00:00Let me just start with the slowdown remarks that we've heard from these top executives, Altman,
00:05Amadei, Musk, saying that a slowdown right now is needed. Do you think that they're genuinely
00:11scared about the pace of development, or is this an attempt to put down competition?
00:20I think they're genuinely worried. I think they see things that we don't see.
00:25They see the models that are going to be released in six or nine months in the future.
00:31And I think they've noticed elements of those models that worry them.
00:36And so they're calling attention to this. And I'm not sure that this really helps them competitively.
00:43I think until this point, there's been a big focus on acceleration, and that's benefited them massively.
00:49So I'm not sure it's really in their benefit, in their best interest to slow things down.
00:53But they see something that worries them, and that's causing them some concern.
01:00And Michael, what do you see? Do you see an actual threat that is currently being posed by artificial intelligence,
01:07or for the lack of a better word, because that's all I've been hearing throughout the weekend,
01:11is AI on its path to pose an existential threat to humanity to kill us all?
01:19Yeah, so you maybe wonder what's different now than, let's say, a year ago.
01:23And I think there's a couple of big differences that we've seen.
01:28The first one is the growth of recursive self-improvement.
01:31Now, this is a very fancy word to mean that the models are getting so good that they can help
01:36themselves to get better.
01:38So essentially, it brings the human out of the loop, so these models can just get better at an amazing
01:44speed.
01:46And the second one is the hacks that we've seen recently.
01:50You may recall the OpenAI hack against Hugging Face.
01:54I mean, this was really a coordinated attack by AI agents that was not anticipated and not able to be
02:03shut down very quickly.
02:05So I think there's some actual facts that are arguing for the growth in these models.
02:13And it's something that at IMD we keep close track of.
02:18So this might be a little bit of a straightforward one, but I still want to hear your thoughts on
02:24it.
02:24Why are these top executives racing to continue building then when it comes to the technology behind AI,
02:32when there's clearly proof, as of now, that it is capable of posing that threat?
02:40Well, I think we're in a competitive situation where it's like a prisoner's dilemma.
02:45If one person stops and nobody else stops, then they fall behind.
02:51So nobody really wants to be the first one to slow down.
02:54And that's just in the United States.
02:57And we're starting to see some agreement between the big AI models in that country.
03:03But that doesn't look at the rest of the world, and in particular China.
03:07And this is a big concern for President Trump.
03:10He doesn't want the U.S. to cede its lead in AI to China.
03:15So if the U.S. slows down and China doesn't, that puts a global competitive situation in place that we
03:21don't have at the moment.
03:24So do you think going forward it's not going to be an argument of as much as man versus machine,
03:29but it will be an argument of the United States versus China, even if a slowdown does happen that is
03:35relevant to the United States?
03:36Do you think China will follow suit?
03:42That's very, very difficult to say, if they'll follow suit.
03:46I think it's in their best interest to also limit AI in terms of biological and chemical weapons or military
03:54uses or privacy implications.
03:58But they've been very clear in that country that they want to be, by 2030, number one in the world.
04:05So it's not like they're gunning for a number two or three.
04:08They want to lead the world.
04:09And they put the resources behind that.
04:12So I think we ideally we want a combination of industry self-regulation, which is what's being discussed at the
04:18moment,
04:19with some kind of multilateral global regulation or at least attempts to monitor and slow this down.
04:29Whether that will happen or not in the current climate, honestly, it looks quite unlikely.
04:36OK, and Michael, just finally, if the slowdown in development does actually happen, despite President Trump's remarks against it,
04:44what breaks first in the AI investment cycle?
04:46Would it be a supply of memory chips?
04:49Would it be valuations?
04:50Would it be data centers?
04:51Would it be something else that we're not really close, closely following?
04:57So I think that the slowdown in the development of frontier models doesn't necessarily need to slow down the development
05:04of AI more generally,
05:07because what most people use, what most organizations use, is not the latest and greatest.
05:13And, you know, the current models are, you know, capable of providing the benefits that most organizations need or individuals
05:21indeed.
05:22So I don't see a necessarily an immediate impact on the ability of these models to bring value to organizations.
05:29Now, if you're asking about the financial markets, it's very difficult to say,
05:32because I don't think they're particularly rational at the moment when it comes to valuation of these companies.
05:38So anything could happen from the perspective of the markets.
05:42But I think we're going to still see, you know, massive demand for energy, for chips, for memory storage.
05:48I don't see that slowing down at any point in the near future.
Comments