Why do you think this is? It seems like a lot of people in AI really want very strong export controls on China, spanning all the way from many of the accelerationists all the way to Anthropic and many of the AI safetyists and seemingly everyone except for Jensen.

Yeah.

Well, look, I think it makes sense.

It is in all those peoples and companies and really the national interest for us to have these kind of strong controls.

But just take it from the company's perspective individually, right? Obviously, the labs want strong export controls because they don't want their active competitors, the Chinese AI labs, to have as much compute.

Similarly, actually, the hyperscalers in the U.S. should also want strong export controls.

sport controls, because where are these chips going? They're going to Chinese hyperscalers.

So what would you say is your policy agenda? Like if you had like a, you know, top five things to do, what would you do?

What is the most capable system that the Chinese have today? How long ago did we have it? And why is it important to maintain that lead?

Sure, so the most capable Chinese system, some debates, depends on the day, but it's probably GLM 5.3, one of their models.

The debate about how far it is behind the US model, probably somewhere between six and 12 months.

I think the Anthropic recently put out a report saying it was on the shorter end.

And basically what that means is the capabilities of that model are equivalent to those that the United States, that a frontier model from a US lab was six months ago when they released them, or whatever that number is.

It's actually very hard to do this evaluation, partly because it's really hard to benchmark AI models, and also because we actually don't know where the frontier is.

But I think generally the estimates are the US is about six to 12 months ahead of China in AI.

Chris, can Besson's proposals be a good start for collaboration?

What is the most capable system that the Chinese have today? How long ago did we have it? And why is it important to maintain that lead?

Sure, so the most capable Chinese system, some debates, depends on the day, but it's probably GLM 5.3, one of their models.

The debate about how far it is behind the US model, probably somewhere between six and 12 months.

I think the Anthropic recently put out a report saying it was on the shorter end.

And basically what that means is the capabilities of that model are equivalent to those that the United States, that a frontier model from a US lab was six months ago when they released them, or whatever that number is.

It's actually very hard to do this evaluation, partly because it's really hard to benchmark AI models, and also because we actually don't know where the frontier is.

But I think generally the estimates are the US is about six to 12 months ahead of China in AI.

Chris, can Besson's proposals be a good start for collaboration?

What, what are you saying? Well, is there an endpoint, and what would that endpoint potentially be if America were to lose?

Well, I don't necessarily think there's an endpoint, um, because this is a technology that's gonna keep increasing exponentially and, and we don't really see a ceiling to it, which is why I think people are, you know, have some, um, concerns about where this is going, but also profound hopes.

Um, I, I think if you think about a world where AI is extremely capable and we fall behind, w- what it would mean is China would have unfettered access to every single US electronic system, every single computer, every single phone, every single industrial control system, every single military vehicle, everything.

Um, that obviously is completely unacceptable from a national security perspective.

Uh, also from an autonomous systems perspective, as militaries transition to, to becoming, you know, largely more driven by drones than by, by, by people, at least on the battlefield, which we're obviously seeing in Ukraine, um, the smarter drone is probably going to win the vast majority of the time.

So the capabilities of the platform are gonna become less important, and the intelligence of the platform is gonna become more important.

And if the Chinese have smarter drones, then their drones defeat ours all the time.

So final thought, when it comes to the US political debate about all of this, and you're not a politician, I'm not gonna invite you to take part in that debate, but when it comes to the ability of the US now to sort this and to kind of come to grips with it, uh, and whether it's Congress or whether it's something else, can you see anything in, that's happening at the moment that makes you optimistic?

Like that, you know, that it's not bad, but that doesn't make me feel seen.

No.

Not all.

No.

And, and if somebody...

And, and here's my thing with this.

Sure, this takes some time.

Yeah

So Chris, I'm curious for your reflections on, on that arc and maybe what it tells you about the future of AI policymaking.

Yeah, I mean, look, the, the particulars of this individual incident aside, the administration very clearly has fundamentally changed its policy on AI over the last few months.

I, I think that Mythos, the Mythos release really did kind of cause a, a sea change in Washington on AI policy.

The, the policy has not been, you know, let it rip, let the private sector rip for, you know, uh, four months or so.

Uh, I think you saw the culmination of that in the AI executive order that, you know, put forward a trans-- a voluntary, um, you know, licensing regime, but, but ultimately or a v- a voluntary, um, um, you know, safety regime, but ultimately it's obviously kind of mandatory.

I think this basically shows that it's mandatory.

Um, so it, it's, it's logical that the admin-- this is a real problem, and I think that the administration is realizing that this is, this meaning like AI regulation and the, the very powerful capabilities that these models have, um, it requires regulation.

it goes to show that there's, there's some real issues over there, I think.

So Chris, I'm curious for your reflections on, on that arc and maybe what it tells you about the future of AI policymaking.

Yeah, I mean, look, the, the particulars of this individual incident aside, the administration very clearly has fundamentally changed its policy on AI over the last few months.

I, I think that Mythos, the Mythos release really did kind of cause a, a sea change in Washington on AI policy.

The, the policy has not been, you know, let it rip, let the private sector rip for, you know, uh, four months or so.

Uh, I think you saw the culmination of that in the AI executive order that, you know, put forward a trans-- a voluntary, um, you know, licensing regime, but, but ultimately or a v- a voluntary, um, um, you know, safety regime, but ultimately it's obviously kind of mandatory.

I think this basically shows that it's mandatory.

Um, so it, it's, it's logical that the admin-- this is a real problem, and I think that the administration is realizing that this is, this meaning like AI regulation and the, the very powerful capabilities that these models have, um, it requires regulation.

it goes to show that there's, there, there's some real issues over there, I think.

So talk a bit about how you and your colleagues in the administration, late 22, going into 2023, How did you kind of rank order the risks from AI? How did you set priorities? Just kind of paint that picture.

Yeah, I think actually it's helpful to take a step back one more year and think back to kind of 2021 and 22.

And the reason for that is that we had a theory in the Biden administration that basically the scaling laws were going to hold.

Scaling laws meaning that AI would get way more powerful.

Would get way more powerful and also that computing power was going to get way more important.

And basically the more compute The more computing power you put into an AI model, the better it's going to be.

Obviously, there are other advances at the same time.

right? That's right.

So what do you think? Fact? fiction, or somewhere in the middle?

I think somewhere in the middle.

There's a few components of this that all of which are different, and also different companies and industries are a little different.

I think you could actually argue that well-crafted export controls increase certainty.

They say very clearly, here are the rules of the road, you go forth after that but like this is where you are not going to be allowed to sell and that was actually part of the the idea behind the the the biden you know administration's position on the ban ai chips and then also the one the trump administration put forward is basically saying here's a line you cannot cross that now obviously that line has moved down several times and the reason for that was because of advancements in in you know ai that that enabled people to get equivalent levels of performance by kind of enabling technical workarounds.

And candidly, it was also that we were learning as we were doing, right? This was hard.

We were regulating on the frontier, and the first control, we made a mistake.

So fact or fiction, AI chip smuggling to China does not happen.

So can you walk us through what exactly has happened over the last 48, 72 hours? Why is this so big, and what should this audience understand on the technical side, sir?

Sure.

Thanks for having me, Steve.

So I...

In yesterday or two days ago, uh, President Trump announced that the, the NVIDIA H200 chip will, could be exported to China.

So this is not NVIDIA's most powerful chip, but it is their second-most powerful chip.

It's the best chip of the previous generation, and most importantly, it is substantially better than anything available in China.

Isn't there some examples, particularly with Huawei, uh, where, where it actually proves our case, sir?

So can you walk us through what exactly has happened over the last 48, 72 hours? Why is this so big? And what should this audience understand on the technical side, sir?

Sure.

Thanks for having me, Steve.

So yesterday or two days ago, President Trump announced that the NVIDIA H200 chip could be exported to China.

So this is not NVIDIA's most powerful chip, but it is their second most powerful chip.

It's the best chip of the previous generation.

And most importantly, it is substantially better than anything available in China.

Is that correct? And there's some examples, particularly with Huawei, where it actually proves our case, sir.

What is the timeline here? What has actually happened and, and how has this news changed our policy on whether the US can or cannot send chips to China?

Yeah, that's a good question.

I'll just walk through quickly the, the history here.

So it was starting in, in October 2022, uh, the United States first implemented policies to say the most advanced AI chips, uh, that were actually drawn along the A100s, a level below this one, the previous generation, anything more advanced than that cannot be advanced to Chi- uh, cannot be sold to China.

Uh, the policy at the time was to maintain as large of a lead as possible over China.

And there are basically two prongs to that.

Uh, one was maximally slowing China down in AI, and the other one was maximally investing at home.

Hmm.

Okay, so before we get into what's happened in the last 24 hours and the last week, like can you give kind of your view on this year and how a lot of these different, um, uh, h- you know, h- how the trade war has been kind of evolving, uh, in your view?

Yeah, absolutely.

So I mean, I think the, the, the administration has very much kind of gone back and forth on, on, on where they are in China policy.

I think that you obviously have had enormous, um, you know, threats of, of tariffs and, and moves to, to kind of really kind of separate US and Chinese economies on pretty fundamental levels.

On the other hand, uh, I think you, you haven't actually seen a lot of implementation there.

There's a continued desire to talk with the Chinese and when it comes to some of the, the, the technology protection measures where I think the, the Biden administration was more focused, um, it's been a, a, a l- there's, it's been a little bit more stop and go.

I, I think that, you know, generally since the first Trump administration, there has been a big pivot in Washington, uh, to move towards a, you know, away from the, the, the policies of Ch- China of let's like deepen integration in order to, to kind of get them to change, uh, and towards a more competitive stance.

What, what, uh, what is the most current debate on the ground in Washington around, like, black wells specifically?

So, that is tangible.

So, I think the, the, the soybean purchases were good.

It's good that we got that back in, in track.

It's good that we kind of rolled back the rare earths.

Um, the things that the administration, you know, gave the Chinese were, were 10% reduction in tariffs for, for fentanyl cooperation, which again, is a good thing.

It's good to have the, the Chinese actually working on that.

Um, we'll see if they actually follow through with it.

These, this type of, um, rivalry and, uh, and retribution is still very much a part of the dialogue today.

So that is tangible.

So I think the, the, the soybean purchases were good, it's good that we got that back in, in track.

It's good that we kind of rolled back the rare earths.

Um, the things that the administration, you know, gave the Chinese were, were 10% reduction in tariffs for, for fentanyl cooperation, which again, is a good thing.

It's good to have the, the Chinese actually working on that.

Um, we'll see if they actually follow through with it.

Uh, suspension of reciprocal port fees and then the, the walk back of the, the 50% rule on export controls.

These, this type of, um, rivalry and, uh, and retribution is still very much a part of the dialogue today.