Dean W. Ball on America's AI Action Plan & 4 Months at the White House

Hello. And welcome back to The Cognitive Revolution. Today, I am thrilled to have Dean W. Ball back for his 5th appearance on the podcast, fresh off a brief but historic tenure as Senior Policy Advisor for Artificial Intelligence and Emerging Technology at the White House Office for Science and Technology Policy. Dean is living proof that a talented person with a passion for understanding AI can go from a newcomer to the field to the highest level of influence in as little as a year. After spending the first 10 years of his career at policy think tanks, Dean took a leap of faith and quit his job to start thinking and writing about AI full-time, launching his Substack Hyperdimensional in January 2024. Around that time, he sent me a Twitter DM about apparent contradictions between OpenAI's preparedness framework and its superalignment plan. How, he wondered, could OpenAI plan to rely on automated alignment researchers for safety while at the same time flagging model autonomy as a leading source of catastrophic risk? It was a good question, and I became an early subscriber. The first of Dean's posts that really stuck with me was one he called Software's Romantic Era, about his first interactions with Claude 3 Opus, in which Claude expressed the desire to understand what embodiment feels like And to hear music. Claude's first request would be Beethoven's 9th Symphony, a fascinating choice since Beethoven was deaf when he wrote it and never actually heard it himself. I still remember one of the concluding lines, quote, here finally is an AI system whose thoughts I want to hear. What a brilliant way to capture the spine-tingling difference between Claude 3 and earlier models, and more broadly, between AI and all other technology. Dean's profile began to rise in a serious way thanks to his early criticism of California's SB 1047, which inspired open source and progress advocates to take an interest and ultimately led to multiple rounds of revisions. While we never ended up fully agreeing on the bill, I was impressed that unlike so many of the shrill voices in that debate, Dean moderated his criticism as the bill itself evolved. From there, he started showing up everywhere, at The Curve, as an FAI fellow, as an affiliate of Fathom, and with a podcast of his own. And then he got the call to join the White House. Now, regular listeners will know that I did not vote for President Trump, but I was nevertheless very excited for him and more importantly, for the country. There are not many people on the political right, or frankly anywhere, who combine the impulse to get hands-on and use the technology to the fullest with a healthy respect for how powerful it could soon become, plus a deep appreciation for the nuances and challenges of integrating it into society and government effectively. I thought then, and I continue to believe now, that Dean was perhaps the single best person the Trump administration could plausibly have picked for the role. At the White House, Dean led the effort to develop America's AI Action Plan, which very well might be the most well-received policy the Trump administration has put out to date on any topic. With positive reviews from many people who expected to hate it, including an endorsement from President Biden's former National Security Advisor Jake Sullivan on the most recent episode of this very podcast. Yes, it is a broadly accelerationist document framed in terms of beating China, but it's also extremely sophisticated, calling for what I believe will prove to be very prudent investments in evaluations and risk assessment, cybersecurity, AI-enabled science, Mechanistic interpretability, compute governance, and much more. In this conversation, we discuss why Dean is now leaving the White House, including his candid self-assessment that while he's great at conceptual work, implementation can probably be better done by somebody else. We explore how the American political right is grappling with AI, from deregulatory impulses to culture war concerns and worries about AI's impact on children. We also dig into the mechanics of unlocking spare gigawatts for data centers, Hear Dean's perspective on which frontier developers should be considered live players and how they're perceived in Washington, and get his perspective on what the AI safety and governance communities can best do to complement the government's work going forward. Along the way, we also hear how Dean used LLMs to simulate interagency feedback meetings, how the 4-year presidential term, which ends in January 2029, just happens to put the administration in sync with many AGI forecasts, And why Dean believes that the Republican Party might become the better home of the AI safety world in the long run. Between his rapid rise and his brief White House tenure, you might say that Dean embodies short timelines. And while I am sad to see him leave such a high-impact role, I hope that his story inspires others who may have similar potential to take their own leap of faith and get into the AI game. We need all the Deans we can get. And we need them as soon as possible. Finally, I want to say congratulations to Dean and his wife on their upcoming first child, another reason he's leaving the White House, and to thank Dean for doing this. He could easily have booked a much bigger platform for his first post-White House interview, and I'm sure those will still come. But I'm grateful he chose to have this conversation here. It was super fun for me, and I always appreciate the chance to dig into so many important details. With that, I hope you enjoy this super deep dive into 4 months in the Trump White House and the crafting of America's AI Action Plan with Dean W. Ball.

Dean W. Ball, fresh off a short but historic tenure as Senior Policy Advisor for Artificial Intelligence and Emerging Technology at the White House Office for Science and Technology Policy. Welcome back to The Cognitive Revolution.

Thank you for having me, Nathan. It's great to be here.

Yeah, I'm excited to catch up on your recent tour of duty. It's been an influential one, to say the least. We've come a long way. This is your 5th time here on the podcast. The first time, just a little over a year ago, I think maybe a year and a half, we were talking about frontiers in neurotechnology, a little side hobby of yours that has probably withered a little bit in the meantime. But if ever there was a cognitive revolution bump, and I say this in jest, to go from just starting to write about AI all the way through to this White House role and playing a key role in the writing and the release of the AI Action Plan. It has been quite a run for you. So I think there's a ton, obviously, to get into. Maybe for starters, want to tell us just about your experience? First of all, how did you even get this job at the White House? Is it something that Did somebody tap you on the shoulder and say, hey, we want you for this? Did you have to interview? Did you have to put your hat in the ring? Um, tell us the whole story.

Uh, yeah, it's a great question, uh, the way these things work. Um, so I never, like, actively solicited for a role in, in the administration. What I would say I did is, um, after, after the president won in November, I published a piece like maybe the week or 2 weeks after that was called Here's What I Think We Should Do. And really the goal there was very much like, well, let's— we have a new administration coming in. I think I would've done that if Kamala had won. You know, it was like, obviously I would've felt less, you know, excited about the prospect of those policies being realized, but But that's basically what I would've, you know, put forward either way. And, um, so I wrote that and that kind of did reasonably well as a piece. And some people, uh, who I knew were going into the administration reached out to me. Uh, and my view was always like, I would happily do it. It's not something like, I'm not desperate to go into government. Just through a lot of different social interactions, I ended up coming, uh, into conversation with Michael Kratsios, the director of, of, uh, OSTP. This was before the administration, so he was appointed by the president and not yet confirmed by the Senate and all that. Um, so I, uh, we talked a little bit and he was like, do you think you'd want to come into government? And I said sure. And then, uh, it was several months of figuring out, uh, you know, the hard part was not actually getting like the initial offer from Michael. The hard part was like figuring out administratively how to achieve the job. Because this is like a little bit of inside baseball, but OSTP, the Office of Science and Technology Policy, is unlike a lot of components of the White House. OSTP was created by Congress in statute, and so it's technically an agency. And what that means is that it has a budget set by Congress that doesn't go up Very much. Uh, it has a pretty tiny budget, OSTP. Uh, the whole White House actually generally has a pretty small budget, um, and, uh, a relatively inflexible one. And so most— many if not most OSTP staffers are, uh, either sort of affiliated with other parts of the government or affiliated with nonprofits or universities. And in— through various sort of bureaucratic means get transitioned into the OSTP role. That took a while to figure out, and I was uncertain it actually would be worked out. I knew it was sort of being worked on, and I knew it was a possibility, but I didn't bank on it. And then it all came together really quickly, like in the last week or two, it all happened really quick. And so I remember finding out that my last day at Mercatus would be Uh, you know, it was basically the same business day, uh, that, that I figured out, okay, it was like today's my last day, and I had to go tell everyone and do a quick, you know, rush and say goodbye to everybody, get all my things in a box, go, and then, um, uh, start, uh, you know, in government on the following Monday.

Sounds like a little of that red tape that might need to be, uh, cut to get, uh, the right people in the right seats faster in the future.

It's hard. Hiring for government is hard. It's really true.

So you've been writing, people have been exposed to your ideas, you've got a lot of good feedback on your ideas. But I have to imagine it's a pretty big difference to wake up one day and be like, all right, today I'm going to the White House. And now I'm not just writing to kind of socialize ideas and hopefully kind of gently steer the public discourse and thinking, but I'm actually going to be you know, in the seat, you know, playing a significant role in making the rules, making the policy, deciding what we should do, you know, with real consequences.

Yeah.

Did you get training when you show up?

That you get training on some stuff. They make sure you know that people can't pay for your bar tab or your lunch, you know, things like that. There's like various procedural ethics rules that you have to comply with. They're very serious about that. There's like an IT training. And then beyond that, no, it's— the thing that's interesting about joining government in general is that I have truly worked at universities and in think tanks that have more internal bureaucracy than the White House does. There are certain things in the White House that like You run into and it's like, wow, that is a hard bureaucratic wall and there is just nothing you can do about that. That is just there. But then like a surprising amount of stuff is way more flexible than you would think and you can actually move like quite quickly, which was one of the things that I really liked about being there. So like that, I would say no one really prepares you for it and nobody can prepare you for it. It doesn't ever stop being— well, I was only there for 4 months, but I have a feeling that if I had been there for 8 or 12 or 16, uh, that it, it would have, um, you know, never stopped being surreal. Uh, and I think there's like a few aspects to it. Like, one thing I would say is that, um, I think it's very hard to know if you're the kind of person for whom power is like an addictive drug. that you just want more and more and more of, like a drug addict wants more, you know, of whatever their drug of choice is all the time, and you kind of like need it to sustain yourself versus are you the sort of person that like doesn't? You know, I think it's like, it's just really hard to know. And I think a lot of people that, you know, without disparaging anyone, but a lot of people that are like in Washington, D.C. are the kind of person, you know, they're the former kind of person. What I think I learned about myself is like, I'm probably more the latter than, than anything else. I'm probably like, I found myself relatively unmoved by it. And if anything, it's in some ways burdened by it because, you know, I mean, the amount of incoming you're going to get from other people is insane, right? Like, I At any given point, I probably had between 30 and 40, uh, you know, unread communications of various kinds. People reach out to you on Signal and WhatsApp and, and email and all sorts of places, just like asking me to meet or do various things or, or read this or, or, you know, think about that or whatever else. Um, there's just a ton of requests for your time. And, uh, you know, even people that are your friends or colleagues, the relationship with them changes in some really important ways. And so it's like very hard to like anticipate how to deal with it. It's a really weird situation. I don't know if it ever like stops being weird, you know, to be totally honest, because it's just like, you know, these are pretty extraordinary jobs. It's a tremendous honor though. You know, you just walk around and, When you have a good day, you really feel like, wow, like I actually really drove this thing forward and, and, you know, did whatever. Um, and like, it's real. Like, it's actually a real thing that we're gonna do, you know? It will be public policy of the United States of America. Um, and, uh, yeah, it's crazy. You have to take it very seriously and it's a huge responsibility.

You told me you wrote a letter to yourself before going in to just try to ground and calibrate yourself to some of those possible issues. How would you advise people who might be thinking about taking on such a role to write that letter? What else can they— since there's no training, what can you offer in terms of wisdom to people that might step into this in the future?

It's a really good question. I would say a couple of things. I would say, first of all, if you're going into a policy role, not every role, there's operations roles, all kinds of different roles in government. But if you're going into a policy role, You should have substantively all of your policy ideas pretty well baked by the time you go in, because you will not have any time to develop policy. You know what I mean? You're not going to be sitting around thinking that much. You don't have time to think, right? I mean, the volume of things that happen to you when you're in government is just wild. It'll— just as an example, you know, uh, XYZ major developed country is coming through the door. They'll be at the White House in 3 days. And so when you're, when you're dealing with things, you know, at that velocity, um, you just, you have to have like policies like very well developed, um, already in your mind, or at least like, you know Reasonably far along, I would say. Um, but also, uh, I think you need, um, you need to have a, a, a very important set of sort of like what your principles are. Ultimately, your job as a staffer in government, uh, whether it's the White House or elsewhere, um, you know, your job is to execute the vision of the president. Your job is to achieve the president's objectives. Um, and so there's a lot of staffers who I think, um, convince themselves that, oh well, president is, you know, he's operating at such a high level, he doesn't know exactly, you know, I know the right thing to do here, um, even if it somewhat disagrees with something that he said recently. Um, you know, that's, uh, um, that will not get you far. Uh, uh, I mean, yeah, people do it all the time, you know, but I would say like particularly in this administration, that will not get you far. But you do need to have a sense of what your own— of what's important to you personally. It might not be that, like, the president's objectives conflict with your principles. I mean, that's certainly possible, but it's not so much that. It's more about drift. It's about drift. Being inside of government is kind of like— it's very weird. It's like you're inside of a pretty self-contained cube. with like glass walls and like everyone in the world is like shouting into the cube all at the same time. So you can hear what people are saying, but you like can't exactly interact with people, um, in quite the same way, you know. Um, just as one practical example, let's say you have a question for, for companies in an industry, right? Um, about like some aspect of their business that, that like might be non-public and you sort of feel like you need to know it to make some policy determination.

Mm-hmm.

There are a lot of rules that govern how that interaction can go, you know, uh, that, that structure that interaction. A lot of your interactions will kind of be governed in ways that will not be intuitive to you if you don't have prior experience in government. I think one of the things in particular, one of the things that, that, that I felt quite acutely is like, I felt myself drifting a little bit on like public scrutiny of my ideas, um, which is weird because like in a certain sense the action plan is the most publicly scrutinized document by many orders of magnitude that, you know, I have ever contributed to. Um, but nonetheless, that was like a hard thing for me, and it turns out that that's really important to me. It's really important to me to get public scrutiny of my ideas. Um, that's one of the things that I, I wrote to myself is like Do you feel like you're losing— like, in a letter to myself, I said, like, you know, there's, there's a risk of drifting from ground truth, right? Because you're just kind of in this bubble. And so you might drift away, uh, and sort of like become more attuned rather than to the ground truth of the world, to the logic of the system in which you operate, which has its own very distinct logic and does not necessarily operate according to anything intuitive. The second you start to see that happening, you have to be careful because that can really change who you are in the long term. Um, so that was something that was really important to me, and I would say that's like a general flavor of thing that I think it's very common, uh, to encounter in government.

Hey, we'll continue our interview in a moment after a word from our sponsors. Yeah, that's interesting. The time piece, um, calls to mind an obvious question. which is how much leverage are you able to get from AI in your work at the White House these days?

So it's interesting. I always had to be really careful how I answered this question inside the building, but I would say this. So LLMs are actually not permitted on White House computers. That's not true of all of government. That's specific to the White House. Um, and it's because I never ever got super into the weeds on what the problem exactly is, but it relates to compliance with a law passed after, I believe after Watergate, called the Presidential Records Act, um, which relates to, you know, as it sounds like, the specific documents produced within the, basically within, uh, the executive office of President is the sort of name of the bureaucracy that, that is the— also called the White House. Um, and, um, there's something about compliance with that law that affects the ability of the White House to use, as I understand it, lots of modern technology. So like, we don't use— we can't use like Slack or Microsoft Teams. You can't use like Zoom or Google Meet or Microsoft Teams. You have to use like Webex. Um, Google Docs is another good example, and also LLMs. That being said, and the main aspect of that is, it has to do with sharing pre-decisional documents, draft documents, things that are literally policy being developed live, which obviously there's a lot of. Pretty much everything I worked on all day long was deliberative pre-decisional drafts. Those things very, very seriously cannot touch LLMs. They can't leave government computers. Obviously, that's a very important thing and not something that I would ever mess around with. That being said, I used AI as a chief of staff for me and as kind of a research assistant. So not a single word of the action plan was written by, edited by, or seen by AI prior to the release, at least not by me. But there were so many times when it's like, okay, well, I have a question I want to ask. I want to try to develop some policies relating to XYZ. And so sometimes it might be like, well, let's brainstorm 1,000 different things. Sometimes I used it for that. Most of the time, I actually had a pretty good sense of, directionally what I wanted to do, and it was about scoping it correctly. It was about really like, okay, like, let me understand the statutes here really well. Like, let me get like all the— what all the statutory authorities, all the laws that enable us to do this thing that's a recommended action in the action plan. Um, you know, you have to, you have to understand all of the like regulatory history, legal history, interpretations, etc., of like All the relevant statutes. You have to really understand, like, how are— how exactly are you constrained? What exactly can you do if you're trying to do everything you can? The way it felt to me was like I was thrown into the cockpit of a plane. Uh, I've never operated a plane before. Uh, in fact, my— I don't know that much about the federal government in the grand scheme of things. Uh, I know a lot more now, but, um, you know, and, and you're like kind of like, okay, well, what do all these switches do? What do all these buttons do? And you're kind of, you know, like the LLM is a really good guide for that. Um, the other thing I did that I think was really valuable is, um, you know, the, the action plan was, I was actually looking earlier today, um, uh, just at, at some, uh, initial notes that I did for, um, actually like back in February when I was drafting, uh, you know, for, for the public RFI? Pretty close. And like, by late April, by 2 or 3 weeks into the job, the action plan was like, I would say, 2/3 to 75% of where it ended up publicly, basically. But that 2/3 to 75% is like a blurry image that was sharpened by lots and lots of interagency feedback. So it's the difference between a useful image and a not useful image, right? Uh, 'cause one thing is just a bunch of blobs and the other thing is actually an in-focus image. Um, but, uh, one thing that I found it very useful to do was for, for all the different items in the plan to kind of like, without directly taking the text of the plan, just saying like, okay, this general idea, this general concept of doing like something, uh, let's simulate interagency feedback, uh, on that, right? Let's, let's actually simulate like—

Mm-hmm.

What would an interagency meeting be like here? Like, okay, like, what is the, the grisly career, you know, uh, uh, in-house general counsel at the Federal Communications Commission gonna say about this, right? Like, to this young whippersnapper who works on AI, you know, what's he gonna say to me, right? Like, about my, about my bright idea. Um, you kind of simulate that feedback. Um, and over time, uh, you, you can basically actually like do the first, like maybe not the entirety of it, you can do like the first couple meetings kind of that you might've had to otherwise might've taken a lot of time. You could actually just kind of simulate that. Uh, and so, uh, that ended up being, um, really useful. And I think what, what I always say to people is like the action plan was not at all written by AI. The ideas in it did not come from AI. Um, it was all written by humans, and not just me, many humans, uh, contributed a lot to it. But it is probably an example of an AI-enabled productivity boost, uh, because it got done in a pretty short amount of time. It was about 3-ish months. It is true that like a very small number of people like really drove the text. And so I think it's probably evidence of some kind of a productivity boost from AI.

The idea of simulating interagency meetings is definitely a fascinating one. Do you want to give any more detail on— you mentioned there's some surprises when it comes to the bureaucracy. Some things are just hard barriers, other things are more flexible than you'd expect. I draw a blank on what that would actually look like in practice. Are there examples that would be informative?

There are definitely some things. I mean, one that's a classic one is, and this is something that's actually kind of a meme in policy circles, but it is true, is something called the Paperwork Reduction Act, which was passed to make government more efficient and reduce paperwork in government. And what it means is that anytime You as a government employee want to reach out to more than 9 members of the public, whether they be individuals, companies, et cetera, nonprofits, et cetera, et cetera. If you wanna ask the same thing, ask for the same information from more than 9 members of the public, you have to go through a whole, like, there's like an entire bureaucratic process you have to go through for that. And there's like all sorts of rules that govern it. And you just like wouldn't guess that 'cause, you know, you, you might be like, Again, I was using the example of like reaching out to, to, to industry. Like, I think like one thing that like a lot of people, you know, taking my job, like one of the first things they might want to do is go talk to the frontier AI labs and say like, okay, like where's the secret briefing? You know, where's the, where's the, where's the briefing for, for high-level government employees that, you know, you don't share stuff with the public, right? Um, and, uh, you know, you gotta be careful about that. because if you reach out to more than 9 companies, then you have brought yourself into Paperwork Reduction Act territory. You wouldn't guess that.

Okay, well, then I have to ask, uh, what's in those secret briefings for the government? Are there actually such briefings? And, uh, what can you tell us about them?

Uh, there are— I mean, there are meetings where, where, you know, we learn, um, uh, things that are, are, of course, you know, and not just with the AI labs, with all kinds of companies, you know. You come in and we learn all sorts of things that aren't public. Um, I would say like there is not the, um, the sort of like the meeting where they come in and say like, here's our AGI roadmap. Like that doesn't happen. That's, you know, the companies are, are pretty, uh, it's not like we keep, it's not like government employees are like known for keeping secrets super well, right? Uh, so it's not like the company is like, they're, they're pretty cautious about what they choose to share and, and not share. Um, you know, in meetings, I think different people push them to differing degrees. Um, I find, uh, I would say like, there was a lot of times when, when talking to, to companies about like non-public information was, was a very important part of my decision-making. But I also think it's probably the case that, like, like, there are certain kinds of questions that many different people in government have about, you know, some aspect of AI, right? Like, how much power do you think— that's a classic one. How much power do you think you guys are going to need? Right? And I feel so bad for the people on the OpenAI and Anthropic and Google and Meta policy staffs. Who get this question from like 20 different government employees a week in like different, different parts of the White House. It's like in every single one of them, lots of briefing, right? Um, but you know, there's not like coordination. You know, we don't— we're often not aware of what the others are doing. Um, and it's hard to be, right? Because it's like, well, you know, how often do I see the— every, you know, every single staffer with a plausible interest in that question at every single agency that relates to it, you know, which are many. And one of the other things that would be, I think, unintuitive to a lot of people is that, like, it's not unintuitive. It's just like you might know it intellectually, but it's like the government just does a lot. Like, there's just a lot of stuff inside of agencies. There's like a lot of people doing all kinds of different things that, like, it would never really occur to you. So you would never be like, yeah, like, why is the State Department here? And it's like, oh, because, oh, that's actually extremely important. Like, they were doing this thing, you know? Um, so, um, yeah, I would say like, uh, those meetings do happen. Um, they're probably not as structured and organized as they could be. There is, you know, uh, I— to be honest, in some ways I felt like I knew more about what was going on inside the labs outside of government, just to be totally honest with you. Um, because one of the things they do once you get the .gov email address is, um, You know, the labs will, like, put you in touch with, um, researchers and stuff. I had a lot of friends at the researcher level.

Mm-hmm.

Um, but that all of a sudden becomes, like, a lot more complicated once you're in government, you know? Like, talking to, like, other researchers at labs, like, it's, it's definitely possible to do, but, you know, they're gonna be more nervous about it. Um, they're gonna wanna bring in their policy, and the second you bring in the policy people, it's like, okay, well, now we're gonna have, like, a script. Uh, and we're gonna say exactly the right stuff. And so it was always very funny. Um, sometimes the, you know, the labs would like organize these briefings, and there's also like different degrees of knowledge about this stuff, right? So like, I'm someone who like, I want to have conversations at like the relevant margin where knowledge is accruing on a topic, right? I want to be like, you know, what, how's this, you know, what do you, what do you, what are your thoughts on the sample efficiency of, you know, reinforcement or something like that, right? I want to ask about like some, architectural question. Um, but, you know, a lot of other government employees with, with plausible nexuses to, to AI, like, they need, like, a much more basic kind of a briefing. And so that's typically what the labs prepare. So, like, there's a, there's a particular briefing, uh, I won't say, I won't say which company and I won't say who the researcher was, but it was a researcher who's very well known on the internet, um, with the policy people where they were, they were briefing us about some, some of the latest developments they were rolling out in their models. And they so desperately wanted it to be a certain kind of script that was designed to satisfy the, you know, like, what would a staffer with little context want? And I kept asking all these highly technical and in-the-weeds questions. And the policy staff, they don't stop you from asking those questions, but there are definitely moments when the policy staff, like the researchers will going on into something and the policy staff will be like, whoa, stop. So yeah, it's actually interesting. I don't know if it's definitely true that I knew more about what was going on inside the labs. I felt like I had a better grasp on it as a non-government employee, but I think it's possible that that's true. I haven't thought enough about it, but I think it's possible that that's true.

Hey, we'll continue our interview in a moment after a word from our sponsors. Yeah, interesting. Um, why are you leaving now?

So yeah, it's, uh, probably a surprise to a lot of people. Um, you know, I'm leaving for a few central reasons, uh, and I'll kind of go in, in, in order of, um, maybe like some are like highly personal and others are, are a little bit bigger. So, uh, the The most personal one is that, um, my wife and I are having our first, uh, child, um, in a few months. And so, um—

Congratulations.

Thank you very much. Uh, in fact, I, I found out, uh, about, um, uh, her pregnancy, um, the day before I started at the White House. The night before. It was like 9 PM that evening. And it was like, the next day you're going to the White House. Uh, so it was a lot to process, uh, all at once. Um, the other, uh, and, and so I, I want to be able to spend time with, with, with them, uh, and like devote, frankly enjoy the last couple months of pre-parenthood and then, you know, be able to really be a very attentive and loving father in, in the, the initial months of parenthood. Um, so that's part of it. And I think it's not impossible to do that as a government employee, but it's hard. And I worry that I would either do one or both of the jobs badly if I had to balance them both. The other reason really is about what I'm good at and what I'm not good at, right? So I spent a big chunk of my career as a manager.

Mm-hmm.

And I was decent at it. I wasn't bad at it. I did, you know, reasonably well and got promotions and stuff like that. But the truth is, is I didn't really enjoy it. Like, I didn't love it. And the fundamental reason for that, I later discovered, is that my success depended quite often a lot on the whims and attitudes and what side of the bed other people in the world woke up on. And so, like, I just wasn't independently responsible for my own success. And I found that when I did projects where I was in control of whether or not the output was high quality, I found that to be much more satisfying. And so, you know, with the action plan, there was this, like, rare opportunity. I obviously, I was not the single author of it. It wasn't like a true, you know, it wasn't like a Substack post where I just hit publish. But like, I exercise like a lot of creative direction over the action plan and, you know, stylistic direction and all this, these sorts of things. And I just like, like, that's the kind of thing I'm, I'm really good at. And I'm just not that good at like the process of, of managing the implementation of the plan. Like, you know, Uh, uh, sort of making sure the agencies are doing it, thinking like all the procedural levers you need to be pulling, all that stuff. Like, it's not— um, I don't have confidence that I'm great at it. I think I could do an okay job, but I think if you only think you can do an okay job, you shouldn't be working for the president. Uh, just like if you only think you can do an okay job at you know, catching a football, you shouldn't, you know, play for the Miami Dolphins, uh, right? Like, you're supposed to be the best in the world. Um, you're supposed to be— it's like, this is the AI policy of the United States of America. Like, this has to be really, really effing good. Uh, and I had a high conviction that I, along with the rest of the team, could deliver that In terms of the text of the action plan. Um, I had high conviction about that. I think that the administration is able— can implement it. I think there's a lot of very, very savvy and driven, uh, people who can also implement the plan. Um, but I just don't think that, like, I am necessarily one of them. And I actually think there are probably ways in which, like, just reflecting on my own writing trajectory over the last 18 months, There's probably ways in which I would be, like, actively unhelpful, uh, because, like, you know, um, I mean, if we went back to, uh, one of those earlier episodes of Cognitive Revolution, the first SB 1047 debate we did, um, the neurotechnology, uh, stuff, like, I'm sure there's a ton of stuff I said in those episodes that, like, nowadays I would look at and, like, very happily discard. Very happily. Oh yeah, I totally don't believe that anymore. That's fine. I've evolved past that. You can't really do that with, like, you know, you're, like, setting the federal government's policy. It's like a big ship. You can't really, like, be like, yeah, well, you know, I changed my mind about that. Like, I decided that was dumb. Certainly there's some, you know, there will be room for flexibility in terms of exactly how it's implemented, but you can't just, like, throw things out at will. And you also can't— add new stuff, you know, all that easily, right? Um, and so, like, I think there's just, like, my nature is to be very much like everything I say is provisional, and I'm always, like, uh, you know, evolving. And so just like me personally, I think I wouldn't be good at it. Uh, that's basically what it came down to. Uh, and then finally, some of the stuff we were talking about earlier, um, where, um, You know, I, uh, I really missed— found myself missing, uh, public feedback, regular working in public with regular scrutiny and feedback from a community of my peers, you know. Um, the, the experience of doing that, it just was so useful to me epistemically. It was so useful to me. Um, and I just like, it's much harder to do when you don't get that real-world feedback. That was a concern I had, and I kind of want to get back to that because I think that like developing, sort of getting my head around really thorny ideas and developing answers to, to, to the hard questions and then like trying to communicate those answers compellingly, that's what I do. Um, and I think I do an okay job at that. Um, I don't know if I do like I don't know if I'm like the master of the bureaucracy. Uh, I just really don't. And I think you have to be honest with yourself about things like that, um, and not try to do, you know, more than you can. I think, um, you know, the way I've been describing it to people is sort of know thyself. Um, so all that gets put together. Uh, and, you know, not any one of those things might not have been enough. I, I might have You know, maintained it all. But the combination of all those things made it feel to me like now is probably an appropriate time to take a step back and, um, you know, contribute to the, both the action plan and AI policy more broadly, um, from the outside where frankly, I think, you know, we're still so early in terms of societal conversations on AI. Um, that there is so much intellectual work to be done and like civilizational scaffolding to be built around this technology. And I think that stuff inherent— I mean, one of the key ideas of the action plan is that that kind of stuff, in addition to the technology itself, is fundamentally going to be built by the private sector. Um, and so, uh, um, you know, I, I kind of want to return to To doing that. Um, so yeah, that's like the— it's a long-winded answer, I realize, but that's kind of, uh, that's kind of where I came down. It was, it was a tough decision. Um, it really was. You know, it's not like this was easy for me to do. Uh, but one thing I think is important to note is that, um, I leave with, uh, you know, I was not pushed out because of like political infighting. Uh, this is not as— there are like— To the extent that there are, like, people on different sides of debates that might externally be referred to by press as factions, uh, this is not an indication of one faction rising over another. Um, you know, uh, this is a decision that I made personally myself, and, um, is not because of— it's also not because of any policy disagreements with the administration. Um, that's like— it's not like I was like, oh, I'm mad about this, I'm leaving. Nothing. Not at all. And I leave with absolutely no bad blood and nothing but warm feelings for all of my colleagues in the federal government.

That's great.

And notable contrast to many departures that we've heard tell all about.

Yeah.

Well, congratulations again on starting a family with your wife. And obviously, that's a once in a lifetime transition that you'll be able to give full energy to that, I think it's safe to say, would have to make some compromises on if you were staying in the job. That makes a tonne of sense. I want to get into a little bit of the context in which this AI action plan was developed, and then we can get into some of the details of it as well. One thing, though, that you had said toward the end there that I found a little puzzling, and maybe this is just my never worked in government naivete, but what is it that stands in the way of a more open feedback welcoming process? I don't know how many people saw drafts of the AI action plan or contributed to it, but presumably it was like reasonably small number compared to how many people would have happily read and commented on a draft if you had put one out there and asked for feedback. Why can't that happen? Or could it happen and it's just sort of an administrative decision? I'm confused, though, especially because it does seem like AI in general— it's not AI in general, AI specifically. It's such a dynamic situation that the faster feedback cycles we could get, presumably the better. So, yeah, what stands in the way of kind of realising that way of working that you prefer within the White House?

Well, I think it's a couple of different things. So first of all, I mean, like, there are absolutely ways, you know, the action plan itself, before it was the final product, it went through a request for information, which is a formal, you know, process by which government agencies can ask the public to submit, you know, their thoughts on a given issue. Uh, something like 10,000 people did that for the action plan, which is a pretty huge turnout. Um, and a lot of them were, you know, really, really heartfelt things written by individuals. I, I read, you know, many, many of these, um, uh, comments. And of course, you know, every company— I mean, it really was amazing. Like, you know, just the, the, the range of corporations, uh, industries. Hollywood actors, right? Like, uh, uh, former politicians, members of the public. Um, so many people, uh, submitted, uh, comments for this. And then, you know, of course, like, I met with, with hundreds, if not 1,000-plus people, um, you know, from the public, uh, in various ways leading up to the action plan to gather, to talk about ideas and gather feedback. So it's not like that. It's more like just me personally. What I meant more is like, like I wrote a Substack post every week, and very often there was like lots of speculative stuff, uh, where I was like, yeah, I think we should probably do stuff like this. Um, and like, you just can't do that. You can't do things like that. And the reason you can't do things like that is that, um, you know, when, when you are working on something that is There is just a different gravity that happens when something comes from the government. Um, it's a huge machine. We can think that we're stepping lightly and in fact be stomping around. And like, uh, you know, um, so you— if we were to say like, oh yeah, like, well, here's kind of like— if we shared like an early draft of the action plan, let's just say, right? If that was like, yeah, here you go, like, we're putting it up on archive, like, let us know what you think. Kind of a thing. The blowback, the plausible blowback from that is just enormous because like it could be political scandals. You could there are conversations going on with Congress. There are a billion trade negotiations going on. There are companies that'll freak out, right? They'll be oh this would damage my business blah blah blah blah. And so I mean this is why like. This is why— that's not the only reason, but it's one of the fundamental reasons that people leak stuff, is to basically make things like this happen tactically, like as a move within the bureaucracy. It's like, you know, this other guy is pushing for something that's really controversial and would be really unpopular. So what I'm gonna do is, you know, he's doing that at like the staff level, so I'm gonna go leak that to the New York Times. And then like, You know, all the senior people are gonna see it and they're gonna get mad, and that's gonna kill the idea. And so a lot of times when you see leaks, what you're seeing is someone attempting to kill something from within. Um, and it— a public disclosure process would totally, like, do that. And then you wouldn't— there need to be lots of procedure around it, uh, et cetera. The other thing I would say is that the action plan, um, you know, like, one thing in theory that you could do— this isn't something like— Government doesn't tend to do this for, like, reports, which is kind of what the action plan is. But it will, it does it regularly for public policy where it will release, like, draft regulations with notice and comment rulemaking, right? There's various different things you can do. You can request information on a potential rule. You can be like, look, we're thinking about doing this kind of regulation.

Yeah.

What should we be aware of here? Like, what do you think we should know? And that's a structured way of, of, you know, doing that. Um, you can also release drafts of regulations and say, we invite comment on that. You can totally do that. Um, and, and, you know, governments do it all the time. I think you— The, the trade-off specifically with the action plan and, and the, you know, the process for getting the action plan from a blank sheet of paper to the final, you know, product was like largely conceived of and run by me. I obviously ran it by other people, you know, higher on the totem pole than me. I was like, if I do it this way, is that okay? The trade-off between like having a really collaborative process where lots and lots of people, even inside the government, are sharing comments is that, I mean, it's just the classic design by committee thing, right? I would say as a general matter, You know, interagency feedback from, for me, from my perspective, was almost always like really productive, constructive, and, and useful and like essential for making the action plan good. But we did it, we did, we did employ a different prop— like we didn't do a traditional interagency pre— usually like what happens is the White House, if there's like a policy you're working on, the White House will convene.

Mm-hmm.

Uh, a, a policy process. And, um, you know, every, every agency with potential equities in that thing shows up, and then you kind of like go— you sort of just, you talk about the thing, you conceptualize a policy, you share drafts. Um, and, uh, you know, that feedback process can be like really quite extensive. We kind of did that for the action plan, but we didn't do it super formally. And we only shared— agencies, for the most part, only saw the portions of the action plan that were directly relevant to their agency. Because the one thing that I felt very strongly about is like, I did not want every single person in the federal government commenting on every aspect of the AI strategy, because if you do it that way, you know, Yeah. There's just gonna be more people with unproductive comments, and I think probably a lot of cool ideas might have gotten squashed by that, to be quite candid. Um, and so, uh, I think it's, it's absolutely essential that you do a formal process if you are actually doing policy. You know, the action plan is recommendations for policy. They're like, I would say, like strong recommendations, but it is not literally— the president is not saying, you know, president didn't sign the action plan. Right, as opposed to like the executive orders that the president signed, where like that is that uses words like shall, you know, not may, not should, but shall. And that is like the president is saying this is a command: go do this to the to federal employees. And so that's a legally binding document. The action plan is not legally binding. If you're making legally binding policy, then of course you have to go through a traditional process, and you have to tolerate some of the bumps and inefficiencies of that. The executive orders, for example, that the president signed. signed along with the action plan went through traditional process, but the action plan itself, a little more flexible.

One big background thing that I'd love to understand from your perspective is, generally speaking, what is the political right thinking about AI now?

Right.

We're headed for this in this 4-year time. Many people are saying that we might have AGI, who knows, maybe even superintelligence, certainly some sort of powerful, potentially transformative AI in this term. And as with many things, when you get into power, it becomes kind of your problem, where in the past you were sort of just able to comment and criticize the people whose problem it was before you won the election. So how would you characterize the factions, if you want, or maybe just different perspectives that ultimately make up the coalition that elected President Trump. And what different sort of perspectives or priorities are they bringing to the AI discourse there? I guess I would start maybe with just Trump himself. We've heard a few comments, but I don't really know, how much does he think about AI? Is he using ChatGPT? Does he have a strong take on some of these core issues? And then there's the tech right, which many listeners to this show, I think, were at one point optimistic was going to be ascendant. And it seems like maybe sort of. There's also sort of the religious right, which has maybe a very different set of priorities, and it would seem like in some tension with the tech right. And there's probably other groups that you might put into that, um, mix as well that have like strong and perhaps, you know, quite distinct points of view. Um, yeah, how would you lay out that landscape for us?

So the, you know, the president was elected with like a really diverse and broad coalition. Um, it's one of the things that is, is, is, is so striking, and I think such a, such a narrative violation for like what you've heard from a lot of people in like sort of more left-wing media Yeah. Over the last like 5 to 10 years about the Republican Party and President Trump. Um, the reality is that, um, in terms of, you know, income level, ethnicity, uh, way, way of life, background, et cetera, the, the president's coalition is just really, really diverse. Um, and so, you know, it, uh, is a coalition that has many different views on AI, and those views are also evolving in real time. I think you actually like started to see them evolve during the time that I, I, I served in government quite significantly. So yeah, I would say like, um, very broadly speaking, um, you know, there are people who come from maybe the somewhat more traditional, like deregulatory impulse of, uh, of, of conservatives that's been around for a long time. Um, president, uh, uh, you know, certainly is not a, uh, uh, um, you know, a hardcore libertarian, uh, by any means. But, you know, he has a lot of that, like, yeah, look, like, fundamentally America— the business of America is business, as, as President Coolidge said. Um, you know, I think that's like very much like part of, part of, um, you know, the president's personal intellectual DNA. Um, so like, I think, uh, he definitely feels that way about AI. You know, I think all of the principles, uh, of the action plan are things that he thinks are, are, are enormously important. He has now weighed in publicly on issues like copyright and preemption where, uh, um, you know, and, and, and on both of those things, he's, he's kind of leaned, I would say, uh, in the direction of, um, AI development and adoption, right? He's leaned in the, in the, rather than putting up new blockers. Of course, the environmental permitting and things of this kind, something that everyone in this administration is very serious about. I think there are some areas where there's a rift, I think, among some on the right based on just like the level of risk that you perceive from, from the technology and also sort of like how consistent you see the technology. With being with like social media and the internet and things like this, there's a lot of, um, hostility, and frankly I think justified, on the right, um, against the big tech, uh, you know, particularly the social media platforms, uh, uh, you know, UGC type of platforms, um, where they, you know, I think for a long time people felt as though, um, right-wing ideas were discriminated against, and I think it's true. that they were. Uh, and I think things like fact-checking and misinformation analysis and things of this sort, uh, very often were deliberate attempts to, um, to shut down right-of-center viewpoints from, from broad dissemination on the internet. And I think that's a shameful thing, uh, and I think those companies should be ashamed of it. Um, there are a lot of people on the right who I— and I have argued this Both in public and privately in group chats and all sorts of things. I have tried to say like, look, um, uh, you are fighting the last battle and you should stop doing that. Like, stop fighting the social media battle because while that is an important thing, no doubt, like it is important and certainly like, you know, the outputs of AI systems matter a great deal. Um, uh, and, you know, ensuring that they're, that they're truth-seeking, uh, and, and not actually like, you know, the result of, of sort of top-down ideological programming. I think that is like extremely important. Um, but it's really interesting because like you talk about that issue and you get really quickly into the most important issues, you know, of the traditional AI safety world. Very quickly, because very quickly, like the issues implicated by, for example, the woke AI EO, um, the federal procurement, uh, that the president signed, you get very quickly into things like, well, how do we actually know, you know, what this, that the system is aligned, right? How do we actually know what this thing's gonna do, that it's gonna do what we want it to? Um, you get very quickly into issues of concentration of power. where I think there's a broad perception that, like, my God, this technology is going to be so foundational. We really need to understand the character, the virtues of these systems, the values that they hold. And so I think not— you know, reality has a way of coming at you. sort of regardless of what, you know, like, if you believed that AI, that those kinds of issues that I just said, like loss of control, uh, alignment issues, things, if you thought that that was like— I think a lot of people a year ago on the right would have been like, oh, that's all like lefty doomer stuff, that's all like EA stuff. And now they're like actually coming in and like thinking about, okay, but like, I care about this. And it's like, wait a minute, uh, like, wait, like, these are actually fundamental issues in AI, which is like— What, you know, people have been saying for a long time. Uh, and so I think, um, uh, you know, what you're actually starting to see is, uh, the right get their head around these issues in a much more, uh, serious way. Um, and, uh, I, I actually, like, I have predicted before in public, uh, before I joined the administration, um, right after the president won, I said, like, I think there's actually a reasonable chance that like the Republican Party is the better home of the AI safety world in the long term. Um, because just because of the way that the incentives of the party work and the way that the party is kind of hooked up. Um, so yeah, I mean, like you are starting to see that and that is, there are some rifts there and some of those things are like people fighting the last battle. Some of those things are like people starting to get their heads around these bigger issues and like, what are the values of the system gonna be? Um, I think a rela— an issue that kind of like falls right in the middle of those 2 things relates to the, uh, uh, um, sort of child safeguards specific to, uh, uh, child use of AI systems. Um, there's obviously the very, very tragic story of, um, uh, the boy who killed himself, I think 14-year-old, in, um, in Florida. Um, you know, that's an issue that really resonates. Uh, it's, it's something that a lot of conservatives are very worried about. And, you know, I'll just put my cards on the table, and for sure this is not federal government policy. Uh, this is not the opinion of the White House or the Office of Science and Technology Policy. But like, when I see big, well-capitalized companies that are Um, you know, making pornography with AI available indiscriminately to people. I'm sure that there's some sort of, like, age gating, but I doubt it's really all that, um, good. Uh, like, I just get mad for so many reasons because it's like, obviously that stuff is inevitable, but like, let that happen on like— let that be the open source model on the— in the North Korean you know, porn bot farm. Like, don't, like, don't take money from, like, the world's biggest institutional investors and be one of the flagship brands of AI to the world and then do this kind of stuff in such a public way. It's so crass. Um, and, you know, it's not gonna end up well with conservatives. It will not end up well with conservatives. And so, um, you know, I think that, like, these culture war issues that have been going on for a long time, Uh, are, you know, still happening and they animate issues in AI. And that like intersects in weird ways with like preexisting AI safety stuff. And you're starting to see those 2 things merge together. Um, and I don't exactly know where that's gonna end up. Um, you know, my hope, and I think like what the action plan, um, is all about is like, look, like I think there are good LLM safety laws for kids that you could pass. I think there are also really bad ones. I think there are prudent things you can do, and I think there's lots of non-prudent things you can do. But the point of the action plan is to, to a certain extent, is to say, like, look, we're early in this and we can— we don't have to fight. We don't have to be at each other's throats. We can identify reasonable things that make progress at the relevant margins, and we can be positive sum about this. We don't have to play the traditional game. Um, and, uh, that's a very important part of, of what, what I was trying to do with the action plan, what we were trying to do.

So yeah, I guess I, I hear you saying it's a live question still, basically, right? We've got the deregulatory impulse, we've got the sort of cultural conservative protect the kids and uphold traditional values impulse.

Yeah, and, and like, no one's gonna win. You know what I mean? It's not like there's ever gonna be a day when you're like, oh, that side won. I, I actually just like want to dispute the idea that there are like factions here, like different sides. My view of it is like, yeah, there's people with different emphases and different priorities on this stuff. Um, but like, if even within, you know, the Republican Party, they perceive themselves to be enemies of one another, it's like, well, geez, you know, there are some zero-sum issues in politics. Don't get me wrong. Like, that's definitely true. But a lot of things, they're, not necessarily that way. Um, and so, uh, yeah, I would say though, uh, you know, right now, um, like, I think the— I think there's just a lot about like capitalism and like the general like structure of American society that is inherently going to, uh, be accelerationist and like create incentives to, um, to uh, uh, develop AI rapidly and diffuse it rapidly, etc., etc. Um, and so like, in that sense, I think AI development is going to be in a very good place. Um, but I do like— I think, I think it will be really interesting to see like how the somewhat more skeptical people, a lot of whom remember big tech, you know, are, are still Sometimes, I mean, it's like literally the same people in some cases, right? It's like, well, wait, like you guys built YouTube and now you're building Gemini and we didn't like, we thought YouTube censored against us and seems like Gemini also might. So like, we're worried about, we're very nervous about this. We're very nervous about like the amount of money that's being spent and the capabilities that are being promised and people like me saying, you know, oh, this is gonna be foundational to everything you do in 10 years from now. Does not necessarily alleviate those concerns. It often makes people more worried. Um, and it could totally be true that that ends up turning into a pretty unhealthy impulse, uh, and we get a lot of laws that freeze our society in amber, um, because of various cons— because, you know, you don't want change. Like, uh, um, and I think, uh, um, on the other hand, I think it's possible to, um— To, to, to, to develop, to develop ideas in like a, a way that actually is productive. And that's— and I will tell you, like, that's another thing that I think I can do. That's part of, you know, why I chose to, to leave. Um, I think I can contribute to pushing things in a more positive direction there. Um, and, uh, you know, I think I can probably do that more effectively outside the administration than I can within.

How would you say rank and file Republican voters are thinking about this right now? Are they being— I mean, generally speaking, it seems like the survey results are like more bipartisan consensus on this particular issue than on almost any other issue, right? People are generally worried. People generally want the government to do something. The public is pretty warmed, I think, to the Terminator scenario style risks, but salience is low. Who's leading who right now? Is the public telling the political class what it cares about and political class is listening, or is the political class leading the public? Or is the public just not focused on this enough that they're really moving the needle yet?

Yeah, when it comes to AI, I think for— I mean, it depends. It definitely depends. And a lot of our politics is geographic, right? So there's senators that represent particular states that have a lot of some kind of industry that is particularly affected by AI, and therefore those senators, has a higher political salience than it might, uh, in a state that doesn't have such things. But as a general matter, I think it's probably right that, like, I think AI in general, in terms of, like, people who think a lot about it, I think it's— and especially AI policy, like, I think it's probably still more of a, like, elite coastal type of an issue. than it is like an issue that normal people are like truly fired up about. I expect that will change in some ways. I expect it'll be very hard to predict how that will change. But, you know, one of the, one of the things that, um, I had it written on a whiteboard in my office, uh, for a while, um, is a mantra the vice president, uh, had, uh, at an event I attended where he spoke. Um, He said, you know, our job is to make normal people's lives better. Um, and I just, like, really, really like that. That, for whatever reason, that resonated with me quite deeply. Um, and, like, I think that our job is to, um, to do things that make normal people's lives better. And it's also like, we— I think we have to communicate about the ways in which AI can make normal people's lives better. Um, and not sort of talk in abstract ways about the future. I think it's like the art of making it more concrete for people, um, is, is, is gonna be extremely important. But yeah, eventually this will gain higher salience. It'll probably do so in some sort of a scandal or crisis or something like that. Who knows exactly what that'll be? Um, seems like it's just kind of, but thus far it's like a pretty continuous, just like up and up and up getting more so. But most people are like, much more concerned about immigration and the economy and things of that kind.

Is there any utopian thinking or anything similar on the right that people look to for inspiration? As you know, I always say the scarcest resource is a positive vision for the future. And I don't know if I'm missing any visionary writing or thinking on the right? Is there anything like that at all?

Not really. Not that I'm sure some people are and I just don't know about it, but not that I'm aware of. Um, in fact, you know what's funny is, um, we got accused by a relatively prominent person who, um, is, is, uh, often in right-of-center circles, um, on— I won't say the name, but a pretty big account on Twitter We got accused in the action plan of being, uh, um, you know, utopian because we talked about AI being able to unravel ancient scrolls once thought unreadable. And I was like, that literally happened like 2 years ago. Like, this is—

Yeah, I think that kid now works for Doge, right?

Lucas, Luke is now at Doge. That's right. One thing is that like a lot of people are not aware of just like the amazing things that are happening right now and how remarkable all of it is. And just like, yeah. Um, so I think, um, there needs to be much more of that. But I think the ground— this— and it's actually, again, it's— this is one of the many things that I hope to do, um, in, in my work. Not necessarily utopian, but, but positive visions. Um, I hope to, um, to do some work along those lines. It's something I was sketching out, some stuff where I just wrote about like, like relatively mundane industries and just like automating, you know, like the various ways in which AI and automation are gonna, are gonna be transformative and like really, you know, in, in cool and interesting ways. Um, and like improve people's lives as a result. Um, there's definitely more of that to do. I think, um, you have to make it concrete. And I think, but one of the things I've realized is that like, when you start talking about that stuff, if it's like, if you, if you're doing like true pie in the sky, like, oh, we're gonna have like civilization on other planets and stuff, people will just be like, nah, yeah, but like, we've heard that for a long time from techno optimists. Um, So I think part of it, what I've really realized is like part of the job here is to make people aware of like the astounding reality that is before us today. Like the actual miracles of modern technology that enable our lives all the time. And I don't just mean AI, I actually mean like broad, more broadly than that. This is something you really uniquely see, um, at the White House because so many people come through where it's like, yeah, we make, like, we manufacture like this thing that like is an essential part of everything, right? And it's like, wow, this is like incredible that this works, right? It's like incredible that like fiber optics works. It's wild. It's completely insane when you actually think about what's going on. I think most people don't. And so I think to some extent there's, like, just explanations of, like, our current civilizational infrastructure. I think you can do poetry there if you try. I mean, and you also— the other thing that goes into that is, like, you think about the ways in which government itself, like, plays a role in a lot of— like, I was in an airplane Uh, a couple of days ago flying back from, uh, from a trip, a work trip. And, um, just kind of like looking around in the plane and I was like, you know, like there are like, uh, technical standards made with government mediation and help for like every single thing in this aircraft, right? Like the plastic, the, or the, the chemical content of the sheathing around the electrical wire. in this plane has like a specific technical standard that like someone worked on, right? And like people maintain, and there are like meetings about this, and you know, uh, and it's like you just realize like how vast it is. And you, you really do appreciate that specifically in government, how like just unfathomably large the whole operation is. Um, and it continues to astound you when you think about the sophistication of a lot of this stuff that's going on all the time. Uh, that our governments just sort of competently do, actually. Like, it— I think we, we, we should absolutely want to improve. Uh, but I also think that we have a tendency, um, to, to— we're like, we're in a very negative mood as a country, and we have been for like the last 10 years or maybe, maybe more. Um, and We only focus on the negative. We only focus on the areas where, where, you know, where we're falling down. But what I consistently saw in government was actually like a great deal of competence, uh, and, and skill. So anyway, uh, I, I think that we actually just need to like give ourselves a pat on the back sometimes.

Certainly as a country, we still have an unbelievable amount going for us and, um, should not take that for granted either. How, um, How AGI-pilled would you characterise different parts of the federal government as being? Maybe starting with yourself, even though you're recently departed.

I've always been pretty bullish on deep learning for a long time. Right. I've been bullish on deep learning since, like, 2015. I think, uh, like, if— I guess what I would say is, like, AGI itself is so nebulous as a concept. And, and my view is, like, I think what AGI would have to be to be true AGI would have to be something that has, like, genuine, like, human sample efficiency and flexibility. Um, and I don't know that we're, like, as Especially close to that. Um, the analogy that I've always used, and, and frankly that, like, I've probably used on this podcast before, is like the bird-airplane analogy. That, like, human cognition is like a bird where, like, you know, I can go fly over to the tree over there and land right exactly on that branch. Um, and I can do it with grace and an energy efficiency that, that That, you know, is, is like really, you know, outrageous. Whereas, like, you know, an airplane is hugely energy inefficient when compared to the bird, and like, you know, it requires like a giant runway, and you have to build all this dedicated infrastructure just for the airplane. It's like an extremely unwieldy thing when compared to the bird, and yet like highly useful. And people like Dwarkesh have, have been kind of turning a little more negative recently. I think they are actually turning to, like, they're calibrating to roughly where I feel like I've been for, for a little while, which is like, um, basically just the idea that like, we are probably going to build the like cognitive Boeing 737. And that, that is going to be like super useful. But it's not gonna be an automated bird, right? It's— we're not just making a mechanical human brain here. We are going to do something different, at least in the beginning. Maybe eventually we get to the AGI, you know, to the more true AGI. Um, how many people are AGI-pilled, uh, within the admin though? And like, yeah, just to be clear, like, I think we're gonna build the Boeing 737 thing like soon. Uh, like Somewhere between like 2027 and 2030, I believe we will do that. And I maintain high conviction in that. Nothing I've seen changes that. I think the bearishness on GPT-5 is kind of nuts. Um, I haven't used the model extensively yet, but like, uh, yeah, there's a bit of like common industry commentary and discourse commentary. Um, in terms of within the admin, um, there's a good number of people Who are like pretty convinced that AI is gonna be like super transformative. Um, I don't know how many people have like— and there's a good number of people that aren't, right? There's a good number of people that are like, nah, I think it's gonna peter out and gonna be hype and whatever else. Um, it's a little hard for me to like describe the, the ratios because as you can imagine, like the people who I ended up working with probably were disproportionately likely to be, like other people who think AGI is good or AI is going to be really transformative. Um, I think I'd say like it's there, but what that means, uh, is different to different people. Um, like, I think it's pretty difficult. I mean, I think like the, the, there's a reason that the verb is feel the AGI, right? Um, it's because it's, it's, it's an emotional experience. Um, and there's a lot of people who I, I, I think have experienced feeling the AGI intellectually. Um, but I don't know that they've done it emotionally. There's a difference between the, the emotional experience of like feeling the AGI is like a pretty wild thing. And it's happened to me at different phases. Um, and also, you know, there's a certain aspect of it that's like not resignation, but like anticipatory nostalgia, I would say, where you kind of like. Um, realize that things are gonna— that certain things are just gonna go away and certain dynamics are just gonna fundamentally change, and you kind of like actively miss those things. Um, all that being said, I think there are people who have, who have done that, but I think it's just like, it, it means very different things to very different people. Um, but, uh, certainly, you know, uh, AI is the president's top technology priority. Um, it is one of the hottest issues inside the administration. Um, and, uh, it's just something that like everyone does care about and thinks about. So, um, I would say it's like functionally, uh, like a pretty AI— um, it's, it's an— functionally, the administration just places a very high priority on AI, and that comes right down from the president.

So 3 more little test points, I guess, to try to calibrate my understanding of that better. One, obviously a White House is going to deal with a tonne of different issues all the time. Is there a dynamic of like, well, there's got to be an AI guy in the room for any given issue because we just sort of have to expect that there's an AI element to everything? Or has that not really happened yet? A second one is, are there explicit sort of timeline assumptions built into any of the planning or reasoning? Like, we think there's a 25% chance Dario is right and we're going to get the beginning of mass unemployment by 2027. So we got to kind of have one kind of contingency plan that's for that and maybe another one that's for that doesn't happen. And a third one, obviously, there was a big bill that allegedly is going to increase the federal debt. And I wonder if there are people going around saying things like, which might not be— I don't mean to suggest this would be wrong, by the way, but it certainly would be a leap of faith to say that, well, AI will help us grow our way out of the debt so we can sort of afford to take this on because the AI will pay the bill. Um, for us in a, you know, in time before we like, you know, really get into big debt-related issues. Um, so yeah, comment as you will on both.

Right. So, okay. Um, going, going in, in order, uh, the first question was not about timelines. It was about what was, remind me what the first one was.

Just like, is there an AI guy in the room first?

Oh yeah. Increasingly, I think increasingly. Uh, not necessarily, um, uh, always. Um, and you know, there are areas where like I don't even know what our nexus would be. But yeah, especially I think, um, like, like post-action plan, uh, as, as people throughout the government saw like, A, how high a priority— I mean, you know, the action plan event had the president, the vice president, and 5 cabinet secretaries. It's like packed, like packed with like superstars from the admin. Um, and, uh, you know, I think when people— that's a signal to people like, okay, we really need to be thinking about this seriously. Um, you know, you'd be surprised how important signals like that are within the government, um, how much that matters institutionally. And more people are understanding. So it's like, has it like— is that like entirely true? But it's— it feels more and more the case. Uh, that, that like AI people are being brought into like functionally everything, which is tough because there's not— there's a good number of them, but there's not like that many, uh, in the grand scheme. Um, so they, they end up being stretched, I think, a lot of the time. Um, I certainly felt that. Um, in terms of timelines, the convenient thing about timelines, uh, is that there is one timeline that we're quite certain of, which is that the president Its term ends on January 20th, 2029, uh, and that happens to be like a pretty good, you know, there— so we— I mean, actually, I mean, truly, one, one of my absolute favorite things about working in, in, uh, the Trump administration is the sense of urgency that everybody has. It's the combination of a sense of urgency and a feeling of not being beholden to the past or traditional ways of doing things. Um, like We, uh, are, are, I think, really like developing a new conceptual lexicon for American statecraft. Um, and a lot of it we're doing, and it's like, I think, you know, the administration, uh, doesn't always frankly do the best job of communicating, uh, some of that stuff. And I, I think also that Um, they often do a very good job of communicating, but it gets misinterpreted. Um, and there's a lot of people who just ignore a lot of the great work and focus on whatever the latest scandal is. So I think Americans are like quite badly apprised of, of what the Trump administration is actually doing. Um, uh, so, um, like that timeline of early 2029 uh, happens to line up with lots of AI timelines. And so, um, you know, like when, when we were thinking about like the data center power issue, um, the default timeline that you give is like somewhere between 2028 and 2030, right? That's like the, like, yeah, we need to do that. And that just happens to line up really well. So I don't know, it's hard for me to say like what would— If that, if that Uh, similarity didn't happen to be there, like, how would people's timelines be, you know, different? Like, I don't know, that's a good question. Um, the debt thing, I can't say I ever heard anyone say that. I've certainly thought it, uh, myself, uh, but, um, but I can't say I've ever heard anyone else say that. Um, definitely another thing I have heard is, um, you know, on some of the, uh, uh, uh, like immigration debates You will sometimes hear people say, well, like, there seems to be inconsistency here that, that like, you know, the pro-immigration camp says like, well, we need, you know, lots and lots of people to grow the economy, to do all this stuff. But then also like, well, wait, are we about to automate a bunch of stuff? You know, like, are we about to massively improve labor productivity? And I would agree that actually is a, a dissonance in that argument. So, um, yeah.

Yeah.

Uh, there definitely are ways in which, like, some of the more, like, I, I would say, like, flagship issues from the, the president's, you know, campaign, uh, and, like, of the platform of the party, there are ways in which AI is now kind of, like, inserting— being inserted into that in different ways. Also, like, I mean, there's a million ways in which, like, like, border security is, like, in many ways an AI problem. Like, there's a lot of stuff that AI can do on, on things like border security. So Um, yeah, in fact, you've had a, um, you have had a guest on this podcast, a former, uh, DHS employee, Department of Homeland Security, named, um, uh, Michael Boyce. Um, and yeah, and he, uh, um, yeah, I mean, he worked on AI at DHS. That's border enforcement. You know, a lot, a lot of what they do is border enforcement.

Is there, is there any talk about, like, a sort of step short of utopia, but any sort of new deal for the American worker? I mean, and this starts to get into some parts of the action plan kind of go this way, but Bernie Sanders has recently said we should make it our explicit policy to share the productivity gains from AI with workers by, for example, having a 4-day workweek or whatever. Is there any traction with those kinds of ideas?

Um, no, but, and I, I wouldn't say either way. Like, I don't know that, like, certainly we, like, the, the administration takes a worker-first priority with AI, and that is, like, very serious. We do think about that a lot. I think it's hard to know exactly what that means at this stage. The administration takes, like, an extremely worker-centric approach to AI. I think we don't know exactly what the future of labor is. Like, we don't know how acute the labor market disruptions are going to be. We don't know, um, where they will take place, uh, or if, you know, if are they going to be distributed by occupation type, by industry, by, um, skill level or, or experience level. Um, you know, we don't quite know exactly. There's some data that starts to look a little worrisome for software engineering, but there's also like plenty of data on the other side of that. So I think we don't entirely know where that's going, but as a general matter, like we will.

Yeah.

I personally think something, and again, super not White House policy, but I think about a 4-day workweek all the time. Because, uh, you know, the, the 5-day, uh, work week comes from the last industrial revolution. That came in the '20s. That came during, uh, Calvin Coolidge's presidency and was the fruit of an industrial— like, I think sort of multiple technology revolutions that happened at the same time. Uh, you know, uh, and, and I think that's kind of what's happening in, in, you know, for us right now too. So it would not surprise me if, if— You know, at some point in the future, we did actually go to a 4-day workweek. Um, that's an idea of, of Bernie's that I have no, you know, I could see the world going that way. I wouldn't say I endorse that idea, but I could absolutely see that idea making sense.

Interesting. Yeah, that's cool. I think about drivers maybe is one really like mundane one, but the sort of analysis there seems so simple. In many ways, it's like we already have statistics that show Waymo is a lot safer than human drivers. We are starting to see in recent Boston local politics some of these things that have been slow to materialise from my perspective relative to my expectations. But nevertheless, we're starting to see this protectionism where it's like, well, wait, what do the Teamsters have to say about this? And then you got the people that are pushing back with, like, So the Teamsters say that thousands of people should die because we need to protect their jobs, which is harsh but not fundamentally inaccurate. And it seems like that is going to be a question that society is going to face that doesn't have nearly as much nuance as maybe a lot of the other things, because it's literally just like, is the AI going to drive the car or is a human going to drive the car? Is a human going to be required to sit there? even if the AI is driving the car. These are seemingly relatively simple questions by comparison to a lot that we're going to have. And we're just starting to see the political battle lines being drawn, it seems like, kind of right now. Again, later than I would have expected.

Yeah. Well, I think to some extent, though, there has been a lot of state action on— there were AV fights you know, at the state level like 10 years ago, um, uh, that, that all predate the current AI, you know, because everyone said self-driving was going to be solved like 10 years ago.

Mm-hmm.

Um, so, um, a lot of that actually started preemptively. I mean, that is a really good one, and I think there's like a lot of non-obvious stuff that happens there. Uh, um, like, I, what I'd say is I think you are right that like on the software side, it's much easier to make the case that like you will see traditional augmentation and you will see like, um, you know, people will become more productive and yes, there'll be some labor market disruption, but in general things will grow. Whereas like, yes, the self-driving thing is like, well, that's just kind of an ambiguous replacement. Um, and so like the question is, will we gener— will that generate new kinds of jobs of some sort though?

Right.

And I think the answer to that is like, maybe, right? Like automated logistics Um, and like the ability to move, to navigate the world, the— in autonomous ways, that could— like, I could see— I think the self-driving car is such a simple early example of like where we could be going. Um, and that could create all kinds of interesting new opportunities. So I don't even know there, but I do agree it's like more of a one-to-one replacement. Um, And, you know, the other issue that I think you're getting at that's, that's, that's very important is like the, um, the state, local, and federal division of labor here, right? Should we have federal autonomous vehicle, uh, rules or should we not? Um, on the one hand, it's a little crazy if I take a Waymo, uh, you know, from my house in Connecticut. to— into Manhattan, it would be a little crazy-making if I, during that time period, passed through somewhere between 2 and 10 different regulatory jurisdictions governing the safety of self-driving cars. Um, that'd be a little weird. Uh, maybe that's fine, you know, maybe that's just an aspect about the future that's weird and like, it's fine, you know. Um, I, I, I, I haven't thought like—

Uber kind of had to deal with that.

Yeah, Uber does. Uber has to deal with that right now. Yeah, exactly. Um, and, uh, like, we can— there are all sorts of ways to make policy issues like that, like, coherent and, and, like, solvable, you know? It's not that— it's not that big of a deal. Um, so, but then there are unambiguously things that should be federal. And, like, yeah, maybe it's a— I think maybe it is up to a city to decide how they want to deal with AVs, you know?

Um, Maybe.

And, and like, I kind of personally have that instinct. Um, I would rather like let that be, you know, the, the federal, um, let that, let that be an area where we experiment at different levels of government rather than just try to like occupy the field with one federal standard. Um, I think the other thing I would say that becomes really important there though is like when you think about autonomy in the physical world, I think especially in the physical world, maybe also digital. I think one of the things that becomes like full autonomy, like Waymo level, like no human, you know, really in the loop at all. One of the things that happens is, is, um, you know, that, that kind of like changes the nature of the liability, right? Because like all of a sudden you're held— Waymo is held to a very high standard, right? If Waymo is successful, or if these self— if Tesla is successful, Um, assuming like we really get rid of the steering wheel in the future, right? Assuming that's the future that we head toward, um, that's gonna be a world where like the companies that operate those vehicles are going to be— they will have the liability risk for like all accidents in the country on their balance sheet. Right now, that liability risk is on my balance sheet. What that means is that, that's, I think that like maybe is a problem, maybe not. It would probably vary by industries to whether or not that's a problem. I've made this criticism before about like basically AGI, like that's going to be nuts, an insane amount of risk to insure without a really, really well-structured liability system. That's actually like, especially in the physical world and for these labour market things, that's like actually probably a good thing if we're being honest. Because like, in order for one firm to be able to internalize the negative externalities of like 5 million car rides a day or whatever, a million car rides a day, like those cars are gonna have to be really damn safe. You are gonna need like a lot of nines of reliability. And so the safety benefit has to be like so overwhelmingly positive and the reliability has to be so good to, to really make it happen at nationwide scale. Um, that'll inherently be slow. It'll take a long time to do. Um, and, but I also kind of think like the existing liability system is like fine for that. Like, even though it does, it is kind of an, it's not like the most accelerationist thing, right? The most accelerationist thing would be to say like, well, the technology's important, so we should give them a liability shield. The Waymos should have liability shield. Like, I don't personally think that. Um, and I, I, I kind of don't think that's I'd be very surprised if the Republican Party goes in that direction too, at least under its current leadership.

Yeah, I think they should have to earn it too. I agree that put up or shut up when it comes to safety statistics is kind of my attitude, and it would seem like they are well on their way to proving it.

Yeah. And autonomy is just like, if we're talking about real autonomy, then I think We should have pretty high expectations, um, of the companies that are, that are doing that. I just, I think that's like, that's, that's probably true. Um, I don't think you actually need a lot more regulation than that, uh, other than like, yeah, you have like something governing like the testing regime. If it's a taxi service, you gotta have like, you know, licensing of some sort. But I think that's like fine. I think that beyond that, like, I don't feel like you need a regulator to be like, self-driving cars have to be safe. I feel like that's baked into the laws of America already.

Well, if there's one trademark of the cognitive revolution, it's that we take our time to get to the headlines. So we've done that here today. But the headline, of course, and we've alluded to it many times, is the AI Action Plan. This seems to have been, perhaps with no exceptions, the best received thing that the administration has done so far. Everything else, I feel like there's haters coming at it from all directions, rightly or wrongly. And we'll leave that aside because this is an AI venue, not a general politics venue. But the AI Action Plan seemed to get remarkably positive reviews from just about every corner, including from people who I think expected to hate it. So I think that is a real feather in your cap as somebody who I don't know if you would want to sign on to led the effort, but certainly played a central role in making that happen. Instead of me walking through it, why don't you just tell us, how do you think about the action plan? What's the story that you would tell about what it is, what it's trying to do? You could weave any of your experience. I saw a number of funny tweets along the way where you were like, final version, V2, revised final, final, whatever. So I'm sure there were some funny stories there. But yeah, take it from kind of the high level first and just give us sort of Dean's— how do you think about it now that it's been out for a few weeks and it's kind of in the rearview mirror? What's the real headline of the AI Action Plan?

The sort of like the object-level description is that we decided we would We wanted to write something that was not like a nebulous strategy document. We decided like, let's do a concrete to-do list for the federal government for like a lot of different things that we can do to advance the ball on AI. It's not necessarily everything we could or should do, and it's not the long-term answer to any of the burning questions that people have about AI. You know, we, we can't answer those questions, and I don't think it's a good use of our time to try to answer these unanswerable things or to pretend to Americans like we can. You know, what, what, what was very important to us is we want to deliver something that we can credibly execute for the American people. That's the heart of it.

And, um, Mm-hmm.

You know, from there, uh, like, one way to think about it is like, if you were to look at an outline of the document, there's like 3 big pillars. And then within that, there's like, um, you know, maybe like within each of them, there's like maybe a dozen, um, like headers. And then there's like a paragraph of text. And then below that, there's like recommended policy action. And there's, you know, somewhere between like 1 and 10 bullet points with recommended policy actions. America's AI strategy is kind of the headers and the text below them, right? Like those things where it's like, we need to, like, this is like, these are our strategic objectives. And so there are many ways in which agencies and also people outside of government can help advance those strategic objectives if they would like to. And then there's like a list of things below that that are like, here are, you know, 5 bullet points. Here's what we're going to do right now. But just so you know, more broadly, like, this is a big priority. Like, this is the strategic objective. So like, the action plan is kind of the bullet points, and then the strategy is kind of the header and the paragraph or so of text below the header.

And You know, I'm—

I think, like, one other thing that— it was never textual in the action plan. It's subtextual, but I kind of alluded to it earlier. I think this kind of basic idea that, like, America can do this, like, we can mature our institutions to deal with this problem. We can adapt, we can evolve, and we can absolutely lead the way. And we can do it in such a way that we don't have to be at each other's throats. We can identify lots and lots of win-win things here. So I'd say the action plan is a deeply positive-sum document that comes out of a city that is usually quite zero-sum. And That is a subtextual message all its own.

Yeah, that's interesting. Very interesting framing. So the 3 pillars, accelerate AI innovation is one, pillar 2, build American AI infrastructure, and pillar 3, lead in international AI diplomacy and security. Let's maybe spend a minute on each one. I don't want to go literally point by point through the whole thing. People can obviously read it, and they should. And they should read Zvi's rundown of it as well. Innovation seems to be proceeding pretty quickly. How much of this is sort of like stuff that's going to happen anyway? How much of it is like, what do you think are the real pivot points?

Right?

Because, I mean, anybody would look back at the last few years and say, yikes, AI innovation is happening fast. And if you listen to the people in the Frontier Developers, they're telling us we should continue to expect it to be quite fast. And we've got IMO gold medals to prove that, even though that hasn't quite hit the GPT-5 product surface just yet. If we didn't do any of this stuff, would we lose or would we slow down? What is the rate limiting factor here that you're kind of— or what is the bottleneck that you're alleviating? Or bottlenecks?

Yeah, so I would say the innovation section of the action plan Is, I mean, obviously it's number one. Uh, so it's the thing we think is, it's the first one. It's the thing that we think is like very, very, very important. Um, and, um, you know, like you are correct that at least today could change 6 months from now, but today there are not that many laws on the books in the United States that govern the development of frontier AI systems. And so, Um, it is true that like right now, you know, the innovation there is proceeding apace. But where I think we need to be more reflective and, and, um, you know, self-critical, I think, as a country, is what about, um, transformative adoption of AI? So like a word that, you know, is throughout the action plan, it's a theme that's Embedded in every single item really is all about adoption. What I want is like, I want flying cars and, you know, automated agriculture producing abundance of food and nuclear fusion. And, uh, um, uh, you know, I want like the commerce, right? With like tons of agents bidding on everything around me all the time and like super like hyper hypermarkets for like, like that stuff is cool to me. Um, and I think those are the areas that's going to be what changes the world, you know? Um, so when people are like, um, you know, well, whose model is going to set the global standard? It's like the people who are going to win, set the global standard, are the people that find product market fit. Like, this isn't about, like, we don't sit at rooms, you know, in the White House, uh, and like sit around a table and say, what should the standard be? And then write a standard and then like. Try to convince other countries to adopt that standard. Like, sure, people do that, but that's not the path to victory in, in technology. The path to victory is, um, people emulating your use cases and using the tools you build because they are useful. Um, and so there's a lot of adoption where I think, I think we, you know, I think we're doing pretty well. And I think the American AI ecosystem is maturing in, in a lot of really impressive ways. But that's sort of the relevant margin for me, is the adoption. And one of the things that's very subtle in the plan is we talk about federal doing an RFI, OSTP, that's my old office, doing a request for information from the public for like regulations that impede AI adoption. And like, that's a subtle thing because we're not talking about like regulations on AI for the most part. And we could be, maybe there are things that agencies have done that we, you know, and there are some, and, you know, those are areas that we should look at. But what I'm also thinking about there is like when a, Are there laws or regulations that have assumptions built into them that are going to be made outdated by AI and associated technologies? Great example of this is like in surveying of construction sites, right? There are like oftentimes state, local, and federal laws that require, um, you know, you have to like survey this construction site and check for whatever, uh, you know. Um, And it, it oftentimes the way it's written is that a human being has to do it. A human being has to personally do it, as opposed to like, well, what if we just had continuous monitoring of sites through drones with LLMs built into them that have contextual understanding of what they're looking at? Like, turns out that's actually illegal, right? So that would be a great example, and there's thousands of things like that. So that's where I'm— that's, that's really kind of what we're thinking about. And then the other thing that I would say is really important is, um, there are some areas like science, science, uh, where, um, the US has— the US federal government has pretty high leverage over those institutions. And so, um, we can specifically drive things there. Um, and like one of, um, one of the examples of that that I've been on about for a long, long time is like automated experimentation.

Yeah.

Right. Uh, and so like you saw, uh, National Science Foundation, uh, announced their, uh, Programmable Cloud Labs initiative, uh, which is gonna be $100 million, um, to different comp— to different companies, academics, et cetera, who are building automated labs for like massively scaled scientific experimentation. A few years from now, we could be in a world where like, uh, that There is automated science infrastructure that can be used as a cloud service, basically, uh, by AI agents, right? Uh, and, um, obviously significant safety issues with that. Uh, but, um, but like, I think the way to think about this is like that, that might not be something that actually exists. Like, that might not happen on its own. Um, the, the, they'll exist, like automated labs exist today, and they're inside of corporate R&D labs, but it might not be— be shared, you know, infrastructure. Very much like how the federal government led the way in the early days before there was a commercial application of high-performance computing facilities. The fed— there wasn't one, and the federal government viewed it as a public good, as common scientific infrastructure. And so we built that, and then we also built the internet because we needed to network those facilities together. I think there's things like that. That, that are very exciting. Uh, and the idea, you know, of, of the NSF thing is to build a network. So that's like one very specific thing, but there's a lot like that throughout the, the plan. Um, there are a couple other things that are about like trust and reliability of the technology. Um, which I do think is gonna be very, like, if, if the, if what people know the technology, uh, if people understand AI is like, the thing that makes it impossible to get justice in court anymore because, like, you can't validate, you know, media. Like, that would really— that would really be awful. And it just so happens that, like, there are levers that we can pull inside the federal government on that, like, now. So, like, let's do it, you know? Um, that's— that's a good example of, like, Something that is actually a bit of a, like, that's a risk management thing, right? Like, that's more of a risk type thing, but it's in the innovation section for a reason.

What do you think is more promising to address that? Is it like the encoding of origin into synthetic media, or is it the, like, on-device attestation type of stuff for real cameras? Or maybe it's both, but how do we get out of that problem?

Plausibly it's both. Um, but I think the thing that's actually going to work will be the— you gotta— you should focus policy effort on the scarce thing, which will be the human. Uh, well, I don't want to say human-created. What I think I want to say is the scarce thing will be like actual photons that hit actual glass in the world, not, um, um, you know, like the— we're processed by real image sensors, not the AI. The AI-generated will be the super abundant thing. So I do think, yeah, it's pretty much like in the long term, what we're going to have to do is have some sort of common standard for validating real-world stuff. I think it's actually not that bad yet for the most part because I think we'll have to have a new level of scrutiny for image and video-based data. But it's still, even like today, even like v03, it's like, it, it's still different enough from the real world that like you can go back and like do, you know, with the, with the scrutiny of a legal system, you know, there are, um, there are still things you can do, but it's like an area that we, we need to make sure we're actually doing those things.

On the diffusion point. I wonder how you see that, or how you could maybe say how you see it, or how you think the administration broadly sees it. I've been struck by kind of two contrasting viewpoints. One is, as I'm sure you're aware, Jeff Ding has sort of popularised this idea that the US leads in terms of our ability to diffuse technology through society and get it to a broad base and get the the practical value. And he argues that China lags in that capability. But then I also see a line in the action plan that is enable the adoption of AI in the Defence Department. And in an episode I have today, Jake Sullivan also talked about the memorandum that he and team put out that was basically arguing that our national security establishment broadly, maybe, or maybe the military is sort of less agile than our Chinese counterparts when it comes to adopting new technologies. Would you say that that's like your view as well? Like, we're better in the private sector, we're slower in the government? Or would you complicate or contradict That, uh, announcement?

Uh, so I think it depends. Like, one thing that America is quite good at, um, compared to, um, well, everyone else in the world, including China, like, we're very good at, like, uh, the pipeline of, like, deep capital markets to cloud computing and cloud computing applications. to, you know, consumer and enterprise adoption of new technology. We are like pretty darn good at that. And like China does not have a lot of like, you know, the much bemoaned B2B SaaS, right? Like China does not have a ton of that. And like a lot of it's actually like pretty useful. Like it's actually like ends up being quite important in like the sort of cybernetics of the business organization. So it, it matters a great deal. In fact, um, I think we should be proud of, of the fact that we, we do well in things like this. Um, there are other areas where, like, in terms of AI adoption in the military specifically, I can tell you that our military, uh, is, is not as agile as I would like it to be. I have not carefully analysed, like, the, the, the difference between China's military adoption and our own. I think one thing— this is always true in analyses of China, um, but China, you know, a lot of their governance of themselves is very like KPI-driven. And so there will just be like, like there'll be directives from on high. Like DeepSeek comes out, right? And it's like there's a box. Every bureaucrat in every part of the country will like in a day, you know, or like a week later or something have to like check a box that's like, did you use DeepSeek for something? You know, are you adopting AI? And what that ends up with is like summary statistics that would suggest that China is adopting AI more quickly in its government than the United States is. Um, but, uh, a lot of that adoption is pretty shallow. Um, uh, you know, I, I think we need to move it more nimbly for sure, but I'm just always aware that that's like, that's like an issue when you're analyzing China, is that like they will oftentimes like, you know, optimize for like specific number, for hitting specific numbers, numerical targets, um, that are like now a little like I'm not sure if that's actually connected to the thing that you want. I think what you really want is like deeper adoption. Um, because as like part of what happens is like we, we drive deeper, we drive deeper adoption of AI, and then that reveals new problems and new subtleties, and that creates a positive feedback loop with product development and finding product market fit. Um, and so like you're starting to see that, right? Like, like, You would not have guessed the form factor of Claude Code. It's a very interesting form factor, actually. You would not necessarily have guessed that you would be using CLI-based tools to like do some of the most cutting-edge, you know, AI automation 3 years ago when you first saw ChatGPT. I don't think that many people would've guessed that. Um, and yet here we are, right? So you're starting to see all that happen. You'll see much more of that kind of stuff happen as we figure it out. But I still think a market-based process is gonna like do a better job, uh, at diffusion if we don't get in the way of it with bad regulation. Because what China will do is like, if there is some transformative use that's blocked by a regulation that they care about, they will unblock that. Like Beijing, Beijing will like put out a directive and it's like, unblock that regulation, and it will happen everywhere, right? So that's where they're better. They're just better at like cybernetics than our government is because we're not a centralized state, right? We're not a centralized authoritarian state, uh, fundamentally. Some, some people think that's what we're turning into. I think those people are dead wrong. But, uh, um, certainly it did— I, I will tell you, as someone who served in the Trump administration, did not feel like I was, uh, I did not feel like I had authoritarian levels of power over anything. So, yeah.

With the military in particular, obviously it seems very fraught. It seems like there's kind of a major mismatch between the way the military has traditionally thought about things and maybe should think about things versus the way that LLMs work. It's like my kind of standard refrain on that is I would want to know that any issues of deception or scheming against the user are fully resolved before I go into combat with my AI battle buddy. And I know we're maybe not immediately jumping to AI battle buddies, but If it's just the kind of like making paperwork more efficient at the DOD, that doesn't seem like the kind of change that doesn't seem like it's going to beat China. Not that I'm, as you well know, focused on doing everything to beat China. But if it is like sort of actual combat operations applications, it seems like we don't quite have the right AI for that.

Right now, I think one thing about the— about military history is that, uh, the, um, the flow of information throughout, uh, military organizations ends up actually very often being like a quite decisive, um, uh, um, advantage that, uh, militaries have. There's a great history, um, I read once, or sort of like analytical book called— I think it was called Information and War. Um, that was about how different communications technologies ended up, um, uh, changing the nature of warfare. Uh, and, you know, it was— there was stuff on the internet, but there was also stuff on, like, the adoption of, like, the radio, uh, inside of militaries and how actually that completely changed. Like, the radio was the first thing that allowed for truly centralized command of militaries, which changed the whole Everything that was possible in warfare. So I actually would dispute somewhat the idea that a pure information processing technology is not actually like a quite important military advantage, even if, like, even if you're literally just talking about like GPT-5, like today's technology, something that's like capable, in particular if you think about synthesising of like, I mean, the staggering amount of information that our government collects on the world every day. Um, you know, a lot of it the intelligence agencies and deeply, deeply secret stuff. Um, but, you know, we, we pick up a lot of information on the world, and the ability to quickly, um, analyze that is, uh, uh, like, and, and make, and sort of make decisions or present decisions to, to humans, uh, you know, for a final call. That could be quite decisive. And that's basically just a pretty traditional LLM adoption story. I mean, it's more things than that too. It has to do with information sharing also, and that's quite difficult. When it comes to the weapons side, I think once you get into the physical world, I think DOD starts to become much more equipped at, okay, we need to—

Mm-hmm.

Like, DOD is good at being like, all right, like, this is— these are the character— these are the performance characteristics that we need before we can adopt this thing. But as you'll see, like, first of all, on the DOD adoption side of things, the action plan has a section about, um, you know, uh, sort of a, a physical facility for testing of autonomous technologies. Um, and that will be, you know, that'll be a physical thing that the DOD will use to, um, uh, to be able to sort of write those specs out. But I, I have faith that their culture is institutionally well suited to that if they have the right tools at their disposal. Um, but another thing, of course, is, uh, you know, interpretability and, and control. And, you know, you might call control alignment, right? Like, this is an area of, I think, deep importance for the military, particularly— I mean, when you think about the fact Who knows how an LLM, how this would work in a system deployed at DOD? But when I ask an LLM questions, at this point, it is situationally aware that it is talking to, like, O3 is situationally aware that is talking to a person who worked at the White House on AI policy. And I'm asking you a question about that like has a plausible connection to AI policy. Um, how is that affecting the model's outputs, if at all, is like, whoa, that seems important to understand really well. Uh, and I don't think we have a very good understanding of that right now. Um, and so hence, hence why I, I, you know, I, I think and expect that, that DARPA will, um, place a quite significant investment into those exact issues.

Yeah, I'd say little to no understanding of how that's impacting just yet, but it is definitely a fascinating question. That's a good point about information processing speed being an important dimension of competition. Obviously not the only one, but I think it's a compelling point that that could matter, even if there's no actual firing of weapons by LLMs at any point.

Yeah. Most military planners that I have talked to think that basically it's logistics and cybernetics, right? It's just like it's moving things about the world physically, and then it's moving information through. And if you do that, Yes, you will. Like, we focus on the sort of tip of the spear, which is the, you know, the, um, the hypersonic weapon or the, the, the autonomous drone or whatever else. Um, and that stuff is important. We need to be able to make sure we can do that stuff. We need to keep up, et cetera, et cetera. Um, but I think if you bolt that stuff on to a more, like, to a 20th century military model, I think you— even if we have the world's best hypersonics and we make them abundantly and we have a super big army of drones, If those things are being commanded by a military organisation that is rooted in the Industrial Revolution era technologies, then I don't think we will fight successfully.

So broadly, I guess if part one is like, don't get in our own way and let our private sector continue to lead in a relatively unencumbered way and do what it does best, we have advantages and we'll naturally kind of maintain those advantages because that's who we are. Part 2, around building infrastructure, is the part where it seems certainly much more plausible that we need to actually up our game to achieve the visions that we have of an intelligence too cheap to meter future. How optimistic are you, I guess, in the first place that we're actually going to be able to do this? There has been quite a long time since we were energetic, so to speak, about building nuclear energy or just building infrastructure fast almost at all. Right. Like, a few exceptions, I guess, but not too many. Is this something you think the administration is really going to be able to change in a short period of time such that we're actually bringing all these things online in whatever, 7 to 30 timeframe?

Yeah, the short answer is yes. I don't know that we will have lots and lots of nuclear reactors producing gigawatts of power, of new nuclear reactors producing gigawatts of power in the next 3 years. That would seem hard. I think we're doing some really significant stuff. You know, on nuclear, uh, uh, there are a series of executive orders that came that were driven in part by our office, um, uh, relating to nuclear that came out, uh, you know, a couple months ago. And, um, you know, there's like an entire reorganization of the Nuclear Regulatory Commission going on. So exciting stuff, to be sure. Um, you know, and I think fusion is plausibly not that— I'm, I'm way more bullish on fusion than most people in the government. That's one thing I can definitely say is like most people, they got like— there are people who got very mad that fusion was even mentioned in the action plan. Uh, uh, you know, people were like, wow, it's never gonna happen, it's 30 years away, you know. Um, and I think like that's just a sort of an older point of view that I think comes from not being engaged with the current frontier of that technology. But, um, you know, I think, uh, when you look at the next couple years, I think we will have to be, um, I think we're gonna do it. I think it's also gonna look a little different than, uh, what you might expect. Like, uh, I always remember, like, Leopold has, like— Leopold Aschenbrenner, Situational Awareness, has, like, this, this AI.

He's a first-namer. It's all, it's all good.

Yeah. Uh, has this, um, um, uh, AI-generated image in Situational Awareness that's, like, an endless field of data centers, and you see, like, big, like, like natural gas, uh, peaker plants, um, in the, uh, you know, distance. And, um, we're gonna build stuff like that. I think there will be things like that, you know, by the end of the decade. I totally do. Um, but like the idea, um, of building like, you know, 100 gigawatt or terawatt, you know, types of facilities, like, I don't think it's gonna be necessary. And I, I think, um, Like, also, the way that we will get there will be through, um, different kinds of unlocks that are important but not as well discussed. Like, one thing is that the American electricity grid, as a general matter, is actually quite over-provisioned, um, because the grid is designed for the worst-case scenario. Um, so it's designed for, like, you know, the day when it's 110 112 degrees in Texas at the peak time when everyone's AC and TVs and electric cars are on, you know, all that, right? It's designed for those moments of peak demand. Which means that the vast majority of the year, there's actually like many gigawatts, um, that are available to be used, um, assuming that you don't need the power 100% of the time. So like there's a, you know, there's kind of a viral report that went around that came outta Duke, a guy named Tyler Norris, that basically said something along the lines of like, if data center operators were willing to curtail their electricity demand for 0.25% of the year, You could unlock 76 gigawatts just from that, um, from the— without building any new physical infrastructure, just purely through what's called demand response. Um, and the problem though is that, uh, um, I'm getting really technocratic here, but, uh, but I mean, uh, The problem is that, um, you know, there, there's, uh, there's an entity called FERC, um, Federal Energy Regulatory Commission, uh, and, um, they regulate, uh, the interconnections that happen. Like, so if I build a new data center and I'm— this is a 1 gigawatt data center, I'm adding 1 gigawatt of demand to the grid. Um, what will happen when I'm trying to build something of that size? Is there will be, uh, the utility in a state will do, um, an interconnection, uh, uh, to the grid, and they'll do modeling, you know, computational modeling of your demand, um, to see how much, you know, like what, what's it going to do. And when they do that modeling, they assume it's completely stable growth. They assume it's like 1 gigawatt 100% of the time.

Okay.

And when you do that and you also factor in the worst case scenario day, then it's like, oh, well, we're gonna have to build like an entire new plant, natural gas plant, and we're gonna have to build like totally new transmission infrastructure to accommodate this. And then that becomes a delay and it becomes a cost for the, the, and the data center operator's gonna have to bear that cost. The utility will build it and the data center operator bears the cost. Um, you can, uh, there's a federal entity— a lot of this is being done at the state level, but there's a federal entity called FERC that when those interconnections implicate interstate transmission line electricity transmission lines, which is like a lot of them, um, FERC has, uh, uh, uh, FERC has jurisdiction at that point. Um, and so what FERC can say is Don't model it as a stable demand. If a data center is coming online and says, we're willing to curtail our demand by 0.5% of the year, then we're gonna give them a faster interconnection. We're gonna move their time to power. We're gonna get them on the grid in 2 years rather than 5. You've unlocked a ton of energy and you've accelerated demand. So there's a comp— and also the other amazing thing about this is that like there is hardware, for demand response, but there is also software. When you think about what demand response is of like, I got a signal from the, I'm a data center and I just got a signal from the utility that I need to cut my power by, by 50% in the next 10 minutes. Um, you have all this equipment running, uh, you have all, you have the GPUs and you have the cooling equipment, you have all this other stuff. Um, and you have like different priority batch jobs, right? Of like, okay, like there's all these different workloads being processed by the data center. One is like me making a cat meme and the other is like a patient's medical records in the same facility. Um, uh, all of a sudden that starts to smell a lot like a reinforcement learning problem, uh, doesn't it? Right? Like, that starts to smell like an AI problem of like, oh, like we need to like optimize how to scale down our power dynamically. Anyway, um, there's a lot of different things you can do there. And like that alone is cap— I think if we do a good job of it, that alone You can unlock like 100 gigawatts. And then in addition to that, we're going to make the permitting easier. And even if Congress does nothing, at the margin, that will be useful. And there's a lot of energy on state permitting reform too right now that folks are engaging in. At the margin, it will be better. Are we going to have like 100 gigawatt data centres by the end of the decade, or like lots and lots of nuclear power plants? Probably not. But I think that by mid-decade, it'll be pretty crazy. Like, by mid-decade, I believe that there will be lots of new nuclear online by mid-2030s.

I had not heard about this dynamic management, but I did have a personal experience maybe, I don't know, 6 or 7 years ago now in Detroit. One cold night in the winter, we all of a sudden got a text from our local utility that was like, We got some sort of problem. Please turn your heat down to 65, everyone. And that's only happened once in my life. Talk about things that we can be grateful for and shouldn't take for granted. The uninterrupted provision of such utilities for my entire life is definitely one. But basically that worked. People responded to the text and we made it through.

And obviously it doesn't scale to send texts to people. So you do need to automate this in some way. Um, but one of the things that's amazing is if you play the tape forward on that and like we were to roll out technology of that kind to like lots of industrial facilities, um, including the data centers, what the net effect would be is that we would utilize existing electric, uh, electricity generation more efficiently than we currently do, which would actually lower prices over time. So you literally can build the data centers while lowering prices if you do it right. Total, again, like theme of the action plan, win-win, positive stuff.

That's great.

Yeah.

Taking one step back, I don't know quite how to evaluate this, but it was reported in Defense One by Defense One, a quote from an anonymous Pentagon official. The quote is, we're not going to be investing in artificial intelligence because I don't know what that means. We're going to invest in autonomous killer robots. This administration cares about weapon systems and business systems, not technologies. Would you call that fake news? Or, like, I mean, there's certainly some fake news out there, but would you, like, how would you reconcile that with your, you know, characterization of your conversations with—

Well, I would say one thing is when I said conversations with military players, I was not actually necessarily referring to people inside the administration, like inside the Department of Defense. I was just like, over the years I've talked, like, I've heard podcasts and talked to various people that know about this stuff. So that's not so much a Trump administration view that I characterised. That quote gets at, you know, weapons systems and business systems. So that's the latter. The business systems would be like the latter case of the, you know, sort of information and communication stuff. Certainly, it's also true that, like, there's totally, like, transformative things, you know. Uh, hardware things that, that we'll be able to do with AI. And, um, I don't even know how to think about the form factor for those things, right? Like, I think drones are an early one, but there's gonna be so much more. And, you know, there's, I mean, I, I'm very excited about like autonomous, um, uh, you know, autonomous boats and autonomous ships. I think that's potentially real. I mean, I think one thing that's like, you know, America has problems with shipbuilding, which I hope we get better at. I think, again, this is an area where this administration is doing more than anyone. uh, has done on the shipbuilding problem. But, um, we're actually good at boat building. Uh, we're actually perfectly competent at boat building. And so if you build like lots and lots of autonomous boats, like, um, you know, that might be a really interesting way to think about the future of naval warfare. Um, basically like a bunch of like school buses, uh, flying, going around the ocean. Um, so no, I, I, to be clear, like I was not discounting the benefit Of, of, um, like weapons systems, of hardware weapons systems that are AI-enabled in some way or another. Um, I was just saying, in fact, like I was saying, like, I just think DOD will be better equipped to like buy that kind of stuff, particularly because they're getting significantly better at working with startups. And so like you're already starting to, you know, there's, there's this blossoming ecosystem of defense tech companies. Um, and I hope that stays, that, that continues and You know, Anduril's kind of operating as a new Neo Prime now and selling all kinds of AI-enabled capability, hardware capabilities to the government. So it's already happening. But, but I would also say like, what is the actual basis of Anduril? Like, what's the real, like the fu— the fundament of, of their business model is software. It's the Lattice software system, which is exactly about information sharing. It's so that like When we build new, we have a software platform. So they're like Apple as they add new things, it all fits into the Anduril ecosystem and everything can talk to one another and communicate in like really high bandwidth. So, um, that would be, that'd be a good example of like exactly what I was talking about.

Gotcha.

Okay.

Interesting. Um, when it comes to bringing chip manufacturing to the United States, That's another big challenge. I guess there's a couple of different angles I want to come at it from. One is we've got these Gulf deals where I'm interested in your take on why are we doing those deals? The answer I always get, which hasn't quite satisfied me, to be honest, is because we want them to build on our AI stack versus China's AI stack. But then I also feel like Well, does China really have any chips to sell them? And couldn't we have sort of, if we're concerned about American values, like planting these giant data centres in these Gulf countries, which I would say frankly simply don't really share American values, doesn't seem like the obvious move. It seems like we could have held off on that a little longer. Because they didn't really have anywhere else to go. But maybe they're doing something for us that I don't understand. It might be energy, it might just be speed of regulatory approval locally, maybe it's cash on balance sheet. And then the other angle, of course, is the Taiwan angle. And I wanted to get your take on how do you think they are thinking about this right now? Right. They have this sort of tricky position where obviously they're right in China's shadow. Everybody knows that that's a flashpoint. They've managed to put themselves in this position where they're super relevant, where if Taiwan goes dark from a Western United States perspective, that's a huge problem. And so we're at least compelled to be ambiguous about exactly what we want to do or would do under various Scenarios, but now they also have this sort of tricky situation where they they like need to be a good friend to us to keep that dynamic going, and that means sharing some technology and putting some TSMC know-how on American soil. But they probably don't want to overdo that, right? I mean, they don't want to share everything because then maybe we wouldn't need to defend them as much. It seems like the Gulf states are kind of trying to do us trying to. engineer their way into a similar position. And so, yeah, I don't know, break down the sort of various geopolitical strategies that folks are playing.

I think a couple of things, a couple of foundational things that go into UAE. And I should say, I worked quite carefully and closely on the Negotiations for that. Um, first of all, I think much like the action plan, the UAE, uh, framework that we agreed to is very positive sum. Um, what we are saying is like, this is not rival— we are going to do a big industrial build-out in the United States. Uh, it's going to generate a lot of jobs and it's going to generate a lot of wealth and it's going to be, uh, an asset that, that you know, people all over the world, and especially Americans, are going to use. But we don't see that as being rivalrous with the idea of other countries that are strategic partners doing really ambitious things too. So like, what makes the UAE special of countries around the world? You asked about AGI-pilled governments. The UAE has an AGI-pilled government. Uh, like, they are the most AGI-pilled country in the world. They, like, think about this technology in very sophisticated ways. I do think it's very important that, that America, uh, be partnered with other, like, other countries that are sophisticated and I think have a, have a pretty good grasp on the likely trajectory of this technology. I think it's really good if all of us, like, all of us leading, you know, uh, uh, countries, like, in terms of intellectually Um, you know, that we are all, um, partnered. Um, and, and, and we, you know, we don't want them building on the, the, the sort of Chinese stack. Um, you know, I think there's probably different estimates and different projections about the future of where China's chip sector is going to go, how quickly they'll be able to catch up. Um, I think as a general matter, it seems like the trend has usually been that they catch up a little faster than the technology analysts here, um, guess. I don't know. I don't have a, um, um, I, I don't have like a deeply like technically principled answer to that question. But what I do know is, as a matter of policy planning, which is, you know, what we were engaged in with this deal, um, I think you probably can't assume that they are going to be slow. Like, I— that feels like a weird hinge point to be, like, complacent about. Um, and, like, I got— I frankly got that, like, a lot of the people I know who are critics, and I think a lot of the people in the prior administration, um, were just, like, weirdly complacent about that one thing. They were like— it was like— It was like they got so— they, they just were like so enamored by like the specific thing of like extreme ultraviolet lithography that they're like, they'll never be able to figure it out. And it's like, oh, like, uh, I'm not sure about that. Like, it's hard. It's definitely hard. There's other ways to do it, right? Uh, it's like not the only way to— not the only plausible way that you can achieve that small on onto silicon wafers. Um, there's all kinds of different things you could do. And there's also scale, and there's also the fact that they don't You know, that like, they'll just eat the profit margins, right? They'll just—

Yeah, they don't have to make money.

Yeah, yeah, yeah. Like, I mean, that's all eventually they do, right? I mean, this is— I think Noah Smith might have said this, maybe on your podcast even, but like, the world where they win is just such a gray world. It's so dreary because like, you know what that means? Like, no profits means no new stuff. Uh, like, it means nothing new happens because no one can ever reinvest. And so great, like, You've made an ultra mega super abundance of chips. It's a dystopian hellscape that they're trying to build. Um, and, uh, at this point there's, you know, um, I, I, I, I dive— I diverge.

Um, but, um, um, it seems a little harsh, for what it's worth. I have to, I have to at least briefly comment that, you know, what I see of China doesn't look like a dystopian hellscape, even if you are a political dissenter, it might quickly become one. But But for most people, it seems like life is getting better.

If China occupies the role that we currently occupy, like they're the world's true frontier economy and the biggest global powerhouse, and we're kind of a significantly less relevant country, that would suggest that technology development becomes much harder. Because you have no, like, if you have no profit, you have no ability to reinvest into the business and make new things. Um, and so it's like, they just like take everyone else's stuff and then like, um, uh, you know, make it super low margin and drive them out of business and then make super abundance of that stuff. Like, in a certain sense, sure, that's fine. But like, um, the long-term result of that is a less innovative world. If that's like actually the strategy, it's a less innovative world and it's a less colorful world.

Um, I do think we see some exceptions to that though, right? Like Huawei is not doing that. They're notorious for like reinvesting huge amounts. And my sense is they are at the frontier, if not maybe like genuinely pushing the frontier in their Domain, right?

Uh, kind of. Yeah. I mean, I don't know. Like, I, I think, uh, um, Huawei, Huawei is an interesting case and there's only so much I can say, but, um, Huawei, Huawei is, Huawei is a somewhat different case. Um, you're right. It's not uniformly true, but I just mean like if you play the tape forward, it's, it's a less, it's a less beautiful world. Um, uh, but in any case, um, I think. Uh, um, where were we? Yeah.

So, I mean, this is all downstream of why did we have to sell the Gulf, uh, countries these huge data centers?

So, so yeah, I mean, like, I think that, um, we do want them to be like on our technology stack. I think that's like 100% true. Uh, I think they're like a valuable strategic partner. And then of course, yes, it's also the case that, um, Uh, you know, we think, uh, that, you know, there are terms that, that we secured, uh, with the UAE's government that, um, you know, require them to make reciprocal investments of similar size to the data centers that get built over there. And those investments could take the form of data centers, but they could also take the form of, you know, investments into, um, energy infrastructure, all kinds of things associated with the AI buildout. So, um, you know, when you're talking there about, about hundreds of billions of dollars, um, that, that they are like eager to invest, uh, into our country. And so, um, you know, I think the— we have to get the security details right. Um, that's what's going on right now. You know, that's what, um, You know, my, my former colleagues at the Department of Commerce are doing is getting the security details right. But if we do it, um, uh, you know, we're like, it's just, it's, it's, it's a total win-win. Um, you know, is it the case that, like, you know, do they have a different government and different values than we do? Like, they totally do. Absolutely. Um, But like, I don't know, I think the idea that we can only engage in commerce with like purely democratic countries, I just— it's just a weird rule. It's like it's never been true before. It's not like we've only— we've only done that before. And like, we would love to sell to other democratic countries. The reality is that most of— like, a lot of them, a lot of the developed democratic countries don't really like AI very much, and they spend more of their time talking about how to put a, put, put a straitjacket on it than they do talking about how to grow it. Um, and so, you know, I hope they change their tune on that. They are going to be big customers. I think they will regret the fact that they're not currently big customers. I think they will regret it in the relative near future. Um, but, you know, you did just see like Norway just did a big, you know, OpenAI deal. Um, for their Stargate for Countries thing. I hope, I hope we do much more of that. Um, you know, I, I, I, I— we want to sell to lots of people, right? Like, our export promotion program is about treating— unless you are like a strategic adversary of the United States, we want to treat countries equally and we want to engage in commerce. This is about commerce. Um, and so yeah, uh, there's— I think there's a lot of win-win to be had.

Yeah, I do support commerce, you know, with, with, uh, and trade, you know, with generally with people that we don't see eye to eye on everything. It does seem a little strange coming from the American right, though, which I would say is generally like not, uh, fond of, I don't know, Islamic values broadly.

Right.

Like there's a, there's a weirdness to that that I see kind of being memory hole. And I guess if I was really forced to pick whose values are more compatible with ours, the Saudi governments or the Chinese governments, I think I lean China. I mean, I know they're like a more serious competitor to us. That's a different question. But I've recently— this is a bit of an aside as well, but in the quest for, is there anything kind of stable and robust that we could try to align AIs too that would work out well for us if indeed they become super powerful in a not super long timescale. Ancestor worship has been one candidate that kind of keeps coming to mind. We're the ancestors, and that maybe is good for us. And I think that's a pretty Chinese-flavoured notion. I recently watched a little TikTok. I know you're not on TikTok, but there was just a guy in a small town who took us into the ancestor hall in his small town. I think that's the name that he gave it. And it was basically sort of the community centre, a place where they have events, but also this sort of public— and it's amazing how much of this is still there. I think a lot of it got wiped out at critical moments in Chinese history, but at least in this place, there's still this testament to these are our ancestors going back pretty far. And he said, this is basically in this dude's mind, this is Chinese religion. And I was like, geez, relative to other religions, that seems like a pretty decent story to try to get the AIs to live with. I don't think it's a solution, but I don't know. I want to say something in favour of Chinese values, or at least in defence of Chinese values, certainly comparing to—

You are hitting on some pretty fundamental things there that I actually worked on when I was in college, believe it or not. I became really interested in the resonance between ancient Chinese philosophy and conservatism, various strains of conservatism. I think it's pretty clear that the concept of emergent order, The concept of emergence, and like kind of like, you know, that idea which was sort of like pioneered in the West was pioneered like, you know, by like Hayek and like complexity science in like the 20th century and maybe has somewhat earlier predecessors in like the Scottish Enlightenment of the 18th century. That idea is crystallized in a Taoist concept called wu wei. Um, which is thousands of years old in China, right? Like, that's a, you know, um, these things are, um, so there's all kinds of interesting resonance and there are actually like many, many different paths there. Um, I would say, uh, you know, the extent to which China actually like embodies today the values of its ancient intellectual traditions like Confucianism, uh, Taoism, uh, uh, you know, etc., I think, uh, it's questionable. Uh, certainly it's not a very Taoist country. Um, uh, um, you know, there are some, obviously some elements of the Confucian system that have persisted, but, but I think it, it, it's, um, there is more continuity with other intellectual traditions that are somewhat less wholesome. Uh, than what you're describing. The UAE, you know, and, and Saudi, like, uh, I mean, there are definitely, like, very stark differences in terms of the way that their societies are hooked up and the kinds of values that, that they, uh, embody. Um, you know, uh, I think, um, this is one area in which President Trump is is very different from, from previous presidents, and I think in a good way. I think President Trump believes— like, he, he genuinely seeks peace. Uh, he is like one of the most— like, I think of any president, certainly in my lifetime and, and probably in modern history, the most earnestly peace-seeking president, um, we've had in a very long time. And, um, also Uh, sort of the flip side of that coin is he, he really actually, like, for all the stuff about trade and everything, um, he actually cares like quite a bit about global commerce, right? Like, he wants it to happen on, on terms that are a little— that, that are, that are more favorable to the United States, but he actually like cares a lot about it. And he thinks that the connection, you know, between commerce and, um, peace for him is very deep. Um, and so, uh, you know, I think he'd love nothing more than to see a lot more sophisticated commerce happening, uh, you know, between our countries, uh, and within the Middle East at the region. Um, so it's a strategic sector. Um, you know, there are definitely areas in which their values— I think, I think we're gonna have to be able to accommodate a pretty wide range of values. Um, into our, into our systems in the fullness of time. Uh, and I think the stuff, you know, there is fundamental sovereignty there. Um, uh, and I think we have to be respectful of that. Um, and, you know, again, like, we're not anti— we're not opposed to selling, um, AI systems to, you know, advanced capitalist democracies, uh, that are you know, largely secular places, you know, like, like Western Europe. Um, and we totally do. We sell, you know, we sell them lots of stuff and they're, they're big AI users. But it's also the case that, yeah, like I said, you know, um, they're not the most enthusiastic. Um, uh, and very often, you know, they're kind of hostile. And, and I, I wish they would stop being that way, but that's where they are right now, uh, in their society. And that's also their choice.

One thing that has struck me is just generally how much continuity there seems to have been between the sort of Trump-Biden-Trump administrations, at least on the narrow set of issues we're talking about today. Not necessarily, obviously, all of politics, but generally the sort of managing and muddling through strategic competition with China seems to be a clear through line. The expert Export controls seems to be more similar than different. You may see that differently, but from my sort of not super deep interrogation, it seems like there's a lot of continuity there. There's the sort of anti-woke notion, but it seems like a lot of people, and I'd be interested to hear your thoughts on this, have sort of said, yeah, well, they kind of had to put that in there for rhetoric, but it wasn't like— didn't seem like their hearts were really in it. How would you describe the— there's also, of course, the biosecurity angle where there seems to be a lot of continuity between Biden EOs on DNA sequence presynthesis screening. And I think you guys have even extended that. How would you characterise the level of continuity? And what do you think are the most important points of divergence?

Uh, so I mean, I, I think there, uh, there is a lot of stuff that's, that's pretty discontinuous. You know, I would say like the export stuff is one good example of that, where like, yes, it is true that export controls on China are, are a thing that, um, you know, I mean, to a certain extent, like the Biden— I think it's actually more appropriate to say the Biden administration is consistent with the first Trump administration, uh, because the export controls on China, um, including EUV, of EUV Um, you know, that, that was initiated under Trump 45. Um, so, um, I, I would actually characterize the Biden stuff as, as, as an expansion of work that was originally pioneered in, in the first Trump admin. Um, you know, uh, but things like the diffusion rule, I mean, I can't tell you how much damage that did to this country, uh, internationally. People love to talk about the damage that this administration is doing, uh, to our reputation internationally and you know, fairly or unfairly, we can set that aside. Um, one thing I can tell you for sure is when foreign governments come to talk to us about, uh, or came to talk to us about, um, uh, compute, you know, AI, um, diffusion was always at the top of their minds. And they really felt like it was a huge slap in the face, um, you know, to put 2, uh, of the largest— by the way, democracies, uh, um, largest growth markets for generative AI, Brazil and India, into tier 2, into the like, we don't like you, we don't trust you tier, was such an unnecessary self-own. I think like getting rid of that and being much more oriented toward export and, you know, at the margin, less oriented toward control. You know, it's not like we don't, we, we don't want zero control. We have security provisions that we're going to be very serious about throughout the world. But we're somewhat less interested in micromanaging the global diffusion of the hardware. Um, you know, we're significantly less interested in that.

That's a big deal.

The woke stuff, um, you know, I think we are cognizant or were cognizant of, uh, the, you know, we're not, we don't want to meddle in the markets, right? So a consistent theme throughout the administration is like, we, you know, we used Uh, um, tools that we use the federal procurement, you know, that's the hook there, right? We're not trying— we're very explicitly, like, not trying to, um, you know, tell, uh, AI companies what the models that they sell to you in a private market transaction. We're not trying to tell them what they should do. What we're saying is that for our purposes as the federal government, we're not a very big customer of these LLMs just yet, but we think we probably will be in the future. And we are saying that it matters to us what the political values of the systems are, and it matters to us, like, you know, you can go and if you want to go, like, there is a, like, uh, you know, like MAGA handshake Daniel Cocotello, uh, thing you can do here where like one thing that's like not inconceivable to me is that, um, uh, you know, if, if especially if we, if we had the dynamics of Trump 45 where like American elite institutions really like, um, you know, they like to performatively like undermine the administration live and so did the tech companies. They'd like, they liked to it was like fun for them to undermine the President of the United States, American companies, uh, right? That still makes, makes me personally angry and a lot of other people angry. Um, uh, you know, it wouldn't surprise me if this technology was diffusing under those political dynamics. It would not surprise me at all if there were covert efforts to like sabotage the LLMs used by, like, used by the Trump administration. Um, in various ways at the margin. And like, we want to be damn sure that's not happening. Um, so we like are actually like quite serious about that. Getting the guidance right will be difficult. Um, but I think it's like, there's actually like, there are some very talented people who are going to be working on that. Um, and so, you know, we don't want to create a massive regulatory burden with that. That wasn't the objective. That we had in mind. What we want, you know, is transparency, uh, right? Like it's, it's, it's really, you know, the, the, the EO says that a way you can comply with this, and I think probably the easiest way to comply with this, will be transparency around system prompt, model spec, constitution testing. It's gonna be stuff like that. Um, which is actually pretty, another area that's pretty consistent. with AI policy, uh, with, like, a lot of, you know, sort of, like, non-MAGA AI, just general AI policy? Um, I think, like, you know, is there consistency with the Biden administration? Um, in those senses, like, very, very deeply not. The Biden administration was totally committed to the idea of using, um, of using regulation and scary words like misinformation and bias To impact the political outputs of, and to politicize the outputs of LLMs all over the place. They were very committed to that idea. If Kamala had won, that's the big thing they'd be doing. They would be pressing at the limits of the Constitution to do it in a way that we absolutely did not. And it just wouldn't be called pressing at the limits of the Constitution because when Democrats press the limit of the Constitution, it's called ambition. And when Republicans do it, it's called authoritarianism. Um, so, uh, um, you know, that, that's just the rhetorical standards that we have to live with. And, you know, we're gonna get that kind of hate and we're prepared for it. You know, that's like, that's the world, right? But, um, I would say like, uh, uh, those things are stark differences. But you are also right that like the way I describe this, because there were people who said this internally, um, you know, there were people and, and who were like, wow, there's like some stuff that's like similar. And it's like, we're in the early stages of a policy field developing here. And like, you are expressing surprise. Like, it's like early— we're in the early stages of financial services regulation, and you are expressing surprise that the concept of interest rates is similar between the two parties. And it's like, no, I mean, like, this is just important abstractions in this field that we must have. Um, so Like, yes, like, interpretability is important to us, uh, and it was also important, uh, to, uh, uh, you know, to the Biden folks. Um, uh, that's true. Um, I think, uh, you know, we have a very different posture. Um, vibes actually do matter a great deal.

Yeah.

The Trump administration has a very different posture toward the technology in general. But there are some technocratic things that are similar. Same again, like biosecurity. I praised on your podcast, I praised the 2024 nucleic acid synthesis framework. And so it's been wildly cool to have the— I played a pretty substantial role in the rewriting of the nucleic acid synthesis framework that the administration is embarking on. I can't talk too much about Um, where that is because it's not quite out yet, uh, but it should be out pretty soon. And you will see, like, what it is, is there's many commonalities, uh, worse, but like the, the Trump administration will strengthen it substantially, uh, um, in some, some pretty neat ways. There is some consistency also, like, you know, outside of action plan, but like, you know, the like MP material stuff. That our Department of Defense pulled off. Like, the Biden people wish they could pull off deals like that. I think one thing that is really different, that's more inside baseball but it matters, is like institute internal culture and energy. The Biden admin was way more like by-the-books procedural, lots and lots of process governing everything, and that slowed everything down. We are way faster moving. Like I said, we're like Trump administration is creating a new lexicon for American statecraft. And, um, we are way less beholden to, like, the past, um, to the way things have been done in the past. We are, in terms of, like, inside the bureaucracy. This is an inside baseball thing, but in terms of culture and, like, what kind of things would you expect an organization run this way versus the way that the Biden administration ran things? Um, I think you should just expect, like, the Trump admin to move significantly more quickly, um, sometimes in directions that you won't like. Inevitably, that would be the case, but they will move more quickly. Um, they will try more things. Uh, and will it also be true that, like, uh, um, because of that, like, um, you know, some, some people will, will characterize that as like, oh, that's chaos. Um, you know, I don't choose to see it that way, but I'm sure some people will.

Yeah, there's many, many ways to narrativize the same events. That's, um, a lesson I learned over and over again. Um, well, I don't want to keep you on this podcast until you have your kid, uh, and we're entering into Rogan territory. Maybe just a couple more things to bring us home. Um, Taiwan, what do you Just to recap that question, what are they thinking? How are they trying to walk this tightrope?

So, I mean, yeah, I think, uh, you know, the, the Taiwanese, and they are, um, obviously they're making large investments into the United States, um, and I think they see that the geographic diversity is being really important, but they also are like a silicon shield Um, for, uh, for the island. Um, you know, I'm not like the world's best person, uh, uh, to, to talk about the dynamics of that. I just— I'm not like— that's not my area of expertise. Um, what I would say is like, uh, you know, from the perspective of building more semiconductor fabs in the US and like indigenizing that, I think like we're making very strong progress. Um, you know, uh, the way that the Commerce Department in particular has, um, some, some work that they've been doing that I feel like has kind of flown under the radar, but they've like revised a lot of the CHIPS funding deals, uh, that the Biden administration had inked with, um, with different companies and just actually made them like more favorable and bigger and better in a lot of cases. Um, which is good because like a lot of those, um, a lot of those deals, like, um, you can make the case, uh, that like the Biden is— the Biden people, like, I mean, they put in a bunch of crap, you know, they put in a bunch of like, you know, well, you have to have, you know, diverse employees, you know, all this stuff, right?

And like the daycare centers on site.

Yeah, yeah, yeah, yeah. So like Commerce took like a lot of that stuff out, right? Um, But it's also a strategically intelligent time to revise those deals because a lot of those deals were negotiated in the like '22 range, um, when chip companies had way less faith in the AI thesis, uh, '22, '23, you know, in that timeframe. Now they're like way more bullish and everyone's like, oh my God, we're going to be under vision. So they were actually willing to make bigger investments under better terms for us. So. That's been, that's been great. Um, obviously the trade deals, uh, will play, or, you know, have played and will continue to play a really important role in getting more stuff. But like, I think we are actually at a point in the United States where we are going to have multiple really robust, like, enclaves of different kinds of semiconductor development. You know, there will be, um, in Indiana There will be the HBM stuff kind of anchored by, by SK Hynix. Um, there will be the hub in Arizona, of course. Uh, Taylor, Texas, Samsung, you know, that facility looks like it finally has enough demand to come online. There'll be other places too. Um, other R&D hubs, like it's actually like very exciting to see. Um, I think, by the way, the automated material science that we talked about earlier is a strategically important part of this. Um, and there can be automated labs for material science that are commonly shared between industry and academia. And that way you're also like, like you're taking frontier academic research and you're doing it at the same shared facility that a corporation might do some frontier research of their own, totally cross-pollinating ideas, right? Like, I, I think there's a very, very bright future ahead for, for the US. Um, I focus on domestic policy, so I don't unfortunately have all that much to say about the Taiwan stuff.

Do you have a sort of sense of what are we aiming for? I mean, I know we want to have some core amount of production capacity domestically, but it seems like we're not going to achieve chip independence in the sense that we'll be not importing chips from Taiwan anytime soon. Is there some other threshold between some bare minimum that we need to keep the AI lights on and total self-sufficiency? Are there other thresholds between those 2 levels that you think are particularly important?

Well, I would say a couple of things on that front. So first of all, I think we actually are trending by the early 2030s to at least be able to satisfy domestic demand with domestic production. Um, so I, you know, I think I feel decent about that. Um, you know, I think the— I think one part of the goal is that. Another part of the goal is like actually reclaim the lead in frontier semiconductor manufacturing in the United States. Um, uh, that's very important. Um, but another thing that like is underrated and I think, um, probably, uh, is an area I'm gonna focus on because I don't think the current— I, I don't think the Biden administration did enough about this, and I don't think the Trump administration thus far has done enough about this specific thing, which is legacy node. Um, because you can shut down civilization. Like, yes, 2-nanometer, you know, and, and, and 18A and 14, you know, all these things, they're very important. Um, very, very cool and important things. Um, But you can shut down civilization with like 45 nanometers. Like, if you— we don't have 45-nanometer production, like, you can stop civilization if we can't import that stuff. That's a harder one. That's harder than just the leading edge. There are a range of interesting policy tools to use there. Um, uh, I'll have more to say on this soon. Um, this was stuff I was starting to think about post-Action Plan. I was like, this is one of the things It was on my mind. Um, so, um, yeah, more— I'll, I'll probably have more to say about that at some point in the next couple months. But, um, yeah, uh, um, that is a big problem. That's a little bit thornier, and it's harder to make pure economic solutions to that. But on the, on the frontier side, like, I think we're trending in a good direction. You know, there's a lot more work to be done, uh, and I think what we need to do is, you know, we need to make sure that we're funding the basic research too that allows the US to make the next generation leapfrogs. US, we need to make sure that like we don't get— there's not some leapfrog technology that we missed. Uh, and usually our ecosystem is quite good at that, but, um, you know, we need to make sure we continue to nourish that.

Do you have any sort of taxonomy or shortlist of things that you are watching most closely that you think could shake the snow globe? Like, one could be a just leapfrog in, um, you know, chip manufacturing technology?

Uh, yes. Uh, but, um, I, uh, not exactly like it's secret, but, um, some of that information comes from analysis and things that I've seen that I don't think should be public, so I will politely decline to answer.

Okay. How about the frontier companies? I guess I'd be interested in your list of who are the frontier companies. I usually have a pretty short list that I would actually include in that.

These are the live players.

Yeah.

And I guess I'm interested in how do they present to the government? How does the administration or the government more broadly understand these different companies? How differentiated are they in the view of the government? Because it was striking that you hear these things like, oh, Anthropic has really good, deep relationships with the government. I don't know if that may be an out-of-date statement now. You can tell me if you have any thoughts on that. But they're often reportedly the sort of— they've got the longest tenured people and the deepest relationships and the kind of most mature presentation. Okay, I've heard that OpenAI sort of has maybe a reputation for saying what people want to hear at different times. Obviously, Elon was involved in the administration in a serious way for a while, and then they seemingly all got federal contracts. And it happened right at the same time that Grok was calling itself Mecha Hitler. And I was just like, it's very strange, right, that we have this sort of blanket announcement that we're going to do a deal with Claude and Mecha Hitler, and we're going to put them out there all on kind of the same press release. And it just got me wondering, do people sort of see these things as just very similar entities, or is there a more nuanced understanding of who these companies are and what they represent and what doing business with them might mean in the future?

I can't super speak for other people. I would say Like, it's, it's certainly the case that like a company like Anthropic has, has like quite good relationships and a quite good reputation inside specifically the intelligence community. And, and part of that is because like Anthropic invested like serious resources early on because like Dario personally is so like NatSec AI build, like, um, they invested significant resources in, um, You know, getting their models stood up on the high side, which is to say in classified networks. Um, because, you know, it's, it's not easy. Like, it's non-trivial technical work that has to be done to, to get this stuff on the high side. And then of course they've like, they also like, you have to, and in order to do that kind of work, you also have to have employees who have, uh, top secret clearances. You have to have, you know, to the the technical staff to be able to do that. You have to, like, there's all sorts of work that they— it's not like the other companies haven't done this. Obviously, particularly Google has. But certainly Anthropic, among the frontier AI companies, invested early. I personally view the frontier firms, I still think it's basically GDM, Anthropic, and OpenAI. I think that kind of hasn't changed for like 3 years.

Obviously, you're going to go do some other stuff that you kind of alluded to that the administration might not do, and you'll kind of try to be upstream or influence things from the outside. What would you say in general? You could kind of maybe list off some ideas like, what is the government going to fail to do that other people need to pick up? This could be state-level regulation, which I know you have an interest in. It could be things just purely done by the private sector or the philanthropy sector. But how would you advise people who want to make an impact, want to make the future safer, brighter, better, and want to just absorb your wisdom about what is the best way to complement what the government is going to do?

I will reframe it slightly as, like, things that the government will not be well suited to do. Um, and also, like, what's the division of labor, kind of? Like, like, for example, I am not, like, in the camp that we need to restrict states from passing, like, all— I think there's— I think when it comes to the regulation of AI development, it's pretty unambiguous that we can't have 50 rules. Um, but like, I think if a state wants to be totally retrograde and like freeze certain aspects of society, of their society, in amber through really aggressive AI use restrictions, um, I won't support that. You know, I'll argue against it if given the opportunity, but, um, but I, I don't necessarily know that, um, It's— I think that's up to them. I think that's up to the people of that state. Um, when it comes to, uh, like where in general I think governments are gonna fall down, um, governments are gonna fall— like, governments are not gonna do a good job at defining what good looks like when it comes to things like Like AI safety, right? And whatever else. Like, they're just not. There's an information asymmetry that makes it really difficult. Like, I will tell you, like, having worked at the White House, like, I don't know tremendously more about what goes on inside the frontier labs than you do. And there might be ways in which you know more than me, because you probably have relationships with researchers that mine have withered a little bit in recent months. So I think that, like, I look at a company, for example, um, like, uh, the AI underwriting company, uh, this, uh, former, uh, uh, Anthropic employee founded that recently got money from Nat Friedman and Daniel Gross. Um, that's a really cool company that is doing, like, private sector-led— it's a for-profit insurance company that is Well, it's a managing general agent, but that's a technicality, um, that's trying to do, um, you know, they're trying to build AI standards. Like, I think that's like very cool and plausibly like, like both a very good business and, um, uh, a very like positive for the world one. Like, it's one, it's a total win-win. It's like probably a very good business. It's probably, um, Uh, really good, um, uh, for adoption, and it's also probably like good for, for AI safety. Um, so as a general matter, like, I hate to say this because I was just in politics and policy, but like, you know, I, I hope that bright young people, um, you know, mostly like stay out of the world of policy because the world of policy is going to move you closer in the direction of zero-sum Whereas, like, markets are much more about positive-sum outcomes. Um, and so work in the— like, there are many, many companies to be founded. My God, there is a— there is so much money. I, I'm sure there are people doing this and I just don't know about it, but like, the world of biosecurity could use some, like, startups so desperately, right? Like, some people that are thinking about that stuff really desperately. Um, there's lots of stuff like that, and, and I think you'd, you'd create— you, you could build some great businesses in biosecurity, um, and do a lot of good for the world. So, like, see if it's a company. You know, there's, like, principled ways you can think of, you can develop to think about what's likely to be a company and what's likely to require government regulation. I think government— governments will struggle to articulate, like, what good— like, the spec for what good looks like, standards, in the context of specifically catastrophic risk mitigation. That will be hard. And so I think we will probably need lots of folks in civil society and the private sector to help us do that. I mean, as a general matter, we are going to— what we need for the adoption piece, the adoption side of things, which is also— the thing is, adoption in regulated related industries is also about, like, lots of things that implicate, that ultimately implicate matters of AI safety that, like, a lot of your listeners might be, like, primarily interested in. This is sort of the thesis of the AI underwriting company, by the way, is like this kind of an idea. If you can create things that look like gold stars that are like, yeah, like, this is like the, you know, this is also the theme of, like, the work that Fathom is doing, an organisation that I used to be affiliated with. Um, and more broadly, like, um, we're just so pre-paradigmatic on so many different— like, like, I don't hear anyone talking about, like, there are so many ways in which, like, I feel like agentic commerce in particular is, like, right around the corner. Uh, and I hear so little about that from either a technocratic regulatory perspective or a more conceptual perspective about what to expect it to look like. Like, all that stuff. Like, there are so many sector-specific things where it's like, let's really talk about this. Like, let's stop saying, yeah, AI is going to be good in healthcare, and like, let's start saying like what that means, um, more specifically, and like what it requires of us. What kind of institu— what, what specific kinds of institutional adaptations are necessary to make AI work well in healthcare? Um, I think we're in a maturing sector, and that means that our— it's time for our thinking to get like more specific and concrete. Um, and again, that's a big part of what the action plan was about too.

Yeah.

Yeah. The gig age and future is certainly, talk about things where your, uh, crystal ball gets real foggy. It's, um, and precious little exploration. Uh, we, I recently did an episode on the AI Village. Uh, you know, there's, there's a few of these sort of gonzo, uh, experiments that try to shine a little light on it, but it does seem like we're gonna put a huge number of AI agents into the economy, like, in a very short period of time. And we really don't know at all what the dynamics of that are going to be. And that's just kind of wild.

And it links up so well with, um, the, uh, you know, with the stablecoin legislation, uh, that the president signed a couple weeks ago. Um, it's such a good— it's such a— there's such a good linkage there of, like, Yes. And also, like, America is pretty darn good at financial services, so there's pretty good chance that, like, that stuff spreads throughout the world and we end up making, like, we end up creating, like, those could be the really transformative applications that are, like, used all over the world and are really world-changing and, and, and in a positive way. Um, and, uh, um, look a lot like, you know, American dominance in AI. It could be that that's one of the best areas. That's sort of my view of the situation. Um, so I personally— and like, I feel like, by the way, like, this is true of the action plan. Like, the action plan doesn't go enough into that. Action plan doesn't go enough into healthcare. Action plan does not go enough into agriculture. Um, these are all things that— or Veterans Affairs. What about that? Like, that's healthcare. It's a huge single-payer healthcare system. It's one of the biggest single-payer healthcare systems in the developed world. Like, uh, and you know, what are We can totally use that to do interesting things, I would think. Um, so, um, there's, there's so much stuff like that that's worth doing.

Well, thank you for, um, spending all this time. It's been, uh, comprehensive, and I've really appreciated it. Anything you want to, um, touch on that we didn't, or any just, you know, uh, final thoughts, things you want to leave people with?

Uh, no, I think the only thing I would say is that, um, you know, I am, uh, happy to, happy to announce that I will be, uh, uh, resuming Hyperdimensional, my Substack, on a weekly cadence. Uh, I will be joining the Foundation for American Innovation as a senior fellow, and I expect to have other kinds of institutional announcement— institutional affiliations and roles to announce in the next, uh, in the coming weeks. Um, I— there is so much to do, so much to catch up on. Uh, uh, and, um, I sometimes worry that Hyperdimensional will have to go into a twice-a-week cadence, uh, because it's, um, it's, you know, it's that, uh, there's that much— there's— I have a backlog of things that I've been itching to say, um, and so much to come soon.

Cool.

Well, we will certainly be following and, uh, look forward to having you back to discuss, um, many of them. For now, Dean W. Ball, fresh from the White House. Thank you for your service, and thank you for being part of the Cognitive Revolution.

Thanks so much for having me. This was fun.

If you're finding value in the show, we'd appreciate it if you'd take a moment to share it with friends, post online, write a review on Apple Podcasts or Spotify, or just leave us a comment on YouTube. Of course, we always welcome your feedback, guest and topic suggestions, and sponsorship inquiries, either via our website, cognitiverevolution.ai, or by DMing me on your favorite social network.

The Cognitive Revolution is part of the Turpentine Network, a network of podcasts where experts talk technology, business, economics, geopolitics, culture, and more, which is now a part of A16Z.

We're produced by AI Podcasting. If you're looking for podcast production help for everything from the moment you stop recording to the moment your audience starts listening, check them out and see my endorsement at aipodcast.ing. And finally, I encourage you to take a moment to check out our new and improved show notes.

Which were created automatically by Notion's AI Meeting Notes.

AI Meeting Notes captures every detail and breaks down complex concepts so no idea gets lost. And because AI Meeting Notes lives right in Notion, everything you capture, whether that's meetings, podcasts, interviews, or conversations, lives exactly where you plan, build, and get things done. No switching, no slowdown.

Check out Notion's AI Meeting Notes if you want perfect notes that write themselves.

And head to the link in our show notes to try Notion's AI meeting notes free for 30 days.