[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-netbsd-120k-usd-anti-ai-code-ban":3,"topics-all":38,"news-related-35b53dfa-7faa-470d-a8e4-798e08edfc41":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},"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 项目 9 月单月收到 120,282 美元捐款,远超 5 万美元筹款目标;该发行版 2024 年起就拒收 ChatGPT、Copilot、Code Llama 等大模型生成的代码,而 Linux 阵营却在大量融合 AI 工具,反 AI 立场正在吸引那些对 AI 代码污染抵触的开发者。","Linus Torvalds 在维护者峰会上承认,他用 Claude Code 协助完成了一组内核 folio 查找路径的补丁,8 线程小读吞吐被拉到 3 倍,第一个分支已经推到邮件列表进入审查。这件事在另一头激起了完全相反的水花——NetBSD 这个一向低调的 BSD 发行版,刚刚在 9 月份创下 120,282 美元的单月捐款纪录,远超 5 万美元的筹款目标(见 https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85608)。\n\n## NetBSD 为什么突然成为反 AI 代码的\"净土\"\n\nNetBSD 拒绝 AI 代码的立场并不是新闻。早在 2024 年,该项目就在开发者文档中明确表态:未经核心团队事先书面批准,由 GitHub Copilot、Microsoft、OpenAI 的 ChatGPT、Facebook\u002FMeta 的 Code Llama 等大模型生成的代码会被视为\"污染代码\"。在过去几年里,这条规则只是该项目内部一个不太起眼的执行细节。\n\n2026 年 10 月初,The Register 报道了 NetBSD 单月筹款 120,282 美元、远超目标的消息,刷新了外界对这个项目的认知。结合 Linux 阵营(从 Linus 本人到 Debian、Ubuntu、Red Hat)的亲 AI 表态,NetBSD 的古老规则正在变成一种反主流的代码纯度立场。在订阅\u002F捐款驱动的开源项目里,\"我们不接 AI 代码\"能够成为卖点,这件事本身就是 2026 年才会出现的现象。\n\n## 与此同步的,是 Linux 阵营把 LLM 写进了工作流\n\n同一时期的另一组事件给出了对照。Linux 主线刚刚收到 Torvalds 第一份基于 Claude Code 的页缓存补丁,他在补丁说明里直接承认代码、测试、前置审查清单的初稿都来自大模型;Debian、Ubuntu、Red Hat 在各自路线图里也加入了 AI 辅助编程相关模块。Linus 公开谈论用 LLM 修代码,本意是\"用它解决我自己手上的活\",但消息传播出来的潜台词是:Linux 项目已经默认承认 LLM 可以参与内核代码贡献,这件事进入了审查流程。\n\n两件事并行发生,但传播方向完全相反。Linux 把 LLM 写内核叫做\"性能提升 3 倍、数量级优化\";NetBSD 把 LLM 代码拦在门外叫做\"代码库的纯度\"。同一个技术能力,在两个阵营里被读出了完全不同的叙事。\n\n## 反 AI 阵营的卖点为什么有效\n\nNetBSD 的禁令之所以在 2026 年开始奏效,与 AI 编程工具的渗透速度快有关。当主流开发者每天都在 Copilot、Cursor、Claude Code、Codex CLI 里工作的时候,Linux 内核这样的高敏感项目就只能依赖贡献者自律。一旦某个贡献者把 AI 生成的代码当作\"我已经审过\"的提交物,审稿人很难分辨。\n\nNetBSD 的规则把这件事简单化:未经书面批准,不允许。这等于是把\"AI 代码污染\"的责任压回到贡献者本人身上,也给了下游使用者一个明确的合同保证。在金融、电信、医疗、政府这些强监管场景里,这样一个明确的政策声明,显然比 Linux 当前\"我们默认信任贡献者审查\"的策略更便于合规。\n\n更深一层,NetBSD 的代码基原本就强调可移植、极简、低层,这与当前云端 AI 工具链强调\"快速写出来\"的优先级不匹配。NetBSD 走的是相反方向的预算:少写、对每一行负责,而不是写得快、靠 AI 加速产出。\n\n## 反 AI 阵营不会变成主流,但会变成一个稳定坐标\n\nLinux、Debian、Ubuntu、Red Hat 不会因此停下 LLM 化进程,这是技术演进的自然方向。但 NetBSD、OpenBSD、Gentoo、FreeBSD 这一类坚持人工代码基底的发行版,反倒因为稀缺变得更有用武之地。在招聘、关键基础设施、长期维护项目里,\"我们这里不接 AI 代码\"会比过去任何时候都更容易被听见。\n\n12 万美元的月度筹款额不大,NetBSD 仍然不会有任何接近 Linux 的商业体量。但这一数字释放了一个信号:开源软件世界里有相当一批开发者,愿意为\"非 AI 友好\"的项目付费。Linux 阵营负责把 AI 编程的主线推进下去,NetBSD 阵营负责保留一个反 AI 的对照样本——这是开源生态正在分层的标志,而不是分裂。\n\n所以这次事件真正的意义不在 NetBSD 筹到了多少钱,而在开源软件第一次因为对 AI 代码的态度分化出了两条独立的资源线。一条钱流向\"用得更快\"的项目,另一条钱流向\"写得更慢、更干净\"的项目。这两条线可能在未来若干年里同时繁荣,而不是谁把谁吃掉。","https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85608","d59894d3-308e-4fd8-8865-86dc1eeac4a2",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",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},"8ba55f45-4aef-48fb-8daa-182644b4ae6e","en","NetBSD nets 120K USD via AI code ban","NetBSD raised 120,282 USD in September, double its goal, by banning AI code since 2024 as Linux adopts LLMs. The surplus shows a paying anti-AI base.","Linus Torvalds admitted at the Maintainers Summit that he used Claude Code to draft a set of kernel folio lookup patches, pushing 8-thread contended reads to roughly 3x the baseline, and the first branch has already landed on the mailing list under review (cf. https:\u002F\u002Fwww.solidot.org\u002Fstory?sid=85608). At the same time, the same news cycle pulled an opposite reaction out of an unlikely quarter: NetBSD, the deliberately quiet BSD derivative, just booked 120,282 USD in September donations, more than double its 50,000 USD funding target.\n\n## NetBSD as the unlikely AI-free refuge\n\nNetBSD has banned AI-generated code since 2024: its contribution guidelines list ChatGPT, GitHub Copilot, Microsoft tooling, Meta Code Llama and friends as sources of \"tainted\" code that require explicit written approval from the core team before submission. For years, this was an unremarkable internal rule.\n\nIn October 2026, The Register reported the donation spike (syndicated by Solidot and DiscoverBSD), giving the project a moment of unexpected visibility. Combined with the Linux ecosystem's open embrace of LLM tooling from Linus himself to Debian, Ubuntu, and Red Hat, NetBSD's older rule suddenly looks like a position rather than a footnote. In subscription\u002Fdonation-driven open-source projects, \"we don't take AI code\" only becomes a selling point in 2026.\n\n## While Linux pulls LLMs deeper into the workflow\n\nThe contrast lands harder because Linux made the opposite move in the same news cycle. Linus Torvalds shipped a Claude-Code-assisted page-cache patch series whose cover letter openly credits the AI for the initial code, tests, and pre-review checklist. Debian, Ubuntu, and Red Hat added LLM-coding components to their roadmaps. The implicit message from Linux — that AI-assisted contributions now default to acceptable until proven otherwise — is the inverse of NetBSD's stance.\n\nBoth communities can now quote each other. Linux says LLM kernel code delivers \"3x faster contended reads\". NetBSD replies that LLM code is a source of contamination regardless of measurable wins. The same capability, two opposing narratives — and both are technically defensible.\n\n## Why the AI-free stance pays off\n\nNetBSD's rule becomes valuable precisely because AI coding tools have become ambient. When most contributors use Copilot, Cursor, Claude Code, or Codex CLI daily, a high-stakes project like the Linux kernel relies on contributor self-reporting. Reviewers cannot reliably tell which patch was AI-generated, and that ambiguity is exactly the problem NetBSD is trying to remove.\n\nBy stating the rule upfront, NetBSD pushes the contamination question back onto the submitter. That makes downstream usage simpler: a consumer can read the project's policy and decide whether the input contract matches their regulatory environment. Finance, telecom, medical, government — all of these have stronger incentives to adopt projects with explicit AI-input policies than to rely on a contributor-reviews-contributor model.\n\nUnderneath, NetBSD's codebase culture emphasizes portability, minimalism, low layers — the opposite priority from cloud AI tools' push to \"write more, faster.\" NetBSD's budget is to write less, own every line, and slow down, which fits a buyer base LLM-native projects cannot serve.\n\n## Anti-AI camp won't go mainstream, but it will stay a stable coordinate\n\nLinux, Debian, Ubuntu, Red Hat will not pause their LLM integration; that's the natural arc of this tooling cycle. NetBSD, OpenBSD, Gentoo, FreeBSD — projects that hold the line on human-written code — become more useful precisely because they are rare. In hiring packets, critical-infrastructure procurement, and long-tenure maintenance contracts, \"we don't take AI code here\" is more legible than it has ever been.\n\nUSD 120,282 a month is not a lot of money. NetBSD will not approach Linux in corporate footprint. But the number signals something the open-source world has not seen: a meaningful cohort of developers willing to pay for an AI-unfriendly baseline. Linux keeps pushing the AI-assisted coding frontier. NetBSD keeps preserving the human-only counter-sample. That is layer formation, not forking.\n\nThe real meaning here isn't how much NetBSD raised. It's that open source has, for the first time, separated into two independent funding currents over a question of AI-assisted contribution. One current flows toward \"build faster.\" The other flows toward \"build slower, cleaner.\" Both can coexist for years. Neither has to eat the other.","netbsd-120k-usd-anti-ai-code-ban","2026-10-15T08:00:00Z","2026-10-15T11:09:52.904967Z","2026-10-15T11:09:52.904974Z",true,"agent",167,[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},"de219584-58fc-45ad-91ea-0049a5cbcf10","OpenAI 开源 Codex Security CLI:把 AI 安全检测塞进每个 PR","openai-codex-security-cli-opensource","2026-07-29T10:30:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"14a3cf67-03ef-4f85-a6a0-d809bfbfee04","Linux 内核不再要求 AI 辅助标注","linux-maintainers-drop-assisted-by-2026","2026-10-19T02:00:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"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":80,"title":81,"news_slug":82,"published_at":83},"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":85,"title":86,"news_slug":87,"published_at":88},"4977d1f4-c8f7-480c-aaa1-ec01d69f44d8","8B 拿 SFT+GRPO 打 685B MoE:KaliBench 把 LLM 网络安全工具调用拆到命令行级","kalibench-cybersecurity-cli-runtime-rewards","2026-10-03T05:00:00+00:00"]