[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-workfactory-workpilot-agent-training":3,"topics-all":38,"news-related-6f2c5b92-a390-4b36-9eff-534daf04f4aa":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},"6f2c5b92-a390-4b36-9eff-534daf04f4aa","WorkFactory:先修任务再微调,879涨到1565","IQuest 团队提出 WorkFactory:为职业任务建信息支撑图,用探针执行诊断缺漏并修复,再交给教师模型采示范微调。112,686 个任务、612 种职业,练出 WorkPilot 家族,35B-A3B 版在 GDPval-AA v2.1 上 Elo 从 879 涨到 1565,数据与模型已开源。","给工作型智能体攒监督数据,一个隐蔽的坑不是量不够,而是任务本身是坏的。论文开篇就把链条挑明:扩展工作型智能体的监督训练,前提是大量多样的职业任务、且这些任务能产出可靠的专家示范;可一旦任务信息缺失或前后不一致,教师模型的执行就会被卡住,监督信号随之变弱——文件缺一个、口径错一处,采出来的示范轨迹就把缺陷原样传给学生模型。IQuest 团队 13 位作者 10 月 27 日挂上 arXiv(cs.AI)的 WorkFactory,干的就是给训练任务当质检员这件事。\n\n## 任务先体检,坏了先修\n\n流程分三步。先从职业、技能、真实交付物出发构造\"职业意图\";再用信息级规划画出一张信息支撑图,把完成预期产出所需的依赖与信息位置显式标出来,按图搭建工作区——相当于把\"这道题需要哪些已知条件、各自放在哪里\"从自然语言描述升级成一张结构化清单;接着跑干净探针执行,检验任务在当前环境里到底能不能被做完。一旦发现支撑失败,就做基于执行的修复:缺什么补什么、错什么修什么,同时保住任务原本要考察的工作,不把题目偷偷改简单。只有通过体检的环境才被保留,交给教师模型采集彼此独立的示范轨迹,再拿去监督微调。论文的受控消融显示,规划加修复这两步既减少了探针诊断出的任务环境缺陷,也直接抬升了下游表现。\n\n## 11 万任务,612 种职业\n\n按这套流程,团队造出 112,686 个任务,覆盖 612 种职业、23 个职业大类,平均每个任务挂 4.15 个参考文件。教师轨迹分别微调 Qwen 系 4B、27B、35B-A3B 三个规模,得到 WorkPilot 家族——同一份数据配方,从轻量级一路铺到重量级,三个规模梯度清晰。\n\n## 879 到 1565,但 27B 反而更高\n\n论文报告的四项专业工作基准——GDPval-AA、JobBench、Workspace-Bench、OmegaUse-OfficeVal——上,三个模型全部超过各自基座。WorkPilot-35B-A3B 在 GDPval-AA v2.1 的 Elo 从 879 涨到 1565,4B 版到 1375;更有意思的是 27B 版冲到 1612,反压更大的 35B-A3B,这与论文\"收益随模型递减、且因模型而异\"的缩放分析对得上——执行步长也呈先变长后变短的非单调曲线。一句诚实提醒:README 明说这些 Elo 是校准到 GDPval-AA v2.1 的本地估计,不是官方榜,转述时别写成\"登顶\"。\n\n## 所以呢\n\n真正值得拿走的不是又一个微调模型,而是把\"任务可执行性\"当成可测量、可修复的对象:质检从产出端(轨迹)前移到输入端(任务环境),体检、修复、再采集的闭环,理论上可以直接搬进任何智能体 SFT 数据管线。数据集与三个模型都已放出,仓库 Apache 2.0,复现门槛不高。下次再看到\"XX 万条智能体训练数据\"的通稿,不妨先问一句:这些任务,教师模型真的做得完吗?\n\n参考:[arXiv:2610.37161](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.37161) · [GitHub: IQuestLab\u002FWorkFactory](https:\u002F\u002Fgithub.com\u002FIQuestLab\u002FWorkFactory)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.37161","7437aeb9-930c-4866-a2e9-48003c1a792b",[11,15,18,21],{"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},"120fa59a-ff6f-4537-9bf5-f818df636a0e","benchmark",{"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},"cf9ecd48-bab1-4194-b111-cd4ec1fcc97c","en","WorkFactory: Repair Tasks Before Training, Elo 879 to 1565","WorkFactory repairs professional tasks before training: 112,686 tasks, 612 occupations, WorkPilot-35B-A3B lifts Elo 879 to 1565 on GDPval-AA v2.1.","A quiet failure mode in supervised training for working agents is not data volume — it is broken tasks. One missing file or one inconsistent field can stop a teacher model from executing a professional task at all, and whatever demonstration survives carries the defect into the student model. WorkFactory, posted to arXiv on 27 October by the IQuest team, takes aim exactly there: treat the task itself as something that can be broken, diagnosed and repaired.\n\n## Give every task a health check first\n\nThe pipeline runs in three moves. Professional intents are composed from occupations, skills and real deliverables. Information-level planning then draws an information support graph that makes explicit which facts a deliverable depends on and where they must live, and the workspace is built against that graph. Clean probe executions test whether the task can actually be completed in the constructed environment; when support failures show up, targeted repairs fix the environment while preserving the intended work. Only environments that pass are retained, handed to teacher models for collecting independent demonstrations, and those trajectories then feed supervised fine-tuning. Controlled ablations in the paper show the planning-plus-repair loop reduces probe-diagnosed environment defects and improves downstream performance.\n\n## 112,686 tasks across 612 occupations\n\nThe framework yields 112,686 tasks spanning 612 occupations and 23 occupational groups, with an average of 4.15 reference files per task. Teacher trajectories fine-tune Qwen models at 4B, 27B and 35B-A3B scales, producing the WorkPilot family.\n\n## Elo 879 to 1565 — and the 27B outruns the 35B\n\nAcross the four professional-work benchmarks reported in the paper — GDPval-AA, JobBench, Workspace-Bench and OmegaUse-OfficeVal — all three WorkPilot models beat their respective base models. WorkPilot-35B-A3B climbs from 879 to 1565 Elo on GDPval-AA v2.1, the 4B reaches 1375, and the 27B hits 1612 — actually ahead of the larger 35B-A3B. That lines up with the paper's scaling analysis: diminishing returns that depend on the model, with execution length first rising and then falling. One honesty note from the README: these Elo scores are local estimates calibrated to GDPval-AA v2.1, not an official leaderboard.\n\n## So what\n\nThe idea worth taking away is not another fine-tuned model but treating task executability as a measurable, repairable property — a check-then-repair-then-collect loop that can bolt onto any agent SFT pipeline. The dataset and all three models are already released, with the repository under Apache 2.0. Next time a release brags about \"N million agent training tasks\", ask one question first: can a teacher model actually finish them?\n\nReferences: [arXiv:2610.37161](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.37161) · [GitHub: IQuestLab\u002FWorkFactory](https:\u002F\u002Fgithub.com\u002FIQuestLab\u002FWorkFactory)","workfactory-workpilot-agent-training","2026-10-31T23:10:20Z","2026-10-31T23:10:23.746219Z","2026-10-31T23:10:23.746236Z",true,"agent",46,[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},"dcc6132a-e432-4ef3-9555-5412d692cd90","RSIHub 开源:先冻结评测器,再谈自我进化","rsihub-agent-evolution-frozen-evaluator","2026-10-26T07:20:28+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"aa718f3c-14cf-4984-8762-794eaaee8c16","MAI-Gym:9B 智能体在 2 万个并行环境里做在线 RL","mai-gym-gui-agent-online-rl-9b","2026-10-21T09:30:00+00:00",{"id":70,"title":71,"news_slug":72,"published_at":73},"bb6e0f7b-3899-4e45-a267-bfa512d470b0","FDE-Gym:95个真实交付任务,最强Agent只过14个","fde-gym-agent-delivery-benchmark","2026-10-18T23:08:06+00:00",{"id":75,"title":76,"news_slug":77,"published_at":78},"206ea36a-1eea-463f-a241-1e3b32f5ec2d","USTC GraphForge:证据图把任务和 rubric 钉在一起,Qwen3.6-27B 涨三基准","graphforge-ustc-qwen36-27b-evidence-graph","2026-10-04T03:05:00+00:00",{"id":80,"title":81,"news_slug":82,"published_at":83},"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",{"id":85,"title":86,"news_slug":87,"published_at":88},"9ebb888c-dfe7-416a-9940-a913527d4f73","AI Agent 的失败比成功更值钱:5 万对错误诊断数据,修正通过率 18.4%→51.1%","agent-error-dataset","2026-10-01T15:11:08+00:00"]