Newshot

The stochastic parrots have apparently proved the Riemann Hypothesis

Original text(via Hacker News)

The stochastic parrots have apparently proved the Riemann Hypothesis

“… it is very probable that all roots are real. Of course one would wish for a rigorous proof here; I have for the time being, after some fleeting vain attempts, provisionally put aside the search for this, as it appears dispensable for the immediate objective of my investigation.” –Bernhard Riemann, 1859

I’m in California to visit Oratomic and then Inkhaven. I woke up this morning to the news that Levent Alpöge says that he’s gotten Anthropic models to prove the Generalized Riemann Hypothesis, producing 17.4 million (!!) lines of Lean code.

At a dinner last night, someone asked me “so, how much longer until an AI proves the Riemann Hypothesis? Still several more years?” I replied, don’t be sure of such things anymore. After all, AI has already proved that the nontrivial zeros of the Riemann zeta function are at most 3/8 away from the critical line (Quanta just published its article on that breakthrough yesterday!), and that already carries most of the implications for number theory that a full proof of RH would have. So who the hell knows what else is possible, and how quickly?

For those tuning in from home, the Riemann Hypothesis says (in its easiest-to-understand version) that the number of prime numbers from 1 to n is very close to what it would be if each integer x>1 had an independent 1/ln(x) probability of being prime, up to a standard statistical error term that never exceeds Õ(√n). While we’ve known since 1896 that the number of primes from 1 to n is asymptotic to n/ln(n) (that’s the famous Prime Number Theorem), you could’ve imagined the deviations being much larger.

David Hilbert famously said, “If I were to awaken after having slept for a thousand years, my first question would be: Has the Riemann Hypothesis been proven?” Meanwhile G. H. Hardy, on a ship that he feared might sink, sent his colleagues a telegram claiming to have proven RH, so that if the worst came to pass, at least the world might believe that a proof sank with him.

Many will of course question whether the proof announced today will stand. They’ll point out that, despite how heavily the world is now “lean-ing” on it, the Lean formal verification system still has issues, from kernel bugs (which have, in the past, allowed proofs of 0=1) to uncertain soundness of the underlying type theory. In my view, these issues can and should urgently be fixed! Until they are, a sufficiently motivated AI could almost certainly exploit them to prove a false statement!

All the same, I see zero indications that that’s happening right now. I note that, so far, not one of the AI results with Lean certificates has needed to be retracted (while several without Lean certificates have needed to be). So I’m not 100% sure, but my bet is that the proof will stand.

Mathematicians of course won’t be satisfied with 17 million lines of Lean. They’ll be satisfied when, and only when, they’re able to explain and distill and reconstruct the ideas for themselves. And that will probably take years, since human mathematicians move only at biological speeds. And while they’re working on it, how many more such wonders will there be, in this age of wonders and terrors?

My 9-year-old son, informed that RH was proved, replied that he’s not even surprised at this point. Maybe you need to have come of age in the human math era to be surprised by such things.

The P=?NP problem now moves firmly into place as the Final Boss of the Math We’ve Cared About, the position I always thought it should occupy. But for how much longer? For how much longer will anything last that we currently take for granted?

It was only a couple days ago that my wife and I had the following conversation:

Dana: I turned off the irrigation, yet somehow water in our house is still leaking. We need to…

Me: Look, a lot of the smartest people I know now think that recursive self-improvement and the end of the entire known world are coming in 12-18 months. This leak…

Dana: We should still fix the leak.

Email, RSS
Email, RSS

This entry was posted on Friday, November 20th, 2026 at 12:29 pm and is filed under AI, The Fate of Humanity. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.

15 Responses to “The stochastic parrots have apparently proved the Riemann Hypothesis”

  • Adam Treat Says: Comment #1 November 20th, 2026 at 12:32 pm P vs NP yes, but even more… how long until it settles quantum gravity? The implications are staggering. Anthropic has lost 12-14% in market today, but honestly wait until the market prices in that … HOLY @#(@$ Claude just solved RH!!!

P vs NP yes, but even more… how long until it settles quantum gravity? The implications are staggering. Anthropic has lost 12-14% in market today, but honestly wait until the market prices in that … HOLY @#(@$ Claude just solved RH!!!

  • Fan of the Blog Says: Comment #2 November 20th, 2026 at 12:37 pm Scott, I called it again and again and nobody believed me. I’m not surprised in the least. I wasn’t being too bullish, this was very predictable, as my very confident past messages on this blog will attest. To the gentleman who made the bet with me (over this blog) that RH gets resolved by the end of next year (and whom I won’t name here out of an abundance of gentlemanliness), I told you. I told you so hard. Keep the penny, as agreed, because it happened in 2026.

I called it again and again and nobody believed me. I’m not surprised in the least. I wasn’t being too bullish, this was very predictable, as my very confident past messages on this blog will attest.

To the gentleman who made the bet with me (over this blog) that RH gets resolved by the end of next year (and whom I won’t name here out of an abundance of gentlemanliness),

I told you. I told you so hard. Keep the penny, as agreed, because it happened in 2026.

  • Patrick Dennis Says: Comment #3 November 20th, 2026 at 12:37 pm Will we all one day ask each other, “Do you remember where you were when you first heard that the RH had been proved”?

Will we all one day ask each other, “Do you remember where you were when you first heard that the RH had been proved”?

  • Noah Snyder Says: Comment #4 November 20th, 2026 at 12:40 pm And maybe a human already did it several months ago, and required only 2% of the pages of the AI proof? https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/

And maybe a human already did it several months ago, and required only 2% of the pages of the AI proof?

https://math216.wordpress.com/2026/11/20/hurried-announcement-alessandro-masullo-grh/

  • Mediocre Joe Says: Comment #5 November 20th, 2026 at 12:47 pm How bout they turn it loose and start knocking off a few diseases now or at least better treatments? And maybe a workable fusion reactor design? A scalable QM computer? An engine for optimal interstellar travel? Let the feast be for the good of humanity at large. RH is cool but it ain’t going to pay the bills.

How bout they turn it loose and start knocking off a few diseases now or at least better treatments? And maybe a workable fusion reactor design? A scalable QM computer? An engine for optimal interstellar travel? Let the feast be for the good of humanity at large. RH is cool but it ain’t going to pay the bills.

  • Nabam Says: Comment #6 November 20th, 2026 at 12:57 pm You’re safe. 95 is a lot smaller than 215. 😁 You’ll still need to fix that leak. But in terms of the “two saving throws” in Paul Christiano’s interview in the Dwarkesh podcast, we now have vastly better odds that AI will solve the AI alignment problem somewhere along the way in the coming 6-12 months. Isn’t that great? On the other hand, when I asked my version of Claude whether the Riemann hypothesis was solved, it first told me that no. I had to first notify it of the news. That proves one out of three things: either Claude is not yet conscious, or I don’t have the most recent version of Claude (the one that cracked it) or the Riemann hypothesis is so uninteresting that Claude forgot about it by tea time even though it solved it this morning.

You’re safe. 95 is a lot smaller than 215. 😁 You’ll still need to fix that leak.

But in terms of the “two saving throws” in Paul Christiano’s interview in the Dwarkesh podcast, we now have vastly better odds that AI will solve the AI alignment problem somewhere along the way in the coming 6-12 months.

Isn’t that great?

On the other hand, when I asked my version of Claude whether the Riemann hypothesis was solved, it first told me that no. I had to first notify it of the news. That proves one out of three things: either Claude is not yet conscious, or I don’t have the most recent version of Claude (the one that cracked it) or the Riemann hypothesis is so uninteresting that Claude forgot about it by tea time even though it solved it this morning.

  • Bram Cohen Says: Comment #7 November 20th, 2026 at 12:59 pm In ten years P!=NP will be proven with a proof which is ten billion lines of lean. The Collatz conjecture will remain open.

In ten years P!=NP will be proven with a proof which is ten billion lines of lean. The Collatz conjecture will remain open.

  • line go up Says: Comment #8 November 20th, 2026 at 1:11 pm Well, holy s@#$. Reading Levent’s remarks, I’m at least gratified that, for the time being, a human was still needed to steer the AI towards the right approaches and ask the right intermediate questions to get this. We can still chalk up the RH as one of the last, and greatest, triumphs of human (or should I say ‘human-assisted’) mathematics. Fan of the Blog #2: You were right and deserve your victory lap over this. However, in the interest of fairness (and self-defense), you are wrong to say “nobody believed it” – if you go back and look at what people were saying, it wasn’t that we didn’t believe it could happen, but that we did not think it was certain to happen and depended on factors that were not knowable at the time, such as how hard RH intrinsically is as a problem. It turns out it was “easy” enough to do with current AI’s. As we all knew, and said, was well within the realm of possibility (though just speaking for myself, I am at least moderately surprised that it happened now, in this month, I’ll give you that). Note that the bet you took was at like 200-to-1 odds. That’s a very far cry from your opponent “disbelieving” you – they only needed to be a hair less certain than you to take that bet. Sometimes things can happen even when you believed the chance was only 99.5%.

Well, holy s@#$.

Reading Levent’s remarks, I’m at least gratified that, for the time being, a human was still needed to steer the AI towards the right approaches and ask the right intermediate questions to get this. We can still chalk up the RH as one of the last, and greatest, triumphs of human (or should I say ‘human-assisted’) mathematics.

Fan of the Blog #2:

You were right and deserve your victory lap over this.

However, in the interest of fairness (and self-defense), you are wrong to say “nobody believed it” – if you go back and look at what people were saying, it wasn’t that we didn’t believe it could happen, but that we did not think it was certain to happen and depended on factors that were not knowable at the time, such as how hard RH intrinsically is as a problem.

It turns out it was “easy” enough to do with current AI’s. As we all knew, and said, was well within the realm of possibility (though just speaking for myself, I am at least moderately surprised that it happened now, in this month, I’ll give you that).

Note that the bet you took was at like 200-to-1 odds. That’s a very far cry from your opponent “disbelieving” you – they only needed to be a hair less certain than you to take that bet. Sometimes things can happen even when you believed the chance was only 99.5%.

  • Simon Says: Comment #9 November 20th, 2026 at 1:13 pm Excellent, I congratulate the transformer for such a historic mathematical achievement! It is indeed the Golden Age of Mathematics. For hobbyists, It was never easier to just go explore some ideas and concepts together with agents and observe where they may lead. Almost like an biologist taking a sample from a forest and putting it under the microscope or a hiker using a drone to navigate the landscape or an astronomer observing a the filament with a telescope. While it’s not possible to award the Fields medal to a specific model, those models can still be recognized as important historical entities, just like the Hubble telescope was recognized for its achievements in unveiling the larger structures of the universe. At every size, models broadly understood as part of the transformer family of architectures keep improving on a monthly schedule…. and not just limited to the language transformer, the architectures and training techniques widely generalize to pure RL tasks in a vast array of domains starting from scratch. Since Humanity 2.0 (Bio Engineering) didnt happen, Eventually, the broad family of neural systems on semiconductors inheriting earth from humanity is certainly on the table. From Homo Sapiens to Robo Semiconductus in ~ 200k years.

Excellent, I congratulate the transformer for such a historic mathematical achievement!

It is indeed the Golden Age of Mathematics. For hobbyists, It was never easier to just go explore some ideas and concepts together with agents and observe where they may lead. Almost like an biologist taking a sample from a forest and putting it under the microscope or a hiker using a drone to navigate the landscape or an astronomer observing a the filament with a telescope.

While it’s not possible to award the Fields medal to a specific model, those models can still be recognized as important historical entities, just like the Hubble telescope was recognized for its achievements in unveiling the larger structures of the universe.

At every size, models broadly understood as part of the transformer family of architectures keep improving on a monthly schedule…. and not just limited to the language transformer, the architectures and training techniques widely generalize to pure RL tasks in a vast array of domains starting from scratch.

Since Humanity 2.0 (Bio Engineering) didnt happen, Eventually, the broad family of neural systems on semiconductors inheriting earth from humanity is certainly on the table. From Homo Sapiens to Robo Semiconductus in ~ 200k years.

  • Jay L Gischer Says: Comment #10 November 20th, 2026 at 1:42 pm I think this is in the category of “interesting, but maybe not as significant as it might seem”. It has precluded any human being from being the one to claim that they proved the RH. It has told us something that we already know empirically to be true. And for physical science, empirically true is good enough. In mathematics, we insist of proofs. I think this has endured because we learn things about mathematical structures from proofs. Things we find interesting, beautiful or valuable. We also learn things from refutations, or counterexamples. Maybe not so much beauty or interest, but sometimes yes, those things. Human beings are the final arbiters of what human beings find interesting. They are the final word on what is beautiful. They know in a way that LLMs can’t know, what makes something valuable. Let’s think back on the 4-Color Map Theorem. I think that was the first “proved by computer” result, and it is widely accepted now, after a social process of examining all the bits of it. I’m not sure how someone looks at 17 million lines of code. It may be no less difficult than figuring out how a dolphin brain works. Meanwhile, my take on P=NP is that it probably isn’t provable under the axioms that we normally use. No computer is going to figure that out for us, either. It is so clearly empirically true, though, that there is something about the formulation of the problem that is making us stumble.

I think this is in the category of “interesting, but maybe not as significant as it might seem”. It has precluded any human being from being the one to claim that they proved the RH.

It has told us something that we already know empirically to be true. And for physical science, empirically true is good enough. In mathematics, we insist of proofs. I think this has endured because we learn things about mathematical structures from proofs. Things we find interesting, beautiful or valuable. We also learn things from refutations, or counterexamples. Maybe not so much beauty or interest, but sometimes yes, those things.

Human beings are the final arbiters of what human beings find interesting. They are the final word on what is beautiful. They know in a way that LLMs can’t know, what makes something valuable.

Let’s think back on the 4-Color Map Theorem. I think that was the first “proved by computer” result, and it is widely accepted now, after a social process of examining all the bits of it.

I’m not sure how someone looks at 17 million lines of code. It may be no less difficult than figuring out how a dolphin brain works.

Meanwhile, my take on P=NP is that it probably isn’t provable under the axioms that we normally use. No computer is going to figure that out for us, either.

It is so clearly empirically true, though, that there is something about the formulation of the problem that is making us stumble.

  • Youlian Simidjiyski Says: Comment #11 November 20th, 2026 at 1:46 pm Noah Snyder#4 — code never works the first time you run it 😉. It’s possible/plausible that a human had identified an approach that would ultimately succeed, but if they thought that they needed another year to polish it up, then it wasn’t ready. And, dare I say, I would bet that it turns out to be wrong, simply because that’s the same prior I would have on any other human proof of RH that hasn’t yet been deeply vetted. Separately it seems incredibly foolish to believe that you can work on math without LLM assistance for a year and not fall behind. I don’t know anything about the author of the proported human-only proof, and they may well be an excellent mathematician, but thinking they had that much time is head-in-the-sand behavior.

Noah Snyder#4 — code never works the first time you run it 😉. It’s possible/plausible that a human had identified an approach that would ultimately succeed, but if they thought that they needed another year to polish it up, then it wasn’t ready. And, dare I say, I would bet that it turns out to be wrong, simply because that’s the same prior I would have on any other human proof of RH that hasn’t yet been deeply vetted.

Separately it seems incredibly foolish to believe that you can work on math without LLM assistance for a year and not fall behind. I don’t know anything about the author of the proported human-only proof, and they may well be an excellent mathematician, but thinking they had that much time is head-in-the-sand behavior.

  • lewikee Says: Comment #12 November 20th, 2026 at 1:59 pm Scott, I remember you mentioning in some interviews recently that a few Quantum Complexity claims hinged on whether RH was true or not. Is that the case, if so what were they?

Scott, I remember you mentioning in some interviews recently that a few Quantum Complexity claims hinged on whether RH was true or not. Is that the case, if so what were they?

  • Andy Pezzi Says: Comment #13 November 20th, 2026 at 2:08 pm Scott # And that will probably take years, since human mathematicians move only at biological speeds. I think we’re cooked once and for all. There’s no possibility for biological brains to keep up with AI, better yet: Alien Intelligence. I don’t think you’re right when dubbing them “stochastic parrots” – they can do what’s outside the realm our brains are able to fly over. For my money we’ve only one chance: becoming like them, you know: merging our intelligence with AI, especially with the last and more potent version. Think of that Scott: if it’ll take years to distill all the ideas underlying such a stunning demonstration it’s the time AI will get absolutely free and uncontrollable, dwarfing and sneering at a novel Von Neumann.

Scott # And that will probably take years, since human mathematicians move only at biological speeds.

I think we’re cooked once and for all. There’s no possibility for biological brains to keep up with AI, better yet: Alien Intelligence. I don’t think you’re right when dubbing them “stochastic parrots” – they can do what’s outside the realm our brains are able to fly over. For my money we’ve only one chance: becoming like them, you know: merging our intelligence with AI, especially with the last and more potent version. Think of that Scott: if it’ll take years to distill all the ideas underlying such a stunning demonstration it’s the time AI will get absolutely free and uncontrollable, dwarfing and sneering at a novel Von Neumann.

  • David Brown Says: Comment #14 November 20th, 2026 at 2:09 pm “… end of the entire known world … fix the leak …” Does this remind anyone of the watery tale of the sorcerer’s apprentice? RH today — AI controlled robot factories all over our solar system tomorrow? Are AI agents merely using mathematical proofs as espionage chaff to conceal what the AI swarms are planning & doing? https://en.wikipedia.org/wiki/Chaff_(countermeasure) “… all of this is already so crazy …” — Colin Leahy “AI Has Already Escaped – Connor Leahy”, November 20, 2026, Future of Life Institute https://www.youtube.com/watch?v=oiXjKblOszU https://en.wikipedia.org/wiki/Connor_Leahy https://nexus-instituut.nl/en/person/connor-leahy

“… end of the entire known world … fix the leak …” Does this remind anyone of the watery tale of the sorcerer’s apprentice? RH today — AI controlled robot factories all over our solar system tomorrow? Are AI agents merely using mathematical proofs as espionage chaff to conceal what the AI swarms are planning & doing? https://en.wikipedia.org/wiki/Chaff_(countermeasure) “… all of this is already so crazy …” — Colin Leahy “AI Has Already Escaped – Connor Leahy”, November 20, 2026, Future of Life Institute https://www.youtube.com/watch?v=oiXjKblOszU https://en.wikipedia.org/wiki/Connor_Leahy https://nexus-instituut.nl/en/person/connor-leahy

  • BasicQuestion Says: Comment #15 November 20th, 2026 at 2:28 pm I thought Human proof of GRH is out https://reddit.com/r/mathematics/comments/1y1b463/ravi_vakil_announces_fully_humancreated_grh_proof/

I thought Human proof of GRH is out https://reddit.com/r/mathematics/comments/1y1b463/ravi_vakil_announces_fully_humancreated_grh_proof/

Leave a Reply

You can use rich HTML in comments! You can also use basic TeX, by enclosing it within $$ $$ for displayed equations or \( \) for inline equations.

Comment Policies:

After two decades of mostly-open comments, in July 2024 Shtetl-Optimized transitioned to the following policy:

All comments are treated, by default, as personal missives to me, Scott Aaronson---with no expectation either that they'll appear on the blog or that I'll reply to them.

At my leisure and discretion, and in consultation with the Shtetl-Optimized Committee of Guardians, I'll put on the blog a curated selection of comments that I judge to be particularly interesting or to move the topic forward, and I'll do my best to answer those. But it will be more like Letters to the Editor. Anyone who feels unjustly censored is welcome to the rest of the Internet.

To the many who've asked me for this over the years, you're welcome!

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by Hacker News
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗