This AI-generated proof OpenAI has come up with does not strike me as very impressive. From what I’ve read, the process involved running 10,000 “AI agents” for 88 hours. Collectively, that’s 880,000 hours of runtime on complex statistical models - Probably well over 100,000 hours of runtime on advanced GPUs. It doesn’t strike me as much of an advance. Given that sort of computational power, I think it’s likely that other ML techniques could accomplish the same task - probably more efficiently, since they aren’t dealing with the overhead of language processing. The biggest change is that there is now a company, OpenAI, with the resources and commercial incentive to do this. OpenAI hopes to use this output as a demonstration of supposed “super-intelligence” to the general public. Strikes me as a more of a stunt.