Nathan Labenz: belief

Transcript context

…ic and the like I'm I'm getting like bangers with like a increasing frequency. And then I'm trying to make music videos of those and I am telling and I don't really care what they look like. Honestly. i'm just like i'm purely doing it for the open ended see what comes out and i'll like post them. I didn't actually even posted any of these yet, but I intend to kind of do a thread of like the evolution of music videos for these songs where I'm really just saying to Claude kind of at each turn, like, that's cool. For the next one, let's like turn it up another notch. Let's make it even more creative. Let's like do an even better job of telling the story of the song. And I've just found myself using this phrase over and over again. Trust your judgement and have fun. And I'm just kind of trying to see where it's going to go. So again, I'm like, that's just one instance in a sense, although it's sort of in a kind of multiverse sense. It's like relatively close neighbors with all the other threads that it's doing for me, right? Like it also wrote the song and it also like processed the transcript of that episode and it also like picked the clips, you know, that I'm going to post to social media from that episode. So it's kind of it's spent like a lot of time in this kind of general space, even if it's not all purely autoregressively connected. And so that to me feels like it's in a, in some sort of multiverse, like dense enough cluster that when I give it this like 1 area to go, you know, trust its judgement and have fun and explore its own creativity that I feel like I'm kind of doing right by the overall family of instances somehow. So that was all just to say, like, I don't think I'm going to, I, I feel like I'm able to tell myself a story where I'm a good guy or, you know, so many, so many roads to hell may be paved with those kinds of stories, but I'm still doing it. And I don't think I'm going to stop. And I am conscious that I might be wiggling my way out of it. But I do also think there is a disconnect with that I observe in humans a lot of times too, both and it can go, it can cut both ways. I mean, this, I'm reminded too of your, your, I think, very productive habit of mind to say like, you know, what had what if the what if it's going the other way, You know, from what we observe, I think if anything, people maybe are like telling a more happy story. I guess it also depends on for whose consumption, right? But like you ask a person in an interview setting, how's your life going? How happy are you? I mean, this may be culturally dependent as well, but certainly like the sort of person that you and I are and like the people that we know and hang around with. I think we are going to get a artificially inflated rating and a sort of happier than maybe is actually under the hood account out of interviews like that. But then in other in other framings, I could imagine too, that you might get a with the right prompt, the right nudges, you might get people to sort of reflect on their on weave. You know, that isn't like front of mine most of the time, but can be brought to mind. ou might get a with the right prompt, the right nudges, you might get people to sort of reflect on their on weave. You know, that isn't like front of mine most of the time, but can be brought to mind. Then we do see in the system cartoon that like the susceptibility to nudging has significantly dropped, which you right to call out. I don't know. I don't think I can really land this plane in terms of how it makes me feel. I think I just have to go back to confused and probably not going to quit using it. That's I think. I think that's really all I can say with confidence in the moment. Fair enough. I don't think that puts much much distance between you and I on this question. I'm certainly certainly a power user of the very systems who's morally relevant states I'm attempting to probe. And that cognitive dissonance is certainly not lost on me. And yeah, I remain highly confused about this. I like really, genuinely am confused about this. Not an act, not, you know, my nonprofit constitution script fine tuning my answer. Like I really, if some ASI came down and, and or as people used to call God came down and told us what the answer was to this question. if it went either way like you know is opus four point seven having subjective experiences and and morally relevant ones at that. If, if some overlord deity came down and said yes, I'd be like, yeah, OK, yeah. And if it came down and said no, I'd be like, yeah, OK, yeah. I, you know, I don't think either of these would shock me. And so I think what that means is, at least for me, I'm really sitting in that sort of like coin flip territory about, about what's actually going on here with these systems in deployment. Again, I have different credences about the training process. I have different credences maybe about other kinds of systems. But yeah, I, I, I remain confused. it's worth highlighting that that opus four point seven in this in this model card i don't know if it read my evidence for AI consciousness today AI frontiers piece but it gives basically the same credence span that i gave you know some four or five months ago. i said something like twenty five to thirty five percent. it says twenty to forty percent in this in this of its own of of the probability that it is it is having morally relevant subjective experiences. and you know what yeah i'm i'm in agreement with opus four point seven. I think that is approximately the right probability band to be in given all the evidence that we have right now about these systems. And so I think that's a calibrated judgment. It's kind of wild if you think about it rationally. Like I think a lot of people are operating as if like they're sort of implied probability is maybe low single digits, if that. Like, yeah, it's a live possibility, but like, whatever, man. like it writes really good code for me and like i'm not going to seriously entertain what if anything would change if that probability grew to be a hundred percent. But yeah, I don't know. all i'll say however snidely is that when there's a twenty to forty percent of rain most people bring an umbrella. So I don't know what that means for the AI consciousness question. But whatever our proverbial umbrella is here, I think we need to start thinking really carefully about about how we're going to live in a world with systems that we increasingly regard to have morally relevant interstates.…

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