This looks crazy? 940 pages and the very first section 1.1.1 is empty. I can see why the author wanted to polish it. Let's just hope somebody clears this up somehow.
But at the same time Brian Conrad is presumably not a crank (per Wikipedia he worked with Wiles on FLT), but at the same time it seems insane to try and keep a 940-page result for a year hidden from public view in the current mathematical climate.
Somewhere around this summer, the world stopped being comprehensible...
Not true - the long proof is the verifiable one (because we have it in Lean), the short human solution is the one that can't be verified without a lot of extra work.
The Alpoge proof, if true, will certainly push mathematics forward a great deal. In the announcement, he lays out the general approach and says that he had 10+ "WOW" moments of insights while working with the AI on it. I have no doubt that there will be tons of follow-on papers clarifying and expanding many of the ideas that the AI explored. This will likely be the case even if the proof is wrong.
So, the work was done without AI, okay. But how likely is it that his colleagues, who were reviewing and advising him, used Claude to do so? Maybe even a year or two ago when they maybe didn't realize that this would end up in the training set?
> It is all human, with no AI. The hope was to bring Alessandro to Stanford for a year to polish the writing and then announce it after running it by enough people. Now it seems best to put out the current draft.
It's a little like the Patch Gap, but for math proofs.
I don't get this argument. Once you have verified it in "ugly" writing, why spend a year polishing it, when any minute an AI-proof could come around the corner.
Only seems to make sense when "polishing the writing" really means "filling the gaps". But then this is not a proof yet. Could still turn out to have gaps that cannot be filled as the approach is just not working.
A timely announcement like this makes me wonder: was the time just "right" for this, in mathematics? What proofs would have come out without AI, in the next few years?
Or, alternatively, did people start working harder in some fashion, even on creative problems that require deep insights, because they felt it was their last chance?
Or, alternatively, are people just (sensibly) starting to post all their work-in-progress results to establish priority (on either the results or approaches used)?
I think a lot of these breakthroughs do seem to happen when the time is just 'right' for them. Just enough progress in all related fields for a bunch of similarly intelligent people to be able to sniff out that progress on an old problem is now possible.
But, AI muddles this a lot given the complete lack of provenance from the companies about where their data comes from and secrecy on the methods and transcripts. So far almost every major result from them has had this issue, and in the absence of increasing transparency from companies, this just pushes me more and more towards believing that the companies are trying to cover up bad behavior in the pursuit of these announcements.
There's a whole term for this! It's called "Multiple discovery" - https://en.wikipedia.org/wiki/Multiple_discovery - "the hypothesis that most scientific discoveries and inventions are made independently and more or less simultaneously by multiple scientists and inventors"
Let's just the shareware approach -- release when you have like 90% of the stuffs but still need time to complete, and promise you will finish it in a year.
The linked proof has been in the works for 12 years and was expected to be worked on for at least another year and was announced early in light of the other GRH announcement.
I don't think anything fundamentally has changed except for the tension introduced by the very real possibility that if you have been working as a human on a math proof for a long time that AI pointed toward the problem you have been working on could scoop you at any time before you get your results out.
It could also be that several mathematicians were already close (after working for decades, like this post). Then AI got its hands on the breadcrumbs, merged them into the weights, and then laundered a path to a solution without a way to trace the credit.
This bugs me and feels much closer to truth than I think OpenAI and Anthropic are acknowledging.
I work as a staff developer at a research institution and it boggles my mind that researchers seem to uncritically use commercial LLM services instead of locally-hosted LLMs.
Could it not also be the case, that mathematicians improved due to AI assistance? Seems far fetched to me, that these highly competitive individuals would not reach for AI assistance in the last few years.
Yes that's probably also true, for the same reason that mathematicians probably improved with the internet.
I guess the main difference is that for a long time it was not acceptable to publish a paper with citations of "I dunno, I wrote this paper by putting literally everyone's paper into a blender".
You're implicitly assuming that frontier LLMs amount to little more than copy-and-paste. I think such a claim requires substantial justification.
I think the most plausible explanation is that LLM swarms are so good at synthesizing disparate information when fed in as context that they are racing (and beating!) humans to things that otherwise would have been solved SOON™.
Agreed. I wonder if mathematicians will become privacy advocates against ACR on TVs and phones transcribing what you're watching and phoning home? We need more vocal people decrying that it is dystopian to try to normalize such privacy invasions.
Do human mathematicians need to publish more incrementally? Seems many are hoping they’ll save up their findings for one big bang finding?
IMO academia could benefit from lower key, more tech blog style discussions of their work, and get away from this attempt to only publish the final result.
Is it common for mathematicians to keep a polished version of their work? I'm surprised by how quickly they are able to put work out after it's been contested, and this is 940 pages of well-formatted text, from the looks of it.
It's common for them to type it up as soon as they have some faith that the proof is valid, because that is the version they send to other mathematicians to confirm before the proof is officially published. Brian Conrad was in the process of reviewing this proof, so it makes sense that it's been written up. They were probably waiting to submit it for publication until multiple mathematicians confirmed its validity, but Anthropic forced their hand.
It is said that others have been reviewing this human-written proof for months, so the publication presumably was waiting for the finalization of the reviews, until the Anthropic announcement forced an immediate publication.
I sincerely hope this is right because I see inherent value in math as a human endeavor. Proving GRH, which we already believed to be true, has no conceivable economic implications: it doesn't improve the productivity of factories, it doesn't let you ship new products, it doesn't allow us to reach the stars. The result matters a lot less than the process of getting there.
Because of this, I find it repulsive that we use this endeavor as a benchmark for AI labs, where they throw tens of millions of dollars just to steamroll a small but prestigious academic field. The damage is already done, but I'm still gonna root for team human. I want us to carve some areas of human activity that LLMs are not allowed to take over. Math, along with art and writing, seems like reasonable candidates.
Can confirm that current GPT (with no guidance) took 5 minutes to find what it claims are errors with counterexamples. (I have nowhere near the level of state-of-the-art mathematical knowledge and experience needed to validate them. I'm posting this solely as additional confirmation of "there may be issues with this proof that an actual mathematician together with LLM assistance would be able to properly evaluate".)
I wouldn't be surprised if these models were aligned to extreme skepticism (to the point of hallucination) when e.g. a user proposes a solution to a Millennium Prize Problem.
Have you tried doing this recently? For my work on software at least, the latest LLMs have a very low false positive rate for bugs. They almost never hallucinate issues. My understanding is the same applies to math.
I would guess that LLMs are much worse at math without having a theorem prover to validate approaches with. Since this draft is not formalized, an LLM is operating purely on the text which is more tenuous.
I'm not arguing that LLMs aren't useful tools for mathematicians, I'm questioning the quality of discourse that amounts to "I pasted this into the LLM and it said there are errors".
There's value in knowing whether a claim is true or false. Plugging things into an LLM doesn't necessarily give 100% certainty or tell us why something is true or false in a human-comprehensible way, but it can move our credences and suggest directions for investigation.
It's being hurriedly announced as fully human work. I wonder if the author had carefully avoided ever running the full argument through frontier LLMs (OpenAI/Anthropic) for fear of it being stolen or incorporated into AI training.
This may be the first time this proof is being proof-read by frontier models
(this is pure speculation, but provides an answer to your question)
I also suspect the author actively tried to avoid using AI at all.
Another possible reason: AI will correct your mistakes by using ideas of others. A proof will unlikely remain human-only, even if you just use AI to check for your errors.
If this is real, and the results on all sides check out, the author made a HUGE mistake by not publishing a few months ago.
Other mathematicians should take heed. If you've got any ten-year-running projects, take them out of the shadows and into the light, or an Anthropic or OpenAI employee is going to solve your problem when they decide to bother with it and have enough spare compute at hand. And then you'll be scooped, with, at best, a "runner up" trophy.
I'd also add that the time is upon us where hobbyists can solve outstanding problems and prove theorems. The short-lived age of the professional mathematician may be genuinely over.
I don’t agree, because his work is considerably shorter and more comprehensible than the Anthropic proof. That will always have more value than some giant lean proof.
While I wholeheartedly agree that shorter and more comprehensible proofs are more valuable, I see no reason to think that LLMs won't _also_ start producing shorter, more comprehensible proofs before long
I'd value the meticulous work. I don't think this was a mistake, for one thing how could he have known? Secondly, mathematics is a human activity, whose reward is insight. I'll value the effort and insight -- or what is the point of priority, to prove you are better than the rest? In that case it's already an apples to oranges comparison. Well, it has to be a matter of personal philosophy on whether it's a mistake or not, I guess. Side note, for the record I have no objection in principle to resolving an unsolved questions using AI, the objection is entirely in what I find valuable besides the answer.
> mathematics is a human activity, whose reward is insight
If the author believed that they wouldn't have published now. They did it because one of the rewards for solving a problem is the recognition from your peers for doing so.
I disagree. Why would you feed years of your work to aid an LLM in beating you if you think you’re close to the answer yourself? Furthermore, many mathematicians have referred to the AI proofs as jumbled incomprehensible messes. Humans should continue their work unabated by the threat of some slop paper that claims a solve.
By publishing half a result, the mathematician guarantees an AI will have the chance to build on their work and beat them. How do you expect these people to get grants if all they do is publish half solutions that machines poach?
LLMs are not humans, they don't 'think' yet we use that term to describe what they do. Using the term poach to describe what AI companies are doing is not being done as a parallel to how that term is used by human academics
With that kind of thinking, why don't we all just give a few companies access to all our personal thoughts? That way we can all get partial credit from our favorite LLM company
This also ignores the fundamental question: how do mathematicians get funding from anyone for publishing half complete problems? They can't and it isn't enough to say academia must change. Academia doesn't change until professors die. You're decades away from this sort of change and the people who it harms the most [young academics] have no say in making the changes people are proposing
Sorry, most mathematicians don't want to look ridiculous by posting something that doesn't hold up careful scrutiny for something as big as RH. See the ABC conjecture and Mochizuki fiasco for example.
> made a HUGE mistake by not publishing a few months ago
Oh no, they'll lose out on some minor press.
Their work will still stand on it's own and for the people that it matters to, they'll know of him and his work.
The broader public probably wouldn't care much anyways, and solving math problems is going to lose a lot of glamour now that llms can do it to large degree. Not to say that it isn't impressive, but there's a large segement that just wants problems solved and doesn't care if it's done by human or otherwise.
I think this is severely under-selling the role of Levant Alpöge at Anthropic in that proof. I am not a mathematician, but he is, Princeton PhD and all. His work was a lot more informed and involved than a bored engineer with an unlimited token budget typing "plz solve GRH" and letting it run.
That doesn't mean the models aren't getting very good at math, but as in every other field, an expert pilot is going to get results a novice will not. Anthropic is hiring a lot of field domain experts and giving them space + compute to chase problems they were already interested in.
Edit to add: I also suspect that if validated, Masullo will receive appropriate recognition in the community for this work, and some grace for it maybe not being as polished as hoped.
Yay, humanity is back in the race! (If the proof checks out.)
It's insane though that he and Brian Conrad thought he had a year to polish the paper without getting scooped by AI. He should have put it out whatever he had on October 7th.
Will we soon learn that this human-made proof somehow made it to Anthropic pipeline and so their LLM later regurgitated (or at least worked off of) it?
As an academic, can I just say that I for one am happy that, irrespective of the pockets of social mayhem caused by AI rollouts, the acceleration of peer review and publication timelines in academia is a true benefit. I hope more scholars feel compelled to get their ideas out faster now that they might be scooped, like, tomorrow by the next frontier AI model. And I doubly hope that the humanities will speed along quicker, too, not just STEM.
I hope that taking the time to produce a legible write-up instead of releasing the proof up front doesn't bite him back. It would be great if, looking back in history, all contributions to solving the hypothesis were attributed fairly; but that's largely up to mathematicians themselves!
I wonder about the cause and effect. Is this only that this hurried announcement is released in response to Alpöge's publication today, or could Alpöge also have known about another mathematician whose work on proving GRH was close to completion — and felt compelled to release first? Especially given the statement in this draft that "[t]he math is complete".
Now there are two proposals to the GRH. One assisted with AI and one that is all human with ZERO AI.
Both should be put to the test and all of them should be rigorously reviewed, just like FLT and the poincare conjecture. This one is 900 pages rather than 250K pages.
You can see how you cannot assume anything yet as both are still claims. It starts to not make sense to do so with the top post in this thread [0] apply appeal to a singular authority:
"This is Levent Alpöge/Alessandro Masullo. If he says he has Riemann, well then, he has Riemann."
Rather than that, instead we should wait for multiple authorities to review both of them and the one that satisfies the GRH and is verified as the correct solution is the one that we go with.
> It starts to not make sense to do so with the top post in this thread [0] apply appeal to a singular authority
That's a bit uncharitable, I read that comment differently - it's telling you that the person who submitted it has a strong track record. That gives valuable context and credibility.
It doesn't matter how "principled" someone is, when it comes to over a trillion dollars no one is "principled".
It is very obvious Anthropic is trying to stabilize their stock. They will stop at nothing to ease concerns of investors, and it is apparent they probably stole results.
I don’t like it when people release work they say they’ve been on for a long time, right after someone else publishes the result. I don’t care whether the other person used AI. That’s going to continue and we’d better get used to it.
It’s annoying because if you’d been working on this all that time, you could have made some effort to tell the wider community and get at least some recognition. Posting after someone else has solved it is like claiming you had the winning numbers once the draw is already public.
Seems like people did know about this in the field… the linked post is not from the author, it's from https://en.wikipedia.org/wiki/Ravi_Vakil who is among other things the president of the American math society. And the paper has been reviewed for "months" by https://en.wikipedia.org/wiki/Brian_Conrad who has presumably been trying to make sure he understands it and helping to clarify the ideas.
It is surprising to me they wouldn't have tried to push out something faster after the kerfuffle last month. Maybe they were trying to do so, but it just takes a while!
> It is surprising to me they wouldn't have tried to push out something faster after the kerfuffle last month.
This is what's so surprising, and annoying, to me! Like you are sitting on a goldmine and still decided not to publish, even after seeing all that happen with Navier–Stokes.
Not everything needs to have daily standup syncs, or X threads.
Many in tech underestimate how much work, and knowledge, exists outside of public internet sources and within communities, exactly like the one Masullo had been interacting with and getting feedback from on the work being shared.
Did you read the post?
Clearly Alessandro Masullo was talking to people in the field. However, you cannot present all parts publicly when you are not finished yet. Otherwise someone might take your ideas.
Mathematicians are perfectionist, they only publish when they are sure it's correct. And Alessandro Masullo does simply not have the compute resources of anthropic, otherwise he probably would be finished earlier.
Wait until people realize what ancient Axial Age technology all LLMS/AI operate on regarding 'matrix' math. That's really going to grind alot of people's gears. There's decades of all sorts of people's work as the basis for how LLMs function, they are a big database of people's work over the ages with connect the dot Axial Age Matrix Connection methodologies. Most people and their work will go unmentioned as it always has in history. A very powerful technology that it doesn't seem most at the forefront of it are being honest with what it is. BTW, There's correspondence across scales as large as Galaxies. Might want to look up what ancient societies did stuff like that from that era. $$$$, alot of fools rushing in.
Nothing more than ego. If you claim to seek the truth, seek the truth. Don't fret if someone finds it before you. I don't care how hard you've worked. The truth is the truth.
Obviously this exact sentiment applies to the Anthropic announcement, so I'm not sure what the point is. If Anthropic cared about the truth and nothing else, then they would have waited until they had a much more succinct proof that could be understood by everybody in the field, so that everyone could be sure of the truth, rather than rushing out a collection of notes to be first in line. If they had no ego, they would have not even attached their (company) name to it at all. It turns out, everybody has an ego and it drives much of our behavior.
[1] https://en.wikipedia.org/wiki/Generalized_Riemann_hypothesis
But at the same time Brian Conrad is presumably not a crank (per Wikipedia he worked with Wiles on FLT), but at the same time it seems insane to try and keep a 940-page result for a year hidden from public view in the current mathematical climate.
Somewhere around this summer, the world stopped being comprehensible...
The Alpoge proof, if true, will certainly push mathematics forward a great deal. In the announcement, he lays out the general approach and says that he had 10+ "WOW" moments of insights while working with the AI on it. I have no doubt that there will be tons of follow-on papers clarifying and expanding many of the ideas that the AI explored. This will likely be the case even if the proof is wrong.
You can start trying to compress the formal result into something understandable.
Please read this article: https://xenaproject.wordpress.com/2026/11/12/and-the-proof-i...
Just because it's in Lean, doesn't mean it's correct, or that the proof actually proves what you think it does
Why hasn't Wolfram gotten into AI? This would be right up his alley.
I'm done.
It's a little like the Patch Gap, but for math proofs.
Only seems to make sense when "polishing the writing" really means "filling the gaps". But then this is not a proof yet. Could still turn out to have gaps that cannot be filled as the approach is just not working.
Or, alternatively, did people start working harder in some fashion, even on creative problems that require deep insights, because they felt it was their last chance?
Or, alternatively, are people just (sensibly) starting to post all their work-in-progress results to establish priority (on either the results or approaches used)?
But, AI muddles this a lot given the complete lack of provenance from the companies about where their data comes from and secrecy on the methods and transcripts. So far almost every major result from them has had this issue, and in the absence of increasing transparency from companies, this just pushes me more and more towards believing that the companies are trying to cover up bad behavior in the pursuit of these announcements.
I don't think anything fundamentally has changed except for the tension introduced by the very real possibility that if you have been working as a human on a math proof for a long time that AI pointed toward the problem you have been working on could scoop you at any time before you get your results out.
I mean, that's literally how AI works.
I work as a staff developer at a research institution and it boggles my mind that researchers seem to uncritically use commercial LLM services instead of locally-hosted LLMs.
I guess the main difference is that for a long time it was not acceptable to publish a paper with citations of "I dunno, I wrote this paper by putting literally everyone's paper into a blender".
I think the most plausible explanation is that LLM swarms are so good at synthesizing disparate information when fed in as context that they are racing (and beating!) humans to things that otherwise would have been solved SOON™.
IMO academia could benefit from lower key, more tech blog style discussions of their work, and get away from this attempt to only publish the final result.
Someone else scoops up and runs ahead with your work.
In an ideal world, where there is no ego and clout chasing, your proposed way is of course much better.
Agent swarms doing research for example don't have these ego issues.
I only write up in LaTeX in almost polished form, the unpolished part is in the pen and paper stage.
Because of this, I find it repulsive that we use this endeavor as a benchmark for AI labs, where they throw tens of millions of dollars just to steamroll a small but prestigious academic field. The damage is already done, but I'm still gonna root for team human. I want us to carve some areas of human activity that LLMs are not allowed to take over. Math, along with art and writing, seems like reasonable candidates.
Maybe repairable, but either way, alpoge/anthropic certainly has the claim to the first proof
AI claims but too few people can check.
Won‘t get better from now on
This may be the first time this proof is being proof-read by frontier models
(this is pure speculation, but provides an answer to your question)
Another possible reason: AI will correct your mistakes by using ideas of others. A proof will unlikely remain human-only, even if you just use AI to check for your errors.
Other mathematicians should take heed. If you've got any ten-year-running projects, take them out of the shadows and into the light, or an Anthropic or OpenAI employee is going to solve your problem when they decide to bother with it and have enough spare compute at hand. And then you'll be scooped, with, at best, a "runner up" trophy.
I'd also add that the time is upon us where hobbyists can solve outstanding problems and prove theorems. The short-lived age of the professional mathematician may be genuinely over.
Keep your work secret and you'll get scooped. Make partial progress public and they'll ingest it into the model and scoop you anyway.
If the author believed that they wouldn't have published now. They did it because one of the rewards for solving a problem is the recognition from your peers for doing so.
Publish what you got, when you got it, or get overtaken by someone who will.
With that kind of thinking, why don't we all just give a few companies access to all our personal thoughts? That way we can all get partial credit from our favorite LLM company
This also ignores the fundamental question: how do mathematicians get funding from anyone for publishing half complete problems? They can't and it isn't enough to say academia must change. Academia doesn't change until professors die. You're decades away from this sort of change and the people who it harms the most [young academics] have no say in making the changes people are proposing
This... what?
Oh no, they'll lose out on some minor press.
Their work will still stand on it's own and for the people that it matters to, they'll know of him and his work.
The broader public probably wouldn't care much anyways, and solving math problems is going to lose a lot of glamour now that llms can do it to large degree. Not to say that it isn't impressive, but there's a large segement that just wants problems solved and doesn't care if it's done by human or otherwise.
That doesn't mean the models aren't getting very good at math, but as in every other field, an expert pilot is going to get results a novice will not. Anthropic is hiring a lot of field domain experts and giving them space + compute to chase problems they were already interested in.
Edit to add: I also suspect that if validated, Masullo will receive appropriate recognition in the community for this work, and some grace for it maybe not being as polished as hoped.
It's insane though that he and Brian Conrad thought he had a year to polish the paper without getting scooped by AI. He should have put it out whatever he had on October 7th.
I wonder about the cause and effect. Is this only that this hurried announcement is released in response to Alpöge's publication today, or could Alpöge also have known about another mathematician whose work on proving GRH was close to completion — and felt compelled to release first? Especially given the statement in this draft that "[t]he math is complete".
Both should be put to the test and all of them should be rigorously reviewed, just like FLT and the poincare conjecture. This one is 900 pages rather than 250K pages.
You can see how you cannot assume anything yet as both are still claims. It starts to not make sense to do so with the top post in this thread [0] apply appeal to a singular authority:
"This is Levent Alpöge/Alessandro Masullo. If he says he has Riemann, well then, he has Riemann."
Rather than that, instead we should wait for multiple authorities to review both of them and the one that satisfies the GRH and is verified as the correct solution is the one that we go with.
[0] https://news.ycombinator.com/item?id=50574968
That's a bit uncharitable, I read that comment differently - it's telling you that the person who submitted it has a strong track record. That gives valuable context and credibility.
This one can't be verified with certainty on a computer (yet) while the other one can be.
It is very obvious Anthropic is trying to stabilize their stock. They will stop at nothing to ease concerns of investors, and it is apparent they probably stole results.
It’s annoying because if you’d been working on this all that time, you could have made some effort to tell the wider community and get at least some recognition. Posting after someone else has solved it is like claiming you had the winning numbers once the draw is already public.
It is surprising to me they wouldn't have tried to push out something faster after the kerfuffle last month. Maybe they were trying to do so, but it just takes a while!
> It is surprising to me they wouldn't have tried to push out something faster after the kerfuffle last month.
This is what's so surprising, and annoying, to me! Like you are sitting on a goldmine and still decided not to publish, even after seeing all that happen with Navier–Stokes.
Many in tech underestimate how much work, and knowledge, exists outside of public internet sources and within communities, exactly like the one Masullo had been interacting with and getting feedback from on the work being shared.
Mathematicians are perfectionist, they only publish when they are sure it's correct. And Alessandro Masullo does simply not have the compute resources of anthropic, otherwise he probably would be finished earlier.
Why did he publish now then?
> And Alessandro Masullo does simply not have the compute resources of anthropic, otherwise he probably would be finished earlier.
I did read the post and it says "It is all human, with no AI", so I don't think your point makes sense here.