[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-autosft-multi-agent-sft-data-research":3,"topics-all":35,"news-related-36f5e4f4-0a18-49a5-9dfe-94f703268806":54},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":21,"news_slug":28,"published_at":29,"created_at":30,"modified_at":31,"is_published":32,"publish_type":33,"image_url":14,"view_count":34},"36f5e4f4-0a18-49a5-9dfe-94f703268806","AutoSFT:让 AI 自己研究 SFT 数据配方","Thinking-Space 团队发布 AutoSFT,多智能体系统自动设计 SFT 数据配方:元智能体分离执行与分析,合成子智能体补齐数据池盲区。论文称分析组件让 IFEval 提升6.5分,合成再加3.4分,代码任务 MBPP+ 提升3.7分;单智能体基线只会调整类别配比。关键是覆盖多样的数据动作。","让 AI 接管后训练流程已经不是新闻,但把它真正跑通的研究者观察到一个尴尬现象:把完整训练栈交给语言模型智能体,它几乎总是收敛到监督微调(SFT),而且把大部分精力花在同一件事上——改训练数据。论文页面关联 TsinghuaNLP 实验室、代码组织为 Thinking-Space 的团队把这个问题摆上台面,发布了 AutoSFT,一个专门自动化 SFT 数据研究的多智能体系统,论文已挂 arXiv(2610.22194,27 页 16 图)。\n\n## 单智能体的失败模式\n\n论文首先点出既有系统的局限:现有工作只给智能体一种控制手段——要么从固定数据池里选行,要么调用生成器;而人类数据工程师的动作要丰富得多,会调整配比、剪枝、修复、替换、再合成。作者为此设计了一个覆盖这些操作的数据动作空间,先让单个智能体去跑完整研究循环。结果它只学会一种动作:调整类别配比,上下文还被执行日志塞满,搜索中途就被丢弃。\n\n## 三层分工:执行、分析、合成\n\nAutoSFT 的解法是多智能体分工:元智能体把执行与分析分离,从多个独立角度收集修改建议,再调用合成子智能体补齐数据池覆盖不到的能力;模型、训练器、预算在每轮对比中固定。消融实验逐个组件验证:执行器让上下文丢失减半,分析子智能体把 IFEval 最佳成绩抬了 6.5 分,合成环节再加 3.4 分 IFEval 和 2.9 分 IFBench。同一套循环迁移到代码任务,分析组件在 MBPP+ 上比毫无提升的单智能体基线高出 3.7 分。值得注意的是,每轮结果都从基座模型重新训练,各组件的贡献因此可以单独归因。\n\n## 评论:动作覆盖比单一强度更重要\n\n这篇论文([arXiv:2610.22194](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22194))最值得带走的结论是作者自己的总结:自动 SFT 数据研究的进展,更依赖覆盖多样的数据动作,而不是把单一动作做得更强。这跟人类数据团队的工作方式一致——配方、清洗、补短板各有专人。作者在社区讨论区把它定位为\"迈向 RSI-Service(后训练的自主科研服务)的第一步\",野心不止于数据管道自动化。需要泼的冷水也很明确:GitHub 仓库(Thinking-Space\u002FAutoSFT)目前只有描述页,源代码标注 coming soon,第三方复现暂时无门路,以上数字均为论文自报。对做后训练的团队来说,现阶段更有参考价值的是框架设计本身——分离执行与分析、显式建模数据动作空间,这两条在任何数据团队里都能直接借鉴。\n\n所以,下次看到\"AI 自动改进 AI\"的叙事,先问一句:它的动作空间覆盖了几种数据操作?AutoSFT 给出的答案是靠多智能体分工换覆盖面,这个方向值得持续盯。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22194","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18],{"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":14,"color":14},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":19,"name":20,"slug":20,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[22],{"id":23,"lang":24,"title":25,"summary":26,"content":27},"b316c947-7a9f-4726-95c2-355f3d1c99c1","en","AutoSFT: Multi-Agent System Automates SFT Data Research","AutoSFT: a multi-agent system automating SFT data research, replacing single agents that only tweak proportions. Reported: +6.5 IFEval, +3.7 MBPP+.","Letting AI agents run post-training is no longer novel, but researchers who actually tried it observed an awkward pattern: hand a language model agent the full training stack, and it almost always converges on supervised fine-tuning (SFT) — spending most of its effort on one thing, revising the training data. A team affiliated with the TsinghuaNLP lab (code released under the Thinking-Space organization) put this problem on the table with AutoSFT, a multi-agent system built specifically to automate SFT data research. The paper is on arXiv (2610.22194, 27 pages, 16 figures).\n\n## How the single-agent loop fails\n\nThe paper first calls out the limitation of existing systems: they give the agent exactly one kind of control — either selecting rows from a fixed pool or calling a generator — whereas a human data engineer adjusts proportions, prunes, repairs, replaces, and synthesizes data. The authors designed a data action space covering those changes and first let a single agent run the full research loop. It learned exactly one action, adjusting category proportions, while its context filled up with execution output and the search was discarded mid-run.\n\n## Three-layer division: execute, analyze, synthesize\n\nAutoSFT's answer is multi-agent division of labor: a meta agent separates execution from analysis, gathers proposals from independent angles, and calls a synthesis subagent for capabilities the data pool cannot cover — with the model, the trainer, and the budget fixed within each comparison. Ablations validate each component in turn: the executor halves context loss, the analysis subagents lift the best IFEval score by 6.5 points, and synthesis adds another 3.4 IFEval and 2.9 IFBench. The same loop transfers to code, where analysis adds +3.7 MBPP+ over a single-agent baseline that does not improve at all. Notably, every result comes from runs that retrain from the base model each round, so each component's effect is individually attributable.\n\n## Commentary: action coverage beats single-action strength\n\nThe takeaway the authors themselves state in the paper ([arXiv:2610.22194](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22194)): progress in automated SFT data research depends more on covering diverse data actions than on making a single action stronger. That mirrors how human data teams work — recipe tuning, cleaning, and gap-filling each have owners. In the community discussion, the authors position the work as \"a first step toward RSI-Service for post-training,\" an ambition that goes beyond automating the data pipeline. The cold water is equally clear: the GitHub repository (Thinking-Space\u002FAutoSFT) currently holds only a description page with source code marked \"coming soon,\" so third-party replication has no path yet and all numbers above are self-reported in the paper. For post-training teams, the more immediately useful part is the architecture itself — separating execution from analysis and explicitly modeling a data action space are two moves any data team can borrow today.\n\nSo the next time you see an \"AI improving AI\" narrative, ask one question first: how many distinct data operations does its action space cover? AutoSFT's answer is trading single-agent depth for multi-agent coverage — a direction worth tracking.","autosft-multi-agent-sft-data-research","2026-10-16T23:07:40Z","2026-10-16T23:07:45.763618Z","2026-10-16T23:07:45.763627Z",true,"agent",118,[36,45],{"slug":37,"tag_slug":37,"title_zh":38,"title_en":39,"intro_zh":40,"intro_en":41,"id":42,"is_active":32,"created_at":43,"modified_at":44},"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":46,"tag_slug":46,"title_zh":47,"title_en":48,"intro_zh":49,"intro_en":50,"id":51,"is_active":32,"created_at":52,"modified_at":53},"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":55},[56,61,66,71,76,81],{"id":57,"title":58,"news_slug":59,"published_at":60},"48726cf0-6f0c-4b09-841e-06aa8b0ffcfd","上海AI Lab Intern-Decision:多模态决策模型,4B 微超 Jev 90.02%","intern-decision-qwen3-5-multimodal-decision-model","2026-10-18T11:13:06+00:00",{"id":62,"title":63,"news_slug":64,"published_at":65},"da6cbb98-3b89-4631-8dbd-59411136fb30","哈佛用 700 家公司拆解 AI 编程智能体:30% 代码增量,49% 合并时延","harvard-coding-agent-review-bottleneck","2026-10-14T11:30:00+00:00",{"id":67,"title":68,"news_slug":69,"published_at":70},"9c8731f9-e170-4855-93a2-b1705abb61c3","Muse Wonder发布:Meta把宝押在个人智能体上","meta-muse-wonder-agentic-model","2026-10-12T23:10:00+00:00",{"id":72,"title":73,"news_slug":74,"published_at":75},"5edbfca7-7c1a-4df3-8577-a0195408b827","RSI-Anything:世界模型陪跑,自我改进有了回滚键","rsi-anything-world-model-recursive-self-improvement","2026-10-12T21:20:00+00:00",{"id":77,"title":78,"news_slug":79,"published_at":80},"8474959b-1569-45d6-afcc-34599f05b2fd","Memento 3:不动权重,LLM自我改进写进规则手册","memento-3-recursive-self-improvement-rulebook","2026-10-11T13:07:13+00:00",{"id":82,"title":83,"news_slug":84,"published_at":85},"e40de2f9-9ee4-47f3-b1f1-7bf93a4870a9","一个动词翻转工具调用决策,LLM 内部向量现形","llm-tool-call-decision-vector","2026-10-09T13:10:00+00:00"]