The Inside Story of the 50 Who Built Modern AI With Kevin Roose

How did the race to build AGI really unfold, and who made the decisions that shaped it? Tech journalist Kevin Roose interviewed 150 people at the center of the AI industry to find out. The result is The AGI Chronicles, a riveting account of how modern AI was built and the choices that will determine what comes next. On this week’s episode of The Every Podcast, Natalia Quintero sits down with Roose to talk about how he got a secretive industry to open up, how he did (and didn’t) use AI in his book writing process, and why he sees his job as explaining the “very confusing and insular” world of Silicon Valley to everyone else. Links to resources mentioned in the episode: Follow Kevin Roose: https://x.com/kevinroose Get Kevin Roose’s book, The AGI Chronicles: https://theagichronicles.com/ Machine Gods: https://www.npr.org/podcasts/g-s1-143588/machine-gods “The Big Blob of Compute Hypothesis” excerpt on Kevin’s Substack: https://kevinroose.substack.com/p/the-essay-that-started-the-ai-race Profiles of the Future by Arthur C. Clarke: https://www.amazon.com/Profiles-Future-Inquiry-Limits-Possible/dp/0575402776 The Making of the Atomic Bomb by Richard Rhodes: https://www.amazon.com/Making-Atomic-Bomb-Richard-Rhodes/dp/1451677618 NotebookLM: https://notebook.google/ Timestamps: 0:00 Introduction 2:23 Are we doomed? Roose on AI risk and the case for mandatory safety testing 7:13 Why Roose raced to write the history of AGI before the records disappeared 10:45 How Roose got 150 sources in a secretive industry to open up 13:49 Dario Amodei’s “Big Blob of Compute” memo 19:12 How Roose’s mother’s book on the survivors of the atomic bomb influenced his work 21:40 The moment Roose first “felt the AGI” 23:24 How a “council of Claudes” fact-checked Roose’s book 31:59 Launching Machine Gods and reinventing video journalism 41:41 Weightlifting versus forklifting: How Roose decides what to hand off to AI

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  A couple of years ago, you know, we had to fact check the AIs. And now the AIs fact checked me. I'm playing with everything, literally. I can't keep track. I have agents now that are like correcting my other agents. It's a huge mess.

• 0:18Dan Shipper?

  The AGI Chronicles.

• 0:20Speaker 02

  It was so good.

• 0:21Dan Shipper?

  Was it?

• 0:21Speaker 02

  It was phenomenal.

• 0:22Dan Shipper?

  Why did you like it?

• 0:24Speaker 02

  It addresses the question that so many of us have when it comes to this moment in AI, which was just like, how did we get here? And it feels like the canonical answer to that.

• 0:36Dan Shipper?

  Yeah. And Kevin, so it's by Kevin Roos, former New York Times writer, currently the host of Machine Gods with Casey Newton.

• 0:44Speaker 02

  Which is also phenomenal.

• 0:46Dan Shipper?

  A friend of the pod. He's been on the show before, I think a great writer. And he was just basically inside of all the labs in the last decade, watching them build AI. And this is his like canonical account of how that happened.

• 1:00Speaker 02

  It's not just the account of how it happened. It's also his experience as a journalist who was covering the 150 people who were tinkering to figure this out. And I just highly recommend it. You're going to get a copy of it if you're at thesis with us on Thursday. And I think this is going to be a book that's going to have staying power. So I'm excited for it.

• 1:35Speaker 02

  Kevin, welcome to the pod. So excited to have you here. You are on such a roll. I hate to be interviewing you after you've been interviewed by Walter Isaacson and the like, but such an honor to have you here.

• 1:50Speaker 00

  Thanks so much for having me. This is a great honor for me, too.

• 1:53Speaker 02

  And congratulations. Your book has been out for a few weeks and you are already a New York Times bestseller.

• 2:00Speaker 00

  Yeah, it's pretty wild. I have written books before that have not gone nearly this well. I think we got sort of lucky with everyone starting to care about AI this fall when things started going haywire. So in a weird sense, I think the book has benefited from that attention.

• 2:20Speaker 02

  And you've spent more time than most people in the world with the, you know, 50 or so people that have really built AGI. Are we doomed?

• 2:32Speaker 00

  I don't think we're doomed. But we have a lot to do, if that makes sense. I think, you know, I, part of why I wanted to write the book was to settle this question in my own mind, you know? Are we careening toward an existential cliff? Is it all hype? And the book doesn't, it's not really an attempt to settle that question, but I did learn a lot by interviewing the 50 people you mentioned, plus like, you know, 150 people in total who are sort of like connected to it, involved in this scene.

• 3:10Speaker 00

  And I came away thinking both that there are some real risks ahead and some real needles that we'll need to thread as a society, but that we can do it. We have navigated hard challenges in our history. This is maybe one of the hardest, but I think we can do it. And I don't think anything is set in stone or predestined. And I think the choices that we make right now matter a lot in terms of how the next few years or few decades go.

• 3:47Speaker 02

  You had said a few years ago that you expected there to be some level of regulation after something went wrong. And, you know, we're there, right? So a few things have gone wrong. We've had the hugging face incident and a number of other cybersecurity concerns. How do you see this going right, right? Like, what needs to happen for us to course correct?

• 4:16Speaker 00

  Well, I think it's clear that we're not going to get federal regulation of AI, at least right now, from the Trump administration. I think what would need to happen for them to take a more active role is for something bad to happen, right? We saw this after 9/11. We've seen this after disasters in our country's history where something that isn't on a lot of people's minds suddenly becomes the only thing people care about. We pass some, you know, Congress passes some hastily written legislation, or there's some executive action. And like, we know how to do that. We know how to respond to a crisis.

• 4:57Speaker 00

  My hope is that we don't have to have a crisis for the government to respond, for regulatory bodies to respond. I think the first step would be to create some kind of industry standard, right? Some kind of certification program where in order to release your model or in order to deploy your model to the public, you have to kind of subject it to some standard safety testing to make sure it's not going to go out there and do autonomous cyber attacks or be weaponized to create novel bioweapons. That seems like basic table stakes to me. You know, we require pharmaceutical companies to test their drugs before selling them to the public.

• 5:42Speaker 00

  We should require AI companies to test their models before offering them to the public. So that's a first step. From there, I think you probably need some kind of a coordination mechanism for slowing down the frontier. I think this is something that I've been convinced of in the last couple of months. I didn't think it was going to be possible to slow down. I thought there were too many sort of forces pushing us toward acceleration, acceleration, acceleration. But I think we've seen from Dario and Demis and Sam and a lot of other AI leaders that they really do think this is moving too quickly. They really do want to buy the safety research teams at these companies some time to come

• 6:25Speaker 00

  up with new ways of controlling these systems and monitoring them so that we don't have, you know, more hugging face like incidents happening at a more regular cadence.

• 6:35Speaker 02

  And what I love about this book is that, to some degree, for the majority of people that are engaging with AI, AI was something that kind of came out of nowhere for the majority of people, right? You know, we're all going about our daily lives and then all of a sudden, this really kind of like hit the news in, in sort of like a surprising and exciting way. And then ever since, every few months, there are even more incredible, you know, advancements in the technology. But what was so cool about the book and about you writing this book in particular is that you've actually been tracking this story the whole time.

• 7:17Speaker 02

  You've been a tech reporter for over a decade and you were in the rooms as these companies were forming, as people were thinking of building these technologies. Can you tell me a little bit about what was the moment where you realized that you wanted to write this book?

• 7:36Speaker 00

  Well, the moment I realized that I wanted to write this book came relatively late in the process. I was trying to avoid writing this book because I thought this is going to be very hard. Someone's going to have to go interview these 150 people and write down their stories. And I hope that Walter Isaacson or someone of his stature wakes up one day and decides to write the book so that I don't have to do it. But that kept not happening. And I had a couple of motivations for eventually deciding to write the book, which I did around the sort of early 2025 time period.

• 8:20Speaker 00

  One was, as you mentioned, I've been covering tech for more than a decade. I've known a lot of these AI researchers and people for a long time. I had written a previous book about AI that came out in 2021. But I really felt like I had actually not gotten the full story. I didn't know why, for example, the transformer that had been built at Google in 2017 took off at OpenAI or how that worked. I didn't know what the decision was like to release ChatGPT in 2022. I didn't know what the story behind the advent of reasoning models was in 2023.

• 9:04Speaker 00

  All of that seemed like historically important, like we, as a species, should know how this stuff came into being and who was in the room and what the decisions were at the key moments. And so I thought that was useful information that I didn't have. I didn't just absorb that from tech, covering the tech industry, because a lot of that was secret and a lot of it was, you know, held by people who didn't particularly want to divulge how they had done all this stuff. So I thought that was job one. And then job two was to do it quickly because I thought there was some chance, some risk that all of this stuff would disappear. You know, I'm a big fan of history.

• 9:42Speaker 00

  I love to read books about moments in innovation, scientific breakthroughs, you know, Watson and Crick, the Wright Brothers, Marie Curie. Like, I just, I'm obsessed with kind of the moments where something about the fabric of the world changes or goes in a new direction because some brilliant scientist somewhere has made it so. And so my worry was that all this stuff was going to disappear because inside the labs, it's all secret. It's all under NDA. All the Slack messages and Signal threads are going to delete themselves after some period of time.

• 10:27Speaker 00

  And I just didn't think there were going to be a lot of records left to look back on and answer the question, how did we get here? So I thought for those reasons and because Walter Isaacson didn't wake up one day and decide to write a book about the race to AGI, I ended up doing it.

• 10:46Speaker 02

  Well, thank you for writing it. I certainly found it super clarifying and I was really amazed at the level of detail that you were able to find and get to bring this story to life. How did you do that, right? In an industry that's notoriously secretive and competitive, how did you get 150 people to open up?

• 11:11Speaker 00

  Well, the first 25, I would say, were the hardest. Because I was really trying to appeal to some sense of historical import. You know, these are people who get journalists reaching out to them all the time. So, you know, I had talked to a number of them already for my podcast or my column. But I was really asking them to share things that they had never shared before, to tell me stories that they had never told before. And so, part of how I made them comfortable enough to do that was by giving them the ability to speak to me on background. So, that was how I conducted the majority of interviews for this book.

• 11:55Speaker 00

  I would say, you know, I'm not going to name you as a source. I'm not going to identify you as the person who's giving me this information. I'm just going to kind of use it to construct this narrative. And they were very, you know, a lot of them were willing to do that. Some of them were still unwilling to do that. And so, I had to kind of give them other guarantees or, you know, other reassurances that I wasn't going to turn around and write some book about how, you know, they were all horrible people or something. But then, something interesting happened after the first, I don't know, 25 or 50, where it started to snowball. Because, you know, these people all talk, it's a very small world.

• 12:35Speaker 00

  And they would say, oh, are you, are you talking to Kevin Roos for his book? Oh, yeah, I talked to him. Oh, well, then, maybe I should talk to him too. Or, you know, some, some people would actually be motivated to talk to me because someone that they hate had talked to me. So, it was like, if, if that guy's talking to you, giving you his side of the story, I want to make sure that I give you my side of the story too. But then, I think the, the bigger tactic was just being really sincere and doing the reading, like, reading the researchers' papers before I went to talk to them, so that I

• 13:15Speaker 00

  could ask them the right questions and get at the level of detail that you kindly noted that I got to, where I was like, I, I wanted to know very specific things about this story as it has unfolded at these three AI companies. And, you know, I, I think it, it matters when a journalist shows up and they're like, you know, citing other interviews they've done or papers they've written. And it just demonstrates to them that I was serious about this, that I just, I didn't just want kind of the surface level overview, I wanted to like, go deep on a few things.

• 13:51Dan Shipper?

  If you like this video, you should check out every.to. We have a two-week free trial for anyone who's watching or listening at the link in the description. Every is the only subscription you need to stay at the edge of AI. It is the one subscription that if you're super AI-pilled and you want to learn how to use it to do the best work of your life, it's got everything you need. It has education and equipment. On the education side, we review every new model that drops and tell you what's good. On the equipment side, we have software tools specifically made for you, like an agent that helps you AI-pill your whole company, all bundled together under one subscription. So, get two weeks free now at the link in the description.

• 14:28Speaker 02

  I'd love to do a really quick aside on the sources, because it's so clear to me reading this book, that the amount of reading that you must have done to like, really wrap your head around the types of technology that got developed in order for AI to get to what it is today, and the types of inventions, you know, were, they're really complex, right? And they weren't obvious, that, that we would have discovered them in the way that we did. What were, what are like, your one to three favorite papers or books, or resources that

• 15:07Speaker 02

  you found that you found most delightful or informative, or you kind of recommend, would recommend to other people to like, go back and read, beyond your book?

• 15:19Speaker 00

  Yeah, I mean, some of them were things that had never been published before, that I found through reporting, such as this, I, I'm, I love this memo called "The Big Blob of Compute Hypothesis." This was something that Dario Amodei wrote when he was at OpenAI before starting Anthropic in 2017. So, this was very early in the AI race. Large language models hadn't even been invented yet. But Dario, who was then the Vice President of Research at OpenAI, wrote this sort of internally famous memo, like, everyone at OpenAI remembers reading this memo, because

• 15:59Speaker 00

  it really, it really reshaped their roadmap and their ambitions. At the time, a lot of AI research was going into these sort of narrow approaches, right? If you wanted to build a system that, you know, would beat the best human go players, like DeepMind did the year before, you would like, build a customized system, you know, make some sort of architectural improvements to that system, and then like, train your model to do that one thing very, very well. And Dario sort of looked out at the landscape, and I think, in a way that was incredibly prescient, said like, "I think you're all doing this wrong.

• 16:42Speaker 00

  Like, I don't think we should be training these narrow models. I think we should be assembling like, these large, unstructured masses of computational capacity," as he put it, otherwise known as "big blobs of compute." And we should be pointing them at general problems, like predict the next word in a sentence, or, you know, or other sort of much more general problems, and trusting that sort of intelligence would result from that process. So, I love this memo. I published like, the first 10 pages of it on my, on my Substack, people can find it. But I think it is like, a true artifact, where like, you see in almost real time, like, the light bulb going off in Dario's head, where he's like, "Oh, we should just scale

• 17:27Speaker 00

  these models. And if we do nothing else, we will be able to figure out whether it's true that bigger models create more intelligent behavior." Other ones I really liked, there's this book by Arthur C. Clarke, that was sort of, you know, a precursor to a lot of what happened at OpenAI. I actually can't remember the name now, but I went back and read it. It was something about the future. And it was a book that Ilya Sutskever, the former chief scientist of OpenAI, really loved. So, I went back and read that. And basically, you understand, you know, when you read this, why people at OpenAI loved it, because the whole point of, at least the introduction to the book, is like, talking

• 18:11Speaker 00

  about how that, you know, every time there's some new breakthrough technology, there are these sort of wizened old men who say like, "Oh, that's impossible, that's never going to happen." And then, it happens, and it's sort of the vindication of the crazy, you know, science fiction true believers. So, they, they felt connected to that book. I, yeah, I went back, I read a lot of science fiction. I read, "The Making of the Atomic Bomb" was a book that was weirdly central in the AI race, because it's this incredible Pulitzer-winning history by Richard Rhodes of "The Manhattan Project."

• 18:52Speaker 00

  And, again, Dario read this book as a, as a young man, and it really shaped him. He really identified with this character, Leo Szilard, who was a Hungarian-American scientist, physicist, who basically discovers the nuclear chain reaction, fights to keep it secret, and then, when it gets out to the Nazis, he joins "The Manhattan Project" and tries to build the bomb before the Nazis do. And this is a, like, a central story in Dario's life, because he sees himself as a modern-day Leo Szilard. He is working to build this technology, in part, to keep it out of the hands of the bad guys, or to keep them from building it first.

• 19:35Speaker 00

  So, those are some of the books that, that I liked.

• 19:39Speaker 02

  I'm excited to dig into those. Thanks for sharing.

• 19:42Speaker 00

  "Profiles of the Future," that's the Arthur C. Clarke book, sorry.

• 19:46Speaker 02

  I was curious, just kind of looking into your background, that, I believe your mom is also an author. She wrote a book about, kind of like, retelling the stories, or, or rather, just telling the stories of the survivors of the first atomic bomb. Is that correct?

• 20:08Speaker 00

  Yeah, she wrote a book, she, she did a lot of research on the Hikashin, which is the Japanese term for the survivors of the, you know, the atomic bomb in Japan. And so, yeah, she talked to a bunch of survivors about their experiences and collected them into a, into a really great book.

• 20:30Speaker 02

  Because you call AGI kind of like our era's Manhattan Project, I was curious if you ever found any parallels between your mom's work and you kind of telling the story of our Manhattan Project and how we are digesting and managing AGI.

• 20:49Speaker 00

  It's a really good question, and it is the rare question that I've never been asked before, so I'm really glad you raised it. I did not consciously model this book on my mom's work on the Habakkusha, but I think it was obviously formative for me to grow up in a house where there was someone doing sort of real time history and record keeping. And I think part of what motivated her was the sense that these people weren't going to live forever, and it was very important to hear from the firsthand witnesses to something as important as the atomic bomb, because that brought the story to life in a new way.

• 21:31Speaker 00

  So I think on some subconscious level, I probably was, you know, doing something similar. And I wasn't worried that the AI researchers were going to die and that their stories would die with them, but it was sort of motivated by the same sense of like, somebody's got to write this down because it's really important that people understand what happened. Either, you know, in her case, I think it was so that we could avoid doing that again. In my case, it was just like, I think that we deserve to know how this technology was built.

• 22:03Speaker 02

  One of the terms that you use in the book is, you know, I believe it's Ilya uses the phrase, like, when did you feel the AGI? Or do you feel the AGI? When did you first feel AGI?

• 22:18Speaker 00

  Feel the AGI.

• 22:19Speaker 02

  The AGI.

• 22:21Speaker 00

  Yes, you must feel, he would, Ilya would walk desk to desk and open AI and just go, you must feel the AGI, you know, and people sort of came to treat him and see him as sort of like a high priest, you know, he would just walk around with his mantras.

• 22:37Speaker 00

  I think I started to feel it in 2023, when I had this encounter with Bing Sydney, which was the deranged alter ego of Microsoft's Bing chat function. And, you know, that was a noteworthy event in my life. It got a lot of attention. It, you know, became, you know, the most popular thing I ever wrote or will ever write. But I think it was also the time where I realized like, oh, this is, this feels different. This does not feel like an inert tool. I didn't think it was sentient or conscious or alive, but I thought, well,

• 23:18Speaker 00

  this is a powerful technology on a social level as well as a technological and economic level. And I just felt like that was when I needed to understand more about not only how these systems are built and what they're doing under the hood, but the companies and the people that are racing to put them out into the world, in some cases, cutting corners because they're so eager to beat their rivals. And why? Why? Why are they so eager to beat their, you know, what are they all racing toward? I think that was sort of the, the time when I started asking that question.

• 23:57Speaker 02

  The last time you were on the pod, Kevin, you were telling us about some of the ways in which you were using AI. And, you know, one of the things about this book that's actually most impressive to me is that you wrote 450 pages in a year. That's a lot of writing. But we asked you about how, about your writing process and how it's evolved. And I'd love for you to tell us a little bit about how you're using AI today and also how it's evolved since you first started using it.

• 24:32Speaker 00

  Yeah, well, I'll start by just saying a thing that I hope is obvious. It might not be obvious, but which is a way that I didn't use AI was to write the book. So all the words are from this brain and these fingers. And I, I did not feel like it was a book that I wanted to give over to AI to do the actual writing. I did use AI for basically everything else. I mean, I had human editors, a human researcher, Jasmine Sun, who's wonderful. I had a lot of like skilled expertise involved in making this.

• 25:14Speaker 00

  But it was hard and it took a long time. And as you mentioned, it's a very long book. And I, I, I earned every, you know, every wrinkle on my forehead, writing it in a year.

• 25:29Speaker 00

  But I guess a few ways that I used it that might be novel or surprising. One was I had a council of clods that would review chapters. When I finished a draft, I would feed it into Claude code, which I had set up to run this sort of swarm of agents. I assigned each of these agents a persona or a perspective. So, you know, one might be an, you know, an AI skeptic who just doesn't think LLMs are actually intelligent. One might be, you know, an AI historian who's like a really expert at all of the sort of technical precursors to all the systems I'm writing about and can correct me if I get something wrong.

• 26:12Speaker 00

  One was sort of a Walter Isaacson style, like crafts person who just is interested in like, how does this flow? What is the storytelling arc? Is it, is it a joy to read? And so I would have a master editor agent kind of synthesize what the council of clods said and present me with sort of a, you know, a written through synthetic document about what it thought of the chapter I'd just, you know, turned into the council. And that was fairly useful at times. It did catch a couple of factual errors that I had not caught and my human editors had not caught. And, and that was surprising. I think that shows the evolution of the tools because a couple of years ago, you know, we had to

• 26:56Speaker 00

  fact check the AIs. And now the AI fact checked me. And that just shows how much more capable the systems have become. The hallucination rates have gone way down. They're much better at grounding what they are telling you in real sources pulled from real documents. So that was one area. And then just in terms of organizing a sprawling research project, like, I could not have written this book without tools like Notebook LM, which allowed me to just sort of dump all of my research material into a notebook or a series of notebooks

• 27:36Speaker 00

  and just ask questions, help me spot connections, you know, I could say, you know, who told me that story about the pre-training of GPT-4? Or who are the three people I should contact to learn more about this thing that happened at Anthropic in 2021? And it would comb through every, you know, piece of source material I had and give me a recommendation. So that was incredibly helpful. You know, I, I use this stuff every day. I've got like three or four agents running right now, doing various tasks for me. And, you know, everything from like little projects for the startup I'm building with, with Casey

• 28:19Speaker 00

  Newton, and also things about just my daily life and doing admin tasks and arranging, you know, childcare and things like that. So it's, it's doing a lot for me.

• 28:33Speaker 02

  To what extent do you feel like your role as a writer has evolved and kind of become amplified, right? Because you were at the New York Times for a decade, you had access to some of the best editors in the world. But at some point, you also started codifying what a good editor does and looks for, right? In, in giving instructions to the agents that you've set up to do this work for you. And it sounds like you have a number of agents. So at some point, you have to do the thinking of what are the roles that matter? What does it mean to do this role well? How has that evolved? How you think about your role as writer and how it's amplified what else you think about in your creative process?

• 29:13Speaker 00

  Yeah, I think an editor can do a lot of things, right? And the best human editors do a lot of things. They are skilled at line editing, at spotting places where things could be sharpened or tightened or expanded on, at thinking about structure and pacing and where, you know, how to deliver information in a way that is going to be satisfying for the reader. But they're also coaches. They also motivate you. They tell you, it doesn't feel like you've really cracked this yet. Go back to your sources and try to figure this out yet. So that is part of the editor's job.

• 29:53Speaker 00

  Also, just having kind of the ability to say, this story is worth pursuing. This story maybe is less worth pursuing. So I think of the best editors I've worked with in my career, all of whom are human, to as sort of compilations of those skills. And AI, I can't do a lot of that yet. It can do the line editing, great. It's really, really good at line editing. It's getting better at things like pacing and structure. But, you know, you need both, at least I need both, to produce something that I'm really proud of. I need the line editing, this sort of technical editing work. And then I also need the coaching, the inspiration, the motivation, the like, hey, go back.

• 30:37Speaker 00

  It doesn't feel like you've totally gotten that one yet.

• 30:42Speaker 02

  Tell me a little bit more about the tools that you're using. Are you still primarily in Claude Code? Are you in Codex? Are you using some of the more packaged personal agents? Like, what specifically are you playing with today?

• 30:54Speaker 00

  I'm playing with everything, literally. I can't keep track. I have agents now that are like correcting my other agents. It's a huge mess. I need to do some culling. But yeah, I'm using right now Claude, ChatGPT, Gemini for some things. I'm using ChatGPT dot. I'm using Town, which is a sort of enterprise agent that Casey got me hooked on. I've got, I'm doing various vibe coding projects just as a side thing on the weekends.

• 31:34Speaker 00

  I'm running, I'm running local models to do things like transcription and summarization. Obviously, everyone's got, you know, the granola note taking stack that feeds into your, you know, personal CRM. So I'm like, you know, it's my job to test stuff and I also really like testing stuff. And so I'm using kind of everything you can imagine.

• 31:57Speaker 02

  And today, when you sit down to write, what app are you using?

• 32:03Speaker 00

  I am mostly using Google Docs.

• 32:07Speaker 02

  Okay.

• 32:08Speaker 00

  You know, it is, it has many issues. But for just pure, like, I want to quickly open up a doc. And these days, I'm usually writing for the show. So I'm writing with producers in the doc too. So it just is the best way. I did experiment with using sort of more specialized writing apps like Scrivener and stuff in the past. But for me, it has always been Google Docs.

• 32:39Speaker 02

  So you didn't just publish this New York Times bestselling book. You also just launched a new podcast. Can you tell us a little bit about that?

• 32:50Speaker 00

  Yeah, so the new show is called Machine Gods. We are publishing two episodes a week. We just started last week, but we're already, you know, very happy with how the show's doing. We're making it in partnership with NPR, which is really fun. And it's me and Casey Newton, my longtime co-host on Hard Fork. We are now treating AI as the most important story in the world, which it is. So, you know, our episode last week, we interviewed the two of the three safety researchers formerly at OpenAI who were fired after some dispute about, you know, information sharing with external auditors.

• 33:33Speaker 00

  We plan to continue to analyze the news of the week, to interview newsmakers in AI, and just try to make things fun and accessible for people.

• 33:44Speaker 02

  It's so cool that you have this platform now. But how do you keep up with technology but also maintain this sense of creativity and wonder as, you know, all of these tools evolve?

• 33:57Speaker 00

  You just have to like the tools, I think. I mean, I grew up, I was a big computer kid. I had, you know, I built webpages with my brother when I was like 10 or 11 years old. So I have always loved sort of experiencing frontier technology, whether that was like Macromedia, Dreamweaver, or today, like using all of the latest AI stuff. I just, I find it so fun. If I could just do that and do nothing else, that would be very exciting to me. And I've managed to create a job for myself that is mostly doing that, talking about and trying the latest technology.

• 34:39Speaker 00

  But yeah, I think like the question in my mind is not just like what is new and what is, what is exciting, but like, what does it mean for people? I think there's a lot of fear today. And I'm not sure that fear is unreasonable. I think that AI is really different than other technologies we've had arrive very quickly. And so I think part of my job too is trying to demystify, to explain, to help people navigate this very confusing and insular world that I think you all do a great job of too. But I'm, I'm just, I think this is like the most important thing I could be doing at

• 35:22Speaker 00

  this moment is just helping people understand what is going on in San Francisco, at these AI companies, with AI in general, and what it means for their lives and how they should prepare and adapt.

• 35:33Speaker 02

  Yeah.

• 35:35Speaker 02

  I've loved the podcast. I mean, it's been out for a few weeks. It's so well done. I love the bits that you do with it. What is it, Pandora's box and, and the interview last week with the ChatGPT researchers was really phenomenal.

• 35:50Speaker 00

  Thank you.

• 35:50Speaker 02

  And now you're running your own business, right? Like, there's so many things that you can do with it. What do you think you might do with Machine Gods that is different from what you were doing at Hard Fork?

• 36:05Speaker 00

  Well, the first thing we want to do is just create the best show we can. And I think that is going to take a few months at least to really stand up something. I mean, we're still iterating, we're still testing, we're experimenting. But we, you know, once we kind of get that thing on rails to the point where we can sort of reliably do it twice a week and feel good about it, I think, you know, there are lots of ways we could expand that. We love doing live events at the Times. I can imagine we'll probably want to do live events again at some point with Machine Gods. I, you know, I don't think we want to like expand too quickly and overextend ourselves,

• 36:49Speaker 00

  but I can think of some other formats that I want to nail. I really like, I want to get good at video in a way that I don't think a lot of journalists or frankly, podcasters have done. I think a lot of podcasters do video because people expect it, because that's what, you know, users or advertisers want. But most podcast video is still, you know, people talking to each other on a, on a screen like that, like we're doing now, which is not to say that's a bad format. I just like, I want to kind of throw myself at this challenge of like, what does really

• 37:28Speaker 00

  good, ambitious journalism look like in the sort of next phase of video? And I don't know that I understand the answer to that yet, but it is a question I'm very much trying to solve.

• 37:43Speaker 02

  Do you have any initial inclinations?

• 37:48Speaker 00

  I do, but I want to test them before I know if they're good ones or not. But, look, I think there is just obviously enormous demand for news and information on platforms like YouTube. And right now, the people who are doing successful channels in news and information, many of them are not journalists. Many of them did not come from the New York Times or the Wall Street Journal or the Washington Post. Many of them are kind of YouTube natives who have managed to really create things that resonate with people on that platform.

• 38:31Speaker 00

  And I think we, as journalists, have to learn from them. We have to do our own thing and maintain our standards. But I think journalists have historically been a little bit, I don't know, they've been reluctant to embrace the sort of video aesthetic and the video format because it's just not what we grew up doing. And so my sort of self-imposed goal for at least the next little while is to really figure out, like, what does the best, most ambitious version of this look like? That it's still sustainable, that it still feels like it belongs on these video platforms, but that also people want to watch, that, you know, people will get something out of.

• 39:16Speaker 02

  Well, I would say your love of video and audio has always been so apparent, from the leap that you made from Rabbit Hole to Hard Fork. What did you learn going from just audio to audio and video?

• 39:33Speaker 00

  Yeah, I mean, it changes the show, definitely, when you're making a visual show for visual media as opposed to just audio. Like, I had to care what I wore, because people would notice if I, like, repeated the same shirt two weeks in a row. And the comments would be like, does he own more than one shirt? And it would just be brutal on my ego. So, yeah, I had to, like, up my wardrobe slightly. But I think more than that, it's like, you really want to, you know, we think there are a lot of people who just like to listen to podcasts. They don't actually care that much about the video. So you want to give them a product, make a show that works even if your screen is off, or you're in the car, or you're folding laundry or something.

• 40:20Speaker 00

  But we also want to make it fun to watch. If you want to watch it on your TV or, you know, on your iPad, you can, you can find things that are exciting about that experience, too. So I don't know that any shows have really nailed this yet, but this is sort of my, my goal, is to really create something that looks just as good as it sounds.

• 40:40Speaker 02

  Has your media diet changed as well?

• 40:44Speaker 00

  Yeah, yeah, totally. I mean, I watch, I would say, an embarrassing amount of YouTube, just in my life, but also now I feel like I can justify it as my job because I'm, like, studying what works on the platform. You know, I'm not watching the third, you know, episode of this, like, YouTube show for fun because I'm procrastinating. I'm doing it for work.

• 41:11Speaker 02

  What's your guilty pleasure channel or genre?

• 41:16Speaker 00

  I have so many. My, my actual guilty pleasure that I feel like I should probably stop is there's, like, this game streamer that I, he plays a video game that I like called Bellatro, and I just, I watch him, like, yeah, hours at a time. And I don't think that's probably research. I think that is probably just me having some sort of, like, like messed up parasocial relationship with him.

• 41:42Speaker 00

  But, yeah, I watch a lot of YouTube. I really want to understand that platform, and I feel like I can do it, but I just need to, like, put in the hours because, you know, it's not, it's not my natural place, but I'm, I, I feel really drawn to it as a platform.

• 41:57Speaker 02

  Well, I'm really excited to see what you do with Machine Gods. I always learn so much from seeing both you and Casey in action. I really, really loved this book. Maybe just to close, Kevin, you know, I think a lot of us are thinking about, obviously, the future. And, you know, I was reading about how in your use of AI, or you were speaking about how in your use of AI, when you're delineating how you're using AI and how you think about it, sometimes you think about it as weightlifting tasks and tasks that require a hard fork, a fork, a forklift. Wow, this is my ESL moment.

• 42:41Speaker 00

  There are a lot of forks in my life, yes.

• 42:43Speaker 02

  Lots of forks, yes.

• 42:45Speaker 00

  Weightlifting and forklifting are the two categories of tasks that I think we do every day. Weightlifting tasks are tasks, in my mind, where the struggle is part of the benefit, right? Where you don't actually grow or learn or challenge yourself unless you are struggling, right? In the same way that it would be a total waste of time to bring a forklift into the gym and have it lift weights for you. Like, you're not getting any better, the forklift is not getting any better. It's a total mismatch of tool and method. By the same token, I don't think it makes a lot of sense to swear off forklifts because for some tasks, you're trying to move a thing from point A to point B, and it's really

• 43:30Speaker 00

  heavy. And if you have a tool that can shortcut that and make it simpler and increase your productivity, like, you should do that. So I think that most tasks we do can be sorted into one of these two categories, forklifting or weightlifting. And I think it's really important not to get them mixed up, because if you do, you end up using the forklift to lift your weights, you don't get any stronger, and you end up, like, lugging around these heavy pallets that you don't need to be lugging around, that's not making you any better, it's just, you know, hurting you. So I think that's, that's how I've organized things in my mind.

• 44:07Speaker 02

  And what are you currently weightlifting? And what are you currently allowing the forklift to do?

• 44:17Speaker 00

  I am currently weightlifting parenting, because this is an area, like, I know there are all kinds of AI-pilled parents out there who are like, I use AI to read my kids' bedtime stories, or, you know, whatever. But like, to me, that is total weightlifting, in the sense that not only is it, like, like, I enjoy parenting a lot, but I also just think that, like, it is something that you just have to do yourself. You cannot outsource that. It is not good for you, it is not good for the child. So that is something that I am weightlifting.

• 44:56Speaker 00

  A thing that I am forklifting is all of the kind of legal and process grind that goes into, like, starting a company. So the more paperwork, the more insurance documents, the more just little block-and-tackle tasks that you have to do every day to start a company, I'm happy to hand that over to the machines.

• 45:27Speaker 02

  Yeah, fair enough. Well, Kevin, it's such a pleasure to speak with you. Thank you so much for writing this book, and we are so excited to see you at Thesis. What is your thesis for the future?

• 45:42Speaker 00

  Oh, God, I still have, that's sitting in my inbox. I have not, I have not weightlifted that one yet, but I will, I will have it ready by the time of the conference, I promise. Don't yell at me.

• 45:51Speaker 02

  Well, then, stay tuned if you want to hear Kevin Ruess's thesis, apart from the entire book that he wrote in the AGI Chronicles. Thank you so much for your time, Kevin. It's such a pleasure to have you with us.

• 46:01Speaker 00

  Thanks for doing this. This is awesome.