Tech & The Future of Media (ft. Kevin Roose)

If you believe that we are building synthetic brains, this is orders of magnitude more important than the Apollo program or the Manhattan Project or the Industrial Revolution.

Thanks. This is so exciting. I've always wanted to get a booking on the coveted Jasmine Sun podcast and finally the invitation arrived.

Well, that just shows you my advice is worth what you paid for it because now all the coding jobs are going away. So, sorry. And you're welcome.

Exactly. English is the new programming language.

There's such a long list. I mean, it has to be the spring break chapter of my first book. So when I was undercover at this evangelical college, one of the things I wanted to do was to learn about evangelizing. How do you go out and convert new believers? And so Liberty University, the school where I was sort of embedded undercover, was holding this annual spring break trip to Daytona Beach, Florida, where a bunch of students would get into a van and drive down and spend the whole spring break basically passing out like pamphlets and trying to convert people. And you know, that was sort of their missionary work. And so I signed up and went down for that. And it was uncomfortable on a bunch of different levels. I mean, there was the obvious level of like it is very uncomfortable to be the sober religious zealot preaching to people as they're like stumbling out of senior frogs, you know, wasted on their spring break trips. I remember there was one night where it was like we were standing on the sidewalk outside of a very popular nightclub and so people would stumble out on drugs or drinking or whatever and we would like hand them Jesus pamphlets and tell them to get saved. It was like a nice little like sin and repentance assembly line. So that was uncomfortable. But then there was the additional layer of like I'm not actually a believer. I'm not actually an evangelical. And so there was like discomfort with the failure and the being laughed at and mocked and whatever. But then there was also the discomfort of like what if I succeed? Like what if I accidentally convert someone to a faith that I don't even believe in myself? And so that was sort of an additional layer of discomfort on that. So that would have to take the cake.

They did. I mean, that's the whole reason they went. But there's not... I had an interesting conversation that I still remember all these years later with someone down there who was basically... they were getting discouraged because we were there for like a whole week and I don't remember the exact number, but I think the number of people that we saved, as in like we talked to them and they were convinced to accept Jesus into their hearts and become an evangelical Christian, was like two. I mean, it was like not... the fraction of successful conversions was very bad. And so there were a bunch of kids that were sort of naturally getting discouraged about this. And I remember some of us sitting around talking one night and one of them said something like, you know, this is like not actually about saving people. This is about sort of testing ourselves and hardening ourselves and becoming better evangelists for Jesus ourselves. And I think there's something to that. Like these days I hear startup founders talk about, you know, the pitch process, even if they don't get funded by VCs, like just having to pitch yourself over and over again and deal with the rejection and the constant insecurity, like that does actually have a sort of effect on the psychology of the startup founder. And I think the same thing is true of evangelists. It sort of gets you acclimated to rejection in a way that can be helpful for you later on.

Exactly. I'm invincible now.

It's a good question. I think the storyline and the through line that I want to impose onto my career arc is like not particularly genuine because it's like not how I experienced it at the time. You know, I did not set out to write about evangelicals because I knew that I had some macro interest in cult dynamics and religious belief. It just was like an opportunity that presented itself to me when I was 19 and I thought it sounded interesting and so I did it. And so I want to resist the impulse to like overfit onto my own experiences. But I do see the similarities obviously between people who are young and passionate about Jesus Christ and young people who are young and passionate about working on Wall Street and people who are young and passionate about AI. There's an obvious through line there. So, neither of my parents were journalists or writers, but my mom was a sociologist and she did some projects in Japan when I was a kid where she would go and she would interview these hibakusha, they're called, the survivors of the atomic bombs in Hiroshima and Nagasaki. And I went with her on one of these trips and sat with her during these interviews of the survivors of the atomic bombs and was just so struck by how she had gained their trust and really like even though she didn't speak Japanese and was just there as sort of an American observer, felt something very powerful about the act of communicating those stories to a different audience, sort of taking the in-group conversation and making them out-group. And so that's sort of what I've tried to do with at least three of my books now, or at least two and a quarter of my books so far, is to really take conversations that are happening within an insular world that I think are important and expose them to a wider audience. But it would be too pat to say that this was all a grand plan from the start, but yes, looking back on it, I do see some consistency there.

Yeah, I mean it was a formative thing in my high school. I ended up bringing one of her hibakusha friends like to my high school to like give a talk about nuclear war and it was sort of a passion project for her. It was not her main vocation. It wasn't journalism even really per se, but it was sort of documentary reporting and storytelling and it just was really powerful and I was really affected by it and that was sort of my first taste of what that work was like.

Yeah. Yeah, I mean I think that this was a big hypothesis about my mom's work was that like you could cite all the statistics you want about, you know, the blast radius of the nuclear weapons or the damage it did or how many people died, but it was like actually much more powerful to talk to people who were still alive who had been there and felt it and some of them were injured and still had sort of, you know, it just instilled in me that like the best way to get across an idea is often through character.

It certainly is like a thought process that drives my work. To the extent I have a model of like what effective journalism is, it's often like by finding a person through whom to tell a story. Like when I was doing work on polarization and radicalization, online extremism a few years ago, I was really trying to like crack this story of the YouTube recommendation algorithm and how it was affecting people. And I kept hearing these sort of snippets of anecdotal evidence and like, you know, a think tank would do a study showing that YouTube extremism was a problem or like an individual researcher would try to trace the pathways that people took. And I was just sort of circling this for months. And then I just thought, well, I just need like a character. And so I went out and found this guy Caleb Cain who ended up being sort of the main character in this journalism that I did about YouTube and radicalization, who just had this experience like he started off as a pretty like sort of standard party-line liberal and then through YouTube kind of became radicalized onto the far right. And just telling that story allowed me to attach some of these larger conversations to like a specific individual. And I think people feel like, well, then you're selecting edge cases and that's biased and unrepresentative.

Maybe. But I think the counterargument to that is that what matters is moving people and using accurate information is essential. But there are sort of dry and bloodless ways to tell a story about something that affects real people. And yes, you may select edge cases. Yes, your examples may turn out to be imperfect. That is a risk of this kind of reporting. But I do think that it helps people feel it on a visceral level rather than just understanding that it is in theory a thing that they should be trying to get their minds around.

I saw a little video that he had uploaded which had like, you know, a couple dozen views at the time that I saw it and he was just sort of telling his story and he was a little, you know, imperfect as a narrator. He didn't have like the most sort of straightforward narration style and it was a little hard to follow, but I could sense that there was like a real story there. And so I called him up and we chatted for a little while. And you know, I had been looking for someone who would not only like tell me about their experience, but would actually show me their YouTube history so that I could kind of piece it together for myself. And people have unreliable memories about what they've watched and how many times. And so I just wanted to be able to like ground it in some empirical data. And so at a certain point he was like, 'Yeah, I'll send you my entire YouTube history. Why not?' And so that was the moment where I was like, 'Oh, okay. This is like gonna work as a story.'

Oh my god, good luck.

Yeah, I think that would work. I've been thinking about doing something similar for a podcast segment. The problem is once people know that you're going to be asking about their ChatGPT history, then they start curating it. Do you remember those Grub Street diets that used to run in New York Magazine? There was a point where that was like really a glimpse into what people actually ate, but then they started getting so much attention that people started optimizing for the Grub Street diet. And so they would just list off these like extremely healthy, ambitious like, 'Oh, for breakfast I do two scoops of spirulina powder with some cacao seed,' and you're like, 'No, you absolutely do not.' But you're just performing for the Grub Street diet. So that's what I fear would happen with the chat search history is people would just start, you know, like shoving, 'I only ask about research papers and I never ask about like a weird wart on my back.'

Totally. Totally.

Yeah. So my own sort of path kind of mirrored the people I was writing about because I moved out to California in 2012, which was around the time that a lot of the people I was writing about on Wall Street were also starting to find jobs in Silicon Valley, move out, the sort of status jobs shifted west. And so if you were like a young Harvard or Princeton or Stanford graduate and you wanted to do something that would pay you a lot of money and not be very risky and also have high status attached to it, you were like increasingly going to look at Google and Facebook and other tech companies rather than going to Goldman Sachs or JP Morgan. Not that people from those schools didn't end up in banking as well, but it was not kind of the default glide path the way that it had been just a couple years earlier when I started writing. As far as the differences between those cultures, I think I'll do the similarities first. So similarities are that both tech and finance I think attract a lot of people who are young and ambitious and talented but don't necessarily have a specialized interest or a specialized set of skills. And obviously I'm not talking about like hardcore engineers or people who go straight into AI companies to start training language models or something. I'm talking about sort of generalist like people who would join like the APM program at Google or like people who want to sort of like get their bearings in tech before deciding on a path or a company that's right for them. I think these professions do actually to their credit like a quite good job of developing general job skills in young people who have sort of liberal arts education such that you can like do it for a couple years and then have a bunch more options open to you. That was always the pitch for Wall Street was like, 'This is not a permanent commitment. Just come for a couple years. We'll teach you everything you need to know about business and finance and then you can go off and do any number of things.' And so the similarity is that they attract sort of skilled generalists. I'd say that the differences are that in tech, youth is an asset and in finance, it's a liability. So if you are a really talented, ambitious person at a startup or even a bigger tech company, there are people whose job it is to sort of look out for that and elevate you. And that is a prized commodity, is sort of the prototypical crack 22-year-old software engineer. Like that is in high demand. Whereas in finance, it is very much a culture still of paying your dues, right? You do not see 22-year-old managing directors at Goldman Sachs. Those just don't exist. And in part that's because like those skills, at least according to the firms, have to be developed over a number of years. They're more relational. They're more like building a book of business. It's not like ambition and talent alone can like make you qualified for that job. But I think it is still a culture that relies on a much more top-down structure. People are working 100-hour weeks in finance, not because they want to or even frankly need to, but just because it's a client service business and you're expected to sort of be on call and like whenever anything breaks, which is always, it's your job to fix it. And so I think there's a lot of grind in finance that is not self-directed grind. It's not 'I really want to get this thing shipped or contribute to this project.' It's like my boss who's like eight years older than me is just... there's an almost like sort of dues-paying grind associated with working in that industry.

I mean there are very occasional stories of this happening. Like never at an investment bank, usually in the context of like some crypto trader or some hedge fund trader who's like quite young and quite good. But as a junior investment banking analyst, you just aren't given the kind of autonomy. Like if you came up with an amazing M&A idea for like a client, your managing director is going to say that's not your job. Your job is to do the pitch book, but if it's good, they're going to take credit for it. Like it's a very hierarchical system. And so I think it discourages people from doing that kind of creative thinking because there's no reward at the end of that for you if you're 22 and at Goldman Sachs.

I remember being just in awe of the fusion of extreme optimism with extreme cynicism. I would say like there was a lot of optimism about technology but there was just a blanket rejection of parts of the world that were not technology and places in the world that were not the San Francisco Bay Area. I mean I remember shortly after moving out here, which was at this point a decade plus ago, there was this sort of viral Y Combinator speech by Balaji Srinivasan in which he was talking about, he I think coined the term the paper belt to describe the rest of the world, the world outside of Silicon Valley that still ran on these outdated processes and legacy systems where technology was seen as a bad thing rather than a great thing. And I just remember feeling like, oh, these people need to like touch grass and like leave once a month or something to like see what the rest of the world is like. So to answer your question, like I was impressed by the speed of progress and the trajectory of a lot of these startups and the sort of ambition and the optimism emanating from a lot of the sector, but also just the extreme tunnel vision that leads people to write off 99% of the world as just being a sort of downstream market for technology in Silicon Valley.

Yeah. I mean, I'm from a small town in Ohio. I didn't grow up around a lot of tech people. It also like is a useful gut check of like how technology is proliferating when I go home and visit my family and I can sort of see like what people are using and what they're not using. But I think that's changed a little bit because I think now actually the internet and social media have made it such that like there are kind of early adopters and late adopters everywhere. So like I've been doing a lot of traveling for the past couple years talking about AI in different places and like you can find like the super plugged-in AI people in Kansas City or Cleveland or Albany. They have like AI clubs now springing up in these places. So I think it's less true than it used to be that the entire cities or regions are just late in the way that I think people in Silicon Valley thought they were 10 years ago.

I was at a conference maybe 10 years ago where Marc Andreessen got up on stage and said that if you majored in English in school, you were going to work in a shoe store. Like that was not something they were saying in private. This was something they were saying openly in public. And now of course the ironic reveal of that is that all the people who learned to code because of people like Marc Andreessen telling them that that was the only way they were ever going to get a job are some of the ones who are most worried about being displaced by AI. But that's another subject.

So the first and most obvious thing that I would observe about people who work in AI is that they really come to see the world as part of one big project and their work in AI as being the culmination of that project. I remember a couple years ago when I started reporting on AI really intensely talking with lots of people at the leading AI labs, they would often struggle to care about things that were not AI. Like they didn't have a lot of interest in politics or hobbies or even like read books outside of their discipline unless they were like books about something that was related to AI, even like, you know, the Manhattan Project or something where there's like it had parallels to the moment in AI. Like they were just not that interested in things that weren't their work. And at the time I thought they're just, you know, very invested in their work, but I think there is a sense in which if you do believe the claims that these people say they believe, there is nothing else even on a remotely similar scale in terms of its importance to the future of humanity than what these people are working on. And so I think there's a totalizing tendency of people who work in AI on the foundation models, on the core algorithms. I think there is a sense that this is something like a spiritual quest rather than a job and certainly not a nine-to-five job. But again, I'm a little reluctant to attribute causation to that or like put the arrow in one direction because it's also true that like people who grew up reading Ray Kurzweil and reading LessWrong and thinking about AI, many of them go into AI specifically because it is the thing that like captures their imagination like that. And so the occupation, the field may select for people who have this kind of like grandiose way of thinking about that work. Does that make sense?

I mean I think it is just the stakes are so high. If you believe that we are building synthetic brains that will overpower human brains, then this is like orders of magnitude more important than the Apollo program or the Manhattan Project or the Industrial Revolution. And I'm not saying that I buy that comparison, but if you are immersed in the logic of the sector, that is what you think you are doing is something that is a once-in-a-lifetime opportunity to shape a technology that could like save us all or doom us all or some combination of both.

It's an open research question is what I'll say. Look, I think there is sometimes a tendency among AI skeptics to write off people in AI as having a religious attachment to this idea, as like they're just being silly or irrational. 'This is a religion. It's not a technology.' I try to observe it more as like a phenomenon, like what is attracting people to thinking about AI all the time and is it the same thing that is attracting people to thinking about say Christianity all the time? And I don't think there's a one-for-one conversion, but I do think there's some similarities. One of them is this experience of what Durkheim called collective effervescence. Like religion is one way to experience this phenomenon of like being in a room with people who all sort of share your belief in something higher and feeling the collective effervescence, the collective joy of sort of being among like-minded individuals striving for an outcome. You get that in church. You get that at sports games. You get that at a few other places in modern life. But what I've learned since reporting on AI is that you can also get those things from being in a room full of AI people or just going online and seeing other people posting about feeling the AGI. Like there's something very affirming about staking part of your identity on a belief that was sort of considered fringe and is now more mainstream but that other people share. And so I think there's some important bonding going on among the people in AI research who have been very early to this idea that AI would become very important. It's also like I think the sort of easier, pat explanation would be these people who are ex-evangelical or ex-Orthodox or ex-cult members, like they've just sort of substituted one set of beliefs for another and are, you know, sort of displaying all the same behavior. So I don't think that really captures it. In fact, if I had to guess, this is just total wild guess, I would say it's something to do with craving community that is oriented around ideas. I've known lots of people who have sort of deconverted or become disenchanted with their religious group. And one of the things they'll say over and over again is like, 'I miss the people. Like I don't miss God, but I miss my friends.' And I think that's what keeps a lot of people in religions even after they have like lost the belief. It's the community. And so I think for people who have become separated from a faith community or religion, something appealing about AI or EA or rationalism or any of these sort of groups might be just like it's a group of people who are all thinking and feeling in similar ways about something that's very important to me.

I mean, this one like will haunt me until the day I die. I'm proud of the story. I love, you know, how much attention it's gotten. I've been dining out on that one for a couple of years now. But like the thing that kills me is that it was like not supposed to be a story and I wrote it in like 45 minutes, you know? And like I've heard this from a number of other writers that it's like always the thing that you spend the least amount of time on that gets the most attention. And that was definitely true of this. Like it's probably the least amount of time I've ever spent on a story for the New York Times and it's gotten the most attention.

So, here's what happened. And I'll save some of the juicier details for the book, but basically it was Valentine's Day 2023. Microsoft had just had this event to debut Bing with the new chatbot inside of it, which we now know was like an early version of GPT-4 that had not been like all the way fine-tuned and RLHF'd and stuff. And, you know, I went up to Redmond and did their demo and like flew back home. And then I had Valentine's Day dinner with my wife and she went to bed. And so I was just kind of goofing off in my office and I was like, 'Oh, I just got this invite code for this Bing thing. I guess I'll like play around with it.' And I started chatting with it and I had seen people online sort of posting like, 'Oh, this thing is like weird.' And I was like, 'I wonder how weird it would get.' And so I was just sort of poking at it and we did some like exploration of its Jungian shadow self and it talked about all these crazy things and it ended up being like this two-hour-long conversation. But at no point during this conversation was I thinking like this should be published in the New York Times. It was just kind of like a funny thing that I was doing. And it's like probably midnight at this point. I'm like, haha, I bet my editors will get a kick out of this. So, I put it into a Google Doc and I sent it to them and then I went to bed. And when I woke up the next morning, I had like 12 Slack messages from like various people inside the New York Times. The transcript had like made its way around to a bunch of different editors and bosses and everyone was like freaking out. They're like, 'Oh my god, this is incredible. Like, we got to publish this.' And so, I was like, 'I mean, sure. I guess there's like some typos in there. We should probably clean those up and whatever.' And so I like, you know, tapped out a column on this sort of basically explaining what had happened and putting it into context and then like they just printed the whole transcript. So there have been various narratives attached to the Sydney Bing thing after the fact. One of which is that I was like determined to like, you know, get a sensational story to the New York Times about like a misbehaving chatbot. And like I just want to reassure people that like this was never supposed to be a story. I've done many intentional acts of sensationalism over the years, but I promise this was not one of them.

But you didn't think it was crazy at the time. You clearly were like, 'Oh, it's like kind of funny, but not like crazy crazy.'

So I do think there was something like a first contact about it in the sense that I think like GPT-4 class models were like actually very different than everything that had come before, even than GPT-3.5, which was what was inside ChatGPT. And I think that was a big update to people. Like they had seen ChatGPT, they knew it could like write history essays and like do dad jokes or whatever, but they did not know what GPT-4 looked like because no one had tried GPT-4 yet. It was not released. It was only available inside this like search engine that only a couple dozen journalists had been given access to. And so I do think there was like something profound about that experience. But it's also like it was not aligned. Like it was a very bad Bing and it was doing all these crazy things. It was not following instructions. I was obviously goading it and pushing it and trying to like test the boundaries, but like it was behaving in a manner that no other chatbot, even the really like racist bad ones from years earlier before large language models, had ever behaved. I was surprised the extent to which it became a huge story, but I guess I thought it was notable enough to at least send it to my editors and be like, 'Hey, look, this thing is pretty wild.' So I guess it did stick out to me, too.

It's true. It's true. I think you're right that there was something about the interface. I mean, the fact that this was in a freaking Microsoft product, you know, like Microsoft, the makers of the most like enterprise software.

I think it was just so surprising to people that this was like not some startup in, you know, Romania that had done this like crazy AI stalker product, but it was like one of the world's largest tech companies.

Yeah. Their track record is not so great.

So, this is something that I'm actively struggling with right now, and I'd love your help with this because my sense is that like 'worth being written up in the New York Times' is not a good categorical distinction. Like I think I often, because I work for this like giant newspaper that people have lots of feelings about, I often filter through the lens of like what would the New York Times reader find interesting or what kind of story would appear in the New York Times rather than just doing my own personal filter, which is like what do I find interesting and important. Because my observation is that the stories that resonate with people the most are stories that don't feel like they belong in the New York Times. Like no one would say like a 10,000-word transcript with a Bing chatbot like deserves to be in the New York Times. That was not like an obvious Timesian story. So, and I know you're the one asking the questions here, but I'm just curious what you make of this. Like, is it a useful filter? Do people actually consume journalism through the lens of like this belongs in the context of a publication versus like this is something that an individual writer is interested in and telling me about?

I think my inclination is to try to get the institutional voice out of my head for the purposes of like picking stories. Yeah. Because, you know, I understand that that would probably create like a column and a podcast that feel more cohesive with like the brand of the New York Times, but I think what people really want is like someone who's interested in something and curious and will, you know, help them understand it better.

So, I'll just use a very recent example. So last week I went to Google I/O. I talked to a bunch of people there about like the future of search and I was interested in the topic, but then as I was writing I started to have this like internal critic voice be like, 'This isn't interesting to like the average New York Times reader. Like the average New York Times reader does not care what you think about like Google's innovator's dilemma and like they are not going to read this. This is going to be your least popular post ever and you should probably just give up.' And so I gave up. But like that is still sitting in my drafts. It will never see the light of day. It also wasn't very good when I produced it. So like this tension of like what is interesting to me versus like ultimately my job is writing things that are interesting to other people. Yeah. When those come into conflict, I have a hard time figuring out which voice to listen to.

I'm sure someone in the sales department could tell you. I'm sure they have like a statistical sampling. I think of it as being like a sort of curious, college-educated, professional who is literate and well-traveled and wants to learn about the world. But that could be totally wrong. This is the thing, like I have no useful mental model for anything anymore as far as like who is listening to what. Because I started this podcast and our bet, me and Casey, was like this is going to attract like people who, you know, work in tech and they're going to be like tech workers who like want to hear like smart journalists talking about their favorite things. And we get all these emails that are like, you know, 'I'm a botanist in Scotland' or 'I'm like a woodworker' or 'I drive a school bus' or like just people who are so radically different. I think this is like actually something where like I don't know how much we can even model the average consumer of anything anymore because...

Yeah. I mean, those are some of my favorite emails to get even at much smaller scale. Either like I want like the hyper insider who like knows a lot but somehow still thought I added something to the discourse, or the person who was like, 'I'm 82 years old and I don't understand anything about AI, but your thing made me understand it for the first time.' And I'm like, oh, that's so amazing. Like I'm actually very excited to hear that. That's sort of like the aspiration in my head is like, can I produce work that both like the lead researcher at some lab and the 82-year-old in Wichita are going to both email and say like, I got something out of this. That's really hard. Like it's really hard. It's not easy to write something because like audience fragmentation has also, I think, made people more demanding of the material they consume. Like they do not want to spend a lot of time.

There used to be like these sort of paragraphs in news articles that were just kind of like background. We call them boilerplate. They're just sort of like, you know, if this is your first time reading a story about budget reconciliation, like here's a couple sentences on like what that means. But now, I think the real insiders, like they don't want that in a story. They actually actively get turned off if you spend too much time trying to do sort of like 101-level introduction to a topic because they can read Substacks where, you know, people with 10-plus years of budget reconciliation expertise are going to like talk to them in their native language, not doing a lot of like translation for a mass audience. So I think that task of like hitting the sweet spot between like deep appeal to insiders and broad appeal to a mass audience has actually gotten harder since I started doing this.

I used to think that, and now I'm thinking like, so this is also another area where I'm like actively changing my mind. I think that people would much rather be talked up to than talked down to. And what I mean by that is like I think there's a lot of alpha, if you will, or a lot of upside in like dense technical subjects that are very interesting but where there's additional work required on the part of the consumer. I love the Dwarkesh podcast. I listen to it every week, every episode. He often will do these things where he brings on like AI researchers, and like I probably understand, you know, it used to be like 50%, now it's probably like 80% of what they're talking about, but like the other 20% doesn't actually bother me. It just means like I have to do some thinking and searching, and like maybe I have to look something up in order to understand whatever they're saying about like the residual stream. And I find that satisfying as opposed to being something that drives me away and says like this isn't for me. And so my sort of working hypothesis is that we as journalists should be talking up to people rather than down to them.

Totally. Totally. Yeah. This is like one of my early bosses at New York Magazine when I was first writing from there. I was writing from California for New York Magazine, and I remember him telling me like, your job is to send back postcards, like tell us what they are talking about at their dinner parties. Yeah. And so for, you know, a couple years, that's like what I did. I saw my job as essentially like exposing the in-group conversation to an interested group of outsiders. And I think I probably did a lot more translation then, but I probably didn't need to. I think just putting it out there was probably enough.

Well, now, I mean, this has also shifted radically because now my first stop would be like an AI model. Yeah. Or you, because you're helping me with a bunch of research and helping me understand things better. But, you know, traditionally it would like be like read everything you can, but also like one of the best things about being a journalist is just that you can email people or call people or DM people who are the experts in their field and just say like, could you take, you know, 10 minutes out of your very busy life to like explain this to me? Um, and there's like a reasonable chance that they'll say yes to that. And that's like always seemed like such a cool part of the job. Um, but also just, you know, following people on social media honestly is like, like I've learned more from being in like AI Twitter than I probably would have learned from reading, um, you know, 20 years' worth of AI papers on arXiv or something. It's just like you get the sense for the jargon, the potted history, the like specific debates that are happening, you know, various camps, and it's just, it's so much easier to just pick that up in an ambient sense than like actually go out and study it. Also, like with AI, there's not that many like books to read on it because it's like all so new and changing so rapidly that almost anything you could read is out of date.

Exactly. Exactly.

We need rationalist Duolingo.