Extrapolating the AI 2027 recursive-self-improvement model, superintelligence-like capability is likely to emerge around 2031, and Anthropic looks like it is pulling ahead due to its internal AI-driven research feedback loop.

102:10year ago. And we built a model of R&D takeoff speeds under the kind of the core AI 2027 scenario where the main way things get crazy is that AI is used more in the development of AI. First by achieving the superhuman coder milestone and then the superhuman AI researcher milestone. And I am unhappy to report that I think that story is generally correct. Um I don't know if the

102:34timelines are exactly right. Um but I my forecast from that process of leading to something that looks like super intelligence around 2031 is roughly stable. I think the things that have happened in the year since AI 2027 come out very much vindicate the theory that the most important thing going on is how useful is AI and improving the productivity of AI researchers within Frontier Labs. uh I've made public

102:59predictions that I thought Anthropic was going to run away with it because they had the best feedback loop of talent and actually using their AI internally. I think that has been you know n equals one but I think it's been totally shown that that's been happening recently. So I think that will continue to happen. So the question is what is the role of forecasting in this? And you can just

103:17imagine what was going on with Dario with Tom Brown like going to the White House trying to get Fable unbanned. What were they doing? what uh moves did they have like for the US government for what it's trying to do? Again, even if you're incredibly cynical and you assume the US government is trying to just kind of score political points, what are those political points and what moves do they

103:35have to do it? There's all these kind of intergovernmental organization things and negotiations going on. To what extent is Anthropic, a company that is working on AI, able to use that AI in order to negotiate a better outcome with politicians? That to me sounds like a conditional forecasting question. if I go and write this letter, if I propose this export control, if I weaken the model, what is the probability that this

103:59person would relent on this thing within the next three days or something like that? So I would say again we don't have any evidence that that's happening. There may not be any evidence that that's happening, but I think in a model of 2029, whether AI forecasting is helping forecast better decisions that is leading to Frontier Labs doing things more effectively is very similar to the way that their improved coding is