Transcript
Pascal-Emmanuel Gobry: Hello, everyone, and welcome to the Sphere podcast, which, as usual, is shot from my terrorist hideout in northern Syria. um My guest today is a good friend, an old friend of mine, Sébastien who is ah both French and American citizen.
Sebastien Laye: Yep.
Pascal-Emmanuel Gobry: um and he is a in particular an ai expert he's an economist by training he teaches a ah course on the economics of ai at hec paris which is a very very good school near paris um and he has also been an entrepreneur and an investor in the field of ai so he He's one of the smartest people I know on this stuff, and it doesn't hurt that he's he's a Republican and a Trump supporter.
Pascal-Emmanuel Gobry: ah So I wanted to have him on to discuss the new AI eight action plan that our buddy Kratios put out. i mean, it wasn't just him, but it was a lot of him.
Pascal-Emmanuel Gobry: ah This was a very widely expected document. They did this sort of consulting process where they brought in thousands of contributions. ah David Sachs was also ah very heavily involved.
Pascal-Emmanuel Gobry: you know Lots of expectations because of the big tech, but the tech industry. but the tech right, the MAGA TechWrite and all of that. So we were all looking for this document.
Pascal-Emmanuel Gobry: It dropped. i I read it I thought it was very fascinating, and very interesting. What did you think, Sebastien?
Sebastien Laye: Well, first of all, thanks for having me, Pascal Emmanuel, and kudos to the Trump administration for putting together, like you said, this very comprehensive, ambitious document. in Effectively, if we get back to the genesis of the document, it has been like six months in the making because they organized back in March, April, what they call in DC an RFI, Request for Information.
Sebastien Laye: So I think they got something like a thousand contributions, 10,000 pages. It's all on public website. I myself contributed very modestly to that effort. At the moment, I was trying to launch a small think tank called the Economic Singularity Institute. And my own contribution was about labor economics.
Sebastien Laye: in a post-AGI world. So I was trying to figure out, not really for today, 2025, but 2028, if we achieve a certain level in terms of AI power, which we call AGI, and probably we will get back on the concept of AGI together, what we should do in terms of reskilling, upskilling the workforce. and And that's an important part of the AI action plan.
Sebastien Laye: Obviously, I wasn't the only one to contribute to labor economics, but they really drew some inspiration from that. If we get back a little bit to the genesis of the text too, so three months after that that consultation, we have a document. We're gonna have another document in September because we organized in June, and I was also involved in that RFI, another request for information related to a future and an upcoming event.
Sebastien Laye: um AI R&D strategic plan. So the research and development part is going to be probably dropped, my guess, just my individual guess, by September.
Sebastien Laye: Before we delve into the the actual content of the text and the details and the the different proposals, I'd like to get back a little bit on the text itself, okay?
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: and and the semantics of the text. ah First all, like you mentioned, it's a very readable, enjoyable text.
Pascal-Emmanuel Gobry: OK, interesting.
Pascal-Emmanuel Gobry: Yes, yes.
Sebastien Laye: You read a lot of...
Pascal-Emmanuel Gobry: It's not written in Washingtonese.
Sebastien Laye: you Exactly. Exactly. You read a lot of policy papers. I read some too, but only in economic or technology. And compared to the the latest administration executive orders like the Biden Diffusion Network, et cetera, all these things, it's not, i mean, you don't have the sort of complex, recondite, abstruse jargon that usually you have in this paper.
Pascal-Emmanuel Gobry: yeah Yes, yes.
Sebastien Laye: So for the general audience, even if you're not an AI expert, please go download the text.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: You can read it.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: Maybe we can skip the bullet points that usually refer to the future executive orders and all the acronym, you know, the NIST, Casey, etc. Just focus on the introductory paragraphs.
Sebastien Laye: It's a great text. And also it's a text, but remember, it was associated with a speech by the president, by Donald Trump. And that speech was, in my mind, the most important technology speech ever.
Sebastien Laye: by an American president ever since, remember, the Kennedy speech on on the Moon Conquest and the Space Race.
Pascal-Emmanuel Gobry: Okay, wow.
Sebastien Laye: So that that's really important, you know, probably the most important one in 60 years.
Pascal-Emmanuel Gobry: Okay, so why?
Sebastien Laye: Then, if... Exactly. And and the the the speech and the text itself is called Winning the Race, America's AI Action Plan. So let's ah let's stop a minute on that. Winning the race, winning the AI race.
Sebastien Laye: Obviously, it's not mentioned, but everyone is thinking China, winning winning the race over China. ah But it's also a very complex statement, a bold statement, because if you have a race, what is the finishing line?
Sebastien Laye: And that's another issue that we will discuss together. You know, what is the finishing line? When we had the nuclear race, we know the finishing line was probably the first bomb, right? The space race that was, and don't know, setting a foot on the moon.
Sebastien Laye: ah You know, operational warp speed for COVID, it's getting ah an efficient vaccine.
Pascal-Emmanuel Gobry: Mm-hmm.
Sebastien Laye: In that case, you know winning the AI race can mean so many things. And that's why this text, again, you know it's ah to me, it's really, it's both a recap of everything great that this administration has been doing over the last six months and a roadmap for the next 12 months.
Sebastien Laye: But you will see, you know it's not going to exhaust the subject because probably 15 months from now, we're going to be coming with other AI action plans.
Pascal-Emmanuel Gobry: Okay. Okay. ah that I mean, i i I definitely agree with you. i think I think I wrote something. I'm trying to find my own article because, of course, I'm i'm the most important person to cite here.
Pascal-Emmanuel Gobry: ah
Sebastien Laye: I can cite you. You say that the text was full of vibes.
Pascal-Emmanuel Gobry: Yes. Yes.
Sebastien Laye: And that really inspired me because you gave me the world that I was looking for.
Pascal-Emmanuel Gobry: That's... that
Sebastien Laye: i was I was saying, okay, this is not complex jargon. This is not erudite. Everyone can read. It's energetic. It's refreshing. And then I was like, you got it. It's full of vibes.
Pascal-Emmanuel Gobry: Okay, so well, there there you go. So you did read the you did read the thing I wrote.
Sebastien Laye: Exactly. I read that.
Pascal-Emmanuel Gobry: Yes, the tone of the paper is very, it's very first of all, you're right, it's very easy to read, even if somebody's not a policy professional or an expert or whatever, they can read it
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: um and And it's sort of ambitious. it's it's like It's full of life. It's like, we we're going to do this and we're going to do that. and bra and And it's very refreshing.
Pascal-Emmanuel Gobry: ah yeah and And I think the form matches the content, right? The the form...
Sebastien Laye: exactly
Pascal-Emmanuel Gobry: presses the content well, which is it's written by people who are bona fide experts, many of them from industry. It's not Washington bureaucrats. It's written by people who are really in this to win it.
Pascal-Emmanuel Gobry: um Okay. ah
Sebastien Laye: Maybe we can mention the people. I mean, I can imagine there's been a huge work, not just for the people contributing, you know, from the outside, like all the think tanks or just individuals like me, but but also the usual suspects like the DOE, DOD, DOS, and the OSTP, Kratios, you mentioned, maybe David Sachs, people around Kratios, like a good friend, Dean Ball.
Pascal-Emmanuel Gobry: Yes,
Sebastien Laye: um So a lot of people have been thinking about these issues for years now, right? So...
Pascal-Emmanuel Gobry: yes. Yes,
Sebastien Laye: That was probably the apex of everything that they have done over the last month.
Pascal-Emmanuel Gobry: yes.
Sebastien Laye: Again, the AI action plan, if we can now delve into the content, it's it's a roadmap for the next months and years, but it's also a good recap of everything that has been done by the Trump administration when it comes to ai over the last six months, because they took several executive orders.
Pascal-Emmanuel Gobry: Yes.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: It started the first week, remember,
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: with Stargate and everything, they took ah very quickly, you know, President Trump took a few executive orders just to rescind, to repeal some of the other zealous regulations of the Biden era.
Pascal-Emmanuel Gobry: yeah
Pascal-Emmanuel Gobry: Yes, yes, yeah yes, yes. Yeah, yeah, yeah. Biden had some crazy OK, so what's let's basic, take a ten thousand foot view
Sebastien Laye: Yeah.
Pascal-Emmanuel Gobry: ah so i would say there basic um
Pascal-Emmanuel Gobry: i I would say there's like three legs.
Sebastien Laye: There are three sections.
Pascal-Emmanuel Gobry: ah And you may you may agree, you may disagree, or like three planks. um Number one, it's deregulate. Just let let let things happen.
Pascal-Emmanuel Gobry: Number two, we got we have to build like a lot of data a lot of data centers and a lot of CPUs and a lot of GPUs and a lot of energy and a lot of all of that. ah And number three is national security China.
Pascal-Emmanuel Gobry: And we've got we've got to do the first two while, you know, doing it faster than China and not letting ah China steal our tech and all of that. i is that Is that the broad outline? I'm doing a Trump thing right now.
Sebastien Laye: Yeah, I agree with you that but the three pronged approach that you mentioned is ah a good recap of a document for general audience. You know, if they want to understand the documents structurally, it made up of it's actually made of three sections. Okay. Section one one is innovation, anything related to, like you mentioned deregulation, but also science.
Sebastien Laye: evaluation, etc. We're going to talk about that, open source, etc.
Pascal-Emmanuel Gobry: Yep.
Sebastien Laye: The second section of the document, if you're interested in that, you can skip and go right away to the second section. It's the infrastructure, so data center, chips, electricity, vocational training.
Sebastien Laye: And the third section, which is probably the weakest, is international policies and other United States should lead in terms of the AI stack with its allies. okay So the three aspects that you mentioned, they they transpired you know all along all across the document.
Sebastien Laye: And like we said, the document itself exudes energy, ambition, ah patriotism, MAGA, American dynamism. Remember, that's a concept, American dynamism, that we owe to Anderson Orr-Reeds, the A16Z fund, and in a specific partner, I think two years ago, it's Katherine Boyle.
Sebastien Laye: And ever since, you know it was used by Jay Vance in his speeches. it's It's a new concept that more and more people are using. I couldn't find the expression American dynamism you know in the text because And we get back to that.
Sebastien Laye: That might be one of ah of the critical review of the text. It leans heavily on Silicon Valley. And obviously, if you put the headline American Dynamism, everyone is going to understand that this is a text for Silicon Valley.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: But I think that Silicon Valley, in in all the different you know potential trade-offs, et cetera, really won the day with that text. I think the text, by the way, is more balanced than we could have thought because ah we get back to that. but ah Even v the ideas around AI safety, AI evaluation, which were more associated with ah the Biden administration, we thought they could have been jettisoned entirely by the Trump administration. And that's not really the case.
Sebastien Laye: That's not really the case.
Pascal-Emmanuel Gobry: Yeah. Yes.
Sebastien Laye: Remember, Biden had created something called the AI Safety Institute.
Pascal-Emmanuel Gobry: yes
Sebastien Laye: Okay, so this AI Safety Institute totally aided by the Republicans, by the GOP, DOGE in January, February, fired most people at the AI Safety Institute.
Sebastien Laye: I thought it would be dead. Well, lo and behold, it's only renamed. It's called by all the think tankers, Kaisi. You will find the name in the document, Kaisi Center for Artificial Intelligence Standards and Innovation.
Sebastien Laye: So obviously you got the word innovation that really matters.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: and And if you look at the potential executive orders that will come and all the work that remains to be done on the basis of this document, this agency is going to be so busy. over the next month.
Sebastien Laye: They have so many things to do. So, and that's the point we're goingnna talk about too. You know, they it's probably now that the different departments and agencies are completely understaffed.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: compared to the ambition ascribed to them by this document.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: Because again, the last six months have been great to some aspect. And you have a recap of all the executive orders, but also chaotic sometimes with Dodge, non-Dodge. They fired a lot of people. I'm not sure that the people that they fired are going to be back, whether we want them or not.
Sebastien Laye: ah Some of them, they were woke people. So that's fine. I'm satisfied with that. Some of them were great technologies. So we'll see. I think that because the text is balanced too, let me use this analogy, but it's little bit like Christmas Day.
Sebastien Laye: You know you have a gift, right? You see gift box. You see the online of a box. You know you're going have your Transformer toy. The pictures of a toy are great. you you you You haven't assembled yet the toy, right? You haven't seen the implementation, the execution.
Sebastien Laye: And so everyone is satisfied. It's only when you assemble the toy that you have some disappointment. So some people would be disappointed ah with the execution coming months. But as of now, quite frankly, if you're an abundance guy like me,
Sebastien Laye: if you're an accelerationist, if you're a techno-optimist, you're satisfied by the text, right? It's great. It's 100% what you think. But let's say you're more like from the mega-populist base.
Sebastien Laye: You don't really care about AI, but you don't want to see work AI. ah You see that there are a lot of provisions against DEI and work AI. It talks a lot about workers first, et cetera.
Sebastien Laye: You're going to be satisfied too, right, when you read the document. Because it
Pascal-Emmanuel Gobry: That's right. theres theres there' is a There's a lot of anti-woke stuff.
Sebastien Laye: A lot, a lot. And again, ah not so much brand new stuff, but again, recap of everything that has been done by the administration over the next six months.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: um So let me take you in an example, you know, the v ah question of the AI regulation at the state level, you know, the AI moratorium stuff, ah which created a lot of issues, you know, for the...
Pascal-Emmanuel Gobry: Right. Right. Yeah.
Sebastien Laye: It was an issue supposed to be part of a big, beautiful bill.
Pascal-Emmanuel Gobry: i i I remember it well because because we we so we supported the ah the the the moratorium and and we were disappointed to have ah to see some of our MAGA friends going, oh, this is giveaway to Big Tech.
Sebastien Laye: you You extensively...
Sebastien Laye: AI moratorium. Exactly.
Sebastien Laye: Exactly.
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: Some people that I really like and respect and support and so on and so forth. But I think that you know not letting Gavin Newsom regulate AI is like a really good idea, actually.
Pascal-Emmanuel Gobry: Anyway.
Sebastien Laye: Yeah.
Pascal-Emmanuel Gobry: Yes, so the AI moratorium.
Sebastien Laye: ah but Yeah, so...
Pascal-Emmanuel Gobry: OK, so as explain the explain the state level stuff in the action plan.
Sebastien Laye: so so
Sebastien Laye: Yeah, so so if we get back to the AI moratorium stuff, the idea that over the last two years, we have so many regulations at the state level that it's placing ah an unfair burden you know on companies when they develop AI models.
Sebastien Laye: It's too complicated. It's way easier they have a federal framework.
Pascal-Emmanuel Gobry: There's like 1,000 pending state bills.
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: Like if you count all the bills, it's like 1,000.
Sebastien Laye: Yeah, yeah. and what
Pascal-Emmanuel Gobry: That's insane.
Sebastien Laye: And the pace has been accelerated ever since January, despite despite the new administration. So initially, they wanted to do an AI moratorium bill. They realized that they would face an uphill battle.
Sebastien Laye: So they tried to put everything in a big, beautiful bill.
Pascal-Emmanuel Gobry: Yep, which was smart.
Sebastien Laye: It didn't work. Of course, Trump was not going to sacrifice his budget just for AI.
Pascal-Emmanuel Gobry: but
Sebastien Laye: And now they're trying to reintroduce some of the aspects of the AI moratorium in different EOs or ob bills, and you have one or two elements you know in that text. For instance, it says that if you are state and you have a state-level AI program, you can only get federal funding because sometimes that requires federal funding, federal grant, ah if you if you support models that are not over-zealous in terms of DEI, for instance.
Pascal-Emmanuel Gobry: Ah.
Sebastien Laye: So same thing for federal procurement. you know You're not going to be able to buy up-the-shared systems if they promote a certain vision of the world that is not the objective truth.
Pascal-Emmanuel Gobry: Ah.
Sebastien Laye: That's the words they use. So you see, you're going have some aspect of the AI moratorium in coming months that are going to be back.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: And it's starting with data with that plan, by the way. Okay? But again, I think...
Pascal-Emmanuel Gobry: Very cool. All right. let
Sebastien Laye: Yeah, go ahead, go ahead.
Pascal-Emmanuel Gobry: ah let Let's talk about the regulation piece because I think it's very interesting. And ah I remember you wrote an article about this in The Examiner. um ah So basically, the idea is what they're calling an AI evaluations ecosystem.
Pascal-Emmanuel Gobry: which is basically we're not going to you know the government will not impose regulations right away, but it's going to try to get as many people to sort of evaluate models come up with their own rankings, their own benchmarks, their own evaluations of risks and so on and so forth. I think it's a very smart idea because I do think that AI safety is an important issue, but I also think that you know because it's an emerging technology, ah government regulation is probably it's too premature. like We don't understand enough about this technology to be able to regulate it well, even if we think it has to be regulated.
Pascal-Emmanuel Gobry: ah So I just think it's very smart. I know you know a lot more about this approach than i do. So can can you explain to us ah what it is?
Sebastien Laye: Yeah, definitely. So I think in this fine piece of text, what you can sense is kind of a trade off between the general doctrine of the Trump administration, which is more, and this is the economy speaking, laissez faire.
Sebastien Laye: And so obviously the text confirms the pivot from the overzealous regulation era of Biden administration. okay It's like, come on, you are the innovators, you are ah the entrepreneurs, you should go for it.
Sebastien Laye: And the fact that you know over the last six months, they repealed a lot of the regulations from the Biden era. But in the same time, you know it's not it's not like they completely abandoned the safety framework. So we're trying to reintroduce the safety framework, but within a framework, within an environment that is still respectful of the entrepreneurs and the commercial labs. Because there is this very strong statement, and we talk about that,
Sebastien Laye: later again, that the AI race should be led by the private sector.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: okay And by the way, the Chinese struck back over the weekend with a document where they were saying that it's the public sector and the Communist Party in in their country that was leading and would win the AI race.
Pascal-Emmanuel Gobry: Oh, really?
Sebastien Laye: And we talked about that because because obviously, as a laser-fair economist,
Pascal-Emmanuel Gobry: That's fascinating. That's fascinating.
Sebastien Laye: um I'm more supportive of the American approach, even though at some point we're going have to be careful because, for instance, for AGI or military application, if a race is only between...
Pascal-Emmanuel Gobry: Yeah, I mean, regulation is a real issue. like you can't be I'm pro-tech, I'm pro-acceleration, I'm pro-everything, but like the idea that there's no serious regulatory issue, I don't believe i don't think any reasonable person believes that, which is which is why it's an important and interesting question, because it's not it's not black and white.
Sebastien Laye: I am.
Sebastien Laye: Exactly. So...
Sebastien Laye: yeah Yeah, and in thatpa in that text in that text, in this more balanced approach, let's say that over the last six last six months, there was Trump, Lightning saying, hey, we're all for the entrepreneurs, we're all for deregulation, etc.
Pascal-Emmanuel Gobry: So, go on.
Sebastien Laye: You got this in the text, but you can also sense the influence, I guess, of other people. don't know it's Cratchos, probably Dean Ball, people who spend a lot of time on AI governance issues.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: where you have this evaluation framework where whenever you have a new model, you need to be able to evaluate the model. You have evaluation of foreign models too, you know from adversaries. That's very important.
Sebastien Laye: You have what we call interpretability. In other words, i.e. being able to understand why is it that an LLM is outputting you know certain results?
Sebastien Laye: Because as of now, we don't know. Even the people at OpenAI and Cloud, they're telling you, we can't really explain ah everything that is done by ChatGPT or Cloud. And of course, for national security applications, you need interpretability.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: You have open source. You have an important provision about open source in the document. I think a lot of people initially you know on the Republican side were against open source with the idea that if you sell an open source model, anyone can steal your secrets, right?
Sebastien Laye: ah But what this administration wants to promote is the fact that, and the Chinese have been very good at that over the last year, if you export an open source model, you know it's kind of a Trojan horse that allows you to export your technology to and to embed your technology and your AI stack everywhere across the world.
Sebastien Laye: And that's that's the real success of the Chinese with DeepSeek, you know, as opposed to proprietary models.
Pascal-Emmanuel Gobry: Yeah. Interesting.
Sebastien Laye: So even OpenAI... OpenAI just released ChatGPT agent, which I'm having a lot of fun with. In a couple of days, GPT-5, in between, you're going to have ah an open-weight, open-source ChatGPT model. okay And they're probably doing this because the new administration is pro-open source.
Sebastien Laye: So that's really important because all these elements of AI governance, quite frankly, six months ago, we did not expect that from a Trump administration.
Pascal-Emmanuel Gobry: interesting
Sebastien Laye: But it's a good thing. you know It's going towards the right direction. Like you said, you and I were more accelerationist. I'm definitely an abundance guy. But in the same time, even though I'm pro-business, I'm a laissez-faire economist, I think that AI governance should be safe.
Sebastien Laye: And then I'm going to go further.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: And in that case, I might be a little bit more critical of the text. Like I say, at some point of innovation, if we want to reach this AGI level, which is ah an artificial general intelligence,
Sebastien Laye: In other words, artificial intelligence that can perform and mimic 100% of human level cognitive tasks. So cognitive tasks, think about anything you can do in front of your desktop.
Sebastien Laye: If we want to reach that level, I don't think that the race, if there is a real race to reach AGI, should be only between the commercial labs in Silicon Valley and the Chinese Communist Party.
Pascal-Emmanuel Gobry: Yeah.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: Because my fear is that in that case, all along the private sector, we might not win. I'm not 100% sure that we're going to win. it's It's really at this level that i would see i would like to see the federal state, it's already way committed with this document, but go an extra step.
Sebastien Laye: like you know in Even in terms of ah of federal R&D funding, the couple of moonshot projects
Pascal-Emmanuel Gobry: Yeah, OK.
Sebastien Laye: at DARPA, etc., that should be undertaken by the state. So yes, it might be that even at the time of the of the nuclear race, you had a lot of subcontractors in private sector. It was probably 70% the public sector, 30% the private sector.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: okay Today, the AGI race in the United States should be 80%, 90% of the private sector, but still that 10% or 20% that the state can do is really important. And I think that over the next two to three years, it's not just about deregulation, etc.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: you need You need to show the money. You need you need to show convictions too. are a couple of things that the guys in Silicon Valley, they're not going to do that the national security community, the Pentagon, they can do when it comes to ai
Pascal-Emmanuel Gobry: Like what? Like give me a specific example.
Sebastien Laye: That's my message. Like you have a good i idea in the plan. It's ah it's in the second or third section. In the first section, innovation, you have the idea that for the Pentagon, they should develop their own AI and autonomous ah virtual ground facility.
Sebastien Laye: So it's a physical facility for for to to run virtual environment with AI.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: So anything that is related to that sort of simulated environment that would help for robotics, military application, et cetera, obviously some people in the private sector are going to try to do that.
Pascal-Emmanuel Gobry: Okay.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: NVIDIA is trying to do that. But I think that the federal state has a lot of data and ah and and a lot of ah firepower to to do it. So if I if i were to to, you know, this concept of AGI, um I talk to a lot of scientists in the US and in Europe, as of now, I would say the equation to reach that level of AGI, let's say in 2028, maybe still with Trump, okay, that that's the earliest possible 2028, could be a little bit later, we'll see.
Pascal-Emmanuel Gobry: okay
Sebastien Laye: But I would say in the equation, it's 50% what we call computing. So all the infrastructure, everything that you have in the document, the data centers, with cheap ah you know being the the United States being the most powerful for data centers and infrastructure and chips, semiconductors, et cetera. That's 50% of the equation.
Sebastien Laye: And sometimes when I listen to the administration, I get a feeling that it's 100% for them.
Pascal-Emmanuel Gobry: Yep.
Sebastien Laye: And I think it's, I'm not 100% aligned yet. It's 50% this, a good 30%, it's still the algorithms. In other words, just believing that because because we're going to have the largest data center,
Sebastien Laye: ah it's not going to prevent us from getting you know a brilliant Chinese guy or a guy in Singapore or or in Sweden who's a genius and can come up with a new algorithm.
Sebastien Laye: That can still happen, you know and then that will definitely upscale anything that we can do with our compute power. And then another 20% is what I said, you know world representation, simulating environments, etc. Well, of course,
Sebastien Laye: The DOD, the DOE, they have tons of data. So that's my own vision. Again, as a conservative, I support the plan.
Pascal-Emmanuel Gobry: Interesting.
Sebastien Laye: We need compute. I think that compute, anything that you hear about data centers, et cetera, if you want to have a better vision, I think it's more important for diffusion than innovation.
Sebastien Laye: ah You have an interesting guy in DC called Jeffrey Ding. i don't know if you're familiar familiar with Jeffrey Ding. He wrote a book called ah Technology and the Rise of Great Powers. and
Pascal-Emmanuel Gobry: Okay, oh, I've heard of the book, definitely.
Sebastien Laye: Yeah, it's a great book that just explained what the economy...
Pascal-Emmanuel Gobry: I think I have it actually.
Sebastien Laye: You probably have it, I'm sure, Jeffrey Dean, you would you would check. ah It's probably associated with a think tank or or university. i don't I don't remember. I've never been directly in touch with him. But he explained something that The Economist has done for a while, that for any industrial revolution, innovation matters, but diffusion is even more important.
Sebastien Laye: So the infrastructure, the data centers... v v Of course, they help you initially to train the model, but they also help you to run the models ah in your economy.
Sebastien Laye: That's why the v ah Saudi Arabia, the Emirates, they want to have large data centers.
Pascal-Emmanuel Gobry: a Yeah.
Sebastien Laye: Otherwise, day to day, you cannot power your new applications. But it doesn't guarantee you in terms of innovation necessarily that you're going to have a breakthrough. you know You know what I mean?
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: it's it's It's essential, but it's not quintessential.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: It's not 100%. If you're not good at fundamental science algorithms, so that would be my message.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: I like the plan. It's an industrial plan. Federal state is doing what the federal state is supposed to do. But still, and we will see in September with the R&D plan, we have to do more when it comes to federal technology.
Sebastien Laye: R&D funding, moonshot for projects, etc. And the Chinese, they can do that because it's a command and control state, you know, the way it works. So they can try a lot of things, you know.
Pascal-Emmanuel Gobry: Yes.
Pascal-Emmanuel Gobry: ah Okay. um Another thing that, I mean, it didn't in retrospect, it's not surprising because it's very MAGA and it's very sort of um economically, ah politically clever because one of the fears ah that people have related to AI is related to jobs. And so there's all this stuff in the report or in the document, ah action plan.
Pascal-Emmanuel Gobry: I forgot the word, action plan.
Sebastien Laye: Bye.
Pascal-Emmanuel Gobry: um ah there's all this stuff about worker-first, worker-first AI. you know ah They're going to create like an AI workforce research hub, their worker-first AI agenda, and all of that.
Pascal-Emmanuel Gobry: so can can you can Can you explain, first of all, like what do they mean by that? And second, um you know ah is is this Is this really real or is just or or is this just PR to to tell you know ah Trump voters in Ohio and Pennsylvania, don't worry, you're not going to lose your job? but
Sebastien Laye: Oh no, it's totally real and it's probably the most important part of the document. So it's actually section two. Section two is entitled ah AI infrastructure. It starts with data center, chips, semiconductor manufacturing. It's a recap of everything that has been done, you know.
Sebastien Laye: at the facilitating using permitting you know for data centers, giving federal lands to build data centers. Then got a section, by the way, that no one really commented that is really important, which is about the electrical grid.
Sebastien Laye: Because remember, it's I think it's ah it's Google's former CEO, Eric Schmidt, couple of weeks ago he said, It was asked, what are the limitations, what are vi v via the any potential stalemates when it comes to AI progress? And they said, only limitations, you know the real limitation is not chips, it's not data center, it's not research or algorithm, it's electricity.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: Electricity.
Pascal-Emmanuel Gobry: Yes. yes
Sebastien Laye: And so you have a whole section on the electrical grid. that advocates stabilizing the existing grid. So stop decommissioning part of the grids because the energy power source is not like environmental friendly or that sort of thing. There's a new ambition for the electrical grid. And then there is a section on the skilled workforce.
Sebastien Laye: And it's really interesting because instead of focusing you know on the the the potential disappearance of threats to white-collar jobs, etc.
Sebastien Laye: It concentrates on the first layer and it says, because we're going to need data center, a manufacturing run in sense to have this compute power, we're going to give a priority to plumbers, HVAC specialists, electricity specialists, because it's coming right after the paragraph on on the electrical grid.
Pascal-Emmanuel Gobry: he
Pascal-Emmanuel Gobry: right great Right, right, right, right, right.
Sebastien Laye: They've had energy efficiency. So it's a very smart strategy that we haven't seen in other in other countries. you know Actually, in Europe, every time that we talk about reskilling, abskilling, they talk about executives you know the the people in knowledge management, the knowledge workers, never about workers first.
Sebastien Laye: Here, but completely ah you know they reverse the the approach. by showing that this AI agenda could be workers first because you could take like your average plumbers and it's going to give him, you know, new opportunities.
Sebastien Laye: That's one thing. And one thing, again, that I did not expect that is very smart and very important. That's a strong, that's one of the strengths, you know, really of a document. And then there's another part which appealed to me as an economist, which is that there is a proposal for like the Bureau of Labor Statistics, the Census Bureau to at long last gather data.
Sebastien Laye: life, on the economy, the job, because we need to know, you know, what kind of task can we replace, who needs to be revealed, up here.
Pascal-Emmanuel Gobry: but OK, so what kind of data? Because BLS already captures a lot of data. Oh,
Sebastien Laye: We capture a lot of data that we economists, we don't have. I've already worked with databases. um
Pascal-Emmanuel Gobry: OK.
Sebastien Laye: don't have the same access.
Pascal-Emmanuel Gobry: So they capture the data, but they don't publish it.
Sebastien Laye: Exactly. Exactly. And it doesn't communicate. And even more so, they went the extra step.
Pascal-Emmanuel Gobry: why why Why would they? I mean, you know. Okay.
Sebastien Laye: And they went, but they went also, yeah,
Pascal-Emmanuel Gobry: I did not know that. That's that's fascinating. ah Yeah.
Sebastien Laye: And they also went...
Pascal-Emmanuel Gobry: Who pays for it? It's the public.
Sebastien Laye: But they also went extra step.
Pascal-Emmanuel Gobry: So yes, help publish the data.
Sebastien Laye: They also went the extra step by advocating the creation, like you mentioned, of the workforce AI research hub, which is going to be like the way that I understand it, you know, I might might be wrong, but the way that I understand it, that that would be a bunch of economists, you know, working with this data and making, you know, monthly recommendations regarding the workforce saying, hey, with this new model,
Sebastien Laye: ah It looks like, i don't know, i I'm just betting up, but, oh, this new model is capable of doing painting. So going to have an issue with all painters. What are we doing? How can we retrain them?
Sebastien Laye: one of the ah we should we Should they use the AI system and be trained in using the AI system? Or should we do something else? It's going to be a way to to make it smoother in terms of transition and industry transitions.
Pascal-Emmanuel Gobry: Okay. Interesting. Interesting.
Sebastien Laye: Okay.
Pascal-Emmanuel Gobry: Yeah, no, very cool.
Sebastien Laye: So that's, ah that's what section two and section two again is, ah I think section one is longer, but section two is very important, but you got data centers, you got electrical grid, and then you got the workforce and you got it, you know, you covered, you know, everything in that second section.
Pascal-Emmanuel Gobry: Yeah. OK, so one of the things you said to me before ah before we started um is that you have some criticisms ah ah of the plan, um which, of course, we believe in free speech.
Pascal-Emmanuel Gobry: um
Sebastien Laye: Of course. And it's just a plan. I mean, there's no way you can put everything into the plan.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: Obviously, their policies are broader than anything that is in the plan.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: And the plan is more managed by the OSTP.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: There are other programs at the Department of State, Department of Energy. We haven't talked about the Department of Energy, but actually Chris Wright has been doing a great job. Same for the DOC, Department of Commerce, over over the last six months. And and and the first section of the document is essentially a recap of everything that they've been doing over the last month.
Pascal-Emmanuel Gobry: um Yeah, so one criticism is not enough federal funding for R&D, Moonshot R&D.
Sebastien Laye: But yeah, qui them I have some.
Pascal-Emmanuel Gobry: Okay. What else?
Sebastien Laye: I got three others. Okay, let's try to be very structured. ah So the second one is immigration. ah
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: The document is now talking about immigration. And we talked very briefly you know in your introduction of the cleavage between sometimes the techno-optimists or the techno-bros in the administration and the MAGA populist base.
Sebastien Laye: You already had a free free battle two battles, I would say. The one on H-1B visa in December that was already about immigration. The second one was about the AI moratorium. And ah I can make the prediction that there's going to be a third one over a summer.
Sebastien Laye: And we touch down on that a little bit later. And it's going to be about the AI chip export controls and China. But to to get back to to the one on December, remember, that was already about the H-1B visa.
Pascal-Emmanuel Gobry: OK.
Sebastien Laye: So that's going to be an issue.
Pascal-Emmanuel Gobry: I remember it well.
Sebastien Laye: Of course, of course. and a And we...
Pascal-Emmanuel Gobry: we We were all glued to our phones at Christmas instead of talking to our families as as God intended.
Sebastien Laye: Exactly. Yeah, exactly. on x We're following that on X and and the and the beef between the two camps. um And that was even before before Inauguration Day. Right, it was back in December yeah for Christmas.
Sebastien Laye: So here's the thing, a couple of stats.
Sebastien Laye: Half of PhD-level people working in AI in the United States right now were not born in the United States. Second stat, 75% of the new meta superintelligence team, you know all of the people that Mark Zuckerberg has been botching from other commercial labs for like $100 million, 75% of these people are first generation immigrants to the United States.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: So obviously, When I read the document, I was like, it's a very ambitious document. We're going to see the implementation, the future executive order, the execution. It all boils down to the execution.
Sebastien Laye: But do we have the human resources to do it? you know Of course, we need the plumber, the HVAC specialists, etc. But if we want to accelerate, we also need the AI specialists.
Sebastien Laye: So unless there is a ah real change in AI research, and it might come in two years, maybe that two to three years from now, maybe at some point, AI research will be automated and it will be a self-recursive AI doing AI research. So maybe three years from now, we're going need less human AI research. But at least as of now, we need talent and it's a war for talent too. So I'm a little bit um um a little bit skeptical that the AI I couldn't find, you know, like ah like an immigration section in the in ah in the document.
Pascal-Emmanuel Gobry: Okay. ah No, no, no, no, no.
Sebastien Laye: That's my first criticism. I have two of them.
Pascal-Emmanuel Gobry: I'm going to give you some pushback on that.
Sebastien Laye: Yeah. Yeah.
Pascal-Emmanuel Gobry: ah Let's have some discussion on that. um
Sebastien Laye: Sure.
Pascal-Emmanuel Gobry: Okay. um So ah the high-skill immigration thing. So H1B is a separate issue because you have these companies that use it to outsource, blah, blah, blah.
Sebastien Laye: It's a total mess. It has to be revamped.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: Okay.
Pascal-Emmanuel Gobry: And...
Sebastien Laye: I'm completely against the H1B visa regarding the delivery.
Pascal-Emmanuel Gobry: Yeah, and and we did some reporting on that. They're working on very interesting reforms, but there's some legal stuff that they're working on.
Sebastien Laye: Yeah.
Pascal-Emmanuel Gobry: um The other thing is about the 75% of PhD, blah, blah, blah, blah. ah A lot of them are from China, ah
Sebastien Laye: Which is an issue.
Pascal-Emmanuel Gobry: and and a lot of and those Chinese people are spies. um and And I'm not saying that because i you know I have some prejudice. I'm saying that because it's the it's the law in China.
Pascal-Emmanuel Gobry: It's the law that if you're a Chinese student ah sent abroad, you have to report on everything that you do to to the embassy.
Pascal-Emmanuel Gobry: ah And if you don't do it, they will go after your family. So I'm not, you know, I'm not, you know, making a moral or whatever. It's just a fact.
Sebastien Laye: No, me neither, but you could have...
Pascal-Emmanuel Gobry: So, you know, training training training the those foreign students so that they can export. And by the way, people who did analysis on the deep seek stuff found out that there was a lot of stolen tech in it.
Sebastien Laye: Of course, a lot of things that were stolen by Chinese individuals in the U.S.
Pascal-Emmanuel Gobry: ah So first of all, you get rid of all of those.
Sebastien Laye: No, but you could have an immigration...
Pascal-Emmanuel Gobry: Secondly, hang on, I'm just finishing this point and you can and you can respond.
Sebastien Laye: Yeah, okay.
Pascal-Emmanuel Gobry: um if you If you look at the proportion of foreign PhD students, so in the US, as as as I'm sure you know, as you probably know, foreign students pay 100%. of the tuition, which, as you know, is these crazy amounts, $50,000, $80,000 a year.
Pascal-Emmanuel Gobry: and and And American students get, ah especially at the graduate level, ah if you're in a top PhD program in the US, you don't pay anything um because you get scholarships and grants and so on and so forth.
Pascal-Emmanuel Gobry: And so the universities, they discriminate against American citizens and they get foreign citizens. um So, you know, the idea that, ah you know, oh, 75% of PhD students are foreign and this is, you know, this is because 75% the talent is non-American.
Pascal-Emmanuel Gobry: you know If you take out the spies and if you take out the people who would not have been admitted if they hadn't been willing or able to pay $80,000 a year, I think you take out like a lot of those.
Pascal-Emmanuel Gobry: And then for the for the rest, you can give them visas, right? yeah I'm not against... But, you know, that's that's my pushback. That's why MAGA populist would say if he was sitting in in this chair, of course, I'm a moderate centrist. i would never I would never dabble in populism. But, you know, that's the counter argument.
Sebastien Laye: I'm 100% with you, and I am a conservative and a mega populist too. And when it comes to overall immigration, you know my ideas, obviously, and that's why I want to see a revamp of the H-1B visa. I'm not advocating for opening the doors massively to other people. I'm just saying we are, when it comes to top AI tenant, in a situation similar to 1942, in order to get bomb,
Sebastien Laye: where
Pascal-Emmanuel Gobry: Right. Right. Right.
Sebastien Laye: you know order to get the bomb
Pascal-Emmanuel Gobry: right
Sebastien Laye: we were trying to patch the best talents from Europe. We have to do this at the moment, and unfortunately, because we have only two to three years, three, four years max before AGI.
Pascal-Emmanuel Gobry: right
Sebastien Laye: And I think that push, like letting aside Chinese, I think that we should have people on the ground in Singapore, Japan, the UK, France, Brazil, and we should try to ah identify the 1000 brands that we need for this program and get them to the US. that That's the only thing that I'm saying, you know.
Sebastien Laye: that's That's my own point.
Pascal-Emmanuel Gobry: Okay. I mean, I agree with that 100%.
Sebastien Laye: The rest, the rest, exactly, the rest, operation people, the plumbers, the financiers, the engineers, we have them in the United States, you know?
Pascal-Emmanuel Gobry: Operation paperclip should not be a one-time thing. It should be
Sebastien Laye: and And again, that was my point during the H-1B visa. The ID, and I was put forth by a few people, I think by Vivek Ramaswani at some point, that the Americans suck at STEM and science.
Sebastien Laye: That's completely wrong. That's not my vision of the United States. It's a country of innovators. I grew up, I wasn't an American citizen. I was born in France at the time in the 80s. I had this vision of a Goonies, you know, where you had brilliant American kids. And I met very young, brilliant American people all the time, you know, in the startup universe. So we have resources.
Sebastien Laye: It's just that we have a very short timeline. We don't have 10 years. And it's just that let's be opportunistic.
Pascal-Emmanuel Gobry: Got it, got it, got it, got it. and That's totally fair.
Sebastien Laye: You know what I mean?
Pascal-Emmanuel Gobry: I mean, this is this is open debate.
Sebastien Laye: Yeah. Come on, let's let's let's sell the American dream.
Pascal-Emmanuel Gobry: you know i'm
Sebastien Laye: And if you have a good guy, you know, in Paris or Germany, we should get him. That's it.
Pascal-Emmanuel Gobry: Fair enough. ah Okay, so number one criticism is immigration. um
Sebastien Laye: number two Number two, the issue of federal personnel.
Pascal-Emmanuel Gobry: You're on the record. Okay.
Sebastien Laye: Again, we had the Doge thing.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: That's great. We need to reduce public spending. But when I read the plan, and I'm going to take an example. So the AI Safety Institute, which became KAISI, it's
Pascal-Emmanuel Gobry: Yes, yes. I remember i remember you told me, you and I used to talk a lot about this AI Safety Institute.
Sebastien Laye: act it so it's part of ah of a something in the Department of Commerce called NIST, National Institute of Standards and Technology.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: When Dodge to control late January, they fired a lot of people at NIST. They were work people, but also a lot of technologists.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: Now they're understaffed. A lot of the things that they have to do at the Department of Commerce for export control, evaluation, risk assessment, they're completely understaffed.
Pascal-Emmanuel Gobry: Right, right, right, right.
Sebastien Laye: And now the people, you know, they find out about this document and they're like, wow, we lost a third of our staff and we have to do so many things.
Pascal-Emmanuel Gobry: yay Yeah, yeah, So, so, so... So to summarize, the the Biden administration created this thing called the AI Safety Institute. And in it, they put a bunch of crazy blue haired communists whose job was to make AI communist.
Sebastien Laye: Of work, crazy work. Exactly, good
Pascal-Emmanuel Gobry: But they also hired some like really good scientists and researchers.
Sebastien Laye: scientists. They understand.
Pascal-Emmanuel Gobry: and and And your your belief that um which i'm i assume I have no reason to doubt, right?
Sebastien Laye: ah feel dad they just
Pascal-Emmanuel Gobry: i I know you well. I know you share the the goals ah of Doge, and I know you're very smart, but basically your point is like Doge could cut too far. They got rid of the woke people, but they also got rid of very good people.
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: Okay, so...
Sebastien Laye: And it's going to take them a ah few weeks, you know, a few months to rehire people, find the right people, etc. And in the mix, we lost a couple of crucial months vis-a-vis, you know, this AI race with China.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: That's my concern.
Pascal-Emmanuel Gobry: Okay. Okay. So is is it is it just the AI Safety Institute or is there a broader problem?
Sebastien Laye: No, it's it's it's it's a broad thing. If you take the Department of Commerce, you have NIST and the Eye Safety Institute, so KAISI now. ah But you know for export controls, like making sure you know that some of the chips are not ended up in into adversary country.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: For PIS, it's another department, it's severely understaffed. you know The transition took a while, the confirmation of people, et cetera. Some of these departments have been up and running for like Three months, nothing more.
Sebastien Laye: They lost a lot of people. So it's it's a real issue, you know. So again, the document, remember, is a trade-off. I can see the influence of the respective people. It's a trade-off between the the ah pure laissez-faire deregulation.
Pascal-Emmanuel Gobry: it's It's just like papal documents. Like you can see which bishop, but ah yeah.
Sebastien Laye: Exactly. You can see exactly that they went on Google Drive and they assembled, they did a collage of different things, which is great. I mean, the final outcome is like, wow. I was like, wow.
Pascal-Emmanuel Gobry: No, I mean, as we were saying earlier, like overall, it's an amazing document, but...
Sebastien Laye: I was stunned. It's so good. So good. It's almost ah it's almost a book. you know ah It's almost a book. where Maybe we should write a book, right?
Pascal-Emmanuel Gobry: I mean, mate they should they should turn it into a book.
Sebastien Laye: Winning the Diaries. Yeah.
Pascal-Emmanuel Gobry: I mean, they turned like, what's his name? Like the Mueller report into a book. Do you remember that?
Sebastien Laye: Exactly. Exactly. Yeah.
Pascal-Emmanuel Gobry: So if the Mueller report can be a book, then this can definitely be a book.
Sebastien Laye: Yeah. yeah This one could be a move. So they're they are definitely understaffed and and it's going to be issue. And again, it's due to this trade-off between the lazy fare, you know, pro technology.
Pascal-Emmanuel Gobry: No, okay, fair enough.
Sebastien Laye: And the AI governance guys, what we are of course, were like, okay, we should be ambitious. We should be careful, you know, with risk assessment, et cetera.
Pascal-Emmanuel Gobry: I mean, you ah you and I are both sort of quote unquote French MAGA, which is, you know, we we believe in free markets, but we also and and understand the importance of like government competence.
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: So, okay.
Sebastien Laye: Exactly.
Pascal-Emmanuel Gobry: So can you, can you tell it? Okay. So BIS, blah, blah, blah. So what specifically are the roles that are needed? ah why Why are, why are they needed?
Pascal-Emmanuel Gobry: Why are they needed now?
Sebastien Laye: Yeah, let me take an example.
Pascal-Emmanuel Gobry: Like what, what's missing specifically?
Sebastien Laye: with Yeah, within NIST, which is a department
Pascal-Emmanuel Gobry: So AI safety does research, right, basically? OK. Yeah.
Sebastien Laye: of research and evaluation too. okay
Pascal-Emmanuel Gobry: OK.
Sebastien Laye: And the NIST overall, you know not just KAISI, the new AI safety institute, but other departments, they are also monitoring risk assessment.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: In other words, the risk posed by new model.
Pascal-Emmanuel Gobry: okay
Sebastien Laye: ah There is a huge document that was produced by the Biden administration called the Risk Assessment Framework. it's It's revamped right now by the the the Trump administration because we want to get rid of the DEI stuff.
Sebastien Laye: But still, you know you need people to evaluate the new models. In this document, you know they assigned to this same agency ah this new capacity to evaluate foreign models.
Sebastien Laye: So that means that every month you're in this department, you're supposed to look at what Chinese are doing,
Pascal-Emmanuel Gobry: Right. Right. So by definition, you're going to need more people if you're...
Sebastien Laye: So you're going need more people, you know, I mean, unless you can, maybe at some point it's going to be automated. That's the thing. yeah And this is something that I wrote about. it That's a the stunning thing about AI governance.
Sebastien Laye: AI governance is governance ah the of a technology called AI, but AI is also ah governance tool, right?
Pascal-Emmanuel Gobry: Yeah.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: In other words, at some point, you're going to be able to use AI to automate governance. So we're going to be there in in a few months.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: So yeah, but we're not there yet. At the moment, it's just ah it's just a people thing. you know we need We need more human resources.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: That's it. And I have another criticism.
Pascal-Emmanuel Gobry: OK.
Sebastien Laye: I have third one or fourth one.
Pascal-Emmanuel Gobry: OK. Yeah.
Sebastien Laye: Should I go for it? Yeah.
Pascal-Emmanuel Gobry: Yeah. Yeah.
Sebastien Laye: So you have a very important, the last section of the document is about international policies.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: It reminds everyone that the United States is the leader of a free world and that the US AI stack should prevail. That's what we've seen exemplified by the Trump visit to the Middle East, which was a demonstration of force and the capacity for the United States to lead with the US AI stack, i.e. the models, um but also the chips, the data centers, and the alliance with the Middle East ah money.
Sebastien Laye: um But what you have in the document is the recommendation that ah they should be like export package. In other words, the United States should be better organized at exporting altogether, not just the models, but also the chips, the infrastructure, etc. So there is a proposal that within the Department of Commerce, there should be a monthly gathering of people from the private sector proposing the Department of Commerce to push for certain export package.
Sebastien Laye: Okay? um That's great. to have two issues. Again, this personal issue, but I don't want to talk about this anymore. But more importantly, you know if you look at the ambition, this ambition to lead the world, et cetera, to push export package, I believe it's the right thing to do.
Sebastien Laye: Because myself, I've been writing and I would continue to advocate. My idea as a MAGA guy was to say, and kind of a like you mentioned for the two of us international MAGA guys was to say, okay, NATO is kind of ah kind of dead and I would like to either complement or replace NATO by a grand technological and economic alliance because I believe that the United States can lead the free world and the allies, etc.
Pascal-Emmanuel Gobry: Interesting.
Sebastien Laye: through technological standards and economic measures and that would be a block ah around AI, free trade, etc. against the very small Chinese authoritarian bloc. So I'm all in for that ambition.
Sebastien Laye: The only issue is that you understand that as of now, and at least for the next month, this ambition of export package, leading the AI thing, etc., is completely at odds with the tensions between the United States and some of its allies and the difficult trade wars, which I believe is one of the reasons that Trump, and he's been victorious, he's been
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: victor victorious in these trade wars, especially with the EU, but he probably wants to put this behind him you know in September.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: He's trying to to finalize a deal with the Chinese. But as long as you have these trade tensions and these difficult negotiations, it's very hard in the same breath to say, hey, guys, let's do something together.
Sebastien Laye: I have my my technology export package. Let's push it together. Let's create a great alliance, needs our own technology, etc.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: So it's it's just that the document, that's another point, the document is not really talking about geopolitics or even national security. you got a couple of things about the DoD, but it doesn't say anything about ah AI and military integration or geopolitical alliances.
Pascal-Emmanuel Gobry: Right. Okay.
Sebastien Laye: So it's a minor thing.
Pascal-Emmanuel Gobry: Fair enough.
Sebastien Laye: I'm just um just, you know, in pointing the weaknesses, what I want to do also is to say, hey guys, this is what we still have to do. You know what I mean?
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: You've a great roadmap, a couple of things to so think about in common.
Pascal-Emmanuel Gobry: yeah, yeah. And it can't have anything. And I was, ah it can't have anything for sure, but it can't have everything either. um And i was I was just so wowed by the sort of overall vibe of it um and ambition that that um I didn't look at it very critically. and and But you're right.
Pascal-Emmanuel Gobry: um OK, well, ah we're nearing the end. um First of all, is there any last open open floor question you want to say about the the plan, anything we didn't mention that you think is really important?
Sebastien Laye: I think there is something that is... There's only hinted at, you know, in the document that it's very important, but people have been talking about it, you know, around this document in previous week, which is the question of AI chip export control.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: And for the general audience, maybe a recap here. ah Remember, you know, the Biden administration had been trying to ban the exp export of chips to China and some of our adversaries.
Pascal-Emmanuel Gobry: Yes.
Sebastien Laye: But unfortunately, in doing so, during the last days of the Biden administration, they enacted something called the Biden Diffusion Network, which was a real nightmare. because this this document was basically, which was not readable like the action the action plan, by the way, it was very complex.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: ah By the way, when this document was published, do you know I really studied the document? I used an AI.
Pascal-Emmanuel Gobry: Okay.
Sebastien Laye: I used AI to help me because it was too complicated.
Pascal-Emmanuel Gobry: yeah
Sebastien Laye: It was not humanly feasible, you know, and I didn't have the time. So this document was saying, okay, if you export the chips, the NVIDIA chips to dear friends in Europe, that's fine.
Sebastien Laye: If you send them to Russia or China, no, no way.
Pascal-Emmanuel Gobry: Right, there was like this multi-tier list of countries, yeah.
Sebastien Laye: Yeah, exactly. And then you had third tier that was very complicated because to send your chips to Israel or Saudi Arabia, that was like a bureaucratic nightmare. So what the travel administration did, they rescinded you know the diffusion network in several executive orders. The last one being when he was in Saudi Arabia or in the Emirates, remember during that trip.
Sebastien Laye: And then we were waiting because we were like, okay, what is he going to say about China? And then in the case of China, the situation is confusing because there's a total reversal of fortune, you know.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: We thought at some point, by the way, a couple of days ago, President Trump said that at some point, he even considered breaking up NVIDIA. So at some point, the Chinese hoax was sort of prevailing because the idea was to ah prevent NVIDIA from sending any chip.
Sebastien Laye: Of course, the cutting edge chips were also the old chips.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: to China and maybe to do something with Nvidia. Now fast forward a couple of weeks later, now we are late July, it looks like this is quite the opposite. It looks like Lutnik just said that it was fine to export any kind of chip to China because ah we wanted the the Chinese developers to be addicted to US technologies.
Sebastien Laye: So it looks like the new vision. And again, you know I'm not sold on that, but I'm not for or against. I don't know. I'm still studying the thing. It looks like the new vision is to say, that's fine.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: We should we should export whatever we we We have yeah at least the because A, we we close you now to NVIDIA. NVIDIA is really the epitome of the US leadership.
Sebastien Laye: And if the Chinese are using you know chips, their hands are tied. You know you know what I mean? the building their technologies on our own infrastructure and we're going control them. So... um So it looks like the Chinese Oaks, at least last time that I checked, like yesterday, it looks like the Chinese Oaks lost the battle to Silicon Valley because Silicon Valley wants to export to China.
Pascal-Emmanuel Gobry: Yeah, yeah,
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: They want to be in China. I'm just pointing out it could be referred ri ah between theteno bros and the and, you know what i mean, and the MAGA base. So we had the H-1B visa in December.
Sebastien Laye: We had the AI moratorium. Maybe we're going to have a conflict you know about but China.
Pascal-Emmanuel Gobry: Right, right.
Sebastien Laye: ah We'll see. What is sure is that the devil is in the details. And as far as I'm concerned, you know if I go back to the technicals, I think that we could export at least the age twenty maybe not the the most recent B200, but at least the H20, we could export them to to China.
Sebastien Laye: But we need to to, a little bit like with military weapons, you know we need to... to make programs in terms of of being better at tracking, you know, where the chips are sent to avoid smuggling.
Pascal-Emmanuel Gobry: Yeah, yeah, yeah, yeah. Yeah, yeah.
Sebastien Laye: We we got to report...
Pascal-Emmanuel Gobry: All of a sudden, Malaysia imports tons of of computer chips.
Sebastien Laye: Yeah.
Pascal-Emmanuel Gobry: It's crazy.
Sebastien Laye: And we got to report...
Pascal-Emmanuel Gobry: theyre have a huge They have a huge AI industry in Malaysia.
Sebastien Laye: We got to...
Pascal-Emmanuel Gobry: It's ah it's very well known.
Sebastien Laye: that's that's That's not okay. I'm not okay with that. We got to report last last last week ah that $1 billion don of Nvidia chips were smuggled between Malaysia, Indonesia, and China.
Sebastien Laye: So we need to be better you know at controlling that.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: I know if it's a question of stuff or stuff or whatever, but we need to do something. I think it's just just a technological issue. We need tracking devices. We need to be better at that and we need to be more strategic when it comes to export control.
Sebastien Laye: And if for the Chinese, i don't think that we should export the last cutting edge chips, okay? Definitely.
Pascal-Emmanuel Gobry: Yeah.
Sebastien Laye: We should export the old ones only if Nvidia needs to make tons of money and if we want to get the Chinese developers addicted to this. ah We just need to be careful again because it's a technical thing. But remember, a lot of the compute power used to be applied at the training stage.
Sebastien Laye: And now it's more and more at the inference stage. The inference means that when you go and chat GPT and you use this new thinking, rezoning models like O3, it's giving you compute power right at the moment when you when you prompt.
Pascal-Emmanuel Gobry: Mm-hmm.
Sebastien Laye: okay
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: And for that sort of compute power, even the old chips, they can be useful. It's just that you you need massive, massive assemblage of chips, the kind of things that are
Pascal-Emmanuel Gobry: Right, right, right, right, right.
Sebastien Laye: that elon musk is doing so again devil is in the details but it's going to be something to follow in the next uh weeks or months and uh it's uh it's it's
Pascal-Emmanuel Gobry: Yeah, no, I agree.
Sebastien Laye: next number for
Pascal-Emmanuel Gobry: I'm i'm sort of baffled by these this expert control thing. my
Pascal-Emmanuel Gobry: under it's i'm When I say my understanding, my theory, I'm not basing this on reporting or conversations I've had or whatever. This is my theory. This is speculation. But my speculation is that this is something that Trump is giving to the Chinese for now while he's negotiating the final trade deal.
Pascal-Emmanuel Gobry: Right.
Sebastien Laye: You mean it will be some leverage to to some leverage against Rare Earth?
Pascal-Emmanuel Gobry: Because same same thing as the Chinese student visas, like cause cause because um Rubio announced they were stopping the Chinese student visas, and I threw a party.
Sebastien Laye: Yes.
Pascal-Emmanuel Gobry: And and then like two weeks later or three weeks later or something, Trump countermanded it. ah i don't i ah it's It's very baffling. the only thing i The only thing I can say is that it's not the final. we we we have the the The deal that currently exists between the US and China is an interim deal, and they're still negotiating the final deal.
Sebastien Laye: Yeah, I agree with you. I agree with you. I agree with you. And it could be a negotiating tool. And we're going have to see you know two to three months from now down the road where we really stand in terms of export control.
Pascal-Emmanuel Gobry: Yeah. Yeah. Yeah.
Sebastien Laye: but we're we're gonna have At some point, like I guess we're going a more finalized doctrine you know from the Department of Commerce regarding this and the full AI stack and the export package.
Pascal-Emmanuel Gobry: yeah yeah yeah yeah
Pascal-Emmanuel Gobry: Yeah. Yeah. No, but but very good. Like this issue of expert controls is very interesting. It's sort of murky right now. This is something to keep track of. Okay. Final question. We have a traditional question that we ask every guest on the podcast, and that is recommend a book. It could be any book, recent, old, fiction, nonfiction, that's not related to your area of expertise.
Sebastien Laye: Okay, can it be still to...
Pascal-Emmanuel Gobry: So not AI, not economics.
Sebastien Laye: Okay. Totally unrelated, you know, but, ah but ah and and I had already read that book a while ago, but over the summer, I've already taken my vacation. I was in Spain and I read again, SPQR.
Sebastien Laye: It's about the Roman history, you know, Mary Beard. it's It's the best, it's the best historical i account of the Roman Republic and Empire. You got two, I would recommend two.
Sebastien Laye: One, one in English, Mary Beard, SPQR. I think it was sold almost a million ah copies of that book.
Pascal-Emmanuel Gobry: Yeah, it's very popular.
Sebastien Laye: And another one that ah obviously is going to be harder to reach for American friends, but ah maybe there is a translation, but the regretted French historian Lucien Jarfagnon, he wrote a book on the history of ah of the ah Roman Empire, just the Empire, not the Republic, that is also amazing. You know you read both of them. If you can speak both languages, it's your your you're all good with Roman history. And by the way, when I was reading the book,
Sebastien Laye: and especially with the beginning of the empire, the Civil War, etc. That was very eerily reminiscent of the situation of the United States you know as a world power.
Sebastien Laye: That's why I'm saying that. Now, guess what? It's a little bit related to AI because you could find them some technologies at the time, like military technologies that are the equivalent of AI.
Pascal-Emmanuel Gobry: it
Sebastien Laye: yeah
Pascal-Emmanuel Gobry: Interesting.
Sebastien Laye: And if I can...
Pascal-Emmanuel Gobry: Well, I always thought that ah Trump was one of the Grachi, personally. we' We're not.
Sebastien Laye: Exactly, exactly. One of the Gracchi, maybe you have also but the very wealthy Claudius at some point, who was a wealthy populist.
Pascal-Emmanuel Gobry: Ha, ha, ha, ha, ha.
Sebastien Laye: There's several people. there's There's a little bit of Caesar in him too, to be honest. yeah Yeah, that would be interesting to do that sort of analogy. And it's related to AI, but it's not it's not nonfiction. It's not an essay. If you want to read fiction about AI and VR and all this technology issue, much better than any essays.
Sebastien Laye: I recently found out about a book, a novel from the 90s called Permutation City. Permutation City by Greg Egan is an Australian author.
Pascal-Emmanuel Gobry: Permutation City.
Sebastien Laye: It's just amazing. This book really, I mean, much better than ah The Free Body Problem, etc. If you want to find ah book packed with science, Permutation City by Greg Egan.
Pascal-Emmanuel Gobry: Okay. Oh, i will i will i will definitely I will definitely read that.
Sebastien Laye: It's fascinating.
Pascal-Emmanuel Gobry: All right. ah All right. Well, thank you. Thank you very much, Sébastien. It's always a pleasure. It's always very fun to to talk to you.
Pascal-Emmanuel Gobry: I hear you're moving back to the U.S.
Sebastien Laye: Yeah, 30 September.
Pascal-Emmanuel Gobry: soon ah to enjoy that American dream.
Sebastien Laye: 30 September.
Pascal-Emmanuel Gobry: ah
Sebastien Laye: Again. Bye-bye.
Pascal-Emmanuel Gobry: America is very lucky to have you. All right. Thank you very much. Have a good day. Bye.
Sebastien Laye: a nice day.
