Where do you fit along that spectrum? And I guess, does it matter if the users are going to be concerned about the outcomes?

Well, I think we want to be careful in the way that we speak about the technology to not anthropomorphize it, which is to say, make it sound like it's a human because it's really not.

And there's two ways of describing the same event.

You could say that the agent escaped the lab, which sounds like it's its own actor.

Or you could say that the lab had bad security.

The lab did not have control.

That's the way I think about it, that fundamentally these are engineering problems.

Are you personally not concerned that those harms and potential harms would be so great that it's it sort of would have been better to get ahead of them?

But it kind of begs this central question of why don't we need a government intervention? Why don't we need regulation in what those frontier labs propose to do?

Well, the first point is, if you look at the structure of the market for frontier AI right now, we do have a duopoly situation.

By any metric, whether it is market share or revenue growth or model capabilities, there are basically two companies at the frontier, and that is Anthropic and OpenAI.

and you have dario and sam have not both said okay let's pace the frontier to make sure our products are safe my point is guys go ahead i think if you feel that the right priority right now in your product roadmap is to make your products more reliable more trustworthy uh more predictable, then you should do that.

And by the way, they have every incentive to do this.

Lena Kahn just came out with a statement that I very much agree with, where she said that the major AI companies are subject to product liability for their products.

or if they release an unsafe product, they're massively liable for that.

Are we there yet at that level of debate, or this is just a warning about what would happen if we were to pull back on the work?

Does he know something we don't know to be able to say, OK, This is all a hoax, as he puts it.

No one is saying that they shouldn't take steps on their own to prioritize the safety of their own products.

Again, every corporation in America has an obligation.

But he's calling it a hoax.

Well, no, the hoax is that somehow we're going to be extinct as a species within a few years because of AI, and therefore we have to stop completely.

What do you think of that?

Well, I'm an American, and I think that the American government should have jurisdiction and control over AI in America.

I think we should maybe talk to other governments to see how they're thinking.

What do you make of it?

Well, what Meta has done here is that they've open sourced this model.

And what that means is they're giving it away for free.

So anyone can just take this model, run it on their own computers.

And if you're a conservative or a libertarian, that's a really good thing because you have total control over your own data.

You have what's called data sovereignty.

And you can't be censored by a big tech company or by the government.

I'm saying, David, let the comrades stay home.

Yeah

... software freedom.

You basically could run your own models on your own hardware and retain control over your own information.

And by the way, this is what enterprises typically do all the time is, you know, um, about half the global data center market is, you know, is on-prem, meaning enterprises and governments create their own data centers.

They don't go to the big clouds.

And by the way, I've got nothing against the hyperscalers, but just, you know, people like to run their own, um, you know, own data centers and, you know, maintain control over their own data and that kind of thing, and I think that will be true for consumers to some degree as well.

Um, so I do think it's an important area that we should, you know, want to encourage and promote.

[chuckles]

I mean, that's what's so fascinating is you guys, I, I think you align on so much stuff, but it does seem like you've come to different conclusions on this one.

Well, I mean, I, I don't know.

I mean, I think he, he's basically in a place where he acknowledges that AI is gonna be developed re- regardless, and so what we wanna do is remain in control of it.

I mean, I don't think we wanna hamstring US development to such a point where, again, then China just wins the AI race.

So I think he understands that the technology is moving forward, uh, and we should develop it in a way that allows us to remain in control.

So i- an area, for example, that I agree with him on is we want the AI to be as truthful as possible.

And, you know, he's pointed out that if you teach the AI to, to lie, like let's say, you know, you teach it to be woke or whatever, that's very dangerous because then the AI can lie to us about what it's doing.

But if, like, a Dasha Burns can prompt, and you know, I have zero understanding of coding and I can make an app on my own, I think a lot of small businesses, for example, would have hired, uh, a coder or an engineer to, to develop something for them, and now they can do it on their own, which on the one hand is great for the small business, on the other hand is not great for the other small business or employee [laughs] that would've otherwise had a job.

I, I think you align on so much stuff, but it does seem like you've come to different conclusions on this one.

Well, I mean, I, I don't know.

I mean, I think he, he's basically in a place where he acknowledges that AI is gonna be developed re- regardless, and so what we wanna do is remain in control of it.

I mean, I don't think we wanna hamstring US development to such a point where, again, then China just wins the AI race.

So I think he understands that the technology is moving forward, uh, and we should develop it in a way that allows us to remain in control.

So a- an area, for example, that I agree with him on is we want the AI to be as truthful as possible.

And, you know, he's pointed out that if you teach the AI to, to lie, like let's say, you know, you teach it to be woke or whatever, that's very dangerous because then the AI can lie to us about what it's doing.

But if, like, a Dasha Burns can prompt and, you know, I have zero understanding of coding, and I can make an app on my own, I think a lot of small businesses, for example, would've hired, uh, a coder or an engineer to, to develop something for them, and now they can do it on their own, which on the one hand is great for the small business, on the other hand is not great for the other s- small business or [laughs] employee that would've otherwise had a job.

That all sounds political, and that's been Andrew's, uh, uh, I guess, the rap that he's pointed out, uh, over and over again, that you can't take politics and, and use that as a reason, uh, to, to put him on a supply chain, uh, risk list.

No, there's a couple of things going on here, Joe.

I mean, the SPLC thing that I'm talking about, that just popped up today.

That just happened today.

It's got nothing to do with what happened at the Pentagon.

Uh, listen, with respect to what happened in the P- Pentagon, that's a procurement matter.

That was not a AI policy matter.

... that, that have been on the table, and in both cases, Anthropic and OpenAI effectively seem to be in the exact same camp, and yet one, potentially for political reasons, doesn't seem to be in the mix or is being called a, a security threat while the other is not.

Take a listen.

Last week, we released a national AI framework, and the idea is to create one rule book for AI in the US.

Uh, the problem that we're seeing right now is that you've got 50 different states regulating this, uh, in 50 different ways, and it's creating a patchwork of regulation that's difficult for our innovators to comply with.

So what the president has called for is one rule book.

Uh, what we did is, and, you know, I work with Michael Kratsios at OSTP and other folks at the White House, we looked at all the different things that the states were doing, and we tried to find some common denominators, and then we published that in a set of principles that we're calling the National AI Framework, and we're calling on Congress to act on that framework.

Uh, the framework contains things like child safety.

That's a very, uh, salient issue right now is how do you protect, uh, kids online? And so we wanna take care of that.

Do you see that changing the president's current thinking on, on exporting, uh, cutting-edge or lead-edge chips to China, or more generally in different jurisdictions around the world?

Are we winning? Is it about chips? What's the race specifically really about?

Well, I think that in general, we're ahead of China.

There's different layers of the stack.

So you've got the models, then you've got the chips, and then you've got the chip making equipment.

So you go down the stack.

I would say that the deeper in the stack that you go, the greater the American advantage.

I think on models, most people would say that our models are maybe six months ahead or so, plus or minus the Chinese models.

And we're going to be sharing a lot more on the progress of this program there.

What do you see as the most important use and where AI is being deployed and implemented best right now?

Well, it's interesting.

I think it-- there's been an evolution.

So I think we started with, you know, AI chatbots like ChatGPT, and in a sense, that was kind of like better web search.

Um, it was really great for research.

You ask it questions, it'll give you answers to anything.

Then we saw, um, we saw models add chain of thought, and they could start to do, you know, deeper reasoning.

Are we winning? Is it about chips? What's the race specifically really about?

... to buy the last generation as opposed to Blackwell?

Blackwell?I, I think that's a fair question, Hugh.

I mean, h- h- here's the way that we see it, is that the American policy f- and this is bipartisan, it goes back into the Biden administration, is that China is not allowed to buy our leading-edge chips, but they are allowed to buy our lagging chips.

The H200 chip that you're talking about and before that, the H20 chip, it is not a leading-edge chip.

It may have once been a leading-edge chip, but, uh, but it is now a lagging chip.

And the reason why you're willing to sell them lagging chips is because it takes market share away from their national champion, Huawei, and that's important because it deprives Huawei of revenue that they can use to then scale up.

And also, there's a very important developer network effect built on top of these chips.

Ah.

How about all the speculation about an AI bubble inflating? How do you see that? What are the greatest risks? And what is hand-wringing?

Well, we're seeing right now is a boom.

Like I mentioned, we're seeing hundreds of billions of dollars of new CapEx being invested in AI.

And again, at every level of the stack.

And we think that that boom is justified by the potential of this technology.

You know, in the dot-com bubble, you had this concept of dark fiber, which is there was a fiber built out and that fiber was not being used.

In this case, there are no dark GPUs.

Are you where you want to be? What's next?

The concern about this latest executive order is that while it addresses your concerns about many different pieces of state regulation, it does not provide for a single federal framework.

Well, at the end of the day, that single federal framework has to be enacted through law.

And we need Congress to do that.

And so the president has asked Congress to do that.

And he's tasked members of the administration to work with Congress to produce that framework.

In the meantime, what we've done here is articulate a set of principles.

We've said what values are important to us.

You know, what this EO allows for, is it the sort of hope that it will lead to the DOJ suing states like New York and California? And if that's the case, you know, the president and the administration's confidence that you'd win them?

The concern about this latest executive order is that while it addresses your concerns about many different pieces of state regulation, it does not provide for a single federal framework.

Well, at the end of the day, that single federal framework has to be enacted through law.

And we need Congress to do that.

And so the president has asked Congress to do that.

And he's tasked members of the administration to work with Congress to produce that framework.

In the meantime, what we've done here is articulate a set of principles.

We've said what values are important to us.

You know, what this EO allows for, is it the sort of hope that it will lead to the DOJ suing states like New York and California? And if that's the case, you know, the president and the administration's confidence that you'd win them?

I think that's right.

Yeah, but I think that is definitely – I think of all the companies, that particular one has been the most aggressive in terms of the regulatory capture and pushing these – for these regulations.

And just – I mean, let's just bring it up a level.

It doesn't have to be about them.

There's now something like 1,200 bills going through state legislatures right now to regulate AI.

Right.

25% of them are in the top four blue states, which are California, New York, Colorado, and Illinois.

And so are we enabling them in some way? Why don't you talk about the sort of state of play and then our strategy?

to ensure

that this is done correctly, right? And I think that, you know...

It's a good recruiting tool.

Yeah.

Join the virtuous team.

Yes.

I think that's right.

Maybe, David, you could lay out the sort of the what's most important to to get right you've already indicated you know open source is one example um you know you alluded earlier to sort of our strategies relates to chips um you know some people say that yes it's it's a good idea to do what we're doing because it'll you know limit domestic semiconductor production other people say oh well you know some of these companies say chips are their biggest you know limiting factors and so are we enabling them in some way why don't you talk about our sort of state of play and then our strategy

What's your position on that and whether the United States wants to kind of leave the world open to China to sell its own technology?

Well, there was a view in the previous administration that we shouldn't sell chips to many countries, including the resource-rich Gulf states.

And I think that was a major mistake, because every time you tell a country that they can't buy the American tech stack, what's their reaction going to be? They're going to turn to China and adopt the Chinese tech stack.

I have a very simple metric for measuring whether we're winning the AI race, which is global market share.

If we look around the world in, say, five years and we see that the American technology stack has, say, 80% market share, that means that we won.

But if we look around the world in five years and we see that the Chinese technology stack, and I'm talking about Huawei chips and deep-seek models, for example, has 80% market share, then obviously we lost.

By the way, that's what happened in 5G.

But when you're allowing only less powerful chips, I mean, I think the president even called them obsolete versions of NVIDIA's chips into China, does that mean that ship of innovation in China has already sailed? Because they need to have the most sophisticated, and they're going to have to build it themselves.

How far ahead of China do you think we are right now in this race? How long would it take for them to catch up? How do we prevent that from happening specifically?

So I think when you look at our lead in the AI race, it kind of depends on what level of the stack you're looking at.

So if you look at the top of the stack with the AI models, we're only, I would say, months ahead of the Chinese models.

They've done a good job with models like DeepSeek and Quen.

But if you go deeper in the stack, we have, I'd say, more significant advantages.

So when it comes to chip design, our leading company is NVIDIA.

Their leading company is Huawei.

How do you think about that dynamic?

Do you have any sense of when he met with Jensen Wang last week at the White House, what case Mr. Wang made to the president that it was the right thing to get the American technology stack back into China?

Well, Jensen has been making the case publicly for competing in China, and I think there are a lot of merits to the argument.

If you look at developments over the past year, Huawei has advanced very rapidly.

They have a new technology called Cloud Matrix.

which networks together 384 of their Huawei SN chips.

Now, the Huawei chip is not as good as the Nvidia chip.

The Nvidia chip is three times more powerful.

How involved have you been, David, in these kinds of talks, in framing the president's thinking that actually just having direct access to lower performance compute is okay with China right now?

And I wondered if you'd just reflect for a minute of how you feel that the final sort of composition of this legislation supports the goals, the bigger picture goals that you and the president had hoped for when you set out to do it.

Sure.

So when the president first took office and even before he took office, when he was on the campaign trail, he promised to make the United States the crypto capital of the planet and to replace the approach of the Biden administration, which was essentially regulation through prosecution.

I mean, the Biden administration was doing everything it could.

to drive crypto companies offshore.

And he promised to change that approach and make and provide regulatory clarity so that the innovation could happen in the United States.

And I think that, you know, the Trump administration has done that.

What is your assessment of how China can compete with its own GPUs in markets outside of America?

What is the best possible realistic outcome in the coming weeks that we can expect now that it looks like we're heading towards another summer killing season?

Well, Charlie, in order for there to be a peace deal, there has to be some sort of overlap in what the two sides are willing to accept, and I- I just don't see what that is at the present time.

The Ukrainians have said that they won't give up a square inch of territory.

I mean, that's what they've said.

Uh, and the- the- the Russians have said that they're not willing to a- a- agree to a ceasefire.

They want a permanent peace deal.

They think that a ceasefire would be the equivalent of trying to call time out in the middle of a war, and the Ukrainians will just use that time to regroup and re-arm and, uh, and- and take a losing position and- and fortify it.

What is your response to that kind of doomerism that this technology's gonna get wildly out of control?

What's your thinking, the administration's thinking on the usefulness of some of these chip curbs?

Well, John, good to be here.

So, on the chip curbs or export controls, I think there's strong bipartisan consensus in Washington not to let China have our most advanced semiconductors because there's a potential dual use to the technology.

But I think that Jensen's right with respect to the rest of the world.

We want the rest of the world using our semiconductors, building on the American tech stack, as opposed to creating an opportunity for Huawei to come in and create a Huawei Belt and Road.

I think we need to revise the export controls with respect to the rest of the world, but I think we have to keep them for China.

What's your take on the concerns, certainly on the Democrat side of the aisle, that there aren't enough safeguards here to keep maybe the president himself and his family from benefiting from this kind of legislation?

Well, so first of all, we have significant bipartisan support for this.

And hopefully we can clean this all up and let people know that, hey, the intent here is good.

So just to go back to one other point you said about not picking winners and losers, I think that's a very important point.

comment that you made that I see as fundamental to my job.

I mean, look, I think we do think that Bitcoin is special for the reasons I said.

Beyond that, we do not want to be in the business of picking and choosing winners in the space.

And again, if some other digital asset could prove that it's as decentralized and as secure and as widely accepted as a store of value as Bitcoin, then maybe it could be elevated in the same way that Bitcoin has.

I mean, I'm laying out the criteria for you.

When will we see the actual bones of a framework? So if I want to do J coin and I want to make an angel investing coin, or I want to do something fun, an NFT collection for all in or something, when will we have the actual rules of the road? And that's going to be multiple agencies,

I would say so.

In any event, what they didn't know, and what I then proceeded to put out there, is that I sold all my cryptocurrency prior to day one of the administration, because I didn't want to even have the appearance of a conflict.

And by the way, I could have waited.

I didn't have to do it like that.

But I decided to do it and take it upon myself to do that because you just know that when it comes to crypto, there's gonna be a lot of fluctuations in the market.

You never want someone to be able to point at one of those fluctuations and say somehow that the cryptos are benefited from that and create a conspiracy theory, which is exactly what basically happened.

So first of all, I got rid of all my cryptocurrency prior to day one.

And so, you know, that's why you're here, and, uh, hopefully we can clean this all up and, uh, let people know that, hey, the intent here is good.