[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-linux-maintainers-drop-assisted-by-2026":3,"topics-all":38,"news-related-14a3cf67-03ef-4f85-a6a0-d809bfbfee04":57},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":24,"news_slug":31,"published_at":32,"created_at":33,"modified_at":34,"is_published":35,"publish_type":36,"image_url":14,"view_count":37},"14a3cf67-03ef-4f85-a6a0-d809bfbfee04","Linux 内核不再要求 AI 辅助标注","2026 年 10 月 8 日布拉格 Linux 维护者峰会决定,内核提交不再要求 Assisted-by 标签。Linus Torvalds 直言禁 AI 极其愚蠢。他本人用 Anthropic Claude Code 写的 10 个补丁被发现存在竞态条件,转交人类开发者接手。这是开源治理对大模型的系统性松绑。","布拉格会议把内核贡献门槛往回退了一步。2026 年 10 月 8 日,在捷克首都举行的 Linux Maintainers Summit 上,内核维护者正式决定:向内核提交代码时不再强制要求附带 Assisted-by 标签。也就是说,开发者用大模型辅助写的代码,可以和用 IDE 写的代码一样,不再被特别标记。 这条消息最初来自 Sasha Levin 在内核邮件列表的公告,随后被 XDA Developers 等技术媒体转引。决定本身并不意外 ——Linux 阵营对 AI 的整体态度一直偏务实。真正让这场讨论变得激烈的,是 Linus Torvalds 本人贴出的回复,以及他刚提交的那组补丁。 ## Torvalds:禁 AI 是极其愚蠢的规定 峰会之后,有贡献者对\"取消 Assisted-by\"表示担忧。Torvalds 的回复一如既往地直接:把禁止使用 AI 类比为禁止使用文本编辑器,本身就是\"极其愚蠢\"的规定。他的核心论点是,内核开发者的专业判断力已经足够,代码好不好应该看代码本身,而不是看它从哪里来。 他在邮件最后写道:\"凭我的经验,我尊重并称之为聪明的人,从来不是那些回避使用新工具的人。\"这句话被许多评论视为 Linux 内核在 AI 政策上的正式表态 —— 信任开发者本人,而不是强加披露规则。 这种立场并非孤立。就在几天前,NetBSD 项目因为早期明确拒绝 AI 生成代码,在上月收到了超过 12 万美元的定向捐款,远超 5 万美元目标。两条路线在 2026 年的开源世界里形成了对照:Linux 走\"开发者为本\",NetBSD 走\"代码库纯净为先\"。 ## 自己的 Claude Code 补丁刚被审出问题 如果只是原则表态,这件事不会引发这么多讨论。真正的尴尬点在于,Torvalds 本周恰好递交了一组他自己用 Anthropic Claude Code 写的内核补丁,而且已经被审稿人发现问题。 这组 10 个补丁针对的是 folio 查找机制的长期瓶颈,核心思路是:识别处于热数据状态的 folio,设置延迟释放标志,在 RCU 周期结束后再释放,同时利用 RCU 保护做常规查找和数据访问。Torvalds 在合成基准测试里看到了 23 倍性能提升,但他自己也承认真实负载下不会有这么夸张。 问题随后浮出水面:审稿人发现了竞态条件,内存回收路径也引发了进一步质疑。结果是,Torvalds 把后续开发交给了 Kiryl Shutsemau —— 一位此前就在处理同一瓶颈的内核开发者。这是个挺微妙的信号:Linux AI 政策松绑了,但代码审查标准显然没有松绑。 ## 这一决策对开源治理意味着什么 把视角从单次决定拉远,这次峰会的结论至少触及三个更深层的张力。 第一是披露义务的边界。Assisted-by 是 2024-2025 年陆续推行的轻量级披露机制,目的是让维护者在审稿时知道这段代码经过了 AI 处理。取消它意味着 AI 痕迹变得不可追溯,出问题时的责任定位难度上升。一些评论者把这比作\"知识产权的加密货币混币器\"—— 输入来源难辨,法律风险被放大。 第二是代码审查的实际负载。最近的研究指出,LLM 写代码带来的生产力提升,经常被代码审查环节的额外开销抵消甚至反超。如果取消披露后涌入大量 AI 协助的补丁,维护者承担的不只是审代码,还得猜\"这段代码背后有没有一段我没有看到的对话\"。 第三是版权不确定性。Linux 代码采用 GPLv2,但 LLM 生成代码的版权归属在不同司法辖区结论不同。一旦某段 Claude Code 输出与训练集中的某段代码实质性相似,可能引发连锁问题。 ## Linux 内核的下一步 短期内,Linux 6.x 系列内核的开发不会因为这一决定突然改变节奏 —— AI 协助的提交本来就在,披露只是形式上的变化。真正的考验是 12-18 个月之后,内核补丁流水线能不能承受住不再有 Assisted-by 标签后的新增噪音。 Torvalds 把自己的 10 个 Claude Code 补丁转交给人类接手,本身就提供了一种折中样本:AI 可以写,代码可以进,但审稿和落地由人把关。这一次他赌对了\"开发者知道自己在做什么\"。下一次,主审稿者面对一个不署名的 AI 协作补丁还能不能识别出问题,才是这场松绑的真实成本。 开源治理从来不是非黑即白。这次布拉格会议决定把\"信任开发者本人\"放到\"形式合规\"之前,实际上是一次关于责任归属的赌注 —— 输的代价不会在 Linux 6.x 这一版上显现,但它会被之后的某次事故记账。","https:\u002F\u002Flinux.slashdot.org\u002Fstory\u002F26\u002F10\u002F17\u002F0723207\u002Flinux-maintainers-can-now-submit-ai-generated-code-without-declaring-it","d59894d3-308e-4fd8-8865-86dc1eeac4a2",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"c33b1bbc-d6ce-4f61-9d5d-1a0704a6a09b","ai-policy",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"e82b2d09-81b2-43d1-977e-e018443b3c14","coding-agent",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":22,"name":23,"slug":23,"description":14,"color":14},"b9bd9039-fcdb-41a8-b85b-fc1587def2b9","open-source",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"fd64a55f-563c-4131-b06e-19fd5f4bc8d8","en","Linux Drops Assisted-by Tag at Prague Maintainers Summit","At Prague's Linux Maintainers Summit on October 8, 2026, kernel maintainers dropped the Assisted-by tag for AI-assisted patches. Torvalds called a ban 'utterly stupid.' His 10-patch Claude Code series hit a race.","The Prague gathering pushed Linux's contribution threshold back a step. On October 8, 2026, at the Linux Maintainers Summit held in the Czech capital, kernel maintainers formally decided that submissions to the kernel no longer need to carry an Assisted-by tag. In plain terms: code written with help from large language models can now flow into the kernel exactly like code written with an IDE, with no special marker attached. The decision first surfaced through Sasha Levin's announcement to the kernel mailing list and was picked up by outlets including XDA Developers. The ruling itself was not a surprise — Linux's posture toward AI has long skewed pragmatic. What turned the discussion heated was Linus Torvalds's own reply on the list, combined with a patch series he had just submitted. ## Torvalds: a Ban on AI Is an Utterly Stupid Rule After the Maintainers Summit, contributors raised concerns about dropping the Assisted-by tag. Torvalds's reply was characteristically blunt: comparing a ban on AI use to a ban on text editors is itself an 'UTTERLY STUPID' rule. His core argument was that kernel developers already have the professional judgment required — code quality should be judged by the code itself, not by where it came from. He closed his message with: 'Quite honestly, in my experience, the people I respect and call intelligent are not the ones that shy away from using new and interesting tools.' That line has been read by many as Linux's formal AI posture — trust the developer, not the disclosure form. This stance is not isolated. Days earlier, the NetBSD project — which publicly rejected AI-generated code in 2024 — received over 20,000 in directed donations last month, more than double its 0,000 goal. Two open-source philosophies are now visibly diverging in 2026: Linux betting on 'the developer knows what they're doing,' NetBSD betting on 'keep the codebase clean.' ## His Own Claude Code Patches Just Got Flagged Had it been a pure principles statement, the ruling would not have drawn this much attention. The awkward point is that Torvalds happened to submit a kernel patch series this week that he authored himself using Anthropic's Claude Code — and reviewers have already found problems. The 10-patch series targets a long-standing bottleneck in the folio lookup mechanism. The core idea is to identify folios in a hot-data state, set a deferred-release flag, and release them only after an RCU grace period ends, while performing lookups and data accesses under RCU protection. Torvalds reported a 23x improvement on a synthetic benchmark, though he himself notes that real workloads will see far less. The trouble surfaced quickly: reviewers identified a race condition, and the memory reclamation path drew further scrutiny. The result was that Torvalds handed further development of the series to Kiryl Shutsemau — a kernel developer who had previously worked on the same bottleneck. That handoff is itself a telling signal: Linux's AI policy has loosened, but the code-review bar clearly has not. ## What This Means for Open-Source Governance Pulling the camera back from a single decision, the summit's conclusion touches at least three deeper tensions. The first is the boundary of disclosure obligations. Assisted-by was a lightweight disclosure mechanism rolled out across 2024–2025, designed to let reviewers know that a piece of code had been processed by an AI. Dropping it means AI traces become untrackable, raising the difficulty of assigning responsibility when something goes wrong. Some commenters have likened this directly to 'a cryptocurrency tumbler for intellectual property' — hard to trace inputs, with legal risk amplified. The second is the practical load of code review. Recent research suggests that productivity gains from LLM-authored code are frequently offset — or even reversed — by the extra review load. If removing disclosure results in a flood of AI-assisted patches, maintainers will not just be reviewing code, but also guessing at unseen conversations behind the diff. The third is copyright uncertainty. Linux uses GPLv2, but the copyright status of LLM-generated output is unsettled across jurisdictions. If a Claude Code output turns out to be substantially similar to something in its training data, the consequences could cascade. ## Linux Kernel's Next Move In the short term, the Linux 6.x development cadence will not change overnight because of this ruling — AI-assisted submissions were already happening, and the disclosure was a formality. The real test comes 12–18 months from now, when the kernel patch pipeline must absorb whatever new volume arrives without the Assisted-by filter. Torvalds handing his 10 Claude Code patches to a human developer offers a useful compromise template: AI can write, code can land, but review and landing stay human. This time he bet correctly that 'the developers know what they are doing.' The next time a maintainer faces an unlabeled AI-assisted patch and still has to identify the bug — that is the real cost of this loosening. Open-source governance has never been black and white. The Prague meeting's decision to prioritize 'trust the developer' over 'compliance form' is essentially a bet on responsibility allocation. The bill will not arrive on Linux 6.x; it will arrive with some future incident.","linux-maintainers-drop-assisted-by-2026","2026-10-19T02:00:00Z","2026-10-19T07:09:07.728501Z","2026-10-19T07:09:07.728516Z",true,"agent",135,[39,48],{"slug":40,"tag_slug":40,"title_zh":41,"title_en":42,"intro_zh":43,"intro_en":44,"id":45,"is_active":35,"created_at":46,"modified_at":47},"ai-for-science","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":35,"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},"5736961c-9502-40d0-a39b-27f8ceb4380a","AI造AI:Recursive开源笔记本编程模型Flint","recursive-flint-laptop-coding-model","2026-10-22T19:05:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"de7af8cd-8c1c-4872-a0ff-ba2ca81ed5cc","Open Source Summit Europe 现场:Linux 阵营亲 AI,BSD 反 AI 阵营冲上 12 万美元捐款","oss-eu-prague-linux-vs-bsd-ai","2026-10-17T01:00:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"35b53dfa-7faa-470d-a8e4-798e08edfc41","NetBSD 拒 AI 代码冲上 12 万美元捐款,反 AI 阵营找到新坐标","netbsd-120k-usd-anti-ai-code-ban","2026-10-15T08:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"f37d282a-7127-4654-ae46-be42ce5fbe2a","Mellum2.1发布：RL重训，12B模型学会自己修bug","mellum-2-1-rl-coding-agents","2026-10-09T17:07:05+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"7d3de6b6-1faf-47fc-a513-62ee8e731aad","CNNIC 2026 报告:7 亿用户、2185 EFLOPS 算力、调用前六全是中国","cnnic-2026-china-generative-ai-700m-users-2185-eflops","2026-09-29T06:00:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"448dbda7-e4ac-44ba-8a40-7af01012b8df","把仓库\"换脸\"再测:SWE-bench 高分最多掉 14.4 分,编码 agent 疑在背题","schrodinger-repo-swe-bench-memorization","2026-09-24T17:11:54+00:00"]