Daily Papers · 2026-09-11
📄 论文日报 | 来源：Hugging Face Daily Papers
🗓 Episode：2026.09.10 | VLM语义动作玩转机器人；可编程世界模型状态可控
1. Show-Harness：仅一个VLM智能体即可玩转机器人
- 英文标题：Show-Harness: Just a VLM Agent Can Play Robots
- arXiv：2609.10522v1
💡 一句话解读：让大模型直接控制机器人很难，因为它的输出和机器人的具体动作对不上。这篇论文设计了一个叫Show-Harness的中间层，把机器人动作拆成一个个语义小单元，让大模型像下指令一样选择，再由机器人自己翻译成具体动作。这样不用专门训练大模型，就能让现成的大模型零样本控制机器人，也能低成本微调小模型。
基础视觉语言模型（VLM）展现出广泛的世界智能，但将其智能转化为机器人控制仍具挑战。我们提出Show-Harness，一种具身化框架（Embodied Harness），通过紧凑的语义接口将意图与动作连接，使VLM能够“玩转”机器人。Show-Harness暴露离散的语义动作单元，VLM可自然对其进行推理，而具身特定的解释器则确定性地将其落地为本地机器人动作，同时让VLM直接负责细粒度的物理决策。通过同一接口，Show-Harness证明了以下可行性：（1）直接解锁闭源前沿VLM进行零样本机器人控制；（2）仅需数GPU小时的微调即可适配小规模开源VLM以实现低成本部署。我们进一步开发了GUMI（GUI操作接口），将相同的语义动作空间扩展到基于GUI的演示收集，使人类和智能体无需专用遥操作硬件即可跨具身“玩转”机器人。大量实验表明，配备Show-Harness的VLM智能体在任务、具身和环境间均能稳健泛化，性能优于代表性的智能体范式和VLA范式。这些结果表明，正确的接口可以从基础VLM中释放出可观的具身能力，而无需增加模型容量或进行昂贵的具身特定预训练。
📄 English Abstract
Foundation vision-language models (VLMs) exhibit broad intelligence about the world, yet translating this intelligence into robot control remains challenging. We present Show-Harness, an Embodied Harness that enables VLMs to “play” robots through a compact semantic interface linking intent to action. Show-Harness exposes discrete semantic action units that VLMs can naturally reason over, while embodiment-specific interpreters deterministically ground them into local robot actions, keeping the VLM directly responsible for fine-grained physical decisions. Through the same interface, Show-Harness demonstrates the feasibility of (1) directly unlocking closed-source frontier VLMs for zero-shot robot control, and (2) adapting small-scale open-source VLMs for low-cost deployment with just a few GPU-hours of fine-tuning. We further develop GUMI (GUI Manipulation Interface), which extends the same semantic action space to GUI-based demonstration collection, allowing humans and agents to “play” robots across embodiments without specialized teleoperation hardware. Extensive experiments show that Show-Harness-equipped VLM agents generalize robustly across tasks, embodiments, and environments, outperforming representative agentic and VLA paradigms. These results suggest that the right interface can unlock substantial embodied capability from foundation VLMs, without requiring additional model capacity or costly embodiment-specific pretraining.
2. 可编程世界模型
- 英文标题：Programmable World Model
- arXiv：2609.10540v1
💡 一句话解读：现有视频世界模型能生成逼真画面，却记不住世界状态、也无法按规则控制。本文把“世界状态变化”和“画面生成”拆开：用程序明确记录每个实体的状态与变化规则，再把这些状态编译成条件信号，交给视频模型渲染成画面，从而做出状态持久、规则可编程的互动游戏。
近期的视频世界模型能生成越来越逼真、可交互的视觉体验，但缺乏在长时间交互中维持持久世界状态并强制执行可编程规则的可靠机制。我们提出“可编程世界模型”（Programmable World Model），该框架将世界状态演化与视觉观测生成解耦：智能体把自然语言指令翻译为可执行程序，用以指定实体状态与状态转移规则，从而直接控制单个实体及其交互；一个轻量级引擎执行这些程序，以更新并维护显式、持久的全局世界状态，包括屏幕外实体和非视觉属性。为连接世界状态与视觉生成，我们引入状态增强的三维有向包围盒（OBB）作为中间表示，它与目标相机轨迹一起被确定性地编译为像素对齐的时空条件信号，供作为生成式渲染器的预训练视频模型使用。该设计让用户能够创建具有预设机制、可单独控制实体、且在整个游玩过程中保持持久世界状态的可玩游戏。我们还提出 CombatStateBench，一个用于评估可编程世界模型的基准。在该基准上，我们的方法达到 94% 的计数准确率和 98% 的状态准确率，显著优于现有交互式视频世界模型，同时支持连贯的长时程生成。这些结果表明，将显式状态演化与生成式渲染分离，对于构建持久、可编程的世界是有效的。
📄 English Abstract
Recent video world models generate increasingly realistic and interactive visual experiences, yet lack reliable mechanisms for maintaining persistent world state and enforcing programmable rules over extended interactions. We introduce Programmable World Model, a framework that decouples world-state evolution from visual observation generation. An agent translates natural-language instructions into executable programs that specify entity states and state-transition rules, enabling direct control over individual entities and their interactions. A lightweight engine executes these programs to update and maintain an explicit, persistent global world state, including off-screen entities and non-visual attributes. To connect world state with visual generation, we introduce state-augmented 3D oriented bounding boxes (OBBs) as an intermediate representation. This representation, together with the target camera trajectory, is deterministically compiled into pixel-aligned spatiotemporal conditioning signals for a pretrained video model serving as the generative renderer. This design allows users to create playable games with predefined mechanics, direct control over individual entities, and persistent world state throughout gameplay. We further introduce CombatStateBench, a benchmark for evaluating programmable world models. On CombatStateBench, our method achieves 94% Count Accuracy and 98% State Accuracy, substantially outperforming existing interactive video world models while supporting coherent long-horizon generation. These results demonstrate the effectiveness of separating explicit state evolution from generative rendering for building persistent, programmable worlds.
3. AgentGrad：干预引导的多智能体系统提示优化
- 英文标题：AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems
- arXiv：2609.08572v1
💡 一句话解读：多智能体系统里每个智能体的提示词设计很关键，但现有自动优化方法有两个毛病：一是改提示词前不验证是否真能解决问题，二是把不相关的错误混在一起优化，导致效果不好。AgentGrad通过逐个干预智能体来精准定位该改谁，再把相似的错误反馈聚类归纳，从而更准更快地优化提示词。
基于大语言模型（LLM）的多智能体系统（MAS）通过使用专门化的多个智能体实现了强大性能，但其表现依赖于每个智能体的提示词设计。在MAS提示词优化中，利用自然语言反馈指导提示词更新的文本梯度方法已成为主流范式。本文指出，现有文本梯度方法在梯度提取和梯度聚合两个阶段存在局限：梯度提取时，先前工作选择目标提示词却未验证修改它能否解决失败，且推导梯度时缺乏对相应智能体中间输出的智能体级监督；梯度聚合时，单个梯度被随机分组并拼接，常混合不相关的失败模式，导致生成的提示词难以泛化。为解决这些局限，我们提出AgentGrad——一个基于顺序干预和语义文本梯度抽象的多智能体系统提示词优化框架。对于每个失败，顺序干预每次修改一个智能体的行为，以识别出修改后能解决失败的目标智能体；目标智能体修改后的输出则作为智能体级监督，用于提取细粒度梯度。语义文本梯度抽象对语义相似的梯度进行聚类，以防止混合不相关的失败模式，并将每个聚类抽象为一个能捕捉共享纠正模式的泛化梯度。实验结果表明，AgentGrad在五个MAS基准上达到最先进性能，且与次快基线相比，平均减少2.5倍的挂钟优化时间。
📄 English Abstract
Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language feedback have emerged as a leading paradigm. In this paper, we identify limitations in two stages of existing textual gradient approaches: gradient extraction and gradient aggregation. In gradient extraction, previous works select a target prompt without verifying whether modifying it resolves the failure, and derive gradients without agent-level supervision over the corresponding agent’s intermediate output. In gradient aggregation, individual gradients are randomly grouped and concatenated, often mixing unrelated failure modes and producing prompts that fail to generalize. To address these limitations, we propose \textbf{AgentGrad}, a prompt optimization framework for multi-agent systems based on sequential intervention and semantic textual gradient abstraction. For each failure, sequential intervention modifies the behavior of one agent at a time to identify the target agent whose modification resolves the failure. The modified output of the target agent then serves as agent-level supervision for extracting a fine-grained gradient. Semantic textual gradient abstraction clusters semantically similar gradients to prevent mixing unrelated failure modes, and abstracts each cluster into a generalized gradient that captures the shared corrective pattern. Experimental results show that AgentGrad achieves state-of-the-art performance across five MAS benchmarks and reduces wall-clock optimization time by $2.5\times$ on average compared to the next-fastest baseline.
4. WearableQA：面向真实世界可穿戴数据的健康推理基准
- 英文标题：WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data
- arXiv：2609.05405v1
💡 一句话解读：现有AI基准很少测试模型能否理解真实用户长期佩戴设备产生的数据。为此，研究者构建了WearableQA，用200名真实用户最多500天的可穿戴数据、血液指标和人口信息，生成4084道十选一健康推理题，并设计16种题型来区分不同推理能力。测试14个大模型发现，成绩从19.6%到72.9%不等，多数低于60%，说明该任务远未解决。
可穿戴传感技术的最新进展使生理与行为信号的连续监测成为可能，但现有基准很少评估AI系统能否对真实用户的纵向可穿戴记录进行推理。我们提出WearableQA，一个包含4,084道十选一选择题的基准，题目由200名真实用户的可穿戴时间序列、血液生物标志物和人口统计学数据构建，每位用户拥有长达500天的每日测量数据。WearableQA保留了包含设备噪声和个体间差异的真实可穿戴数据分布。为评估不同的推理能力，我们引入16种题型，沿两个互补维度组织：数据推理与健康推理，区分对纵向测量的计算与生理学解释；单信号与跨信号推理，区分对单个信号的推理与多信号整合。为大规模构建可靠题目，我们采用双重锚定框架，将基于文献的生理学发现与经统计验证的群体生理模式相结合，从而捕捉真实可穿戴数据中有意义的关系。对14个专有和开源大语言模型（LLM）的评估表明，WearableQA能有效区分模型能力，性能范围从19.6%到72.9%，而随机基线为10%。此外，WearableQA远未被解决：大多数模型的准确率低于60%。总体而言，WearableQA为评估LLM在真实世界可穿戴数据上的推理能力提供了一个现实且具有诊断性的基准。
📄 English Abstract
Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user’s longitudinal wearable record. We introduce WearableQA, a benchmark comprising 4,084 10-option multiple-choice questions constructed from the wearable time series, blood biomarkers, and demographics of 200 real users, each with up to 500 days of daily measurements. WearableQA preserves authentic wearable distributions that include device noise and inter-individual variability. To evaluate distinct reasoning capabilities, we introduce 16 question types organized along two complementary axes: data versus health reasoning, which distinguishes computation over longitudinal measurements from physiological interpretation; and single- versus cross-signal reasoning, which separates reasoning about individual signals from the integration of multiple signals. To construct reliable questions at scale, we adopt a dual-grounding framework that combines literature-grounded physiological findings with statistically validated population-grounded physiological patterns. This enables the capture of meaningful relationships observed in real-world wearable data. Evaluation of 14 proprietary and open-source LLMs demonstrates that WearableQA effectively differentiates model capabilities, with performance ranging from 19.6% to 72.9% against a 10% chance baseline. Moreover, WearableQA remains far from solved: most models achieve accuracies below 60%. Overall, WearableQA provides a realistic and diagnostic benchmark for evaluating LLM reasoning over real-world wearable data.
5. SWE-Bench Pro Verified：面向软件工程智能体的可靠基准
- 英文标题：SWE-Bench Pro Verified: A Reliable Benchmark for Software Engineering Agents
- arXiv：2609.08149v1
💡 一句话解读：这篇研究要解决的是：现有评测软件工程 AI 的基准 SWE-Bench Pro 不太靠谱，题目可能泄题、题目描述有误导、测试范围也不准，导致 AI 成绩虚高。作者的做法是堵住泄题漏洞，并小幅修正有问题的题目，做出更可信的版本，结果发现一些 AI 的真实水平其实比之前报告的低不少。
SWE-Bench Pro 已成为评估软件工程智能体处理高难度仓库级任务的标准基准。然而，我们的分析工作表明，其评估受到两类不可靠因素的损害：一是奖励黑客（reward hacking），源于金标准解决方案或隐藏评估信息的泄露；二是任务质量问题，包括误导性的问题描述和范围不当的测试。这些问题会抬高基准成绩，掩盖智能体真实的编码能力。我们提出 SWE-Bench Pro Verified，这是 SWE-Bench Pro 的验证版本，可同时解决上述两个问题。我们的方法结合了反黑客（anti-hacking）防护措施——在不破坏智能体正常功能的前提下消除主要泄露渠道——以及任务精修（task refinement），对有缺陷实例内部的不一致进行最小化修正。在 SWE-Bench Pro Verified 上的评估显示，一些模型的表现远低于此前报告，这表明现有 SWE-Bench Pro 上的结果可能高估了真实的软件工程能力。SWE-Bench Pro Verified 为评估软件工程智能体提供了一个更可信的基准。
📄 English Abstract
SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: \textbf{reward hacking}, enabled by leakage of gold solutions or hidden evaluation information, and \textbf{task quality issues}, including misleading problem statements and improperly scoped tests. These issues can inflate benchmark performance and obscure agents’ true coding ability. We present \textbf{SWE-Bench Pro Verified}, a verified version of SWE-Bench Pro that addresses both problems. Our approach combines \textbf{anti-hacking} safeguards that eliminate major leakage channels without disrupting normal agent functionality, with \textbf{task refinement} that minimally corrects inconsistencies within flawed instances. Evaluations on SWE-Bench Pro Verified reveal that some models perform substantially worse than previously reported, suggesting that existing results on SWE-Bench Pro may overestimate real software engineering capability. SWE-Bench Pro Verified offers a more trustworthy benchmark for assessing software engineering agents.
6. SAEScientist-Bench：AI智能体能否开展自主SAE可解释性研究？
- 英文标题：SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?
- arXiv：2609.09113v1
💡 一句话解读：这篇研究要解决的问题是：AI能否像科学家一样，自己动手用稀疏自编码器（SAE）工具去搞懂大模型内部到底学到了什么。为此，作者搭建了一个测试平台，让AI智能体在13万多个特征里自主设计对比实验、寻找最佳可解释特征，并用专家标注结果来打分。结论是：AI有一定自主发现能力，但还明显不如人类专家。
尽管关于递归自我改进（RSI）的研究主要是在自动化模型训练流程，但可靠的自主开发还缺少一个关键支柱：用于理解模型所学内容并确保安全对齐的事后监测与审计。机制可解释性工具是弥合这一缺口的关键，其中稀疏自编码器（Sparse Autoencoders, SAEs）作为基石，能够分离出可解释特征以用于模型检查和引导。本文提出 SAEScientist-Bench，用以评估 AI 智能体能否像科学家一样利用 SAE 工具进行自主机制发现。给定一个目标概念，智能体会设计对比探针，并在 Gemma-2-9B-IT 的 Gemma Scope 字典中导航 131K+ 个特征，以发现最优特征；评估依据是 Neuronpedia 上经过整理的专家参考特征，涵盖激活排名、对比文本上的概念选择性以及因果引导三个维度。在 10 种智能体配置和 20 项任务中，前沿智能体展现出真实的发现能力，并在不同评估维度上领先，但仍明显落后于专家基线：在将目标概念与对比控制区分开方面接近专家水平，而在因果生成引导方面则差距较大。进一步分析表明，尽管智能体能够设计对比以排除虚假候选，但它们经常误读实验测量结果。这些结果确立了将实验性模型理解作为闭环自主 AI 研发中一项可衡量能力。我们的代码见 https://github.com/Trae1ounG/SAEScientist。
📄 English Abstract
While research on recursive self-improvement (RSI) has predominantly automated model training pipelines, reliable autonomous development demands a missing pillar: post-hoc monitoring and auditing to understand what models learn and ensure safe alignment. Mechanistic interpretability tools are essential to bridge this gap, among which Sparse Autoencoders (SAEs) serve as a cornerstone by isolating interpretable features for model inspection and steering. In this paper, we introduce SAEScientist-Bench to evaluate whether AI agents can act as scientists utilizing SAE tools for autonomous mechanistic discovery. Given a target concept, an agent designs contrastive probes and navigates a Gemma Scope dictionary of 131K+ features in Gemma-2-9B-IT to discover the optimal feature, evaluated against curated expert reference features anchored on Neuronpedia across activation rank, concept selectivity on contrastive texts, and causal steering. Across 10 agent configurations and 20 tasks, frontier agents demonstrate genuine discovery capabilities and lead different evaluation dimensions, but remain well behind the expert baseline, approaching expert levels on separating target concepts from contrastive controls while lagging substantially in causal generation steering. Further analysis reveals that although agents can design contrasts to rule out spurious candidates, they frequently misinterpret experimental measurements. These results establish experimental model understanding as a measurable capability for closed-loop autonomous AI R&D. Our code is available at https://github.com/Trae1ounG/SAEScientist.
7. 仅凭分数不能证明发现：用于审计 AI 研究代理的发现认证协议
- 英文标题：Scores Alone Do Not Prove Discovery: The Discovery Certification Protocol for Auditing AI Research Agents
- arXiv：2609.09219v1
💡 一句话解读：这篇论文要解决的是：AI研究代理声称有发现时，光看分数高不能证明它真的发现了新东西。作者提出一套认证协议，用可执行的恢复测试和反馈测试来审计这些声明，确保结果不是靠记忆或走捷径得来的。
AI研究代理结合先验知识、公开来源和实验反馈来产生有用结果。发现认证协议（DCP）将这些结果的声明转化为可执行的恢复测试和反馈测试。第一关（Gate 1）在密封评估上验证有用改进；第二关（Gate 2）向匹配的代理提供注册的起始信息和观察到的网络内容，同时隐藏目标研究历史。每个达到数值目标的有效方法都提供一个恢复见证并触发核心否决。DCP核心要求足够的控制、零观察到的恢复，以及在一个新的注册回合中对恢复的有限样本界。可选的第三关（Gate 3）从共享检查点测量真实反馈相对于指定中性策略的平均效应。DCP证据在独立零校准和注册效应边际之后加入这一效应。两项受控审计在不同模型下对SQLite优化和虚拟催化剂控制执行了完整协议。每项在96个回合中产生零恢复，上限为0.0468。每项配对研究产生30个真实恢复和零个中性恢复，并通过了60对零研究。其他案例测试了核心、已恢复和审计不完整的决策。一个确定性、无大语言模型的验证器从冻结证据中重现决策。DCP为AI研究中的有用结果、替代路径和反馈效应提供了一种通用证据语言。
📄 English Abstract
AI research agents combine prior knowledge, public sources, and experimental feedback to produce useful results. The Discovery Certification Protocol (DCP) turns claims about these results into executable recovery and feedback tests. Gate 1 validates useful improvement on sealed evaluation. Gate 2 gives matched agents the registered starting information and observed Web content while withholding the target research history. Every valid method reaching the numerical target supplies a recovery witness and triggers the Core veto. DCP Core requires adequate controls, zero observed recoveries, and a finite-sample bound on recovery in one fresh registered episode. Optional Gate 3 measures the average effect of truthful feedback relative to a specified neutral policy from a shared checkpoint. DCP Evidence adds this effect after independent null calibration and a registered effect margin. Two controlled audits exercise the complete protocol in SQLite optimization and virtual catalyst control under different models. Each produced zero recoveries in 96 episodes, with an upper bound of 0.0468. Each paired study yielded 30 truthful recoveries and zero neutral recoveries, with passing 60-pair null studies. Additional cases exercise Core, recovered, and audit-incomplete decisions. A deterministic, LLM-free verifier reproduces the decisions from frozen evidence. DCP provides a common evidence language for useful outcomes, alternative routes, and feedback effects across AI research.
8. Puppeteer：基于物体锚定与姿态感知的共语手势生成
- 英文标题：Puppeteer: Object-Grounded Posture-Aware Co-Speech Gesture Generation
- arXiv：2609.00369v1
💡 一句话解读：现有语音驱动手势模型只关注音频与动作对齐，忽略了姿态约束和周围物体，导致生成的手势与物理空间脱节。Puppeteer 将长手势拆成基本单元，用因果变分自编码器编码成按时间排列的潜在表示，再在潜在空间做条件扩散，结合语音、动作历史、初始姿态和物体几何来合成手势，并支持中间补全和手势续写。
生成时间连贯、语义上与语音对齐、且与周围物体有合理关联的共语手势仍具挑战。现有语音驱动手势模型强调音频-手势对齐，但未显式考虑姿态约束或周围物体，无法捕捉身体手势与物理空间之间的内在关联。我们提出 Puppeteer，一种在因果潜在空间中运行、姿态感知且物体锚定的共语手势扩散模型。我们将长手势分解为结构化基元，并学习一个因果变分自编码器，将其编码为按时间排序的潜在令牌，每个令牌仅依赖于过去。然后，我们直接在因果潜在空间中进行条件扩散，以语音信号、运动历史、初始姿态参考和物体几何为条件，合成物理上一致的手势。这种按时间排序的潜在表示支持显式时间控制，并支持手势中间补全和手势续写等任务。为了更好地评估共语手势合成（超越现有指标），我们引入了针对该任务定制的新评估指标。我们还创建了 SceneGes，这是首个精心整理的、包含具身共语手势及对应 3D 物体的合成 3D 数据集，支持物体锚定的手势生成。实验表明，Puppeteer 比先前方法生成更多样、时间同步更好的手势，同时支持物体锚定的手势合成。
📄 English Abstract
Generating co-speech gestures that are temporally coherent, semantically aligned with speech, and grounded with surrounding objects remains challenging. Prior speech-driven gesture models emphasize audio-gesture alignment but do not explicitly account for posture constraints or surrounding objects, failing to capture the inherent correlation between body gestures and the physical space. We present Puppeteer, a posture-aware, object-grounded co-speech gesture diffusion model operating in a causal latent space. We decompose long gestures into structured primitives and learn a causal variational autoencoder that encodes them into temporally ordered latent tokens, each depending only on the past. We then perform conditional diffusion directly in the causal latent space, conditioning on speech signals, motion history, an initial posture reference, and object geometry to synthesize physically consistent gestures. This temporally ordered latent formulation enables explicit temporal control and supports tasks such as gesture in-betweening and gesture completion. To better assess co-speech gesture synthesis beyond existing measures, we introduce new evaluation metrics tailored to this task. We also created SceneGes, the first curated synthetic 3D dataset of embodied co-speech gestures and corresponding 3D objects, enabling object-grounded gesture generation. Experiments show that Puppeteer generates more diverse and temporally synchronized gestures than prior methods, while enabling object-grounded gesture synthesis.
9. DianShi-RxnDB：面向研究人员与AI智能体的、通过全自动流程构建的大规模细粒度有机反应数据平台
- 英文标题：DianShi-RxnDB: A Large-Scale, Fine-Grained Organic Reaction Data Platform Built via a Fully Automated Pipeline for Researchers and AI Agents
- arXiv：2609.06703v1
💡 一句话解读：化学AI研究需要大量高质量的反应数据，但这些知识散落在专利的文字、图片和反应式里，难以直接使用。本研究用全自动流程把这些专利内容提取、整理成一个大型数据库，收录约2400万条反应记录，并提供网页工具和供AI调用的接口。
高质量的结构化有机反应数据对发展化学人工智能（AI4Chem）至关重要，但这类知识大量分散在专利文本、图像和反应式中。我们提出 DianShi-RxnDB，这是一个通过整合专利文本、图像与反应式的全自动提取与规范化流程构建的大规模细粒度有机反应数据平台。其语料涵盖美国专利商标局（USPTO）与欧洲专利局（EPO）1976至2025年间公布的有机合成专利，产出约2400万条反应实例，其中约1480万条（61.7%）通过自动合格性检查。每条实例代表一次具体的单步实验，记录参与物、角色、用量、温度、反应时间、产率、实验步骤以及与来源专利的溯源链接。在对1300条抽样的合格实例进行人工评估时，字段级微平均准确率为92.95%。与 Pistachio 的匹配对比进一步显示，该平台在去重记录数、表示粒度及字段级完全一致率方面具有优势。该平台提供用于检索、筛选、比较和来源核验记录的 Web 研究工作台，以及为 AI 智能体提供可组合结构化检索工具的模型上下文协议（MCP）服务。DianShi-RxnDB 可通过 https://dianshi.opendatalab.org.cn/ 访问。
📄 English Abstract
High-quality structured organic reaction data are essential for developing artificial intelligence for chemistry (AI4Chem), yet much of this knowledge remains dispersed across patent text, images, and reaction schemes. We present DianShi-RxnDB, a large-scale, fine-grained organic reaction data platform built via a fully automated extraction and normalization pipeline integrating patent text, images, and reaction schemes. Its corpus covers organic synthesis patents from the USPTO and EPO published between 1976 and 2025, yielding approximately 24 million reaction instances, of which approximately 14.8 million (61.7%) pass automated qualification checks. Each instance represents a specific single-step experiment recording participants, roles, quantities, temperatures, reaction times, yields, experimental procedures, and provenance links to source patents. In a manual evaluation of 1,300 sampled qualified instances, the micro-averaged field-level accuracy was 92.95%. A matched comparison with Pistachio further indicated advantages in deduplicated record counts, representation granularity, and field-level exact agreement. The platform provides a Web research workbench for searching, filtering, comparing, and source-verifying records, and a Model Context Protocol (MCP) service offering AI agents composable structured retrieval tools. DianShi-RxnDB is available at https://dianshi.opendatalab.org.cn/ .
10. SyncWorld：视觉校准使世界模型成为零样本模拟器
- 英文标题：SyncWorld: Visual Calibration Enables World Models as Zero-Shot Simulators
- arXiv：2609.09155v1
💡 一句话解读：机器人世界模型在跨环境时，同样的动作指令会因摄像头、机器人外观等变化而产生不同视觉效果，导致模型难以泛化。SyncWorld 通过一段“视觉校准”片段（展示所有可控动作的配对画面）让模型现场理解当前设置下的动作与画面关系，从而无需额外训练就能在新环境中准确模拟动作结果，甚至用于改进策略。
世界模型正越来越多地被用作策略在环的想象环境，其中可靠的推演需要对底层机器人动作具备细粒度可控性。在机器人领域扩展此类模型的一个关键障碍是，动作在像素空间中并非通用语言：视觉环境、相机视角、机器人位置或形态的变化会改变同一数值动作在视觉上的表现，导致混合训练下监督信号冲突，并在部署时泛化脆弱。我们提出 SyncWorld，一种以动作为条件的 world model（世界模型），无需任何额外训练即可作为跨未见环境的零样本模拟器。SyncWorld 利用一段视觉校准片段——展示所有可控自由度的配对帧与动作——在上下文中指定特定设置的动作-视觉映射（Action–Visual Mapping）。使用视觉校准上下文进行训练，教会模型通过视觉证据解读动作，并在没有显式校准时利用交互历史。实验表明，SyncWorld 能在先前未见的设置中准确模拟动作结果，并且其模拟推演的能力支持无需训练的测试时策略改进。
📄 English Abstract
World models are increasingly used as policy-in-the-loop imagination environments, where reliable rollouts require fine-grained controllability with respect to low-level robot actions. A key obstacle to scaling such models in robotics is that actions are not a universal language in pixel space: changes in visual environment, camera view, robot placement, or embodiment alter how the same numerical action manifests visually, leading to conflicting supervision under mixed training and brittle generalization at deployment. We introduce SyncWorld, an action-conditioned world model that serves as a zero-shot simulator across unseen environments without any additional training. SyncWorld leverages a visual calibration episode—paired frames and actions that showcase all the controllable degrees of freedom—to specify the setup-specific Action–Visual Mapping in context. Training with visual calibration contexts teaches the model to interpret actions through visual evidence and to leverage interaction history when explicit calibration is unavailable. Experiments show that SyncWorld can accurately simulate action outcomes in previously unseen settings, and that its capability of simulating rollouts enables test-time policy improvement without training.
11. Φ-Bench：大型语言模型能否工程化支撑自身的基础设施？
- 英文标题：$Φ$-Bench: Can Large Language Models Engineer the Infrastructure That Powers Them?
- arXiv：2609.10226v1
💡 一句话解读：现有测试只考大模型写零散代码或按指定目标优化，无法衡量它们能否独立完成整套AI基础设施的长期工程。为此，研究者构建了Φ-Bench，从真实代码库和前沿研究中提炼出从局部函数补全到系统级优化的多难度任务，系统评测大模型工程化AI基础设施的能力。
大型语言模型（LLM）在推理和代码生成方面展现出卓越能力，这使人们期待它们能协助开发和优化支撑其自身运行的基础设施。然而，现有基准测试主要关注孤立的核函数、预定义算子或预先指定的优化目标，因此无法评估LLM进行开放式、长周期LLM基础设施工程的能力。为填补这一空白，我们提出了Φ-Bench，一个用于系统评估LLM工程化LLM基础设施栈的基准。Φ-Bench源自前沿研究中研究的优化问题，并以真实世界的代码仓库为基础，广泛覆盖LLM基础设施栈，涵盖从局部核函数级函数补全到长周期实现和端到端系统优化等不同复杂度的任务。对前沿LLM的大量实验揭示了它们当前在工程化复杂LLM基础设施方面的能力与局限，为未来AI基础设施自主优化道路上仍存在的挑战提供了洞见。
📄 English Abstract
Large language models (LLMs) have demonstrated remarkable capabilities in reasoning and code generation, raising the prospect that they could assist in developing and optimizing the very infrastructure that powers them. However, existing benchmarks mainly focus on isolated kernels, predefined operators, or pre-specified optimization targets, and therefore fail to evaluate the ability of LLMs to perform open-ended, long-horizon LLM infrastructure engineering. To address this gap, we present $Φ$-Bench, a benchmark for systematically evaluating LLMs on engineering the LLM infrastructure stack. Derived from optimization problems studied in frontier research and grounded in real-world code repositories, $Φ$-Bench provides broad coverage of the LLM infrastructure stack and spans tasks of varying complexity, ranging from localized kernel-level function completion to long-horizon implementation and end-to-end system optimization. Extensive experiments on frontier LLMs reveal their current capabilities and limitations in engineering complex LLM infrastructure, offering insights into the challenges that remain on the path toward autonomous optimization of future AI infrastructure.
12. 重新审视后训练中的完整推理轨迹
- 英文标题：Revisiting Complete Reasoning Traces for Post-Training
- arXiv：2609.07103v1
💡 一句话解读：这篇研究要解决的问题是：训练大模型做推理时，是否真的需要喂给它完整、冗长的推理过程？作者发现完整过程帮助有限，而截断后的部分过程反而有效，因为中间步骤大多是冗余的。他们提出用起点和终点（端点）来训练模型，让模型自己补全缺失步骤，这种方法在监督微调、强化学习和同策略蒸馏中都有效。
大型语言模型（LLMs）通常在后训练阶段使用预先收集的推理轨迹来提升推理能力。这些轨迹由于路径复杂交织而往往很长，其中常包含通往答案途中的绕路。然而，在后训练（如有监督微调，SFT）中，LLMs 是否真的能从学习完整轨迹中获益，这一问题尚未得到充分探索。基于我们的初步研究，我们发现完整轨迹带来的收益有限，而部分轨迹即使经过大幅截断也仍然有效。我们通过基于注意力的分析和受控的 token 移除研究，分析了推理轨迹中的冗余，两者均表明中间 token 对最终推理质量的贡献极小。这表明，避免冗余信息可能让 LLMs 在已知轨迹端点的情况下，从其内部知识中推断出缺失步骤，从而在内部构建连贯的替代方案。此外，我们表明使用端点训练 LLMs 会导致推理行为的一致变化，并且它也有益于基于强化学习或同策略蒸馏的后训练方法，这凸显了重新审视完整推理轨迹的必要性。代码见 https://github.com/naver-ai/revisiting-trace。
📄 English Abstract
Large language models (LLMs) are often post-trained on pre-collected reasoning trajectories to improve their reasoning capability. Such trajectories tend to be long due to complex, interwoven paths, which often include detours on the path toward the answer. However, it has been underexplored whether LLMs indeed benefit from learning complete trajectories in post-training, such as supervised fine-tuning (SFT). Starting from our pilot study, we find that full trajectories provide only limited benefit, while partial trajectories are effective even under heavy truncation. We analyze redundancy in reasoning trajectories through attention-based analyses and controlled token-removal studies, both of which show that intermediate tokens contribute minimally to final reasoning quality. This suggests that avoiding redundant information may allow LLMs to internally infer coherent alternatives by inferring missing steps from their internal knowledge, given known trajectory endpoints. Furthermore, we show that training LLMs using endpoints leads to consistent changes in reasoning behavior, and that it also benefits post-training methods based on reinforcement learning or on-policy distillation, highlighting the need to revisit complete reasoning traces. Code is available at https://github.com/naver-ai/revisiting-trace.
13. 训练更智能，而非更费力：主动学习中的切换信号引导训练
- 英文标题：Train Smarter, Not Harder: Switching Signal-Guided Training in Active Learning
- arXiv：2609.06806v1
💡 一句话解读：主动学习里每次拿到新数据后，是「从头重训」还是「接着微调」一直没人认真研究。本文发现早期重训更有用、模型稳定后微调更划算，于是提出 HybridAL：实时监测模型是否稳定，稳定了就自动从重训切到微调，既省时间又不牺牲效果。
训练策略——即从头重新训练还是从上一个检查点微调——是主动学习中被忽视的决策变量。我们表明这一选择具有可利用的结构：在早期轮次，重训最有用，因为每个批次都能显著重塑已标注数据的分布；而一旦模型轨迹趋于稳定，微调就变得更安全。我们提出 HybridAL，一种自适应训练调度方法，它监测一个在线稳定信号，并在持续稳定后从重训切换为微调。两个互补信号——基于权重的谱指数变化 Δα 与基于验证集的准确率变化 ΔAcc——分别位于时间-校准权衡的不同位置。在三种编码器骨干和六个文本分类任务上（每个任务五个随机种子），HybridAL 的最终宏 F1 在 0.010 的裕度内不劣于重训和微调，最多节省 49% 的重训时间，并恢复了重训在校准方面（以负对数似然 NLL 衡量）优势的相当一部分。与在预先设定的轮次进行切换的调度相比，HybridAL 以适度的额外开销获得更低的 NLL，表明依赖轨迹的切换比固定的提前切换能提供更强的时间-校准权衡。
📄 English Abstract
Training strategy, namely whether to retrain from scratch or fine-tune from the previous checkpoint, is an overlooked decision variable in active learning. We show that this choice has exploitable structure: retraining is most useful in early rounds, when each batch can substantially reshape the labeled distribution, while fine-tuning becomes safer once the model trajectory stabilizes. We propose HybridAL, an adaptive training schedule that monitors an online stabilization signal and switches from retraining to fine-tuning after sustained stabilization. Two complementary signals, spectral exponent change $Δα$ (weight-based) and accuracy change $Δ$Acc (validation-based), span different points on the time-calibration trade-off. Across three encoder backbones and six text-classification tasks (five seeds each), HybridAL keeps endpoint macro-F1 non-inferior to retraining and fine-tuning at a 0.010 margin, saves up to 49% of retraining time, and recovers a substantial fraction of retraining’s calibration advantage as measured by negative log-likelihood (NLL). Compared with schedules that switch at a pre-committed round, HybridAL obtains lower NLL at moderate additional cost, showing that trajectory-dependent switching provides a stronger time-calibration trade-off than fixed early switching.
14. 为什么视频仍然如此昂贵？视频与音视频大语言模型中的推理效率机制综述
- 英文标题：Why Is Video Still So Expensive? A Survey of Inference-Efficiency Mechanisms in Video and Audiovisual LLMs
- arXiv：2609.10355v1
💡 一句话解读：视频大模型理解视频时，处理帧数越多、上下文越长，计算和内存开销就越大，导致难以在手机或实时场景中使用。这篇综述系统梳理了各种降低推理成本的方法，比如减少采样帧数、压缩视觉和音频标记、优化模型计算流程，并对比了不同方法在精度和开销上的权衡。
视频理解正快速向视频大语言模型（VideoLLMs）演进：这类系统将视频表示与预训练大语言模型耦合，并基于文本提示进行条件生成。它们在字幕生成、问答、检索和时间定位等任务上表现强劲，但计算和内存开销随帧数和上下文长度增长，限制了在实时、移动和资源受限场景中的部署。本综述涵盖视觉和音视频VideoLLMs的推理效率机制，这些机制报告了在参数量、每次输入的浮点运算次数（FLOPs）、延迟、内存或视觉与音频标记数量上的具体降低。我们分析了帧采样、模态编码、连接器级标记缩减以及LLM预填充和解码等环节的瓶颈。我们按方法在流程中所处的阶段进行组织，覆盖2022年末以来开发的VideoLLMs，以及当前流程中仍作为组件的早期帧采样和视觉编码器机制。我们汇总了在相同宿主模型和输入协议下文献报告的精度-成本对比（如有），将其与异构的跨论文证据区分开，并指出了音视频效率和标准化评估方面的空白。我们在https://github.com/momentslab/awesome-efficient-videollm维护了一个代码库。
📄 English Abstract
Video understanding has rapidly evolved toward video large language models (VideoLLMs): systems that couple video representations with pretrained large language models and condition generation on a textual prompt. Their strong performance on captioning, question answering, retrieval and temporal grounding comes at a computation and memory cost that grows with frame count and context length, limiting deployment in real-time, mobile and resource-constrained settings. This survey covers inference-efficiency mechanisms for visual and audiovisual VideoLLMs that report concrete reductions in parameter count, FLOPs per input, latency, memory, or visual and audio token count. We analyze bottlenecks across frame sampling, modality encoding, connector-level token reduction, and LLM prefilling and decoding. We organize methods by the pipeline stage at which they act, covering VideoLLMs developed since late 2022 together with earlier frame-sampling and vision-encoder mechanisms that remain components of current pipelines. We assemble literature-reported accuracy–cost comparisons under shared host models and input protocols wherever available, distinguish them from heterogeneous cross-paper evidence, and identify gaps in audiovisual efficiency and standardized evaluation. We maintain a repository at https://github.com/momentslab/awesome-efficient-videollm.
15. 协同演化智能体运行框架与模型：同策略修正帮助较弱模型在模仿失败之处追赶
- 英文标题：Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails
- arXiv：2609.09134v1
💡 一句话解读：小模型做专业任务时，靠一套“外挂配置”（提示词、工具、流程脚手架）能省钱又出活。但研究发现，若直接让小模型照抄强专家的完整操作轨迹，反而会破坏它与这套配置的契合，七个任务全部退步。作者改为只让专家修正小模型自己操作中出错的那一步，既保留其原有思路，又同时收获配置进化和模型微调的好处。
智能体运行框架（agent harness，即模型周围的系统提示、工具集、执行钩子与上下文管理脚手架）是决定智能体任务成败的关键因素。自动化的运行框架演化能让较小模型以远低于前沿模型的成本在特定领域任务上表现良好。由于运行框架与模型权重共同塑造行为，我们探究二者应如何结合。在七项企业级智能体任务上，我们先用较弱模型演化出一套运行框架，随后发现更强的专家模型往往能更有效地使用它，这提示专家监督或可弥合剩余差距。然而，在演化后的运行框架下用专家的完整轨迹训练较弱模型却适得其反：在 Qwen3-Coder 与 Gemma 4 上，七项任务性能全部回退 4 至 30 分，尽管同样的做法在未演化的运行框架下是有益的。分析表明，模仿虽能传递知识并提升脚手架使用率，却破坏了模型与运行框架的契合：较弱模型采纳了专家的规划策略，却不具备执行该策略的能力，也不再匹配围绕其原生规划风格演化出的运行框架。为此，我们提出一种同策略专家修正流程，由元层 MLE 智能体自动完成：在较弱模型自身的 rollout 中定位失败的那一回合，仅请专家重写该回合。这样既保留了模型的规划风格，又结合了运行框架演化与模型适配的收益。我们的结果识别并化解了运行框架更新与权重更新之间的一处冲突，为特定领域企业任务上的经济型协同演化提供了保持兼容性的方案。
📄 English Abstract
Agent harnesses (the system prompt, tool set, execution hooks, and context-management scaffolding around a model) are a critical determinant of agentic task success. Automated harness evolution can enable smaller models to perform well on domain-specific tasks at a fraction of frontier-model cost. Since both the harness and model weights shape behavior, we ask how harness evolution and lightweight fine-tuning should be combined. Across seven enterprise agent tasks, we first evolve a harness with the weaker model, then find that a stronger expert often uses it more effectively, suggesting expert supervision could close the remaining gap. However, training the weaker model on the expert’s complete trajectories under the evolved harness backfires: performance regresses on all seven tasks by 4 to 30 points across Qwen3-Coder and Gemma 4, even though the same procedure helps under the unevolved harness. Our analysis shows that imitation transfers knowledge and increases scaffold usage, but disrupts model-harness fit: the weaker model adopts the expert’s planning strategy without the competence to execute it and no longer matches the harness evolved around its native planning style. We therefore develop an on-policy expert-correction pipeline, automated by a meta-level MLE agent, that localizes the failing turn in the weaker model’s own rollout and asks the expert to rewrite only that turn. This preserves the model’s planning style and combines the gains of harness evolution and model adaptation. Our results identify and resolve a source of contention between harness and weight updates, yielding a compatibility-preserving recipe for economical co-evolution on domain-specific enterprise tasks.