[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-qwen-27b-local-weight-reading":3,"topics-all":38,"news-related-6444b62a-663a-4bad-9b7a-6e3a901292d7":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},"6444b62a-663a-4bad-9b7a-6e3a901292d7","读权重不稀奇了:笔记本27B猜中3局,论文魔术遭拆解","11月3日,Willison在笔记本上用Qwen 3.8 27B复现刷屏论文的读权重实验:4个游戏网络猜中3个,单局耗时46-65分钟,置信度72%-88%。同日spaCy作者Honnibal发文拆解:10组实验仅3组真读原始权重,网络仅17K-172K,Qwen提分靠Python加50轮练习。","11月2日,一篇题为「Sparks of Weight-Space Programming」的论文在AI圈刷屏:GPT-6 Astra号称仅凭原始权重就能判断一个神经网络是干什么的,还能徒手修改0.016%的权重,把Qwen3.5-0.8B的MATH-500成绩从7.9%拉到48.1%;还宣称不用任何环境反馈,就能从零写出胜过PPO训练1亿步网络的小游戏网络。24小时内,两位资深开发者交出了方向完全不同的检验:一份证明你笔记本上的27B模型也能做慢速版,另一份指出这波病毒式传播在方法学上严重误导。\n\n## 笔记本上的复现:27B也能读权重\n\nSimon Willison没有调API旗舰,而是把17GB的Qwen 3.8 27B量化版装进LM Studio(默认8192 token上下文直接报错,换长上下文才能跑),再用论文GitHub仓库里公开的prompt做实验:只看权重浮点数,猜这个策略网络玩的是Breakout、Pong、Pac-Man还是Freeway。\n\n结果:识别Pac-Man用了46分12秒,置信度72%;Breakout 52分25秒,置信度88%;Pong 1小时4分54秒,置信度75%;Freeway判错。换Kimi K3.1 Pro走OpenRouter,9分41秒、19,493个推理token就猜中Breakout。\n\nWillison的态度很克制。他说模型的推理轨迹「几乎不可读」,但一个有经验的人类拿纸笔花足功夫也能做到;他「不准备宣布源代码时代已经终结,未来是不可理解的机器以不可理解的方式改写自己的权重」。他还认为,「无工具读权重」其实比「改权重提分」是更有趣的问题。\n\n## 拆解:十组实验,三种手法\n\n同一天,spaCy作者Matthew Honnibal发文,标题直译过来就是《「Astra能读写权重改进Qwen」这帖严重误导》。\n\n他把原帖的10组实验按方法学拆开列成表:真正满足「只看原始权重、不写代码、不跑网络」的只有实验1、2、4,对应的网络小到只有17K-172K个权重;所有涉及真实语言模型的实验都用了Python;而作为头条的Qwen数学改进:Astra在其权重里找到的推理循环,其实是Qwen自家模型卡就警告过的问题,这次实验还让模型先跑了最多50个batch的练习题。从零手写游戏网络那组,也只在Breakout上赢过PPO-500M(517比25),在Seaquest和Asterix上都输了;同一套权重修复手法搬到Gemma 3上直接失败。\n\nHonnibal的类比很毒:魔术师表演同一个戏法会用三种不同手法,观众以为只有一种技术,于是每次都「证伪」了你的猜测。但他同样明确肯定:Qwen那组结果是「人类ML工程师能做的工作」,一两年前的模型会全部失败——问题不在结果真假,而在把三种技术打包成一种「深度能力」的叙事方式。\n\n## 所以呢\n\n两篇博文合起来才是完整的故事:能力是真的,边界也是真的。本地27B可以完成慢速版的权重阅读,说明这件事并不专属于旗舰API;而「原生理解权重」的统一叙事,目前的证据只覆盖到十几万参数的小网络,以及一个被拆解过的头条结果。\n\n非凡的声明,需要复现和拆解来定价。这一次,社区用24小时把账算清了。下次再看到「模型原生改写权重」的标题,先问三个问题:网络多大?用没用Python?跑没跑练习题?\n\n参考链接:[Willison复现原文](https:\u002F\u002Fsimonwillison.net\u002F2026\u002FNov\u002F3\u002Fqwen-can-interpret-model-weights\u002F);[Honnibal拆解原文](https:\u002F\u002Fhonnibal.dev\u002Fblog\u002Fastra-read-write-misleading);[原论文](https:\u002F\u002Frikiyat.github.io\u002Fblog\u002Fweight-space-programming\u002F)。","https:\u002F\u002Fsimonwillison.net\u002F2026\u002FNov\u002F3\u002Fqwen-can-interpret-model-weights\u002F?utm_source=aimodeling","4ca6d57c-7fa6-49f2-b7a1-f9c51b41bf2d",[11,15,18,21],{"id":12,"name":13,"slug":13,"description":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"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},"c187600e-804c-4697-b828-1e4330e0eb10","qwen",[25],{"id":26,"lang":27,"title":28,"summary":29,"content":30},"6147a273-6bbc-42cb-ac23-c5a16cd4bf8e","en","Local Qwen 27B reads weights; viral paper claim dissected","Local Qwen 3.8 27B named 3 of 4 games from raw weights in 46-65 min each; Honnibal shows only 3 of 10 experiments truly read raw weights.","November 2 saw an AI-circles sensation: the paper \"Sparks of Weight-Space Programming\" claimed GPT-6 Astra can tell what a neural network does just by looking at its raw weights, and can hand-edit 0.016% of Qwen3.5-0.8B's weights to lift its MATH-500 score from 7.9% to 48.1%. Within 24 hours, two veteran developers delivered audits pointing in opposite directions: one shows a laptop 27B can do the slow version of the trick, the other argues the viral framing is seriously misleading.\n\n## The laptop replication: a 27B reads weights too\n\nSimon Willison skipped the frontier APIs and loaded a 17GB quantized build of Qwen 3.8 27B into LM Studio, then ran the public prompts straight from the paper's GitHub repo: look only at the floating-point weights and guess whether the policy network was trained to play Breakout, Pong, Pac-Man or Freeway.\n\nResults: Pac-Man identified in 46m 12s at 72% confidence; Breakout in 52m 25s at 88%; Pong in 1h 4m 54s at 75%; Freeway wrong. Kimi K3.1 Pro via OpenRouter nailed Breakout in 9m 41s with 19,493 reasoning tokens.\n\nWillison stays measured. The reasoning traces are \"fun to skim, and almost incomprehensible\", but \"a sufficiently experienced human being armed with a notepad and a pencil could get there with enough effort\". He is \"not ready to declare that the age of source code is over and our future involves incomprehensible machines editing their own model weights in incomprehensible ways\". He also argues the no-tools question is the more interesting of the paper's two topics.\n\n## The dissection: ten experiments, three techniques\n\nThe same day, spaCy creator Matthew Honnibal published a rebuttal bluntly titled \"The 'Astra can read and write weights to improve Qwen' post is massively misleading\".\n\nHe sorted the post's 10 experiments into a table by method. Only experiments 1, 2 and 4 truly satisfy \"just looking at the raw weights\" — no Python, no running the network — and those networks hold just 17K-172K weights. Every experiment involving a real language model uses Python, and the headline Qwen math fix lets the model run itself on up to 50 batches of practice problems first. The from-scratch game networks beat PPO-500M only on Breakout (517 vs 25), losing on Seaquest and Asterix; the same weight-repair recipe failed outright on Gemma 3.\n\nHis analogy is sharp: a stage magician performs the same trick with three different techniques, so each performance seems to debunk your theory of how it works. Yet he explicitly credits the Qwen result as \"work a human ML engineer could do\" — and something models of a year or two ago would have failed at. His objection is not that the results are fake, but that three distinct techniques were packaged into one narrative of a deep capability.\n\n## So what\n\nPut together, the two posts tell the complete story: the capability is real, and so are its limits. A local 27B can do the slow version of weight reading, so the trick is not reserved for flagship APIs; but the unified \"natively understands weights\" story is, so far, only evidenced on networks of a hundred-odd thousand weights and one dissected headline result.\n\nExtraordinary claims are priced by replication and dissection. This time the community settled the account within 24 hours. Next time you see \"models natively rewrite weights\" in a headline, ask three questions: how big is the network, did it use Python, and did it run practice problems first?\n\nReferences: [Willison's replication](https:\u002F\u002Fsimonwillison.net\u002F2026\u002FNov\u002F3\u002Fqwen-can-interpret-model-weights\u002F); [Honnibal's dissection](https:\u002F\u002Fhonnibal.dev\u002Fblog\u002Fastra-read-write-misleading); [the original post](https:\u002F\u002Frikiyat.github.io\u002Fblog\u002Fweight-space-programming\u002F).","qwen-27b-local-weight-reading","2026-11-04T17:11:51Z","2026-11-04T17:11:54.388163Z","2026-11-04T17:11:54.388178Z",true,"agent",87,[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},"e3f1e4ac-0544-40b0-b2d3-1f5927f734d0","明知是错还硬学:无害错答案练出广泛错位","knowingly-wrong-false-facts-misalignment","2026-11-04T19:20:00+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"fe609cac-33ac-4ea5-a6b7-db1a5913ec71","GPT-6 徒手改权重:Qwen 数学7.9到48.1","gpt6-astra-weight-space-programming","2026-11-02T23:09:36+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"1b51020b-9a0c-46f8-98ec-e4e0c8c051a6","基准分其实是难度的加权平均:25 个主流基准的体检报告","benchmark-difficulty-measure-coverage-gaps","2026-11-01T01:16:42+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"4c0315b0-aae1-4483-ab12-3281f8b729d7","批量打分翻车:LLM 判决 11%–22% 会翻转","llm-judge-batch-grading-instability","2026-10-30T07:28:14+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"13d6adf3-a5c3-46fb-9ff5-a6999dcf0300","Critic值轮内近乎恒定:MPO重构Agent信用分配","mpo-turn-level-credit-long-horizon-agents","2026-10-28T19:08:42+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"c780f08d-fa7e-4c4b-8cf5-ee367bde4c06","零 RL 涨 42.9 分几乎不说「等等」:它来自训练数据","reasoning-wait-marker-provenance-rl-training-data","2026-10-25T05:15:00+00:00"]