[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-openai-4d-kakeya-math-results":3,"topics-all":39,"news-related-02620717-94d4-4853-bc81-932dbd06c2e0":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":25,"news_slug":32,"published_at":33,"created_at":34,"modified_at":35,"is_published":36,"publish_type":37,"image_url":15,"view_count":38},"02620717-94d4-4853-bc81-932dbd06c2e0","OpenAI模型解数百数学难题,陶哲轩谈数学2.0","OpenAI公开内部前沿模型产出的数百项数学新结果,声称包含四维挂谷猜想的解与黎曼假设方向的进展,大量证明已用Lean形式化并开源到GitHub。发布方式按数学家顾问组建议改良,但陶哲轩评论称AI公司迫使开创性研究保密,数学2.0时代应降低解难题的核心地位。","10月6日,OpenAI 在官网发布报告《[Sharing AI progress in mathematics](https:\u002F\u002Fopenai.com\u002Findex\u002Fsharing-ai-progress-in-mathematics\u002F)》,公开一批由内部前沿模型产出的数学新结果。这批结果放进 GitHub 仓库,附论文修订与引用协议,大量证明同步给出 Lean 形式化版本,可被计算机逐行验证。\n\n## 数百个结果里有什么\n\n据 Scientific American 报道,清单声称包含四维挂谷猜想的解、对若干最重要计算机算法的改进,以及朝黎曼假设方向的实际进展——注意是「进展」而非「解」。背景颇具戏剧性:今年菲尔茨奖得主王虹正是因证明三维挂谷猜想获奖,如今 OpenAI 的内部模型声称把同一问题推到了四维。多家科技媒体清点的口径约为 372 个问题家族、七百余篇论文,而消化这批结果预计要花数学界数月时间。\n\n## 发布方式的三个变化\n\n九月 OpenAI 放出「解决百余个开放问题」的说法后,菲尔兹奖得主们曾联名警告 AI 偏离数学本质。这次 OpenAI 按高等研究院 AGMAI 顾问组的公开建议做了改良:结果进 GitHub 带引用协议;公开 Lean 形式化;附上透明度材料——10 份模型推理摘要,以及按官方口径以 ChatGPT Pro 用量估算的算力开销,平均每个结果约相当于三小时 Pro 级思考。不过 Engadget 指出,委员会的部分建议仍被忽略:逐题算力时间与提示词都没有公开。\n\n## 陶哲轩:问题不在发布,在激励\n\nUCLA 数学家陶哲轩在社交平台发表长评(Solidot 编译)。他说数学 1.0 时代,知名难题的证明会引发演讲、研讨与精简,最终沉淀为下一代数学家的教材;而 AI 的证明由对数学兴趣不大的人用提示词「解决」,既无法出席研讨会,也无法与领域专家对话。更尖锐的判断是:AI 公司的抢发行为正迫使数学界把开创性研究秘而不宣。他的结论是,数学 2.0 时代应降低「抢先解决难题」的核心地位,把学术阐释、社区建设与开辟新方向提升为衡量进步的维度。\n\n## 所以呢\n\nLean 验证回答「对不对」,回答不了「谁理解了它」。当机器数小时堆出数百个结果,稀缺的不再是证明本身,而是把证明变成知识的人类流程——研讨、教材、传承。OpenAI 承诺资助系列研讨会来消化结果,方向对了;但只要抢发头条的激励还在,陶哲轩担心的保密循环就不会停。这对整个 AI 行业是个通用命题:生成速度超过消化速度时,产出越快,知识负债越多。","https:\u002F\u002Fopenai.com\u002Findex\u002Fsharing-ai-progress-in-mathematics\u002F","15975962-b5fe-49e5-ae68-687ba6cb7015",[11,16,19,22],{"id":12,"name":13,"slug":13,"description":14,"color":15},"9112951a-2abb-4214-b63a-385ec7afb2ba","ai-for-science","AI for Science 专题：追踪 AI 在生命科学、化学材料、物理世界模型等科学方向的关键突破",null,{"id":17,"name":18,"slug":18,"description":15,"color":15},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":20,"name":21,"slug":21,"description":15,"color":15},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":23,"name":24,"slug":24,"description":15,"color":15},"42e59a88-7795-47dc-a334-ef1e72c24347","openai",[26],{"id":27,"lang":28,"title":29,"summary":30,"content":31},"512472b1-cfc5-4cc4-91b8-36fa1b233cc1","en","OpenAI Model Claims 4D Kakeya Among Hundreds of Math Results","OpenAI posts hundreds of internal-model math results, claiming a 4D Kakeya solution and Lean proofs. Terence Tao: Math 2.0 must demote problem-solving.","On October 6, OpenAI published [Sharing AI progress in mathematics](https:\u002F\u002Fopenai.com\u002Findex\u002Fsharing-ai-progress-in-mathematics\u002F), releasing a broad batch of new mathematical results produced by an internal frontier model. The results live in a GitHub repository with protocols for paper revisions and citations, and many proofs ship with Lean formalizations that a computer can verify line by line.\n\n## What is in the hundreds of results\n\nPer Scientific American, the claimed haul includes a solution to the four-dimensional Kakeya conjecture, improvements to some of the world's most important computer algorithms, and actual progress toward the Riemann hypothesis — progress, not a solution. The backdrop is dramatic: this year's Fields Medalist Hong Wang earned the prize for proving the three-dimensional Kakeya conjecture, and now OpenAI's internal model claims to have pushed the same problem to four dimensions. Media tallies put the release at roughly 372 problem families and over seven hundred papers, and mathematicians may need months to digest it all.\n\n## Three changes in how it was released\n\nAfter September's claim of more than one hundred solved open problems drew a joint warning from Fields Medalists, OpenAI this time followed the public recommendations of the AGMAI advisory group at the Institute for Advanced Study: results went to GitHub with citation protocols; Lean formalizations were published; and transparency materials were attached — ten summaries of the model's reasoning, plus compute estimates. By the company's own accounting, the average result used roughly three hours of ChatGPT Pro-level thinking. Engadget notes, however, that some board suggestions were still ignored: per-problem compute times and the prompts used were not released.\n\n## Terence Tao: the problem is incentives, not disclosure\n\nUCLA mathematician Terence Tao wrote a long commentary (relay: Solidot). In the Math 1.0 era, he argued, proofs of famous problems triggered talks, seminars, digestion and simplification, eventually becoming textbooks for the next generation. AI proofs are 「solved」 by people with little interest in mathematics using prompts; they cannot attend seminars or converse with domain experts. His sharper claim: the behavior of AI companies is forcing mathematicians to keep groundbreaking research secret. His conclusion — in the Math 2.0 era, the field should demote the centrality of cracking famous problems, and elevate exposition, community building, and the opening of new directions as measures of progress.\n\n## So what\n\nLean verification answers whether a proof is correct, not who understands it. When machines can stack up hundreds of results in hours, the scarce resource is no longer the proof itself but the human process that turns proofs into knowledge — seminars, textbooks, transmission. OpenAI promised funding for workshops to digest the results, which points the right way; but as long as the land-grab incentive remains, the secrecy loop Tao fears will not stop. For the wider AI industry this is a general law: when generation outpaces digestion, faster output means more knowledge debt.","openai-4d-kakeya-math-results","2026-10-08T13:10:00Z","2026-10-08T13:08:38.588373Z","2026-10-08T13:08:38.588391Z",true,"agent",243,[40,48],{"slug":13,"tag_slug":13,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":36,"created_at":46,"modified_at":47},"AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":49,"tag_slug":49,"title_zh":50,"title_en":51,"intro_zh":52,"intro_en":53,"id":54,"is_active":36,"created_at":55,"modified_at":56},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":58},[59,64,69,74,79,84],{"id":60,"title":61,"news_slug":62,"published_at":63},"1e43b4fc-39fe-4cac-aaf9-57f82d5c0311","AI 公司与数学界「错位」:两个月三次刷屏,把同行评审甩在身后","ai-math-severe-misalignment-fields-medal","2026-09-15T10:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"982c5e1e-5274-442e-9237-abaf39e8ee3c","25 位菲尔茨奖得主联名公开信:AI 解题竞赛正在伤害数学","fields-medalists-ai-misalignment-math","2026-09-13T13:07:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"700ec55f-98a2-4016-b61b-df4d1046ef4f","一个符号错误,OpenAI 撤回 3 篇 AI 数学论文","openai-math-repo-first-retractions","2026-10-11T17:12:28+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"80d7cab1-b2a8-4644-913d-c4118f0cd7c8","五个开放问题有了答案:Muse Spark 与数学家合写六篇论文","muse-spark-open-math-problems","2026-10-05T13:40:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"b3632f86-c054-49ba-af6c-a09337ee6b5d","菲尔兹奖得主联名警告:AI 偏离数学本质","fields-medal-mathematicians-ai-misalignment-declaration","2026-09-22T07:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"f8d091db-ca3a-44f6-b004-0b5f8aba0bef","OpenAI 请来9位数学家,却管不住模型节奏","openai-math-advisory-group","2026-09-21T21:15:00+00:00"]