00:22:37 · Ruby Justice Thelot, Herbie Bradley
[00:22:37] S1: Right. [00:22:37] S0: From the labs of exactly what types of productivity uplifts they're seeing. Can we get more high resolution employee surveys? What are the latest unreleased models? I think in August, Anthropic had a system card thing where they reported, oh, we have this unreleased model, and we have seen this amount of productivity uplift from it. The responsible scaling policy in this case has, I believe, this threshold that they refer to in their S-1 of a two x overall productivity uplift, meaning they're making progress two x faster than they would be, I think, pre AI with this latest ML model, or I should say they expect to achieve this threshold with a near future model or imminently with the term I believe in the S-1. And so that 2x basically translates into, if I recall, around the kind of 10% researcher productivity improvement per unit of the Epoch capabilities index benchmark, which still, there's various different people trying to model RSI. It translates into an R of around 0.5 if you have some scale where like R equals one is a fully self sustaining acceleration loop. [00:23:48] S1: Right. So wouldn't that imply that we don't have a fully self sustaining acceleration? [00:23:53] S0: Yeah, that's correct. But they did make this threshold as concerning enough to report as an RSP thing. Right. [00:24:02] S1: What do you think it would take to get from where we are now to full self sustaining acceleration? Why do you think Anthropic thinks that they will go from 0.5 to R1 or greater?
00:24:02 · Ruby Justice Thelot, Herbie Bradley
[00:24:02] S1: What do you think it would take to get from where we are now to full self sustaining acceleration? Why do you think Anthropic thinks that they will go from 0.5 to R1 or greater? [00:24:16] S0: I expect just having internal theories about exactly how different aspects of the research pipeline can be automated. You know, one example is John Schulman, I believe, on the podcast, proposed that researchers could sort of distill their high level taste and knowledge about experiments into models by doing LHF or sort of becoming human raters for different ideas. I'm sure Anthropic have a whole roadmap internally where they have gamed out a bunch of this stuff and kind of see the rate of improvement and see that it's getting there. I mean, we know that they published this blog post a few months ago with the graph of different levels of RSI and sort of making progress towards more and more autonomous research. I personally still think it's quite likely that even if you do get a period of self sustaining acceleration or you can in theory close the whole loop, that it will be desirable for various reasons, including to avoid good hearting or to avoid safety issues to keep some researchers in the loop, because at some point, the speed of the overall process doesn't end up being bottlenecked by the researchers. It ends up being bottlenecked by compute Right. The time it takes to do experiments and data and this kind of thing.
00:24:16 · Herbie Bradley, Ruby Justice Thelot
[00:24:16] S0: I expect just having internal theories about exactly how different aspects of the research pipeline can be automated. You know, one example is John Schulman, I believe, on the podcast, proposed that researchers could sort of distill their high level taste and knowledge about experiments into models by doing LHF or sort of becoming human raters for different ideas. I'm sure Anthropic have a whole roadmap internally where they have gamed out a bunch of this stuff and kind of see the rate of improvement and see that it's getting there. I mean, we know that they published this blog post a few months ago with the graph of different levels of RSI and sort of making progress towards more and more autonomous research. I personally still think it's quite likely that even if you do get a period of self sustaining acceleration or you can in theory close the whole loop, that it will be desirable for various reasons, including to avoid good hearting or to avoid safety issues to keep some researchers in the loop, because at some point, the speed of the overall process doesn't end up being bottlenecked by the researchers. It ends up being bottlenecked by compute Right. The time it takes to do experiments and data and this kind of thing. [00:25:23] S1: Well, we're coming up on time. There's so much more to get to, but we're out of time, unfortunately. Herbie, thanks so much for joining us on MTS. [00:25:30] S0: Thank you. Very fun as always.
00:25:30 · Herbie Bradley, Ruby Justice Thelot
[00:25:30] S0: Thank you. Very fun as always. [00:25:32] S2: A huge thanks to MTS sponsors. Blitzy, autonomous software development for enterprise code bases. Ship five x faster. Blitzy.com. Adquick, make your brand a billboard out of home advertising as easy to scale as digital. Adquick.com. Arena, measuring AI performance in the real world. Arena.ai. Support for the show comes from VCX, the public ticker for private tech. The US stock market started history's greatest wave of wealth creation. From factory workers in Detroit to farmers in Omaha, anyone could own a piece of the great American companies. But today, our most innovative companies are staying private longer, which means everyday Americans are missing out. Until now. Introducing VCX, a public ticker for private tech. Visit getvcx.com for more info. Getvcx.com. Carefully consider the investment material before investing, including objectives, risks, charges, expenses. This and other information can be found in the innovations fund prospectus getvcx.com. This is a paid sponsorship.