Venture Europe: Entrepreneurship | Technology | Venture Capital | Eu/Acc
00:12:35 · S0
[00:12:35] S0: Yeah. Yeah. Well, I think, you know, there's there's two there's two groups within AI. There's the ones who, as I was saying, you know, see this as a special thing, and they are concerned about the responsibility, and they're basically good guys. And, you know, it's a bit like, you know, Oppenheimer creating the nuclear technology and being conscious that this is really this is big stuff. And then there's another group that I would say is kind of in denial and says it's just a normal technology. And this is the sort of the open weight, open model system, open source. And it's like, well, you know, it's just a technology, guys, and don't freak out. You know, we can just distribute it to everybody. We don't have to confront the danger of bioweapons or the danger of autonomous lethal warfare and so forth. And, you know, it doesn't... We don't need to control it particularly. I think that's going to be proved wrong in the end. I think there's going to be some disasters which force politics to get even more anti AI than it already is. And at that point, AI is going to be regulated. But for now, the Jensen Huang Group, one could call it, are kind of all in on open weight. I think that's maybe a little bit less virtuous than the ones like Dario who take it seriously, the danger. But look, everybody is a human being. Everybody is complicated. Everybody has some white and some black, some good and some evil. Unfortunately, this is real life.
00:14:05 · Calin Fabri
[00:14:05] S1: It's real life. To your point, I had just been to LA three weeks ago, and on the stage it was Jensen Huang, Satya Nadella, and Elon Musk. You could definitely see them splitting into two groups. On one hand, we need to be responsible, we need to pace a little bit, the Frontier Labs in particular, and on the other hand, Jensen were like, well, we need to keep going, this is not something special. What do you make of the argument that the ones that they try to pace the frontier labs is more like a regulatory capture strategy?
00:14:43 · S0
[00:14:43] S0: Yes. Would say, look, I do have one criticism of the frontier, and particularly the anthropic approach, which is I think occasionally there are people who exaggerate on that side, they say, you know, more than ten percent risk of human annihilation in the next ten years or by 2030 or something like that. If you think about it, that's a bit crazy. Right? I mean, first of all, we have no data to say what is the probability. I think you can say it's a non zero probability. That's what Demis says, and I think that's the right way to say it. We don't know how big the probability is, but it's non zero. But to say 10%, I mean, that is crazy high. I mean, first of all, the most likely outcome is going to be that there is a sort of mid level disaster. Think about Chernobyl blowing up. Think about 3 Mile Island. Think about the dam breaking and New Orleans being flooded. Right? Something very bad. Okay? But it's not human annihilation. Come on. So that's going to happen first, and then I think there will be a political reaction, and then we will take a lot of safety precautions. But the regulatory capture claim, Jensen Huang makes, which David Sacks, the former White House adviser makes, this is just nonsense. It's crazy nonsense. If you know anything about the history of someone like Dario Amodei, he was sincerely writing about serious AI risk, you know, before OpenAI was founded, so more than ten years ago. He was writing safety papers when he was a researcher at Google. And so this is deep in how he thinks about the technology long before there was any kind of commercial product, any kind of race, any kind of regulatory capture possibility. So no, he's been consistent. He's just saying what he believes.
00:16:38 · Calin Fabri, S0
[00:16:38] S1: Right. I mean, also, Musk makes the same argument, because if Dario Amodei right now, before their IPO, if they are saying that the probability is kind of high, that humanity is at risk, this is too much of a four d chess model, right? Because you just can't say that if you are just ahead of an IPO, meaning that he probably just generally believes that. [00:17:01] S0: This is the impressive thing about Dario. Even if you can't say that before an IPO, you know what he does? He says it. [00:17:07] S1: And and how, like, how would you regulate that? Also thinking that now you have, like, a global phenomenon. Right? If you regulate in Europe, US is going to go ahead in China. If you regulate in The US, China is going to go
00:17:19 · S0
[00:17:19] S0: ahead. Yeah. Look, I mean, I think the core problem in AI regulation is this thing you point out of you need to coordinate between The United States and China. I mean, I know we're in Europe, and I'm a European, but brutally, if The US decides that we're going to have AI regulation, Europe is going to just have to go along with it, because all European AI connected companies, they want to sell to America, they want to raise money in America, they want to be able to visit America, so you have to kind of pay attention to American law. ASML may not want to go along with the American chip export controls on China, but it has to. Right? So I think what this means is that Europe is going to follow, other parts of the world will follow, but China is the only country that cannot be coerced into following. So China and The US have to talk to each other about a joint approach to some of the safety stuff. And at the moment, you know, I go to The US a lot. I spend maybe more than half my time in The US. I lived there for eighteen years. I work at an American think tank. I have all these friends who do geopolitics and national security in Washington and New York. These are my friends. And so I talk to them all the time about, you know, we need to talk to China. And they're like, Sebastian, you don't understand. We've tried to talk to China. We can't talk to China. It's impossible to talk to China. And I say, you're wrong. I just went there in March. I talked to all the AI leaders. You know, they care about safety too. They don't want cyberattacks from criminals and terrorists either. You know, if you think it's difficult to talk to China, what about speaking to Nikita Khrushchev of the Soviet Union in the nineteen sixties? We had the Nonproliferation Treaty for Nuclear Weapons in the nineteen sixties, and we had the Cuban Missile Crisis. So don't tell me that you can't talk to China. You can talk to China if you want to talk to China. So we have this fight all year. I say I'm winning a little bit. Like, you know, Trump, you know, met with Xi recently. He's meeting him maybe two more times this year. There is a special dialogue between the treasury secretary, Scott Bessent, and his Chinese counterpart about AI safety.
00:19:29 · S0, Calin Fabri
So we are starting to see the beginnings of the conversation, but we're a long way from a deal where China and The US agree that, you know, there are certain things about controlling GPUs that we both should agree on. It's just not happening soon. [00:19:44] S1: One last question. It's like, if we focus on Europe, we know that DeepMind sold to Google. We know the reasons. You mentioned them in your book why that was just inevitable. What do you think is the opportunity in Europe for the entrepreneurs in Europe in AI? Because we clearly can't compete at a model layer, but arguably the model gets commoditized.