logoalt Hacker News

trio8453 • today at 4:18 PM • 8 replies • view on HN

We need to give Levent a lot of credit for this. As much as we talk about Claude proving it, Levent seems to have a rare talent for getting the most out of it.


Replies

kingstnap • today at 5:15 PM

Should we though? Is working at anthropic and using unrestricted unreleased models really fair?

If everyone got to do that then we would really know how rare his talent is. But we don't, and he doesn't post his prompts or workflows. So how can you be so sure.

If he was at least posting his prompts we could acknowledge that. But honestly I'm not going to give him the benefit of the doubt on this. Highly accomplished and successful yes, but uniquely talent who knows.

➕ show 1 reply
riemannzeta • today at 6:06 PM

There were certainly others at OpenAI and Google that were actively searching for a proof with frontier models. That he got there first is a credit to the intuition he had built up from decades of work on related proofs and deep understanding of related mathematical toolsets known to be relevant to different parts of it. That much is clear from his 4 page remark.

jeremyjh • today at 5:09 PM

Yes, I would love to have access to his conversation logs. And it’s a real shame the reasoning traces are encrypted even if he shared them.

I wouldn’t understand any of it but I might catch some glimmer of what he brought to this. My intuition is that is quite a lot.

isotypic • today at 8:34 PM

I mean, do we? For all we know all he has done is had access to unlimited compute on higher power models than available to the general public. His general complete unwillingness to divulge any information about any of the results he has obtained (AFAIK, we are still waiting on any details for the Jacobian conjecture counterexample or high rank elliptic curves or Hadamard matrices or S^6 complex structure or...) combined with the general ability of current models to make progress with minimal prompting means the burden of proof is on him.

If he truly did have non-trivial mathematical input to this it should be trivial for him to give talks on at least the overall strategy of the proof. Before LLMs this was a bare minimum expectation for basically all mathematicians and their results, and I see no reason why we should give up on it now, especially so since the previous primary artifacts (the papers themselves) are now on their way to being completely automatable.

➕ show 1 reply
alethic • today at 8:09 PM

What? He just works at Anthropic, lol. Any monkey with a PhD in mathematics can do what he is doing given access to Anthropic's internal models... and they'd probably be doing a better job at exposition, too. Alpöge seems to be uniquely uninterested in explaining his results, for someone who calls themselves a mathematician. His approach has been pretty clearly harmful for the social structure of mathematics.

➕ show 1 reply
kurtis_reed • today at 7:13 PM

For all we know he just said "prove the Riemann hypothesis" to an internal model and got approval for the expense

➕ show 2 replies
riemannzeta • today at 6:03 PM

To that point, it seems that he enjoyed interacting with Claude. Sometimes I wonder how many of the bad experiences result from what Jung would have called projections of the user's self onto the AI through their prompts.

➕ show 1 reply