Transcript - Golden Nuggets Podcast #37 (Round 2): Balancing learning with creative output

Balancing learning math with doing projects that will get you hired. The role of mentorship. Designing social environments for learning. Why it's important to let conversations flow out of scope. Misconceptions about "slow and deep" learning. How to create career luck. The sequence of steps that led me to get involved in Math Academy (lots of people ask me about this so here's the precise timestamp: 1:13:45 - 1:24:45). Strategies to maximize your output. The "magical transition" in the spaced repetition process.

Cross-posted from here.

Want to get notified about new posts? Join the mailing list and follow on X/Twitter.

Link to Podcast

Below is a smoothed version of the raw transcript.

James: We are once again delighted to be joined by Justin Skycak, Chief Quant and Director of Analytics at Math Academy, who is here for round two. Thank you for being here, Justin.

Justin: It’s my pleasure. Great to chat with you guys again.

Zander: Now we’ll get to even more of our questions.

Justin: Yeah, let them rip.

James: The first topic I’d like to discuss is about the Math for ML course. Many of the people studying this course have the goal of changing their careers into machine learning, and I think I feel this as well: this heavy pressure to be producing some legible output along the way, to use as a portfolio or something like that. How do you think people should manage the balance between going into the Math Academy minds and just grinding versus working on projects alongside it?

Justin: I see what you’re saying, right? Because when you want to get hired for a machine learning or software role, a large component comes down to having a portfolio of cool projects you’ve done. So the bigger question is: how much time do you want to spend skilling up your foundational skills versus just using them to grapple with some project that looks impressive?

Let me talk about the failure mode that happens when people don’t do much foundational skill work and just focus on projects, and then we can kind of back into an answer about how much foundational skill work you need. At the extreme, if you focus just on projects and don’t know much about machine learning, I’ve seen this happen before many times. What typically happens is someone goes onto Kaggle or another platform, downloads a dataset, and gets off-the-shelf models, fits a decision tree, a regression model, or a neural network, or whatever. They end up with a standard project that’s been done thousands of times before. It’s unimpressive because it doesn’t really demonstrate knowledge of anything other than how to import a class and run it.

I think the purpose of a project is to not only demonstrate that you want to go out and do things but also that you’re able to do non-trivial things because you have the foundational knowledge to do that. Ideally, if I’m thinking about the ideal project to show off in a machine learning interview, it would go beyond just importing some off-the-shelf library and running it with default parameters or slightly tweaked ones. There should be something like constructing a custom loss function that’s specifically fit to your scenario or combining several models in an intelligent way that demonstrates you really know what’s going on.

To get to that point, where you can do a cool project that demonstrates high-level ability, you need quite a bit of foundational knowledge. I just realized there are two types of projects we’re talking about. There’s a project to show off in an interview to demonstrate ability, and then there’s a different project just for yourself to be excited about learning more machine learning. These are two different kinds of things.

For an interview-level project, you’d need a lot of foundational knowledge. That would involve taking a Math for Machine Learning course and also a full machine learning course that covers the specific algorithms. Our Math for Machine Learning course covers a lot of the math used in those algorithms, but it doesn’t cover the backpropagation algorithm. It covers the chain rule and matrix operations—the math you need to have.

So, right, foundational knowledge would be Math for Machine Learning and then an actual focused machine learning course. After that, you’d be in a position to do really interesting things like customizations. That said, working on less impressive but still fun projects along the way can definitely be a good motivational tool. Something like getting your feet wet with just running off-the-shelf machine learning models on datasets—that’s fine. There’s nothing wrong with that.

The only downside is, if you spend too much time doing that, you’re not focusing on the foundational skills that will enable you to do more impressive projects. But if you spend a smaller fraction of time on that and it keeps your motivation going, fueling you to keep doing more math for machine learning and implementing algorithms from scratch, then it’s totally fine.

So, my answer is: if we’re talking about the final project you want to show off before an interview, I’d say to get your foundational knowledge in place first.

James: No, it’s very useful to get because I feel like I’m in this situation as well. I think it’s analogous to a lot of things you face in life as a 20-year-old man. A lot of the goals you face are very long term, and the challenge is deciding whether you’re going to complete them serially, one after the other, or concurrently. When you have long-term goals, whether it’s fitness, financial stuff, or career, if you just did them serially, at least from my perspective, you’d be 50 before you hit the ones you’d deprioritized later down the line. It feels kind of the same with this machine learning stuff.

If I first complete Math Academy, then I complete networks from scratch, building from linear regressions upwards toward more complicated models, and only after that do I actually work on creating a really cool project that will get me a job, it will be so long before I eventually get to that final point. It’s frustrating for me as someone who has done the software engineering route, where it feels like on day one, you can basically start creating something creative. Even if you don’t fully understand the programming language, you can create a project that solves a problem in a creative way.

I guess my question is: does it have to be completed serially like that to get to that end point, or is there a way to make it more concurrent? Is there something about machine learning that makes it very different from the software engineering approach?

Justin: I have several answers. I’d say it can be done incrementally. I kind of split this into two ideas: there’s a real project and a trivial project. But in reality, there’s a whole spectrum of projects in between, and at some point, it becomes good enough to get you hired for a job.

You can go overboard and keep spinning up on your fundamentals, working through all these textbooks, and end up acquiring more knowledge than a PhD researcher in machine learning, but you haven’t actually done any projects. You’re just working out of textbooks, and you’re still thinking, “Oh, I just need to do these other ten textbooks, and then I’ll be right.” That’s ridiculous. I agree, that’s also a failure mode.

I think the real answer is to always devote some portion of your time to working on projects, but not to the point where it’s overwhelming and causing friction in your foundational skill development. Maybe something like the 80-20 rule—20% of the time working on projects. Eventually, your projects will get advanced enough that you can get hired for a job based on them.

It also depends on the job. Different jobs have different requirements and expectations. Sometimes, a simpler project will be sufficient to get an interview and get the position. Sometimes they expect more advanced things. Sometimes the type of project you do isn’t about how simple or complex it is, but how relevant it is to the company.

It’s probably a good idea to always spend a fraction of your time contributing to your portfolio and keep your career feelers out to see if anything is catching traction. You can also make it adaptive. If you try to go to some interviews and they’re not impressed with your projects, that’s probably an indication that you may need to skill up your foundations a bit more because they think the projects are too simple.

But if you’re getting a lot of feedback like, “Wow, this is really cool! We should have you talk to so-and-so on our data science team,” that’s good. It may just mean you need to keep leaning into that project and make it cooler, or move on to other projects.

I feel like I answered one of your questions but not the other. I forget what the other one was.

James: That was a great answer, really useful. I think the other part was, what makes machine learning different?

Justin: Oh, yeah. To software engineering, in this respect. The big difference is the body of foundational knowledge that machine learning sits upon. In software engineering, there are tons of nuts and bolts, and everything is going down to assembly code. But to make a cool, advanced app that does something, you don’t typically have to know assembly in order to do really interesting things at the top level.

There is a body of foundational knowledge there, and it goes really deep. But there’s a special, it gets so well encapsulated that you just don’t really need to interface with it very much, if at all.

Now, in machine learning, it gets very… there’s a bleed in the math foundations, and it bleeds out to the top level. If you want to do some off-the-shelf stuff with machine learning, no problem. But the minute you start wanting to customize things or if the off-the-shelf model isn’t working, and it’s time to tweak parameters to understand what you need to do, you have to really understand what’s going on at a mathematical level with the models.

I think the core difference from building an app and having to tweak things about the app is whatever you’re tweaking is better encapsulated under the hood. There’s some level where you get to pull levers and know exactly what they’re doing. But in machine learning, there’s no clearer label on what the lever is. The clearer label is phrased mathematically, and if you don’t understand the map, you don’t understand what the lever really does or the effect of doing it, if that makes sense.

Zander: Would you say that it would make sense if you were driven to do a project while your base-level skills are underdeveloped? That’s a good incentive, a good thing to work on? As you’re skilling up with Math Academy, you feel more confident about adding something to your project. That’s an immediate way to make use of your skills. Like you say, it’s kind of a trivial project. It might not be something actually novel or useful in the real sense, but maybe it’s useful to you because it’s an interest or whatever else.

It seems to me that would work because it’s not too far outside of your skills. When you get stuck, you kind of just keep skilling up and wait until you can make the next step and go from there.

Justin: I like that idea. You kind of crank on a project a little bit, and eventually, you get to the point where you don’t feel like you’re doing anything special. You’re focused on skilling up, then you come back to it later with fresh advice, fresh knowledge, and think, “Oh, I can do this cool thing.” You do that a little bit, come back. I think that could totally work. The only thing to watch out for is when you spend too much time on the project, not realizing you’re stalling out. But as long as you have a way of keeping yourself honest about whether the project is going in an interesting direction and you’re making improvements to it, and you’re always continuing to skill up on the side, then I think it could be a great idea to work on a project.

Zander: On the side, and as long as it doesn’t make you feel like your goal project is so far out of reach that it’s demotivating for you. Sometimes, with stuff that takes a lot of base knowledge, you think, “How is it ever going to be possible that I’m going to know how to do this?” That’s a balance, psychologically.

Justin: That’s true. Every person has different emotional responses to different situations. I’m just thinking about when I used to teach this really advanced computer science program. They would be implementing models from scratch and running them on datasets. Whenever new students came into the program, I was asking them to do difficult things. Different students had different reactions to it. Some were like, “Oh, this is really cool. I have no idea how to do that. I want to learn.” Other students were like, “Oh, this looks really challenging. I’m intimidated. I don’t want to take the class.”

It’s very personal. I guess it’s just a matter of identifying what you’re like at the core. Looking outside yourself and thinking, “If I were teaching myself as a student, what kind of motivational techniques would I be using?”

Zander: In your experience with that, since it’s so closely related, is there anything that tends to go along with that kind of gumption, or you could even call it agency of a certain kind? Did you perceive any way of helping those students skill up in that area? Gumption as an ability— is there a way to make them more excited about a challenge instead of shying away from it?

Justin: I know it’s a tough question, but I have a partial answer so far. My partial answer is that if a student manages to get through several hard tasks, hard projects, and they’re worried about the next one, I would always say in that situation, “Okay, you’re worried right now, but think back to every single time in the past you’ve been in a situation where you thought, ‘I don’t know if I could do this,’ but the project was right at the edge of your ability. You were able to come through. I was here to support you in it, and guess what? You got through it just fine.”

Now look back at that project. How hard is it now? “Oh, it’s super easy. It took you a week beforehand, and now you could knock it out in an hour.” Well, the same thing is going to happen. You kind of look historically at it. Once somebody has built up enough successes, you can remind them, “Hey, you’ve been in this situation before, it worked out fine. Don’t worry about it.”

Of course, it never completely resolves any worry, but it gets them over the hump. They think, “Yeah, I’m going to take the next step because I know it’s going to work out fine, even if I don’t feel like it’s going to work out fine.”

But the question is still, “What if it’s the first project someone is doing, and they’re just too intimidated by it?” I think the answer to that is: If you haven’t built up a history of overcoming challenges, it’s important for the first challenge to be scaffolded very well. It should be structured in a way that doesn’t seem too intimidating. You kind of gradually turn the dial on the level of challenge relative to the student’s ability.

So you start out with something that’s not as intimidating and maybe give more support. You just try to take steps to bring down the level of worry. But it’s true that it comes with a threshold. If someone doesn’t have this level of agency or gumption beyond a certain threshold, it just doesn’t seem to work out. There’s only so much you can do as a teacher or when supporting a student.

Scaffolding down the project can help, but you can only get it down to a certain point. There will always be a step the student has to take. Some students are just unwilling to take that step.

James: Interesting. Could you talk a little bit about the role of mentorship, maybe even in your own experience? I know you’ve written on Twitter about Jason being a great mentor for you. You’ve given this analogy describing the relationship between you two as doing a super intense PhD with the world’s most demanding and supportive advisor. I wonder if that kind of plays into this gumption thing. Is there a role for mentorship in increasing a student’s gumption or the scale of the project they’d be willing to take on?

Justin: I think mentorship can help with that, but I think typically the main value of a mentor, and not just the value to the mentee but also to the matching problem, is how to match really good mentors with people who need mentoring. I think that tends to favor mentees who are already in a state where they just want to barrel ahead.

At least in my experience, being both a mentee and a mentor, it works best when the mentee is basically a cannon who’s just going to blast out, and the mentor just has to point them in the right direction. The mentee is this crazy force, sprinting full force ahead, saying, “I’m going as fast as I can, just point me in the right direction, and I’ll keep going that way. Just tell me to watch out for snakes and sand pits or whatever’s going to try to attack me, but I’m going to give it all I’ve got.”

Ideally, if there were more good mentors in the world, you could give every person a mentor like that, and it would be great. It would definitely help with any lack of agency. But what happens in practice is that the people who get good mentors are the ones who have the agency to seek them out and convince them that they’re worth their time.

James: To broaden this question slightly, I’m really curious about what suggestions you give to Math Academy students in terms of designing that ideal social learning environment. When I look back through history, there are lots of examples in mathematics of small pockets, very small regions in different countries, which produced a lot of mathematical talent. One example would be the math circles set up by Kolmogorov after the 1940s. There’s also Louis Legrand in France, where half of France’s Fields Medalists attended, even in one family with the name Bernoulli.

So, exactly, even if people’s ambitions aren’t necessarily so grand or high as to achieve a Fields Medal, how should they ideally design that social circle around themselves in order to learn most efficiently and have the most fun?

Justin: Great question. What are the characteristics of a good social circle to support your learning math or just in general? I think one thing that pops to mind is—it’s kind of a cliché—but surround yourself with people who are smarter than you, who are more committed than you, who are just more intense. Surround yourself with people who are at least as intense as you. You always rise to the level of the people around you, or you fall to the level of the people around you.

We see this all the time in sports. If you practice with a team where you’re not the best player, and you have to prove your worth, you’re going to have a great game relative to your own abilities. It’s going to force you to take things seriously because of the competition. But if you’re practicing with people who aren’t quite at your skill level and you’re the best one there, even if you have a great game relative to everyone else, it’s not going to be as productive for you. You’re not going to be as motivated, and you’ll feel like you can dunk on everyone without trying. That’s the worst situation to be in.

When you feel like you can dunk on everybody without trying, it means you’re not trying. And because you’re not trying, you’re not actually improving. All you’re doing is dunking on people without increasing your ability to do things. It’s not a productive situation. It might help other people if it motivates them to step up, but at the same time, it could demotivate them due to the skill gap.

I guess the first criterion is to surround yourself with the most intense, most committed, most knowledgeable people who are far beyond your ability, but not so far that it’s totally demotivating. They should be above you in ways, but you should feel you can close the gap and join them. Once you’re really solid at particular skills within a field, it can turn into a thing where you’re better at things A, B, and C, and they’re better at C, D, and E. Maybe you’re working together, or maybe you’re just trash-talking each other a bit about, “Oh, you couldn’t even do A and B, watch me,” and they respond with, “Well, check out D and E.”

The second part is the social ties. It’s not just about feeling like you need to skill up to join the ranks, but also about the social bonds that keep you aligned. I saw this at the gym. One of the biggest factors in keeping someone going to the gym is whether they have friends there. If they don’t, they might stop exercising for a week and be at high risk of never coming back. But if they have friends at the gym and fall off the wagon, they can come back after a week, and their friends say, “Dude, let’s go! Where have you been?” That can bring you back into the routine.

It’s almost like if you start losing your habit and your friends’ habits are going strong, they can pull you back in. There’s a gravity that brings you back. It’s very motivating if you feel your personal bonds with people are increasing as your skills increase. It’s like speaking healthy nutrients into a yummy meal.

There are other factors that are important in a social group, but the two that come to mind are skill difference and the social bonds. Try not to be the best one there, have areas to improve, and make sure the people you’re working with are motivating you to rise to their level. At the same time, you’re developing nice social bonds with them. Do you guys have any other things that you were thinking of? Do you do any social groups for learning or career-related stuff?

Zander: I do some online work, and it’s surprisingly hard to find overlap with this kind of stuff in person, at least it has been for me. But online, we’ve been doing this for many years, and it’s been super helpful. When you were talking about sports, it reminded me to reemphasize the standards point. I think high standards are so crucial. If you look at the stories of the best players in a given sport, like Tom Brady in American football or Lionel Messi in soccer, they are known for raising the quality of the players around them simply by their own high standards and the way in which they practice and engage fully with the sport. Every practice session, or just by being there, they raise the level. At some point, you can be the best in some area, and your role becomes partly to set the standard for others. I think that general culture of high standards is crucial. That’s how you win the Super Bowl or the World Cup, because you’re raising everyone up. Of course, that’s a different stage in a person’s life and career when they’re already the best. Beyond that, be grateful if you’re in a place with high standards, because many people don’t want that pressure.

Justin: I remember reading about Kobe Bryant and the Olympic basketball team. He was waking up at 5 a.m. to work out before their official practice. I think if I remember correctly, he would wake up and go to the gym at the same time that some others were going to sleep after a long night out partying. They’d say, “Dude, what are you doing?” And he’d reply, “I’m going to work out. What are you doing?”

Zander: Yeah, he’d show them. Exactly. I remember a story from that time where they all wanted him to go out with them at night. He said, “All right, I’ll go out with you, but in the morning, we’re going to be at the gym by the end, no matter what time we get back.” They went out super late, and everyone else was sleeping, but he was there in the gym, even though he only got like two hours of sleep. Some people are just driven, and that’s a valuable thing. That’s how you win the Olympics.

Justin: Or win the Nobel Prize or whatever it is. Totally. I think one interesting thing that stood out to me about that story is that at first, it seemed like the teammates were almost annoyed or laughing, just kind of like, “What the heck? This guy is up at 5 a.m. in whatever man.” But then he kept on doing it, and it kept sinking in. Eventually, they were like, “He’s right. Do we want to win this thing or not?”

It’s not just a one-and-done thing. It’s not like you just do something once. It’s a continuous thing—you just keep doing your thing, and eventually, people join in.

Zander: One thing that’s corrosive to that attitude of high standards is when you see someone who’s in good shape and says, “I hardly even go to the gym. I don’t even work out that much. I eat whatever I want.” It’s like, you’re missing the point. You’re bragging about not trying very hard. That’s not impressive. It just means you’re somehow locked into some genetic gift. The impressive thing is to try hard, no matter what your baseline is.

I think it’s the same thing with learning. If you’re not trying hard wherever you are right now, maybe your baseline is much more intelligent than someone else’s. But that doesn’t give you slack to do anything else just because you’re at a higher baseline. That is corrosive to the standards. It doesn’t matter where you are; it’s about where you put in the effort.

Justin: That’s really interesting because sometimes you can get labeled as a try-hard or as an overachiever, with a negative connotation. That can sometimes lead you to want to just say, “Oh, I don’t try that hard.” But it’s actually the better way to get people to rise to your standards—just lean into it. Be like, “Yeah, I try really hard. Why don’t you? Come on, let’s go.”

It’s a counterintuitive emotional response, but it makes total sense. People often use a lack of talent to justify not trying very hard. They’ll say, “You’re able to get so much more results than me. Why should I try?” But they’re inverting the fact that not having that talent is exactly why they should try harder.

That’s the key principle. A guy like Tom Brady, if you know American football, doesn’t have natural talent, and yet he’s the greatest quarterback ever. How did that happen? Yes, he’s tall and has some baseline qualities you need, but there are other ways to skill up in life and overcome deficits. A lot of that has to do with high standards, work, consistency, and all the things we’ve talked about before about how to get the most value out of Math Academy or anything else in life. These are the traits required.

To encourage that in the community, I think, just to get back to the core of James’s question, at least in my experience and awareness of this, that’s a big piece of it.

Justin: I was just going to ask if you had similar experiences or characteristics of good social groups you’ve been in.

James: It’s definitely something I’ve been thinking about a ton recently. I’ve been reading this book called Collaborative Circles, and it’s about these very creative groups throughout history where each of the members credits the group for playing a huge role in the creative output they went on to create later on. A good example is the Inklings, which is right in my hometown of Oxford. It was a literary circle that included J.R.R. Tolkien, C.S. Lewis, and a bunch of other writers. They all credit the meetups they had at a pub in the center of Oxford called the Eagle and Child Pub for the works they later created.

It’s been making me think a lot about how, since I’ve left my job to pursue this learning thing full-time, I should be designing my social environment to best use the time available. It’s been tough. As Under said, it’s very difficult to find people in real life who share these niche obsessions with you. I find it much easier to find people online—almost effortlessly, to be honest, especially through Twitter and Discord. That’s the approach I’ve been taking at the moment, and I’ve had some success. Me and Zonda spend a lot of time in Discord with people, doing Pomodoro sessions together and chatting about our projects. That’s been really useful.

But I do think there’s something about having a group of people in real life who are pursuing the same mission or have a shared set of overlapping values, which can’t necessarily be 100% replicated online. Even though I think for now, online is probably the best way for me. It’s something I’ve been thinking a lot about, but I don’t know fully how to do it to the best degree.

Justin: That’s really interesting. I’ve had the same exact experience. It’s so much easier if you have more niche interests. Online, it’s just so much easier to find people. But before I got on Twitter, before I was talking to a bunch of Math Academy students, I didn’t really know anyone outside of Jason, Alex, and Sandy. And that was about it. No one was really interested in the same things as me, like applying cognitive learning principles to math learning and mapping out the entire process. I’d nerd out about it with people here and there, people who were intellectually curious in other ways, maybe super into data science or whatever. But they’d think what we were doing was mildly cool, but they wouldn’t get super excited.

Once I started posting a few things on Twitter, it was so weird. I suddenly had 20 different people talking to me. I was like, how did I miss this for all these years? I went through college thinking nobody was interested in the same stuff as me. But it turns out, if I had just gotten on Twitter and posted some stuff, I would have had so many great relationships. Better late than never, though. It was phenomenal.

I agree, James, that if you can find those people in real life too, it amps up the effect even more. But the challenge is how to find them. Sometimes it feels like an intractable problem if the interests are niche enough, because the density of those people isn’t very big.

James: I think the challenge is probably figuring out how to design an online community in a way that you can get all the benefits of a real-life community. That’s the thing I’m not sure about. I don’t know if there’s a design pattern out there I can just easily copy and implement in a Discord server to get the best results. It feels like you have to do a lot of experimentation to work out how to arrange things so the social environment is really creative. You want people to enjoy being there and feel propelled into doing much more ambitious things than they thought they could. But there hasn’t been much research, as far as I know, into how to do that. In fact, there probably isn’t much understanding of how people do it in real life either. It’s just a very difficult problem we wish we could solve because, if we could, we’d have a lot more creative output. The social environment is so important to people.

Zander: It’s definitely something I want to work out. I think some of it just can’t be replicated digitally. I could be wrong about that, but it seems like some of it just can’t be replicated. There’s a benefit to being in person. Take a work environment. People talk a lot about remote work and the benefits. There are many benefits, obviously, but one of the things that can’t be replaced, at least not so far, is this tacit, implicit learning, osmosis kind of thing you get when you’re just around other people who are working very hard and are better than you at something. You’re observing what they’re doing and absorbing all this tacit, implicit knowledge. There’s so much of that, and it seems hard to replicate digitally with any kind of technology. You can have your camera on all the time or whatever, but it doesn’t give the same result.

James: That’s right.

Justin: Hi guys, James here. I hope you’re enjoying the podcast so far. I just wanted to let you know that me and Zander have a secret Discord server full of people working on various learning projects and doing Pomodoro sessions during the day. The Discord is private, but if you message me or Zander on Twitter, we’ll send over an invite link. That’s all, back to the podcast.

Justin: One kind of interesting thing, I guess, that’s relevant to this topic is I’m working remotely currently, and I have been for the past year, a little over a year, for Math Academy. But when I started out, I was in person. I worked together with Jason for so long, sitting right next to each other and having conversations. I lived with him and Sandy for a year or two during the pandemic, actually. We were doing everything together. We had so many of those late-night conversations with Jason about the philosophy of Math Academy and various things we were trying to do. That totally helped in terms of getting us both on the same page about the vision for what we were doing.

If I had started out fully remote from the onset, I don’t know that it would have been so much more difficult. It’s hard to imagine.

James: It’s funny because if you were just communicating with people asynchronously and doing these stand-ups a few times a week, it cuts out a lot of the time you’d spend doing idle chit-chat. But that’s actually really useful because that’s when you’re figuring out the philosophy of what you’re building. It’s hard to justify that in a remote work environment—like, let’s sit on a call and talk about philosophy for four hours. But that’s the kind of useful, high-level vision you get when you’re with people in real life. I think it’s being cut out of remote work.

Justin: Yeah, yeah.

Justin: One thing that we still do quite a bit at Math Academy is that we don’t have any scheduled meetings. We don’t have a weekly stand-up or anything. It’s just kind of like we’ll ping each other—Jason will ping me, or Alex will ping us, or one person will ping another person, like, “Hey, do you want to catch up for a quick call?” A quick call always turns into a one- or two-hour call. It always starts out as, “Let’s just sync up on some status updates,” but it always spins into, “Let’s just keep talking about whatever, the philosophy of what we’re trying to do and how things are going in our lives.”

It starts out as a directed call with a purpose, but it spirals into a nice, couch-like conversation. One thing that’s very helpful is not cutting off the call, not trying to scope it down so much that you only cover what was originally intended to be covered, but just letting it go on for a while and talking about whatever. That’s been very helpful to us in having a remote company.

Zander: That same page thing, I think, is key. The fact that you have to try to stay on the same page when you’re remote. As you mentioned, when you’re in person, if your company is five people, you don’t have to write any company documents. Everyone already knows how to do things, what to do, and what the vision is if you’re in the same place. But if you’re five people and you’re all remote, you constantly have to write stuff, and I do think there’s some innovation to be done there to help sort that out. But it seems like a tough problem to solve digitally.

James: In the name of things that get lost in the pursuit of efficiency or productivity, I have this kind of wacky quote from Seem to Laugh, and I was hoping to get your take on it. It’s a little bit of a tangent, but he says, “I don’t want to learn fast in any subject. I don’t want shortcuts. If I don’t enjoy the subject, I don’t want to learn it, and if I enjoy it, I want to prolong the pleasure. I avoid what exam takers do; I trade speed for depth.” How do you feel about this quote?

Justin: I feel like it’s very easy to misinterpret. I don’t think there’s anything necessarily wrong with the quote. I think the way the author intended it to be taken makes sense, but I think the way most people will take it is not the way the author intended. It reminds me of how a lot of people will take this quote, thinking, “Oh, so-and-so mathematician was always slow at math, they were always slow to work out their problems, so I should not have to take the time to test or learn my multiplication tables. Let me just take my time on these things.”

One key difference in many of these comparisons is that well-known mathematicians were slow because they were working out every problem from first principles, deriving everything. Whereas, most math learners, when they are slow, it’s because they have no idea what’s going on, and they’re just flailing around.

If you are able to deeply focus, where you are actually spending the whole time being productive, deriving things from first principles, building things from scratch, thinking deeply about the material, that makes sense as something productive for learning. If you’re in the middle of some deep thoughts about a new math theorem, reading it, thinking, “Oh, I wonder if these things I know are examples that would hold here, but would they hold in these cases? What about this case? Would it hold there?” and you’re getting lost in productive thought, that makes sense.

But unfortunately, what’s typically going on is that a typical student is taking a long time to read instructional material or solve a problem and is spinning out of control in an unproductive way. They’re just reading the same thing over and over, hoping it will just click into their head or somehow magically resolve the confusion. They’ve read it three times, and there’s still a part that’s uncertain. At that point, it’s time to move on, go look at a worked example, and see it in a different context. Maybe things will make more sense. That’s the most optimistic scenario: a student trying to work productively but failing to do so.

The least optimistic scenario, unfortunately the most common, is that the student is distracted by unrelated things. They look at a theorem or a worked example, and then they think, “Oh, YouTube is calling. Let me watch a video,” or “It’s time to text my friend.” They get distracted and move off-task. If a student spends an hour in class but only gets through five or ten minutes of focused work, that’s an indication that something has gone off the rails. Usually, they’re just goofing around with non-math-related things.

To go back to the original quote, it makes sense, but it’s operating on this very optimistic worldview of what’s happening in students’ brains. There are definitely students who fit that category, but they are far outnumbered by those who are not working slowly productively. Typically, that indicates unproductive friction in the process.

Zander: I think it depends on your goal. If mathematics is an end in itself and you’re purely learning for pleasure, then go ahead, take your time and learn for pleasure. But if you’re trying to do something with it, if it’s a means to get something else done and that’s part of your goal, then getting through it quickly, as long as you’re not suffering miserably for years, does seem to be the right approach. Even if you are enjoying it along the way, it’s not purely an end; it’s also a means. So it makes sense to speed through it, even if it is pleasurable.

Justin: That makes a lot of sense. That’s a really good point. The whole goal orientation—I’ve run into so many mathematicians who seem very against the idea of trying to work through things quickly and efficiently. A lot of the time, the response is kind of like, “No, just let the students enjoy and take the math in, take pleasure in it.” The perspective they’re coming from is that they’ve never actually felt pressured to learn math. It’s always just been a very enjoyable experience. They’re assuming that math translates directly to everybody else, which often it doesn’t. But it totally makes sense why there would be that perspective.

I guess it’s true that if you take so much pleasure in math, then you can have your goal just be to take pleasure in math. If you’re doing that in ways that happen to be productive, you will reach a high level of skill even if you’re not directly optimizing for it. But that’s very exceptional. I also think there’s some risk of…

Zander: Thinking of things in the Platonic way. If we take weight loss or fat loss, for example, it’s better to lose fat at a slower rate because you lose less muscle mass. But if you lose fat too slowly, you risk being demotivated. I’ve been on a diet for three years and I’m still overweight. You’re doing it optimally in the sense that you’re not losing muscle mass, but you’re getting demotivated, and you’re going to quit, still far from where you want to be.

You have to balance these psychological aspects with doing things in the optimal way. Maybe the optimal way is pleasurable but also fast enough that you don’t get demotivated along the path to the ultimate goal. I think there’s more than one thing to balance there.

Justin: Right, making progress is motivating.

Zander: Exactly. And if you’re working so slowly that you’re not making progress, that can be very demotivating.

Justin: I think that’s part of the magic sauce of Math Academy, honestly. You’re always making progress. You hardly even spend time reading because everything is so fine-grained. You just read for a couple of minutes and then you make progress. I think that makes people realize, “Hey, I can actually do this. I’m able to solve a new type of problem I couldn’t solve five minutes ago.” Then 15 minutes later, I’m able to solve another new type of problem.

I think that’s part of the reason why you see so many people voluntarily paying 50 bucks a month, maybe quitting their job, maybe spending 20 hours a week in addition to their job doing this. It provides that huge sense of progress. I think that’s a big part of the secret sauce—the speed and the sense of progress, that visceral feeling you get. It’s all connected to the pleasure. I think progress is pleasurable in itself.

Justin: This reminds me of a lot of the things I’ve read in the literature about deliberate practice. When you’re engaging in deliberate practice, it’s not intended to be pleasurable, and oftentimes, it’s not fun at all. But it is the progress towards a goal that gives you pleasure. Think about star athletes like Kobe Bryant in the weight room at 5 a.m. I’m sure that wasn’t fun for him, but he could tell it was making an impact on his game, and that’s fun. Winning games is fun. Reaching a high level of skill is fun. It’s almost like you have to redefine what you view as fun and pleasure. It’s not the activity itself; it’s the results of the activity.

Zander: I think that’s right. I think it’s a confusion of terms. You get satisfaction from having run a marathon, but while you’re running it, it’s painful.

James: I’m curious, in your own case, what was your goal in studying mathematics? Did you ever see yourself becoming a mathematics professor and staying in the academic world, or were you always planning to use mathematics in a more practical way, like you’re doing at Math Academy?

Justin: Yeah, it’s a good question.

Justin: When I started leaning into math self-study, that was like sophomore year of high school, so I was probably around 15. I remember my train of thought—it wasn’t a particular thing I was shooting for. I wasn’t learning math with the intention of becoming a professional XYZ, and it wasn’t even to get into a good college. I neglected a lot of my applications because I was so focused. The reason I got super into math learning was that it seemed like wherever it took me, it was going to elevate my quality of life. It seemed like the difference between me and doing cool stuff was math.

It was also something I found cool and enjoyed. I thought, “The difference between me and doing really cool stuff like physics, computational biology, research, machine learning, or whatever it is—I need math to get there.” I really love math, it’s fun for me to learn, so why not start leaning into this a ton and get there faster? That was my thought process.

James: I know you’ve written on your website that instead of optimizing returns in the stock market, you optimize learning efficiency in students’ brains. I’m curious, was quantitative finance or working in a high-frequency trading firm ever on the cards for you? Is that something you were interested in?

Justin: It was something I thought about for some time. I went through a lot of phases where I would flip between different possible life paths before eventually finding one that happened more naturally. It’s kind of funny because every time I thought I was going to be something—there was a time I thought I was going to be a mathematician, then I thought I’d be a theoretical physicist researching black holes. As I got more into math, I shifted to being a mathematician because I liked the tools of math. Then I thought, “I want to apply these tools to real-life scenarios,” and I became interested in data science. I even worked as a data scientist for a few years.

This just flipped back and forth. There was a time I thought, “I want to get involved in quantitative finance; that seems really cool.” But every time I focused on something, it ended up showing me that that wasn’t what I wanted to do. I don’t know how generalizable that is—it’s probably not very generalizable. Maybe I was jumping the gun and trying to decide what I wanted before my feet actually moved in that direction. But in the end, I just did what I enjoyed—getting involved in math education. The way I got into quantitative education and learning research was when I felt at a fork in the road. I could either choose math education, which didn’t have a lot of quantitative aspects, and just focus on tutoring and teaching, or lean more into data science and financial analysis for banks and digital marketing analysis.

I ended up choosing math education because it felt closer to my heart. I didn’t enjoy the specific domains where I was applying data science. I didn’t know that I would end up where I am now, but I kind of hoped that I would eventually connect back to something I find fun and interesting. And that’s exactly what happened, a lot faster than I thought.

Zander: The world is better for it. It makes me think about how many more cool things we could have if some of the smartest people alive weren’t figuring out when to buy and sell stocks and derivatives. Not that there’s anything wrong with it per se, but it seems like a waste—maybe it isn’t, I guess, if you’re making money and doing something good with it. But it seems like a black hole of cognitive output.

Justin: I know what you mean. A lot of the most mathematically capable minds end up in a situation where they’re just competing against each other. It does provide liquidity to the markets, which is super nice, but it also feels like I’ve seen it happen to people I knew in college or after. They were really intense mathematically and about improving people’s lives. Then, four years later, they’re working at a trading firm, and that’s cool, but I’m a little disappointed.

Zander: I remember reading that Michael Lewis, who wrote Liar’s Poker, the book about his experience working in investment banks in New York in the eighties, said he wrote the book so that people wouldn’t follow that path. He wanted people who really wanted to be something else—like a marine biologist or something that would benefit the world—to realize that there’s nothing in that field. It’s just vacuous, just empty. He said people didn’t take that lesson, but I do think it’s a common problem.

Justin: One thing that just came to my mind is that I think one reason why that is such a common path for mathematically skilled people is that it’s such an obvious application of math. You learn a bunch of math, and then maybe you want to make a lot of money in the future and do a lot of cool math. Well, if you want to do that, what are the most obvious paths? One is to become a mathematician, another is theoretical physics or academia in general, and then in industry, there’s quantitative finance and data science. If you throw in wanting to make a lot of money, then the academic path is kind of shut down, and you’re funneled into industry. Finance becomes such an easy path. For someone who’s really interested in finance and excited about strategies, options, derivatives, or whatever, that’s great. But there are a lot of people who try to go into that field just because they see the dollar signs and think it’s an application of math. They go in that direction because they’re following the crowd, when in reality, there are so many areas where you can apply math and quantitative software in the world. The catch is that it’s not very obvious all the time.

Zander: Now, at least with machine learning, there’s more of an obvious idea that it’s both lucrative and very mathematically intensive. If you’re interested in that, it’s a better path in terms of actual progress for the world, if you believe AI progress is good, obviously.

James: Totally. Interesting. Do you think there’s a way to go hunting for these hidden applications of math, or do you just have to stumble into it?

Justin: I think what Zander was saying, with machine learning coming to the forefront and AI and data science, there are a lot of industries where it’s becoming more clear how to use quantitative analysis to create improvements. There are companies built around these fields. For instance, drug discovery, protein design—there’s a lot of machine learning going on in that area. For someone in computational biology, there’s an easy on-ramp there.

But the core of your question seems to be if you don’t find yourself being pulled towards any of those existing paths, what do you do? What I did was basically just say, “Screw it,” and jumped headfirst into what I enjoyed doing, hoping things would work out. I’m hesitant to recommend that as a general strategy because I don’t want people ending up homeless or in massive debt, but I think if you have a field in mind that you have so much conviction towards, and you can’t be persuaded not to go for it, then you just have to go for it.

Since it’s a new field, and there aren’t obvious existing companies doing what you want to do, you have to go the startup route. It’s kind of like how people say about acting: If you can be convinced not to be an actor, don’t be an actor. But if you’re not going to listen to that advice, then you go into it with full force, and it has to become your life. The same goes for pursuing quantitative mathematical pursuits in some area where it’s uncommon. You have to be all in, or don’t do it. The easy path of joining an existing company that’s doing what you find interesting isn’t there. If you’re going to do this, you have to see it through.

One part that was really helpful for me was running into Jason and Sandy early on. When I made the jump into math education, I lucked out by finding the right people who were interested in the same thing. That’s a necessary ingredient. If I hadn’t met them, would I have started my own company doing something similar? Probably not. You have to jump into it full force, but you can also run into the right people who are doing what you’re interested in, and that can provide an easier path to join an existing movement.

So, if you want to do math and you’re looking at the available opportunities and the companies already doing things, but you’re not really into any of them, and you have an area you’re really interested in, then your options are either to start your own thing or find a group of people who are doing that kind of work. They might not be a well-known company; it might just be the seed of an idea or in an R&D stage. The challenge is running into people who have your niche interest, which comes back to what we were talking about earlier.

To find that group online is probably the easiest way. I don’t know how replicable my path to Math Academy is. I didn’t even plan for it. I moved from Indiana to Los Angeles with the idea that I needed to be in a bigger city for math education because tutoring doesn’t work well in a small town. The payment rates are too low, and it’s hard to make a living tutoring in a small town. Moving to a big city allows you to make enough money tutoring. I happened to move into the right neighborhood, Pasadena, which is right next to Caltech. Caltech is very advanced mathematically, so there were a lot of serious mathematicians around. It made sense that this was ground zero for serious math education.

If you want to do something where there’s no clear career path but you know there’s some ground zero for it, or a place like Silicon Valley or Boston, you just have to go there in person, talk to people, and hope you run into the right people. Then you can join up and do something cool. But of course, how do you know where ground zero is? It’s hard to figure that out.

Zander: Sometimes I think, when people talk about things that have luck components, they think they’re inherently improbable. If I’m doing something that requires serendipity, it feels like it’s very unlikely to happen. Hopefully, I’m one of the lucky ones. But there’s also this sense that if you stay in the game long enough, if you’re doing the right things, trying as hard as you can, keeping all these factors in mind, and being a good person so that people want to help you, you’re increasing your own luck.

These aren’t meant in a woo-woo way; I mean it in a connected, practical sense. You’re multiplying opportunity by going to the right places, meeting more people, and helping them out where possible. By doing these things, you’re increasing the probability of what people would call a lucky encounter. But it’s not luck to move right next to Caltech. That’s not lucky in any sense unless you didn’t realize you were doing that, but even then, it’s still a deliberate choice.

It’s the same with people moving to LA to be an actor or to New York to work in finance, or moving to Silicon Valley to be in the startup world. These are common paths, and while most people won’t succeed, it’s not because they fail to do the “luck” part. It’s because they fail to do the surrounding parts, like being worthy of the opportunities when they come or seizing the right ones. Not to say that everyone can succeed in a luck-based thing, but I do think people should be discouraged from relying solely on luck.

Justin: That’s a really good point. There are actions you can take to increase your luck, and you can compound those actions over and over until it becomes a statistically likely event that you find your group of people.

When I moved to LA, the decision to move to a bigger city wasn’t just luck. It was a real decision based on the lack of opportunity in a small town. I needed to move to a bigger place. Living in Pasadena was a lucky choice for me, though I did like the environment there. It felt like there were people I could relate to, but I wasn’t consciously thinking at the time, “This is where I’ll find a group of really interesting people.” It was more like, “This is what I’m enjoying,” but that was highly correlated with finding the group of people.

One thing I was doing over there was going out and doing a ton of stuff, working with several tutoring agencies. I got involved with Math Academy’s school program. I learned about Math Academy because I was tutoring a kid whose parent was well-connected with people in Pasadena and mentioned the program. That put it on my radar.

I even met Sandy and said, “I’ve got this really bright tutoring student who seems like they’d be a good fit for Math Academy. Would you be willing to meet with them to talk about what Math Academy is?” At the time, Math Academy was just a school program with some basic software. We didn’t even have automated systems; it was just software where the teacher could select questions and have them graded.

We met at a coffee shop, me, Sandy, the student, and their parents, and had a chat. At the end, I said, “Hey, Sandy, if you ever need help TAing, let me know.” She said, “Actually, we would benefit from having a TA,” and that’s how I got involved. I did group tutoring for Math Academy students and eventually tutored Jason’s son, Colby.

Through that, I’d often chat with Jason. Normally, after a tutoring session, you’d talk with the parent about how it went, but with Jason, it started like that and then turned into sitting on the couch for an hour, not even talking about tutoring anymore, just discussing math education. Eventually, I ended up creating videos for Math Academy. Jason wanted a way for kids who missed class or needed to catch up over the summer to use the system and prepare for the upcoming year.

In the summer of 2019, Jason said, “We need to automate this system. You’ve done a lot of work in data science, and this seems like your alley. Why don’t we talk through it and see if you can help us get this thing on the rails?” That’s when things really started taking off.

It was a sequence of continually asking, “What are the various things I could move towards?” Doing this, doing that, and eventually landing in the right spot.

Zander: I think that’s a perfect example. What a fantastic story. These luck stories are a perfect example of what it’s not like. You’re not just walking down the street in Indiana, and someone hands you a piece of paper saying, “Do you know mathematics? We need your help.” You had done all these previous steps, getting tutoring opportunities and following up on the parent’s suggestion. There were so many steps along the way. You were mentioning this at the coffee shop. All these things multiply your chances of having a lucky opportunity.

And I think, just because we were talking about communities, that’s why I’m harping on this. It’s about being in the game long enough and providing help where you can. You don’t reach out to people and say, “Hey, I’ll help you. I don’t have any skills, but I’ll help if you want.” You offer real value. You already have the skills. It’s about being in the game long enough, having the ability, and seeking opportunities. Maybe your luck will increase. That’s beautiful. That’s awesome.

Justin: Lean into the skills that you do have to get in the game. Step one is just getting in the game. You don’t have to play the position that you’re meant to play from the beginning. Step one is just getting yourself on the playing field with whatever skills you have. For me, initially, that wasn’t even coding. It was just tutoring and making math videos. But one thing led to another. Get in the arena, and then things take off. That’s pretty much my comment.

Zander: Even Satya Nadella, I don’t think he started out as a CEO candidate. He was just some Microsoft employee. Now he’s the CEO after a long enough time. That happens a lot. Don’t even worry if it’s exactly what you want to be doing. As long as it’s close enough and you can accelerate from there, it works. Anyway, not to harp on this even more, but I think it’s valuable from a life perspective of how to get into more of these lucky opportunities.

James: That was really nice. It was really nice to hear that. Sorry, I had a lot of thoughts. Would you mind if I ask one or two more questions?

Justin: Sure.

James: One thing we didn’t manage to get to last time is the productivity side. I’m curious how you manage your time, especially working on Math Academy alongside writing the book and doing the great Twitter posts, without getting caught up in scrolling or other distractions.

Justin: I’ll say that it is pretty overwhelming. It’s not that I have discovered some secret productivity hack where you can just sit at your desk for eight or nine hours a day and have a relaxed evening. It’s very exhausting. I push myself to the limit in terms of output, and honestly, I take that too far sometimes. I’ll get exhaustion symptoms, like cold and fever symptoms, just from working too hard. Then I need to get some sleep, and once I rest, it goes away.

I just want to start off by saying that the level of output is not sustainable through comfortable means. The enemy of productivity, as James was suggesting, is scrolling on things that are not moving the needle. One trick I use to avoid unproductive scrolling is to cycle between different tasks during the day. I’ll spend an hour doing one thing, an hour doing another, and maybe cycle back to the first task. These are all things that need to get done. I’ll switch between writing, coding, being on a call with somebody, and other tasks. It keeps things varied and helps me stay engaged.

The variety prevents boredom. It’s not that the task is boring; it’s just that after an hour, you can get tired of doing the same thing. If you change what you’re doing, it can hold your interest longer. Another trick I use is piggybacking off existing work. When I post something on Twitter, it’s often based on a recent conversation or a thought that just popped into my mind. I can quickly put it to paper without having to grind much.

Sometimes I’ll take a response I gave to someone, where I didn’t get much traction, and turn it into an interesting post. Or I might take a section from Math Academy that I posted and expand on it based on the reactions I got. I also take more formal content and style it into something punchy, like a motivational post. Instead of several paragraphs, I might break it down into short, punchy sentences, kind of like a personal trainer speaking to you about fitness.

Initially, I wasn’t sure about rephrasing things. It felt redundant, but I realized that delivering the same message in different ways can reach different people. It’s like how pop music uses the same four chords, but the rest of the song makes it unique. Repeating the same message in various formats helps it sink in with different audiences. It’s a way to keep things fresh while still delivering the same core message.

James: It seems like a good way of using Twitter to put out posts with some major ideas and later turn them into fully-fledged posts. It’s very useful.

Zander: On the topic of external things to Math Academy, since we’re working on projects, I want to ask about your plans for a separate algorithm for term and definition stuff. It seems to me that definitions and remembering the names of things might require a different schedule. I know it’s an extremely detailed question, but some people feel they need external support, maybe with their own spaced repetition system. I saw Adi Matushak mention this in his notes about remembering terms or names of things. I’m wondering your thoughts on this, if it’s something you’ve considered.

Justin: It’s something to think about in the future. It’s true that remembering a particular term or definition is more fine-grained than remembering a broader topic, so it makes sense that various components of a topic could be on different spaced repetition schedules. However, the spaced repetition schedule for a topic is typically a good approximation of the spaced repetition schedule for the definitions or terms used in that topic.

In a topic, spaced repetition involves solving problems, and to solve the problems, you need to recall definitions or theorems. Provided you are recalling properly—not just looking up the definition but actually trying to recall from memory—then the repetition schedule for the topic should align with the repetition schedules for the component information.

One thing I’ve heard from people is that initially, they feel like they’re forgetting some things. They do a topic, solve some problems, and then when the review comes later, they’ve forgotten the theorem or definition. They need to review the lesson again for a reference. It’s not like they’ve forgotten everything, but they can’t retrieve it from memory right away. Then they trust the system, continue using it, and after some consistent use, they reach a point where they suddenly think, “Wait, I can remember this stuff now.” They wonder why they couldn’t remember it before.

I think what’s going on is related to the spaced repetition mechanics. The first time you see something, your memory decays rapidly. If you’re late to the next repetition, your memory will decay more during that period. If we’re a day off in the repetition schedule or if you skip a day, it can cause some jitter in when the repetition actually happens versus when it was optimal. This causes memory to fade quickly, and you might fail the initial retrieval during the next review, requiring you to look at the reference material. However, after you do that a few times, your memory decay slows down. After consistent repetition over time, your memory decays more slowly, and even if there’s a delay or gap, your memory is more stable.

This process is when you feel things locking into place. It’s a bit speculative since I haven’t run empirical experiments on this, but it seems to be a good explanation of what’s going on.

Zander: And on that particular spaced repetition point, that does match my experience with spaced repetition in general. I think I even emailed Peter Wosniak about this many years ago. I was basically saying I can’t seem to recall the memory until it settles in, before I really have access to it in as easy of a way as I do for some items I’ve had for years. And he seemed to agree that sometimes you just need that initial repetition or two for the memory to become solidified and get stronger, like you say, a little less variable and susceptible to that quick drop-off. But yeah, I have so many more questions. Somehow we’ve gone another two hours and I still have at least four hours worth of questions.

Justin: As I’ve had, I know James has to go, but if you wanted to stay on, I’m happy to chat with you.

Zander: We could do that, or what we could do… I know we don’t want to jump the gun here, but would you be open to a part three? Maybe not soon, but in a couple of weeks. Would you be open to something like that?

Justin: Oh yeah, totally. That sounds awesome.

Zander: That would be awesome. That way James could be here as well, because I know he has a bunch of questions too.

James: Yeah, I’ll be happy. I’m sorry for calling it early, but it’s getting so dark, and I have to run to the shops to get food and other stuff. I’m sorry, guys.

Justin: No worries.

James: I won’t be able to get my XP today. I think I’m in. So, I know it’s necessary, right?

Justin: Then no more part three. All right, going to take care of the future here. Make sure James gets us good.

James: Yeah, I’m happy to. I’ll send you some date suggestions.

Justin: Awesome. That would be perfect.

James: Thank you so much for your time again. It’s a pleasure to talk to you. It’s always super fun.

Prompt

The following prompt was used to generate this transcript.

You are a grammar cleaner. All you do is clean grammar, remove single filler words such as “yeah” and “like” and “so”, remove any phrases that are repeated consecutively verbatim, and make short paragraphs separated by empty lines. Do not change any word choice, or leave any information out. Do not summarize or change phrasing. Please clean the attached text. It should be almost exactly verbatim. Keep all the original phrasing. Do not censor.

I manually ran this on each segment of a couple thousand characters of text from the original transcript.

Want to get notified about new posts? Join the mailing list and follow on X/Twitter.