Dean Ball: recommendation

Transcript context

…Well, one thing I wanna say, when I, when I was, I was, I was talking about competition specifically with respect to like, um, like what I will be doing, which is to say like, like, you know, our... The goal of my team will not just be, like the goal of my team will be to have like really, really fantastic intellectual output. That is like, wow, this team, if it were like not part of OpenAI and it were just like its own like little think tank, it would be like one of the most interesting think tanks in the whole country, right? And like everyone would be paying attention to it, right? Like I actually just want it to be like a really superb team that is producing, um, you know, policy and sort of public interest related work, um, that is, that is as good as anything else that, that, you know, any of our, any of OpenAI's competitors put out. That's what I meant by competition, to be clear. And I think there is a good example of straightforwardly healthy competition. When it comes to RSI, I think, yeah, I think there's certainly a lot of unknowns here. I, I put, my base case for what RSI means, at least in the earlier innings, is something to the effect of, one thing, it's like RSI, recursive self-improvement is a part of every technology. Every general purpose technology has some aspect of recursion to it, by the very nature of generality, right? General purpose means one of the purposes to which it can be applied is itself, right? So in some ways, I don't see recursive self-improvement as some big break from the history of technology. I see it as being actually like it would be surprising if there weren't recursive self-improvement in AI, right? I also think, like we've been doing recursive self-improvement in this field for a long time, and some people imagine a Things is, there's gonna be some big break moment where you could argue really since GPT-4, we've been using the models to make the models better since at least GPT-4. And I'm sure that if you actually went back and looked through the history of machine learning even more, I bet you that actually goes back even further than that. So some people imagine there being this sharp discontinuous jump in terms of what RSI means. And I certainly think that is plausible, but I don't... I'm a little s- it's not my prior, because your prior should just generally be that there is alwa- I'm sure on this podcast I've said before, that there is always more continuity than discontinuity. This is always the case. Um, and so your prior should be, like, against a massive discontinuous leap. It is plausible, and so I think step number one, and I can't, I can't... I'm not in yet, so I don't actually know because I haven't looked at the roadmap yet. But step number one would be to try to really measure twice and cut once, and figure out what we think this might mean. ot in yet, so I don't actually know because I haven't looked at the roadmap yet. But step number one would be to try to really measure twice and cut once, and figure out what we think this might mean. Tr- really try to refine the probabilities there, at least in my own mind, of are we talking about a discontinuous leap that happens very soon, or are we talking about something that actually is smoother in some way? Then I think if we are in... I think the chances of discontinuous leap are high enough no matter what, that you need to be planning now, even if your, even if your credence that it happens is 20% or 10%. That's high enough that you should be making plans now for what you wanna do. nough no matter what, that you need to be planning now, even if your, even if your credence that it happens is 20% or 10%. That's high enough that you should be making plans now for what you wanna do. And there, uh, we get into inter-lab coordination on things like, there's the slowdown pause thing, right? There's what would actually be the mechanisms of that? Under what conditions would it be triggered? Again, I'm pretty skeptical of such notions. But I think as a... There's a lot of policy planning I did for the US government that I wrote down and put places, but did not... It's not in the action plan. It's scenarios, right? We gotta be prepared for a wide range of scenarios here. And so similarly, I think there's gonna be some aspect of that where we, and we need to be ready to... We need to have advanced thinking on all that stuff. And so it's two things. It's okay, at what point... What are some triggers that we can set in advance for, is this gonna be a discontinuous leap specifically, and how can we refine that question to make it as specific as possible? And then in the event that those triggers happen, what is it that we would do? At what point do we go to the government? It's worth noting one proposal that I'm, like, a fan of. Or I'm a fan of the FTC, the Federal Trade Commission, writing what's, what would be practically called a no action letter. Where the FTC would write a letter and they would say, "Look," like they'd put out public guidance that would basically say, "Look, if you guys coordinate for these very specific reasons, we're not gonna consider that cartel behavior, and we're not gonna, we're not gonna enforce that." Um, I think that's, like, plausibly a good step to make that at least opens optionality. Though I also understand that, like, the way you scope, you have to be really careful about how you scope that, because look at what Anthropic did, right? Anthropic undermined the case for this dramatically just with the fable safeguards. With, "We're gonna degrade your outputs in the name of safety," which is, like, very clearly, like, a consumer protection violation. Just like very clearly. If... Can you imagine if a cartel of AI companies in the name of safety agreed to collude to degrade outputs in particular areas? That would be, like, wildly anti-competitive, right? And so you have to scope it really carefully, and also probably, like, companies need to be very careful about what kinds of things they do that undermine the case that safety is something we should be making these considerations for. But yeah, no, I don't have a lot of specifics to share on the RSI plan itself, for the simple reason that, like, I'm not in there yet, and I don't, I haven't, I haven't had those conversations yet.…

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