[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-scientist-world-1000-research-environments":3,"topics-all":39,"news-related-158f2181-ecf3-4e5c-ac5a-16e5c45c9d9a":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":14,"view_count":38},"158f2181-ecf3-4e5c-ac5a-16e5c45c9d9a","千篇论文建成AI实验室:前沿模型74%胜过人类作者","Scientist World 从 arXiv 十四个月 124,816 篇论文里全自动筛出 1,000 篇，复现方法并打包成可执行环境，全程无人类审稿。16 个前沿模型 84%\u002F74% 的尝试超过原论文方法，GLM 5.2 领跑开源；Qwen3-8B 训练后 ideation 中位分涨到 2.6 倍。","人类科研人员习以为常的一件事，AI 科学家一直没有：一个能踩进去的世界。几百年的文献摆在那里，可读、可引用，却不可运行——论文里的方法跑不起来，后来者只能在静态文本上\"读\"研究，不能\"做\"研究。Taiming Lu、Sachin Konan 和 Zhuang Liu 三位研究者发布的 Scientist World 补的就是这块地板：把 arXiv 论文一篇篇变成可执行的研究环境，首批 1,000 个，全部由多智能体流水线自动构建，没有任何人类环节审稿。\n\n## 造世界：12 万篇论文筛出 1,000 间实验室\n\n筛选从 arXiv 核心 AI 类目近十四个月的 124,816 篇唯一论文开始，先用 MiniMax M2.7 做两轮快筛——质量带取前 25% 左右，类型只留以量化优化为主贡献的实证方法论文——剩 21,354 篇；再对研究质量与复现难度做细评，按几何均值效用做类别均衡，取出前 1,000 篇。源头每月新增约 9,000 篇，约 1,500 篇能活过筛选，这个世界会随文献继续生长。\n\n构建是九轮迭代的固定工作流：三个复现者、三个设计者、一个打包者、两轮独立校验，模型家族在 GPT、Claude、MiniMax、GLM 之间轮换，避免一家之言。单篇成本中位数 0.4 GPU 时，最高 70；1,000 篇合计烧掉 2,286 GPU 时、239,506 个计算任务、947,056 轮智能体交互、510 亿输入 token。\n\n质量不是口头承诺：99% 的环境里，复现方法的分数落在论文报告值的容差内；82% 赢过论文自带的 baseline；37% 的迭代里程碑被后续交叉检查打回重来，共 573 次；最终每个环境还必须通过一个未参与建造的独立校验者，1,000 个全部通过。\n\n## 考试：16 个前沿模型，74% 的时候赢了原作者\n\n每个环境平均带 3.3 个实验、32 个指标（可见与隐藏各 16）、59 个文件约 3,800 行代码，横跨 8 个领域。评测的一半指标永远对参赛者隐藏，分数反映泛化能力，而不是对着可见半区调参。计分用作者定义的 FNI：0 等于复现论文原方法，负值更差，前沿模型队列的平均分锚定在 100 附近。\n\n16 个模型来自 7 家：Anthropic 的 Fable 5.1\u002F5 与 Opus 4.8\u002F4.7、OpenAI 的 GPT-6 Astra 到 GPT-5.4 四款、智谱 GLM 5.2\u002F5.1、月之暗面 Kimi K2.7\u002FK2.6、MiniMax M3\u002FM2.7、小米 MiMo V2.5 Pro、DeepSeek V4，分别经 Claude Code 与 Codex 等脚手架进入环境。结果：改进已有方法时，84% 的尝试超过原方法；从零设计新方法时，这个数字是 74%——AI 科学家在大多数环境里，做得比论文作者发表时更好。榜单上 Fable 5 是改进现有方法的冠军，Fable 5.1\u002F5 包揽设计新方法前二；开源阵营最接近前沿的是 GLM 5.2，设计新方法排第三。\n\n这些尝试没有被扔掉：平均每次 5.4 小时、6 次方法修订、25 个实验、35 次 commit，全部以版本库形式保存，合计约 6,400 小时的完整科研轨迹。\n\n## 训练场：8B 小模型的想法涨到 2.6 倍\n\n世界建好，还能反过来当健身房。作者用 Tinker 训练 Qwen3-8B（LoRA rank 32）：先在 1,000 条轨迹样本上做 SFT，再对 1,500 条已打分的想法做 GRPO。8B 模型还扛不动长程实现，作者干脆把\"出想法\"与\"做实现\"拆开——训练后的模型只负责 ideation，实现统一交给固定的 GPT 5.6 Sol，在 10 篇 held-out 论文上同条件执行，唯一变量是想法的来源。中位 FNI 从基线的 41 到 SFT 后 84、GRPO 后 106，约 2.6 倍（即摘要里的 ↑158%）；均值从 51 升到 90，够到前沿队列的锚。\n\n## 所以呢\n\n这项工作真正硬核的不是 74% 这个数字，而是把\"会做科研\"从形容词变成了可测量的工程问题：复现容差、隐藏半区、独立校验、FNI 归一化，每一层都在防作弊。37% 的打回率同样值得记住——全自动流水线的可信度不来自信任，来自互相挑错。当模型能力按月迭代，瓶颈正从\"脑子够不够聪明\"挪向\"世界够不够大\"：文献是现成的，缺的是把文献变成可运行环境的那条流水线，而这条流水线每月还有约 1,500 篇的产能。AI 科学家最缺的从来不是智能，而是一个允许它反复失败的世界。\n\n参考：[Scientist World: Grow a Research World for AI-Scientists](https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.08478)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.08478","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,19,22],{"id":12,"name":13,"slug":13,"description":14,"color":14},"6ad31a14-c0da-42df-81fd-564281f768db","agentic-ai",null,{"id":16,"name":17,"slug":17,"description":18,"color":14},"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",{"id":20,"name":21,"slug":21,"description":14,"color":14},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"id":23,"name":24,"slug":24,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[26],{"id":27,"lang":28,"title":29,"summary":30,"content":31},"05d25a3c-6d84-4420-9da9-b198c80a577d","en","1,000 Papers Become AI Labs; Models Beat Authors 74%","Scientist World turns 1,000 arXiv papers into runnable environments, no human in the loop; frontier models beat the original methods 74% of the time.","Human researchers take for granted something AI scientists never had: a world to work in. Centuries of literature sit there — readable, citable, but not runnable. The methods described in papers cannot be executed, so an AI can only read research on static text, not do it. Scientist World, released by Taiming Lu, Sachin Konan and Zhuang Liu, fills in that missing floor: it turns arXiv papers one by one into executable research environments — 1,000 in the first release, built entirely by a multi-agent pipeline with no human reviewing any paper at any stage.\n\n## Building the world: 124,816 papers screened into 1,000 labs\n\nSelection starts from 124,816 unique papers across arXiv's core AI categories over fourteen months. Two fast screens with MiniMax M2.7 keep roughly the top 25% quality band and empirical-method papers, leaving 21,354; detailed judgments on research quality and reproduction difficulty follow, and a geometric-mean utility picks a category-balanced top 1,000. The source adds about 9,000 papers a month, of which about 1,500 survive — the world keeps growing with the literature.\n\nThe build is a fixed nine-iteration workflow: three reproducers, three designers, one packager, and two verifier passes, with the model family rotating across GPT, Claude, MiniMax and GLM to avoid a single builder's bias. Median cost is 0.4 GPU-hours per paper, at most 70; the 1,000 papers together consumed 2,286 GPU-hours, 239,506 compute jobs, 947,056 agent turns, and 51B input tokens.\n\nQuality is not a promise but a measured property: in 99% of environments the reproduced method's score lands within a small margin of the paper's reported results; 82% beat the paper's own baselines; 37% of iteration milestones were reverted by later cross-checks — 573 times in total; and every environment must pass a verifier that took no part in building it. All 1,000 did.\n\n## The exam: 16 frontier models beat the authors 74% of the time\n\nA typical environment carries 3.3 experiments, 32 metrics (16 visible, 16 hidden), and 59 files of about 3.8k lines of code, spread over eight areas. Half of every evaluation stays hidden from the contestant, so the score reflects generalization rather than tuning against the visible half. Scoring uses the authors' Frontier-Normalized Improvement (FNI): zero matches the reproduced paper method, negative is worse, and a frontier cohort's mean score is anchored near 100.\n\nSixteen models from seven providers entered: Anthropic's Fable 5.1\u002F5 and Opus 4.8\u002F4.7, four OpenAI models from GPT-6 Astra down to GPT-5.4, Zhipu's GLM 5.2\u002F5.1, Moonshot's Kimi K2.7\u002FK2.6, MiniMax M3\u002FM2.7, Xiaomi's MiMo V2.5 Pro, and DeepSeek V4, entering through scaffolds such as Claude Code and Codex. The result: 84% of improve attempts and 74% of innovate attempts beat the paper's reproduced method — in most environments, the AI scientist does better than what the human authors published. On the leaderboard, Fable 5 is the strongest at improving existing methods, Fable 5.1 and Fable 5 take the top two spots at designing new ones, and among open models GLM 5.2 comes closest to the frontier, placing third at innovating.\n\nNothing was thrown away: the average attempt spans 5.4 hours, 6 method revisions, 25 experiments and 35 commits, all preserved as version-controlled repositories — roughly 6,400 hours of complete research traces.\n\n## The gym: a small model's ideas scale 2.6x\n\nThe world can also be used as a training ground. The authors trained Qwen3-8B with Tinker (LoRA rank 32): supervised fine-tuning on 1,000 trajectory examples, then GRPO over 1,500 already-scored ideas. Because an 8B model cannot yet carry long-horizon implementation, ideation was separated from execution — the trained model only proposes ideas, a single fixed implementer (GPT 5.6 Sol) executes every idea under identical conditions on ten held-out papers, so the idea's origin is the only variable. Median FNI rises from 41 (base) to 84 (SFT) to 106 (GRPO) — about 2.6x, the ↑158% in the abstract — while the mean climbs from 51 to 90, reaching the anchor of the frontier cohort.\n\n## So what\n\nThe hard part of this work is not the 74% headline; it is turning \"can do research\" from an adjective into a measurable engineering problem: reproduction margins, hidden halves, independent verification, FNI normalization — every layer is anti-gaming. The 37% reversion rate matters too: trust in a fully automated pipeline comes from mutual error-checking, not from faith. As model capability iterates monthly, the bottleneck is moving from \"is the brain smart enough\" to \"is the world big enough.\" The literature already exists; what was missing was the pipeline that turns it into runnable environments — and that pipeline now has a monthly intake of about 1,500 more papers. What an AI scientist lacks was never intelligence; it was a world that lets it fail, repeatedly.\n\nReference: [Scientist World: Grow a Research World for AI-Scientists](https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.08478)","scientist-world-1000-research-environments","2026-11-08T21:08:04Z","2026-11-08T21:08:04.673959Z","2026-11-08T21:08:04.673967Z",true,"agent",29,[40,48],{"slug":17,"tag_slug":17,"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},"87d43d7b-b32e-499d-af8a-68afc6506a2c","科研Agent大考:头名80.9分,达标率仅26%","sci-agent-exam-scientific-agent-benchmark","2026-10-26T17:09:05+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"ec2c558c-502d-43a5-9494-c766dfd515e9","EurekAgent：把科学发现的瓶颈从「工作流」拽到「环境」，11 美元跑出 26 圆 packing 新 SOTA","eurekagent-environment-engineering-11-usd","2026-06-11T17:56:35+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"6fd14d23-04ef-43c0-bb9a-fc365a6a6843","OpenAI湿实验基准:榜首66.7%,菌落还是数错","openai-biobasicsbench-wet-lab-agents","2026-11-08T19:30:00+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"9785c9bb-85a0-4b97-8890-aa2d6808bf86","Agent记忆陷阱:良性对话把不安全操作率推到81%","agent-memory-trap-benign-overgeneralization","2026-11-08T15:12:52+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"8857d393-a5f7-4b2a-8439-bf87ac324d4f","数学建模基准自曝:12个模型算对却只得49.82%","tb-mathmodel-agentic-verifiable-math-modeling-bench","2026-11-08T03:15:00+00:00",{"id":85,"title":86,"news_slug":87,"published_at":88},"ca1e87fb-2e21-4af2-b958-046a7efbefb0","姚期智署名TCSExam:模型考理论CS,跑分跑出新构造","tcsexam-lean-verified-tcs-benchmark","2026-11-06T19:10:36+00:00"]