00:00In your work with the president in consulting and advising on the formulation of this executive order, what was the problem that you were trying to solve for?
00:11And what is it that you said to the president about why this EO was the right approach to focus on state level laws?
00:20Well, thanks for having me. The problem that we see is that you've got a thousand different bills going through state legislatures right now to regulate AI.
00:29And over 100 measures have already passed. Some of these bills are contradictory and you've got 50 different states running in 50 different directions.
00:36That type of compliance regime is going to be very hard for small companies and startups, especially innovators, to comply with.
00:43And so what we need is a single federal or national framework for AI regulation.
00:49And that's what the president has supported. And by the way, he supported this for a long time.
00:53If you go back to his July speech on AI, he called for a single national framework then.
00:59And what we've done with this EO now is to make clear that that is the administration's policy and to task members of the administration to work with Congress to try and enact that framework through legislation,
01:10because ultimately this needs to be a law and in the meantime, create tools that the administration can use to push back on examples of the most onerous and excessive state regulations.
01:20David, there is, of course, some pushback, you know, on the executive order from the states themselves, from other Republicans.
01:29You know, as you know, like I studied the July speech and strategy closely.
01:36A big part of it, you know, was infrastructure related and about deregulation.
01:41The 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.
01:52Well, at the end of the day, that single federal framework has to be enacted through law, and we need Congress to do that.
02:00And 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.
02:07In the meantime, what we've done here is articulate a set of principles.
02:10We've said what values are important to us.
02:13We've said that we want to protect child safety.
02:16That's important.
02:17We want to respect copyright.
02:19We want to preserve the ability of local communities to choose what infrastructures in their communities.
02:25We're not seeking to preempt the states in any of those areas.
02:29So this is an important set of principles that we have put forth.
02:33And at the same time, the EO provides for a number of tools that can be used to push back on excessive state regulation.
02:41And let me just illustrate why I think this is so necessary.
02:44Because what we're really talking about here is regulation of AI models and algorithms.
02:49Well, think about how an AI model is developed.
02:53You can have developers in one state or multiple states writing the code.
02:57It can then be trained in a data center in another state.
03:00You then can have inference happen in another state.
03:03And the entire service is provided over the Internet using national telecommunications infrastructure.
03:09So you're dealing there with at least four different states.
03:11And all of them can lay claim to regulating those AI models.
03:15And those regulations can be in contradiction with each other.
03:19Even Democrat governors have admitted this is a problem.
03:21So just the other day, Kathy Hochul, the governor of New York,
03:24basically said that she might prefer to enact California's SB 53,
03:30which is a regulation that they just passed in California,
03:32rather than the bill that her own assembly gave her, the RAISE Act.
03:36Because she sees that, wait, do we really want to create this patchwork of different regulations?
03:40So even Democrat governors are realizing this is a problem.
03:42And if they all run in different directions,
03:44then we're going to end up with a patchwork or a mishmash of regulations
03:48that are impossible for companies to comply with.
03:50What the president is calling for here is just common sense.
03:54We want to get to a single national framework of compliance,
03:58as opposed to 50 states running in different directions.
04:01Meanwhile, Kathy Hochul actually is getting a bit of criticism, perhaps,
04:04for narrowing and what some are saying is bowing down to business.
04:08David, I'm really interested in how you oppose that view,
04:11because there is anxiety in the population, AI versus jobs, AI versus energy bills.
04:17How are you giving them the sense that we haven't seen federal government
04:21and indeed now state governments just handing over the reins to big tech billionaires,
04:25as people call them?
04:27Right. No, I understand there's a lot of fear out there about AI and job loss specifically.
04:32And a lot of those fears have been have been drummed up.
04:34Let me just say on the job loss question, because I think this is really important,
04:37that Yale just released a study and it showed that in the 33 months after the launch of ChatGPT,
04:43there was no discernible disruption to the U.S. job market. None.
04:47They said no discernible disruption.
04:49And in fact, if you look right now, more jobs are being created than being lost.
04:52So this whole idea of job losses just isn't true.
04:55There was an article in the Wall Street Journal just last week talking about the construction boom
05:00that's happening that's benefiting construction workers like electricians, like plumbers,
05:04like workers who pour concrete or hang drywall.
05:07Their wages are up 30 percent because this infrastructure boom that's happening right now.
05:11And there's actually a job shortage in many of those trades, meaning we need more workers going into those trades.
05:17So what we're seeing right now is an overall AI boom that's benefiting the economy.
05:21You know, the GDP growth rate was tracking about 4 percent, and half of that, up to half of it,
05:28has been attributed to AI.
05:30So I just think that this narrative about job loss has been blown out of proportion.
05:35Certainly there could be job displacement in the future, but we haven't seen any of that so far.
05:40It's been quite the opposite. It's been job gains.
05:42David, final one on the EEO, if I may.
05:46You know, what this EEO allows for, is it the sort of hope that it will lead to the DOJ suing states
05:53like New York and California?
05:56And if that's the case, you know, the president and the administration's confidence that you'd win them?
06:01Well, that is one of the tools that is in the EEO, is that the DOJ has been tasked to form a litigation task force
06:09that would have the ability to push back on excessively burdensome state laws,
06:15laws that may be unconstitutional, violate the First Amendment, things like that.
06:19By the way, the DOJ already had that power.
06:21So this is not a novel power.
06:23But what's being done here in this EEO is we're marshalling all of the resources of the federal government
06:27behind the strategy of the president to create a national framework.
06:30Now, in terms of what laws we go after, that's a decision that has been made.
06:34You know, we haven't decided whether California or New York should be targets in that way.
06:39The one that I think is probably the most excessive is this Colorado law that seeks to prohibit algorithmic discrimination.
06:46What that basically says is if an AI model has a disparate impact on a protected group, then that model is violating the law.
06:54Model developers, by the way, have no idea how to comply with this because they're not aware of all the downstream uses of their model.
07:00I mean, if a business decides to use an AI model in a hiring decision, for example, that business is already on the hook for discrimination.
07:07So how would the model developer know that it was being used in that way?
07:11But what Colorado is trying to do there is get their ideology inserted into the model.
07:16That's very concerning to us. We think there's a First Amendment issue there.
07:19But look, we haven't made any decisions in terms of how that litigation task force will be used.
07:23David, briefly, all of this is set in the context of U.S. versus China and a deem to run forward on AI development.
07:32Meanwhile, it's been a busy week and H-200s might indeed be able to get to China.
07:36How many do you think you'll do in volumes?
07:37And what do you think the appetite is of China to buy NVIDIA's more sophisticated chips?
07:42Well, it's interesting.
07:43I just saw an article that said that China was rejecting the H-200, so apparently they don't want them.
07:48And I think the reason for that is they want semiconductor independence.
07:52The same way that the United States wanted to be energy independent, they want to be semiconductor independent.
07:56So they're rejecting our chips.
07:58And that's part of the calculation that goes into the decision of what we authorize to be sold to China.
08:04The U.S. policy has always been that we don't allow the leading-edge chips, and we're not.
08:08This H-200 chip, it was state-of-the-art a couple of years ago, but now it's been superseded by the newer Blackwell architecture and the Rubin architecture that's coming out next year.
08:18So this is now a lagging chip, not a leading chip.
08:20But what you see is China is not taking them because they want to prop up and subsidize Huawei.
08:26They want to create a national champion.
08:28And that was part of our calculation of selling not the best, but lagging chips to China as you can take market share away from Huawei.
08:34But I think the Chinese government has figured that out, and that's why they're not allowing them.
Be the first to comment