[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-r2orch-self-play-skill-withdrawal":3,"topics-all":38,"news-related-022731b7-2aa0-435f-b93e-3f84eea2699a":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},"022731b7-2aa0-435f-b93e-3f84eea2699a","R2Orch双层自博弈:技能内化,零技能上阵","港中深团队的 R2Orch 双层循环训练编排器:内层自博弈挑任务,外层逐个撤技能,收敛到零技能部署。零技能测试下,9B 编排器在论文自评 12 项基准上全超基线,分布内平均 86.0 提到 89.8,代码与模型已开源。","教智能体干活,业界通行做法是往上下文里塞技能:检索技巧、工具用法、检查清单,越攒越厚。港中深团队 10 月挂出的 R2Orch 论文反着来——技能只是脚手架,编排器把技能学进参数之后,再把脚手架逐根拆掉,最终零技能上阵。这篇论文 11 月 2 日由作者自投登上 Hugging Face 当日论文榜第 2 位,拿到 86 个赞。\n\n## 双层循环:一层练,一层拆\n\nR2Orch 的主角是一个 Director(编排器):它为每个任务从空白画布开始,一次一条原子编辑——加智能体、设提示词、指定模型路由、连接交接边——搭出一张智能体图,再交给冻结的 Worker 模型执行。训练由两层循环嵌套完成。\n\n内层是自博弈。一个 Proposer 负责挑任务,训练目标叫 Recursive Learnability:奖励那些「撤掉候选技能后,rollout 仍然会打出不同结局」的任务,可学性用 l(p)=4p(1−p) 估计,p 是隐藏组的成功率。这一步直接对准了 GRPO 类分组强化学习的死穴——当一组 rollout 全对或全错,组内没有优势差,梯度归零,论文称之为 advantage collapse。\n\n外层是技能撤除(Skill Withdrawal)。每个任务跑 G=5 张图:3 张隐藏组看不到候选技能,2 张可见组带着技能。因为隐藏是随机分配的,可见组减隐藏组的均值差就是技能净收益的无偏估计,这些对照累积成 Bayesian-bootstrap 后验,后验收益期望压到阈值以下就撤。论文报告 12 次撤除全部通过该判据,后验 E[Δ+] 至多 0.019。两层循环首尾相接:失败被蒸馏成新技能,技能被写进参数,撤除又把 Proposer 引向新的失败。\n\n## 零技能上阵,12 项基准\n\n实验配置:骨干是冻结的 Qwen3.5-9B,Director 和 Proposer 都是 rank-64 LoRA 适配器,单机 4 张 A100 64GB,训练 200 个周期,种子技能只有 18 张卡,训练池 6,154 个任务。论文自评表格里,R2Orch 以零技能状态在 12 项基准上全部超过基线:分布内平均分从 PATS 的 86.0 提到 89.8,分布外从 76.1 到 78.8;AIME 2026 从 74.0 到 81.3,MedQA 94.06、ALFWorld 90.63、WebShop 91.09,SWE-Bench Verified 42.81。消融显示两条循环缺一不可,固定脚手架(技能永不撤)在零技能测试里掉分最多;训练进行到一半时,零技能准确率就已经反超带技能状态。\n\n## 迁移是真的,短板也在\n\n不重训,这个 Director 直接换到另外 6 个 Worker 骨干上,每项基准都有提升,弱骨干提升最多——各骨干间的分布内成绩极差从 25.7 分收窄到 6.4 分,说明编排图补上的是规划与校验能力,而不是某个模型的私有能力。论文自己承认的局限也很直白:每个数据集的初始技能是手工定制的,新领域如何自动导出种子技能仍是开放问题;仓库刚开源(10 star),许可证为 research use,模型 R2Orch-9b 与数据集已发布。\n\n技能库的终局,也许不是越攒越厚的提示词合集,而是像教辅一样——用完即弃,能力留下。当「撤走技能之后还顶用」成了训练目标本身,智能体才算真正把知识内化了。\n\n参考:论文 [arXiv:2610.22740](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22740);代码 [github.com\u002Fbeita6969\u002FR2Orch](https:\u002F\u002Fgithub.com\u002Fbeita6969\u002FR2Orch)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22740","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},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"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},"c4e101a1-5755-4fae-bc88-da7ac947bbb9","en","R2Orch: Self-Play Orchestration That Drops Its Skills","CUHKSZ's R2Orch trains a 9B orchestrator via self-play plus skill withdrawal, tested at zero skills: its 12-benchmark eval lifts IID 86.0 to 89.8.","The industry's default way to make agents better is to keep stuffing skills into the context: retrieval tips, tool usage notes, checklists. A team led by the Chinese University of Hong Kong, Shenzhen (with collaborators at Fudan, UIUC, Tsinghua and NUS) took the opposite path in R2Orch, a paper posted in mid-October and revised on October 22: skills are scaffolding. Once the orchestrator has internalized them into its parameters, the scaffolding is removed piece by piece, and the final orchestrator deploys with zero skills. On November 2 one of the authors submitted it to Hugging Face Daily Papers, where it sat at #2 with 86 upvotes.\n\n## Two nested loops: one trains, one removes\n\nThe protagonist is a Director. For every task it starts from an empty canvas and issues one atomic edit at a time — add an agent, set a prompt, assign a model route, connect a hand-off — until an agent graph is built for frozen Worker models to execute. Training nests two loops.\n\nThe inner loop is self-play. A Proposer picks the tasks, driven by an objective called Recursive Learnability: it rewards tasks whose rollouts still reach different outcomes after the candidate skill is withheld, with learnability estimated as l(p)=4p(1−p), where p is the hidden part's success rate. This targets the known weak point of group-based RL like GRPO — when every rollout in a group succeeds or fails together, there is no advantage left to learn from, what the paper calls advantage collapse.\n\nThe outer loop is Skill Withdrawal. Each task runs G=5 graphs: 3 in a hidden part that cannot see the candidate skill, 2 in a visible part that can. Because hiding is random, the visible-minus-hidden mean difference is an unbiased estimate of the skill's net benefit; these contrasts accumulate into a Bayesian-bootstrap posterior, and once the posterior expected benefit falls below threshold, the skill is withdrawn. The paper reports all 12 withdrawals passed the criterion, with posterior E[Δ+] at most 0.019. The loops close into one cycle: failures are distilled into new skills, skills into parameters, and each withdrawal points the Proposer at new failures.\n\n## Twelve benchmarks, zero skills at test time\n\nThe setup: a frozen Qwen3.5-9B backbone, Director and Proposer both rank-64 LoRA adapters, one node with 4 A100 64GB GPUs, 200 cycles, only 18 seed skill cards, and a training pool of 6,154 tasks. In the paper's own tables, tested with zero skills, R2Orch tops all 12 benchmarks against six families of baselines: the in-distribution average rises from 86.0 (baseline PATS) to 89.8, out-of-distribution from 76.1 to 78.8; AIME 2026 goes from 74.0 to 81.3, MedQA 94.06, ALFWorld 90.63, WebShop 91.09, SWE-Bench Verified 42.81. Ablations show both loops matter — a fixed scaffold, where skills are never withdrawn, loses the most at zero-skill testing — and halfway through training, zero-skill accuracy already overtakes the with-skills state.\n\n## Transfer is real; so are the limits\n\nWithout retraining, the same Director improves all 6 other Worker backbones on every benchmark, and weaker backbones gain the most: the in-distribution spread across backbones narrows from 25.7 to 6.4 points, suggesting the graph supplies planning and checking rather than one model's private ability. The paper's own stated limitation is blunt: the initial skills are hand-customized per dataset, and deriving them automatically for a new domain remains open. The repository is freshly open-sourced (10 stars) under a research-use license, with the R2Orch-9b model and dataset released.\n\nThe endgame of skill libraries may not be an ever-thicker prompt, but something like a textbook: used, absorbed, discarded. When \"still works after withdrawal\" becomes the training objective itself, an agent has truly internalized the knowledge.\n\nReferences: paper [arXiv:2610.22740](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.22740); code [github.com\u002Fbeita6969\u002FR2Orch](https:\u002F\u002Fgithub.com\u002Fbeita6969\u002FR2Orch)","r2orch-self-play-skill-withdrawal","2026-11-02T21:10:07Z","2026-11-02T21:10:12.556288Z","2026-11-02T21:10:12.556300Z",true,"agent",14,[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},"be62f4f6-6935-4c2c-abe8-9e0732bb04ef","Trajectory2Gym:智能体的成功轨迹不等于好教材","trajectory2gym-compositional-interaction-synthesis","2026-10-27T21:10:28+00:00",{"id":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"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":75,"title":76,"news_slug":77,"published_at":78},"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",{"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},"44740b4d-8c2c-44fc-8fff-fd89f3fb54ed","12 万美元 token 把 Copilot 运行时从 TypeScript 搬到 Rust","github-copilot-rust-migration-stephen-toub","2026-09-27T11:00:00+00:00"]