Noam Brown#30
I think people underestimate the effect GPT-5.x is already having on research in mathematical fields. 5.5 Pro has been proving many theorems for me, but I don’t talk about it much because I want to publish those results with my name on it and that’ll take time to write up.
Positive users praise GPT-5.5 Pro for proving many nontrivial theorems in math research while negative users criticize its imprecision for academic papers and foresee future problems.
I don’t really expect that I’ll do many original proofs anymore going forward, similar to people not coding anymore. I hope I don’t run out of questions to ask, but I already struggle coming up with new ones at sufficient speed.
Many of the key mathematical ideas in the last paper I’ve written were already contributed by 5.4 and 5.5 (which we have acknowledged because for now this is an interesting thing to say; soon it will be obvious and cliché) https://arxiv.org/pdf/2605.06194
At the moment papers still need to be 99% human written because models are so bad at English. But once they are superhuman at that too, it’ll be time for another existential crisis.
@DominikPeters most people just hate ai and don't want to look. the proof is lean formalized, maybe not the whole thing is correct, but there is clearly value here
@DominikPeters Do you mean people don't like their English?
I think objectively, they're fairly decent at English.
@DominikPeters How can we verify the mathematical proofs generated by AI models? I’ve observed that AI models incorporate certain mathematical concepts from the 1900s and apply them with recent research, even when there are studies highlighting the limitations of those older concepts.
@DominikPeters Hey, do you see a big gap between Plus and Pro?
@DominikPeters cool
@DominikPeters Idk I guess I just don't see how this is much different from how it's affecting other industries - using it to help navigate a massive corpus in an amount of time that is much shorter than we could ever do it ourselves.
@DominikPeters would you please take a look at my work? this was a long collaborative effort with 5.4 and 5.5 https://github.com/alegator-cs/infinite_twin_primes
@avg_wrng_ans @DominikPeters From my experience, they are too imprecise in their formulation for an academic paper.
@DominikPeters Agree, I think 5.5 Pro still needs some 'taste' of math, but it proved so many nontrivial things
@DominikPeters What do you think your role once they become superhuman?