because one misapprehends other stamina. Okay, so in that particular situation, how do I explain why three hours of studying, et cetera? First off, social. So if I were alone and studying that book for three hours and I weren't effectively trying to perform for you,”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“There are still going to be instances where the naturalistic reinforcement isn't going to work. So, for example, the linking that you brought up, one issue for doctors is rare diagnoses. So, if it's only going to be once every couple of years that you see a patient that's going to present with these symptoms, that's not going to be frequent enough to naturally reinforce your memory of that. And you're going to need some out-of-band mechanism. And unfortunately, I think for many kind of creative leaps and creative insights, that may be closer to the regime that we're in.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, right. So again, the answer is going to be it depends. I think that maybe the most familiar example of what you're talking about. Likewise, a great way to become a better musician, all of the best pianists at sight reading that I knew in university played with churches. They were so good at sight reading because they had to show up every Sunday and they're playing a different thing. Knew him every Sunday, right? So this is immersion also. And over time, they're learning all these cadences and these things that are really common and whatever. But you can't show up and be the church pianist every Sunday in the first place if you don't already have some decent foundation. So this is sort of a bootstrapping argument that one role for explicit practice of this kind is to get yourself into a position where you can more naturalistically reinforce.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“And engineering in general. I had this really rich network for stuff to slot into. So learning stuff is easier when you have other stuff to connect it to is a nice principle. Metacognitive load on me was lighter because others were figuring out what we should be doing. So just like by contrast, now I'm doing research, like I'm trying to discover things that are not known. I'm trying to make things that didn't exist. The hard questions that I answer are mostly like, what should be done or what should I do? And that question is not just a technical one of like, how should I implement this feature that needs to get built, but what intervention on a reader should be taken? That requires synthesizing Lots of different unfamiliar literatures.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Familiar and like it was mostly flow all the time every day. I was kind of like I was just in it. I knew the stuff that I needed to know. I was very well practiced. And the space didn't change that much. Engineers, most engineers at Apple most of the time are not pushing the frontier of what is known, like trying to discover. They're doing very difficult technical work, mostly applying things that they already know and understand quite well to problems which are usually, not always, moderately well and pretty well understood, let's say. Memory was essential to me doing that job well, but I had already built most of it by the time I got there. I'd already built just tons of stuff for Apple's platform. And I had to learn a lot of stuff. I learned a ton of stuff about the internals of those systems. But because I already had such a rich understanding both of Apple's platforms and of computer science.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“It was just totally different. Let's talk about this a little bit. This feels very, very juicy for me. Most of what I was doing was engineering, some of it very difficult engineering, but mostly engineering, mostly on things that were fairly well understood. So I wasn't trying to decide what should be done usually. Sometimes I was from a technical perspective, but certainly rarely from a product perspective was that a irrelevant question for me was kind of like a somewhat design-minded engineer and I did a bunch of kind of engineering and designish things on tasks which were set out for me. At that point, I had been programming for a really long time, I don't know, 13 years maybe by the time I joined Apple, longer. And programming in Apple's ecosystem for probably two-thirds of that time, three quarters of that time. So everything was just really...”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“No, that's a great question. So, I mean, like, here's the fun thing. I was much better at what I was doing then than I am at what I'm doing now. That's pretty funny.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Don't necessarily exactly. And so when you repeatedly access something, when you practice retrieving it, the prediction of the utility of the thing goes up. And when you do it in a variety of situations, it goes up across a broader distribution.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Have this weird cacophony on the network, and in particular, what's salient, what to do next, which response seems most appropriate to this question, like you might answer those kinds of things very ineffectively because all this stuff is coming up for you that is like much less relevant. One of the theories about how well we remember stuff in what circumstances is actually called 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.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“No, it's a good question. So we're getting to a part of the cognitive science space that I'm less familiar with and also that I suspect we simply know less about. But let me just riff a little bit. One of the things that we sort of know is this idea of spreading activation. So when you go to try to look something up or when you try to deal with a particular situation, there's something almost kind of like DNS exchanges or like routing on a network or something where, okay, we start from some point that it's like a stimulus and just speaking very informally, like we kind of expand outwards from there. And there are effectively weights on those connections. And by tuning those weights effectively, we route the packets on the network effectively. Memory is encoded in these weights, at least partially. So if you didn't forget things, then you might...”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“And some of these memory champions have memorized truly absurd orders of magnitude, maybe two orders of magnitude more things than I have practiced. Certainly people who are multilingual have really, really absurd numbers of things memorized. So there isn't a resource management argument.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah. So, I mean, another weird thing about memory is that as far as we can determine, memories aren't lost exactly, at least not completely. There's a bunch of interesting experiments that people have used to demonstrate that decades later, things are still there. If you can queue them right, people can bring things back, even things that they feel are lost. And of course, you can also queue people in ways that are hallucinatory. So we need to be careful about that. I guess the reason why I bring that up is that it flies in the face of this view that like there's a limit. One of the things that I think is kind of weird about this memory system stuff or memory championships or something like that is people are like, oh, if you do, if you do these things, will you start to forget other normal human stuff? And what's weird is no. I've been doing this memory system stuff for years. I just know more stuff now.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, it could be. So memory is very connected to attention. And we can't attend to everything. So one of the roles of memory is to help guide us to the things that are important. So maybe I happen to know the magnitude and energy of an electron volt. That's something I can draw on because of the memory system stuff. But I also don't like, I don't want that to be like front and center in my mind all the time. I don't want it to be hyper salient the way that I don't know some very important design principle is to me. So yeah, there's some role there. There's some theories that the reason we have forgetting is that our environment, our ancestral environment was very traumatic. And so our episodic memory in particular, we would like to maybe not be all that faithful. I actually don't know the status of those theories.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Personal, if it feels very inquit, I often haven't been able to describe textually the constituents of the thing very clearly. There's just kind of a feeling that something in this general direction is connected with something in that general direction, or there's a tension. And so that makes me a little hesitant. LLMs depend on our ability to externalize things and to make them legible. Back to the learning point about the role of memory. If what you're trying to do is to understand something pretty difficult, your ability to understand that thing is still absolutely going to be bound on your memory of the constituent material.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“In a way that others can't. So, anyway, this and other things are kind of just part of my rant about people, I think in general, we underappreciate the role that memory has in our lives. So just to come back to the question, explicit memorization or explicit making sure that you can recall the thing reliably. We can test it against these things. So for the case of the creative instinct, for instance, noticing the contradiction, noticing the connection, I imagine that we will have future notebooks that will do some of this noticing with us and that will decrease our need to be able to rely on our own sense of salience or something like that. But I guess I don't know how much. I'm skeptical that my own experience coming up with weird ideas that feel very new is that it feels very”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Now, that's only possible if the other thing is in your memory. If you have to think to look up the other thing, then the newspaper wouldn't seem so salient to you. Likewise, in just really boring ways, early on in my time in Khan Academy, I just learned in a very thorough way using memory systems just a whole lot of details about the education market. And this let me be in high-level executive kind of conversations where we're trying to figure out strategy stuff and somebody would propose a particular direction or what about this? What about that? I could say like, well, the total budget spending for instructional materials is this and like 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. So basically like on the fly I can evaluate ideas.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Even in the course of a text is memory bound. Some of that's working memory, but 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. Likewise, in creative work, there's a bunch of studies trying to catalog case studies of how it is that people have flashes of insight. You know, it's a little hard to talk about that. But one of the things that's a pretty consistent source for insight for people is noticing a surprising connection or a surprising contradiction probably feels pretty familiar, right? Like you're reading through the newspaper and you see people have finally figured out how to do X and you're like, wait a minute, that means if I combine it with this other thing, we'd be able to do Y or something like that.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Maybe this is a good time to talk about what memorization is or what it's for. So we could use that word to refer to the practice of learning more trivia, for instance. So for instance, a thing that I and some people I know have done is gone through this book, Cell Biology by the Numbers, which says all of these things, like how big exactly is a nucleotide? volume does it take up it's kind of helpful occasionally to know that it's about a nanoliter and that that can help you model things so you can just like commit all of those things to memory right that's one kind of memorization we could talk about how lms affect that but I just want to make the case that so much of what you do and experience day to day is memory bound or is memory influenced in important ways so just for instance your ability to understand a difficult argument”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“And he's got decades of practice thinking out loud and like writing down a decent take the first time. And so he gets something pretty good the first time much of the time. And that works for him. So that kind of is a note, right? doing the thing that I guess like your initial thoughts on the subject is kind of what you would write in a note.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Tyler's a good example. I think he's actually a little easier than some others we might discuss. So let's talk about Tyler for a second. One of the other things that's so interesting about Tyler is his writing obligations. So this is a man who's blogged every day since, I don't know, 2007 or something like this and has a Bloomberg column, I think weekly, something like 1,500 words, and also has published something like a book a year for, I don't know, a decade or more, and occasionally publishes some academic articles. Plus a bunch of other collateral. So that is notes. And I think it's also important to note that the way that Tyler writes these blog posts and the way that Tyler does these columns and even the books is very different from the way that many other book authors work. The blog posts often have this real first draft mentality.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Or actually understand material at a higher level, then you could have before because less of yourself is engaged in this kind of question of how do I make the questions from the material.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Well, yeah, I mean, I expect the main lane, which you're getting better, is actually not really about any of the details of those materials. I think it's about your practices as an interviewer, the way that you generate questions, like you probably have a bunch of patterns, whether you know it or not. You read a thing that a person has written in hopes of generating good questions about it. And even though you don't have this habit for textbooks yet, maybe of constantly demanding things of the textbook, you have, I think, started to develop this of essays or blog posts that interesting people you're interviewing have read to point to this Anderson theory, in the course of repeatedly doing that, you've made it automatic, parts of it automatic, so that you don't need to do it consciously. You can focus more on the material. You can probably take on more difficult material.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Anderson's theory is just that he has a bunch of specific claims about what's necessary for knowledge compilation to happen and what you'll be able to do as a consequence of certain degrees of knowledge compilation. I think he'd probably respond to this by saying something like actually in order to effectively compile things that you've learned into schemas that will match feature scenarios effectively then you need to be exposed repeatedly to those things. You need to use them to do a variety of things that will basically show your brain that it is relevant to apply these things in combination. And simply reading probably won't do that. But if you read and you have a lot of conversations and you're in a context where it's kind of the manding and it's drawing on what you read, then you may naturally do that kind of compilation step.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Fine. Very practically speaking, it's funny that he uses the word compile because one of the prominent theories of cognition, that is how we come to know and learn things is this theory called Akdar by John Anderson. And a key part of it is this process that he calls knowledge compilation. This is the process by which we take individual facts and turn them into kind of higher level patterns that we can generalize and apply in more contexts. And I think that's what Paul is gesturing at, that you read a book, it contains like a story or case study. And by reading it, you learn to generalize to some extent and you apply it in other contexts when it seems relevant. And the reason why I bring up”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, yeah. I mean, what he's saying is true to some extent whether or not that extent is sufficient is going to depend a great deal on the situation and on what you need. If your aspiration actually depends on having a deep detailed understanding of the material, then the kind of imprint on your worldview or on your automatic responses or something like that made by the book may not be sufficient. On the other hand, if what you want is to absorb a lot of different ways of looking at the world, knowing the details of these isn't necessarily important, maybe just want to know, well, Confucius emphasizes community and society as a moral patient and contrast maybe to the individualism of a bunch of humanist philosophers. And that's kind of the level that you feel like you need to make decisions in that domain.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“That's true. It's also the case that you don't even know all the stuff there is. And this is another key problem. There's another reason why we outsource stuff There's a fundamental tension in unschooling, for instance. Let the kids pursue what they're interested in. And like, that's cool. There's a lot of good things about that. But also say that a kid's true passion turns out to be, I don't know, ocean geology or something and they're in a landlocked country and there's just no one around them that talks about ocean geology, then they're missing out on some great opportunity. But if the school had a program where they are bringing in guest speakers or whatever, and then there's a special lecture on ocean geology from this person and it lights up the kids' world, even if they wouldn't have chosen that lecture, that's a good thing.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Not in the same way for most subjects. And I think that's mostly just because the amount of cognitive demand that's placed on me by the subject is just much lower than it is for other subjects. So much of it is familiar already that I can deploy my own planning more effectively as I go. But it's also the case that because I know so many things about the subject, I can do a better job from the get-go of making a plan. Making a plan requires kind of modeling a path or predicting a path or saying, well, I guess I'd need to see how this connects to that or something like this. And 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 kind of infer pretty accurately.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Where the metaphor breaks down is that although scaffolding is ubiquitous in education, we give you simpler versions of questions first. That's a kind of scaffolding. We partially work the answer first. That's a kind of scaffolding. We give you worked examples first where we might ask you to predict the next step of the worked example. That's also a tenant scaffolding. Where the metaphor breaks down is that once you become more capable, we try to remove the scaffolding. It's called fading. The idea is that once you have solved a lot of calculus problems, you don't need half of it worked out and you're just filling in one of the blanks anymore. And in fact, doing that would not be as effective a learning experience. So the application with the syllabi might be something like, if I'm studying something in computer science, which is a domain that I know really well, I don't need those syllabi.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Less on it. And you probably should. I think like a very common mistake that people make is to. Feel like they need to do the thing the right way, and that is exhaustive and completionist or something. If they fail because they find themselves bored or unmotivated because the material doesn't actually seem to relate to what they want to know, but they're just kind of going on faith that like, well, if I follow what the author says, everything will be good. Anyway, they find themselves having trouble for that reason, and then they just stop. So this is bad. And they would be better off just skipping around according to their interest and continuing. One other thing I'll say about this is that the role that these syllabi play is as a scaffold, this is sort of a term of art from learning science, but it's actually, it relates to the thing we're familiar with. If you want to get higher up a building, you may not be able to climb it yourself, but you can build some scaffolding around it. And then suddenly you can reach that top shelf or the top of that building.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“This is, I think, like a pretty classic issue for learning in general. You have this problem where you have to sort of boost yourself in a domain, you have to outsource the question of what is necessary to know. You might know, for instance, that you really want to build a model that can generate images, given descriptions, or something like that, like Midjourney, but you don't even know what you need to study to do that. So you pick up some textbooks on machine learning. You're kind of outsourcing the answer to this question to the author. Like, what is necessary to know, to build things? Maybe you can find a book that's actually labeled What You Need to Know to make an image generating model. But even then, you're outsourcing your answer to the author. So you can take that answer as a start and treat it as tentative and revise it iteratively. And as you become more skilled, you can”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Of making you kind of reflect on what you've just learned. And in addition to the questions being asked, you might find yourself pondering, well, I'm being asked about this, but why does this matter?”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“That you simply didn't even notice. And so not only does that give you feedback, so it tells you, oh, maybe you need to go reread that specific section, but it may also change your behavior towards future sections. So in interviews, readers told us, for instance, that after they reach the first set of questions or a particularly difficult set of questions, they found themselves slowing down and reading more attentively or realizing actually that they're reading practices were ineffective in general, kind of in the way that you were mentioning towards the start of the conversation. There's been a bunch of research on the key phrase here is something like adjunct questions, questions that go along with a text. That's kind of like what I was just talking about. And they have all kinds of effects. So the adjunct questions have the kind of effects on forward material I was just describing. And they also have the effect of”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source
“Every say 1500 words or so in this text on quantum computation. And our primary intention in doing this was to help people remember what they read. And we had this theory that part of what makes it hard to learn a complex subject is that there's all these new definitions and notation and terms and things being thrown at you at once and you're being asked to combine these things which are still unfamiliar. And so you're constantly having to retrieve these elements and struggling to do it either it's taking a while or your success rate is low. So anyway, that was our motivation. But it had this other metacognitive benefit that was really important that you read 1500 words and now you're being asked these questions. That 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.”
2023-07-12 · Dwarkesh Podcast · Andy Matuschak — The reason most learning tools fail · IDENTIFIED FROM THE TRANSCRIPT · source