{"slug": "ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything", "title": "AI Just Cracked the Hardest Problem in Mathematics. It Will Change Everything", "summary": "Anthropic employee Levent Alpöge published a PDF on November 20 claiming to have solved the 167-year-old Riemann Hypothesis with Claude, releasing 17 million lines of Lean code that passed multiple independent verification checks but no human-readable proof — his attempt at one ran past 250,000 pages, which he called an \"unreadable travesty of slop.\" Mathematicians split on the result: Stanford's Brian Conrad told The New York Times it is \"frankly completely outrageous\" to release the proof without fully understanding it, while Clay Mathematics Institute president Martin Bridson called the announcement of \"historic significance\" and Fields Medallist James Maynard told Nature it could rank among \"the absolute pinnacles of mathematical results.\" NYU's Maksym Radziwiłł told Quanta Magazine the work does not answer mathematics' core question of \"why.", "body_md": "On November 20, Anthropic employee Levent Alpöge published a PDF where he claimed to have [solved the Riemann Hypothesis](https://alpo.ge/remark.pdf) (RH) working together with Claude.\n\nHe released the Lean code (17 million lines), which has already [passed multiple independent](https://math.mit.edu/~drew/grh/) [verification checks](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/), but not a human-readable demonstration: his attempt to produce one reached 250,000+ pages of what he called an “unreadable travesty of slop.” Given that RH concerns zeros of the zeta function, it’s fair to say that Claude wrote much ado about nothing. But even if we don’t know the “hows” or “whys” of the Riemann Hypothesis, it seems safe to say that now we know it’s true.\n\nWhat does this mean for mathematics? Or, well, the world writ large?\n\nIt *changes everything*. This is as historic as it gets.\n\nYou might think I’m exaggerating. “If every event changes everything, then nothing changes at all.” But for real, *how am I to blame for the fact that AI is indeed breaking the ground at every step it takes?* Anthropic’s IPO? Biggest in history. The AI boom? Most expensive. And now RH, which, at 167 years old, was widely considered by mathematicians the most important unsolved problem in the field. These things exist in the superlative; “changes everything” is merely an objective descriptor.\n\nAnd so while clickbaiters enjoy a field day, mathematicians witness a field die.\n\nAlpöge agrees that this is an inflection point but disagrees with the defeatist tone: “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.” Maybe he’s not the one to make that judgement call, though. He’s accustomed, naturally, to deciding what’s good or bad for humanity from the privacy of the datacenter where he lives.\n\nLet’s see what others think.\n\nCritics and skeptics will never go out of vogue when it comes to AI-powered mathematics.\n\nMaksym Radziwiłł, professor of mathematics at New York University, [told Quanta Magazine](https://www.quantamagazine.org/update/169569/) that “The object of mathematics as the rest of science is really the question ‘why?’ and this doesn’t answer it at all.” Andrew Granville, professor of mathematics at the Université de Montréal, [told Nature](https://www.nature.com/articles/d41586-026-03812-7) that “What he released is garbled,” and argued that Alpöge should have prepared an intelligible mathematical exposition before announcing the result. Brian Conrad, professor of mathematics at Stanford University, [told The New York Times](https://www.nytimes.com/2026/11/20/science/anthropic-ai-math-riemann-hypothesis.html) that it is “frankly completely outrageous” that he released the proof without understanding it fully.\n\nOthers are more measured, some even enthusiastic.\n\nFields Medallist Terence Tao [says](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/) the proof will take time to digest, although he notes that at least there’s a human author to consult. James Maynard, also a Fields Medallist, [told Nature](https://www.nature.com/articles/d41586-026-03812-7) that, once the proof is properly understood, it could rank among “the absolute pinnacles of mathematical results.” Martin Bridson, president of the Clay Mathematics Institute, [called it](https://apnews.com/article/anthropic-claude-math-solution-ai-4fe8b69a891f160f9e7f0c85af8c55c8) an announcement of “historic significance,” potentially transforming our understanding of what mathematical research can achieve. Cornell mathematician Steven Strogatz offered [The Washington Post](https://www.washingtonpost.com/technology/2026/11/20/using-anthropic-ai-mathematician-claims-have-solved-riemann-hypothesis/) an appropriate description of Alpöge and Claude: “A very skilled AI jockey riding a very fast AI horse.”\n\nThe mathematics community is, as with all things AI, divided. And I can see why because the impact of AI cracking RH implies that the future is totally unclear and perhaps not quite human.\n\nPerhaps mathematics will become a [pastime](https://www.thealgorithmicbridge.com/p/millennium-pastimes), pursued for the pleasure of thinking, as people still play chess knowing they cannot beat the machines. Perhaps it will [cease to exist as an academic career](https://www.thealgorithmicbridge.com/p/new-survey-math-students-are-giving) altogether. Perhaps we should accept that [no human pursuit is eternal like the Alps](https://www.thealgorithmicbridge.com/p/i-have-been-writing-about-ai-for). Or perhaps the opposite will happen: the machine will fling open entire wings of the cathedral of mathematics, and those with the curiosity, ambition, and nerve to venture inside will discover that there is more to explore than anyone, Riemann included, ever imagined.\n\nIn terms of immediate utility, just as with Navier–Stokes (whose “harder” form was [also solved this week](https://github.com/tristanbuckmaster/navier-stokes/) by Alpöge, Tristan Buckmaster, and Claude), proving RH doesn’t seem like a big deal. Mathematicians have been assuming it true for decades, so the main consequence is that some proofs can finally dispense with the fine print: *conditional on RH being true*. Then again, mathematicians aren’t known for their concern with real-world utility. If they were, they’d have stuck to three dimensions and made a living cleaning windows or fitting pipes.\n\nAs for the existential impact on the profession, however, the best summary is perhaps this: mathematicians, novices and veterans alike, might soon find themselves cleaning windows and fitting pipes. Well, perhaps that’s a slight exaggeration. After all, only they can make sense—if anyone can at all—of these discoveries. No one better than human mathematicians.\n\nThey will become the connective tissue between humanity and the machine, capable of speaking both languages and thus translating for us. Or, conversely, they will become *cuckulators*, to use the derogatory term—albeit essentially accurate—that’s making the rounds online, which suggests that they will prompt the machine and sit back while it performs intellectual feats worthy of a fantasy hero lacking the most basic verisimilitude amid a bunch of unintelligible and unparseable slop cuckulations. Then, they will clean up the mess, stripping down the haystack to find the needle.\n\nThere’s a silver lining, though. I wrote it down in August.\n\nI’ve been chronicling the fall of mathematics into the maws of the artificial beast for months now. In August, I published “[The month AI conquered math](https://www.thealgorithmicbridge.com/p/the-month-ai-conquered-math-the-full)” and even then, I joked about the RH in a way that, even as satire, felt far-fetched: why am I talking about the Riemann Hypothesis as if this was not some otherworldly challenge that some aliens left for us in this obviously manufactured universe? Well, turns out I was not bold enough. In the age of AI, reality is indeed stranger than fiction. David Hilbert [would be rolling on his grave](https://x.com/fermatslibrary/status/2123804859291877644). I wrote:\n\nAs the most fervent advocates love to say—and I agree with them on this point—this is the worst AI will ever be. The Riemann hypothesis lies on the horizon, like a final boss.\n\nWe’ve arrived at this horizon in barely 100 days. I thought maybe 100 months, if we were lucky. Even if understanding the proof will take longer, the final boss seems to have indeed fallen ahead of schedule; surely much sooner than even [Tao himself thought possible with AI](https://x.com/LinkofSunshine/status/2123800866738024731). But I had also preempted a “cope pill” for this event. I wrote:\n\nSolving the Riemann hypothesis would be a breakthrough of such a magnitude that it would dwarf everything we’ve seen so far. And yet, solving the Riemann hypothesis doesn’t make one Riemann, for he had to come up with it in the first place.\n\nAha! Eureka! Gotcha, clanker!\n\nYou might be smarter than all humans combined but I bet you can’t conjure up a problem that will keep yourself busy for nearly two centuries. Riemann himself was not even close to solving his own hypothesis (although he correctly assumed it to be true) but he nevertheless came up with it.\n\nWhich is harder, I ask you, creating or solving problems? Is it harder to invent new math or using optimally the math that already exists? The correct answer is, naturally, a matter of *what kind of creature* one is.\n\nFor us humans, obviously the latter is harder for both questions. On the one hand, whereas only one human could imagine RH, none could solve it (machines are better than us at math but even we can see that 1 > 0). On the other hand, no one could figure out what was needed to solve RH but in the meantime we’ve collectively invented entire branches of mathematics, from set theory and topology to category theory. Newton wanted to understand motion, so he invented calculus. And, as much as it might have pained him, he was only human. As far as we know—maybe I’m wrong—it’s exactly the other way around for machines.\n\nMaybe they will someday beat us at everything, as the AI industry hopes, but I’m not convinced today’s architectures, inductive biases, and training objectives are equally suited to every form of intellectual discovery. As I asked in the August essay,\n\nWhy does [AI] have such strong deductive and inductive skills and yet it lacks in abduction? Why does it tend to find counterexamples rather than proofs? Why is it so good now at scouring the abandoned corridors of the cathedral of mathematics . . . and yet it would have no clue whatsoever about how to build a new wing?\n\nMaybe the endgame of mathematics is that people like Alpöge will use AI to solve every single math problem humanity has ever found but will in turn fail to figure out new ones because those left to be encountered lie beyond what human eyes can see.\n\nIn his words:\n\nClaude 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.\n\nIt would be not so much a state of “proof abundance,” then, but of problem scarcity.\n\nApparently, Claude used a lot of math to solve RH—in Alpöge’s words, “An astounding amount of mathematics is present in the argument”—in ways no human could, but it doesn’t seem to have invented entirely new kinds of it, which mathematicians thought of as a requirement to solving it.\n\n[Here’s Tao on Dwarkesh Podcast in March 2026](https://x.com/dwarkesh_sp/status/2123966918688403756):\n\nPart of the reason we prize problems like the Riemann hypothesis is that we’re pretty sure a new type of mathematics has to be created. We don’t even know what the shape of the solution is, but it doesn’t feel like a problem that will be solved just by exhaustively checking cases.\n\nIt seems that, whereas we humans dig new holes to find new treasures, AI can simply wormhole through multidimensional space from existing holes into the golden chest. That is incredible and also an *inhuman* way of doing things, with some flaws as well as important advantages. If we take Tao’s remarks at face value, this solution to RH, insofar as no new math was actually needed which is a stretch to assert given we know so little about how it was done, won’t be as useful even after mathematicians have finally connected the dots.\n\nMaybe, after all, *we do make a good team*.\n\nLike a knight and his steed (who is whom here, though?). Like a centaur even: half a powerful body, half a creative mind. The specifics of this unlikely alliance are yet to be described accurately by the taxonomists of our time. Is it creativity that we excel at? Maybe taste? Sensibility to beauty?\n\nUnfortunately, while a fair number of people can ride a horse, it’s not quite as easy to ride a machine. Maybe Alpöge can or Tao. Riemann could for sure as could Von Neumann, already himself half machine if we are to take at face value the anecdotes of those who knew him. The rest of us, as it always was, won’t have much room in this new intellectual arc.\n\nOh, but it has always been like this! What am I grieving if not a vicarious loss that will in no way affect my actual life? That’s why most people are happy—or, at least, not sad—that mathematics has fallen, because they see the spectacle not from atop the ivory tower crumbling under their feet but from the ground level, on the dirty street. They watch the demolition of the last symbol of an old world of “betters” and “worses.” AI is the great equalizer, they think, an inexorable democratization mechanism. “If you value intelligence above all other human qualities,” [Ilya Sutskever said](https://x.com/ilyasut/status/1710462485411561808), “you’re gonna have a bad time.” The rest of us are, as usual, strolling through the mud in the mire.\n\nThey are terribly wrong.\n\nI mean it literally: not only do mathematicians like Alpöge enjoy the gift of way above average smarts but now they also enjoy the access to the means of intellectual discovery. Thinking in terms broader than math, we’re living in a *silicracy*: those with access to the frontier models will beat everyone else at everything.\n\nNot to go on a tangent, but I read earlier today an insight about Nvidia’s relationship with Anthropic/OpenAI and I couldn’t agree more: for the chipmaker to survive it will need to become a modelmaker faster than the modelmakers are becoming chipmakers. They can throw Claude or Astra against the problem of designing a cutting-edge chip and they might break all the records.\n\nThat is so amazing for progress.\n\nExcept that it’s also a new form of class struggle.\n\nThese rich people—Anthropic just IPOed so most employees are millionaires, some of them billionaires—took without permission the accumulated knowledge and know-how of human civilization, built a machine, and are now preparing unfathomable concoctions to find truths restricted to ordinary humans not only cognitively but also materially. Alpöge didn’t use a public Claude model but a private one; other mathematicians could’ve beaten him? Probably not. After all, Tao—arguably the best living mathematician—uses the top public models all the time and he thought RH still had some 20 years to go. No one but an Anthropic employee could have solved RH.\n\nMath has been a matter of having pen and paper for five hundred years but now we’re back at the pre-printing press stage of science. The amount you can contribute to the world, regardless of the size of your brain or your scientific prowess, is measured in token spending. AI is not a mechanism of democratization but the exact opposite: we are deep in a new form of feudalism.\n\nMoving forward, the progress of science could be determined by something as arbitrary as whether Dario Amodei likes you. I’m not even kidding. Alpöge explicitly says he contacted Amodei to ask whether he could announce his discovery. *[Redacted] model has been used as [redacted] to help me and [redacted] solve [redacted] using techniques such as [redacted] and [redacted] that you will be able to use soon for only $199.99/month.* That’s what it sounds like.\n\nThink about the absurdity. The greatest mathematical discovery in history needs a CEO’s permission to be shared with the public and will be scheduled for release according to the will of the markets and the private interests of Anthropic. “Did you add the ultra-safe-for-stupid-people guardrails, Alpöge?” No wonder Amodei thinks Anthropic will be more powerful than the US government.\n\nI wouldn’t be surprised if he thought he, rather than the machine, was the Second Coming.\n\nThis silicracy may be temporary, however, for a reason its beneficiaries haven’t considered: money can buy one access to a superior intelligence, but it cannot buy one the intelligence required to keep up with it. Eventually, and perhaps for the best, math will go the way of chess. Not even machine riders will remain, for the machine is itself as human as it is horse; body and mind both superior. The era of the centaur begins today but it, too, shall pass. A beautiful liminal phase of cooperation that will end the instant the last human becomes a hindrance to the machine.\n\nMagnus Carlsen would fail to win 1,000 out of 1,000 games against the latest Stockfish version and that shouldn’t surprise anyone at this point, but what is not well-known is that he cannot quite contribute at all in a Stockfish vs Stockfish game. His suggestions, suboptimal; his every move, one step closer to defeat. It’s not hard to see how mathematics will go down the same way. It is as much a sign of cope to say that creating new math is beyond the machine as is to say that humans will always remain relevant in some way. No one escapes the unsavory aftertaste of the Bitter Lesson.\n\nIf Claude solving RH is not enough to convince you of this, then nothing will. Not even reality.\n\nAs for me, I see a world submerged not in a sea of paperclips but of 250,000-page slop pdfs. What will happen when the machine’s proofs are not understood even by the humans using it?\n\nNothing will happen.\n\nAlpöge doesn’t fully know what’s going on. RH is hard because the human mind can’t quite crack it. But if a different mind does, it doesn’t immediately make it more legible to us. The machine will have to resort to analogies and ELI5s to include us.\n\nEventually, neither they nor we will be able to bridge the gap and no connective tissue will remain to hold together the human world and the machine world. They will drift away, in a sort of parallel but unconnected existence, like the Milky Way and the rest of the universe.", "url": "https://wpnews.pro/news/ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything", "canonical_source": "https://www.thealgorithmicbridge.com/p/ai-just-cracked-the-hardest-problem", "published_at": "2026-11-21 18:59:24+00:00", "updated_at": "2026-11-21 19:12:36.290492+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "large-language-models", "ai-safety"], "entities": ["Anthropic", "Levent Alpöge", "Claude", "Riemann Hypothesis", "Terence Tao", "James Maynard", "Brian Conrad", "Clay Mathematics Institute"], "also_reported_by": [], "takeaway": "Claude produced a 17-million-line Lean proof of the Riemann Hypothesis that passed multiple independent verification checks, but the 250,000-plus-page human-readable exposition Alpöge attempted was judged unreadable, showing machine-checkable formal proofs can outrun human comprehension.", "recommendation": {"advice": "Pair formal, machine-checked proofs with a human-readable exposition before announcing results; reviewers rejected Alpöge's 250,000-page write-up as garbled despite the Lean code passing verification.", "playbook": "output-verification", "playbook_url": "https://wpnews.pro/playbooks/output-verification"}, "alternates": {"html": "https://wpnews.pro/news/ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything", "markdown": "https://wpnews.pro/news/ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything.md", "text": "https://wpnews.pro/news/ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything.txt", "jsonld": "https://wpnews.pro/news/ai-just-cracked-the-hardest-problem-in-mathematics-it-will-change-everything.jsonld"}}