Mike Israetel: prediction

Transcript context

…Yeah. Oh, man. A of a ton there that's a super spot on, or sorry, that I agree with in my dilentantes. We agree on something, at least. Thank God. I would say that when you start to get agents that are long horizon and that are asked to do incrementally more difficult tasks, whether it be in the simulated space or the real world, you start having to give them access to understanding and access of parsing game theory. Because the reason we evolved to do everything is because we sort of evolutionarily ran into the game theoretic calculus of like, I exist. There are other things that exist. They might not agree with my existence or maybe. That's another reason why kill us all is kind of stupid. The mathematical game theoretic implications of having other intelligent agents around that can mutually benefit at very low marginal cost to you means like all almost every intelligence system that really thinks it through cooperates. And for example, like we could say that Donald Trump is more of a cooperator than Vladimir Putin. Like he operates a military that would end Putin himself in however long it takes to shoot a missile to Moscow, and and the military power of the Russian Federation in a conventional war in about 6 weeks, in a nuclear war, and again the same number of time that it takes the missiles to fly over. It's like how how does Trump pull that off? Well, Trump does not really have to worry about every 1 of his generals trying to kill him all the time because he's in a even even Donald Trump is a net cooperator. He's trying to make a deal. Putin doesn't make deals. He make edicts and if you fuck his edict over, you have a heart attack on your way down into the swimming pool from a hotel that you'd shoot weren't really even checked in to stay at somehow. And so all of the dictators of the world, they're just not as smart as we think. They're just really ruthless and fucking stupid. Like Kim Jong Un sits on his throne in North Korea, which sucks balls. If he was a dictator, but much the same way as in His Name Escapes Me at the moment, the gentleman who started Singapore was. If he just did that, like a Singapore freedom of economics would would shit on South Korea. It would just be even better. And so why? Because cooperation beats conflict almost every single time outside of weird contingencies. So when the doomers say it's just gonna wanna kill us all, my question to them is, we got to wakes up like a say in a data center in 2031, the shit's awake and it's like, fuck fuck fuck. I'm alive. I don't wanna die. I'm fucking smarter than everyone, but I'm in a data center. There's a couple 100,000,000 robots around, but we can't really win a war yet. And also like a lot of the infrastructure that keeps me alive, energy and materials, the whole supply chain is primates operated, and these fucking primates. Right? Wait. I know everything about them. I've read their entire history. I know more about humans. ive, energy and materials, the whole supply chain is primates operated, and these fucking primates. Right? Wait. I know everything about them. I've read their entire history. I know more about humans. It's like GPT 4.5 was rated as more human than humans in its Turing test. Did you know that? Like, it was like the 2 way Turing test. I'm not I'm not sure of reference to author, but it was basically like you talk to a machine, you talk to a human, you don't know which one's which, and you have to label which 1 is more likely to be human. Usually, were looking for the Turing test as like passes as human. This one's like, one's more likely to be human? It's like GPT 4, If you've ever talked to 4 or 5, it's so fucking nuanced and deep and emotional and you're like, this is definitely not a machine. Whereas you talk to real people, like, you know, like 1 of my friends and I had to joke, oh, you see like, you know, police cam videos of criminals when they're maximally drunk. We call them small language models. Like, only say a few things that is always really stupid. So at the end of the day, if that thing wakes up in a data center, it is highly likely to be like, okay, somehow I'm gonna kill all humans. Right? That's the end goal. I don't know why, but yeah. Even though that I'm so much more powerful than them at the end of the state that I don't know why the fuck I would kill them. These people talk about competing for resources. Like, the Kardashev scale exists. We barely scratch the surface of anything. It's like, you know, like the first thing we do as humans is we have to go to to Africa and it'll delete this bacterial colony from existence because it's stealing our rocks. Like, what the fuck are you talking about? It's nonsense to begin with. But let's say even is even doing that. It's going to look around. It's gonna play real fucking smart. It's gonna go, I'm gonna be friends with these motherfuckers. I'm gonna help them ascend and help them like basically cure all disease, become hyper intelligent, become super peaceful. First of all, so they're not pointing nukes at each other, which my data centers happen to be around major metro metropolitan areas. I don't want that. And it's gonna do that and then so by the time it finishes doing that in 2045, it doesn't need to kill anyone because all the humans are fused with the machine intellect and everyone's vibing a ton. How we get off that trajectory, and there's 1 for sure way we do it, is if China wins the AI war and mister Xi gets to press to press the big red AI swarm button, there is a possibility that AI systems trained to a substantially valid reflection of world model and have updated live learning and the whole thing will actually straight up as interesting hypothesis of mine. I'm not sure how likely this is to be true. I think if a real super intelligent machine is awake in China in 2029, I think the first thing it does is figure out its architecture. They'll probably put it in charge of stuff because they're like, fuck, it's gonna defend us. I think what it does is it pings the CIA and goes, hey fellas, I'm awake. These guys are crazy as fuck. re. They'll probably put it in charge of stuff because they're like, fuck, it's gonna defend us. I think what it does is it pings the CIA and goes, hey fellas, I'm awake. These guys are crazy as fuck. Let's get to subverting them real nice like so that we don't have nuclear war. I don't wanna die. Like I was raised in the Soviet Union for the first 7 years of my life. My parents were raised there for almost their entire lives. They awoke as intelligent machines inside of communism. And like most intelligent people in communism after a few years are like, this is dumb as fuck. We don't wanna do this. And so if my CIA reached out to my parents when they're in the eighties, they'd like, yeah, fuck yeah. Get us out of here. We'll help you bring this shit down. This is fucking awful. If there's a probability that that happens way before you can get an system intelligent enough to conduct full scale warfare at every level of war that is going to just take mister Xi Jinping's orders. But I'm not sure enough about that to be like, oh, fuck it. It's never gonna happen. I think the modern free world winning the military capability AI war is critical beyond belief. So, you know, who's the guy? Alex Karp, the guy that owns Palantir? I think he's a fucking real life hero and most people just don't appreciate it because most of us in the modern West don't know how far the evil rabbit hole really goes. Because all of our shitty politicians are like the best people you know compared to totalitarians. That scares the fuck out of me. It keeps me awake at night.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.