[0:00:00] Hi listeners, this is the 80,000 Hours podcast, where we have unusually in-depth conversations about the world's most pressing problems, what you can do to solve them, and whether free surgery from someone without a medical degree is a good idea. I'm Rob Wiblin, head of research at 80,000 Hours. Last week's episode was Chris Ola's first podcast ever, and today we're releasing his second podcast ever. Because we ended up having so much to cover, it made the most sense to do more than one recording session and split it into two quite different interviews.
[0:00:26] Last week's episode focused on Chris's technical work, but this one is about Chris's life and experiences so far. We cover issues like how he rose to the top of the ML field without a university degree, how he got his foot in the door at Google Brain and OpenAI, the academic journal that Chris founded called Distill, and his views on how to explain complex things really well, how to write the best cold emails, micro marriages, micro best friends, and much more besides. All right, without further ado, I bring you Chris Ola.
[0:00:57] Today, I'm speaking with Chris Ola. Chris is a machine learning researcher currently focused on neural network interpretability. Until last December, he led OpenAI's interpretability team, but he recently left with some colleagues to help start a new AI lab focused on large models and safety. Before OpenAI, he spent four years at Google Brain developing tools to visualize what's going on in neural networks. And he was hugely influential at Google Brain, being the second author on the launch of the Deep Dream article back in 2015. And I guess the Deep Dream images are something that just about everyone has seen at this point.
[0:01:26] He also helped pioneer feature visualization, activation atlases, building blocks of interpretability, TensorFlow, and even co-authored the famous paper Concrete Problems in AI Safety. On top of all of that, in 2018, he helped found the academic journal Distill, which is dedicated to publishing clear communication of technical concepts. And Chris is himself a writer who is popular among many listeners to the show, and his blog has attracted millions of readers by trying to explain cutting-edge machine learning in highly accessible ways.
[0:01:51] And he's managed to do all of this without a degree because he dropped out of college in 2009 to defend a friend against bogus terrorism charges. And in 2012, Chris took a $100,000 Thiel Fellowship, a scholarship designed to encourage gifted young people to go straight into research or entrepreneurship rather than go to university. What an intro. Thanks for coming on the podcast, Chris. Thank you for having me, Rob. That is an extremely flattering introduction. I should say my role in TensorFlow was very minor. But all of the rest of it stands.
[0:02:19] Yeah, you know, this is my first podcast ever. So if I get terrified of the medium and never do a podcast episode again, everyone will know who to blame. It'll be Karen and me to blame. Yeah, or alternatively, if things go well, we'll have a fantastic scoop and get lots of new subscribers. That's the dream. All right. I hope you get to talk about your research into AI interpretability and your many unusual life experiences as just described. But first, what are you working on at the moment and why do you think it's important? I think one of the craziest things about machine learning.
[0:02:49] And so the thing that I've always felt.
[0:03:15] It has sort of just been this question that I've been obsessed with and just sort of feels like the burning question in machine learning to me is how in the wide world are these systems going and doing all these crazy things that we don't know how to do? And I care about that for safety reasons. And honestly, I just care about it because it seems like this incredibly crazy thing about the world that I just want to understand.
[0:03:36] Yeah, that makes a lot of sense. I guess you've been on a, it sounds like looking at your CV, something like an eight-year journey working on this problem, trying to pick a way at it and take neural networks from being these black boxes to things that we can probably understand and build on. Yeah, I mean, it's not the only thing that I've done for the last eight years, but it's definitely been the biggest one. And I've tried lots of things. A lot of the things I tried early on didn't work very well, but over time, I think we've really developed.
[0:04:04] And not just me, but lots of other people and lots of collaborators that I've worked with have really been able to get to a point where we can actually very significantly understand neural networks. And I can actually just sort of look at their weights and read entire algorithms for doing things that we didn't really know how to do before off of them. And that's been really cool to see. All right. Well, we'll return to some of those techniques later on.
[0:04:24] Let's talk now about, I guess, the very unusual and circuitous career path that you've ended up going on. And I guess we can see if people can learn any lessons from it, or maybe whether it's just too weird to be generalizable. Obviously, you now do machine learning research really professionally, but unusually, you don't have a PhD or even an undergraduate degree. I guess, what's the story of how that ended up happening? Gosh, well, it's a bit of a personal story. But when I was in high school, I became really involved in a community technology space.
[0:04:53] And so the G20 came to Toronto and one of our members was doing security research where he'd do things like he was recording where temporary cameras got put up. And the police thought this was really suspicious. And they raided his house and they found a hobby chemistry and electronics lab. And they decided he was making bombs. One of the police officers conducting the raid had served in Afghanistan and thought he could recognize explosives and misrecognized.
[0:05:23] Chemicals as explosives. And yeah, it was a really, you know, obviously a really awful situation for him. I didn't know him that well. I knew him a bit through Hack Lab. But I guess a lot of us really rallied in supporting him. And I had a lot more flexibility than other people because I was a university student in my first year. And it just seemed really important to me to try to support him. And so I started going.
[0:05:49] And going to court whenever he had bail hearings and stuff like this and trying to do court support type stuff to just help him and help his family a little bit. And I took notes and things like this. And then he was in under house arrest for one year. And in the second year, I took a year off university so I could go and be at his trial full time. And he was found innocent of all charges in the end. But that led to me initially dropping out of university.
[0:06:15] Yeah, I read a bit about this records from, I guess, 2008 on your personal website. It is an astonishing and pretty tragic story. You must have been, I guess, I'd imagine that I would be as well, just like so outraged and disgusted in order to basically drop out of university and just, it sounded like you were focusing a lot of your time on tracking this case. And I guess trying to help in as much as you could to make sure that this guy didn't get some long, like terrorist prison sentence for something that he absolutely hadn't done. Yeah. I mean, I don't know that I actually helped very much.
[0:06:45] Really, at best, I saved him a little bit of money by going and doing some things that saved his lawyers a bit of time here and there. I think the highest impact thing I did was I transcribed an interrogation session and put it on YouTube. And then his lawyers could go and there's this nice feature where you can click on lines in a transcript and it will jump the video to that point. I think that was probably the highest impact thing I did. But I think actually it wasn't primarily outrage at first. There was a lot of outrage. But I think actually the first thing was fear.
[0:07:13] I was scared for this person. I was scared for myself and for all of my friends that maybe they were going to now come after all of us as, I don't know, some kind of co-conspirators or something like this. And so it was just a really scary time period. And I think, yeah, maybe the main thing was just it seems really important when sort of people are systematically sort of stepping away from someone or abandoning someone, it seems really important for other people to rally and to try and support them.
[0:07:42] Or that was sort of a strong emotional intuition I had. And so that's, I think that was probably the main motivator for me. Yeah, I guess I hadn't thought about that angle as much that I suppose it's actually on some level required bravery, or I guess bravery or foolishness. Because by taking such a big interest in this as someone who's an associate now of an accused terrorist, you're like potentially painting a target on yourself to have the police come after you or like investigate you or potentially try to engage in reprisals if they're frustrated with what you're doing. Imagine your parents were
[0:08:11] Maybe they're on board or maybe they were just horrified at the risk that you were taking? No, I think all the adults in my life were very, very deeply worried about this and pushed me really hard to not do it. I think realistically, any jeopardy that I was in would have already been from pre-existing associations. But I think I didn't know anything about law at that time. I didn't know anything about how to think about this type of thing. And so it just seemed very, very generally scary.
[0:08:40] Yeah, I guess it sounds like you, it's totally understandable why you did this, but it sounds like you think you didn't make that much of a difference. In retrospect, do you wish that you hadn't dropped out of university or hadn't done this stuff? Or is it just that then it's just maybe an impossible hypothetical to consider because then you'd be a different person? Yeah, it's pretty hard to think about it. I do think that there's a way in which it was altruistic, but not especially effective. On the other hand, it's like hard to feel too bad about, like, I think probably
[0:09:08] Probably this person felt more supported and his parents probably felt more supported. And that definitely felt good. And I don't know that my time was super high leverage at that point in my career. But it's hard to think about. I mean, it also just put me on a trajectory that made me do a lot of different things. And maybe in some ways made me more effective later on by sort of becoming more unusual. So it's hard to think about. Yeah.
[0:09:35] It makes sense that kind of when someone's accused of a crime like that, and even if people all know that it's a false charge, that people are inclined to just run away from them. And then I guess that means like someone has to take the hit of being willing to stand up and associate with them in order to prevent that person just being completely ruined or having like no social support and feeling like they're being completely abandoned. Yeah, it's a difficult situation. I suppose, yeah.
[0:09:58] It's hard to think about it through an effective, virtuous lens, but I extremely admire and understand why you ended up doing that. Maybe I've just become inured to stories of misconduct by police. I suppose we don't want to go into all the details here because this isn't ultimately an episode about criminal justice. But this must have shaped your view of national security services somewhat negatively. Yeah, what did you think at the time? And maybe how has your opinion of that evolved since you were 18? Yeah, it definitely...
[0:10:27] Soured my opinion of national security agencies. There was a lot of really awful stuff. They threatened his wife at one point to try to get him to confess to something. There was just a lot of stuff where in his trial, I thought the prosecution was deeply disingenuous. And I should say, you know, in my sort of role as trying to be a supporter, I don't think that I was always the epitome of intellectual integrity. But I think that if you are a prosecutor who's trying to put somebody in jail, going and making disingenuous arguments, like one that...
[0:10:57] I remember is they argued that the reason he had chemicals that could burn bright colors was he was planning to make a rainbow bomb. So after they found out he didn't have explosives, it became that he had chemicals that could be used to make explosives. And they found out that he had chemicals that could burn bright colors. And yeah, this argument that it could be a rainbow bomb. And it just, it's so like, I don't think that anyone who sort of is approaching this in an honest way would make that kind of argument. And they really caused, I think, severe harm to this person's life.
[0:11:24] And so that really deeply soured my views. And yet, on the other hand, I think over time, especially as I've thought more about X-risk and bio stuff and even AI, I've gradually had more sympathy also for national security type concerns. And yeah, I don't know. I guess where I sort of wind up is something like just being really disappointed because I guess I really want to be able to trust these organizations.
[0:11:53] They clearly are systematically falling short and are also doing something important. And also just sad, like, even from their perspective, they're just going and burning goodwill left, right, and center by doing all of these things. And it's sort of from one, I put on one hat and I'm sort of really outraged at the, just the appalling ethics of it. And I put on the other hat and I'm just really sad at the needless waste and...
[0:12:22] And burning of goodwill. Yeah, it's extraordinary that so much time was spent on this case, which like rapidly became apparent that there was nothing actually to investigate here. I guess, I mean, I know that there's like amazing people with the national security, like some of the people with the greatest integrity and incredible work ethic. And I guess it's just national security and law enforcement. It's like any other industry where there's some people who are fantastic and there's some people who engage in misconduct all the time.
[0:12:48] I guess it's just an area that is so important and where there's so much power that when people do engage in misconduct, the consequences can be extremely unjust and extremely harmful. And it's also such a large space that I think it's harder probably to go in and have a high bar and to sort of really filter things. I don't know. I'm not an expert. I don't want to opine too much.
[0:13:07] Yeah, well, I mean, the police are like one of the largest employment groups. So it is a lot of people. And I suppose, yeah, as you're saying, it probably is difficult to have such an extremely high bar such that only like the people who have demonstrated the greatest integrity in their life can possibly become police officers because it's just so many. Let's maybe go back to the path that you were taking in your career then. So you've been going to university, but you left in order to do this. And I suppose, yeah, your like natural, your boring, boring track in life had been somewhat derailed. But you could have probably gone back to university after that, but you decided not to. Yeah, why was that?
[0:13:36] Yeah, so I guess when I was doing this sort of court support type work, there were lots of months where there were lulls and there just wasn't anything to do. And so I had lots of free time. And I'd been interested in 3D printers for a while. So I got really involved in working on 3D printers. And first I was designing 3D printers. And then I was working on a startup with a friend for open source software to go and design objects for 3D printers.
[0:14:01] So a lot of this so-called CAD software that you use to design objects, 3D objects, and we were trying to create open source tools because we felt that was very important. So I applied for the Teal Fellowship, which is a program that provides financial support for people under the age of 20 to go and work on sort of ambitious projects or do unusual things.
[0:14:23] I got it and I sort of was like, well, you know, I have two options. One is go back to university and the other is I can work on whatever I want for two years. It turns out that wasn't a super difficult decision. Yeah, but I guess a lot of people who are considering not going to university and, you know, starting a business or doing something else, they feel like they face a lot of pressure, most understandably from other people, from society, from themselves. They have to, you know, you have to go to university, you can't not.
[0:14:49] Yeah, was it a difficult decision to make on some level, even if you kind of preferred the path of not going? Well, I have a lot of experience sort of doing things, cutting against pressure from the previous stuff. Yeah, that's fair enough. But I mean, I think at the time I framed it and especially framed it to other people as well. You know, I can do this for two years and then I can still go back to university. And that sort of seems like, you know, an amazing opportunity. I guess one other thing that sort of comes into play here.
[0:15:15] It's actually much easier to do unusual things when you're validated by a third party. Like I think people, when they hear about the TFL show, they're like, ah, the high value things that they're providing funding. And that's certainly part of it. But I think that actually the higher value thing was actually like adults in my life totally came around once I was getting, you know, given $100,000 to go and work on stuff in a way that they really were not supportive beforehand. So I think there's also just a really big effect in terms of legitimizing an untraditional path, making it easier.
[0:15:45] Yeah, I guess it speaks to the signaling component of going to university, where I guess you need some credential to show that you're a legitimate person who's not crazy or not irresponsible. And I guess the Thiel Fellowship was providing that as a substitute. I think there's an element to that. I think there's also just an element of when you do a non-traditional thing, it's scary for people who care about you. And having some signal that, in fact, you're doing something reasonable, I think can help a lot.
[0:16:10] So you have this essay on your website where I guess you go into your views on the general path of not going to university and going and doing something else instead, which is probably the best resource that people can go to if they want to properly learn all of your views on that. Yeah, do you just want to summarize briefly? Is this something that a lot more people should be considering? I get a lot of emails from people asking me if they should go to university. I think it's maybe the single most common question I get asked. And I think for almost everyone emailing me, they should go to university.
[0:16:40] And the reason that I think that is, I think that if you want to benefit from going and doing something else, you have to have a lot of, I think, sort of self-discipline and willingness to go and work hard on things and self-motivation to work hard on things without an external forcing function. And I think that often people don't have this and then this kind of thing doesn't work as well for them. On the other hand, I think for the people...
[0:17:08] And maybe to give some more context, I think a lot of the people who I saw really thrived in the PL fellowship. I mean, some had already, before the age of 20, done undergrad degrees. So there were those ones. But I think a lot of people had done really significant personal projects involving software or science or something like this. And I think that's actually a pretty good test. If you have sort of been able to, out of self-motivation, go and do your own large personal project.
[0:17:35] And, you know, obviously, if you are in a position to be able to support yourself, then you're likely to be able to do well in something like the TL Fellowship or taking a year off or taking a few years off. But if you aren't, it'll be much more challenging. Yeah, I guess, to what degree is this a path just for really, really talented people? I think it's like fair to say that, you know.
[0:17:58] 20-year-old Chris Ola had an awful lot of potential. And I guess some of the other people who I've seen thrive not going to university, it just seems like they were kind of on a rocket ship to begin with. And yeah, it makes me wonder, if you're someone who's merely very talented, do you need to go to have those credentials in order to get your foot in the door in the kind of careers that you want? My guess would be that the self-motivation thing is bigger than anything else. But yeah, I feel like I also just don't feel that qualified to like opine on how useful university would be to people who are very different from me.
[0:18:28] For example, I can imagine a world where actually like for many people who are not going to pursue sort of really high, I don't know, challenging careers, maybe actually going to trade school might be more effective or things like this. I could imagine there being like different reasons why not going to university might make sense for people who are very different from me. But it's harder for me to speak to that. Has not going to university like ever created issues with being taken seriously later down the line? Or I mean, possibly immigration could be a concern as well. Yeah, so I think immigration is an underrated concern with this.
[0:18:58] Almost all visas for the US require you to have an undergrad degree. I'm on this really weird alien of extraordinary ability visa, which just doesn't enlist it. But I think that's really the main option you have if you don't have an undergrad degree. I think that there's also some social challenges where I think actually something that I undervalued when I decided to not go to university was it makes it so much easier to have friendships and to have romantic relationships as a young person. And I was sort of...
[0:19:27] And I think that was actually a really significant cost that I didn't understand at the time. Yeah. So tell us a bit more about the Teal Fellowship and I guess to what extent having the social group there substituted for universities, perhaps. Well, it might have done more so if I'd been able to live in the Bay Area. I visited at that time periodically, but I was still living in Canada since I didn't have a visa and couldn't live in the US. Yeah. So it might have done more if I had been able to.
[0:19:56] I did make some friends and I think that was valuable. And I don't know, I mean, I guess more generally, it was just a really incredible period of being able to go and pursue things that I was excited about. So I started with 3D printers and I also was just doing lots of random side projects. And then I guess after a while, about a year in, I guess I switched from 3D printers to machine learning. And that was a fairly big pivot. Yeah, what caused you to do that?
[0:20:25] Well, I guess there were a few things. So one is just sort of I had the opportunity to do so because someone I knew in Toronto, Michael Nielsen, was writing a textbook on neural networks and ran a seminar series sort of to practice for writing his textbook. And so I got exposed to a lot of, you know, I guess to a lot of ideas about learning, just as the field was really just on the verge of taking off right after the exciting computer vision results from Krzyzewski in 2012.
[0:20:51] So I sort of had the opportunity. I sort of was aware that it was this really exciting thing, and I became very excited about it. And my main collaborator on the 3D printing stuff dropped out. I guess another thing that contributed a little bit was I actually happened to meet Holden Karnofsky of GiveWell and the Open Philanthropy Project. That would have been pretty early on. Yeah, this was, I guess, in 2012 or 2013. The reason that I met him was I knew, well, I was friends with Dario, Dario Amodei.
[0:21:20] Who at that time was a grad student. And Dario, somehow, despite being on a grad student stipend, I don't know, was probably giving a fiscally irresponsible amount of stipend to GiveWell. And that led to him meeting Holden. And so I ended up getting dinner with Holden and Dario. And during the dinner, I sort of pitched Holden on how, you know, I was working on 3D printers. And, you know, I thought that if we can have open source tools for 3D printers, you know, we'd be able to go and bring 3D printers to everyone.
[0:21:49] And maybe we'd end scarcity and stuff like this. And Holden just really shot it down. And at the time, I was very miffed. And actually, the thing that Holden said was, I don't think this is valuable, but it seems like a great way for you to develop skills. At the time, I was really miffed. But in retrospect, it was actually a very useful thing for him to say. I think this is a common theme of talking to Holden that I feel kind of miffed at him sometimes. And then very often, I'm very grateful for the thing that he said.
[0:22:19] Yeah, what was his case against working on open source? Sorry, what do you call it? Like local manufacturing? What's the term for this? I mean, I was writing open source CAD tools. I guess I don't remember the exact argument that he made. I think he was just skeptical of a lot of things. And I think that there were a lot of things to be skeptical of, not the least of which was, you know, like whether 3D printers would actually be able to become really widespread at low cost in the way that I was envisioning. And I think also just whether these open source tools were actually...
[0:22:47] Yeah, people have been talking about 3D printers for a long time. And I've always kind of thought, I just don't know how many things like that I want to make. How many things that can be 3D printed? I'm like that interested in producing for myself. Maybe I'm going to be proven wrong one day. I think the really compelling story is, you know, if you can create a 3D printer that can print itself, and it can also print a wide variety of objects, you can sort of have this feedback loop of making it really easy to go and spread them and then go and...
[0:23:15] Anything that you can print, you can provide sort of freely to everyone. And you have these printers that you can like 80% print or something like this, right? And then there's the remaining vitamins that you can't print. But the vitamins are all the hard parts. And so it's sort of an illusion that you're sort of making progress at this, I think, to a significant extent. I don't know. I haven't been involved in this space for a long time. But at one point, I had a lot of fantasies about this. And I designed a 3D printable vacuum cleaner and worked on how you could make microscopes with 3D printers.
[0:23:44] It wasn't, in retrospect, wasn't very useful. Yeah. What do you think of filling this role where you kind of shoot down someone's plans? It's a difficult one because you come across as potentially kind of an asshole. Maybe you're not really in a position to confidently tell people that the idea is bad. I think onlookers don't like you when you're saying that someone's plans are bad and the person themselves might object and you run the risk of being wrong.
[0:24:09] And yet it does seem like sometimes it's just really important to have someone say, no, this is a dumb idea. It's a balancing act. I think it is really important. It's something that I wish I was better at. I think personally, I really struggle to get people negative feedback. And when I'm in a management role, I really push myself to do that, but it's hard. And I think, yeah, I don't know. I really admire some people who can just be like, yeah, that's nonsense. And just really...
[0:24:38] Really, frankly, say it. And I think, you know, ideally going to couple it also with support that, you know, Holden's message wasn't just, you know, this is a dumb idea. It was, you know, it doesn't seem like a very good idea, but you're developing lots of good skills and, you know, you'll probably do something useful in the future.
[0:24:55] Yeah, I've tried to play this role a handful of times. I've heard someone's plan and just been like, no, this is terrible. I really got to change it because I don't see this having any impact. I've gotten a mixed reception, I think it's fair to say. I'm maybe a little bit more cautious about doing that now. I don't know. Maybe I should go back. After this anecdote, I should go back. I think it probably took Holton a few years for me to totally come around on him not just being a little bit annoying at the time. Yeah. Did he pitch you on going into machine learning? Was he like, this is machine learning is the future? No, not at all.
[0:25:23] And I was completely separately excited about machine learning from other stuff at the time. Okay, so Holden, among other factors, convinced you that 3D printing maybe wasn't the most important thing to be working on. And then independently, you were kind of getting excited by machine learning, which I guess was beginning to take off at that time? Yeah, or deep learning was beginning to take off at that time. Yeah, interesting. And yeah, I guess the thing that I just couldn't get over and I was really obsessed with was we have these systems that...
[0:25:51] Can do these things that no human being knows how to write a computer program to do. And no one knows what's going on inside of them. And that question just really hooked me. And I couldn't get it out of my head and sort of became a big motivator for me. Interesting. Yeah, I suppose I've just always taken it for granted that neural nets are these combinations of nodes and weights and like it's a bunch of numbers and it does some stuff and we don't understand how it works. But that is, from my point of view, that is crazy. It's crazy. Yeah, no, it's completely bonkers that we have these systems.
[0:26:21] In any other area, you'd say that it's harder to go and create a system that automatically builds your systems for you than to go and get the system that can do the thing that's hard. But in this case, we have no idea how to go and produce the system. And people have certainly tried. People have tried for essentially all of the machine learning tasks we talked about. People have tried to go and hand-build systems that do this. We can't build systems that can go and do the things that neural networks do. Even when we try really hard to do it by hand, yet somehow they automatically form.
[0:26:50] It's just truly a crazy fact of the world. Yeah. So I guess we're making them indirectly by programming a computer to say, well, try to do this thing. And then when it fails to do it, we're like, well, you would have done a little bit better if these numbers were a bit different in this way. So then we'll try again and then we'll change the numbers again and just keep going until it kind of works.
[0:27:08] But yeah, at the end, we end up with a whole bunch of numbers. And we're just like, well, we don't really know how to make sense of this. We don't even know like what, or at least in the past, as we'll get to, we didn't know what all the different parts did. Yeah, somehow those numbers correspond to a computer program. What's going on inside it? I want to know. Sounds good. Okay, so you were particularly... And 19-year-old Christopher really wanted to know. Nice, okay. Hasn't really changed. And what did you do next in order to get into this? I suppose...
[0:27:38] The idea that someone now could get a job at one of these AI labs just with a high school degree. I mean, maybe, I don't know whether it was easier then or whether you just particularly stood out, but how did you manage to get your foot in the door? Because people would often think that that's the step that's potentially really hard if you've taken this unconventional path. Yeah. Now I should say, I'm not the only one who sort of has less of, you know, doesn't have credentials and is at these labs. Like I think Alec Radford doesn't have a PhD. I'm not sure that he has an undergrad degree.
[0:28:05] And there's a number of other researchers who are like us, so I'm not unique in it. For me, I was really lucky that I got to work with Michael Nielsen. And I think he just basically treated me like one of his grad students for six months or a year. And that was a really helpful experience and I think just helped me mature a ton as a researcher. And then I started cold emailing different labs and asking if I could visit. Joshua Bengio put out a call for PhD students and I wrote him, I sent him out a week writing him.
[0:28:35] Yeah, he ended up considering me for, and actually I ended up getting accepted. I didn't end up going. And so I started, yeah, I started visiting different research groups and giving talks on research that I'd done. And I'd got some, I think, very modestly interesting results. And I think in a lot of ways, the field was, it was much smaller and it was growing, you know, I think relative to its present size, it was growing very rapidly. And so that also did make it easier to get into.
[0:29:04] Yeah, interesting. So I guess your first serious role was an internship at Google Brain. Is that right? Yeah. So I gave a talk at Google and Jeff Dean sort of very kindly offered to take me on as an intern. So I ended up doing an internship at Google and it went on and on and on. And I ended up being an intern for an entire two years. Is that common? No, apparently I'm not.
[0:29:33] The person who did the longest internship in Google's history. But I think I was a good competitor for that position. Were you doing this extended internship kind of in lieu of doing other formal training? So was this the place where you kind of learned all the things you need to know in order to get properly hired? Yeah, I think I learned a lot through it. I also learned a lot from Michael and from self-study and stuff like this. But yeah, I think there's probably a significant element of that. And I mean, I was just so lucky to be surrounded by so many.
[0:30:03] Really, really wonderful people who were really generous with their time with me. And especially with sort of doing different stuff. So Google Brain was about 30 people at the time. And rather than like building neural networks to go and do different things, I mean, I did some of that and I got to be involved in lots of projects and some work with generative models and contributed a little bit to TensorFlow and things like this. But the main thing I was doing was just, you know, trying to figure out what was going on inside these neural networks. And really...
[0:30:31] Very few people were going and thinking about that at that time, I guess. Matthew Zeiler did a little bit of work, although he sort of stopped. Alex Mordvinsdev, who would become a close collaborator of mine, was doing some work in this space. And I just spent a lot of my time writing blog posts about machine learning and trying to visualize them and trying to understand what was going on and trying lots of things that didn't work very well. Yeah, what were some of the highlights or the most interesting things that you did? And I guess you were at Google Brain for six years or something like that.
[0:30:59] Yeah, yeah, five-ish, five-something years, maybe six. Yeah, what were some of the highlights? Probably the craziest thing was Deep Dream. So the really early Deep Dream results, Alex Mortensen discovered, and he gave a talk, and I was just so excited. I sort of dumped all the projects I was doing and got involved. And it was just sort of electrifyingly exciting.
[0:31:23] So I guess the idea is you just try to make the neurons in a layer of a neural network activate by going and optimizing the input image. And you can have these sort of, if you haven't seen them, these sort of hallucinogenic images that are full of dog slugs and psychedelic colors and things like this. And neural networks seem like magic, but this seems like really crazy magic. And I just spend hours and hours and hours trying different things and trying to visualize different layers and trying to go and fiddle with what we were doing in different ways.
[0:31:51] And I found that there was this, you could apply this to going and visualizing individual classes. And we discovered that like part of how it recognizes a barbell is looking for arms attached to the barbell. But it was just this like really, you know, exciting moment where just of like, I guess just of like really intense mystery. And that was really cool.
[0:32:10] Yeah, I feel like everyone has seen the Deep Dream images. I imagine the audience for the technical paper was a lot smaller than the audience for the images, but it was an incredible hit. Do you want to just describe how those images are produced? It seems like you're kind of reversing the neural network and then getting it to spit out an image rather than classify something. Yeah, so the idea is that you go in, you sort of fill in with the input image to get an image that causes the neurons in a given layer to fire a lot.
[0:32:39] And as neurons begin to fire, you try to go and cause those neurons to fire more and more and more. I see. Later on, we'd come up with this. I guess one of the things that I always feel a little sad about with Deep Dream is we got all of this attention initially, but I feel like we didn't really understand the results in some important way. We didn't really understand what we had discovered. And I sometimes think about it as like...
[0:33:00] At some point in the invention of the microscope, somebody must have like, you know, found a distorted piece of glass and realized that they could go and hold up things and they sort of see these distorted images, but the distorted images have these parts that are enlarged and they can see things that they couldn't see before. And I sort of feel like Deep Dream was that stage. And then later we turned it into feature visualization and sort of developed, and of course this wasn't just our work, we were building on work that other people did as well, that allowed us to sort of have these lenses that allowed us to very clearly see and sort of really understand.
[0:33:28] Be a very powerful tool for understanding different parts of neural networks. But we didn't really get that. That came after the attention from Deep Dream. Was the flood of attention helpful at all for getting taken seriously or getting more resources? I don't know to what degree the ML community respects creating a tool to make amazing images that people can stick up on their social media. I think people were really excited for a month or two. The thing that I remember most was getting lots of emails from people in psychology.
[0:33:55] Who wanted to investigate this with us. I remember getting an email from some really serious looking professor about how they wanted to investigate these themes similar to psychedelic images. Could we do some sort of joint collaboration to figure it out? And never followed up on that. But I think that a lot of it wasn't as helpful. And maybe just because we, from my perspective at least, didn't yet really understand the results deeply. We weren't really able to tunnel it into something that was really helpful for...
[0:34:24] Advancing an agenda of really understanding neural networks, I think. What was it like working with so many people who I imagine were a whole bunch older and more senior than you? It was an interesting experience. Yeah, so for a while, I was the youngest person because even the interns were PhD students who were quite a bit older than me. And actually, at the time, it was only 50 people and several of my colleagues had children who were my age. That was also kind of fun. But actually, just like everybody was super sweet and was just really...
[0:34:53] Really lovely to work with. I think the thing that was slightly challenging, and it became more challenging after I became full-time, was trying to interact with my peers. So I'd have experiences, for instance, where I'd try to hang out with PhD students, and then they'd be like, ah, Dr. Ola, because everybody assumes that I have a PhD. Can I be your intern? Or something like this. And it actually, I don't know, if you're going into an environment really hoping that you can sort of have...
[0:35:20] Have pure style interactions and become friends with people. It's actually really disappointing when that happens in a pretty strange experience. Interesting. I suppose if you spend all your time hanging out with 30 and 40-year-olds as a 20-year-old, do you just end up talking a lot about, I don't know, male patent boldness and how to buy a house? You're like, hey kids, let's talk about our kitchen renovations or something like that. I feel like the things that people talk about are potentially quite different. I think...
[0:35:48] I don't know. Mostly I spoke with people about research. I tried to have a few conversations about effective altruism. I think people were just kind of confused by me. Nice. Okay, so you're spending these five years at Google Brain. And I guess you had a whole lot of very widely read articles there. And you also founded this academic journal, Distill. Do you want to flesh out maybe the rest of what happened at Google Brain and what's happened since then? Well, I think my career trajectory probably became less unusual and interesting.
[0:36:17] Probably around that point. But yeah, I guess after a few years at Google, I started to have a much stronger vision for how interpretability research could be pursued and sort of a very idiosyncratic one. And I really wanted to try and build up a research group doing that type of interpretability research. And I guess I also was very interested in this idea of sort of open notebook science, or I didn't really see any downside to being very public about interpretability research. And I just thought, you know.
[0:36:46] It would be nice if we could just share it as we're doing it and try to involve other people. And I talked a lot with my manager and Jeff, and they were both actually really incredibly supportive. And we just sort of all agreed that actually Google probably wasn't the best place for me to go and pursue that. And so I ended up going to OpenAI and founding a team, the Clarity team, going and doing this kind of interoperability research there. And that was a really positive and...
[0:37:14] Yeah, it seems like you've kind of tracked, I suppose, this clarity or interpretability field didn't exist kind of when you joined. You've like virtually founded it, or at least you were there kind of around the founding. And now you've lived to see it become a meaningful fraction of research into ML and something that's like, I guess, widely accepted and people are interested in. Well, I think right now, a lot of people sort of are doing lots of different things called interpretability research. And lots of other people contributed to sort of, I think,
[0:37:42] You know, creating lots of different branches and directions in the space. But I, yeah, I guess I've sort of had a very particular vision for what the type of work I'm most excited about in the spaces. And I've been really fortunate to be able to help build that out. And I think just really lucky to be involved early on. And so I, yeah, I got to work with a bunch of really amazing collaborators at OpenAI. And we sort of pursued what we called the circuits agenda, which was the sort of, yeah, attempt to go in.
[0:38:10] And sort of fully understand neural networks. Yeah. All right. Let's pick up and look at your, I guess, the kind of unconventional career track that you've been on in general, I guess, from 2008 through 2015. Yeah. What do you think we could learn from this experience? Are there any lessons that listeners can draw as to their own experience? Or is it maybe they should go and defend one of their friends from terrorism charges? Is there anything that people could learn? Or is it just too weird? Well, okay. So I think probably the most useful thing I've extracted has been thinking about
[0:38:39] I think you had the Pareto frontier of skills. So for example, a lot of my early contributions to machine learning were basically being able to create these really helpful illustrations of complicated ideas. What skills did I need to do that? Well, I needed both to understand machine learning and I needed to go and be able to draw. And I wasn't an exceptionally good artist or scientific illustrator. And I wasn't exceptionally knowledgeable about machine learning. But very plausibly for a while, I was the person in the world.
[0:39:09] Who was the best at the intersection of machine learning and drawing? And so if you sort of think about, if you sort of think of these two-dimensional plots of different skills or three-dimensional plots of different skills, and you think about the Pareto frontier, very often society is good at producing people who are optimized for a particular skill or set of skills that society has really validated as useful. And we like create entire like pipelines training people. But I think that often if you can find useful intersections of skills that aren't sort of these couple of standard skills.
[0:39:36] There can be a lot of value and it's much easier to go and have a big impact and often have a big counterfactual impact. And so I often think a lot and when I'm talking to people about their own careers, I often try to frame it in terms of, you know, what are the skills that they're cultivating and sort of what do we think the Pareto frontier with regards to these skills looks like? And do we think that there's places where rather than going and becoming the world's best at one skill, they could produce a lot of value by...
[0:40:04] Yeah, that's really interesting. I guess thinking about it theoretically, I suppose part of the reason is just that there's like so many combinations of two different things that you could throw together. And so that the space of possible combinations is like vastly larger. And so you have a lot more to choose from. And it also means that you could be like the only person who's interested in X and Y if you choose two things that are sufficiently distant. And then you have like a truly unique skill set and you might just stumble on something that no one else has even tried to find. Exactly.
[0:40:33] And now the problem is the space is exponentially big. And you want to not just find an intersection, but the intersection has to be useful. So you have to have some taste in sort of picking the skills that you develop. But I think that there are lots of opportunities like this. And that often, yeah, it's just, it's much less competitive than going and being good at one of the skills that society already really values as a thing to optimize for. Yeah, so I guess, so you could choose the wrong combination, just whether, yeah, there aren't that many complementarities between the two options.
[0:41:02] I guess you also kind of might fall between the cracks, I suppose, of existing disciplines or like a common complaint that people have about doing interdisciplinary work is that, you know, if you're doing, you know, philosophy of economics, the economics department doesn't like you and the philosophy department doesn't like you and no one really feels like you're one of them. Yeah, I guess I want to distinguish this a little bit from interdisciplinary work, which I think is something slightly different. Like when I was doing this scientific illustration of machine learning, it was really a pure machine learning contribution. Like it was something that was...
[0:41:29] Valuable to the machine learning community and targeted at the machine learning community. And so I think that there is a distinction, I think, between... Skills and maybe bodies of knowledge. Yeah. Or you could also do something that was more cross-disciplinary. You could use machine learning for scientific drawing or something like this. But I think that's not really what I was doing. Yeah. Okay. So I suppose we might be an example in that where at the intersection of being knowledgeable about...
[0:41:56] I have a favorite example of this phenomenon, which I'll probably both slightly mistake, and I actually owe to Michael Nielsen. But my understanding is that Richard Feynman, I guess when he...
[0:42:24] I don't know much about physics, but he had to do all this work, and I guess physicists were doing all this work, going and solving really complex integrals. And the usual set of techniques was to go and use tricks from complex analysis, analytic extensions and stuff like this, and try to go and solve the integrals that way. And Feynman didn't really know these complex analysis tools very well, but he had all of these weird tools around fractional calculus and stuff like this, and he used those instead. And by having a sort of...
[0:42:52] Maybe it's not even that weird a skill set, but in having a different skill set than his colleagues, he was able to have more counterfactual impact and go and solve problems that other people couldn't. And it's not that his tools were better. It's just lots of people were already trying with the other tools. And he brought a different set of tools to the table.
[0:43:07] Yeah, so you found some money lying on the sidewalk. I guess some options might be, you know, knowing a lot about some technical area, plus say knowing about operations in organizations or knowing about how to do business or knowing about money, knowing how to manage people. Maybe those are like combinations. Yeah, I think people management plus technical skills is a huge superpower. I think it's something that I am trying to become good at. I think the people who I see who are really good at it, I think are, yeah, it's a really amazing thing. Yeah, they end up very sought after.
[0:43:34] Yeah, I think any sort of communication plus technical skill. I think web development plus science is actually really underrated. I think that often being able to build interactive interfaces can allow you to go and, well, I guess the basic pitch is, I think a lot of scientists are drawn towards very reductionist, and maybe this is more true in machine learning than other fields, I'm not sure, but they tend to go and look for summary statistics because they can sort of easily work with summary statistics and make line plots and things like this.
[0:44:02] And I think if you instead are able to go and create interactive tools and explore things, you tend to just sort of interact with the data in a different way. And I think there's actually just something where, at least in machine learning, I think there's a lot of value that gets left on the table and I suspect elsewhere. Yeah. Do you think that this might be slightly a Bay Area phenomenon? The thing is like people really appreciating people who have quirky skills that are combined. And that maybe if you're in a more conservative social situation, maybe it'll be more risky or is that right? Yeah.
[0:44:32] That might be true. Although I think in a lot of cases, if you can demonstrate that your intersection of skills produces value, I think that's really the critical thing. Once an organization is getting value out of your intersection of skills, whether it's weird or not, probably isn't going to be the critical thing. Yeah, the make or break issue. Yeah, I guess that's maybe one other lesson is that if you can show people that you can do stuff directly, then often you can route around credentials. And I think that is...
[0:45:00] That is quite often true. The thing is, it actually is potentially quite hard to figure out how to produce value and how to do a good job without having the training. So it requires someone who really has either just a lot of raw ability or a lot of focus or a lot of discipline to do things outside of a structured environment. Absolutely. And I think it also depends a lot on the discipline. Practicing law without credentials isn't something that's going to fly. So I think depending on the discipline that you're working in, and I guess this is sort of cutting the other way from my previous answer, I think...
[0:45:29] How flexible the discipline is. But I think this is actually a little bit of a different question than the intersections of skills thing. You can get a PhD and have an odd intersection of skills, perhaps. Yeah, I think probably people aren't just going to learn medicine through apprenticeships. For understandable reasons, it's a somewhat credential-filled field. I would be very nervous to go and have a doctor who did not have an MD. Yeah, I guess it reminds me of that you can get a free haircut if you're willing to be someone's first victim as a receiving haircut.
[0:45:58] Perhaps it's more difficult in a surgical environment. Yeah, we've talked a bunch about whether people should go and do undergraduate degrees, but how do you feel about going to grad school? Is the picture very different? Yeah, so again, I don't know how much my advice generalizes between fields, but at least for machine learning, I usually encourage people to just ask where they can go and do the best research. So I guess for context, the question that I usually hear people...
[0:46:21] Asking when they're considering going to and doing a PhD is they're like, well, should I just like go to an industry lab and do research there? Or should I go and do a PhD and go and develop my machine learning there? And I think at least for them, the essential thing isn't whether it's a PhD or not. The essential thing is how much do you think you're going to learn from the people you're going to be around? And just how good a research environment is it going to be? And so for some people, I think quite often going and doing a PhD will get them more mentorship.
[0:46:49] And get them into a research environment that's more suited to their tastes than the opportunities they have in industry. But for other people, it's the other way. And I think it's probably better to try and compare concrete questions. I mean, I guess a very classic advice that people give about doing PhDs is to think really hard about the group that you're going to be working with and try to understand them.
[0:47:10] It's just like you should compare specific research groups when you're thinking about doing a PhD. You should also, if you're considering industry as well, consider the particular industry group that you would be working in and what you would get out of working there. How snooty is ML about whether you went to the right university or whether you were working at the best lab? I think not very. I mean, it probably depends from institution to institution. But I think most institutions, if you have impressive results, that's the main thing they're going to care about.
[0:47:39] I actually had one university encourage me to apply for a professorship at one point. And I was very surprised by this because I don't have an undergrad degree or a PhD. And, you know, I sort of think of universities as being very, very traditional. And so I think, you know, if universities are willing to consider stuff like that in light of, you know, when people do good research, I think that sort of is a pretty strong signal that, yeah, that a lot of institutions are willing to focus on people's research rather than their employment history or where they did their PhD.
[0:48:08] Yeah, it seems to be like maybe ML as an academic discipline is pretty unusual. It seems like it's routing around a whole bunch of the typical norms. Like most research is now just presented at conferences, right? And then just people... That's not unusual for machine learning. That's just all computer science. Ah, that's all the computer science. I see. Okay. Yeah, computer science by and large doesn't use journals as much and relies a lot more on conferences. Ah, interesting. Yeah, do you know how that happened? Not sure.
[0:48:36] Yeah, I do know it leads to strange situations in academia where somebody who's looking at the career of a computer scientist might be like, oh, they just publish in conferences. They're not doing good work. In fact, that's just the norm. Yeah, interesting. I think there's some areas of physics. Machine learning, I think, also has a significant fraction of people just publish on archive and not publish in a venue at all. But I think there's areas of physics that have taken it much further. Wow. My impression is there's some areas of physics that...
[0:49:05] Just like people aren't really using, like by and large, aren't using conferences or journals or any peer-reviewed venue and are just putting things on archive. Wow, they're living the dream. They're living the dream! So I've sometimes said, you know, I'm really not sure what value academic publications are providing because it doesn't seem like peer review is doing that great a job at clearing out bad papers.
[0:49:27] Presumably does something. But from what I understand, the test retest validity of peer review isn't that strong. So even if you submit the same paper to a journal twice, it's like there's a high chance it'll be rejected once and accepted another time, which is a bit of a red flag. Yeah, NeurObs did this test where they peer reviewed some subset of papers twice. And they found that if one review process accepted it, the other review process had a 50% chance of accepting it.
[0:49:53] Which because they're accepting smaller, less than half of the papers. So it's not as bad as it sounds, but it's still a very noisy process. Yeah. So, I mean, it seems like if these fields have kind of dispensed with traditional journal system, then we could maybe learn from them how much value the journals were providing in the first place by seeing whether this has been a disaster. Maybe this is something I should look into. Yeah. So I know you've written this article about how people can develop good...
[0:50:21] Taste in what to research and how to go about it. I actually didn't have time to read that one to prepare for this interview. Yeah, maybe could you give people a summary of your views there? Sure. So first of all, I don't consider myself at all an expert on this. This is just sort of my essence for what's worked with me and when I've been mentoring people, things that I've found helpful. But I think it's often helpful to divide being a good researcher into two parts. One is taste. So your ability to go and pick good problems.
[0:50:48] And go and pick good avenues to attack those problems and things like this. And the second you might call technique or execution. And, you know, I think, you know, maybe if you like picture a chemist working with vials and pipettes and weird things, it's sort of pretty clear that there's like a sort of a whole technique to going manipulating that laboratory equipment. And I think that it's subtler in other fields, but I think that there is something in certainly in machine learning of the technique of training models and even just of being a good programmer and doing very...
[0:51:17] Minute things of manipulating your code editor or going and manipulating distributed systems and stuff like this. And so I think that's sort of a question of how do you develop both of those skills? And for taste, I think that's probably the hardest one to develop. I tried to come up with a sort of list of exercises that one could do. And an example on it, I think probably the most useful one, is just write down a list of problems that you think might be important to work on. And then have somebody else, ideally your mentor, go and just rate them.
[0:51:46] One of the really hard things about developing taste is that you have such a slow feedback loop on learning lessons because you have to go and do the entire project normally. And so what you want to do is use a mentor or use somebody else as sort of a cheat proxy for getting feedback. And then if you disagree with their feedback, you can either talk to them about it or maybe you even want to go and do that experiment. So I think that could be useful. And I think there's lots of other things. I think reading about the history of science is helpful. I think going and trying to write just about why you think things are important is helpful.
[0:52:14] So in any case, I think there's a bunch of exercises there. And then on the technique side, I actually think the most valuable thing here is working closely with people who have good technique. And I think actually, at least in machine learning and probably other computer science disciplines, going and pair programming with people is immensely valuable. I think that there's a lot of stuff that's hard to communicate in other forms that gets passed along when people are pair programming. And so I think for developing technique, often pair programming is the highest leverage thing to do.
[0:52:43] It's really interesting. I'd noticed, so I guess in fields of work, it kind of tasks where they have a physical embodiment. We're actually moving things around in the physical world. People get to see one another and they get to learn from watching them what they're doing and figure out how to do it better. And for work on computers and laptops, that exists much less. Exactly. Because there just seems to be a culture in general that someone, like when you arrive at an organization, you're trying to get training, for example, from your manager.
[0:53:10] You don't literally sit behind them all day and watch what they do. I mean, that would be like, maybe that would be sensible. But you can't just sit behind them and watch their screen and then see like, how do they move the windows? How do they reply to people? That doesn't happen. And I guess that means that it's possible for people to just miss really basic stuff potentially. And it sounds like maybe in programming, there is this peer programming thing in part to fill this gap because it's maybe such a severe problem there.
[0:53:34] Yeah, I mean, I think there's an increasing culture at a lot of organizations of programming. I feel like I hear people talking about it a lot more. And I, yeah, I've found that it's really helpful for passing this sort of stuff along. And I myself am constantly learning from other people when I work with them. And I, you know, I hope that they're learning from me as well. I think your point about how it's hidden by, you know, when you're not, when it's not physical, sort of technique gets hidden by that is a really good one. Yeah, I think that culture exists in part because people are worried about,
[0:54:04] Well, I guess two reasons. One is sheepishness, perhaps, about people disagreeing with how they're going about their work. It's easy to just hide it and never have people watch your screen. Another might be that you're worried about confidentiality. You don't want other people looking over your shoulders and reading your emails. There's a real norm of not looking at other people's screens in general. But it seems like maybe we should think about ways to work around that. You'd have a specified time when someone literally is just going to watch you work, and then you try to not do anything that would be too sensitive, where they shouldn't be looking at the screen.
[0:54:29] I think I'd be fascinated to see how my colleagues just go about their day. Do they switch windows as often as I do? I don't know. It's key sometimes, some of these basic skills to your productivity that could be worthwhile. Well, I think it's also not just teaching. I think often you can just push through things faster if you have a second person with you. I guess Jeff and Sanjay are famous for being extremely impactful at Google and pair programming all the time. Ah, so they sit together with their computers next to one another and they just work together on a problem. Yeah.
[0:54:58] That's really interesting. Yeah, do you have a theory for why that's not more common? It seems like it might be just a really sensible way to get things done. I think it is modestly common in programming and software engineering. And I guess the thing that I'm trying to highlight here is that it's a really useful, like I think people sometimes feel just like, oh, it's a nice way to work, I think, or it's like an effective way to work. But I think, especially if you're trying to develop technique, I think it's the best way to go and transmit it that I'm aware of. Yeah.
[0:55:25] I've seen in your articles that you're also just generally a big proponent of writing cold emails to people. You found that that's worked pretty well for you. Do you think that generalizes to others as well? I get a lot of cold emails. 99% of them are terrible. They are, can you do my homework for me? Or like, you know, can you answer this basic question that I could Google for like one minute and answer? And so I think people get this impression that cold emailing doesn't work because of course, if you send emails like that, people are overwhelmed and aren't going to respond. Or even if you just like very generically are like,
[0:55:55] You send a nicely written email and you're like, I'm trying to get into machine learning. Can you do a half an hour phone call with me to talk about how to do that? Even that, you're not very likely to get a response from. But I think the thing that people miss is that if you write really good cold emails, it's actually not that hard to be the best email I received that week. And I think that if you're willing to invest energy in understanding what a researcher or a group is working on, and you're specifically referring to their papers and...
[0:56:24] You have, you know, you have thoughtful questions at things. Yeah, I think that people pay a lot of attention to that. And I think that it will, it very often works well. And so I think there's a big gap in what people mean when they talk about quality emails. And I think that if you're willing to put in the work, and I mean, if you just genuinely really care about what somebody is doing and have put in the work to understand it and can talk about it really intelligently, that's going to come through. And is, you know, it's a much more compelling reason for the person to talk to you than other things. Right. So it sounds like...
[0:56:53] You don't think that people should write tons of cold emails to all kinds of people. But if there is someone whose work you're really into, whose work you really understand, then you should not be sheepish about emailing them. Because even if they're getting other emails, your one is really going to potentially stand out if you can demonstrate that you have actually read their paper. Well, and I think the other thing is, I think there's a lot of people who are trying to look at how to get into machine learning. And what they do is they send lots of emails to people or they email famous people. I think what you should actually be doing is trying to figure out who you would be really excited to work with.
[0:57:21] And really understand their work. And ideally, do somebody who's a little bit less famous, maybe. And then reach out to that person with an email where you've put a lot of work into sort of explaining it being clear that you've read their work and like connecting your interests to theirs and things like this. And so there's a number of emails that have been really important for me where I spent like a week writing them. And I think that was a totally worthwhile investment. And I think that's not how people usually think about cold emails. That's so interesting. Yeah. How do you feel about...
[0:57:48] Length. Maybe we're going to get a little bit into the weeds of email technique here, but go on. A lot of, yeah, I think a lot of the most impactful emails I've written were only a few paragraphs long, you know, less than one page. But I read like five of that person's papers beforehand. And, you know, that, yeah, I think that comes through in subtle ways. And I didn't integrate it super ham-fistedly, but I just, you know, I was writing to them because I genuinely was invested and cared about their work and had shared interests with them.
[0:58:18] Yeah, I think that's very, very different. Yeah, I think the main lesson that I've learned, well, I suppose this is a different class of cold emails is kind of asking people for small favors or for feedback or offering advice. It's that the shorter it is, the more likely people are to, certainly the shorter it is, the more likely people are to answer. And maybe also if you can really condense down the information that you want to convey into like just a couple of sentences that people are much more likely to absorb it. Because people are flicking through their email inbox pretty quickly and often they have like other things going on.
[0:58:46] If they open an email and it's a wall of text, then you definitely run the risk that they're just going to close it and then never get back to it because it's just too much. And they don't yet know whether it's really worth investing the time in. Yeah, I think you really want to, if you're writing something longer as well, you want to really optimize the introduction for that reason. Yeah, yeah, yeah. Yeah, do you have any thoughts on how to be a successful researcher? I guess there's one last thing that I find pretty helpful, which is thinking about research as a market. And I don't think this is very novel to me, but...
[0:59:15] I do find it a really powerful frame. And so I guess the general idea is you think of researchers as investing in different research ideas. And if the research idea pans out and other people don't sort of grab it before them, then they get some reward from that, maybe more resources or just they get a payoff from that in some way. And so you can sort of see there as being this sort of competition to go and grab promising research ideas.
[0:59:45] And I think there's roughly two strategies that you can sort of play in this market. One is you can sort of work on things where everyone really agrees that they're important and that are really popular. And what you're doing when you do that is you're going and making that little area of the research market a tiny bit more efficient. You know, you're going and making it so that ideas that sort of are important get done a little bit more quickly. And I think that is actually genuinely a valuable thing to go and do.
[1:00:13] Like if the thing that you're doing is really important and you make it happen in expectation a week earlier or a month earlier, that's really great. But the other strategy you can do is you can try to beat the market. You can try to work on things where you just can see that something is undervalued relative to what most of the community thinks. And I guess that's the thing that I try to do a lot of the time. And there's lots of reasons why you might be able to beat the market. It could be that you just care about things that other people don't. So if you care about safety,
[1:00:42] Or in other areas, if you care about animal welfare or if you have weird goals or different goals than a lot of people, you might be able to beat the market in that way. I think another way, though, is just sort of if you have some insight that you really believe is true about a problem and that's not a widespread insight, that could be really helpful. And that can be, I sort of, yeah, I feel like that's a lot of what I'm doing. And sort of me personally, I think that you can genuinely understand neural networks if you're willing to put enough energy into it.
[1:01:12] So trying to figure out what's going on. And it's like a big bet that I'm making that most other people aren't making. Yeah. I guess to translate this into 80,000 hours speak, one option would be to go with the thing that kind of everyone agrees is most important or like has a particularly large impact if you can make progress on it. The problem there is that it's probably not going to be neglected because everyone is onto it already. Another option would be to take a punt on something that...
[1:01:35] is currently really neglected that other people aren't working on. But of course, there's a possibility that other people who have chosen not to work on it may be right, that it's too hard to make progress or it doesn't really matter even if you do. And I suppose, yeah, the ideal would be that you have some underlying insight that other people have missed. Maybe they just don't have time to engage with it because there's so many ideas out there. Lots of things get ignored. And then you can find something based on that insight that is important and neglected and easy to make progress on. And then if you're lucky enough to succeed in that, then you have a maid, right? Exactly.
[1:02:04] Yeah, do you think you kind of struck gold in that way with working on interpretability? It seems like it was a field ripe to develop. Well, I think I did. And I think it also played a lot to my comparative advantages. But yeah, I think I did. Other people might still disagree. I think the jury's still a little bit out on how valuable it is. But I think I did. Yeah. Let's push on and talk about explaining complex things really well, which I know has been a passion of yours for many years.
[1:02:34] To start, why is it essential for a field to invest in really good explanations of things? I think in many fields, achieving a research level of understanding is like climbing a mountain. You know, there's all of these ideas that you have to understand and build up towards before you can go into research. And, you know, mathematics, I think, is a really striking example of this, where there's just years and years of ideas that you're probably going to go and spend climbing to the point where you can go and do research, because there's just so much that has been piled on top.
[1:03:03] And then when you get to the top, you go and you pile some more results on top and you make the mountain higher. And I think a lot of people are proud of this because they're sort of like, the fact that it's like this long pilgrimage to go and get to the point where you can do research. And that means that it's especially profound and it reflects all of the work that's been done to date. But I think that actually it's often a reflection that we haven't put enough work into explaining things and going and building up really good infrastructure for learning.
[1:03:31] For learning about that field. And I think this can come in lots of forms. It can be poor expositions. It's just not good explanations of things. Sometimes it's just undigested ideas. It's an idea that's important, but it hasn't been really refined to the completed version of that idea. I think it's very common for there to be bad notations or just bad definitions of things that make things more complicated. And all these things make it harder to go and understand the topic.
[1:03:57] One analogy that I like is sometimes in software engineering, people talk about technical debt, which is like, you know, you go and you move really fast to go and get to that point where you can like go and shift some feature or something like this. And in the process, you write lots of bad code and it's really messy and gross and you have bad variable names and it isn't documented. And then it's hard for other people to build on top of. And I think something analogous, a kind of research debt is sort of endemic in science. Yeah. So I guess this problem is kind of...
[1:04:24] Probably just getting worse over history because... Yeah, it's just getting worse over history because we're just piling the results on. I mean, sometimes people do write beautiful tutorials and textbooks that make it a little bit better. But I think on net, it's getting worse. And it tends to be that the older a field is, the worse this problem is. Yeah. So I guess the...
[1:04:44] Like at this point in many academic fields, if you want to really reach the frontier, you have to wait until you're 30 and you've done PhD and then a whole bunch of extra stuff. I guess in some fields, maybe it's like 35 or 40 before you can actually start contributing. And that's a reflection of the fact that, one, there's just like a huge knowledge base. And in some cases, in some fields, I think we're kind of pushing up against what it's possible for the human brain to do. And so you just have to really reach like your peak capability before you can add in anything new.
[1:05:09] And it seems like the basic concepts end up being distilled really well because they're taught in primary school, in high school. People figure out how to communicate those, but then the closer you get to the peak, the more it's just a mess because no one's really figured out. There is no textbook maybe about this, about the most recent results. And so it becomes slower and slower maybe to reach up to the top because everything becomes inscrutable. And I think it's often less incentive to go and work on exposition. This is my impression. There's a really nice article by Thurston before he passed away.
[1:05:39] Where he talks about basically, in my framing of it at least, sort of killing a field by just going and picking all the low-hanging fruit and then dumping the field and just like all of these results that he didn't explain and sort of making it totally inaccessible. And, you know, I think, I guess that's sort of an extreme example, but I do think this kind of thing can, you know, genuinely hold back fields. Hold on, so you're saying that someone could just dump a whole bunch of, well, they'll dump the next step, but explain it so badly that people are repelled from even comprehending it.
[1:06:09] You have to do several next steps, but you can just pile on a huge amount of progress and not communicate it at all. And now it's like sort of a dead end where like you can't go and get sort of credit for going and redoing that section of the field and like all the fruits have been picked and you can't build on top of it because it's really hard to understand. Yeah. That is a fascinating possibility. We're trying to hear your first question for a second.
[1:06:34] I think that's another thing to be said about why it's important to invest in good explanations, or there's something about why good explanations are valuable, which is the impact of an explanation is often nonlinear, where as you go and you write a better explanation, not only does it provide more value to its readers, but you also have more consumers. And so I think something that's sort of surprising is you might think that in a lot of things, as you make something better, there's really sharp diminishing returns.
[1:07:03] And of course, there are diminishing returns where it becomes harder and harder to make something marginally better. But I think with explanations, there's also a nonlinear thing where as you go and you make something, an explanation better and better and better, and it becomes better than any existing explanation. Actually, its value significantly increases in this very nonlinear way. Because that becomes the default reference that everyone's going to read. And so they're all saving a bunch of time because it's better.
[1:07:28] And maybe also more people are like, oh, I could plausibly understand this because now it's been explained properly. So you end up expanding the interest in the topic in aggregate. Sometimes I feel a little bit sad about this, but the thing that I've written that's been most read is this tutorial on LSTMs. And it's been read several million times. And so if you just like, if I wrote it a little bit better, such that I saved every reader one second going and reading it, it actually adds up to like non-trivial amounts of time that I saved lots of people. And so just like, yeah, when you have a million multiplier on something that actually...
[1:07:57] That really has a big impact. And the difference between having a million multiplier on something and having a hundred or a thousand multiplier on something can actually be sort of a pretty sharp transition where you manage to go from being a good explanation that sort of is sort of comparable to existing explanations to an explanation that is better than the existing alternatives. Yeah. Yeah. The key idea here to me as an economist is kind of fixed costs versus variable costs. And here we have a fixed cost in and making an explanation better, where it's like.
[1:08:26] That's just a cost to the person who's writing the article. They're going to have to spend more time revising it. But then the benefits are potentially variable depending on how many people read it. So as the number of readers goes up, the amount of time that it will be plausibly justified to spend improving the explanation just balloons out. If you're going to have a million readers on a textbook, then you really want to make sure that it's very, very good. So it saves them all time. While we're thinking of this in a sort of economics-y way, I think another thing that's interesting to think about is if you think about N people in a field interacting, and you increase the... As you vary the size of the field, you can ask...
[1:08:55] How much effort goes into explaining things and how much effort goes into understanding things if you want everybody to understand all the work that everyone else is doing? And the effort to explain things grows linearly because just each person has to explain their work. But the effort to go and understand things grows quadratically because each person needs to understand every other person's work. And so if you have the option to go and shift some of the, like sort of change the coefficients on both of those where people produce better explanations.
[1:09:24] And then the consumers go and do a little bit less work. You can sort of change the size that a field can reach before it fragments, I think. Where there's some maximum size that a field can have where everybody actually understands everything that's going on in that field. And it's determined by how people explain things because of this linear quadratic cost thing. Yeah, interesting. Okay, so the idea there is if people are bad at explaining what they're doing, then it means that...
[1:09:49] I feel it all fission because people feel like they have to specialize more because it's just so much work to understand what those other people at the other end of the field are working on. Yeah, it's all Greek to them. Yeah, I think that's the natural reaction. Yeah, so what is your philosophy of what makes a good explanation? And I guess what's missing from most typical explanations that makes you think that they're not as good as they could be? Yeah, so I think that there are two parts to a good explanation. One is just having a really clear way of thinking about the topic. And one is executing the explanation well. And it's...
[1:10:18] Hard to give any advice on how to go and have a sort of clear, nice way of thinking about a topic. The thing that works for me is I just get really annoyed with my understanding of things until they feel nice. And so that's what works for me. But I think there is more that one can say about how to execute an explanation well. Okay, so here's something I find kind of mysterious. It's often the case that you have people who are extremely knowledgeable about a topic, and they put a lot of effort into writing an explanation, and they produce an explanation that's really hard for other people to follow.
[1:10:48] And then it gets worse. People tell them that it's hard to follow and they try to make it better. And the result is actually that the explanation becomes progressively worse and harder to follow. And when they look at their explanation, they're like, oh, it's so easy to follow. It's a really good explanation. So yeah, it's very mysterious. Why is that? And I think my hypothesis would be that they have the benefit of two resources that their reader doesn't have. So first...
[1:11:11] They don't have to go and store things in working memory when they read their explanation. They've already memorized all the terms and all of the ideas. And so they can go and write. Oftentimes, as they try to make their explanation better, they add more and more details, and they're trying to be really, really thorough and so on. And the end result is that they're loading up the reader's working memory. I mean, I'm not a psychologist. As I understand it, the reader only has seven or so slots of working memory, and they fill it all up. And then it's really hard for the reader to go and continue sort of pulling pieces in and connecting them together.
[1:11:40] And then similarly, they have all the motivation already. They understand why you should care about the ideas and why you should push through the hard parts. But the reader doesn't have that. And of course, it's easy for them because they've already done the hard parts. And the reader doesn't have that. And so I think these two deficits, the not having the same working memory benefits that the author has, and the not having the motivation that the author does, are often the reasons why people think that they've written a good explanation, and it's not a very good explanation.
[1:12:09] That would be my theory of why explanations fail. So I guess in the first one, what's happening is people say, I couldn't follow this. And what they do is maybe add more detail. They like add even more to the explanation, but then that is actually making it even harder to follow potentially. Perhaps because what they have to do is simplify it rather than make it all exactly precise. Is that kind of one way that things go right? Yeah. Although I think it's not exactly simplification. Like I think you don't want to do is dial an idea down and sort of not actually explain the important thing.
[1:12:37] But what you do want to do is sort of think hard about how you can reduce the strain on somebody's working memory. And I think there's lots of tricks you can do with this. I think diagrams are often really helpful because you can spatially arrange things. Annotating equations can be helpful. Just thinking about like how you structure your explanations that there's sort of fewer long distance interdependencies can really help. More like having little things in the margin where you remind people of important things. So I think there's lots of things like this that you can do. Just like asking if you really need to introduce terminology.
[1:13:05] And I think, yeah, all of those contribute. All right. So you weren't merely annoyed by this. You decided to actually try to do something to help contribute to fixing this problem. Tell us a bit about Distill. Yeah. So something that I find especially frustrating about this is there's lots of really great explanations that people just write as blog posts. And it isn't treated like a real academic contribution. It doesn't help them in getting promoted or...
[1:13:34] Getting hired or things like this usually, at least in academic jobs. And so it seemed to me that sort of the production of this kind of really valuable work, I think explanations and also just sort of interactive visualizations of things, which is another thing I really care about, just wasn't rewarded. And that if we could create a way to go and reward it and support it, we'd get a lot more of this kind of work. So how can we go and create more, you know, how can we go and incent it?
[1:14:00] Well, the idea behind it still was, you know, if we could create an academic journal that could sort of be an adapter between weird artifacts that aren't sort of normal scientific papers and the traditional academic system, that that maybe could allow for people to go and get these career rewards and support that doing sort of more traditional academic contributions can get them. And so we, yeah, we created a journal and that was, yeah, that was our thesis for why it would help.
[1:14:27] Yeah. And how has it gone? Do you think that your thesis was broadly right? Or have you learned something from the experience of actually trying to fix the problem as people often do? Yeah, I think a lot of parts of this thesis were wrong. So I think that the two parts that I don't believe anymore are one that sort of being rewarded is the primary blocker on this type of work being done. And I think it's more the case that people don't do this kind of work because it's hard. And that people who do do it, it's a passion project where they just sort of...
[1:14:57] You know, feel very deeply that they need to go and do this or they're just really excited to do it. And then it's, I don't know, there's just like no realistic world where the, or it's very difficult to get to a world where the career incentives would be sufficient for people to do this kind of work for that reason instead. Like even if you could get this work treated like a normal academic contribution, if your tutorial is treated like a typical paper but takes, you know, five times more effort to go and produce, that's not something that people are going to go and do.
[1:15:25] Okay, so if I understand it, you think you were right that there aren't like sufficiently strong academic or prestige, maybe motivations to work on doing really good distillation of ideas. That was discouraging people somewhat. But the thing that was really discouraging people was that they struggle to do it and it takes them ages. And so if it were easier to do, they might be willing to do it just for the love of it, just because they want to bring knowledge or want to bring water from the fountain of knowledge to everyone else. Well, I think I would say that the people who are doing it right now.
[1:15:53] Are doing it because they have a very strong internal motivation to do it. And then on the present margin, it's hard to create a large enough incentive that you'll actually really change things. Interesting. There's a second reason why I think this doesn't work. So I think previously I thought just like institutions wouldn't reward this kind of work if it wasn't published in a peer-reviewed venue. And it wasn't sort of published in a legitimate academic journal or conference or something like this. I think now it's more that there's some institutions that are really flexible.
[1:16:23] And we'll reward non-traditional work. And they don't care whether it was published in a journal. They're just going to evaluate it on its merits. And there's more traditional institutions that just are going to look at the still and be like, this is too weird for us. We're out. And it's actually a pretty small group that sort of is in the middle enough that this actually makes a big difference. And so I think both of those are sort of make the case for an institution like the still a lot weaker than I initially thought it was.
[1:16:50] Right. So are you guys going to shut it down and try a different method? Or do you think it's like, there's still a niche for Distil and we should keep it around even if it's not going to solve the problem in total? Yeah, I do think there's other ways Distil has provided value. I think that just sort of being an example of what you can do with explanations and with interactive visualizations and interpretability has been really valuable. It's just useful to show what's possible if you try really hard in the space. And I think it's also been useful sort of as a laboratory for meta-science where...
[1:17:19] It's sort of given us a little bit of credibility to do things like organize. We did this weird discussion article thing where there was a somewhat controversial paper and we just got a bunch of people to, instead of doing peer review, they just wrote papers discussing the paper and then we compiled them together and summarized them. And like people spend months going and writing these, what sort of are effectively reviews. And so I think that was a super cool experiment where we're doing this thing that we call spreads, where we go and collect, you know, a series of articles, sort of very incremental articles building on one topic.
[1:17:48] And those seem to have also been quite successful, I think, and really interesting. So those are things that I'm glad about with Distil and that I think are positives of it. But it also has really large costs to run. And I think something that I didn't appreciate enough when I was starting it was just how political it could get, where there's a whole component of people who are upset with us because they sort of think it's corrupt that we publish in our own journal. So because it's such a small niche community, a lot of people who are doing this work.
[1:18:17] The people who are sort of going and investing lots of effort in creating interactive scientific papers tend to be a slightly tight-knit community and either are involved in running Distil or know people who are involved in running Distil. And from the outside, that kind of looks incestuous and corrupt, I think, and people are unhappy about it. And yeah, I don't know. That's like a thing that isn't super fun to navigate.
[1:18:45] Yeah, that sounds tricky. All right. Well, it's, yeah, very interesting. And I think not uncommon that you've gained a lot more insight into what is the actual nature of the problem by trying to solve it. And that maybe, yeah, that's a very different technique now, knowing everything that you do. But it still looks really cool to me. I mean, I was looking around the website in preparing for this, and it's much better than the papers that I normally have to read. Just like visual quality and quality of explanation. So it definitely has achieved that goal. Thank you. Yeah, it's been...
[1:19:15] It's been really wonderful to be part of. And I think something I feel really proud of is in addition to the papers that I've worked on, being able to support lots of people in going and producing these kinds of papers has been really cool and special. All right, well, we're almost out of time for this session. But I guess, yeah, to return to some personal stuff, to get to know you better, if you just had to completely change careers and you somehow became totally indifferent to making the world a better place, what would be the most self-indulgent or most enjoyable career for you to pursue instead, if you thought about this?
[1:19:44] Oh, gosh. Well, one thing I think I'd really love to do is just teach young children math. Yeah, I both often feel, yeah, often wish that I could interact more with young children. And I find teaching really delightful. And I think I would just get a lot of joy out of going. And I sometimes like to daydream about like, oh, how would I go and teach? Like, can you turn group theory into a board game for young children?
[1:20:12] There's these Cayley diagrams and you could go and use them. Or sometimes I'll babysit friends' children and I've done experiments with making knots and then trying to get them to guess whether it's the unknot or things like this. And so I think you could have a lot of fun. And I think I'd have a lot of fun with that. Teaching children is a real job, Chris. This is the most self-indulgent thing you can imagine doing. It's a lot of work. I don't know. I'd want to do it in a very, in a way that's...
[1:20:41] I guess if I was maximizing just for my personal enjoyment, it would probably be a very part-time thing or something like this. Yeah, what's that XKCD comic where it's like, people ask me, what's your dream job? I'm never sure how literal to be. And the person's actual dream job is, I think, removing some lint from a lint cover and then using a lightsaber to damage a door and then retiring to a life of luxury for the rest of their life because they've been paid an enormous salary for these two tasks.
[1:21:10] Well, if I was being selfish and self-indulgent about your potential alternative career, I'd really love to read kind of Chris Ola, the blogger or sub-stacker or whatever it would be, tweeter, tweeter maybe, although you do do Twitter. Yeah, there's just all these really amusing, delightful posts on your website that I had the luxury of being paid to read during the prep for this interview. Yeah, well, when it stands out is people are probably familiar with this idea of the micromort, which is like a one in a million chance of death.
[1:21:38] And then I guess I've noticed recently people have been applying this to more and more concepts. So folks might know that some people involved in the effective altruism community made this wonderful website called microcovid.org.
[1:21:49] Where you can put in all of these, specify all of these things about something that you did, you know, in what country, like how many people, for how long, you know, how crowded, was it indoors, outdoors, in order to figure out how many one in a million chances of getting COVID did you incur. So, you know, if you meet someone who has COVID indoors, then you clock up 100,000 micro-COVIDs, which is a 10% chance of getting COVID. But more day to day, you clock up these, you know, one, two, three, four, five, six micro-COVIDs. And people can use that to kind of judge how much risk are they willing to take.
[1:22:18] Make these judgments that are often so hard to do and these risk reward judgments in a more coherent way. Yeah, and you had a neat application of this kind of one in a million chance thing with the micro marriage where you're talking about, well, you know, as you're going about your social life or just your life in general, you can imagine that the more people you meet, you're clocking up these one in a million chances of meeting the love of your life potentially, someone who you can have a really fulfilling relationship with. Can you lay that one out for a second for us? Yeah, I guess this trick of
[1:22:46] I've going and inventing units to think about things is a trick that I really like. So before we dive in, I guess some context. I wrote that post kind of as humor, but some context is that I actually often find it pretty hard to motivate myself to go to social events. I find them when I don't know lots of people at an event and I sort of can't fall back on just talking about research. I often find that kind of stressful. And yeah, and I often don't want to do it. Yeah.
[1:23:16] Totally. Yeah. But I, you know, I also, I also really want a family someday and find a partner and, you know, that requires you to be, be social. And I guess, you know, going to a dinner party that a friend's hosting or going to some other social event or even going on dates or, you know, dating someone, you know, a lot of these things, the, the, the party doesn't lead to you, to you meeting anyone or any of these other things don't.
[1:23:43] don't lead to where you were hoping they would go. And that can be really discouraging and really hard. And so I, yeah, I've sometimes thought that it's useful to sort of try to be like, well, you know, even though this time it didn't work out, there was actually a chance that that could have led to something. And yeah, you can use that to try to motivate yourself and also maybe to just sort of emotionally smooth over or help yourself see that you're making progress even when these things can be kind of discouraging.
[1:24:12] Yeah, yeah, that makes a lot of sense. I suppose it's a bit of a cliche that people who are dating seriously because they're trying to find someone to build a life with, it can be very, very demoralizing because for that, the bar is naturally pretty high. It's like, if you're really going to commit to being with someone for decades and raising children with them, then you kind of want to do your research and actually check that this person really is a good fit to build a life with you. But then that means that maybe the best thing is just to go on a lot of dates, like even have, you know, the beginnings of many different relationships, which eats up a lot of time.
[1:24:40] Optimally, most of them are going to fail, but it's a little bit hard to keep in mind that things are actually going well, what it seems like in the narrow sense. You're making no progress. There's kind of this disconnect between the thing that you know, which is this is perfectly normal and expected and what this should look like, probably. And yeah, the day-to-day experience of often trying things that are difficult, breakups are really painful.
[1:25:09] Feeling like you haven't made any progress. And so, yeah, I guess that's a lot of the reason why I think this is kind of sometimes a useful way to think about things. Yeah. Yeah, just taking a step back and thinking about this kind of micro X concept in general, I suppose it seems like the case where it's really useful is with low probability. Well, I guess like events that are low probability, but like not so low probability that you should ignore them. I guess I'd specify, I think they're useful for low probability, high impact things. Right, yeah. And so there are cases where...
[1:25:39] You can sort of get yourself a little bit confused because the probabilities involved are so small that if you think of it in terms of probability, yeah, somehow that lens by itself can confuse you. But the impact is so large. And it's only by thinking about the multiple of those two is that you're actually sort of, you have a unit that's sort of coherent to think about. Yeah. Yeah. It helps us like one in a million chance. And I suppose for alternative things, you could do one in a thousand. Yeah. I sometimes use milli units for things as well.
[1:26:08] Yeah, so why is this helping me think about these things? I suppose...
[1:26:19] One thing is that when you're thinking about it in terms of probability or just intuitively, the difference between 100 micromorts and 1,000 micromorts doesn't feel like very much. Or, you know, 100 micromarriages or 1,000 micromanages, it all just kind of blurs into like, this is very unlikely. But actually specifying it out in like millionth chances makes it feel more measurable and real. Like you're clocking up these risks of death or these risks of something really good going on. And you can like see the incremental progress and also like weigh.
[1:26:44] Well, I think that's right. But I think that's another thing that happens, which is really interesting, which is as you spend more time thinking in these units, just like, you know, you've spent so much time thinking at distance, but like a foot and a meter and a kilometer, or I guess for a lot of people, maybe a mile, are sort of intuitive units. Similarly, I think if you think about things like this, you know,
[1:27:13] 100 micromerges or 1,000 micromerges start to feel like, in some sense, meaningful things. To me now, I'm like, gosh, 1,000 micromerges? That's a lot of micromerges. 10,000 micromerges, that's insane. And these start to feel like meaningful things. And if you think that you went to an event and you met a bunch of people and you, as you say, clocked up this many micromerges, there's a scale. I don't know, maybe it's just the scale of what you typically get or something where you can feel...
[1:27:43] Yeah, that went really well, even though there's in some sense no tangible result. Yeah, and in the case of the micromanage measure, I guess it also made me update how large the benefit might be at a specific social event. Because I guess most people end up getting married, usually to someone they like quite a bit, and they stay with them a while. So how many social events does it take?
[1:28:03] I mean, most people don't really have time over their youth to attend more than a thousand social events where plausibly they could meet someone. It might be a bit closer to a hundred for many people. So actually like the chance of meeting someone who you might like end up forming a serious relationship with at any particular social event could be, is maybe somewhere between one in a hundred and one in a thousand, which may be, so on a gut level, it doesn't feel super likely in any specific instance, but actually you go to a bunch of events that really clocks up pretty quickly, which is why lots of people end up paired up.
[1:28:31] Yeah, I mean, I think that assumes that going to social events is the primary mechanism by which people meet their partner, which it might not be. It could be that it's through introductions by friends or online dating or things like this. But yeah, I think these things can be really significant. And I think it's just such a life-changing thing and such a potentially dramatically positive thing. Yeah, I don't know.
[1:28:52] Maybe, not even maybe, I'm definitely a pretty weird person. So maybe this isn't useful to other people, but I'm often surprised how much time, I don't know, say that I've spent thinking about linear algebra and trying to really deeply think about linear algebra or other topics that are sort of these intellectual topics, compared to the amount of time that I've spent thinking really hard about these things that are going to shape my future in really dramatic ways. And yeah, I don't know, it seems intuitive to me to try to think.
[1:29:21] Think really carefully about those things. Yeah. Okay, so we want to apply this concept to low probability, high consequence things. We've already got the micro more. So death, that's a big deal. I guess we've got micro COVID. COVID's not quite as bad as death, but it's unpleasant. We've got, yeah, micro important relationship. What else could there be? What are the other big life events? I suppose like potentially finding a really fantastic job or career? Yeah, I think that would be totally a plausible thing. Yeah, I think that totally makes sense. I've sometimes found this useful.
[1:29:51] For thinking about pain. So I think about the most painful thing that I've experienced, and then thousands of that. And then I can use that to think about the amount of suffering that would be involved in something. So I sometimes find that useful. Yeah, that's a little bit dark. But what about micro friendships? I think one of the things that's delightful about the micromanaged marriage is that it's supplying this thing that...
[1:30:16] So far, I'd only seen applied to negative things to positive stuff as well. Yeah, no, I think, well, I think it's, I guess a related thing is thinking hard about extreme upsides rather than just extreme downsides. Like I think we often are sort of focused on avoiding extreme downsides, but there's these really extreme upsides that matter as much or more. And I think we often don't think about very much. And so, yeah, finding an outstanding job, finding a partner. I think one thing that's a little awkward about the...
[1:30:46] That or especially about the friend thing is I think we just have less of a vocabulary for like describing the like, like there's sort of a difference between somebody who's going to be your best friend for the rest of your life. And, you know, or like a very, very close friend. And it's going to be kind of life changing for you from somebody who's sort of a casual friend. And, you know, you enjoy meeting them, but it's not it's not really life changing in the same way. And I think marriage hopefully is sort of like pretty clearly putting you in the category of this person who's.
[1:31:13] Who's life-changing and hopefully really dramatically changes their life for the better. And yeah, I think the same thing with the job where you sort of like need to specify the like really outstanding job. I think you've sort of, you like, I don't know, like micro best friend or something is actually a really interesting unit. It's a sort of a cumbersome thing to say, but I think these really, yeah, these, you know, the closest friends that I have have really dramatically changed my life for the better and are extremely precious to me.
[1:31:42] I'd go through a lot for a one in a thousand chance to go and make another friend like that. Yeah, yeah. Yeah, I think as a society where maybe underrate the value of friendship, there's a lot of things about how people's lives are structured that caused a lot of their friendships to atrophia in their 30s and 40s and 50s and for them to end up, if they don't make an active effort, with a pretty small friendship group. But I guess that's potentially a topic for another day. My guest today has been Chris Ola. Thanks so much for coming back on the 80,000 Hours Podcast, Chris.
[1:32:12] Thank you so much, Rob. It was lovely being here. If you enjoyed this one and haven't listened to last week's episode 107 with Chris yet, I strongly encourage you to go back and give it a chance. It explores issues like what is interpretability research and what is it trying to solve, how neural networks work and how they think, multimodal neurons and what implications they might have for AI safety work, digital suffering, scaling laws, and how nice it would be if Chris's research agenda could succeed.
[1:32:39] If you're interested in using your career to work on safely guarding the development of AI like Chris is trying to do, or working to solve any of the problems that we frequently discuss on the show, then you can apply to speak with our team one-on-one for free. We've made some hires and removed our waitlist to apply for one-on-one advising, so our team is keen to speak with more of you regular 80,000 hours followers. They can discuss which problems to focus on, take a look over your plan, introduce you to mentors, and suggest roles that would particularly suit your skills, among other things.
[1:33:07] Just head to 80000hours.org to learn more about that service and apply if you feel like it. The 80,000 Hours podcast is produced by Kieran Harris. Audio mastering is by Ben Cordell. Full transcripts are available on our website and produced by Sophia Davis-Vogel. Thanks for joining. Talk to you again soon.