Interview with Ben Kuhn about productivity, his college experience, job choice, the value of an inside view, the EA communitycreated: ; modified:
This is a transcript of my interview with my friend Ben Kuhn (@benskuhn). Ben Kuhn is the CTO of Wave, a company building financial services for unbanked people in sub-Saharan Africa. Wave is hiring.
The video of the interview is available here: https://www.youtube.com/watch?v=5NfpjbAP-uM
- 00:35 How Ben spent his winter
- 02:40 How being a CTO is different from being a software engineer in the context of productivity, weekly reviews, day-to-day happiness, and long-term satisfaction
- 17:56 On being a good manager and channeling one’s neuroticism productively
- 26:19 On doing 1:1s with Ben’s partner vs doing 1:1s at work
- 39:08 How finishing undergrad in 3 years worked out for Ben and what college is for
- 47:51 Is college a good place to explore your interests?
- 53:03 What Ben learned from starting the Harvard EA Chapter
- 57:08 Where do people who you talk to these days come from?
- 1:02:46 Interning at Jane Street and the (non-)importance of solving hard technical problems for making an impact
- 1:13:03 Interning at GiveWell, trying to be a researcher vs being a programmer
- 1:21:07 Trying to start a “startup”
- 1:22:38 Working at Theorem LP
- 1:29:38 Finally learning about Wave, realizing it’s a big opportunity, and getting a job there
- 1:43:25 What Ben disagrees with his friends about: the value of an inside view, the inefficiency of markets, the base rates
- 1:51:20 What Ben thinks about the state of the EA community
- 00:40 Your room can be as bright as the outdoors
- 03:00 I apparently got 50% better at my job last month
- 06:12 My weekly review habit
- 26:36 The unreasonable effectiveness of one-on-ones
- 39:08 Why I left Harvard early
- 41:16 College advice for people who are exactly like me
- 1:03:22 You don’t need to work on hard problems
- 1:23:12 Career choice: Evaluate opportunities, not just fields
- 1:37:17 Why and how to start a startup serving emerging markets
- 1:49:35 Where are All the Successful Rationalists? (Applied Divinity Studies)
Alexey: Hey, Ben.
Ben: Hi. How’s it going?
Alexey: Everything’s great. How are you?
Ben: Great. Excited.
00:35 How Ben spent his winter
Alexey: Cool. So how has your winter been? I remember this post from fall of 2019, where you mentioned that you installed this really, really, really bright lamp in your office. And that it almost, I don’t know whether it blinded you, but it made your room a lot brighter. So has your mood been better this winter?
Ben: This winter, I actually had a different strategy. So my partner and I decided to become nomadic because we were paying way too much for a crappy apartment near Harvard square, which is a reasonable decision when there are things going on in the Cambridge area and not a reasonable decision when there’s a pandemic and nothing is happening. I am currently in El Paso, Texas, which actually is a super awesome place at least to stay while you don’t want to spend any time with people. I have no friends in El Paso. If I would otherwise be interacting with friends, I might not like it as much, but anyway, I’m staying in El Paso, but I’m keeping East Coast hours so that I can continue working with the people that I worked with in Senegal and on the East coast, which means that I wake up slightly before sunrise and I go to bed slightly before sunset.
And that combined with our house, having a very large number of windows, sometimes in silly places, but it lets in a lot of light. It means that I actually haven’t, I haven’t been missing my ultra bright light that much. I actually, I did take the ultra bright light with me, but it turns out that it’s kind of hard to set it up, in Airbnbs, in a place where it both illuminates the room properly, but also isn’t blinding, but I keep hoping somebody will start a company that makes versions of this light that are actually good. And as in like, you can put them in a room and it’s easy to place them in a place such that daylight illuminate the room, but do not blind you, but nobody’s done it yet.
02:40 How being a CTO is different from being a software engineer in the context of productivity, weekly reviews, day-to-day happiness, and long-term satisfaction
Alexey: So the first thing that I wanted to ask you about, I guess it’s sort of continued, well, it’s directly continuation of the topic of productivity, but I guess it’s more direct, but one of the posts of yours that, your blog is just really good, but one of the posts in particular that I keep thinking about is the one on you managing to increase your productivity by 50% in one month, by noticing that you’re spending a lot of time on Slack and that you’re spending a lot of time doing random things during the day when they should be batched and other things like that. And this was written five years ago, right? How has this post aged?
Ben: I think that if my output were mainly code, it would still be pretty much accurate fortunately or unfortunately my output is no longer mostly code. My primary output is now a well-functioning engineering team and it’s a lot harder to figure out what actions are productive or unproductive, along that axis. For instance, if my primary output is technical work, then it makes a lot of sense to try to batch all of my meetings to particular times of day so that I can have as much time heads down programming as I want. But, if my output is a well-functioning engineering team, then actually the cost of delaying a meeting to a point where it might be bashed to other things could be quite high because it takes me. That means that I can’t use the information that I learned in that meeting for the rest of my day.
So it’s a lot harder to optimize, optimize productivity by just looking at RescueTime stats and being like, “How do I make these numbers go up than it was when I was mainly a programmer?” Also, RescueTime is much better at telling you, are you, or are you not programming? Like 70% of my time is communication and scheduling. What does that mean? I don’t know, RescueTime can’t even really tell… RescueTime. Yeah. If you’re in a meeting, but you’re looking at another document, right. It completely screws up your RescueTime statistics, for instance. But even if the statistics weren’t screwed up, I have problems: “Oh, some of my meetings are incredibly useful and some of them are useless and stuff”. It’s a lot harder to take a purely statistical view of my productivity, I think, than it used to be.
Alexey: Right. This sort of has the vibes of the whole Kegan’s thing of rationality and meta-rationality, where you try to just maximize the thing versus now everything is so hard you just try to get a taste of what’s important and try to do these things.
Ben: Yeah, actually, so maybe I would say the aged version of that post is my blog post on my weekly review habit, which has largely, I think weekly reviews have largely replaced RescueTime as the thing that tells me am I being productive as in, I can’t just look at statistics anymore after I spend a lot of time, staring at the ceiling and being like, “Was this a productive week? What did I actually do?” And RescueTime, can’t tell me that anymore. If there is a habit that has increased my productivity by 50% recently, it would definitely be weekly reviews. Gosh, I guess my subject matter hasn’t really changed in the past five years.
Alexey: How have your weekly reviews or? Well, there’s so many different directions, but how have your weekly reviews evolved since you started doing them?
Ben: Quite a lot. For a long time, they were mostly focused on habit improvements and system improvement stuff. And that… I think that type of weekly review actually synergized quite well with looking at RescueTime, because you can look at RescueTime and then be like, “Why was this number so high? What can I do to get it lower?” And it turns out that I should be keeping my phone in different rooms that it doesn’t distract me or something. I think that recently I’ve noticed that they have started taking a lot longer, I think largely because I spend less of them on that type of low hanging fruit, habit tweak and more of them just thinking about stuff. Where stuff is, or I don’t know, I guess a lot more of my job now involves solving problems where I can’t automatically apply a thing that I already know how to do. And so I just need to spend a lot of time thinking about, “Oh, we need a new manager for this team. Let me go through all of the candidates in my head and figure out, whether I think they would be good or bad.” Or “Oh, I need to improve this system at Wave or this process at Wave.” I’m not very good at designing processes yet. So it takes me a long time to think about it.
Anyway, I guess, I spend a lot less of my time in weekly reviews now, brainstorming random habit improvements and, the format has shifted more towards, I write down a couple of things that are on my mind and then I go for a really long walk and think about them. I think theoretically if there’s enough time in my schedule that I could devote, I could spend time during the week, thinking about stuff also. And in fact, I do, but somehow… I guess maybe this is a similar phenomenon to what happens in engineering, but somehow having time to think on the weekend makes my brain less likely to slide off of a difficult problem and go, “Oh, what am I going to say in the meeting that’s happening in four hours or something like that.”
Alexey: I definitely feel very similarly about weekly, thinking during the weekends versus thinking during the weekdays. Yeah. I think for me the version of this, I guess is that I do my weekly reviews on Sunday and then Sunday, I’m just explicitly prohibited from doing any work on the main project that I’m working on. So there’s all of that because I know all of the things that cannot possibly be as important as this weekly review, none of them are urgent. It makes focusing on these things easier. I feel like there’s a pretty strong connection to how are you thinking that in the shower for me is much easier because I know that if I come in 30 minutes there, I just, I know that I’m going to be able to just think only about the thing, the problem at hand and not be distracted by anything else. Sundays kind of work like this as well.
Ben: It’s funny. I posted an essay or I wrote an essay recently where I mentioned having my best ideas in the shower or something. And I posted on Twitter and when one of the replies was like, “Oh, have you ever considered taking baths instead?” And it was a joke, but also I hadn’t considered it. And I feel like I probably should have considered it. I actually, I didn’t start taking baths, but I did start taking a lot more walks. And I think that was pretty good for getting more time, during the week to think about tricky things. Also, another reason that having mountains in your backyard is great. It makes taking walks a lot more enjoyable.
Alexey: I guess I don’t need to reference one of your old the posts in every question, but you’ve always been an engineer and you’ve always been doing technical work and let, I guess hands-on technical work and you’ve transitioned to being a manager over the last, I guess, couple of years. And how has this transition been working for you? Did they expect that you are, are you enjoying it as much as being a software engineer?. Did they expect when you were graduating college, that this is how things would turn out?
Ben: I definitely did not expect it to turn out this way. I think in terms of enjoyment, I would say that… So in happiness studies people will often differently measure day-to-day happiness versus life satisfaction. I would say that my life satisfaction is much higher, but my day-to-day happiness has probably gone down somewhat… I don’t know. It’s really hard to beat the dopamine fueled feedback loop haze of writing a bunch of code.
Alexey: And instead you need to think about hard problems that do not have defined solutions all the time, rather than being into this loop of coding.
Ben: However, I don’t know. Overall I think the impact that I’m having on the world is much higher because I’m working, I’m building the team that builds the thing, instead of just building the thing I have, there’s a big force multiplier from doing that and doing it effectively. And that means that, when I look back on any given period of time, I’m like, “Wow, Wave really did a bunch of awesome stuff.” And I’m really proud of that. And, to be clear, I was proud of my, my programming work for Wave too, but… I don’t know, being able to see the effects of, having an entire team of people that’s amazing and being able, I actually, I’ve been recently thinking about this a lot because it was performance review season at Wave. And one of the funny things that I realized about, we explicitly prioritize hiring people who are very excited about our mission, but also very low ego. And this means that one of the most frequent pieces of feedback that I give during performance reviews is to tell people that they they should have more confidence in their own ability, which is just, obviously this has downsides, right? If you hire a bunch of people who are extremely competent and have very little ego, this is basically a recipe for people to have imposter syndrome because they see everyone around them, who’s incredibly competent. They don’t realize that they are equally competent. And in fact, in some sense a downside of having this value, but it does make for performance reviews, just actually a really joyful time, because I can just feel like my job is just to tell people that they’re more awesome than they think.
Anyway, being able to see this entire team of people become more awesome over time is quite satisfying. Even if, day-to-day, it involves a lot of being in meetings. I guess I would also say, I think the one of the most striking things about the transition between programming and being a manager is that there was a much stronger dip in my, day-to-day happiness initially, which then corrected over time, which largely, because when I first started being a manager, I did not have any instinct at all for which activities am I doing today that will actually be valuable because this is the thing that people commonly say about management versus programming is that your feedback loop goes from like hours from getting your tests to pass or days for getting something into production. It goes from hours or days to weeks, months, and years. Initially I was just like, “Ah, I have no idea whether anything I’m doing is useful.” I feel like I’m flailing around in it and it’s pointless. But as I gained more experience and saw those initial decisions play out over the full, feedback loop time horizon, I ended up getting a much better intuition for, “Oh, this meeting where I gave somebody feedback that they found really useful.” this actually was an incredibly productive management activity and it’s going to make them slightly more effective over the course of the next six months. And so I can sort of back propagate the dopamine hits from six months down to a shorter timescale. I guess there have actually been multiple of these transitions where basically every time, your time horizon lengthens, you end up feeling that you’re flailing around for a little while until a period of n time horizons passes. And then you start to figure out and be able to predict the impact of your decisions that you make in the moment.
17:56 On being a good manager and channeling one’s neuroticism productively
Alexey: So my last post was about management and how people who we now think of great leaders, we’re not necessarily good leaders when they started out and even Musk used to just have no idea what to do with management. Steve jobs ended up being, or I guess he ended up not being the CEO. And then the guy who had a ton of experience came in to Apple pretty, pretty early, and then hired everyone. And then it took Steve jobs 10 or 15 years to actually learn how to manage people. So your transition is interesting to me in light of all of this. And again, from the perspective for someone technical, but shifting into management job and feeling like “I’m flailing around and have no idea what to do” and then “No, it actually makes sense.” And it says that I’m getting pretty good at it and it’s actually satisfying.
Ben: I think many of the people that you mentioned in this post and also some other managers that I of heard and I guess like myself is specifically being interpersonally difficult to work with and learning how to… specifically the reason that they’re a bad manager is just that they’re kind of a jerk, right? Like Steve jobs is the classic example of this…
I think Elon Musk, I’ll say is kind of famously difficult to be managed by because he can be a jerk or at least was at various points. I haven’t, I have not been extensively following people’s reviews of his management. Anyway. I guess this makes me curious, maybe the underlying traits, these people are very neurotic, I think, and learning how to, what they do over time is they learn how to productively channel their neuroticism. I guess I wouldn’t compare myself or you don’t want to my overall abilities to Steve jobs or Elon Musk obviously. But I do think this has been my number one challenge with learning how to be, I guess, considered an okay manager by my reports and which I think was also true for Drew the CEO of Wave. I have loved working with him as a manager, but, one thing that we’ve struggled with that I also see and saw in that post is basically people learning how to, use their neuroticism productively rather than just flipping a shit at everyone.
Alexey: Right. What are the aspects of being a manager, do you feel, you still do not have a good grip on?
Ben: Oh, tons of, I still am far from perfect at being neurotic, like productively instead of non productively. As a programmer, I am very fast at reading an error message and being like, “Oh, I bet I know what is causing this.” And sometimes other engineers that I work with are, not as fast, perhaps because they have not been practicing that skill for six years or whatever. And so sometimes people will ask for my help for something. I think it’s like clear from the air. They’re like, “Oh, how do I fix this error message?” And to me it seems clear from the error message, what’s going on. But to them, it’s not because they’re less good at interpreting arcane error messages than I am, which is, understandable.
When I was a younger programmer and a very bad manager… I would often go… Well, I was going to say, I would often get annoyed. I think that’s a little bit of a strong description, but I would often be momentarily annoyed. I’d be like, “Oh, why can’t everybody interpret these error messages.” But it turns out that’s not a productive thing to be neurotic about.
If I just bugged everybody, whenever they asked for help on an error message that I think they should have been able to interpret correctly. That’s not really productive. It just is going to make them annoyed, because there are more important things to coach them on. However, on the flip side, if I never coached anyone on how to, if I just let it go anytime somebody did something that I thought they could have done better at, I wouldn’t be a good coach for them. I wouldn’t be helping them improve.
And say, figuring out the balance of… That’s what I mean by productively channeling your neuroticism is, I guess, part of the reason that I have become a good engineer is that I pay attention to those things when I’m doing them and get really annoyed when I do them wrong, but I can’t be as annoyed at everybody else’s like I am at myself. And so figuring out the balance of, when to actually give people feedback on something and when to just be like, “This, isn’t the most important thing to talk about.” This I think one of the things that I’ve been learning a lot about, and I’m still learning, and to be clear, it’s not just the axis that you need to work on is not just like, how often do you give people improvement feedback? There’s a second axis to just, how often do you give people positive feedback, which for many high neuroticism people, the answer is almost never, right? Because you doing things right as, yeah. That’s what I’m paying you for. You wouldn’t be here if you didn’t do most things.
But if you only tell people when they do something wrong, then they have a bias, you’re giving them a bias sample of what you think of them overall. And so that’s something that I’ve been trying to improve at recently. Anyway. Yeah.
26:19 On doing 1:1s with Ben’s partner vs doing 1:1s at work
Alexey: you’ve asked not to be compared with Steve jobs and Elon Musk, but I have to know that your example is remarkably similar to the example Elon Musk gave when he was asked about his personal struggles of being a manager, noticing a technical problem earlier than someone who he was a manager too. And he just learned to let people solve the problems and guide them rather than just blurt out the solution, because it turns out that there’s no way to learn if you just do this.
So you have this transition from being a software engineer to being a manager. But one of the things that I think you’ve been doing for a while that is usually associated with the being a manager is conducting one-on-ones. You had this post which I also think a lot about doing one-on-ones with your partner. And you mentioned that you notice that grad school is just terribly structured, totally not optimized for actually making people feel good and accomplish things. And that just by starting to conduct weekly one-on-ones with your partner about her thesis, you helped her a lot with it. And this, did the things that you do in the process doing one-on-ones with your partner, carry over to the things they ended up doing as a manager.
Ben: I think it’s been… That’s a good question. I think that my partner doesn’t have the same type of personality as many people that we hire to wave as software engineers. Some of them have definitely been the same, right. Part of one-on-ones is just getting more comfortable talking to the other person so that you can have difficult discussions together. That happens with everybody. But for instance, my partner was pretty shy when we started dating, one of the most useful, questions that I could ask her in one-on-ones is like, “Hey, so have you sent that email to that person yet?” And she’d be like, “No, I should probably do that.”
Ben: And I think compared to the typical wave employee, she’s probably 95th to 99th percentile shyness or something. So I spend a lot less time, nudging other wave employees to send people Slack messages. Not zero time, but a lot less. I think also the one-on-ones that we had while she was working on her thesis were actually closer to a status meeting than some business focused one-on-ones as in she would just, we’d just talk about her thesis. One of the rules that people often give you for, one-on-ones in a company context is you should actually not talk about current projects because you should have other meetings for that. It should be your time for your report to talk about whatever they want. In relationships that’s kind of flipped. Most of the time you’re talking about whatever you want. I don’t think they were, I guess they were, they were fairly different on that axis.
Alexey: Right. So your partner finished her PhD and I think you mentioned on Twitter that she is now, I guess, changing careers.
Alexey: Through this transition, did you continue doing one-on-ones? Have they been as helpful? If yes?
Ben: Yeah. So currently we are actually spending less time in one-on-ones focusing on her career plans because we realized that basically… We we’re generally quite different in various ways. And for a while, I was the only person whom Eve was talking to about this. And I guess it was hard for me to do the correct amount of stepping back and not being opinionated basically. Because we were at such different career stages, where Eve is like trying to figure out which among lots and lots of different possible options to do. Basically, I think figuring that out requires a lot of introspection that probably shouldn’t be biased by what one particular person will think of the outcome. I think it was hard for me not to bias her if we talked about it too much or at too high of a frequency. And so, we actually stepped back from talking too much about that in our one-on-ones.
My basic rule is, I shouldn’t be talking to her about it disproportionately more than other people, I think, because there are ways in which she’s quite different from me and it’s hard for me to simulate or imagine or give good advice taking those into account. There might be other people that are better at giving advice or better role models for her. As you mean, she does not want to become like the CTO of a software company and which I don’t think she does.
Alexey: Yeah, I feel like I have something similar with my wife where I think, where we’ll also clearly… It seems to be enjoying legit different kinds of work.
Ben: Mm-hmm (affirmative).
Alexey: And it’s like actually difficult. It’s surprisingly difficult to empathise with that and with the thing that I’ve been encouraging. I have this thing where I’ve been sending people cold emails or I’ve been hanging out on the internet since I was like 12 and I’ve been sending people cold, like really poor cold emails but still cold emails since I was like 16 and I have this habit where I was just like, if I see someone interesting on the internet, if I see someone who I want to meet, I can just email them and there’s a reasonable chance that they actually want to talk to me. And one of the things that I now encourage, try to encourage Nastya more is to just, is there a person who you wanted to meet? Who you would enjoy talking to? Who you might want to share career advice or who might be helpful to you? Whom you just want to become friends with? And let’s just sit down and write to that person because they might actually be helpful with whatever it is you’re trying to do, which I probably have absolutely no insights with. We probably should have started doing this later.
Ben: You mean, you should have started doing the… Basically, the meta level of, “Oh talk to other people”, instead of just trying to talk it through yourselves. Yeah, I strongly agree with this. I think if you try to give your partner to, or at least for me when I was trying to give Eve too much explicit advice, I ended up giving her advice about how to become a mini version of myself instead of the best version of her own self. And so, I’m sure that I also do that in work one-on-ones, right? But in work one-on-ones, it makes more sense to do that at least to some extent because most of my reports are trying to become more effective Software Engineers and/or managers.
Ben: Whereas Eva’s not, so that’s another relevant dis-analogy, I guess.
Alexey: Yeah. So Eve finished her PhD in like philosophy and it took quite a few years, do you think it was… Given the fact that she now is transitioning to a different career, do you think it was a mistake to finish the PhD?
Ben: I’m guessing that… So at the time that I actually tried or I raised the possibility to Eve that she should drop out. I think a couple of years before she finished, or maybe it was only one year before she finished. At that point, I think that her take was that… Basically it was worth finishing because there were only a couple of years left. But if she had known when she was looking for PhDs, what it would entail? What she basically… What she knew now that it like wouldn’t have been worth it to finished.
So, I think the reason that it was worth finishing at that point was… So if you don’t care about going into academia, getting a PhD, at least in philosophy its actually a pretty sweet job and if you have a low cost of living. You don’t have that high of a workload. You make a reasonable around the median income depending on where you are and you can hang out with smart people a lot. And so your hourly rate, if you’re not basically putting every waking hour into your dissertation or succeeding at the job market is actually pretty good.
The main cost of doing a PhD, I think the main cost for Eve for doing a PhD for seven years or whatever, I think it was six years of actually being in school and she took one gap year but the main cost was not being able to iterate on finding a thing that she was more excited about doing permanently for those six years. And that was the fault of academia not preparing you for anything except to academia. I guess, yeah. I don’t think it was a terrible decision given what she knew when she was 21 or whatever. I think in an ideal world, she would have known a lot more when she was 21.
And then, at the point at which she actually decided based on good reasoning and knowing a lot more about what a PhD was like on the job market, she was like, she could drop out at that point but she was quite close to finishing her dissertation. It was still a good job and having the credential is plausibly helpful for other things. If it takes you one year to get the credential because of a bunch of sunk cost that you already have instead of six years, then I don’t know, maybe it actually is worth it. And it certainly got her parents off her back. Her dad is a philosophy professor, so…
39:08 How finishing undergrad in 3 years worked out for Ben and what college is for
Alexey: Yeah. So you finished undergrad in three years and you were so impatient with college that you’re like literally forego the last year of undergrad. And so, was this a good decision in retrospect and how did this happen?
Ben: So I think this decision was clearly great ex post. If I had made a different decision, I probably would not have ended up working at Wave. I think I’m unusually good fit working at Wave. I think I was an unusually good fit working at Wave and it’s allowed me to have a lot bigger impact than I would at other randomly selected jobs I could have ended up at at the same time.
So yeah, ex post, I think it was a great decision. I think if I had been more effective at using college – I don’t think I was very effective at using college – if I had been more effective, it might not have been the right decision. I didn’t have a great idea about what college was supposed to be for. I still don’t have a super great idea about what college is supposed to be for, but my guess is that the best way to use it is to make a lot of really smart friends. And I think I was okay at that, but not great.
And if I had been great at it, then maybe it would have been worth the extra year to have more bonding time and kind of hosting a synchronized schedule with them or something. But yeah, I don’t know. It’s hard to say. I haven’t spent a lot of time thinking about… Well, I guess I spent enough time thinking about what I should have done in college to write a blog post about it, but without having actually having used it effectively. All of this is kind of guesswork.
Ben: So, I’m not sure.
Alexey: Yeah. This was sort of the thing that I was also thinking about. My sort of impression of the college is to have the ability to explore your interests by taking any nature of science and any other course you want and do you want to do Physics? Or do you want to do Math? Or do you want to do Chemistry? Do you just want to do Computer Science? And to meet people who you will hopefully interact for and talk to for the rest of your life. And one of the things that I think you probably did very right was… I think you were the person who started EA chapter at Harvard, right?
Alexey: Or you were one of the people who could do this.
Alexey: In this way you collected a bunch of people who are really smart and really aligned with your point of view, in contrast to the median person in Harvard. So, do you have ideas on what could you’ve done better to meet more, really smart people?
Ben: I actually think Harvard specifically is quite bad for this. For a weirdly specific reason which is their housing policy. So the housing works is that you get randomly assigned to one of the, I think, 12 upperclassmen dormitories during your freshman year. You can choose up to seven classmates, to be randomly assigned in a block to the same dormitory. Your blocking group is the only set of people that you are guaranteed to live nearby for the next three years and you have to pick them sort of like midway through your freshman year while you’re also…
So that means, in practice what this means is that around, I forgot of the exact time that, maybe December through like February of your freshman year, you’re just running around in a panic, trying to figure out who are the people that you want to be best friends with for the next three years? While simultaneously trying to pass your ridiculously hard courses that you’re taking because this is your one chance to take a bunch of ridiculously hard courses that they offer for freshmen. Anyway, it’s not a very fun process and it means that you make this one incredibly high stakes decision that basically determines your friend group for the next three years.
Alexey: That doesn’t quite some true because you can hang out with the people outside of your living block.
Ben: I think this might be another way that Harvard is worse than many other places is that people tend to be very busy and the thing you got from people in your dorm that you don’t get elsewhere is the ability to spontaneously hang out.
With other people, you have to coordinate it. You have to coordinate with them. And it’s kind of annoying because they have 13 different extracurriculars and they’re too busy being captain of the Quidditch team to hang out with you or something. Anyway, I think this is also maybe a cultural thing that is somewhat specific to Harvard, but even so, the dorms were kind of spread out and it’s kind of annoying at least at Harvard. And my sense is college students generally are very lazy and don’t really want to travel, to have lunch with you when they could spontaneously hang out with the people in their dorm instead. In fact, Harvard students are sufficiently lazy that it’s generally viewed as a disaster if you got randomly allocated to one of the three dorms that’s like a 10 minute walk from campus instead of a two minute walk from campus.
Yeah. It doesn’t completely determine your social circle, but it does lock you in quite a bit. It’s not guaranteed that you’ll find anyone in your dorm. In fact, pretty unlikely that you’ll find anyone in your dorms to be roommates with you who’s better fit with you than the people that you walk with.
The other thing that’s annoying about this system is that because blocking groups are randomly assigned to dorms, dorms don’t get to have their own personalities. And so, it’s much harder to navigate yourself to a group of larger than eight people that shares a bunch of interests or ways of thinking about stuff with you. And instead, they all sort of have this bland standard Harvard culture. I didn’t expect when I was picking colleges that housing policy would have such a large effect on the environment, but in retrospect, that might mean, maybe that would have been something to keep in mind.
Sorry. I feel like you maybe asked and there was an original question that I’m not sure I completely answered because Harvard’s housing policy is one of my hobby horses and I kind of went off on a tangent.
47:51 Is college a good place to explore your interests?
Alexey: The original question was, what could you have done better? One of the things that I personally discovered, unfortunately way too late was – in Russia, we enter a specific program where I entered the program in Mathematics and Economics and I couldn’t really take any course I want or I had a specific program. One of the things that I discovered during my third year was that I could actually take any course I want. I just have to go to the study office and they just kept lying to me over the years that I couldn’t do this. Then at some point I discovered that I could do this and then I just started going to the Computer Science department and hanging out there much more. In this way, I met people who, I think, I should have hung out much more than my first and second year and the first half of third year.
I mentioned that a version of this for you could’ve been that, instead of just going to classes at Harvard, just spend a day at MIT and hang out there and meet people there.
Ben: Yeah, definitely. I don’t know how that particular plan would have worked because I think geographical proximity is so important, but I think the most effective thing that I could have done on that axis was just choose somewhere with a better housing policy. I think, in general, being better informed about… I’m not sure to what extent this is something I could have done better or not but I was super poorly informed about how the world works when I was choosing schools and, if I had been better informed, I could’ve made much better decisions.
One way that I think was accessible to me in which I could have been better informed was that I could have spent more time doing things that weren’t college before going to college. This is related to another point that you made or it sounded like you thought that, at least in the US, where there was not this restriction on what you can do, right, college is like a good place to explore your interests.
I actually think college is a pretty garbage place to explore your interests because you find out what it’s like to take classes about your interests, but that’s not actually what your interests are going to look like for the rest of your life. Unless you’re a grad student or something, unless you want to go into academia. But if you enjoy taking physics classes, okay, what does that mean? It means maybe you should go to physics grad school. What else does it mean? You don’t know because you’re a college student, right? What things can you do in the real world that use the same skills as doing physics? I don’t know. I would argue maybe actually, one of the things is maybe site reliability engineering and technology company, but you’re not going to know that because nobody’s going to tell you that who hasn’t actually done that and told you like, “Oh yeah, it’s interesting like physics in certain ways.”
Alexey: Right. No, these are actually really good points because you learn what things you enjoy studying. But not the things that you would enjoy doing as your career.
Ben: Yeah. So I think that if you didn’t care at all about the signaling benefits from a college degree, I think it would make much more sense to not front load all of your studying about things. And then all of your trying to actually do things. And then backload all of your trying to actually do things. Because there’s a feedback loop between these two things where by trying to do things in the world you learn what are your knowledge gaps that you then want to go fix by studying things or something.
The only way in which this is socially allowed is by taking a gap year. I do think that taking a gap year, at least for everyone I knew who took a gap year between high school and college, they enjoyed it. I do think that the set of things you can learn about while you’re taking a gap year is pretty limited. If you happen to know how to code, then you can get a programming job and that will teach you a lot about what programming is like. I don’t really know what else you can do. I know some other people sometimes spend their gap year hiking around Chile, which is really fun, but nobody will teach you that much about what you might enjoy doing for the rest of your life. So I feel like this is something where I don’t have a great sense of what should replace the thing that people normally do, but it seems like it would… interesting to think about how to optimize.
53:03 What Ben learned from starting the Harvard EA Chapter
Alexey: Right. Yeah. How has starting the EA chapter at Harvard affected your life? How would your life be different if you had not done this?
Ben: Good question. I think it’s hard to say to what degree any of the effects have been like persistent, other than, I don’t know, I got a bunch of people interested in Effective Altruism. Otherwise, they wouldn’t have been and they’re off doing cool things. And say maybe one of the ways in which it’s affected my life is maybe it caused a very tiny reduction in the chance that I get turned into paperclips by an unaligned artificial intelligence or something like that.
But in terms of the more noticeable, personal effects, I think it’s kind of hard to say. I think there were a couple of ways in which it was an interesting experience that I can claim, maybe they affected me later, but it’s unclear to what extent that’s true. One is, I was doing something that I was very clearly not the best person to do, but I was there and nobody else was. I don’t think I was particularly good. I wasn’t good at logistics. I wasn’t good at remembering to create Facebook events for speakers before the day of the speaker coming. I wasn’t good at giving introductions to the speakers that we had. I wasn’t good at coming up with events, but whatever. I was the person who was there and motivated.
And so I guess if I took a generalizable lesson away from that, it was like, “Oh, even if this thing feels like it kind of sucks, if you keep putting one foot in front of the other, you’ll eventually at least become passable at it. And then, hopefully, you can hand it off to people who are more than passable at it” which is what happened at Harvard Effective Altruism and it’s now run by people who are far more competent than I ever was. Thank God. Because it would definitely be dead if it was people of equivalent competence running it for the next seven years. Anyway, I think this is actually remarkably similar to what happens in startups.
And I guess one specific thing that I actually was even more remarkably similar to – was a lot of what I did was I sent a bunch of cold emails to people asking them if they wanted to give talks. Because it turned out that having people give talks was our best way to get new people interested in Harvard Effective Altruism.
And what this meant was that I would just find large numbers of people who seem even possibly interesting and I would just crank out cold emails to those people. 90% of those cold emails would get no response or the person wouldn’t be interested and 10% of them, the persons would come give a talk and we’d get some new members. It turns out this is exactly what hiring for a company looks like. You do the same thing over and over again and 90% of the time it fails and it sucks and it’s super fricking boring. Then 10% of the time you like hire an amazing person that you love working with. So I guess that was a good object lesson in what it feels like to sample from a heavy-tailed distribution. And then it turns out it’s just like being willing to draw many samples from a heavy-tailed distribution. It’s like a good thing to have. Also, writing blog posts. That’s another thing that’s like very similar to that.
Alexey: Yeah. A couple of blog posts generate half of all page views over the years.
57:08 Where do people who you talk to these days come from?
Alexey: You went to Harvard which is the single most prestigious university in the world, you went to a pretty good high school in Boston which was also like academically demanding. Then you’re hanging out a lot with the Effective Altruism community and it seems that all of these groups are very selected on just being really smart. They’re variously selected for interests aligned to yours, but these days, do you talk to a lot of people from your high school? What’s it like for being smart, but not for interest alignment? Do you talk with a lot of people you went with to Harvard? Or do most people you talk to, like your coworkers and people who are involved with EA or you met hanging out on Less Wrong?
Ben: I think the people that I talk to the most are from a pretty random selection of, or a pretty random set of those things. My high school was not very large. So my class is 35 people of whom I’m still in touch with one-ish. Harvard, I’m still in touch with a few people from Harvard. A few people that I’ve met from like Less Wrong, a few people that I’ve met from like other things. A few people that I met through the Contra dancing community… I don’t think any community that I’ve participated in was an excellent filter for people that I wanted to stay in touch with… Which is interesting to think about.
I think interacting with people on the internet or interesting people on the internet is probably the one thing that seems like the best filter but it’s also not clear to what extent, anything other than the interaction itself is doing the filtering. Maybe I, there are lots of people that I like interact with once on the internet and most of them I don’t want to interact with again. But I do think randos on the internet is like the best source of like interesting people.
But specifically it’s not Less Wrong or anything like that. Just the internet in general.
Alexey: Right. Yeah. This to me is extremely interesting because first, I feel like it had a similar experience. But second, because at some level, it seems bizarre that there has not been such filtering invented by someone with interests seem like style of general attitude towards to life and thinking styles that would filter for these kinds of things and instead there are sort of these kinds of filters and these kinds of filters and random people on the internet.
Ben: I don’t know. I think it’s interesting to think about why this hasn’t happened. I don’t think you could put together a 6,000-person university of weird people from the internet. Maybe, but I think you’d have to get all the weird people from the internet. Sorry, specifically 6,000 people between the ages of 18 and 22. So, the type of filtering that Harvard needs to do… Yeah. I don’t think they could. You could imagine, I think a high school could do a better job of this, but then it’s also in a specific city and nobody’s going to move to go to a high school. So, then you have a much smaller pool of people that you’re drawing from.
Maybe, now that it’s possible to do so many more things over the internet we’ll see this change with more niche groups. Also, what Harvard wants to filter for, is people who will be very legibly successful and therefore, improve their brand. That’s how they continue being the most prestigious university in the world. It would be a big risk, at least, for them if they started filtering for like weirdos on the internet instead.
And then, if you’re a weirdo from the internet, symmetrically, by not going to Harvard and instead going to weirdo-from-the-internet university, you’re also taking a big risk because you’re forgoing the number one most prestigious thing. So, it seems like there’s some sort of equilibrium that’s quite hard to break, at least in universities specifically.
1:02:46 Interning at Jane Street and the (non-)importance of solving hard technical problems for making an impact
Alexey: So, the next broad area of things I wanted to talk about was your professional career. And we sort of started with this, but you worked at like a bunch of places and then you interned at a bunch of places–
Ben: Interned at a bunch of places. I worked at one other place that wasn’t Wave, but yes.
Alexey: So the first internship that you had, or like one of the first was Jane Street. And you wrote in one of your blog posts that at the time you thought that the most satisfying thing to do is to work on hard technical problems and they’d give you some like really hard integrals to solve. And afterwards, you’re like, “Well probably just solving the most difficult technical problems is not the way to optimize for in the job.” But the question that I’m coming towards is: what exactly happened? Do you remember your thought process during your job of first realizing that there are probably like a lot of interns there who were very technically capable. And there were other people do work at Jane Street and at other firms like Jane Street and they do end up just choosing this as a career and they do end up enjoying this job. And what was the difference in how you thought about your career that made you decide not to choose it? Because there are like good EA reasons to get a job at Jane Street and a lot of people in the EA community as well get such jobs.
Ben: Yeah. Okay. So I do think that if you want to make a lot of money, one of the most effective ways to do that is to solve hard technical problems that nobody else can solve. There, the relationship is clear. Like, you have a very rare skill, which is like you can like solve integrals that other people are like, “Oh, you can only do this numerically” or whatever. So, I think if you want to make a lot of money, for instance, because you like money or because you want to spend it all on bed nets or whatever, I think solving hard technical problems is a pretty good way to do that. I don’t know if it’s the best way, but certainly you’ll make a lot of money with like pretty low variance. That seems pretty good. Assuming you have the ability to solve the hard technical problems.
I feel like when I wrote the blog post which I think you’re obliquely referencing, which is “You don’t need to work on hard problems”, I mainly meant if you want to have a big impact on the world. I think the thing that I assumed when I was interning at James Street and actually kept assuming for quite a while after that was, basically I was over-applying the efficient market hypothesis or something. I was like, “It can’t possibly be the case that you can just write this app that like subtracts a number from one row in your database, and adds a number to another row in your database and it turns out this is like insanely useful to millions of people.” Actually it turns out that you can. This is like what Wave does. But I didn’t really realize that it was actually possible to build things that had a really large, positive impact on the world without solving some hard problem that nobody else could solve. And I didn’t actually realize that until after I started working at Wave or I didn’t fully realize it until after I started working at Wave. Whereas before then I was like, “Obviously somebody would have picked up that $500 million bill lying on the street.” Yeah, and that was in fact false.
Alexey: Yeah. I feel like this fits perfectly into my everlasting crusade against efficient market hypothesis being applied to your personal life as an individual. I feel like a similar thing sort of happened to me and I feel like a similar thing can happen to a lot of analytically minded people. Like, efficient market hypothesis makes so much sense, it must be like really true in life where–
Ben: And it’s so powerful. Somebody can ask you any question and you can be like, “What does the efficient market hypothesis have to say about this?” And it has something interesting to say, but it turns out, in fact it also is often wrong because the market is in fact not efficient and like–
Alexey: I feel like the explanation that I think is most likely for that is that like one crucial assumption of EMH is that there is enough capital in hands of smart investors with good information and that they can correct prices by using this capital. And one problem with real life is that there’s just a lot of problems and not enough people, like way not enough people to actually work on picking up all of the fruits and–
Ben: Also, transaction costs, right? There are like incredibly high transaction costs.
Alexey: And the transaction costs, yeah. And sort of the thing that I have, is that the number of problems increases at a higher rate than the number of people who could be working on these problems. But this is as speculative as the exact extent to which EMH actually applies to anything.
Ben: So you studied economics, maybe you know about this. I had been curious whether there have been any attempts made to, I guess, quantify what happens, even in like toy models sort of when you break assumptions of the EMH? Like how inefficient should I expect the market to be, if I don’t know, like transaction costs are on the order of the price of the item? Which I feel is approximately true for, I don’t know, buying things off Amazon. I spend like approximately as large an amount of time doing research and trying to understand what the thing fricking is, as I do the money that I spend, like actually buying the thing. And so how inefficient should I expect the market to be like, based on this? Or if there’s like a bunch of dumb money, how much should I expect, given the amount of dumb money relative to smart money to start the market? Does this research exist? Because if it does, I could just spend all day telling my efficient market hypothesis partisan friends that they’re wrong with it. Instead of just only having anecdata.
Alexey: Right. Yeah, I’m not really aware of such kind of research, but my intuition is that once you introduce significant transaction costs, everything just breaks down. But my intuition is biased against economic models being robust to assumptions you choose. So I might be wrong here.
Ben: I think you’re right that it breaks down. So your intuition is that it breaks down very quickly, as soon as the thing is like, the transaction costs are present at all.
Ben: Yeah. At one point I considered trying to write a blog post in which I actually built a toy model, in which like things traded good or something, but I’m not good enough at figuring out like the right framework for a model such that I thought it would actually tell me anything interesting.
Alexey: Yeah. Maybe someone listening or viewing this video will end up coding this up and telling us the result.
Ben: Or just put it on paper. And I feel like some enterprising economist must have been like, Oh, let me investigate the limitations of this idea that everyone in my profession believes.
Alexey: That’s exactly the type of reasoning that does not hold. Yeah. Okay. So this was Jane Street and then you interned at GiveWell, right?
1:13:03 Interning at GiveWell, trying to be a researcher vs being a programmer
Alexey: How was interning at GiveWell and why didn’t you stay there?
Ben: I think that I wasn’t very good at doing research, at least relative to how good I was at, for instance, programming. I think the things that I did at GiveWell were like, I forgot what their term was for it, but basically like an investigation into whether interventions to promote breastfeeding in developing countries could plausibly be highly effective. So for background, breastfeeding instead of giving your children, like baby formula, is expected to have various benefits for the child. I think in particular, it’s expected to be immunoprotective because there are various things in breast milk that it’s hard to put in formula. But the prevalence of breastfeeding in developing countries is very low because of extensive marketing campaigns by infant formula suppliers, like Nestle or whatever.
Anyway, so I wrote this report. I found it very exhausting because of the percentage of my time that was devoted to answering the question of, “Oh, what is the next most important thing to look at?” versus the percentage of my time that was devoted to actually doing things, which was like very high. It was kind of difficult for me to do it. It wasn’t very conducive to like deep work, having to always run in the back of your head. Is this the most useful thing that I could be reading right now to answer my next question? The quality of the studies was not great. So it was just like a super fuzzy question. The criteria, by which one might even answer it were not clear. And I ended up writing a report and the conclusion of the report was like maybe.
Yeah, I don’t really know what got done with that information, if anything. But anyway, I felt like if I were programming in the equivalent amount of time, I could have, I don’t know, built a cool app that people would use. And I would’ve known almost all the time, like what is the next most important thing for me to be working on? So, I don’t think this was like, research is more demoralizing than like programming and stuff, like abstract sense, but maybe if I were an equivalently good researcher as I was a programmer, I would have spent that entire month in a flow state. And it would’ve been like, yeah, this is super awesome. But I don’t think that I was a very good researcher and I was like a very good programmer, so I decided that it wasn’t my competitive advantage or something.
I did still get a lot out of my internship, in particular, mostly from hanging out with Holden who I guess became one of my general role models for like how to do a good job of thinking about things. And I guess one that’s sort of vague, one particular way in which it was very useful to have spent time interacting with him was that, I think he was like a good antidote to many of the memes that were going around the effective altruism community at that time that were the result of taking very simple models way too seriously.
I think Holden was very good at basically taking these ideas and being like, “Wait, this seems ludicrous, I don’t believe it,” just based on common sense, but then actually figuring out the actual case for not believing supposedly the ludicrous seeming thing that the effective altruism community was very excited about. So I think, for instance, that was a time during which like 80,000 Hours was really pushing the idea of replaceability that is, when taking a job, you should consider the counterfactual, like who would have the job instead of you? If that person would do an equally good job to you, then maybe it doesn’t actually make any counterfactually adjusted impact whether or not you take the job. It turns out that if you look at a really simple model, then it’s true that consideration is important. But then if you layer in more and more things about like how the real world works, that are not accounted for by the very simple model, then it suddenly becomes like absurdly unclear to what extent any job is actually replaceable or not.
So it turns out that most of the time, replaceability should not be one of the primary things that you’re thinking about when you choose a career. And to 80,000 Hours' credit, by the way, they updated, they stopped emphasizing that consideration nearly as much and it’s not really a core part of their messaging anymore, but I think they agree in retrospect that it was a mistake to focus on it so much. But it was like contrarian, clever idea that is good for getting press, but it’s not actually the right way to think about the problem.
Alexey: This is kind of like the story with earning to give.
Ben: Yeah. Replaceability is one of the [reasons why] you might think that earning to give is like much better than various other things, right? Because when you donate money, it’s clear that whoever would have that job in your place wouldn’t be donating that money. And therefore you’re like, your donations are not replaceable. It is a really robust case that you’re having a positive counterfactually adjusted impact when you donate the money compared to, if you try to do a job that is like being a doctor, where you like directly save lives or something.
Yeah. Anyway, so I think that there were a lot of this type of like memes flying around the effective altruism community. Basically, because it was the early days, people hadn’t thought deeply about what’s the actual right way to think about these things yet. I think this is actually similar to the efficient market hypothesis. This is also the time when the EA community was really obsessed with the efficient market hypothesis. This is again, one of those things where it’s like really easy to say contrarian things that seem vaguely like they might be true by applying EMH. And then it turns out they’re not actually the most important things to say, but you can quickly get immediate traction by saying them.
Anyway, I think Holden was good at being immune to those things and that was really useful.
1:21:07 Trying to start a “startup”
Alexey: Yeah. Cool. So then I think at some point you tried to start a startup or something?
Ben: Oh Gosh.
Alexey: It does not count?
Ben: I would say trying is a generous term. I participated in something. Startup is also a generous term. Yeah. I guess that sounds silly enough that maybe I should describe what actually happened; but like Harvard had a startup competition and I, and some friends entered the competition and won like second place or something. And so one of my summers at Harvard, I decided to try to work on this thing even more, work on this, like startup, full-time, this pseudo startup. It had no customers, it never got to the point where it was even plausible for it to get customers. I don’t know. We were very bad at doing startups, so mostly I just sat around sort of vaguely programming, but not really. But yeah, it could technically have been described as starting a startup, but only very technically.
1:22:38 Working at Theorem LP
Alexey: Yeah. Okay. And then you took your first full-time job with Theorem LP.
Alexey: And how was the job and why didn’t you stay there?
Ben: Yeah, that was, let’s see. I think it was fine–
Alexey: Oh, sorry. So the reason I’m especially interested in this question is because again, in one of your old blog posts, you mentioned how the job was at least twice as good as it’s the next best job. And it was interesting that despite this, you ended up leaving after a year and not stay in there for longer term and capitalizing on this-
Ben: Well, so that’s the thing about power law distributions, right? Is that usually the top item in your sample is going to be like twice as good as the next item, but then if you keep sampling, you’re just going to have the same thing happen again. And in fact that is what happened. I would say the difference between Wave and Theorem was probably more like 10X than 2X, mostly based on me being an unusually good fit for that role.
So to get back to, how was it actually. So Theorem was a company that used machine learning to do credit scoring, basically. This was during a time when a couple of companies that did like peer-to-peer lending were getting very successful. So they were called like Prosper and LendingClub, and what they would do is you could apply to them for a loan, but instead of them giving you a loan, like with their own funds or something, individual investors would sign up for a Prosper and LendingClub account, and then they could buy like slices of your loan or fund slices of your loan.
And at least for a while, I don’t know if this is still true, but at least for a while, basically what Prosper and LendingClub were doing was they were like… I’m not very confident in this because I didn’t really dig in that much depth, but I suspect what they were doing was basically like arbitraging the fact that a lot of people have credit card debt because credit cards are one of the things that make it easiest to rack up debt. But credit cards also have to charge super high interest rates because of the fact that it’s a revolving loan. So when you default on a credit card, you’re likely to have maxed out the balance, right? So defaults on credit cards are really bad and you lose most of your money and you have to charge really high interest rates to make up for it.
So if you refinance your credit card debt into a fixed term loan, people are not likely to default on their first payment of a fixed term loan. They’re more likely to default like halfway through or something and then you only lose half the money. So it’s much better. You can charge a much lower interest rate. And so Prosper and LendingClub, basically, I think most of the reason they were successful is because they became like a good way to refinance your credit card debt onto something with a lower interest rate.
And then it turned out that they could ask you for a bunch of information that I believe for regulatory reasons, they weren’t allowed to use in their credit scoring algorithms, but investors, what they would do is they would underwrite the loan and then later investors would buy it. And so the investors could look at all this extra information that was not able to be used in the underwriting decision. And you could get some amount of alpha from that information. And so what Theorem did was they basically ran a hedge fund that bought the best loans on these platforms and then made pretty good interests for relatively little risk. There was still some macro economi risk that we had no idea what it was because the economy had never been through a downturn while these peer-to-peer lending platforms were in business. So we didn’t know how correlated would defaults be when the macro economy became very bad. So now they probably actually know that, I guess I should probably go back and look at what happened to Prosper and LendingClub during COVID.
At least at that time, aside from macroeconomic risk, it was just like, “You can make a really good return by being good at credit scoring.” So I did two things. One is I worked on improving our machine learning models. And then the second was like, I worked on putting the machine learning models into production. It was pretty fun, I learned a lot about machine learning. I got to write some cool blog posts about machine learning. I think ultimately what I later realized was that it was not very motivating because ultimately the company was not constrained on its ability to do good machine learning. And it turns out that if you threw a halfway decent model at this problem, you would just do pretty well.
And then the amount that you could improve on that was limited by the fact that defaults were very rare. And so your return was basically carved by the average interest rate of the loan that you were buying. And so model improvements were kind of useful, but not really that useful. And in reality, what this company needed was more people who are good at hedge funds sales. And so, I spent a little while trying to do hedge fund sales. I sort of almost convinced one guy to invest in our hedge fund, but that was close to the time that I left. And I think after I left, he didn’t end up investing in the hedge fund. Also, it was a guy that I previously knew. I was garbage at actual hedge fund sales. I was doing the same thing that I did at Harvard Effective Altruism of like, Oh, “I’m clearly garbage at this thing, but nobody else is around to do it, so let me do it.”
Anyway, in retrospect, I was kind of demotivated because of realizing that the company was not actually constrained by the thing that I was supposed to be working on, but I didn’t realize that at the time, because I had never worked at a place that was good at allocating resources only to the things that it was constrained on.
1:29:38 Finally learning about Wave, realizing it’s a big opportunity, and getting a job there
Alexey: So it took you like four attempts at working at different kinds of jobs to finally get to Wave, where you’ve been for like five years or something.
Ben: It depends on how you count, but approximately four.
Alexey: The number is interesting to me because I feel like the reason that you ended up finding Wave was because I guess you were like really not satisfied with just working on machine learning somewhere, being a software engineer somewhere.
And I feel like it’s common to expect that you will figure the first or second or the third job will definitely be the job that you should be working on if you have not. And even the second or the third job that you’re doing you don’t feel like working on it for more than half a year, then there’s must be something wrong with you. And your story is interesting just from the perspective of it took you four or five years since your first internship before you found the job that you’re really good fit for.
Ben: In fact, I think it’s even worse than that because I didn’t realize that my problem was that I didn’t feel like working on it until I knew what it felt like to work on something that felt actually important.
There were many such things. Similarly, I didn’t realize how important it was to have a good manager until I started working with Drew, the CEO of Wave because at least for me, he was an incredible manager who gave me by far way more useful self-improvement feedback than anyone else that I had worked with up until that point. But I didn’t know how good any of those things would have been or even when I joined Wave, I didn’t realize that to what degree it was a better place for me than the place that I was previously working.
The way that I realized that I should leave Theorem and join Wave was actually fact that – so I was not Wave’s first engineer – their first engineer was my then girlfriend, whom I had referred there. And then I kept looking over her shoulder and being, “Oh, huh. It seems like the thing that you’re doing is much more fun than the thing that I’m doing.” And then after about nine months of that, I was like, “Man, fuck this.” And I simultaneously realized that earning to give was like a bad strategy because I fully internalized the fact that Dustin Moskovitz had $8 billion and whatever money I made was like not going to be like $8 billion.
In fact, I didn’t even realize the extent to which I should have been dissatisfied with all of the other things that I had done until afterwards. In fact, I’m still not sure of the degree to which I should have been dissatisfied with there is, right? I don’t know whether everyone should expect to be able to find a company Wave within their first five attempts at changing or whether I got super lucky and, in fact, other people should expect to take so many attempts to find a company that’s equally good that, in fact, should stop searching earlier. Yeah, it’s a very tricky problem that I don’t have a great answer to.
Alexey: To me, it’s not obvious why Wave is such a great opportunity.. It’s a financial services company, right? It works on transfers. And again, by EMH, we would expect that there would be a bunch of companies already, doing money transfers. In fact there are a ton of companies doing money transfers, both in developing world and in developed world, between the US and Africa and within the US and Africa probably. And so, why is Wave such a great opportunity?
Ben: That’s a good question to which I also don’t have a full answer. I should clarify that the thing that Wave is doing now, which I think is the best and that I’m most excited about is: we are building mobile money systems within various countries in Africa. So these things are basically, we provide the same services to people that banks provide, but instead of having bank branches, we have agents that are scattered all throughout the country that are like shopkeepers who use their spare cash to service Wave customers' transactions. You’ll go to an agent, deposit your money, and then you have balance in your Wave account that you can send on to somebody else or something like that. And so, because it’s a lot easier to start to open up a new agent than to open up a new bank branch, we can provide dramatically better service at a lower price point than banks can.
Alexey: It’s like regulatory arbitrage essentially?
Ben: Calling it an arbitrage is strong – there are explicit regulations that you must comply with in order to be like a mobile money provider. It’s just like countries in Africa realized that you can’t start a good bank in Africa because building branches is too expensive relative to the average account size. And so they decided that in addition to banks, they would allow people to build like mobile money things.
There were regulations specifically you must comply with in order to be mobile money. And the question is, “Why aren’t there other like good mobile money systems in Africa?” And I don’t know the full answer to that question. It feels to me you probably could have started effectively Wave four years before Wave was actually started and done a reasonably good job. You would have had a slightly harder time because fewer people would have smartphones, but nothing super fundamental would be different. So I’m not sure of the full answer to this question. Part of the answer I think is clearly the efficient market hypothesis is false.
Ben: So I actually wrote a blog post that was sort of about this, which you may or may not have been obliquely referencing, but it’s called Why and how to start a startup serving emerging markets. The theory that blog post outlines is that there are very few people who have both spent a lot of time in one of the clusters of the world that produces great technology products and then are also happy to instead of continuing to live there, go live in a place in Africa where they would learn enough of local contexts to be able to build a great product that solves people’s problems in those countries.
I think that is one of the key differentiators of Wave is that… And it’s not there are zero people with this trait to be clear. Of all of the people who like live in Silicon Valley, right? And are excited to start a tech company. I would guess that probably under 0.1% of them would be like excited to live in Africa for four years, Drew the founder of Wave has, right?
Maybe it’s higher than that actually now, because YC is now admitting a substantial number of companies from Africa where the founders are from Africa, although that wasn’t the case when Wave was started.
The original mobile money systems were all built by telecoms. And if you’re a telecom and you’re building a mobile money system, this mobile money system is by far the most complex product you have ever put in front of a consumer. The default is you’re going to do a terrible job at it because you have never had to make something simple before, or take something complex and make it simple. And there’s one telecom that did a really great job of this, which is Safaricom the company that ran the original, the OG mobile money system, M-Pesa and our current theory about why that works was that basically M-Pesa was run effectively as a startup, without very much interference by Safaricom.
Whereas all of the copycats and other countries were run as business units of the telecom. So they like took one of their like telecom execs, and they were, “Hey, you like start a mobile money system.” And the exec was, “Okay, let me hire you a bunch of engineers, buy something off the shelf from like Ericsson or Huawei or something.” And then if you want to change your software, you have to file a ticket with Huawei. And then 12 hours later the Huawei engineers respond to your ticket and they’re like we can’t do this or whatever. Anyway so it was much harder for them to execute well than it was for Safaricom. So I guess that’s the reason why the EMH is not a good argument that Wave wasn’t a good like opportunity.
Sorry. I should first have given the actual argument, why is Wave a great opportunity? The reason Wave is a great opportunity is because it allows people to transfer money instantly at much lower prices than they previously could. If you’re, for instance, a fish trader in Senegal, this is life-changing awesome because, for instance, you send fish to a market somewhere else in Senegal, people will sell the fish, get money and they send the money back to you. And as soon as you get the money you buy your next load of fish. So if the money is coming back on a bus, then like it takes another day for you to get your money back.
If the money comes back by Wave, you get it instantly. Then you can buy your second load of fish instantly. And that means that if it takes you one day to buy the fish and then one day to get the money back, Wave is effectively letting you buy twice as many fish with your given stock of capital. And therefore you have doubled your income. Okay, it turns out people like doubling their income. Therefore, Wave is like a really useful product to them. So sorry I guess that part should have gone first, but that’s why. And so it was a great opportunity because like we built a product that people really love and therefore it grew really fast. And it is now used by over 25% of all the adults in Senegal, for instance. And it doesn’t seem like there’s any reason why we should not succeed equally well elsewhere. So it’s exciting to be able to potentially build the economic rails of like much of Sub-Saharan Africa. And then the first half of my answer is why nobody else did this four years ago maybe.
Alexey: Yeah, now this sounds really impactful more than just international payments or something. Oh, I guess it wasn’t international payments in the first place, but–
Ben: So our first product was international payments. Yeah, so we were sending money from the US to Kenya and we would deliver the money using M-Pesa the OG mobile money system that I described earlier. The reason that we started Wave was actually that we tried to expand that to other countries. And that was when we realized that it looked like they had mobile money systems. If you looked at them up on Wikipedia, they had mobile money systems, but the mobile money systems were all like shit. They were really bad. And so we were like, “Oh, this sucks for our product because it’s not growing as fast in these other markets, because we’re delivering to much worse mobile money systems.” But it was great for us because it meant we could just build a better mobile money system. And that was, that’s a much more exciting product than just international money transfer because theoretically it could be used by everyone in the country and not just people whose relatives managed to immigrate.
1:43:25 What Ben disagrees with his friends about: the value of an inside view, the inefficiency of markets, the base rates
Alexey: Yeah. So the next couple of questions that they wanted to ask here and less about your job, but like these general questions that people usually ask in interviews, where you try to learn something really insightful from the person you’re interviewing rather than a bunch of details of their lives. But yeah so the first thing of this kind of is, what do you find yourself disagreeing with your friends about?
Ben: I guess, not taking the efficient market hypothesis very seriously. I think there’s some communities that I’m part of that take it way too seriously. But I think even with respects to many friends, I am somewhat more likely to try to take an inside view and less likely to put weight on efficient market hypothesis considerations.
I think one example where ex post I was clearly right and many friends were clearly wrong was that I thought Wave’s equity is going to be very valuable. And I tried to tell people that this was true. And they were like, “Ah, efficient markets and startups and venture capitalists and whatever.” and I was like, “No, VC markets are extremely inefficient, whatever.” Anyway, I don’t know if like ex ante–
Alexey: But the base rates.
Ben: Yeah. So I don’t know if ex ante I was wrong about this, but ex post I clearly was right. Wave’s equity has gone up a lot more than the people who thought that it wouldn’t.
Alexey: It might not be the best thing to share in the thing that will be published on YouTube.
Ben: I think it’s fine to share directions. I will not disclose how much it’s gone up, but I don’t know. For instance, I don’t think any of those people’s predictions was compatible with us building a mobile money system that was used by 25% of people on Senegal.
But more generally, like I said, I was fine. I was much more comfortable than these people that were being like, “Okay, it looks like we’re going to succeed at building this mobile money thing.” Then if you look at how valuable of a business that should be, gosh, it looks really good. Okay. Yeah. We’re raising it this much lower valuation. And that’s because venture capitalists don’t understand Africa and the VC markets are silly and we’re trading off speed for trying to optimize valuation. And they were, “Why would somebody ever trade off speed for trying to optimize valuation and blah blah blah.”
To be clear, I don’t want to use first principles reasoning like that very much. Even assuming that I was right about it ex ante, which is hard to say, maybe I got lucky, but even assuming I was right about ex ante, it requires a lot of context and specific knowledge to get right. And that context and specific knowledge is hard to acquire. But I think at least for career decisions, I think you want to get to a point where you can use an inside view instead of being, “Oh, the base rate is X or something like that.”
Alexey: But this thing about base rates and you disagreeing with your friends about this, how much of this was the fact that you were inside of this thing and saw the details versus your just general, thinking actually working differently. And if it’s in significant part of your thinking at the time being different, the style of your thinking, how did that come about?
Ben: I think the causation went in the other direction. That is, I saw Wave clearly just destroy the efficient market hypothesis. And then I was, “Oh, wait, wow, the efficient market hypothesis here was so wrong. Maybe I should stop paying much attention to efficient market hypothesis-type arguments.” So the update was caused at least in part by witnessing first-hand, an absurdly inefficient market.
Alexey: Okay, this makes sense.
Ben: And witnessing, comparing the type of reasoning that the founders of Wave employed to find the idea and then turn it into a great company versus the type of outside view reasoning that EMH partisans used to evaluate whether something is worth doing or not.
When I joined Wave, I had a number of conversations with Drew where I said something about base rates and then he was basically like, “Fuck base rates man.” I was like, “No, but don’t you understand like base rates?” And he was like, “No.” And then eventually I realized that was part of why Drew was effective.
There’s a really good blog post on Applied Divinity Studies where he hypothesizes that the reason there are so few successful entrepreneurs in rationality or effective altruism communities is that these communities train people to put a lot of weight on outside type view arguments. I think that’s pretty much correct. And the reason I stopped doing it was that I saw firsthand that “Oh, the effective entrepreneurs that I am interacting with every day do not reason in this style at all.”
I think it’s pretty good for having a low variance, moderately positive outcome. I think ignoring outside views is very high variance. A lot of the time, you’re wrong. For instance, before they built Wave, the founders of Wave built 10 unsuccessful social mobile apps. And they were doing the thing: they were sampling from a heavy-tailed distribution over and over again. And then finally they got a sample that was like a gazillion instead of zero. But at first they drew 10 zeros.
Alexey: Were you not bothered by the 10 unsuccessful social media companies that came before Wave when you were joining?
Ben: No, Wave was obviously succeeding at that point.
1:51:20 What Ben thinks about the state of the EA community
Alexey: What are the things that you disagree with the EA community at large about, today? Over the years there was this thing with earning to give, how exactly valuable that is, but today the EA community shifted focus significantly towards long-term risks and long-term things and you’re working on this thing that on the one hand the EA community would never consider super valuable because it’s not like super explicitly long-term directed, but at the same time, it’s direct work in a way that–
Ben: So I think the EA community believes that given a worldview on which it is valuable to work on global poverty, working at Wave is one of the best things that you can do.
For instance, we’re EA-approved enough that our job postings appear on the 80,000 Hours job board. They don’t appear on the list “we think these are extra great” section, but they appear in the “other opportunities that might be promising” section or something. I think even the global poverty wing of the EA community has shifted away from earning to give as the be all and end all and towards a more pluralistic, “you should actually try and do things yourself” route.
In terms of what I disagree with the current EA community about… I don’t know. I’ve stopped following it as closely or engaging with it as much as I did in the past. So it’s a bit harder for me to say than it would have been. Presumably the EA community believes that the EA community is more interesting than I believe the EA community is, in the sense that there’s a bit less there than there was in the early days of the EA community.
It’s just like a bunch of people working on long-termism stuff that’s valuable under this one worldview. And then a bunch of people working on global poverty stuff that valuable according to this other worldview. And a bunch of people working on animal welfare stuff that is valuable under this third world view. And the world views are quite different.
It’s unclear that they get that much from engaging with each other. They have very different ways of doing epistemology and course prioritization. And so what is there actually left to effective altruism overall? It’s hard exactly to say.
If you strip out the particular parts of each of these worldviews that are pretty different, is there some sort of common core other than like these people hang out with each other a lot? I don’t know. The community isn’t very monolithic.
Yeah. I guess it’s hard to see what’s going on there that’s super interesting right now, divorced from the work on the particular course areas that a lot of EAs think are important. There’s various interesting stuff… if you’re interested in AI risk, there’s a lot of interesting AI risk-related stuff going on. But general EA stuff? I don’t know. It just seems kind of weird that it’s a thing.
Alexey: Cool. So I think I actually asked you everything I wanted to ask here.
Alexey: So, do you think there’s anything that we can talk about that you expected that I would ask you or that you thought that it would be fun to talk about that we did not end up talking about?
Ben: To be honest, I just expected you to ask interesting questions and you did, I didn’t spend a lot of time preparing for what I imagined you might ask about because I figured you’d probably ask better questions than I could come up with yourself. And you delivered. They were great questions. Super fun to talk and yeah, thanks for having me.
Alexey: Yeah, thanks for taking the time and I’ll edit this at some point and it will appear on YouTube at some point.
Ben: Sweet. I’m excited for that. Thanks again.
Alexey: Yeah, it was great talking and see you, bye.
Ben: You too, bye.