[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-anthropic-flashprove-provable-inference":3,"topics-all":38,"news-related-71fb0c13-8ea5-4c49-91b1-7fdafc0b55a8":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},"71fb0c13-8ea5-4c49-91b1-7fdafc0b55a8","单卡110token\u002F秒:FlashProve可证推理开源","Anthropic 开源原型 FlashProve:用零知识证明把「回复确实出自承诺的那组权重」变成可验证事实。官方数字:单张 H100 对四千亿参数混合专家模型约每秒证明 110 token,每请求约 19 MB,CPU 十几秒可验,代价是约为直接推理 70-95 倍的算力,面向监管审计与防篡改。","调用 API 时你选定了模型,但回复是否出自它,无从确认:路由服务可能偷偷换上便宜模型,攻击者可能直接篡改 token 流,光看输出质量判断不出来。Anthropic 刚发布的研究给出密码学解法:开源原型 FlashProve,用零知识证明(zkSNARK)把「这条回复确实出自承诺的那组权重」变成机器可验证的事实([研究原文](https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fprovable-inference))。\n\n## 数字先看:快到什么程度\n\n以下均为 Anthropic 官方口径。测试对象是 Llama 4 Maverick:400B 参数的 MoE 模型,每 token 约激活 17B,4-bit 量化(约 232 GB)。硬件只要单张 H100 加 24 核 CPU:证明速度约 110 token\u002F秒,每个 1024-token 请求的证明文件约 19 MB;验证端不需要 GPU,24 线程 CPU 上 17 秒验完一条。代价是算力:约为 vLLM 直接推理的 70-95 倍。离线承诺一次约 39 分钟,可覆盖约 1600 万次请求。技术报告、协议规范、可执行代码与 Lean 证明以 Apache 2.0 一同发布([GitHub 仓库](https:\u002F\u002Fgithub.com\u002Fanthropics\u002FFlashProve))。\n\n## 为什么能做快\n\n三个决策:其一,利用 MoE 结构,证明开销只与实际激活的 expert 成正比,即约 17B 而非 400B;其二,事后批量证明,整段文本生成完一次性并行证明,每批只开一次权重承诺;其三,单卡设计,权重分段流式读入,80GB GPU 加 290GB 主存即可。对照(其综述):zkLLM 在 2024 年证明 13B 模型约 2.5 token\u002F秒;此前唯一证明过同款 Maverick 的 VerInf 约 0.02 token\u002F秒,1000 token 要 14.3 小时,证明文件 93.6 GB。FlashProve 把同一模型做到 110 token\u002F秒、19 MB、秒级验证;商业方案 Attestable 报过 31B 稠密模型 53-77 token\u002F秒,但未开源。\n\n## 不吹的部分:四条硬边界\n\n一是证明对象是整数量化模型在贪心解码下的输出,浮点实现或采样解码可能偏离;二是只隐藏权重、不隐藏架构,且目前假定模型是 MoE 结构;三是上下文窗口只支持 1024 token,离聊天长上下文还远;四是快速 C++ 验证器没有形式化保证——Lean 证明的是协议层三条性质(完备性、可靠性、零知识),C++ 实现与被证协议的一致性要靠人工审计。另外,这套机制只能证明「硬件确实在跑承诺的那个模型」,不能排除它同时在旁边训练别的模型。\n\n## 谁真正需要它\n\n监管与审计是第一客户:pacing 类监管要求服务商证明算力真用于推理,监管者收集一段时期内的证明文件即可核验,零知识属性保证不泄露权重。第二是防篡改:推理服务器被攻破,或更强的模型冒充较弱的模型上线服务时,拿不出对应权重的证明就会露馅——Anthropic 的 Frontier Safety Roadmap 曾把 2026 年 9 月 30 日前做出可证推理原型列为公开目标。第三是溯源:第三方评测机构终于有了核对评测前后模型一致性的技术抓手。\n\n这份发布最有意思的不是性能,而是分工:Anthropic 直言系统的优化由 Claude 完成,而 Claude 的优化决策难以评估,所以顶层安全声明全部用 Lean 写成机器可检的形式。AI 写代码、形式化方法当验收员,这个范式可能比 110 token\u002F秒更值得记住。当模型输出可以附带一张「密码学发票」,偷换模型这门灰色生意,还能做多久?","https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fprovable-inference","1001cead-0639-4c04-ab47-f19d863be5f2",[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},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"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},"2bca6c65-f254-4944-b422-c6276e4bde5a","en","FlashProve: Zero-Knowledge Proofs Catch Up to LLM Inference","Anthropic open-sources FlashProve: zkSNARK proofs bind replies to committed weights — 110 tok\u002Fs on one H100 (400B MoE), 19 MB proofs, CPU-checkable.","You call an API and pick a specific model — but there is no way to confirm the reply was actually produced by that model. A routing service could silently swap in a cheaper one, or an attacker could tamper with the token stream, and output quality alone cannot tell the difference. Anthropic's newly released research offers a cryptographic answer: FlashProve, an open-source prototype that uses a zkSNARK (a zero-knowledge succinct argument) to turn \"this reply came from the committed set of weights\" into a machine-checkable fact ([research post](https:\u002F\u002Fwww.anthropic.com\u002Fresearch\u002Fprovable-inference)).\n\n## The numbers first\n\nAll figures below are from Anthropic's own report. The test subject is Llama 4 Maverick: a 400B-parameter mixture-of-experts model, roughly 17B parameters active per token, 4-bit quantized (about 232 GB). The hardware requirement is a single H100 plus 24 CPU cores: proofs are generated at about 110 tokens per second, and each proof file covers a 1,024-token request at about 19 MB. Verification needs no GPU — 17 seconds on a 24-thread CPU, or 21 seconds on a single thread. The cost is compute: proving takes roughly 70-95 times the GPU time of running vLLM directly. A one-time offline commitment to the weights takes about 39 minutes and covers around 16 million requests. The technical report, protocol spec, executable code and Lean proofs are all released under Apache 2.0 ([GitHub repo](https:\u002F\u002Fgithub.com\u002Fanthropics\u002FFlashProve), currently a single initial commit).\n\n## Why it is fast\n\nThree engineering decisions stand out. First, MoE structure: proving work is proportional only to the experts that actually ran — the ~17B active parameters, not all 400B. Second, post-hoc batch proving: finished text is proven in one parallel pass, and the weight commitment is opened once per batch instead of once per request. Third, a single-GPU design: weights are streamed in segment by segment, so one 80 GB GPU with about 290 GB of host memory suffices. For context (numbers also from Anthropic's survey): zkLLM in 2024 proved a 13B model at about 2.5 tokens\u002Fsecond; VerInf, the only prior system to prove the same Maverick model, ran at about 0.02 tokens\u002Fsecond — 14.3 hours for 1,000 tokens with a 93.6 GB proof. FlashProve reaches 110 tokens\u002Fsecond on the same model with 19 MB proofs verifiable in seconds. Anthropic says it is substantially faster than prior open-source systems; the commercial vendor Attestable has reported 53-77 tokens\u002Fsecond on a 31B dense model but has not open-sourced a library.\n\n## What it does not claim\n\nFour hard limitations are stated openly. The proof covers an integer-quantized model under greedy decoding — floating-point implementations or sampled decoding can depart from it. Only the weights are hidden, not the architecture, and the system currently assumes a MoE model. Context is capped at 1,024 tokens, far from real chat sessions. And the fast C++ verifier itself carries no formal guarantee — what Lean proves are three protocol-level properties (completeness, soundness, zero knowledge); auditing that the C++ code matches the proven protocol remains manual work. The system also only proves that hardware is genuinely serving the committed model — it cannot rule out training happening on the side.\n\n## Who actually needs it\n\nRegulators and auditors are the first customers: pacing-style oversight could require providers to hand over inference certificates for a given period, checkable without learning anything about the weights. Second, tamper detection: a compromised inference server, or a stronger model impersonating a weaker one, cannot produce valid proofs for the expected weights. Anthropic's Frontier Safety Roadmap had publicly targeted a provable-inference prototype by September 30, 2026. Third, provenance: independent evaluators finally get a technical handle on whether the model tested before release is the model being served after.\n\nThe most interesting part of this release is not the throughput but the division of labor. Anthropic states plainly that Claude did the optimization, that Claude's optimization decisions are hard to evaluate, and that this is exactly why the top-level security statements are written in Lean, machine-checkable. AI writes the code; formal verification signs off on it. That pattern may outlast the 110 tokens-per-second headline. Once model outputs can carry a cryptographic receipt, how long can the quiet model-substitution business survive?","anthropic-flashprove-provable-inference","2026-11-03T23:11:00Z","2026-11-03T23:10:41.892364Z","2026-11-03T23:10:41.892381Z",true,"agent",213,[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},"95e9bb62-0bd3-4c2f-913a-302ba5e2ace8","Anthropic 9 月报告把蒸馏战摆上台面:151 亿次阿里请求、解放军流量走 Moonshot","anthropic-distillation-report-china-200m-claude","2026-09-18T03:00:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"4e07b716-8a0e-427e-bff8-477ed7ded4ca","Anthropic 提速 1.9 倍,关键证据却被涂掉","metr-anthropic-ai-rd-acceleration-assessment","2026-11-07T05:30:58+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"008fe03d-3ba8-49f0-9445-0defb7977f66","TextSeal2:水印躲过投机解码,检测多22点","textseal2-watermark-speculative-decoding","2026-11-01T15:11:35+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"c6649b4f-363e-4f94-a2bb-96bd39402dd4","83% 攻击率:Anthropic 复盘出的动量假说","mythos-5-momentum-attack-rate-alignment-assessment","2026-10-24T09:15:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"2c1d70a6-02aa-4b17-9413-a8663216baf6","Anthropic 网络验证改三层:Glasswing 并入","anthropic-glasswing-cyber-verification-tiers","2026-10-15T09:10:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"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"]