Mike Israetel: prediction

Transcript context

…We'll talk about that later. But you know, like for example, I'll I'll train with her. And ironically, what I get out of that has got almost nothing to do personal training. She she's an incredible woman. Really, really amazing woman. You know, sometimes she taps me on the shoulder. We laugh and we joke together. And it just enriches my life in so many ways that are not captured by these abstractions. So all of us have different abstractions. But then you're also talking about this kind of monolithic view of AI that understands everything. That's not the way that our evolution works. Evolution is all about this path dependence I was talking about. And we have many agents. And we all take our own paths, which means when we do creativity, we continue the epistemic lineage that we are on. And this is this unfurling process. What is not the case is that there's this kind of gray goo where all of the knowledge is convolved together and creative steps could happen from any part. I think any future super intelligence that we do create will resemble our intelligence. So it'll be very kind of diverse and and collective. I think at first that's the case. Then when super intelligence is smart and capable enough through a few different ways to rearchitect itself, it's going to leave behind our abilities quite quickly. And it's also going to leave behind most of the very narrow conceptions that we have about intelligence. I think Andre Karpathy, if I'm saying that correctly. You know, read enough things on Twitter and you're just, don't know how to say that person's real name. Said recently that something that kind of attends to your view a little bit. He's like, you know, we humans I'm very much paraphrasing Andre. I'm sorry if you ever see this. It was an ex post, know, as far as those go. You know, humans, we sometimes think that like we are somehow the seat of intelligence and that our cognitive domain is like what intelligence really is. But we're learning with AI according to him that like we have a niche in intelligence that's like to your point, path dependent evolutionary derived. I was gonna say highly functional, but functional enough to get us here and build all this cool stuff and build cameras and stuff. But only occupies a very small fraction of of the entire universe of possibility, the whole vector space of intelligence. And most of vector space we actually don't know. And so 1 of my things and and I'm not saying you're saying this at all by the way, it was curious to me when people say, I don't think because this used to be a majority view and even AI research 20 years ago, is that machines will never be more intelligent than humans. And like the gall to think that fucking primates who mostly scratch their testicles and sometimes do theoretical physics have some kind of like global optimum and inclusivity on all of intelligence, it's fucking preposterous. And I think when artificial superintelligence with agency truly kicks off, I think it'll be a miracle when it does. It's going to recursively re architect itself and start expanding into all of the domains of intelligence we don't know anything about. And then we will see really what intelligence is and just to use that really quick just by analogy, like if wolves could talk and even abstract, which they can't, but let's give them 2 magical abilities, they would pick like the smartest wolf in their wolf pack and be like, you know, Bob the wolf. He's a fucking man. He shit he fucking knows. Right? He knows when the bears are around, he knows when they're not. And then he'd be like, you know, okay. Well, here's like, I don't know, Stephen Hawking. Be like, so he's your Bob. Like, yeah. You know, in a sense. Like, okay. So like he really kind of knows where the bears are, but like he does a lot cooler shit than that. And you try to explain to the wolves what that is, and there's no idea how to contextualize that whatsoever. I think that's the flavor to use a very technical term. Though I did just learn that in AI training taste is a technical term and I was like, what the fuck? idea how to contextualize that whatsoever. I think that's the flavor to use a very technical term. Though I did just learn that in AI training taste is a technical term and I was like, what the fuck? And it was explained to me and I was like, I'm not smart enough to understand that by a long shot. But the the flavor of artificial super intelligence is verticality of ability, like doing math at this level, this level, that level, and then expansion across domains to domains we understand and like orders of magnitude more domains we haven't even conceived of.…

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