# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/openai-withdraws-3-math-papers-50003107 # OpenAI Withdraws 3 Math Papers 331 points · 315 comments [Full discussion]() [Read original]() Category: [Research & Evaluation]() ## Skept-o-meter & Hotness Skept\-o\-meter: Low\. Estimated from 10 comments\. 14 comments for the summary\. Peak rank: \#2 Time in Top 10: 24\.0 hours Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\. 42 recorded rank observations from 2026\-10\-08T14:03:30\.113384\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\. Hotness — latest 42 recorded Hacksnap ranks: 2026\-10\-08T14:03:30\.113384\+00:00: rank \#7 2026\-10\-08T15:02:54\.444758\+00:00: rank \#7 2026\-10\-08T16:03:13\.067807\+00:00: rank \#6 2026\-10\-08T17:03:47\.92465\+00:00: rank \#4 2026\-10\-08T18:01:50\.360431\+00:00: rank \#4 2026\-10\-08T19:02:55\.719274\+00:00: rank \#3 2026\-10\-08T20:01:36\.359022\+00:00: rank \#2 2026\-10\-08T21:03:34\.000719\+00:00: rank \#2 2026\-10\-08T22:01:13\.459732\+00:00: rank \#3 2026\-10\-08T23:01:25\.318392\+00:00: rank \#3 2026\-10\-09T08:01:53\.314021\+00:00: rank \#3 2026\-10\-09T09:02:59\.93426\+00:00: rank \#3 2026\-10\-09T10:02:22\.652527\+00:00: rank \#3 2026\-10\-09T11:02:04\.807282\+00:00: rank \#3 2026\-10\-09T12:02:16\.769295\+00:00: rank \#3 2026\-10\-09T13:02:30\.35034\+00:00: rank \#3 2026\-10\-09T14:01:34\.215134\+00:00: rank \#90 2026\-10\-09T15:01:27\.564392\+00:00: rank \#89 2026\-10\-09T16:02:06\.561056\+00:00: rank \#90 2026\-10\-09T17:03:06\.673648\+00:00: rank \#92 2026\-10\-09T18:01:02\.851052\+00:00: rank \#92 2026\-10\-09T19:01:11\.250291\+00:00: rank \#92 2026\-10\-09T20:01:32\.91907\+00:00: rank \#92 2026\-10\-09T21:02:08\.376144\+00:00: rank \#94 2026\-10\-09T22:01:19\.253075\+00:00: rank \#96 2026\-10\-09T23:01:31\.990912\+00:00: rank \#96 2026\-10\-10T08:01:00\.593376\+00:00: rank \#95 2026\-10\-10T09:01:45\.527975\+00:00: rank \#97 2026\-10\-10T10:00:53\.799353\+00:00: rank \#97 2026\-10\-10T11:00:35\.65851\+00:00: rank \#96 2026\-10\-10T12:01:11\.169647\+00:00: rank \#96 2026\-10\-10T13:00:54\.008377\+00:00: rank \#95 2026\-10\-10T14:01:43\.208766\+00:00: rank \#97 2026\-10\-10T15:01:24\.825188\+00:00: rank \#97 2026\-10\-10T16:01:10\.213328\+00:00: rank \#96 2026\-10\-10T17:00:48\.84189\+00:00: rank \#95 2026\-10\-10T18:00:57\.55132\+00:00: rank \#95 2026\-10\-10T19:00:52\.379463\+00:00: rank \#94 2026\-10\-10T20:00:21\.993912\+00:00: rank \#94 2026\-10\-10T21:01:10\.657738\+00:00: rank \#95 2026\-10\-10T22:00:50\.433569\+00:00: rank \#94 2026\-10\-10T23:01:00\.95027\+00:00: rank \#95 OpenAI withdrew three math papers after a sign error invalidated a shared construction and revised 14 others; the debate centers on whether external verification is healthy correction or an unpaid burden\. ## The brief OpenAI's math history page reports that a sign error in a paper on Weil classes invalidated a stabilization\-trace cancellation argument and a construction used by two dependent papers, prompting withdrawal of three manuscripts\. The page says 14 other manuscripts were revised with proof repairs and clarified hypotheses, and 13 more were updated to cite the revised editions\. It also reports six new formalizations and five other additions, bringing formalized top\-line results to 300 of 719, roughly 42%\. - A sign error in “Algebraicity of Weil classes on split abelian eightfolds” invalidated a stabilization\-trace cancellation argument and the construction used by two dependent papers\. - OpenAI withdrew three manuscripts: the Weil classes paper, “Algebraicity of Kuga–Satake Correspondences for K3 Surfaces,” and “The rational Hodge conjecture for products of K3 surfaces\.” - Withdrawn papers carry notices explaining the gap and linking to archived manuscripts; updated papers retain prior editions via README version notes\. - OpenAI revised 14 other manuscripts with proof repairs, corrected statements, clarified hypotheses and dependencies, and one obsolete citation fix\. - The fixes prompted 13 additional manuscripts to cite revised companion editions, and OpenAI added 6 formalizations plus 5 other additions\. - Total top\-line results formalized now stand at 300 of 719, about 42%\. ## Discussion themes Analyzed: 2026\-10\-08T17:03:17\.194417\+00:00 Analysis sample: Based on 32 of 52 usable stored comments. Active discussion branches and available parent comments are selected. The analysis input was further shortened to fit its context limit. This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is. ### Error rate and verification of AI\-generated math papers Commenters debate the significance of a few flawed papers among roughly 400, with one calling 3 mistakes a good hit rate and another noting retractions are career\-damaging for mathematicians\. Replies question the comparison given the 48\-hour volume and note withdrawals came only after external review, including an Astra review, raising concerns about pre\-publication verification\. Others argue autonomous verification already exists at scale, while a reply disputes that much mathematical knowledge is unwritten or unformalized\. Sources: [Comment 50004733]() · [Comment 50004887]() · [Comment 50004919]() · [Comment 50004961]() · [Comment 50006183]() · [Comment 50006231]() · [Comment 50006253]() ### Formalization and review burden for AI\-written proofs Commenters argue that a large dump should be fully formalized because there is too much material to review by hand, and AI\-written write\-ups are hard to read\. One reply says the write\-ups are garbage and could have been improved, while another asks why not share flawed work early; a counter says one can prompt ChatGPT for endless nonsense to review\. The thread also disputes whether autonomous verification can replace human checking, and one reply notes programming\-language abstractions are deterministic while LLMs are not\. Sources: [Comment 50004980]() · [Comment 50005168]() · [Comment 50005172]() · [Comment 50005561]() · [Comment 50006183]() · [Comment 50006231]() · [Comment 50006253]() · [Comment 50006524]() ### AI automation and the math/software career pipeline Commenters worry that automated math threatens the thriving mathematical community that catches errors, and one cites Tao on that risk\. Replies debate whether automation makes the community unnecessary or impossible to fully automate, and compare it to software: AI has reduced junior hiring, raising questions about how future senior engineers will emerge\. Others say companies will still need people to steer AI and make domain decisions, or that junior developers can still learn by building systems; one says the hysteria is overrated\. Sources: [Comment 50004661]() · [Comment 50005157]() · [Comment 50005815]() · [Comment 50005950]() · [Comment 50006289]() · [Comment 50006338]() · [Comment 50006362]() · [Comment 50007100]() · [Comment 50006558]() ### Responsibility for publishing flawed AI\-generated math One commenter calls the company irresponsible and says it should not be allowed to continue; replies ask how withdrawing a paper is irresponsible and argue the sentiment is motivated by fear of economic impact\. A counter\-argument says irresponsibility means not caring about wasted time or flooding the field with slop, not verifying Lean proofs against natural\-language proofs, and mining open problems; another says withdrawing after a flaw is how science is meant to work\. Sources: [Comment 50004921]() · [Comment 50004933]() · [Comment 50005030]() · [Comment 50005466]() · [Comment 50005062]() · [Comment 50004961]() ### Human mathematical community vs autonomous verification The thread debates whether a thriving human math community is needed to catch errors or whether autonomous verification suffices\. One asks for a link to someone pointing out an issue and is told an LLM found it; a reply says that is not the same as relying on a community\. Another argues OpenAI rushed without completing autonomous verification for every paper, while a reply calls the claim that verification already exceeds human math ridiculous because much knowledge is unwritten; a further reply disputes that and says AI systems outperform human mathematicians\. Sources: [Comment 50004661]() · [Comment 50004723]() · [Comment 50004774]() · [Comment 50004807]() · [Comment 50006183]() · [Comment 50006231]() · [Comment 50006253]() ### Unpaid verification and the open\-source abandonware analogy One commenter compares releasing flawed AI outputs to tech giants open\-sourcing abandonware, saying problems become others' to fix and unpaid human effort may verify AI outputs\. A reply challenges the unpaid premise, saying qualified analyzers are mostly paid researchers and that releasing abandoned source is a gift, citing id Software\. Another reply rejects gatekeeping around who can analyze proofs, saying anyone with a sufficiently advanced model can check them\. Sources: [Comment 50004803]() · [Comment 50005052]() · [Comment 50005181]() ## Sources & coverage AI-generated summary · 2026\-10\-08T14:02:05\.068813\+00:00 Based on 14 of 15 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. The model input was further shortened to fit its context limit. Article text may also be shortened. Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.