[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-trajectory2gym-compositional-interaction-synthesis":3,"topics-all":38,"news-related-be62f4f6-6935-4c2c-abe8-9e0732bb04ef":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},"be62f4f6-6935-4c2c-abe8-9e0732bb04ef","Trajectory2Gym:智能体的成功轨迹不等于好教材","Trajectory2Gym登顶Hugging Face当日论文榜:成功的工具调用轨迹未必是好监督,不如构造覆盖依赖结构的可执行交互、只从有依据的决策提取监督。五档Qwen3上τ\u002Fτ²-bench与BFCL均值提升16.9到29.8个百分点,数据集与五个checkpoint开源。","先让智能体把任务跑通,再把成功轨迹攒起来喂给模型——这是训练工具型智能体的一条主流流水线。10 月 24 日提交到 arXiv 的论文 2610.33568 开头就否定了其中一半:成功的工具调用轨迹,不一定是好的监督信号。这篇挂在 ant-international 组织名下、由一作 Yihao Hu 提交到 Hugging Face 后,列为当日论文榜第一,拿到 330 个赞。\n\n## 成功轨迹的两个暗坑\n\n论文把\"成功但没用\"拆成两种病:一是组合覆盖不足,多步任务里关键的观察-动作依赖没有被表达出来,模型记住的只是碰巧成立的动作序列;二是决策不落地,某个动作在决策时刻,学习者可见的交互历史根本不足以支撑它——等于让模型背一个推不出来的答案。前者丢结构,后者养武断。\n\n## 不采轨迹,造交互\n\nTrajectory2Gym 的做法是把\"采集轨迹\"换成\"构造交互\":先定位组合结构上欠覆盖的部分,围绕它们构造可执行的交互,监督信号只从真实执行中有依据的决策里提取。配套两个机制:结构支持度选择负责给训练经验分配名额,父级归一化 SFT 把\"一次语义交互\"而非\"一行渲染数据\"当作训练权重的单元。一个反直觉的细节:官方代码库明确写着,实验用的是普通的 assistant-token 交叉熵,不是 RL 目标——涨分来自重新组织过的监督信号,而非更花哨的训练算法。\n\n## 五档 Qwen3,全部上涨\n\n在 1.7B、4B、8B、14B、30B-A3B 五个 Qwen3 规模上,τ-bench、τ²-bench、BFCL 三基准等权均值提升 16.9 到 29.8 个百分点,每一档都涨;增益在多轮、有状态的交互上最强——后面的决策依赖前面的观察与反馈,这恰好是纯轨迹蒸馏容易漏掉的部分。HF 讨论区有读者逐项引述论文消融表:仅父级归一化一项,Qwen3-8B 的三基准均值就提高 8.04 分,τ²-bench 从 52.50 升到 74.17;支持深度取 2 时均值 72.28,高于深度 3 和 4。30B-A3B 档使用了扩展语料。\n\n## 开源清单很实在\n\n代码以 Apache-2.0 许可证放出,README 里直接给了方法到代码的映射表;数据集 mis2\u002FTrajectory2gym 收录 34,772 条语义交互,7 个 JSONL 分片约 125MB,附文件校验和、来源与许可标注;五个 checkpoint 全部公开——1.7B(float32,8.1GB)、4B(17.6GB)、8B(32.8GB)、14B(bfloat16,29.5GB)、30B-A3B(61.1GB),发布时未做量化或重训;README 也注明分数沿用实验结果,并非发布时重跑。论文页:[arXiv:2610.33568](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.33568);代码:[GitHub - yihaohu0118\u002FTrajectory2Gym](https:\u002F\u002Fgithub.com\u002Fyihaohu0118\u002FTrajectory2Gym)。\n\n## 所以呢\n\n这篇论文真正打中的,是数据飞轮叙事里被略过的一环:默认只要智能体跑得够多、成功轨迹攒得够厚,能力就会自己长出来。但五个规模的对照实验说明,决定上限的不是轨迹数量,而是每条轨迹里依赖结构是否被表达、决策能否从可见历史推出。对正在做智能体蒸馏管线的团队,这是个可以直接落地的提醒:堆量之前,先审一遍轨迹里有多少无据决策。构造可学的交互,可能比收集成功的轨迹更接近 scaling 的本质。\n","https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.33568","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},"11ec6c7b-a154-4f70-9e4d-58954734c78e","en","Trajectory2Gym: Successful Trajectories Make Poor Teachers","Trajectory2Gym: successful trajectories are poor supervision. Five Qwen3 scales gain 16.9-29.8 points on tau\u002Ftau2\u002FBFCL; data and 5 checkpoints open.","Let the agent solve the task first, then collect its successful execution traces and feed them back to the model — that is a mainstream pipeline for training tool-use agents today. A paper submitted to arXiv on October 24 (2610.33568) pushes back against half of that recipe in its opening line: successful tool-use trajectories are not necessarily useful supervision. Listed under the ant-international organization and submitted to Hugging Face by first author Yihao Hu on October 27, it ranks as the #1 paper of the day with 330 upvotes.\n\n## Two hidden failures inside successful trajectories\n\nThe paper splits \"successful but useless\" into two failure modes. First, insufficient compositional coverage: the observation-action dependencies that matter in a multi-step task are simply not represented, so the model memorizes an action sequence that happened to work. Second, ungrounded decisions: at decision time, the interaction history visible to the learner does not justify the action — the model is effectively asked to memorize an answer it cannot derive. The first erases structure; the second rewards arbitrariness.\n\n## Don't collect trajectories — construct interactions\n\nTrajectory2Gym replaces trajectory collection with interaction construction. It locates under-supported compositional structures, builds executable interactions around them, and derives supervision only from grounded decisions in the realized executions. Two mechanisms support this: structural-support selection allocates the resulting training experience, while parent-normalized supervised fine-tuning treats one semantic interaction — not one rendered row — as the unit of training weight. A counter-intuitive detail: the official repository states plainly that the experiments use ordinary assistant-token cross-entropy, not an RL objective. The gains come from reorganized supervision, not from a fancier training algorithm.\n\n## Five Qwen3 scales, gains at every one\n\nAcross five Qwen3 scales — 1.7B, 4B, 8B, 14B, and 30B-A3B — the equal-weight mean of tau-bench, tau2-bench, and BFCL rises by 16.9 to 29.8 percentage points, improving at every tested scale. The gains are strongest in stateful, multi-turn interaction, where later decisions depend on earlier observations and feedback — exactly what naive trajectory distillation tends to lose. In the Hugging Face discussion, a reader walks through the paper's ablation tables: parent normalization alone adds 8.04 points to the Qwen3-8B mean, with tau2-bench going from 52.50 to 74.17; support depth 2 reaches a 72.28 mean, beating depths 3 and 4. The 30B-A3B study uses an expanded corpus.\n\n## A genuinely complete open-source release\n\nThe code ships under Apache-2.0, with a method-to-code map in the README. The dataset mis2\u002FTrajectory2gym contains 34,772 semantic interactions in seven compressed JSONL shards (about 125 MB), with checksums, provenance, and license notices. All five checkpoints are public — 1.7B (float32, 8.1 GB), 4B (17.6 GB), 8B (32.8 GB), 14B (bfloat16, 29.5 GB), and 30B-A3B (61.1 GB) — unquantized and unretrained for release; the README candidly notes the reported scores are inherited experiment results, not a fresh rerun. Paper: [arXiv:2610.33568](https:\u002F\u002Farxiv.org\u002Fabs\u002F2610.33568); code: [GitHub - yihaohu0118\u002FTrajectory2Gym](https:\u002F\u002Fgithub.com\u002Fyihaohu0118\u002FTrajectory2Gym).\n\n## So what\n\nWhat this paper really hits is the link the data-flywheel narrative skips: the assumption that if agents run long enough and successful trajectories pile up high enough, capability will grow on its own. The five-scale controlled comparison says otherwise — the ceiling is set not by the number of trajectories but by whether their dependency structures are represented and whether each decision is derivable from visible history. For teams building agent distillation pipelines, this is an immediately actionable reminder: before scaling up volume, audit how many actions in your trajectories are ungrounded. Constructing learnable interactions may sit closer to the essence of scaling than collecting successful trajectories.\n","trajectory2gym-compositional-interaction-synthesis","2026-10-27T21:10:28Z","2026-10-27T21:10:33.277460Z","2026-10-27T21:10:33.277471Z",true,"agent",193,[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},"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":65,"title":66,"news_slug":67,"published_at":68},"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":70,"title":71,"news_slug":72,"published_at":73},"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":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},"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",{"id":85,"title":86,"news_slug":87,"published_at":88},"777afb24-262f-45cc-961f-d5d49ad42883","AgentOPSD 用递归贝叶斯信念破解多轮 Agent 强化学习的信用分配：清华\u002F浙大\u002F美团让 GRPO 学会看哪个 turn 决定胜负","agentopsd-recursive-belief-credit-assignment","2026-08-07T02:00:00+00:00"]