--:-- Dwarkesh Patel: I feel like people are just way too unambitious in general. Not in the "I want to be better than other people" way, but they just don't think of big projects. They don't work on them. They don't have big dreams to do cool things. Or if they are, it usually just boils down to something like social status. Like, "I want to be the person that does this, that's better than other people." I don't know how you change that. But I do think rewarding a culture where you want to do ambitious, original things that are interesting, where you don't know where they're going to lead—I think having that in you is rare. And I think cultivating it is probably good for yourself and society.
--:-- Dwarkesh Patel: Okay, today I have the pleasure of speaking with Scott Young, who is the author of the book *Ultralearning: Accelerate Your Career, Master Hard Skills, and Outsmart Your Competition*. So Scott, I'll ask you some practical questions in a second. But first, let's talk about Einstein and Newton. They both had an *annus mirabilis*, a miracle year, within which many of their important contributions were concentrated. What explains this phenomenon of miracle years?
--:-- Scott Young: Well, I don't know. I think whenever you look at these sort of outlier people like Newton and Einstein, you have to realize that most people never have a year where they accomplish anything like that. I think a lot of it is a selection effect: you have a smart person who just happens to be working on the problem that will lead to a huge breakthrough. We could have lived in a world where Newton spent a lot of time on alchemy and then discovered the way to turn lead into gold, and that works. But that's not the world we live in. His work on physics and the *Principia* and stuff like that was what led to the breakthrough.
--:-- I think Einstein's a little rare in that he had a couple of key insights that led to physics. He discovers, or sort of proves through Brownian motion, the existence of atoms. Then the photoelectric effect, which is the thing he actually won the Nobel Prize for—not his relativity, which is what he revolutionized physics for. I mean, it started quantum mechanics, so you can't downplay it too much. But then special relativity, and then he struggles with the math for like eight years to get general relativity. So Einstein is a bit of an exception in that he did have multiple huge breakthroughs. When people talk about lists of geniuses, sometimes that list gets populated by people who don't really deserve to be there, but Einstein is definitely not like that. He's an accurately rated genius who is seen as being extremely smart and important, and actually is extremely smart and important.
--:-- Dwarkesh Patel: Right. But as you mentioned, there were different problems. Unless there's a deeper principle which I'm missing, I am having trouble understanding why special relativity, the photoelectric effect, and Brownian motion happened in the same year.
--:-- Scott Young: I don't know. I think for Albert Einstein's case, the kinds of problems he was working on were also amenable to his style of thinking. I'm a big fan of the Isaacson biography of Einstein. I talked about it and did a little summary post on my blog. You can see that Einstein is really one of the great intuitive physicists. He's very much a spatial, visual person. His thought experiments are really the mechanism he's using to generate these insights.
--:-- Putting the photoelectric effect aside for a second, special relativity is this: we have these weird experimental results showing that the speed of light doesn't vary depending on which direction you measure it. That's weird because if you think about a wave going through some medium, if you're moving relative to the medium, the speed should change. But it doesn't seem to do that. Working through the geometric implications of that, you get to this idea that lengths will contract as you go faster. These are all mind-bending, but they come from rigorously working out the intuitions. You can see that as being somewhat different from the more mathematical physicists who were very strong at advanced math. It was less about "what's my physical intuition about this?" and more about "what is a way of representing this?"
--:-- I may be getting a little outside my comfort zone, but I'm imagining people come up with matrix mechanics. It's like, "Oh, this is an interesting pattern. This is one way you could do it that makes the math easier." I believe it was Minkowski who Einstein worked with on the tensor stuff, because he was a little more limited there. Not that Einstein wasn't also brilliant at math, but that's definitely not what led to his huge breakthroughs. He had some intuition, and then he worked to formalize it. For other people, it might be, "Oh, this is an interesting mathematical pattern. I wonder if it would work for this particular problem." So I think it may just be that these types of problems were uniquely suited to his style of doing physics. If you look at the different time periods, it may just be that there were a few insights that required someone with this kind of skill set to unlock them.
--:-- Dwarkesh Patel: Yeah. This also relates to an essay you wrote called "The Narrow Path of Success," where you explain that you kind of have to be following the right track to be successful in many ways. Doesn't the example of Einstein confound that explanation? You have somebody who's a patent clerk and can't get an academic job many years after he did special relativity. He still couldn't get an academic job.
--:-- Scott Young: Yeah. Well, when I'm talking about the surprisingly narrow path to success, this is contrary to this romantic idea that in high-ambition fields, you can just do whatever you want and someone will recognize you and it'll work. If you actually get data on how these careers work, it's clear that's not how they work. The motivating example that caused me to write that post was reading Jason Brennan's book *Good Work If You Can Get It*, which is about a data-driven approach to analyzing how careers work in academia. The picture he paints is stark. There are way more people going into academia than there are academic jobs. The actual process of getting those jobs is quite rigid, and the filters are quite rigid.
--:-- Einstein is a bit of a counterexample, but he kind of isn't. He had some bad things on his resume and had a really hard time getting a physics job. The right way of looking at that is that Einstein struggled to get through because he didn't have the right resume, and you are not Einstein. There are always these exceptions where someone had the wrong start and then had a truly spectacular, one-in-a-million result that brought them stardom. But you can't count on that. That's the lesson here.
--:-- We romanticize the Einstein stories, but if you're looking at the typical path for people and how 99% of people find success in this field, that's what you should be betting on. You shouldn't go into a poker game thinking, "If I get a royal flush, then I'll be really good." You need to bet on, "Given that I probably have an average hand, what's the way I should play?" If you are in the top 0.01%, don't listen to me because you're smarter than me. You know what you should be doing. But if you're like everyone else trying to figure out what you should be doing, understanding the path to success in most fields and what the typical path is, is so important. Generally, if you are outside of that path, you're going to be facing strong headwinds. If you're a remarkable Nobel Prize-winning genius, maybe you will see some success. But if you are not, you're stacking the odds against you for no reason.
--:-- One of my favorite examples from that post is about non-fiction book publishing. There is an established path. That path is: you first get an agent, because getting an agent is easier than getting a book deal. Then you work on a proposal, which is partly what the book is going to be, and partly a business plan to show that you've thought out why this book will actually sell enough to interest the publisher. Then you pitch it, you get a book deal, and then you write it. The writing comes way late in this process. What most people do is they write the book first and then look around for people who will publish it. That screams "amateur" to publishers, and they don't like that. It's very difficult, even if your book is pretty good, to get a book deal that way. There are exceptions, but it's difficult.
--:-- If you're serious about becoming a nonfiction author, the first step would be to try to get an agent. Obviously, if you're self-publishing, none of this applies. But if you're trying to go through a traditional publisher and have a published book through a serious publisher, that's what you want to do. I remember I wrote this and someone was upset, like, "Why do I have to do that? Why do I have to do it this way?" It's not that you *have* to do it that way. It's just that if you're not doing it that way, you're making it harder for yourself. It's always fine to do something weird and creative, but you should know that's the strategy you're picking and that you're making it hard for yourself in certain ways.
--:-- This narrow path to success ties into one of my major points, not only in *Ultralearning* but in my philosophy in general: people really ought to do more research about what the typical way is that these kinds of things succeed before they embark on projects. Even if you decide, "I'm not going to do it that way, I've got my own path," you should know what the status quo is. You should know what works. You shouldn't just say, "Oh, I didn't know that's how people did this, or that's how you got success in this field." This research is not even that hard to do for most things. It doesn't require super secret knowledge. Most people just don't know how to do it, or they don't care to do it, or they don't want to find out the answer because maybe it's not something they want to hear.
--:-- Dwarkesh Patel: Right. Okay, so maybe we shouldn't be studying the exceptions that hard. But let me ask you one more question about how Einstein's way of learning contrasts with ultralearning. He seems to have lacked the sort of structure, discipline, and organization that an ultralearning project kind of requires. Is that evidence for a more fun-based, excitement-based, and curiosity-based form of learning?
--:-- Scott Young: Well, I don't know whether that's true. Again, I'm going off the Isaacson biography here. But Einstein did an enormous amount of deep thinking on hard problems when it wasn't necessary. He's the patent clerk working through some of these hard ideas. So I don't know whether it fits this mold.
--:-- I think it's important to keep in mind that when you write a self-improvement book like *Ultralearning*, you do want to create it in a somewhat structured way. You're taking what were, in some ways, spontaneous intuitive approaches that successful people used, and you're trying to turn it into a structure that someone who wouldn't do that automatically could follow. Similarly, if you were to analyze any domain of importance and relate it to someone, you have to make it into a format of, "Okay, you need to follow these steps."
--:-- In my mind, Einstein was one of the people I researched for the book. I would have considered him as one of the vignettes if he wasn't way too famous and people have already heard too much about him. It wasn't because I thought he wasn't suitable. There are certainly people who break the mold because they're doing a lot of things that cognitive science seriously considers bad practice, yet they're really successful. I kind of give Terrence Tao as the counterexample—someone who just seems to be really, really, really smart, and he's not really doing anything particularly special. There are other people I talked about as mild counterexamples. But in my mind, Albert Einstein definitely isn't one of them. He seems to be someone who followed a lot of the ideas, even if he wasn't doing them in a structured approach. They just came naturally to him.
--:-- Dwarkesh Patel: Interesting. What's the relationship between age and ultralearning? Is there a prime age where you're at the peak of both your plasticity and your maturity to be able to learn things?
--:-- Scott Young: There are kind of two questions here. One is: are the principles and techniques I talk about in *Ultralearning* age-specific? I would say probably not. It seems to be the case that retrieval works better than review. I don't see a reason why that would change as you get older. Similarly, feedback is going to be important whether you're 79 or 17. I don't think those things are going to change. So you could read *Ultralearning*, and if you're thinking, "What practical step should I take to learn X or Y?" I don't think the advice would change too much.
--:-- The other point you're making is about ultralearning as this ambitious learning project. How does that change with age? That certainly does change. We know that fluid intelligence probably declines from your early 20s, kind of straight line downward. Things like working memory also decline. I did an essay where I dove into some of the research on aging and learning—it's a really interesting subfield. I did it after the book, so I didn't include any of it.
--:-- One key finding is that the frontal areas of the brain seem to be the parts that deteriorate faster as we age. The two main things that tend to be harder as you get older are related to the frontal areas. Think of this as your executive control: dictating where you should focus your attention, being able to switch rules for things. If I give you a puzzle where you have to apply a certain rule, but then I switch it suddenly and you have to apply a different rule, there's more perseverance. Older people have a harder time switching because the habit they want to override is harder to switch back and forth from. This suggests that if you're older, you need to pay more attention to having an environment conducive to focus. If you're 21, maybe you can have the television on in the background and just tune it out. But if you're 75, you may find it impossible to keep from being distracted because of that frontal area stuff.
--:-- There's also stuff on chunking. There seem to be difficulties with the medial temporal lobe areas, which are involved with binding information. That's a big part of the intuition chapter, where I talk about chunking—assembling pieces of information together so they can be attached. It seems that older people have more difficulty doing that, which would obviously impact learning. It's also why someone who is older might recognize a person but can't remember their name as easily, because the name and the facial recognition don't bind as strongly.
--:-- There might be a benefit to being more explicit in how you put information together so you can bind it more easily. For instance, there are studies showing that if you're making flashcards and you have to do more manipulation to show how the parts connect—like in this example, learning Chinese words. There are two characters and two pronunciations. If you put them side by side, you have to take the first part of one and link it with the first part of the other, and the second part of one with the other. It turns out this makes it somewhat harder to remember than if they're on top of each other with a natural, easy visual link. If you're older and struggling more with these issues, organizing your material better, making the connections between things you have to learn more explicit, and doing that pre-processing work might be a little bit better. But those are also things that would benefit people who just struggle with learning more. If you feel like you're more distractible or struggle to understand concepts, those would also apply.
--:-- The broader thing that matters is that when you're learning something, using the correct strategy is mostly going to be the same. I don't think there's some strategy that works for really smart people and some strategy that works for people that don't. The same things probably work. It's just going to be easier if you're more intelligent or you're younger or you have those advantages.
--:-- Dwarkesh Patel: Sure. Yeah. Although the principle of transfer seems to be injured by your explanation that your capacity to look beyond superficial differences and recognize deeper principles is harmed by age, right? Or to connect different concepts.
--:-- Scott Young: Yeah. I don't know whether that's the case. There are two issues there. The issue of being able to look at what the deeper principle is, I think, is part of this problem: you need a lot of exposure to a field and to the knowledge in the field to build up a repertoire of patterns so that when you see it, you can actually see what the principles are. There's a classic cog sci study where they took physics novices and experts and showed how they look at problems. The novices focus on stupid stuff like, "Oh, this involves a pulley or an incline ramp," whereas the experts are like, "Oh, this is a conservation of energy problem." The naive way is thinking, "We should just tell people how to recognize whether it's a conservation of energy problem." But the problem is that the conservation of energy aspect is an abstract property of the problem. Whether it involves a pulley is an obvious aspect. What you need to do is look at all those obvious aspects and then figure out what the abstract or higher thing is. It seems like you probably do this through chunking. You learn all these smaller patterns and build them up so that when you see it, it all comes together and you can see what the situation is.
--:-- I do think it's probably the case that if you're learning something with a lot of abstraction, really spending time to familiarize yourself with the ins and outs of the more basic pieces allows those things to lock together faster. The vignette I chose for that one was Feynman, just because he spent so much time playing around with math and had so much familiarity that he was like an encyclopedia of weird math trivia. Random stuff comes up, and he can come up with the answer because he has all these patterns stored in memory.
--:-- That was a part of the book I was thinking about a lot. There was that part of intuition which explains that your memories in area one can influence and help your knowledge in area two. But then the chapter on directness talks about the failures of transfer learning. I would just try to put those two things together.
--:-- Dwarkesh Patel: Yeah.
--:-- Scott Young: If we're visualizing it a little bit, the principle of intuition is recognizing that because of this hierarchical structure of chunking, you have this building up of layers of abstraction on top of each other. What the thing on transfer is showing is that when you go from one domain to another, the problem with the transfer idea is that when we talk about learning skills, we tend to use fairly general labels. We say, "I'm going to get good at X," and X is just this broad category of skills. What's missing is that to actually perform those skills quite well, you have to do something very, very precise. It seems to be a general feature of the brain that it learns things quite specifically. That's how it works. It's not just a defect—that's why we're smart, because we can make very fine-grained discriminations between things.
--:-- You can get transfer if you're thinking about it in terms of: "I understand domain one well enough that I can see an abstract pattern here, and I understand domain two well enough that I can see an abstract pattern here, and I can see that these are the same abstract pattern and make that linkage." But the problem is that if you haven't chunked the first domain enough to get up to that pattern, and you're trying to talk about the next, they don't match because superficially they're quite different.
--:-- This is the issue: yes, if you understand physics problems well enough that you can see that conservation of energy is a principle that holds, and then you start learning a new domain and realize, "Oh, this is actually like conservation of energy," you get it up to that level of abstraction, and then yes, you can do that transfer. But the problem is that most people don't have this richly abstract, principle-based reasoning about things. They just see all these superficial details that have nothing to do with each other. If the superficial details of one don't have to do with the superficial details of another, there's zero transfer there.
--:-- One of the books I read that was a major source of the research on transfer was called *Transfer of Cognitive Skill*. I'm blanking on the author's name—I think it's Haskell, mentioned in *Ultralearning*. His point later in the book is that his idea about how we overcome this transfer problem is that we teach more theory. Ideally, if we have this richer, more abstract ideas about things, then we're getting it to a level where it can transfer above these superficialities. I'm a little bit more skeptical of that because I feel like that's kind of what universities do—teach theory—and it doesn't seem to work very well. So you're suggesting what we're already doing.
--:-- But the point I want to make in the directness chapter is: presumably you're reading this book and there's something you want to be good at. If you want to be good at it, make sure that those specific skills, the micro skills that need to be in place, are the ones you actually need in the real situation. If there's a mismatch, your performance is going to go down considerably. There are lots of real-world situations where just having a theoretical insight is not enough. You need to actually perform all of these small sub-skills. If you don't have the sub-skills, your performance is zero. If you're doing some training that doesn't work on the sub-skills you need, you're not actually going to be able to perform.
--:-- I'm talking about this very abstractly, but language learning is the example I use. If you only learn how to recognize sentences so you can't actually recall them, that's useless for you when you're speaking, because you're not recognizing sentences—you're actually speaking. I'm very critical of Duolingo there because a lot of the exercises they do are not recall. They're just doing multiple choice from a word bank. But actually speaking involves recall. It also involves pronunciation and working around words you don't understand. The more you analyze the skills you're trying to learn, you realize, "Oh, this is why this thing doesn't work, because the actual thing I need to do in the performance situation is not what I've been training."
--:-- Dwarkesh Patel: Hmm. This makes me wonder if this explains why very widely read people and very broadly knowledgeable people don't seem that much smarter than just generally widely read people. I had Tyler Cowen on the podcast and I asked him, did he learn more in the last 10 years than he did between the ages of say 15 and 25? And he said, obviously 15 and 25. And I asked him, if you take the concept of compounding growth of knowledge seriously, shouldn't you expect to be learning more in the last 10 years?
--:-- Scott Young: I don't think compounding works generally. Compounding is a very seductive idea where you just get more and more returns. But really, the areas where there is true nonstop exponential growth are vanishingly rare. When they apply, they totally transform the situation you're dealing with. Startup growth is an example of compounding return, where one guy in his basement can rule the world after 10 or 20 years of grinding. That situation is vanishingly rare. Most things have regions of exponential growth and then regions of diminishing returns.
--:-- Classic economic thinking is that diminishing returns tends to be the default way of viewing things. You get most of your growth in the beginning. There are some basic mental models or concepts which are very fundamental, and once you understand them, you get the 80/20. You get quite a bit of understanding, and then you get into the esoterica of academia where they're debating the finer points of some BS problem that doesn't really matter. Spending a lot of time studying that may be necessary to advance the field, but from a utilitarian standpoint of where you're going to get the most benefit from your learning, it was in that first phase.
--:-- Maybe that's the thesis of ultralearning too. You can capture 60% of the value or utility of a field in a relatively short period of time if you plan your learning out very effectively. I don't want to say that's always the case. For professional careers, you often need to be in the top 1% of a skill for it to matter at all. But definitely, if you wanted to become good enough as a programmer to do a lot of your own programming for personal tasks, but you're not a full-time programmer, you can easily do that in a year. The way people often think about it is, "I have to go to school for four years, and then maybe work in an office, so I have to do it for a decade to be good at it."
--:-- Similarly with language learning. If you want to be able to lecture in a language or speak so fluently that your entire life is lived in that language, it's going to take you probably a decade. But you get to a level where traveling in the country is pretty frictionless after maybe a couple months. That's the interesting zone. You could get decent at these skills in a relatively short period of time. How does that change your calculus about what kinds of skills you pursue or what kinds of things you invest in?
--:-- Dwarkesh Patel: Does this diminishing returns apply to consecutive ultralearning projects as well? I think you implied in the book that there's a cumulative advantage of doing one ultralearning project, and then you have the meta skills and confidence to go from one to another.
--:-- Scott Young: They're probably S-curves. One of the main points I try to make is that there is this kind of compounding confidence curve, especially at the beginning, which is where I'm focused. If you've never done aggressive self-directed learning projects before and you try to do this, and you're reasonably smart and have the stick-to-itiveness to actually get a project done, and you actually do it and follow through on it, then you can often get to this sort of, "Oh wow, I didn't know I could do that."
--:-- A lot of the people I interviewed who went through this, like Tristan Amonabella, are great examples. His outcome was pretty extreme. He was a reasonably competent, smart guy, but he'd never tried to do things this way before. Then he does it and gets this huge result, and he's like, "This is crazy. I could think of all these other types of projects I could tackle now if I was really serious about them."
--:-- There's this improvement not only in your confidence but in your overall strategy. If you've learned a couple languages, then learning more languages becomes a routine activity. It still takes the same amount of time to get to that level of mastery. But the way people feel and talk about, "Oh, I really want to learn Spanish, but I haven't made any progress in five years" — they don't deal with that. They say, "Okay, I'm going to work on Spanish, and I know I'll be able to get to a level where I'm having a conversation after a few months," just because that's how it works.
--:-- Now, if we're talking about people who become smarter and just become this eclipse level, like the Johnny Depp movie where he becomes the universe computer — obviously not. Once you've learned and mastered the edge of what we know about performance and memory, you're doing little tweaks. It's a little bit like athletics. If you've never gone running before and someone says, "There's this thing called jogging, and if you put your shoes on and run for a bit," you can probably get quite a bit better for a while. There's quite a bit of gains. But once you're at the level where you're regularly competing in marathons, you've run the Boston Marathon, now you're shaving seconds off your time. So there's probably an S-curve there.
--:-- The argument I'd like to make is that most people are before the gains. They're the people who have maybe jogged a couple times in their life but never taken it seriously. For the intended audience of this book, that's true. But whenever we talk about compound growth, we just have to keep in mind that unending compound growth is a very scary thing in the world. It implies that one person controls the entire universe or is smarter than every other person on earth. It has to diminish at some point.
--:-- Dwarkesh Patel: It's interesting. I had Robin Hanson on the podcast, and he wrote a blog post about the long view. The idea is that you can't just invest in the stock market and a thousand years down the line be the richest person in the world by a big factor. He asked, why is it not the case that some organization says, "It doesn't matter if we die, we're just going to put a couple million dollars in this fund and we're going to control the universe in a few thousand years"?
--:-- Scott Young: The classic reason for that is that it's one of those engineering fallacies. Obviously, once an organization started to have that kind of power, not only would you have corruption from the inside where the human agents controlling the agency would start diverting the resources and breaking the company's mission, but everyone from the outside would start to dominate. You don't just dominate the world without also having an army, wide popular appeal, and doing that kind of thing.
--:-- I could see some scenario where that happens, where you're creating an organization like a new church that becomes the mass religion and then controls the universe. But there's no secular, no political ambitions organization that manages to concentrate so much wealth for thousands of years and people are like, "Oh yeah, we'll just honor that contract."
--:-- No disrespect to Robin Hanson. I think he's a brilliant thinker. I really admire him because he points out these sorts of, "Why don't people just do this?" He has the right kind of mind to consider things that seem absurd to normal people. But that's the issue for something like that. I think it's probably true for all kinds of compound growth. At some point, it has to stop. You can't just go on forever. But if there's some meaningful range where you get compound growth, even if it's a short range, it can still be super important to pay attention to.
--:-- Dwarkesh Patel: Okay, let me intersperse some practical questions here. Following your advice, I've been on my own ultralearning projects. When I'm learning, I sometimes have a lot of questions about how this works on a deep level. Is there value in first getting the whole context, kind of building a rough map, or should you, as you're learning, question each piece of knowledge as you're accumulating it?
--:-- Scott Young: I think the way I view projects and learning projects in general is that they all have to start with, "What does success look like? What is the outcome I'm trying to generate from this learning project?" Analyzing things abstractly is hard because if you're optimizing for a different goal, the answer might be different.
--:-- Even taking language learning as an example, for the goals I had, trying to figure out the etymology of words or spending a lot of time figuring out where words come from quickly hits a point where you're wasting a lot of time. What I actually need to do is just memorize more words. I don't need to dive deep on that.
--:-- That being said, that's not a general principle. There are other subjects where the tendency is to not do enough depth, to not really go into it and understand things. In my mind, that's just because the specific constraints of learning a language suggest that's the optimal strategy. Whereas learning physics, for instance, the problem is that most people don't get nearly enough depth. They understand the concepts too superficially. They're seeing the pulleys, they're not seeing the conservation of energy.
--:-- When you're asking about the right way to approach a subject, it's really hard to do that in a goal-free setting. It's hard to say, "I want to learn physics the best way possible." I don't think there's really an answer to that. There's an answer to, "I want to be able to pass these types of exams in my university class the best way possible," or "I want to create groundbreaking physics research," or "I want to be able to discuss physics intelligently with other physicists." All these different goals suggest different ways of building around this.
--:-- There are more robust strategies. A strategy that works better if you have multiple goals. If you not only want to pass your class but also want to be able to invent things, you have a wider set of goals. I do think there are strategies that are more robust.
--:-- If I'm learning Chinese and my only goal is to look really good for 10 minutes of an audio thing I can prepare, then obviously the strategy is to script it and rigorously practice. It's to actually not learn Chinese at all. But that's not a very robust plan. I would never do a project that way. That's not really what you want to do, even if that was one of the outcomes you wanted.
--:-- If you're learning Chinese and your goal is to talk to people versus read Confucius, you're talking about completely different strategies. Even if there's some overlap, the thing you do to read ancient Chinese texts and the thing you do to chat with people on WeChat are totally different goals. One of my main points is that you can't even begin to think about optimizing unless you say, "Optimizing for what?" Learners tend to discount that because they're like, "I want to know X" or "I want to learn this topic." Sometimes those assumptions are the problem.
--:-- Dwarkesh Patel: Out of curiosity, take the example you gave of trying to be a groundbreaking researcher. Say you're Cal Newport and you're trying to make a new proof. What is a strategy you use to learn at the cutting edge?
--:-- Scott Young: One of the key things would be to specialize. You see this at top level researchers. They specialize because it's much easier. Your benefit of getting a breakthrough is going to come from ideas that are closely related to what you're doing, and you need to have them at an extremely high level of fluency. There's a drop-off curve of the value of ideas the further away they go. If you have a really high level of specialization in a field, you're probably at the cutting edge for techniques and stuff, and you can find problems to work on.
--:-- That's a little bit different from, "I want to be an Albert Einstein." That is part of it, but I think it's just luck. You have to be in a fertile territory. But you're also looking for something competitively unique because everyone else is specializing.
--:-- This is another consideration. Everyone else in the world is also pursuing some sort of strategy. Sometimes doing something that's objectively suboptimal but nobody does gives you an advantage. My own life is an example. Obviously, it's better if you can actually go to MIT and get an MIT degree and the alumni network. But the fact that nobody was doing the MIT Challenge gave me a claim to fame. If only one person is doing it, that makes me more unique. The value to my career has probably been higher than getting an MIT degree. But if every single person was doing the MIT Challenge, hundreds of thousands of people, then maybe the difference changes.
--:-- There is a sort of strategic role in how you learn things. One of the things I talk about in the book is, given the goal I'm optimizing for, how do successful people do it? Go to academia and say, "How do I make things that inch forward the field?" You have to specialize. You have to figure out what problem you will understand the best in the world, and then find ways to make improvements over it. That's how our whole academic system is organized. There's no surprise there.
--:-- That may not be what we want as a society. Maybe we want these weird cross-pollinating insights that no one saw coming out of left field. But as a pure strategy on your own, you're definitely going to get into situations where you learn two things and there's no obvious overlap, and you've mastered them and it's been kind of useless. It's hard to pick those two things. It's hard to figure out what insights, when you breed these two ideas together, give you a super idea and not just some garbage idea that's actually much worse than what people are currently doing.
--:-- Dwarkesh Patel: The MIT Challenge, is that an example of the failed simulation effect? I'll let you explain what that is. Is it that because it's a rare thing to do, it just seems much more impressive? It's hard to simulate in your mind.
--:-- Scott Young: Yeah, the failed simulation effect comes from a good friend, Cal Newport. It's from his book *How to Become a High School Superstar*. The idea is that the impressiveness we assign to things is not based on how much work that thing requires, but on how hard it is for us to imagine doing it.
--:-- His example is that if you're in 14 different clubs, president of the debate team, on mathletes, took a bunch of AP courses — that's objectively a ton of work. To be the high school valedictorian and do those things is a lot of work. But we can imagine ourselves just being a grind doing all of that. Whereas when we see someone like a mutual friend of ours, Manish Sethi, who published a book for programming for teens through an actual publisher that sold fairly well when he was in high school — that's the kind of thing I can't imagine doing now. How does an 11th grade kid do that? That makes him seem much more impressive, even if the amount of work is strictly less than what it takes for someone else to do.
--:-- I think the MIT Challenge is a little bit of that example. The reason people found it impressive is that when they imagine doing it, they can't imagine doing it themselves. Maybe I got the benefit of that.
--:-- There's a technique Cal talks about where you make these inroads into a field and then find new opportunities. There's all this serendipity that pushes you forward to an outcome people wouldn't have expected. I'm not sure whether that's true with the MIT Challenge. I can explain how I did it, and it's not like someone else couldn't attempt it. There wasn't a lot of, "I just happened to find the right guy at MIT who gave me access." I just did it.
--:-- But in terms of why people find it interesting, certainly it's because of that. I also benefit from the fact that if I had done just some middle-of-the-road university's computer science curriculum, people would be ho-hum about it. Most of the work is just doing the curriculum. It's not the fact that it's MIT. I only picked MIT because they post their material online for free. If the University of Wisconsin's computer science program uploaded their material, I could have used theirs too, and people would have been less impressed.
--:-- I'm really kind of sneakily leveraging the fact that people associate MIT with being super smart. That's because of how strange their application process is, and that's the very thing I didn't do when I did this project. Looking back on it, I'm definitely proud of the project, but it's kind of funny that that's the thing people fixate on. I also really like the language learning project, but the MIT Challenge is the one that captivates people for some reason.
--:-- Dwarkesh Patel: I'm studying computer science at UT, and our curriculum has classes where the programming assignments are hard to code up, so they literally just copy them off of the Stanford programming assignments. If you're the kind of student who wants to copy answers off the internet, you'd just look up Stanford CS whatever and find it there.
--:-- Scott Young: That's another thing that came in my favor. MIT's computer science program is very math and theory based, which means it's intellectually more difficult. It's harder to understand the MIT curriculum. When they're doing analyses, they just assume everyone knows calculus. In other universities, sometimes they'll go easy on you. They won't make you do the calculus. They make you do the calculus in every MIT class.
--:-- You take the intro microeconomics class. When I did it in my school, they never made you do the calculus. They were always like, "Find the intersection on this graph." Whereas at MIT, it's like, "Now we take this integral and find the Laplacian of it." They just do that because they assume everyone has a very strong math background.
--:-- However, the way that works to your advantage is that in a lot of other universities' CS programs, they have these really long, tedious programming assignments. MIT hits the sweet spot of being difficult but not a long amount of work. When I did the project initially, the idea was just to do the final exams. Nothing else. That was my justification for whether you could do it in a year. You have to learn it and do a hard final exam, but you don't have to do all the busy work you get in school. You don't have to do every single problem set, essay, or group project.
--:-- I was getting some flack in the beginning, like the first week or two, about why not the programming assignments. It turns out you can add the programming assignments to that challenge and it doesn't materially change the amount of work. I probably only added an extra week or two to my schedule to also do the programming assignments, just because the programming assignments at MIT are very tight. People were questioning me because they went to different CS programs and they'd say, "I spent months working on this program, there's no way you could do it." It's like, that's not how MIT does their course. It's all proof by induction, drawing graphs on pencil and paper.
--:-- Dwarkesh Patel: From a practical point of view, do you want to do the heavy programming assignments to become a real programmer?
--:-- Scott Young: Yeah, probably. But I think if your goal was, could I get the kind of high theory, sort of conceptual education of MIT, then I think it works. And to be honest, because I'm not a practicing programmer, I just program for fun. The conceptual stuff is really where all the value was for me. Like knowing how to design websites is not super valuable for me, but understanding how information works in general actually translates really well when you're understanding cognitive science and stuff. So I'm actually glad that I did it that way. But my preferences are probably different from people who study CS so they can get a computer science job and be a programmer. That was just a different situation for me.
--:-- Dwarkesh Patel: Can you explain how the knowledge about computation, how to encode and transmit computational information, how does that relate to the research you do or the research you read in cognitive science?
--:-- Scott Young: Cognitive science explicitly overlaps with computer science. Cognitive science as a discipline is usually philosophy, psychology, neuroscience, and computer science, right? Artificial intelligence and stuff. When you're reading papers that talk about a computational model for how chunking works, if I didn't have a background in understanding computer science, this would be a very difficult paper to read. He's talking about how you make these nodes and then you make these links and that's how you represent a data structure of conceptual information. If you've only done psych classes, this is a hard paper to understand. But if you spent classes doing algorithms where you're trying to design systems that do just that, then it makes sense to you. It's like a natural language. You're like, "Oh yeah, that's a linked list," or "Oh yeah, that's a tree structure," or "Oh yeah, that's a graph."
--:-- Similarly, I was just doing a lot of research on motivation, and there's all this stuff in neuroscience about the motor loop, which is this basic circuit motif that runs through the basal ganglia. It's how your brain decides to do one and only one thing when you've got a billion neurons firing all at the same time. It's this pattern that basically just don't lock everything except the one thing that you want to do. This is a kind of neuroscience idea, but you can really visualize it in terms of, "Oh, if you had to design a circuit to do this, this is very similar to electrical engineering." It's like, we need to amplify one pattern and run it through a filter, and there's this center surround pattern. There's lots of stuff like that. But to me, that's not the "how do I program stuff in JavaScript." That is understanding this abstract theory level, which we were talking about low ability for transfer, but it's more that you have to get these abstract ideas in both domains for it to work. Most people don't do that.
--:-- Dwarkesh Patel: Right, right. It also sounds like OS and time scheduling.
--:-- Dwarkesh Patel: Going back to transfer, because you did mention that, it does seem that your career in terms of the learning you're doing, you're doing it explicitly because you see a high potential transfer between the research you're now doing and the computer science you did before. Why are you pessimistic about transfer in general when it's...
--:-- Scott Young: Well, we have tons of studies showing that people who attend classes don't transfer a lot of what they learn to the outside world. It goes back to my Einstein point. If you see data which says that X doesn't work for almost all people, you should be skeptical of that as a general method for being successful at X. It raises the question of whether or not this is successful. It's not just that this only impacts dull, uninspired students. One of the studies I quoted was that people who were honors level physics students could not solve problems that differed superficially from the ones they had studied. This specificity with the brain comes down to the idea that yes, there is this ability for transfer, but that's the magical bonus that happens sometimes when you're lucky. The base reality is that when you learn most skills, they adhere quite specifically to the context you learn them in.
--:-- If you are budgeting for a learning project where you're trying to accomplish a particular goal and your flowchart is "do this thing that's unrelated to what I want to be good at, then the magic transfer happens, and at the end I'm able to perform this skill," that's a really bad flowchart. What you should do is be like, "Okay, how do I line these up?" I think learning broadly, learning for the ability to have these mental models or concepts or deeper understandings of things, I'm a big supporter of that. I do a lot of that. I learn lots of different subjects on the off chance that I'll be able to get one of those deep insights. But if we're learning for a concrete purpose, it's really useful to think of it in that way. What is it that allows me to perform in the way that I want to be able to perform? As soon as I'm learning for a task, I adjust how I'm approaching things for that task. If I'm doing a research paper, I know what I'm trying to get out of things and how I'm trying to read stuff. I'm not doing something that's totally unrelated.
--:-- The directness idea goes to the fact that yes, it is good to learn lots of things. If you do learn lots of things, just through sheer volume, you're going to get some of these nice transfer effects. I'm not even really against learning broadly. A close reading of Ultralearning could be seen as an anti-David Epstein "Range" kind of thesis, but that's not really my point. My point isn't that you should only learn narrow, specialized things. Simply that once you decide on what a goal is, it always makes more sense to approach it directly. Having this broad base to build off of is also good in general, but that's a little bit different from "I'm trying to accomplish X, so do the thing that leads to X" as opposed to "I don't want to do that, that's too hard, I'm going to do this other thing." That's nonsense.
--:-- Dwarkesh Patel: Two questions. First, when you were doing the MIT Challenge, did you have the goal of having these transfer effects, or were you just trying to learn about programming and computer science directly?
--:-- Scott Young: A lot of the assumptions I went into the MIT Challenge with did not bear out. There's a weird bittersweet quality to that. The thing people fixate on, that I did it, is the part that I think was pretty successful. You could look through my exams and deem that my scores weren't high enough for your standards, but in the general sense, I think I accomplished that. But my idea going into the project was that I just finished university. School had been my life. When you're in university, all you know is school. It seemed like this was a huge opportunity. Why aren't more people doing this? This was pre-reading about all the signaling stories and my general level of cynicism about university. I was just like, instead of spending hundreds of thousands of dollars, you can do it for free. They're just giving this away, and you could get this knowledge and learn it.
--:-- I was interested in learning computer science. I had already learned some programming before, but I didn't know academic CS. I was thinking about going back to school to study it, going back as a mature student and doing another undergrad in computer science. This just seemed like the perfect opportunity. Because I had a blog and was commenting on students and educational issues, it felt like, "Is this just me? Am I crazy? Why has no one done this before? If I'm going to be the first person doing this, presumably there's going to be millions to follow." It seemed very exciting. This was going to be the road to how everyone's going to do education in the future.
--:-- Now, nothing has changed. I know of very few people who have done anything like the MIT Challenge to that scope. I know a handful of people have done their own similar projects, and I'm sure there are people I don't know of, but certainly not the massive waves I had predicted. In that sense, it didn't quite work out how I wanted it to. But at the object level, not my implications of the project, but what the project actually was, I think it was pretty successful. I did learn a lot of programming. I did do programming after. But if my goal had been to get the entry level computer programming job that would be the path to a career there, I would have done it a little differently. The MIT education primes you more for working on harder programming problems, not the routine stuff most programmers work on. It would have been a good jumping off point if I wanted to get into certain branches of programming where it's complicated and the math is hard. I think it would have not been a bad way to get a foothold into a job the same way that getting an actual CS degree is a good foothold into a job, even though the skills are not exactly the same.
--:-- One of the benefits of having done it is that after you see the mental models and the ideas, you have these abstract patterns that you can apply. But economics also has a lot of those patterns, and I didn't do a degree in that, but I feel like I get a lot of value out of those too.
--:-- Dwarkesh Patel: When you were learning, was there a specific way you geared your education to maximize the potential of transfer learning?
--:-- Scott Young: The more you understand things, the better you get at transfer. This goes back to the intuition chunking idea. The way transfer works—and I'm using a model, I'm not saying this is a scientifically proven thing, this is just how I visualize it abstractly based on what I've read—the idea is if you think of this hierarchical model where at the very low level there are superficial features of the things you're encountering, and then there are increasingly abstract features about the problem. Chunking is finding more and more abstract patterns. You're like, "Oh, this problem of this type is an example of that," which isn't obvious unless you have all these base patterns to notice what's going on. Understanding is this process of building hierarchies of information that go deeper and deeper. Those tend to be the things that transfer better because the more you understand a quite abstract feature of a problem, the more you can spot it in other domains that you've understood enough to see that pattern.
--:-- If I were to contrast a different way of learning that was just focused on high exam performance, just memorizing the most common solution types and how to do them without understanding why it works, I think in MIT classes that's probably a bad strategy because there's too much diversity of problems. In some schools, they do actually teach that way because the students can't perform novel problems on exam solution sets. MIT did use some problems that were seen in a previous assignment set, so if I had done that problem I had a leg up, but most of the exams give you novel problems, so you have to have a minimum level of understanding.
--:-- My general approach to learning is to focus on understanding over memorization. There are some exceptions. In language learning, for instance, I focus on performance over understanding. But generally, if you're focusing on understanding, that's probably going to lead to better transfer than if you're focusing on memorization. That had long been a bias in how I approached learning things. There is probably a case to be made that if you were strictly concerned with performance on an immediately subsequent exam, you could sometimes get better performance with just memorizing. But even with a long curriculum, they assume you understand the Fourier transform in later classes. If you just memorize the equation and don't understand what it does, you get to something later and it doesn't make any sense to you. You have to go back down and understand it. The more you really understand ideas and the principles behind them, if you have that as an aim, then you're learning for better transfer. It happens to be the optimal approach for doing complex exams like MIT, but certainly not for all schools. I generally recommend it anyway because what's the point of just getting a good mark on an exam if you don't know anything useful for anything else?
--:-- Dwarkesh Patel: Let me ask you about focus. You say in the book that you're agnostic about whether focus can be trained as an ability in general. Unless I'm misunderstanding this out of context, this seems to contradict Cal Newport's book on Deep Work. Am I missing something?
--:-- Scott Young: Cal and I are good friends and we co-instruct a course on focus. There's a certain sense where we are in agreement on the trainability of focus. Where I might disagree with Cal somewhat is that I think Cal views it from a capacity point of view. He's using the muscle metaphor for focus. All the research I've done on transfer really suggests that's not the right way of thinking about the brain. For instance, doing brain training games that improve working memory don't help your working memory for things that aren't brain training games. If you did a muscle analogy and said, "I'm going to improve my working memory," you're kidding yourself because that's not how it works. You can improve your working memory for probably pretty specific things, like chunking is an example, or by developing quite specific strategies.
--:-- When we're talking about focus, what kind of thing are we talking about? If we're talking about the capacity for you to focus as a kind of cognitive ability, I think it's doubtful that you could robustly improve that particular faculty just by doing specific training. You could probably improve it in specific ways, but they're going to be specific. You could improve your ability to focus on certain types of problems because you've built up specific cognitive strategies for doing that.
--:-- That being said, why even talk about focus at all if I don't think it can be improved? Because when we talk about focus casually, we're not just talking about a cognitive faculty. We're also talking about all the habits, routines, strategies, and affective processes that influence focus. The reason most people can't focus—is it because their brain is incapable of focusing? It's because their phone's buzzing, or because they're like, "Oh, this is boring, and I'm getting distracted, and I don't enjoy what I'm learning, or I don't find it interesting." People lump a lot of things together.
--:-- If I were writing a scientific paper and making a bet on whether focus training programs improve the ability to focus in the way cognitive psychologists typically measure, I'd be a little bit pessimistic because they're probably measuring in a tightly controlled experimental setting. If I were talking about whether there are ways to robustly reduce procrastination, reduce distraction, reduce the ability to get frustrated and give up or waste your time on something else, then yeah, I think that's totally trainable. I think that's something you probably could improve on. Cal Newport tends to take more of a facultative approach with some of these things, and I'm more skeptical of that. But for the average person, the message is you ought to focus more. There are probably aspects that a normal person ascribes to being the ability to focus that you could improve.
--:-- Dwarkesh Patel: This is very interesting. But when Cal Newport says in Deep Work that the other things you do in your day after you're done with work, your phone, your TV, that those diminish your faculty to focus on your actual work, is this model wrong? Am I free to do whatever I want?
--:-- Scott Young: The way it's stated is probably too crude. As a practical consequence, maybe. I don't know the research so much on that. I think it is potentially possible to have tons and tons of focus in your working hours and then just be in this buzz on Twitter all day. I'm a little bit less pessimistic than Cal that it's going to destroy your brain in some way. I don't know that that's true. What I will say is that from a motivational perspective, Twitter is the variable reinforcement schedule. It's the Skinner box that gives you rewards intermittently, so you're willing to press it constantly and perpetually. If you're choosing between Twitter and reading this hard book, the motivational gradient is going to be to go on Twitter all the time. When we're talking about focus there, is that a cognitive capacity, or is that just your affective ability to choose this harder thing?
--:-- If Cal's talking about focus in that way, which I think he is, then I kind of agree with him. If you're not on Twitter, if you don't have these variable ratio reinforcement schedules that are constantly like, "Ooh, it would be more fun to be on YouTube right now," or "Ooh, it would be more fun to check my phone right now," if you don't have those things constantly as active primed habits in your behavioral repertoire, it is easier to focus on reading a hard book. I think that's where you could get better at reading a book.
--:-- As for the ability to like—for instance, I'm skeptical of people who say, "I meditate for an hour a day, and therefore I'm going to have enhanced focus capacity." To me, that sounds like the brain training thing, which we know to be false. But if we're talking about removing a lot of these distracting options from your list of habitual responses, will that allow you to sustain endurance and persistence on harder activities? I think that's probably true.
--:-- Dwarkesh Patel: A way to think about the analogy here is that you have certain forms of exercise like cross training that don't seem to be highly correlated with other physical activities or performance, but then you have stuff like weightlifting which seems to be strongly correlated with your ability to play football or soccer. Is meditation, if there's a distinction there—I don't know if there is—but if there is one, can an activity like meditation be in the productive sector?
--:-- Scott Young: I think meditation can be good, but it's important to know what meditation is for. I feel like a lot of the research on meditation is quite poor. I say this as a non-researcher, so I'm certainly going to get in trouble for it. The problem with meditation is that the people who like meditation—and I've been on multiple meditation retreats, I've done meditation daily for periods of months, so I'm not just some guy who didn't get it, I understand why people meditate—but the problem is that it is very tied in with essentially religion, Buddhism. I don't like attacking anyone's spiritual beliefs. There is a really strong philosophical component to meditation and what the right way to live life is. I don't want to say, "Oh, you shouldn't meditate," or "People who meditate, that's bad." It's more that sometimes you need to question the very specific claims as stated. One of the specific claims is that because I meditate for an hour every day, I'm going to be much more focused. I don't know whether that's true. If it is, it's definitely not the mental model I have of how the mind works.
--:-- When we're talking about whether Twitter destroys your ability to concentrate, I think the mechanism for a statement like that to potentially be true is that Twitter is super appealing. It is super enticing. If it is one of the default habitual responses you have at any moment, you have to put in a lot of energy resisting it. In an era before television, people could regularly sit and listen to long radio programs. As a regular thing, sit around the table and listen to radio. We don't do that anymore. Why? I think probably because visual media is a little bit more compelling. When novels were the only form of entertainment, people would binge read novels who maybe now would be binge watching reality TV or on Facebook. As technology has developed, we've developed increasingly enticing options. The more you engage in those habitually, the harder it is to reduce that impulse. If you eat junk food all the time, it's going to be hard in a particular moment to be like, "I'm going to eat this kale salad rather than a hamburger," because your context is like, "Oh, you know what would be better right now?"
--:-- Scott Young: The hamburger. Whereas if you never eat junk food or you only eat it in very limited contexts, it's just like, well, I just eat kale salads for lunch. It just makes sense for you to do that. It doesn't require willpower. So I think of it more in this way. But Cal has his own mental model of how this works, and I'm not really a debating type. If he disagrees, he disagrees with me. This is just my layman's interpretation of what I've read.
--:-- Dwarkesh Patel: Can we go back to the MIT Challenge? I remember that article you wrote about why people are getting so much better at Tetris. You talked about how somebody did the four-minute mile, and now hundreds of people have beaten it, even in high school. Why hasn't that happened with the MIT Challenge? They see somebody can get a STEM degree at MIT online, and now everybody's like, "I can do it in six months."
--:-- Scott Young: I have some regrets about that article. The regret is that I should have just focused on one explainer, but I threw in the banister thing. Now I regret it because someone pointed out to me that the banister thing apparently wasn't true. If you look at the trend lines for performance, it wasn't the case that they needed an example. It was just that we were generally innovating in mile-long running, and it was generally going down like this. I don't know the full argument, but they were refuting the idea that just needing an example to do something is enough.
--:-- There seems to be something to it, at the very least, that a lot of innovation is the difficulty of coming up with an idea. But once you're aware an idea exists, it's like, "Oh, that's one of the things I could do." But I don't know if it's the case, as I was perhaps mistakenly claiming, that a known barrier—just knowing it has been surmounted—allows people to perform better, even without knowing how they did it.
--:-- What I wish I had focused on in that article, and only made it about, is the speedrunning community. I'm not super into it—I don't speedrun and I don't even really play video games—but speedrunning as a cultural phenomenon is fascinating to me. It's extra fascinating because it's seen as low status, like a bunch of guys in their basement on Twitch playing games. But I think people should be studying this. There should be academic papers written about it.
--:-- The Tetris example I gave: Tetris was really popular in the late 80s, early 90s. The whole world was playing it, and people were bad at it. Now there's a much smaller group of people playing who are insane. They're playing at a level where the game glitches out so you can't even see the blocks falling, and they're beating those levels. It's crazy.
--:-- I think this shows how innovation happens, how we make learning progress in society, through these network structures. A few key innovations unlocked an explosion in performance growth for a field. Originally, speedrunning worked through magazines or old-school 1990s websites where someone would verify a time and publish it. The problem is you can't learn from that.
--:-- What changed the entire domain was that the proof for a particular speedrun time was to post livestream footage of you doing it. This was a game changer because now anyone who wants to see what you did can watch it. It's like the patent system: to get the patent, you have to show how it works, allowing people to innovate on it.
--:-- I'm very interested in this sociological aspect of learning. I'm a big fan of progress studies—the idea that we are not innovating at the rate we should be to solve big problems like global warming, hunger, poverty, disease, and pandemics. We ought to be innovating a lot more. This wasn't something I talked about in *Ultralearning*, which was all about psychology and cognitive science. But the sociological aspects of innovation and learning are really fascinating to me.
--:-- Another book I recommend is David Wooten's *The Invention of Science*. He's a historian who goes through the scientific revolution. He points out that there were cultural shifts triggered by very particular events that created the right mechanism to create science. It's weird to think of science as an invention because it's obvious to us now, but he rigorously shows how European languages didn't have a word for "discovery" until Christopher Columbus discovered the New World. Culturally, it was assumed the ancients knew everything.
--:-- The printing press changed the culture from a manuscript culture to a book culture. Being a book culture created facts because what was written in a book would be fixed and stay for all time. You could point to so-and-so writing this in a particular book. The ancients were often dubious of experimental evidence because things change. You need an indelible record of what someone said to even make that process possible.
--:-- I think speedrunning is a culturally underrated force. It's a real innovation that to claim credit for a particular accomplishment, you must post a video record of you doing it. That's not true for most fields. Why don't we have that in a lot of fields? Even within a company like Google, why aren't the best programmers competing to be the best by constantly livestreaming what they're programming?
--:-- Dwarkesh Patel: Yeah.
--:-- Scott Young: There are certain factors about speedruns that make them amenable to this, so maybe there are additional innovations I'm missing. I'm sure people at Google have thought about this. But my feeling is that this sociological pattern of how we gain knowledge as a society is fascinating, and we haven't even scratched the surface of all the ways technology could assist that. That's a different problem from disseminating the college curriculum to a larger group of people. That's a totally different problem from how we actually innovate and improve things in society.
--:-- Dwarkesh Patel: That's fascinating. By the way, you get a 10 out of 10 on an interesting tangent there, going from speedrunning to the Enlightenment. Have you heard of George Hotz?
--:-- Scott Young: No, maybe you can tell me.
--:-- Dwarkesh Patel: He's a guy who does exactly this. He was the first kid, at 17, to jailbreak an iPhone, first person to jailbreak a PS2. Now he does livestreams on YouTube all the time. He's folding COVID proteins, doing LeetCode problems. There's a nascent community of Twitch streamers and livestreamers for certain things.
--:-- Scott Young: Part of me suspects there are additional ingredients necessary to make this apply beyond normal problems. I even did a livestream project for quantum mechanics because I was so excited about this. Admittedly, I was a little lukewarm on the result. It would have been nice if livestreaming had been a thing when I was doing the MIT Challenge. But the major problem is that it's very difficult to parse what's happening. A livestream for a speedrun could be four minutes, so everyone can easily watch and study it. If I've got 400 hours of me doing a task, it's impossible for you to watch and easily learn from it.
--:-- There are probably additional things necessary to make that broadly applicable. Maybe livestreaming alone is not enough. That's maybe why we haven't yet seen an explosion—there are one or two more innovations we need to lock that in. But I'm bullish on the fact that we don't really understand innovation at a sociological level. This is an area where if we had a better theory of innovation and how we could do that culturally and socially, rather than just explaining how we've innovated up to this point, we could find additional social innovations that would themselves prime more innovations. I think there are lots of opportunities for technology that we haven't scratched yet.
--:-- Dwarkesh Patel: It's interesting how peer groups play into this. This podcast is named the Lunar Society after a group in England during the Industrial Revolution that was fantastically productive, including Charles Darwin's grandfather, James Watt, and a bunch of important industrialists and scientists. It's an interesting question why these kinds of congregations that defined very productive cultures and time periods, like in Florence, like in England, haven't been happening across the internet where people are connected all the time. Where are the modern Michelangelos and other artists meeting up on the internet?
--:-- Scott Young: I don't know. I've become persuaded by the thesis that progress and innovation have stalled in the world. The first half of the 20th century had more innovation than the second half. We're not even halfway through the first half of the 21st century. One argument is that we see all this innovation in computers, but in the first half of the 20th century or the second half of the 19th, there was huge innovation everywhere, in every single aspect of life. Innovations have been confined to a single industry.
--:-- Even startups are kind of like, "Let's apply the internet business model to industry X." It's hard to say that Airbnb and Uber represent separate innovations. They're both like, "What if we take the hotel industry and the taxi cab industry and run them like software industries?" It turns out that works. It's hard to think about computation and how many of those are discrete innovations.
--:-- There are arguments that we have a different culture than before, that there's been an increase in barriers to innovation. It could be related to low-hanging fruit. Advances in physics in Einstein's day could happen by a guy in a patent office. Now, figuring out the Higgs boson required billions of dollars at CERN to build the largest machine on Earth. We're a long way from when he figured out Brownian motion on the back of an envelope, just by looking at dust in a water glass. We may have found a lot of the easy innovations.
--:-- Or it's possible there are unexplored spaces, but we haven't found the first frontier innovation there. Alan Turing writing the first paper on the computer opened up an entire explosion of fields, but it hinges on someone opening that crack, and then you have an explosion of growth leading off that idea. I don't know the reason why things have seemed to slow. But if we had a better understanding of the principles of innovation and how it works sociologically—and I say this not being an expert on that literature—it could lead to more experimentation in how we think about group structures, networks, prizes, incentives, and cultures of innovation.
--:-- Innovation is the most positive-sum game there is. If some part of the world became a new Silicon Valley for something new, that would be very, very good for the world. And yet, what are the things preventing that?
--:-- Dwarkesh Patel: For the people listening, I recommend my podcast with Jason Crawford about the roots of progress, and also with Murray Rice and Caleb Watney. Jason Crawford is a great guy. I had him on the podcast a little while ago.
--:-- Let's talk about Paul Graham's new essay that just released yesterday called "Early Work." He talks about how the products of your early work on a project aren't going to look that impressive, and he gives various methods for getting over that hump and ignoring how unimpressive it seems. What do you think of that?
--:-- Scott Young: I think it's true. I've written about it before, perhaps less eloquently than Paul Graham or with less authority. I think one of the things—and I feel this myself—is that there's an opportunity cost problem. This might itself be one of the things that deters innovation. Once you get good at something and it starts producing rewards, it's very difficult to climb down that gradient and do stuff that's quite unrewarding.
--:-- When you're a kid, you're bad at everything. Kids are objectively terrible at all things. I say this having a son, so apologies if he listens to this one day, but they are not good at anything. We praise them for doing stuff, for trying things, for exploring their talents. Kids try and do lots of things. It's normal as a kid to do art, every single kind of sport, all these things. We don't say, "You're not good at this, you better stop now." But as adults, we have this threshold of, "You need to be at least this good to even be doing this."
--:-- I think that prevents us from doing a lot of things. Some of that is probably wise. If you spent all your life doing things you're not talented for, you won't spend time doing the things that are good. In a limited lifespan, there probably is an exploration training phase, and then you work on the best opportunity for 15 years to make your mark. I think there's merit in that. Robin Hanson made the point that your 20s and maybe your 30s are for training, your 40s and early 50s are for making your mark, and then you're done. You have to think about your career life cycle in that way.
--:-- Even for me, I feel like I'm transitioning more into the exploit versus explore mode of my career. When I was 23 doing the MIT Challenge, I was exploring this huge, vast terrain of areas. I started at the university, learned languages, did all sorts of different things. Now I feel like, "These are my strengths and weaknesses, what I've done well. Where can I really apply that to make an impact?" So there is some benefit to that.
--:-- But part of the challenge is that people are way too unambitious in general. Not in the "I want to be better than other people" way, but they just don't think of big projects. They don't work on them. They don't have big dreams to do cool things. If they do, it usually boils down to social status: "I want to be the person that does this that's better than other people." I don't know how you change that. But rewarding a culture where you want to do ambitious, original things that are interesting and you don't know where they're going to lead—having that in you is kind of rare. Cultivating it is probably good for yourself and society.
--:-- Dwarkesh Patel: You inadvertently answered my next question, which was what kind of advice would you give to a 20-year-old? You said we're talking about being more ambitious. Is there anything else you would say?
--:-- Scott Young: People always ask me what I would give myself as advice for a 20-year-old. It sounds really arrogant to say this, but when I go back, there's not actually a lot that I would change. Part of it is that I actually have respect for my younger self. As you get older, you get more complacent and soft in ways. There were things I did where I was really ballsy that I don't know if I would have the guts to do now, which is a weird way of thinking about it because we often think in terms of strictly increasing capacities.
--:-- Doing things under uncertainty is different. It's easy to look back with hindsight and say, "I wouldn't have worried so much about X," but maybe because you were worried about it, it actually happened. So I tend to avoid that.
--:-- If I were to think generally, my first thing would be that when I was younger, I had a stronger compulsion to monetize—or prove things would be successful. The MIT Challenge and the year with that English were a bit of a dip in that. They only happened because I'd already gotten my blogging business underway. But I recognize now that if you have that kind of personality where you have prospects, you're a little bit clever, you're willing to work on interesting, innovative projects, just working on really interesting, innovative projects and pouring your whole heart into them is probably what you should be doing.
--:-- Instead, you see people sub-optimally trying to optimize for things like, "Here's how I can make a little bit of money doing X." It's like, no, you're working on the coolest thing, and now you want to find some way to do something else.
--:-- I'll give an example. I know someone who was planning on going into medicine. Very smart. But then she gets a waitressing job and is earning good tips, good money for her at that time. That steals away energy from studying. She doesn't go to medical school, and that path closes. To me, optimizing for waitressing salaries is totally the wrong thing to do at that point in your life, especially if you're bright and have ambition. You should be all in on the medicine path because it has way higher value in the long run.
--:-- When you're 20 and your family's not super rich, and you see yourself making a lot of money and think, "I could have a house at 23," it's very tempting to go down that route because you have limited experience. What I would change about my own perspective is that if you have some ability, and you know this about yourself—if you're listening to this, you know you're reasonably clever and hardworking, you're listening to podcasts like this, you have ideas—then you should be trying to earn enough money to live. You can't always have people prop up your ambitions. It makes sense to do things with some reward potential because doing things that are totally crazy with no basis in the real world might be a waste of time. But prematurely trying to take the value right now out of an opportunity seems to be something super talented people do mistakenly.
--:-- Dwarkesh Patel: That's very good advice. Just keep going down doing kind of crazy, interesting projects that build your skill. You're basically increasing the quality of problems you can work on in your 30s and 40s. Working on really high-quality problems when you're 35 or 45 is where you want to be, rather than finding a way to make a little bit of money when you're 22.
--:-- Scott Young: It's hard to do. I also didn't come from super rich parents. I paid my way through university and scraped all the money when I was doing the MIT Challenge. But there is a difference between "I am paying the bills" versus "This is an okay immediate opportunity to make a bit of money or get a bit of status, and I'm going to run for that" rather than investing in longer-term ambitions.
--:-- Dwarkesh Patel: When I ask for advice for 20-year-olds, I'm always doing it under the pretense of asking for advice for me. But that's my advice for you then: work on stuff so that...
--:-- Scott Young: I think there's a bit of status anxiety when you're young. You want to prove yourself, you want people to respect you, you want to earn that social credit right now. It depends on your reference group. If you have a high-quality reference group, they'll push you in ambitious directions.
--:-- I grew up in a small town. Small towns are notorious for ambitious people leaving. There was very much a "crab pulling the other crabs down out of the pot" situation in high school. Then I went to a sort of middling university where again, there weren't a lot of people around me with that perspective. When that's the surrounding culture, there is more anxiety and pressure to prove yourself right away, to show you're doing the right thing.
--:-- Maybe the value of schools like MIT is that if you surround yourself with people at MIT, the idea of doing some world-changing innovation and building up the knowledge and capital to reach that point feels more natural than it might at some other university where it's like, "Why are you doing this?"
--:-- Scott Young: I'm making all this money doing this side gig right now. And you should be more like me or something like that. Very good advice.
--:-- Dwarkesh Patel: I actually have a vivid example in my mind of how that's directly applicable to me. Final question. You said you now have a son. The last chapter of your book was very fascinating, it was my favorite chapter because it's like transitions, like a Malcolm Gladwell book. I want to let you explain the example of Judith Polgar.
--:-- Scott Young: Yes, the Polgar example.
--:-- Dwarkesh Patel: Have you been utilizing that example of her teaching for your son? There's no way that I would do that.
--:-- Scott Young: I think for me, the Polgar example – and I do have a little bit of regret now because the Polgars are kind of overused as a story. It's a little bit like the reason I didn't put Einstein in the book wasn't because he wasn't a good example, but because he's too well known. The Polgars have been in a lot of books in this genre right now. I only really realized that belatedly, but I found it very interesting. I definitely thought it was a kind of interesting experiment.
--:-- In the same way that nowadays I don't really think the average person should try to do the MIT Challenge, but it's a kind of "this is something that could be done" and use it as a guidepost of things that are interesting. For me, and maybe this is my own cultural biases, I definitely don't have the kind of attitude toward parenting where I'm going to engineer my son to be like something.
--:-- I think what I want to do is give him access to experiences so that the kind of person he wants to become is influenced by what kinds of things I showed him were possibilities. This is the attitude I take with my readers and my blog. I don't like the manipulative approach of "here's how I twist and prod you into the thing." Rather, I always feel the best way to do things is to set an example, give people resources, show them how it's done, and allow them to become their own person. That sometimes means making their own mistakes too. You think they should do X and they're not doing it.
--:-- I'm giving advice to my younger self, but when my son is in his young twenties, he may see things differently from me. I'm certainly not going to be in a position to say you need to do this or that. The main lesson in my mind of the Polgar example is just how far you could go in changing someone's base nature. It's definitely a provocative experiment. It's certainly a counterweight to a lot of the ideas from psychology and behavioral genetics that most of us are fixed and have no changeable traits. It seems like a counterexample – it's interesting that you just decided you're going to have a chess prodigy and you had three.
--:-- But as an example, the idea that we are authors of our own life is super central to my value system. I would never want to rob my son of that feeling. Whatever he chooses to do and excel at and believe in life, that is something that he is the agent of, as opposed to "I'm living my dad's life plan for me, chapter three." That's my stance on it. I know people certainly differ, but that's how I think about it.
--:-- Dwarkesh Patel: Well, Scott Young, thank you so much for being on the podcast. This was the most fun podcast I've done and certainly the most personally interesting and useful. It was a great pleasure to have you on. You have a great guest list, so I'm very proud to have ranked so highly. Thank you.
