Terence Tao a month ago:
"The AI companies have clearly... I'm sure they have secretly devoted much more time to the Riemann hypothesis than what they've disclosed. And they couldn't touch it. I mean something beyond current AI is needed. So maybe some combination of humans and AIs in 2050 or something will make further progress. Who knows?"
Really interesting to note from his timeline, he marks the starting point as the Stepanov determinant, which he earlier mentioned relates to OAI's "quiet inclusion of the Siegel argument ... I immediately recognized it as Stepanov without multiplicity". I believe this was one of results in their 700 solution dump.
I think this gives a partial counter to the arguments that this AI-driven math over obscure problems is a wasteful benchmark and doesn't help anyone. Turns out these are in fact important building blocks, and OAIs published results helped someone else do some great new math. I think that's nice.
These proofs really are quite huge. The proof is 1.3GB, ~15M lines of code:
mkdir -p riemann_proof && cd riemann_proof
# fetch parts 01–14
for i in {01..14}; do
curl -fSL -o lean.tar.gz.part$i https://alpo.ge/lean.tar.gz.part$i || { echo "failed: part$i"; break; }
done
# join and decompress (creates ./lean)
cat lean.tar.gz.part{01..14} > lean.tar.gz
tar xzf lean.tar.gz
# size, file count, line counts
du -sh lean | cut -f1 | xargs printf '%s on disk (lean folder)\n'
find lean -name '*.lean' | wc -l | xargs printf '%s .lean files\n'
find lean -name '*.lean' -print0 | xargs -0 cat | wc -l | xargs printf '%s lines total\n'
find lean -name '*.lean' -print0 | xargs -0 cat | grep -cv '^[[:space:]]*$' | xargs printf '%s non-blank lines\n'This title seems weird. Why just "anthropic employee", not specifying that hes a mathematician, or just including his name? Without the clarifications in the replies, many will likely assume this was a random Anthropic employee just throwing tokens at the problem, not a very well known mathematician working with the model. Probably not the intention, but these kinds of titles feel like they're trying to shift the credit from the person to the company and their models.
The way this is getting announced via a human mathematician humbly presenting a result is great, as opposed to the marketing influenced OpenAI post.
I am still a bit sad though. I had hoped the Rieman hypothesis would fall mostly to a human mathematician rather than 60/40 AI/mathematician that my layman's read of that note interprets.
Woah, if this proves to be correct it'd be massive, for maths, computer science...and AI too, I guess!
The worry I have right now is this though:
> The argument is incredibly long at the moment, and while its formalization is finished and it is routine to produce a huge, unreadable document containing sufficient detail, in my view of my path I have just taken a step.
this feels like we're moving away from the old world of 'science', where we could actually read an experiment, understand its steps, and, with enough time, possibly reproduce it.
And we're back in the world of 'magic' and 'belief'. Yes, there are formal ways to prove that this is correct, but if the proof is so convoluted and hard to read, is it actually something a human will ever find the time to go through and understand? We're going to start delegating (oh well, we're already doing that, I suppose) the understanding to AI, so it can just help us condense it down to smaller pieces that are digestible.
Still, very strange dizzy feeling, much more than Navier-Stokes
It's not a random Anthropic employee. It's claimed by Levent Alpöge. The guy is on fire and already has like 3 or 4 very interesting verified result this year. I've showing for months his wikipedia page to my coworkers in the math department in my university.
This is Levent Alpöge (https://en.wikipedia.org/wiki/Levent_Alp%C3%B6ge), a supremely qualified mathematician and one of the best in the world at guiding AI towards new discoveries in mathematics.
If he says he has Riemann, well then, he has Riemann.
Personally, I'm deeply discomforted by the idea that the few mathematicians who have aligned themselves with these companies are able to race ahead of the rest of the mathematical community like this. It is unrealistic to expect Anthropic to immediately share their monetary wealth with the rest of the world (though this is their stated intention in the long-run via UBI), but math is just math. Does Levent really need to take the crown jewel of open math problems for himself before the models are even publicly available? Did OpenAI really need to launch a swarm of 10,000 agents and scoop Tristan Buckmaster, a university professor who had worked on the Navier-Stokes existence problem for most of his career?
This is a time of deep uncertainty and anxiety for working and aspiring mathematicians. OpenAI and Anthropic gain little by announcing results like this ahead of model release. So if they can't restrain themselves on such economically useless things as pure mathematics, it does not bode well for the rest of us.
That was fast. I thought we’d have more time.
I wonder if we’ll have room temp superconductors and fusion this time next year.
If this holds up it seems as good a day as any to call the end of one time and beginning of another.
I wish I could be fully happy about this new era instead of worrying if I’ll be alive in 5 years or not. If there was any time for Anthropic to try to coordinate a pause, it feels like now or never, they seem to have a pretty strong lead at the moment. But probably they won’t and can’t, and we will just put the fate of our species to a coin flip.
Because these theoretical discoveries are coming thick and fast, it is getting to point where we are dismissing the gravity of some of these milestones. The next big milestone would be bridging this gap between theoretical discoveries and real world impact.
I am hopeful we will see cheap abundant nutritious food and clean water, more efficient solar batteries, economical space travel, cancer treatments, climate change mitigation etc. Most importantly, I hope people can still find purpose in their lifespans. Maybe a few centuries from now it would be thought of as barbaric that people we forced to sit on a desk for 8 hours a day just to afford a livelihood.
> In October 2024 I attended a workshop at Harvard University ... “An AI that can prove the Riemann hypothesis is not one I’d want to meet,” said Andrew Sutherland, a number theorist at the Massachusetts Institute of Technology. “If that happens, mathematicians having jobs will be the least of our problems.”
Scientific American, May 19, 2026.
https://www.scientificamerican.com/article/the-riemann-hypot...
David Hilbert:
> If I were to awaken after having slept for a thousand years, my first question would be: Has the Riemann hypothesis been proven?
What I found surprising. How many absolutely groundbreaking breakthroughs with AI are in Mathematics. And at the same time I can't remember anything similar in other fields. For example there's nothing groundbreaking in Software. Yes, Claude can write some PoC code faster than me. Great. What about creating some software which wasn't possible before? What about creating operating system without a single bug? What about inventing some video compression which is miles ahead of competitors? What about creating Rust compiler which runs 10x faster than rustc?
I think the closest thing is formal verification of Signal protocol, I guess. Which is insanely cool, but still not as cool as millenium problem.
Here is a crazy speculation: I think Anthropic has a Model 5 (more context on numbering: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleratio...)
I'm checking the timelines of the progress, initial progress is between Oct 6-20, then there is nothing for 3 weeks. Finally, everything is wrapped up in 2 days between 11/15 and 11/16? If Levent got access to Model 4 in early October or late September, then he could have made the initial progress with Model 4. The crazy jump in the last two days could have been with Model 5. If that's actually the case, the next few weeks are going to be crazy.
> The argument is incredibly long at the moment, and while its formalization is finished and it is routine to produce a huge, unreadable document containing sufficient detail, in my view of my path I have just taken a step. I intend to do my best to understand and impart understanding of this argument and its certain simplifications, to speak about it, to write it properly.
Okay, I am anxious whether this holds true or not, but he claims that the formalization is finished.
Why not just release the formalization in the meantime?
This is probably the biggest day in math in recent memory.
Edit: nevermind, the Lean formalization is linked.
This should be the wake-up call to pause development of models any bigger, enforced globally just like nuclear proliferation.
Even without considering rogue AIs, our society relies on individuals having limited destructive power. A dictator can ask the armed forces to bomb an entire country, or a lone terrorist can bomb a small number of targets, but no lone terrorist can bomb an entire country.
Here current AI has enabled one person to solve what is arguably the most famous problem in mathematics. What happens if a jackass uses the next AI to design a supervirus, or break AES, or develop an unpatchable hardware attack (a la rowhammer) that bricks computers remotely? The last month of non-stop vulnerabilities is just the tip of the iceberg as models improve and people have more time to work with them.
It's not too late, but we must stop now.
Is the length of the proof coming more from having a large number of branches that need to be analyzed and merged (like for 4-coloring problem), or more from having a large number of (new) intermediary results/lemmas falling in the deduction chain? I understand that we may find shorter versions of the proof with time.
A live compilation/check of the proof/Lean code http://106.14.11.164
Probably late to the party but I downloaded the source and tried having claude go through and create a manim explanation: https://github.com/harttraveller/riemann/blob/main/assets/fu...
https://news.ycombinator.com/item?id=50572355
The two stories were next to each other earlier and made me laugh, that's all.
Wow, so I guess this is what it feels like to be in the "foothills of the singularity" as Demis put it. I have a feeling of awe and dread combined that's hard to explain. Things are about to get very interesting indeed.
Anyone know if this was Mythos/Fable or some as of yet unreleased internal model?
“Often, great scientific achievements will be examples of one extraordinary individual’s ability to reduce a complex mass of information to a single pattern.”
— John Barrow, Impossibilty: The limits of science and the science of limits, 1998
(Presented without comment.)
One reason the formalized proof is so large is that the proof depends on a wide variety of previous deep results, and because the proof is self contained all those previous deep results are also formalized inside the formal proof.
Factoring out those results into separately formalized chunks would make the proof shorter.
I want to congratulate Anthropic.
Secondly, I want to quote one article, how easy it is for humans to misunderstand difficult things AI does, as something supposedly easy:
> They mostly came away from Claude Code with the sense that this is a genuinely useful tool that they didn’t previously know, but the gut-level reaction here is: “I never knew that building software was so easy” rather than “It’s very impressive AI can do this.” [1]
It should become very clear for everyone how to adjust their views. However I doubt many will, even after this, they need something physical, like humanoid robot to be impressed.
[1]: https://www.lesswrong.com/posts/ySXuvJcqRindQwAk7/how-my-stu...
And, thus at some point with 250K+ (and, below) page proofs, we ought to realize we are hitting an edge where we begin to be materially incapable of verifying anything: This itself is a sign that the complexity of the problems being solved - the scientific and methodical aspects of managing/engaging with that complexity - are beyond us.-
And, yet, at some point, the theory coming out of such efforts will transfer to material, engineering and design deliverables, and ... work.-
And then we will cease to undertand our technologies themselves. Our machines, our devices. Built on an understanding of the world we cannot reach, by machines we can build ourselves no longer.-
And so begin the Dark Ages of mathematics, I suspect.
Many people here have an idealized understanding of math as something incredibly profound. But it was mostly something that nerds practiced for the joy of it, in a nearly zero-pressure setting, the results rarely applicable to real life.
After the dust settles, I suspect there will be a lot fewer people interested in a job that now boils down to prompting and then staring at 250,000 pages of LLM output. And there will also be a lot less incentive for anyone to spend tends of millions in compute to produce these proofs just for the sake of it. I suspect we'll end up doing less math, not more.
Alpöge is moreover claiming the Generalized Riemann Hypothesis. The written argument is over 250,000 pages long and not provided.
The toolbox here contains a surprising, to me, amount of algebraic stuff.
Whether the proof holds,
The most exciting time, and the most dreadful time.
Glad to live through the 90s/2000s so that I can entertain my son a bit about human history. Double glad that I’m over 40 and closer to retirement than the younger generations.
He provides a lean proof in the first footnote. I downloaded it and took a look, he seems to use the RM conjecture defined by the community years ago, and he doesnt seem to introduce new assumptions (although to fully check this requires running the proof, which on my laptop will take days). Because of this, we only need to wait for someone to run the proof checker, and if it passes, then thats it, the only uncertainty left is bugs in Lean
I know this thread will be full of people in denial about full recursive self-improvement / intelligence explosion / superintelligence (the original meaning of the word) within months or a year, so just quoting Scott Aaronson:
"Experience has shown that, even now, there will still be people explaining in patronizing tones why none of this is real and none of it counts. If such people were capable of being impressed by anything that happens in the empirical world, of updating on anything, they would’ve already been impressed and already updated several years ago, long before things had reached the point of an actual Mathocalypse."
Does anyone know if the proof is constructive (perhaps telling us something new and usefull about the prime numbers) or is just establishing the fact the RH is true, but with no new insight (e.g. no new factoring algorithm falls out)?
No idea how much it cost, how many tokens were burned etc. M
For Navier-Stokes, it was, at market prices, something like 30m USD. A lot of money. I’ll assume it was more here, given the size of the proof.
It’s interesting for a few reasons. First, although I suppose you could spend the money on postsocs and PhD students (this is kind of 600 academic-year money) and maybe even achieve the same result over a few years, LLMs have accelerated this. You can spend the same effort/money in a few weeks.
But the other observation is that it is VC, and now public investor money as well they are burning, for no obvious reason other than marketing. Sure, this is sort of furthering human understanding of the universe, but is that what Anthropic investors want for their money? And there has been a lot of projects like this. What’s Anthropic’s total spend on their own research?
I find that a bit confusing. Why do they do it? Is there nothing better they can spend the money on, in terms of RoI? What do they think they are buying with it?
While I’m not in any way denying the power of AI, I find the relentless ramming of “AI is now everything” a bit offputting. It should be selling itself. There shouldn’t need to be an industry of influencers telling me to go all in on Claude and agents running agents.
So why are these firms spending so much money on proving the obvious?
if this is true (and i suspect it is), this will be the first agent/model result to surprise me. I thought it'd take another 2 years (at least) for reimann to fall. truly a remarkable time to be alive. i can only hope that we can all share the progress & accomplishment and not live in a world where you're not considered intelligent unless you work at one of 3 companies that are able to push scientific progress. i fear many of us will spend the rest of our lives as cattle performing the manual labor of distributing SI to the world.
Was this solved in a chat interface? It's not clear, but it seems like it from the document. The most relevant quote seems to be:
> Claude and I have had a great time together with such breadth in recent months, but this time Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line. I had us on the line of combining two quite different ideas present in a toy setting, one of which was already new and shocking, making very simple one of the shadows of Riemann over which we mathematicians felt most triumph. I guided us to the right situation analogous to that toy setting, and then Claude began to leave me in awe.
If indeed this was simple 1:1 chat, that's quite important, in and of itself, in my opinion. Presumably, OpenAI and others have been throwing autonomous agent swarms at this problem without success, but a single human + Claude seems to have cracked it.
We may be in the "centaur chess" era of superhuman intelligence.
> once we saw multiplicity 2 on O(2, n) via the "natively-automorphic" determinant ... meaning holo/automorphy over a function field
So if I understand correctly this proof has proofs for additional things that are steps to the main proof. That would be similar to the proof of Fermat's Last Theorem that required proving the Taniyama–Shimura conjecture [1].
[1] https://en.wikipedia.org/wiki/Fermat's_Last_Theorem#Subseque...
Note that the pdf contains links to the lean formalization
https://alpo.ge/lean.tar.gz.part01 through https://alpo.ge/lean.tar.gz.part14
Can someone who knows the math take a deep breath and decide if knowing the Riemann Hypothesis is true is useful? Like if aliens transmitted it to us and we had no hope of verifying it independently within 5 years.
Every time one of these AI proofs drops, people say the real value is the human journey, and the subfields it opens up, and the follow-on hypotheses it generates. Was that really why this finding had a million dollar bounty on it? Just the journey?
From the link:
It is natural to feel that this is an end, in my case a loss of a principal guiding light, but what I saw during the course of progress so amazed me that I believe it is one dawn across science of something far greater.
If the proof holds up, I suspect that will become a classic quote.
I hope we don't run out of interesting math puzzles to solve!
I've always enjoyed learning about these sorts of open problems (when they were open, that is), and the feeling of mystery and intrigue they invoke, of mathematics and of the universe at large.
It's like living at a time when parts of the world map were still uncharted - for now.
I'm wondering what kind of model and framework these guys use, and how much compute can they burn. The question is, can a mathematician these days keep up without being an employee at these frontier labs or having privileged access to their thinking machines? It's a bit disheartening to think of a future in which progress in knowledge work is conditioned by access to frontier AIs, which can think orders of magnitude faster than we can.
He casually mentions it also solves the Generalized Riemann Hypothesis and thinks it will take down the Extended Riemann Hypothesis.
Generalized: superset of the Zeta function.
Extended: I'm fuzzy on these details, but I think of it as Riemann over number fields, not just the complex plane.
I am elated to read about this, it's amazing. Sounds like there's some genuinely interesting and new math in there, which is also awesome.
Is there any hope of a mathematician actually understanding the proof? Millions lines of code and hundreds of thousands of pages for the proof seems unfathomably large to be actually comprehended by a human.
It's been almost exactly a year since GPT-5 first started solving Erdős problems. What a time to be alive!
I have never heard enough of the Riemann hypothesis. Based on the comments here, it sounds like a big deal for this to be proven as it validates many dependent proofs.
It is an impressive accomplishment. Im also wondering if there are any material changes to any non theoretical work as a result of this.
Does this make any algorithms faster? Does manufacturing become better somehow? Can engineers use this to make better designs of anything?
I genuinely know nothing about the hypothesis beyond what is mentioned in the first paragraph of the paper, so I'm having a hard time knowing how this changes real world things. I already thought AI models were more intelligent and capable tools than anything else. This proof didn't change that.
For the non-mathematicians and folks interested in real-world applications, what is the impact of the Riemann hypothesis being proven? Are there major breakthroughs in cryptography or other areas of computer science or physical science that flow from proof of this hypothesis that was generally already considered true?
Even when it turns out to be incomplete or wrong, it just shows how much sprawl we have in humanity's collective knowledge. Maybe we are just beginning to coalesce it into a manageable hierarchy, and if nothing else, that's worth spending time and money on, instead of staying in our knowledge silos and pushing forward without the full picture. A pessimistic look would assume that wouldn't be enough, but imagine what could be done once you can start with a clean slate and a clearer learning path.
There seems to be a completely human proof also: https://math216.wordpress.com/2026/11/20/hurried-announcemen...