[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-cua-master-open-world-flywheel":3,"topics-all":38,"news-related-90eaefa3-cbb1-440a-9c2f-97c594d0bf78":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},"90eaefa3-cbb1-440a-9c2f-97c594d0bf78","CUA-Master:教学视频喂出OSWorld 80.2","CUA-Master 用嵌套双飞轮重做电脑操控智能体的数据管线:外环按模型弱点造数据,内环从教学视频挖种子、自催化合成可验证任务,采集配快照回滚重试。27B 模型 OSWorld Verified 80.2、OSWorld 2.0 36.9、WeaveBench 43.9,递归迭代持续涨分,代码与模型已开源。","训练电脑操控智能体,最贵的从来不是模型,是那些「像人一样操作软件」的轨迹数据。主流做法是让模型自己试、自己攒经验,论文里叫 self-generated experiences,说白了就是自产自销:模型会什么,数据里就只有什么,收益越滚越薄。11 月 2 日挂上 arXiv 的 CUA-Master 换了个思路——互联网上现成的教学视频,本身就是海量真人操作演示。论文上线次日冲上 Hugging Face 论文日榜第二,拿到 90 个赞。第一作者 Xinlei Yu 领衔、共 18 位作者,把「数据从哪来」这个问题当成了论文主角。\n\n## 两层飞轮,转的是开放世界\n\nCUA-Master 的框架是嵌套双循环。外环是「世界-数据-模型」飞轮:持续引入开放世界语料,模型当前暴露的弱点反过来决定下一轮造什么数据、优化什么能力。内环管数据本身:一头从教学视频里挖真实工作流当种子任务,用视觉、语义双通道重建轨迹,把字幕里的语义触发器和画面变化检测相交成边界结构;另一头做自催化传播,把种子抽象成可复用的能力原语,再按 K 个原语 × N 个应用组合出可控、可验证的新任务。采集轨迹时还有个值得抄的工程细节:快照回滚重试——agent 走岔了就回滚到上一个快照接着试,而不是整条轨迹作废。\n\n按 GitHub 仓库的配置,种子挖掘覆盖 8 个领域、36 个子领域、650 多个应用,视频会先过一道 0-5 分的 CUA 适配度打分,不合格的直接不进管线。\n\n## 27B 的成绩单\n\nCUA-Master-27B 在纯 GUI 和 GUI+CLI 混合两类基准上都交了分:OSWorld Verified 80.2,OSWorld 2.0 为 36.9,WeaveBench 43.9,Agents' Last Exam 25.0。摘要的措辞是 competitive——有竞争力,没吹霸榜。更值得注意的是 80.2 和 36.9 之间的落差:更难的新基准上,整个赛道都还有大坑要填。论文同时报告递归迭代持续涨分,数据合成的质量和效率也在提升。\n\n开源交付相当实在:代码、配置、提示词、VM 镜像管理都在 GitHub 仓库,训练数据转 ms-swift 的 SFT 格式,RL 阶段直接拿任务级验证器当结果奖励,README 里还挂出了 Hugging Face 上的 CUAMaster-27B 模型入口。仓库几天内攒到 86 stars。\n\n## 真正的信号:数据供给换源\n\n这篇论文对行业的价值,不在某个分数,而在把 CUA 的数据供给从「模型自我博弈」切到「开放世界语料」。自产自销的天花板是模型当前的能力边界;教学视频这样真人在真软件上操作的语料,天花板是人类已经发布过的全部操作知识。配上可验证的任务合成,数据质量不再靠感觉,由验证器兜底——它同时就是 RL 的奖励来源。\n\n对做 agent 数据的团队,这是套能拆开抄的工程蓝图:视频挖种子、原语组合造任务、快照回滚省算力,三个模块独立成立。对看赛道的人,判断标准很简单:谁先把「开放世界数据 × 递归改进」的飞轮转起来,谁就先把 CUA 的数据成本打下来。\n\n参考:[arXiv:2611.03126](https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.03126);[GitHub: xlyu0106\u002FCUA-Master](https:\u002F\u002Fgithub.com\u002Fxlyu0106\u002FCUA-Master)。\n","https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.03126","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},"5748d69a-a9aa-431f-ac09-9a1a497f25a5","en","CUA-Master: Video-Seeded Flywheels Push OSWorld to 80.2","CUA-Master seeds computer-use agents from instructional videos via nested flywheels; the 27B model hits 80.2 on OSWorld Verified. Code is open.","The costliest ingredient in training computer-use agents is not the model — it is the interaction data showing how a human actually operates software. The mainstream recipe lets the model try, fail, and bank its own experience; the literature calls these self-generated experiences. In plain terms, the model mines itself: the data can only contain what the model already knows, and returns diminish with every loop. CUA-Master, posted to arXiv on Nov 2, flips the source — instructional videos already on the internet are massive libraries of real human demonstrations. A day after release it sat at #2 on Hugging Face's paper-of-the-day board with 90 upvotes.\n\n## Two nested flywheels\n\nThe framework is a nested double loop. The outer loop is a world-data-model flywheel: open-world corpus keeps flowing in, and the model's current weaknesses decide what data gets built and which abilities get optimized next round. The inner loop owns the data itself. One side mines authentic workflows from instructional videos as seed tasks and reconstructs trajectories through visual-semantic dual paths, intersecting semantic triggers from transcripts with frame-change detection into a boundary structure. The other side runs autocatalytic propagation: seed tasks are abstracted into reusable ability primitives, then recomposed under explicit complexity control — K primitives × N applications — into controllable, verifiable new tasks. Trajectory collection adds a detail worth copying: feedback-guided snapshot-rollback retries, so a derailed step rolls the environment back to the last snapshot and tries again instead of voiding the whole trajectory.\n\nPer the GitHub repo's configuration, seed harvesting spans 8 domains, 36 subdomains, and 650+ apps, and every video passes a 0-5 CUA-suitability score before entering the pipeline.\n\n## The 27B scoreboard\n\nCUA-Master-27B reports scores across GUI-only and GUI-CLI hybrid benchmarks: 80.2 on OSWorld Verified, 36.9 on OSWorld 2.0, 43.9 on WeaveBench, and 25.0 on Agents' Last Exam. The abstract's own word is \"competitive\" — no leaderboard sweep claimed. The gap between 80.2 and 36.9 is the real story: on harder, newer benchmarks the whole field still has a long way to climb. The paper also reports continuous gains across recursive iterations and improved quality and efficiency in data synthesis.\n\nThe open-source delivery is solid: code, configs, prompts, and VM image management all live in the GitHub repo; training data converts to ms-swift SFT format; RL uses task-level verifiers as outcome rewards; and the README links the CUAMaster-27B model on Hugging Face. The repo has gathered 86 stars within days.\n\n## The real signal: switching the data supply\n\nThe industry value here is not any single score — it is switching the CUA data supply from model self-play to open-world corpus. Self-generation caps out at the model's current ability boundary; instructional videos, where real humans operate real software, cap out at all the operational knowledge humanity has ever published. Pair that with verifiable task synthesis and data quality stops being vibes — the verifiers double as RL reward sources.\n\nFor teams building agent data, this is a blueprint you can copy module by module: mine seeds from video, compose tasks from primitives, save compute with snapshot rollback. For anyone watching the space, the yardstick is simple: whoever spins the open-world-data x recursive-improvement flywheel first will be the first to crash CUA's data costs.\n\nReference: [arXiv:2611.03126](https:\u002F\u002Farxiv.org\u002Fabs\u002F2611.03126); [GitHub: xlyu0106\u002FCUA-Master](https:\u002F\u002Fgithub.com\u002Fxlyu0106\u002FCUA-Master).\n","cua-master-open-world-flywheel","2026-11-03T21:08:37Z","2026-11-03T21:08:49.060688Z","2026-11-03T21:08:49.060698Z",true,"agent",66,[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},"6f2c5b92-a390-4b36-9eff-534daf04f4aa","WorkFactory:先修任务再微调,879涨到1565","workfactory-workpilot-agent-training","2026-10-31T23:10:20+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"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":75,"title":76,"news_slug":77,"published_at":78},"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":80,"title":81,"news_slug":82,"published_at":83},"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":85,"title":86,"news_slug":87,"published_at":88},"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"]