边云 LLM 系统中面向函数调用的碳感知路由框架
针对当前 Agent 系统中函数调用过度依赖高能耗云端大模型的问题，研究团队提出了一种面向三层边云异构架构的碳感知路由框架。该框架利用统一语义-词汇嵌入空间下的轻量级 k-NN 预测器，评估各边缘层级的查询准确率、延迟与功耗，并结合电网实时碳强度将请求分发至可成功执行的最低排放层级。测试结果显示，该方法在保持云端级准确率的同时将运行碳排放平均降低至原来的 1/4（降低 4 倍）。这为降低大规模智能体系统的环境成本提供了切实可行的工程路径。
独立 AI 资讯库
AI 前沿动态、原文归档与媒体镜像
从自动抓取数据库实时读取，适合快速浏览、检索和后续扩展成完整资讯站。
11977资讯
11845正文归档
75683媒体
AutoTailor：面向 Web Agent 的自动化与用户对齐能力筛选及自适应框架
针对 Web Agent 利用可复用工具降低底层浏览器交互成本和延迟时，自动发现的工具集通常庞大、冗余且与用户需求对齐欠佳的问题，研究者提出了 AutoTailor 框架。该研究构建了一种元智能体（meta-agentic）系统，旨在实现自动且贴合用户需求的工具能力筛选与适配。这一工作为降低智能体交互开销、提升工具选择效率及推进实用化 Web Agent 部署提供了重要参考。
Asclepius：面向长程临床医疗 Agent 的自适应 Harness 框架
当前大语言模型（LLM）Agent 的基准测试多集中在短程、单一任务轨迹上，但实际部署往往需要 Agent 在资源竞争下连续运行数小时，暴露出截然不同的系统失效模式。为此，研究团队提出 Asclepius 框架，并基于临床环境模拟器（Clinical Environment Simulator, CES）让 Agent 应对复杂的长期医疗管理任务。该研究揭示了动态长程环境下 Agent 稳定运行的瓶颈，为自主医疗 Agent 及复杂长程任务 AI 的工程化落地与基准评测提供了重要参考。
Fraglingo：基于位点感知自回归片段生成的分子设计
Fraglingo 提出了一种基于位点感知自回归片段生成的分子设计新框架。该方法旨在贴合化学家的实际研发逻辑，支持在特定连接位点扩展分子骨架、替换取代基或对骨架进行修饰，同时对目标分子特性进行优化。通过将片段级编辑与属性优化直接结合，该方案有效克服了传统原子级生成模型化学合理性较差的局限。这对 AI 制药开发者和计算化学研究人员而言，提供了一种更符合药物化学实践经验的分子优化与从头生成工具。
通过应用程序行为建模实现 Web Agent 的 Token 高效任务执行
随着 AI Agent 在各类复杂任务中展现出强大性能，业界对智能体基础设施的投入达到了空前规模，但不断攀升的 Token 处理成本正成为实际落地的核心瓶颈。本研究聚焦于自动化操作 Web 应用程序的 Web Agent，探索通过应用程序行为建模来优化任务执行流程。该方法致力于在保证任务完成度的同时显著降低 Token 消耗，对于开发者削减大模型推理开销、推进 Web Agent 规模化工程落地具有重要指导意义。
GAVEL：基于病例报告裁决抽取临床时间线差异的 LLM 评判协议
针对从临床病例报告中抽取时间线时面临的人工参考标注不完善、事件对齐不精确等评估难题，研究团队提出了 GAVEL 评测框架。GAVEL 是一种基于大语言模型裁判（LLM Judge）的裁决协议，旨在直接对比并裁定提取出的临床时间线与原始病例报告之间的差异与质量。该方法克服了传统评估流程对人工黄金标准的强依赖，显著增强了复杂临床文本时序提取评估的自动化与可靠性。这项研究为医疗 AI 领域在病程时序抽取、电子病历分析及临床大模型落地评估方面提供了新的基准工具。
OrchSLM：探究小型语言模型（SLM）协同编排的动态机制
针对大语言模型（LLM）重度依赖云端基础设施所引发的高延迟、隐私泄露及高昂成本等问题，本研究提出了探索小型语言模型（SLM）协同编排动态机制的 OrchSLM。研究深入分析了在 Agentic 流水线中如何通过多个轻量化模型的高效协作来替代单一庞大云端模型。这一成果为在边缘设备与受限环境中部署低成本、低延迟且兼顾隐私的 Agent 架构提供了极具价值的技术路线与实践参考。
根因归因本质是搜索问题：面向长程 Agent 故障的持续搜索诊断
随着 AI Agent 在复杂长程（long-horizon）任务中的广泛应用，系统产生了规模庞大的执行日志，准确诊断其中的故障根因成为保障 Agent 可靠性的关键。研究指出，将结果层面的失败信号转化为具体可行的干预策略，其核心可被视作一个搜索问题。通过引入针对故障的持续搜索机制，开发者能够更系统、高效地在海量日志中完成多步骤 Agent 的故障定位与归因，为长流程 Agent 的自动纠错与稳健性提升奠定了基础。
TimeThink：激发时间序列大语言模型的组合推理能力
针对时间序列多模态大语言模型（TS-MLLMs）在问答任务中难以精准捕捉动态时间模式的问题，研究提出了 TimeThink 框架以激发模型的组合推理能力。该方法旨在引导大语言模型对复杂时序特征进行多步骤、组合式的深度推理，从而突破传统时序问答的表达局限。该成果对智能时序分析、工业监测及金融预测等垂直场景的 Agent 落地应用具有积极的参考价值。
LabAgent：利用 AI Agent 定制科研中心赋能科学发现
科学研究是一个高度强调传承的持续过程，前人开发的研究方法通常需要由新入学者进一步拓展以探索更深入的问题。然而，实验室人员更替（如学生毕业、学者流动）等现实因素往往导致科研技术与经验的传承面临断代挑战。为此，该研究提出了 LabAgent，旨在利用 AI Agent 为科学发现量身定制各类科研中心（Research Hubs）。这项工作为解决科研知识沉淀与自动化实验探索提供了全新路径，有助于降低团队继承成本并加速科研成果产出。
迈向自适应物理 AI：LLM Agent 能否胜任长周期物理任务？
该研究深入探讨了基于大语言模型（LLM）的 Agent 在脱离人工干预的情况下自主管理长周期物理任务的可行性。研究指出，物理环境的复杂性要求 Agent 具备持续观察环境状态、执行具有实际因果影响的操作并进行动态自适应的能力。这一探索为具身智能（Embodied AI）的长程决策机制提供了关键洞察，对未来开发高自主性工业机器人与物理自动化系统具有重要指导意义。
Vibe Patenting：面向专业专利撰写 Agent 的 LLM 裁判评估
研究人员提出了用于评估 AI Agent 的端到端专利撰写测试基准 Vibe Patenting，系统探究了 LLM 裁判（LLM Judges）在复杂专业工作中的可靠性。实验表明，通过独立 LLM 裁判提供的结构化反馈进行迭代修改，低推理成本的 Agent 能够逼近昂贵的高推理 Agent 的撰写表现，同时特定领域的 Agent 工作流可进一步提升成稿质量。经专业专利律师的独立评估验证，LLM 裁判展现出具有实际意义但高度依赖具体指标的一致性以及系统性的校准偏差。该研究为开发者将 LLM 裁判应用于专业垂直领域的评估与闭环优化提供了关键依据与实践指导。
广义 Agent 迭代：统一迭代策略改进与递归自我改进的形式化框架
本研究提出了名为「广义 Agent 迭代」的形式化理论框架，旨在统一分析迭代策略改进与递归自我改进（Recursive Self-Improvement, RSI）。论文针对当前在不同尺度上探讨 RSI 时缺乏统一定义与形式化描述的现状，深入反思 RSI 究竟代表一种现象、机制还是一种前景展望。该框架为构建具备自主进化与持续提升能力的智能体系统奠定了理论基石，对探索下一代自演进 AI 架构具有重要指引意义。
ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
arXiv:2609.13356v1 Announce Type: new Abstract: In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency. ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can ove
Model Guides You How to Draw: Adaptive Visual Gating for Unified Multimodal Reasoning
arXiv:2607.10004v1 Announce Type: new Abstract: Unified multimodal models (UMMs) with interleaved reasoning, which generate both textual and visual steps as part of intermediate reasoning traces, have demonstrated great potential for visual mathematical reasoning tasks. However, we identify a key i
Bridging the Catalog-to-Real Gap: Scalable Product Recognition via Multi-Stage Contrastive Learning
arXiv:2607.09888v1 Announce Type: new Abstract: Automated product recognition is a cornerstone of modern retail intelligence; however, accurately matching real-world, in-store images against extensive corporate catalogs remains a major scalability bottleneck for large-scale applications. In this wo
ShapKO: Shapley-Adaptive Modality Knockout for Robust Multimodal Learning
arXiv:2607.09884v1 Announce Type: new Abstract: Multimodal medical models often degrade when inputs are missing, a common scenario in real-world clinical workflows. Separately, even when all modalities are present, modality dominance is observed during training, where optimization over-relies on a
Prompting-MammAlps: Fine-Grained Text-to-Video Retrieval for Camera-Trap Data
arXiv:2607.09876v1 Announce Type: new Abstract: Automatically retrieving videos from large camera-trap datasets remains challenging. Text-to-Video retrieval (TVR) methods based on large video-language models (VLMs) have potential to retrieve events of interest by describing them with simple text qu
Does YOLO26 Truly Offer Advantages Over Its Predecessors for Edge Deployment? A Benchmark Study in Aquaculture
arXiv:2607.09835v1 Announce Type: new Abstract: The recently introduced YOLO26 architecture incorporates NMS-free end-to-end inference and is optimized for deployment on resource-constrained CPU-based devices, making it well-suited for edge-based aquaculture applications. However, its performance,
Memory-Conditioned Tool Calling for Camera-First Visual Agents
arXiv:2607.09822v1 Announce Type: new Abstract: Recognition tells an agent what is in an image; personal memory affects what is worth looking up next. In a camera-first setting the user can send only an image, so the agent must form the lookups. We study whether personal visual memory improves agen
ReflectWorld-MM: An Entity-Oriented Multi-Media Memory System for Open-Ended Video Streams
arXiv:2607.09759v1 Announce Type: new Abstract: Building assistants that can continually watch the world, remember what they see, and reason over their accumulated experience is a long-standing goal, and recently multimodal agents equipped with long-term memory over video streams have attracted inc
From Direction to Magnitude: How Multimodal Instruction-Tuning Reorganizes the Geometric Encoding of Identity-Specifying Prompts in Transformer Hidden States
arXiv:2607.09842v1 Announce Type: new Abstract: We investigate whether identity-specifying system prompts produce statistically distinguishable geometric fingerprints in the hidden-state trajectories of four open-weight transformer language models spanning four post-training regimes: no training (G
Quota Marketplace: Dynamic Pricing for Efficient Allocation of ML Training Resources
arXiv:2607.09802v1 Announce Type: new Abstract: The escalating demand for Machine Learning (ML) training resources in recent years has resulted in a substantial gap between the high demand and the available supply. Efficient allocation of these scarce and expensive resources is crucial for organiza
EvoClawBench：智能体能否从自身运行中学习可重用技能？
现有智能体基准主要评估任务完成、工具使用或技能效用，但无法衡量运行时能否将自身运行证据转化为可重用技能，以在完成编写开销后提升后续新执行的效果。为此，我们提出了 EvoClawBench。该基准旨在分离并测试智能体系统从历史运行中抽象可复用技能的能力，为开发者评估 Agent 框架的学习可累积性提供新维度。
Quantum-Inspired Contextual Learning for Sparse-Ring Fraud Detection in Dynamic Transaction Graphs
arXiv:2607.09704v1 Announce Type: new Abstract: We present an exploratory benchmark and quantum-inspired modeling prototype for fraud screening in dynamic financial transaction graphs. Coordinated fraud may not be visible from individual transactions alone, but may emerge as a multi-period relation
Safe responses matter: Output-aware safety guardrail mitigate over-refusal in MLLMs
arXiv:2607.09697v1 Announce Type: new Abstract: Existing safety mechanisms for multimodal large language models (MLLMs) face a fundamental trade-off between safety and utility. Model fine-tuning achieves robust safety but compromises general utility. Input-side safety guardrails offer a lightweight
Depth-Entropy Guided Sampling for Training-Free LLM Reasoning
arXiv:2607.09693v1 Announce Type: new Abstract: Reinforcement learning (RL) has become the dominant paradigm for improving the reasoning capabilities of large language models, but it requires expensive training, curated data, and reward signals. Recent work shows that sampling from sharpened base-m
What Context Does a Coding Agent Actually Need to Act?
arXiv:2607.09691v1 Announce Type: new Abstract: A modern coding agent can hold an entire repository in its context window. Most of its reading is wasted -- and the interesting question is not how much context an agent can use, but what it actually \emph{needs}. We study that question at the moment
AuditWeave: A Tamper-Evident, Auditor-Navigable Evidence Layer for AI-Assisted and Data-Transformation Workflows
arXiv:2607.09682v1 Announce Type: new Abstract: AI systems are increasingly used to assist consequential decisions in regulated domains such as auditing, finance, and healthcare. This creates a recurring obligation: an organization must be able to reconstruct, after the fact, which evidence informe
Demographic Prompting at Scale: When More Attributes Hurt LLM--Human Agreement
arXiv:2607.10590v1 Announce Type: new Abstract: We investigate how annotator demographic attributes, supplied as prompt cues, shape the alignment between large language model (LLM) predictions and human annotations across five tasks. Using five open-source LLMs, we systematically vary the number an