评分理由: 论文核心研究如何通过轻量级微调方法（\policy）训练小型语言模型（SLMs）成为有效的搜索代理，以解决SLMs在复杂多跳推理任务中搜索工具调用不足和幻觉倾向的问题。因此，与"Small Language Models (SLMs)"、“LLM Agents”、“Tool Use”、“Retrieval-Augmented Generation (RAG)“高度相关（10分）。论文也涉及从大型语言模型（LLMs）蒸馏代理行为，与"Large Language Models (LLMs)“高度相关（10分）。论文提出的微调方法属于监督微调（SFT）范畴，与"Post-training/SFT"相关（8分）。研究任务涉及多跳推理，与"Chain of Thought/CoT Reasoning"相关（8分）。目标之一是减少幻觉，与"Hallucination Mitigation"相关（8分）。其他关键词如MoE、Scaling Laws、RLHF等未在论文中提及或仅边缘相关，给予0分。

配备搜索工具的智能体已成为解决知识密集型任务的有效方案。尽管大语言模型展现出强大的推理能力，但其高昂的计算成本限制了搜索智能体的实际部署。因此，近期研究聚焦于将大语言模型的智能体行为蒸馏至小语言模型中。通过对复杂多跳推理任务的综合评估，我们发现尽管小语言模型具备的参数知识较少，但其调用搜索工具的频次更低，且更容易产生幻觉。为解决这一问题，我们提出\policy方法——一种轻量级微调策略，通过显式训练小语言模型，使其能够基于检索证据进行可靠检索并生成答案。相较于从大语言模型进行智能体蒸馏的方法，我们的方案在Bamboogle数据集上提升了17.3分，在HotpotQA数据集上提升了15.3分，在各基准测试中均达到大语言模型级别的性能。进一步分析表明，小语言模型中的自适应搜索策略往往会导致性能下降，这凸显了稳定搜索行为对可靠推理的必要性。

Agents equipped with search tools have emerged as effective solutions for knowledge-intensive tasks. While Large Language Models (LLMs) exhibit strong reasoning capabilities, their high computational cost limits practical deployment for search agents. Consequently, recent work has focused on distilling agentic behaviors from LLMs into Small Language Models (SLMs). Through comprehensive evaluation on complex multi-hop reasoning tasks, we find that despite possessing less parametric knowledge, SLMs invoke search tools less frequently and are more prone to hallucinations. To address this issue, we propose \policy, a lightweight fine-tuning approach that explicitly trains SLMs to reliably retrieve and generate answers grounded in retrieved evidence. Compared to agent distillation from LLMs, our approach improves performance by 17.3 scores on Bamboogle and 15.3 scores on HotpotQA, achieving LLM-level results across benchmarks. Our further analysis reveals that adaptive search strategies in SLMs often degrade performance, highlighting the necessity of consistent search behavior for reliable reasoning.

评分理由: 论文核心研究LLMs在问答任务中的认知校准问题，通过监督微调（SFT）实现三动作框架（回答、询问、弃权），直接涉及LLMs、SFT、RAG和幻觉缓解等关键词，并间接关联到推理、对齐、自我反思和智能体等概念。其他关键词如MoE、量化、科学AI等未在论文中提及。

大语言模型（LLMs）在问答和检索增强生成（RAG）任务中表现出色，但它们通常隐含地假设用户查询是完整且可回答的。在实际场景中，查询往往不完整、模糊或缺失关键变量，导致模型产生过度自信或虚构的回应。

本研究探讨了信息不完整情况下的决策感知查询解析问题，即模型必须决定是直接回答、请求澄清还是选择弃答。我们发现，标准及增强的RAG系统均未能可靠地展现这种认知意识，即使在信息不足时也倾向于生成答案。

为解决这一问题，我们提出了PassiveQA框架，该框架通过监督微调将模型行为与信息充分性对齐，包含回答、询问和弃答三种决策。我们的方法整合了结构化信息状态表示、基于知识图谱的上下文，以及一个经过微调的规划器，该规划器能显式建模缺失变量和决策推理过程。

在多个问答数据集上的实验表明，在计算受限的训练条件下，经过微调的规划器在宏观F1分数和弃答召回率上均取得显著提升，同时降低了幻觉生成率。

这些结果为以下观点提供了有力的实证依据：认知决策能力必须在训练过程中习得，而非仅在推理阶段强行施加。

Large Language Models (LLMs) have achieved strong performance in question answering and retrieval-augmented generation (RAG), yet they implicitly assume that user queries are fully specified and answerable. In real-world settings, queries are often incomplete, ambiguous, or missing critical variables, leading models to produce overconfident or hallucinated responses. In this work, we study decision-aware query resolution under incomplete information, where a model must determine whether to Answer, Ask for clarification, or Abstain. We show that standard and enhanced RAG systems do not reliably exhibit such epistemic awareness, defaulting to answer generation even when information is insufficient. To address this, we propose PassiveQA, a three-action framework that aligns model behaviour with information sufficiency through supervised finetuning. Our approach integrates structured information-state representations, knowledge graph-grounded context, and a finetuned planner that explicitly models missing variables and decision reasoning. Experiments across multiple QA datasets show that the finetuned planner achieves significant improvements in macro F1 and abstention recall while reducing hallucination rates, under a compute-constrained training regime. These results provide strong empirical evidence that epistemic decision-making must be learned during training rather than imposed at inference time.

评分理由: 论文核心研究智能体（Agent）的多步工具使用（Tool Use）和推理（Reasoning），用于解决结合视觉线索和开放网络证据的地理定位任务。因此，与"LLM Agents/Autonomous Agents/Agentic Workflow"和"Tool Use/Function Calling/API Tool Use"高度相关（15分），是论文的核心。论文涉及多步推理和验证，与"Chain of Thought/CoT Reasoning/Multi-step Reasoning"和"System 2 Thinking/Slow Thinking/In-depth Reasoning"高度相关（10分）。论文的智能体工作流涉及知识检索和整合，与"Retrieval-Augmented Generation/RAG/Retrieval-Generation"有一定关联（8分）。论文背景涉及大模型驱动的智能体，因此与"Large Language Models/LLMs/Foundation Models"有一定关联（8分）。其他关键词（如MoE、SFT、量化、AI for Science等）在论文中未直接涉及，评分为0分。

深度研究智能体通过多步骤工具使用整合碎片化证据。BrowseComp为此类智能体提供了纯文本测试平台，但现有多模态基准很少同时要求弱视觉线索组合与BrowseComp风格的多跳验证。地理定位是一个天然测试场，因为其答案依赖于组合多个模糊视觉线索并通过开放网络证据进行验证。为此，我们推出GeoBrowse——一个将视觉推理与知识密集型多跳查询相结合的地理定位基准。第一层级测试碎片化视觉线索的提取与组合，第二层级通过注入长尾知识和混淆关键实体来提升查询难度。为支持评估，我们提供了智能体工作流程GATE，包含五个图像思维工具和四个知识密集型工具，并发布了基于可验证证据的专家标注逐步轨迹，用于轨迹级分析。实验表明GATE优于直接推理和开源智能体，这表明无工具、仅搜索或仅图像的设置均不充分。性能提升源于连贯的、针对特定层级的工具使用规划而非更多工具调用，因为这些规划能更可靠地抵达标注的关键证据步骤，并在整合至最终决策时产生更少错误。GeoBrowse基准与代码发布于https://github.com/ornamentt/GeoBrowse。

Deep research agents integrate fragmented evidence through multi-step tool use. BrowseComp offers a text-only testbed for such agents, but existing multimodal benchmarks rarely require both weak visual cues composition and BrowseComp-style multi-hop verification. Geolocation is a natural testbed because answers depend on combining multiple ambiguous visual cues and validating them with open-web evidence. Thus, we introduce GeoBrowse, a geolocation benchmark that combines visual reasoning with knowledge-intensive multi-hop queries. Level 1 tests extracting and composing fragmented visual cues, and Level 2 increases query difficulty by injecting long-tail knowledge and obfuscating key entities. To support evaluation, we provide an agentic workflow GATE with five think-with-image tools and four knowledge-intensive tools, and release expert-annotated stepwise traces grounded in verifiable evidence for trajectory-level analysis. Experiments show that GATE outperforms direct inference and open-source agents, indicating that no-tool, search-only or image-only setups are insufficient. Gains come from coherent, level-specific tool-use plans rather than more tool calls, as they more reliably reach annotated key evidence steps and make fewer errors when integrating into the final decision. The GeoBrowse bernchmark and codes are provided in https://github.com/ornamentt/GeoBrowse

评分理由: 论文核心研究对齐训练后语言模型中的稀疏路由机制，直接聚焦于对齐（Alignment）和可解释性（Mechanistic Interpretability），因此这两个关键词得最高分（15和10）。论文明确研究对齐训练（Post-training/SFT）的模型，涉及RLHF（作为对齐方法之一），并分析大语言模型（LLMs）的内部机制，这些关键词得10分。论文提到稀疏路由机制，与稀疏模型有一定关联，得5分。论文涉及预训练与后训练的区别，与预训练关键词有5分关联。论文研究拒绝有害内容，与事实性/真实性有5分关联。其他关键词如小模型、量化、推理加速、科学AI等与论文内容无关，得0分。

我们发现，在对齐训练的语言模型中存在一种反复出现的稀疏路由机制：一个门控注意力头读取检测到的内容，并触发下游的放大器头，从而增强拒绝信号。通过政治审查和安全拒绝作为自然实验，我们在来自6个实验室的9个模型中追踪了这一机制，所有模型均在120组提示对语料库上进行了验证。该门控头通过了必要性和充分性互换测试（p < 0.001，置换零假设），核心放大器头在自助重采样下保持稳定（Jaccard指数0.92-1.0）。三个同代缩放模型对显示，路由机制在规模扩展时分布更广（消融后效果减弱高达17倍），同时仍可通过互换检测识别。通过调节检测层信号，我们能够连续控制策略强度，从强硬拒绝到引导转向事实遵从，其路由阈值随主题而变化。该电路还揭示了意图识别与策略路由之间的结构分离：在密文编码下，门控头的路由贡献崩溃（Phi-4模型中下降78%，n=120），而模型则以解谜而非拒绝的方式响应。即使深层探针分数表明模型已开始表征有害内容，该路由机制也从未激活。这种不对称性与预训练和后训练的不同鲁棒性特征一致：预训练形成广泛的语义理解，而后训练产生的策略绑定则更狭窄，在输入变换下泛化能力较弱。

We identify a recurring sparse routing mechanism in alignment-trained language models: a gate attention head reads detected content and triggers downstream amplifier heads that boost the signal toward refusal. Using political censorship and safety refusal as natural experiments, we trace this mechanism across 9 models from 6 labs, all validated on corpora of 120 prompt pairs. The gate head passes necessity and sufficiency interchange tests (p < 0.001, permutation null), and core amplifier heads are stable under bootstrap resampling (Jaccard 0.92-1.0). Three same-generation scaling pairs show that routing distributes at scale (ablation up to 17x weaker) while remaining detectable by interchange. By modulating the detection-layer signal, we continuously control policy strength from hard refusal through steering to factual compliance, with routing thresholds that vary by topic. The circuit also reveals a structural separation between intent recognition and policy routing: under cipher encoding, the gate head’s routing contribution collapses (78% in Phi-4 at n=120) while the model responds with puzzle-solving rather than refusal. The routing mechanism never fires, even though probe scores at deeper layers indicate the model begins to represent the harmful content. This asymmetry is consistent with different robustness properties of pretraining and post-training: broad semantic understanding versus narrower policy binding that generalizes less well under input transformation.

评分理由: 论文核心研究LLMs中Chain-of-Thought推理的完整性和可靠性，提出结构化提示框架来增强推理控制、减少幻觉、提高可解释性，并应用于网络安全分析。因此，与LLMs、CoT推理、幻觉缓解、可解释AI高度相关（8-10分）；与系统2思维、自我校正、上下文学习有一定关联（5-8分）；与小型模型（提及本地部署）有部分关联（5分）；其他关键词如MoE、缩放定律、训练方法、代理、量化等未涉及（0分）。

思维链提示已被用于增强大语言模型的推理能力。然而，其在安全敏感分析任务中的可靠性仍未得到充分检验，尤其是在结构化人工评估下。模型缩放和微调等替代方法可用于帮助提升性能，但这些方法通常成本高昂、计算密集或难以审计。相比之下，提示工程为引导大语言模型推理提供了一种轻量、透明且可控的机制。本研究提出了一个结构化提示工程框架，旨在增强思维链推理的完整性，同时提升本地大语言模型部署中安全威胁与攻击检测的可靠性。该框架包含16个要素，归入四个核心维度：(1) 上下文与范围控制，(2) 证据锚定与可追溯性，(3) 推理结构与认知控制，以及(4) 安全特异性分析约束。该框架并非启发式地优化提示措辞，而是引入了显式的推理控制机制，以减轻幻觉、防止推理漂移，并增强安全敏感场景下的可解释性。以软件定义网络流量中的DDoS攻击检测为案例研究，在结构化与非结构化提示条件下对多个模型家族进行了评估。帕累托前沿分析与消融实验表明，该方法带来了持续的推理改进（在较小模型中提升高达40%）以及跨模型规模的稳定准确率增益。具有高度评分者间一致性（Cohen’s k > 0.80）的人工评估证实了其鲁棒性。研究结果确立了结构化提示作为一种有效且实用的方法，可用于实现可靠且可解释的人工智能驱动网络安全分析。

Chain-of-Thought (CoT) prompting has been used to enhance the reasoning capability of LLMs. However, its reliability in security-sensitive analytical tasks remains insufficiently examined, particularly under structured human evaluation. Alternative approaches, such as model scaling and fine-tuning can be used to help improve performance. These methods are also often costly, computationally intensive, or difficult to audit. In contrast, prompt engineering provides a lightweight, transparent, and controllable mechanism for guiding LLM reasoning. This study proposes a structured prompt engineering framework designed to strengthen CoT reasoning integrity while improving security threat and attack detection reliability in local LLM deployments. The framework includes 16 factors grouped into four core dimensions: (1) Context and Scope Control, (2) Evidence Grounding and Traceability, (3) Reasoning Structure and Cognitive Control, and (4) Security-Specific Analytical Constraints. Rather than optimizing the wording of the prompt heuristically, the framework introduces explicit reasoning controls to mitigate hallucination and prevent reasoning drift, as well as strengthening interpretability in security-sensitive contexts. Using DDoS attack detection in SDN traffic as a case study, multiple model families were evaluated under structured and unstructured prompting conditions. Pareto frontier analysis and ablation experiments demonstrate consistent reasoning improvements (up to 40% in smaller models) and stable accuracy gains across scales. Human evaluation with strong inter-rater agreement (Cohen’s k > 0.80) confirms robustness. The results establish structured prompting as an effective and practical approach for reliable and explainable AI-driven cybersecurity analysis.

评分理由: 论文核心研究LLM智能体的记忆系统，与"Large Language Models”、“LLM Agents"和"Retrieval-Augmented Generation"高度相关（10分），因为论文明确研究LLM智能体、使用RAG并改进其检索。与"Context Window Extension”、“Chain of Thought”、“System 2 Thinking"和"Hallucination Mitigation"有一定关联（5分），因为论文涉及长上下文推理、多步查询策略和事实保持（减少幻觉）。其他关键词如MoE、量化、对齐等未在摘要中提及，评为0分。

大型语言模型（LLM）智能体需要持久性记忆来维持个性化、事实连续性及长程推理能力，然而标准上下文窗口与检索增强生成（RAG）流程在多轮会话交互中会出现性能衰退。本文提出MemMachine，一个开源记忆系统，其在保持真实对话原貌的架构中整合了短期记忆、长期情景记忆与用户画像记忆，通过存储完整对话片段并减少基于LLM的有损信息提取来实现优化。MemMachine采用情境化检索技术，通过扩展核心匹配项及其周边上下文，有效提升了相关证据跨越多轮对话时的召回率。在多项基准测试中，MemMachine实现了优异的精度-效率平衡：在LoCoMo基准上使用gpt4.1-mini达到0.9169；在LongMemEvalS（ICLR 2025）的六维度消融实验中取得93.0%的准确率，其中检索阶段优化——包括检索深度调优（+4.2%）、上下文格式化（+2.0%）、搜索提示设计（+1.8%）和查询偏差校正（+1.4%）——显著优于摄入阶段改进如句子分块（+0.8%）。当配合优化提示时，GPT-5-mini较GPT-5提升2.6%，成为最具成本效益的配置方案。在与Mem0的对比中，MemMachine在同等条件下减少了约80%的输入令牌消耗。配套的检索智能体能够自适应地在直接检索、并行分解与迭代查询链策略间路由查询，在随机噪声环境下于HotpotQA-hard和WikiMultiHop数据集上分别达到93.2%和92.6%的准确率。这些结果表明，在保持情景真实性的基础上结合自适应检索层，能为个性化LLM智能体构建鲁棒且高效的长时记忆系统。

Large Language Model (LLM) agents require persistent memory to maintain personalization, factual continuity, and long-horizon reasoning, yet standard context-window and retrieval-augmented generation (RAG) pipelines degrade over multi-session interactions. We present MemMachine, an open-source memory system that integrates short-term, long-term episodic, and profile memory within a ground-truth-preserving architecture that stores entire conversational episodes and reduces lossy LLM-based extraction. MemMachine uses contextualized retrieval that expands nucleus matches with surrounding context, improving recall when relevant evidence spans multiple dialogue turns. Across benchmarks, MemMachine achieves strong accuracy-efficiency tradeoffs: on LoCoMo it reaches 0.9169 using gpt4.1-mini; on LongMemEvalS (ICLR 2025), a six-dimension ablation yields 93.0 percent accuracy, with retrieval-stage optimizations – retrieval depth tuning (+4.2 percent), context formatting (+2.0 percent), search prompt design (+1.8 percent), and query bias correction (+1.4 percent) – outperforming ingestion-stage gains such as sentence chunking (+0.8 percent). GPT-5-mini exceeds GPT-5 by 2.6 percent when paired with optimized prompts, making it the most cost-efficient setup. Compared to Mem0, MemMachine uses roughly 80 percent fewer input tokens under matched conditions. A companion Retrieval Agent adaptively routes queries among direct retrieval, parallel decomposition, or iterative chain-of-query strategies, achieving 93.2 percent on HotpotQA-hard and 92.6 percent on WikiMultiHop under randomized-noise conditions. These results show that preserving episodic ground truth while layering adaptive retrieval yields robust, efficient long-term memory for personalized LLM agents.

评分理由: 论文提出DARE框架，专门针对扩散大语言模型（dLLMs）的后训练和评估。核心内容与多个关键词高度相关：1）论文明确研究扩散大语言模型（dLLMs），是LLMs的一种新兴变体，因此与"Large Language Models"高度相关（10分）。2）论文聚焦于后训练（post-training）流程，包括监督微调（SFT）、参数高效微调（PEFT）、偏好优化（如RLHF）等，因此与"Post-training/SFT”、“PEFT/LoRA”、“RLHF/DPO”、“Instruction Tuning/Alignment"均高度相关（各10分）。其他关键词如MoE、SLMs、RAG、推理加速、科学AI应用等，论文未涉及或仅作为背景提及，因此评分为0分。

扩散大语言模型（dLLMs）正逐渐成为主流自回归模型的有力替代方案，它通过迭代去噪和并行生成机制取代了严格的序列化标记生成方式。然而，其开源生态系统在模型家族之间，尤其是在训练后流程方面仍处于碎片化状态——强化学习目标、推演实现和评估脚本通常以论文专用代码库的形式发布。这种碎片化现象延缓了研究迭代速度，增加了复现的工程负担，并导致算法间的公平比较难以实现。本文提出DARE（dLLMs Alignment and Reinforcement Executor），一个用于训练后调优与评估dLLMs的开放框架。基于verl框架和OpenCompass评估体系构建的DARE，将监督微调、参数高效微调、偏好优化以及dLLM专属强化学习统一整合到适用于掩码扩散与块扩散语言模型的共享执行栈中。在涵盖LLaDA、Dream、SDAR和LLaDA2.x等代表性模型家族的实验中，DARE提供了广泛的算法覆盖、可复现的基准评估及实际加速方案。大量实证结果表明，DARE可作为可复用的研究基础平台，用于当前及新兴dLLMs的训练后方法开发、比较与部署。

Diffusion large language models (dLLMs) are emerging as a compelling alternative to dominant autoregressive models, replacing strictly sequential token generation with iterative denoising and parallel generation dynamics. However, their open-source ecosystem remains fragmented across model families and, in particular, across post-training pipelines, where reinforcement learning objectives, rollout implementations and evaluation scripts are often released as paper-specific codebases. This fragmentation slows research iteration, raises the engineering burden of reproduction, and makes fair comparison across algorithms difficult. We present \textbf{DARE} (\textbf{d}LLMs \textbf{A}lignment and \textbf{R}einforcement \textbf{E}xecutor), an open framework for post-training and evaluating dLLMs. Built on top of verl~\cite{sheng2024hybridflow} and OpenCompass~\cite{2023opencompass}, DARE unifies supervised fine-tuning, parameter-efficient fine-tuning, preference optimization, and dLLM-specific reinforcement learning under a shared execution stack for both masked and block diffusion language models. Across representative model families including LLaDA, Dream, SDAR, and LLaDA2.x, DARE provides broad algorithmic coverage, reproducible benchmark evaluation, and practical acceleration. Extensive empirical results position that DARE serves as a reusable research substrate for developing, comparing, and deploying post-training methods for current and emerging dLLMs.

评分理由: 论文专注于低资源机器翻译，提出MERIT框架，核心涉及监督微调(SFT)和基于奖励的优化(GRPO，类似RLHF/DPO)，因此与"Supervised Fine-tuning"高度相关(10分)，与"RLHF/DPO"相关(8分)。论文使用大语言模型进行翻译，与"Large Language Models"相关(8分)。研究强调数据质量和针对性数据整理，与"Scaling Laws AND Data Quality"有一定关联(5分)。框架涉及领域适应和语义对齐，与"Pre-training/Domain Adaptation"和"Instruction Tuning/Alignment"有一定关联(各5分)。其他关键词如MoE、SLMs、RAG、CoT等未在摘要中提及，评为0分。

从中文到低资源东南亚语言的神经机器翻译（NMT）仍受到两大关键制约：洁净平行语料的极度匮乏，以及现有挖掘数据中普遍存在的噪声。这一长期存在的短缺不仅阻碍了有效的模型训练，还导致其性能与高资源语言方向存在巨大差距，使得老挝语、缅甸语、他加禄语等数百万使用者所依赖的翻译系统质量持续低下，尽管大规模多语言模型近期已取得进展。我们提出了多语言专家奖励引导调优框架（MERIT），这是一个统一的翻译框架，它将传统的以英语为中心的ALT基准转化为一个以中文为中心、涵盖五种东南亚低资源语言（LRLs）的评估体系。我们的框架将语言特定标记前缀（LTP）与监督微调（SFT）相结合，并引入了一种新颖的、由语义对齐奖励（SAR）引导的组相对策略优化（GRPO）。这些结果证实，在低资源语言到中文的翻译任务中，有针对性的数据策展和奖励引导的优化策略，其效果远超单纯的模型规模扩展。

Neural machine translation (NMT) from Chinese to low-resource Southeast Asian languages remains severely constrained by the extreme scarcity of clean parallel corpora and the pervasive noise in existing mined data. This chronic shortage not only impedes effective model training but also sustains a large performance gap with high-resource directions, leaving millions of speakers of languages such as Lao, Burmese, and Tagalog with persistently low-quality translation systems despite recent advances in large multilingual models. We introduce \textbf{M}ultilingual \textbf{E}xpert-\textbf{R}eward \textbf{I}nformed \textbf{T}uning (\textbf{MERIT}), a unified translation framework that transforms the traditional English-centric ALT benchmark into a Chinese-centric evaluation suite for five Southeast Asian low-resource languages (LRLs). Our framework combines language-specific token prefixing (LTP) with supervised fine-tuning (SFT) and a novel group relative policy optimization (GRPO) guided by the semantic alignment reward (SAR). These results confirm that, in LRL{\textrightarrow}Chinese translation, targeted data curation and reward-guided optimization dramatically outperform mere model scaling.

评分理由: 论文核心研究大语言模型（LLMs）的知识蒸馏（KD）方法，提出SODA（半在线蒸馏对齐）框架，用于将大型教师模型的知识高效迁移到小型学生模型（SLMs）。因此，与"Large Language Models”、“Small Language Models”、“Post-training”（蒸馏属于后训练技术）和"Alignment”（论文强调分布对齐）高度相关（10分）。其他关键词如MoE、Scaling Laws、RAG、推理加速等，论文未涉及，给0分。

大型语言模型的黑箱知识蒸馏面临严格的权衡困境。简单的离策略方法（如序列级知识蒸馏）难以纠正学生模型固有的错误。完全同策略方法（如生成对抗蒸馏）通过对抗训练解决了这一问题，但引入了众所周知的训练不稳定性和极高的计算开销。为应对这一困境，我们提出SODA（基于对齐的半同策略蒸馏），这是一种高效替代方案，其动机源于前沿教师模型与更小规模基础模型之间固有的能力差距。由于紧凑型学生模型的自然零样本响应几乎严格劣于强大教师模型的目标输出，我们可以通过将教师模型的最优响应与学生模型一次性静态输出快照进行配对，构建出高效的对比信号。这表明，让小型学生模型接触其自身静态的次优行为已足以实现高质量的分布对齐，从而无需昂贵的动态展开过程和脆弱的对抗平衡。在四种紧凑型Qwen2.5和Llama-3模型上的广泛评估验证了这种半同策略范式。SODA在16项基准测试结果中的15项上达到或超越了现有最优方法。更重要的是，它在实现更优蒸馏质量的同时，训练速度提升10倍，峰值GPU内存消耗降低27%，并完全消除了对抗训练的不稳定性。

Black-box knowledge distillation for large language models presents a strict trade-off. Simple off-policy methods (e.g., sequence-level knowledge distillation) struggle to correct the student’s inherent errors. Fully on-policy methods (e.g., Generative Adversarial Distillation) solve this via adversarial training but introduce well-known training instability and crippling computational overhead. To address this dilemma, we propose SODA (Semi On-policy Distillation with Alignment), a highly efficient alternative motivated by the inherent capability gap between frontier teachers and much smaller base models. Because a compact student model’s natural, zero-shot responses are almost strictly inferior to the powerful teacher’s targets, we can construct a highly effective contrastive signal simply by pairing the teacher’s optimal response with a one-time static snapshot of the student’s outputs. This demonstrates that exposing the small student to its own static inferior behaviors is sufficient for high-quality distribution alignment, eliminating the need for costly dynamic rollouts and fragile adversarial balancing. Extensive evaluations across four compact Qwen2.5 and Llama-3 models validate this semi on-policy paradigm. SODA matches or outperforms the state-of-the-art methods on 15 out of 16 benchmark results. More importantly, it achieves this superior distillation quality while training 10 times faster, consuming 27% less peak GPU memory, and completely eliminating adversarial instability.

评分理由: 该论文核心研究LLM驱动的智能体（Agent）在多步工具调用环境中的安全漏洞（间接提示注入攻击）及防御策略，与"LLM Agents/Autonomous Agents/Agentic Workflow”、“Tool Use/Function Calling/API Tool Use"和"Multi-agent Systems/Agent Coordination"高度相关（10分），因为这些是论文研究的核心对象和场景。论文也基于多种LLM骨干进行评估，因此与"Large Language Models/LLMs/Foundation Models"高度相关（10分）。论文未涉及其他关键词所代表的具体技术（如MoE、量化、推理加速、科学AI应用等），因此这些关键词评分为0分。

开源框架的快速部署显著推动了现代多智能体系统的发展。然而，扩展的行动空间——包括不受控的权限暴露和隐藏的系统间交互——带来了严峻的安全挑战。具体而言，间接提示注入攻击通过将恶意指令隐藏在第三方内容中，可在智能体正常操作期间触发数据窃取等未授权行为。当前的安全评估主要依赖孤立的单轮基准测试，而这些智能体在复杂动态环境中的系统性脆弱性仍未得到充分探索。为弥补这一空白，我们系统评估了针对四种复杂间接提示注入攻击向量的六种防御策略，覆盖九种大语言模型基座。关键的是，我们的评估完全在动态多步骤工具调用环境中进行，以捕捉现代自主智能体的真实攻击面。超越二元的成功率指标，我们的多维分析揭示出显著的脆弱性：高级注入攻击成功绕过几乎所有基线防御，部分表层缓解措施甚至会产生适得其反的副作用。此外，尽管智能体几乎瞬时执行恶意指令，其内部状态却表现出异常高的决策熵。基于这种潜在犹豫现象，我们探索将表征工程作为鲁棒的检测策略。通过提取工具输入位置对应的隐藏状态，我们发现基于表征工程的断路器能在智能体执行前成功识别并拦截未授权操作，在不同大语言模型基座上均实现了高检测准确率。本研究揭示了当前间接提示注入防御的局限性，并为构建具有韧性的多智能体架构提供了高度实用的范式。

The rapid deployment of open-source frameworks has significantly advanced the development of modern multi-agent systems. However, expanded action spaces, including uncontrolled privilege exposure and hidden inter-system interactions, pose severe security challenges. Specifically, Indirect Prompt Injections (IPI), which conceal malicious instructions within third-party content, can trigger unauthorized actions such as data exfiltration during normal operations. While current security evaluations predominantly rely on isolated single-turn benchmarks, the systemic vulnerabilities of these agents within complex dynamic environments remain critically underexplored. To bridge this gap, we systematically evaluate six defense strategies against four sophisticated IPI attack vectors across nine LLM backbones. Crucially, we conduct our evaluation entirely within dynamic multi-step tool-calling environments to capture the true attack surface of modern autonomous agents. Moving beyond binary success rates, our multidimensional analysis reveals a pronounced fragility. Advanced injections successfully bypass nearly all baseline defenses, and some surface-level mitigations even produce counterproductive side effects. Furthermore, while agents execute malicious instructions almost instantaneously, their internal states exhibit abnormally high decision entropy. Motivated by this latent hesitation, we investigate Representation Engineering (RepE) as a robust detection strategy. By extracting hidden states at the tool-input position, we revealed that the RepE-based circuit breaker successfully identifies and intercepts unauthorized actions before the agent commits to them, achieving high detection accuracy across diverse LLM backbones. This study exposes the limitations of current IPI defenses and provides a highly practical paradigm for building resilient multi-agent architectures.

评分理由: 论文聚焦于医学影像（胸片）的时间序列分析，属于AI for Science（生物信息学）应用，得10分；涉及vision-language pretraining（医学基础模型），与LLMs/Foundation Models相关但非核心，得8分；明确包含pretraining和fine-tuning（SFT）技术，各得10分；其他关键词如MoE、SLMs、Scaling Laws、Alignment、RLHF、RAG、Reasoning、Agents、Compression等均未涉及，得0分。

视觉-语言预训练的最新进展催生了强大的医学基础模型，但多数模型仍孤立分析放射影像，忽视了对比既往与当前影像以评估时序变化这一关键临床任务。对于胸部X光片（CXRs）而言，捕捉时序变化至关重要，因为放射科医生不仅需要评估影像表现的静态特征，还必须追踪其随时间的演变过程。我们提出TILA（时序反转感知学习与对齐框架），这是一种简洁而有效的框架，其使用时序反转（即反转图像对的顺序）作为监督信号，以增强现有时序视觉-语言模型对方向性变化的敏感性。TILA在预训练、微调和推理阶段整合了反转感知目标，通过显式学习时序顺序来补充传统的外观建模方法。我们还提出了一套统一的评估协议，用于衡量模型在时序反转下的顺序敏感性与一致性，并构建了MS-CXR-Tretrieval检索评估集——该评估集基于通用构建协议，可应用于任何时序CXR数据集。在公开数据集和真实医院队列上的实验表明，当应用于多种现有架构时，TILA能持续提升疾病进展分类的准确性和时序嵌入的对齐效果。

Recent advances in vision–language pretraining have enabled strong medical foundation models, yet most analyze radiographs in isolation, overlooking the key clinical task of comparing prior and current images to assess interval change. For chest radiographs (CXRs), capturing interval change is essential, as radiologists must evaluate not only the static appearance of findings but also how they evolve over time. We introduce TILA (Temporal Inversion-aware Learning and Alignment), a simple yet effective framework that uses temporal inversion, reversing image pairs, as a supervisory signal to enhance the sensitivity of existing temporal vision-language models to directional change. TILA integrates inversion-aware objectives across pretraining, fine-tuning, and inference, complementing conventional appearance modeling with explicit learning of temporal order. We also propose a unified evaluation protocol to assess order sensitivity and consistency under temporal inversion, and introduce MS-CXR-Tretrieval, a retrieval evaluation set constructed through a general protocol that can be applied to any temporal CXR dataset. Experiments on public datasets and real-world hospital cohorts demonstrate that TILA consistently improves progression classification and temporal embedding alignment when applied to multiple existing architectures.

评分理由: 论文提出TimeRAN，一个用于无线接入网（RAN）时间序列建模的统一多任务学习框架，其核心是一个轻量级的时间序列基础模型，并进行了大规模预训练和有限监督下的高效适应。因此，它与"Foundation Models”（8分）高度相关，因为论文明确提出了一个基础模型；与"Pre-training"（10分）高度相关，因为论文进行了大规模预训练并开源了数据集；与"Supervised Fine-tuning"（8分）相关，因为提到了任务特定的微调；与"AI for Science"（5分）有一定关联，因为RAN优化属于AI在特定领域（通信网络）的应用。其他关键词（如LLM、MoE、RLHF等）主要针对自然语言处理或特定技术，与论文的时间序列模型和RAN应用领域无关，故得0分。

无线接入网（RAN）正演变为一种可编程、解耦的基础设施，日益依赖原生人工智能算法进行优化与闭环控制。然而，当前的RAN智能系统仍主要由针对单一功能设计的任务专用模型构成，导致模型碎片化、任务间知识共享有限、泛化能力差以及系统复杂性增加。为应对这些局限，本文提出TimeRAN——一个面向RAN时序建模的统一多任务学习框架。TimeRAN采用轻量级时序基础模型配合少量任务专用头部，通过学习可迁移的表征，能够在有限监督下高效适配多种任务。为实现大规模预训练，我们进一步整理并开源了TimeRAN DataPile，这是迄今为止最大的RAN分析时序数据集，涵盖多样化遥测源、协议层和部署场景，包含超过35.5万条时间序列和5.6亿个测量点。我们在涵盖异常检测、分类、预测与填补的完整RAN分析任务集上评估TimeRAN，结果表明其仅需极少或无需任务特定微调即可达到先进性能。最后，我们将TimeRAN集成至概念验证型5G测试平台，证明其在实际场景中能以有限资源需求高效运行。

The Radio Access Network (RAN) is evolving into a programmable and disaggregated infrastructure that increasingly relies on AI-native algorithms for optimization and closed-loop control. However, current RAN intelligence is still largely built from task-specific models tailored to individual functions, resulting in model fragmentation, limited knowledge sharing across tasks, poor generalization, and increased system complexity. To address these limitations, we introduce TimeRAN, a unified multi-task learning framework for time-series modeling in the RAN. TimeRAN leverages a lightweight time-series foundation model with few task-specific heads to learn transferable representations that can be efficiently adapted across diverse tasks with limited supervision. To enable large-scale pretraining, we further curate and open-source TimeRAN DataPile, the largest time-series corpus for RAN analytics to date, comprising over 355K time series and 0.56B measurements across diverse telemetry sources, protocol layers, and deployment scenarios. We evaluate TimeRAN across a comprehensive set of RAN analytics tasks, including anomaly detection, classification, forecasting, and imputation, and show that it achieves state-of-the-art performance with minimal or no task-specific fine-tuning. Finally, we integrate TimeRAN into a proof-of-concept 5G testbed and demonstrate that it operates efficiently with limited resource requirements in real-world scenarios.

评分理由: 论文提出AI Trust OS治理框架，核心关注LLM、RAG和多智能体工作流在企业环境中的治理、可观测性和合规性问题。摘要明确提到"large language models, retrieval-augmented generation pipelines, and multi-agent AI workflows"，因此这三个关键词高度相关（10分）。论文属于大模型应用领域（企业治理），但未涉及其他关键词的具体技术原理或创新，因此其他关键词评分为0分。

大型语言模型、检索增强生成流程以及多智能体人工智能工作流的加速普及已引发结构性治理危机。组织无法监管其不可见之物，而为确定性网络应用构建的现有合规方法论缺乏相应机制，用以发现或持续验证那些在工程团队间自发产生、缺乏正式监管的人工智能系统。这导致监管机构要求的人工智能治理成熟度证明与组织实际可提供的证据之间出现日益扩大的信任鸿沟。本文提出“AI信任操作系统”——一种面向持续自主人工智能可观测性与零信任合规的治理架构。该系统将合规性重新构想为一个全天候运行、由遥测数据驱动的操作层：通过可观测性信号发现人工智能系统，借助自动化探针收集控制断言，并持续合成信任凭证。该框架基于四大原则：主动发现、基于遥测证据而非人工证明、持续状态监测而非时点审计、架构支撑的实证而非政策文件的信任。其通过零信任遥测边界运作，瞬时只读探针在此验证结构性元数据，无需接入源代码或载荷级个人身份信息。人工智能可观测性提取代理会扫描LangSmith与Datadog的LLM遥测数据，自动注册未记录的人工智能系统，从而将治理模式从组织自我报告转向实证化的机器观测。通过对ISO 42001、欧盟《人工智能法案》、SOC 2、GDPR及HIPAA等标准的评估，本文论证了遥测优先的人工智能治理模式代表了企业信任生成与验证方式的根本性架构变革。

The accelerating adoption of large language models, retrieval-augmented generation pipelines, and multi-agent AI workflows has created a structural governance crisis. Organizations cannot govern what they cannot see, and existing compliance methodologies built for deterministic web applications provide no mechanism for discovering or continuously validating AI systems that emerge across engineering teams without formal oversight. The result is a widening trust gap between what regulators demand as proof of AI governance maturity and what organizations can demonstrate. This paper proposes AI Trust OS, a governance architecture for continuous, autonomous AI observability and zero-trust compliance. AI Trust OS reconceptualizes compliance as an always-on, telemetry-driven operating layer in which AI systems are discovered through observability signals, control assertions are collected by automated probes, and trust artifacts are synthesized continuously. The framework rests on four principles: proactive discovery, telemetry evidence over manual attestation, continuous posture over point-in-time audit, and architecture-backed proof over policy-document trust. The framework operates through a zero-trust telemetry boundary in which ephemeral read-only probes validate structural metadata without ingressing source code or payload-level PII. An AI Observability Extractor Agent scans LangSmith and Datadog LLM telemetry, automatically registering undocumented AI systems and shifting governance from organizational self-report to empirical machine observation. Evaluated across ISO 42001, the EU AI Act, SOC 2, GDPR, and HIPAA, the paper argues that telemetry-first AI governance represents a categorical architectural shift in how enterprise trust is produced and demonstrated.

评分理由: 论文核心是提出一个LLM驱动的多智能体仿真框架（LLM-MAS）用于优化服务运营，直接高度相关于"Large Language Models"（LLMs是核心工具）、“LLM Agents”（使用LLM作为智能体）和"Multi-agent Systems"（多智能体仿真框架）。其他关键词如MoE、SFT、RAG、推理加速等涉及具体技术细节或应用领域，论文未涉及，故评分为0。

服务系统性能取决于参与者对设计选择的响应方式，但由于人类行为的复杂性，对这些响应进行建模十分困难。本文提出一种基于大语言模型的多智能体仿真框架，用于优化服务运营。我们将该问题构建为具有决策依赖不确定性的随机优化问题：设计选择被嵌入提示词中，并塑造了基于大语言模型的智能体交互所产生结果的分布。通过将关键数值信息嵌入提示词并从大语言模型生成的文本中提取该信息，我们将这种不确定性建模为一个受控马尔可夫链。我们开发了一种轨迹上学习算法，该算法在单次仿真运行中，同时构建零阶梯度估计并更新设计参数，以优化稳态性能。我们还引入了方差缩减技术。在一个可持续供应链的应用中，我们的方法在性能上超越了多种基准方案，包括黑箱优化、使用大语言模型作为数值求解器或作为角色扮演系统设计者的方法。一项基于真实行为数据的最优竞赛设计案例研究表明，LLM-MAS既可作为已知设计方案的性价比评估工具，也可作为一种探索性工具，能够发现被传统方法忽视的优秀设计方案。

Service system performance depends on how participants respond to design choices, but modeling these responses is hard due to the complexity of human behavior. We introduce an LLM-powered multi-agent simulation (LLM-MAS) framework for optimizing service operations. We pose the problem as stochastic optimization with decision-dependent uncertainty: design choices are embedded in prompts and shape the distribution of outcomes from interacting LLM-powered agents. By embedding key numerical information in prompts and extracting it from LLM-generated text, we model this uncertainty as a controlled Markov chain. We develop an on-trajectory learning algorithm that, on a single simulation run, simultaneously constructs zeroth-order gradient estimates and updates design parameters to optimize steady-state performance. We also incorporate variance reduction techniques. In a sustainable supply chain application, our method outperforms benchmarks, including blackbox optimization and using LLMs as numerical solvers or as role-playing system designers. A case study on optimal contest design with real behavioral data shows that LLM-MAS is both as a cost-effective evaluator of known designs and an exploratory tool that can uncover strong designs overlooked by traditional approaches.

评分理由: 论文核心研究如何通过轻量级微调方法（\policy）训练小型语言模型（SLMs）成为有效的搜索代理，以解决SLMs在复杂多跳推理任务中搜索工具调用不足和幻觉倾向的问题。因此，与"Small Language Models (SLMs)"、“LLM Agents”、“Tool Use”、“Retrieval-Augmented Generation (RAG)“高度相关（10分）。论文也涉及从大型语言模型（LLMs）蒸馏代理行为，与"Large Language Models (LLMs)“高度相关（10分）。论文提出的微调方法属于监督微调（SFT）范畴，与"Post-training/SFT"相关（8分）。研究任务涉及多跳推理，与"Chain of Thought/CoT Reasoning"相关（8分）。目标之一是减少幻觉，与"Hallucination Mitigation"相关（8分）。其他关键词如MoE、Scaling Laws、RLHF等未在论文中提及或仅边缘相关，给予0分。

配备搜索工具的智能体已成为解决知识密集型任务的有效方案。尽管大语言模型展现出强大的推理能力，但其高昂的计算成本限制了搜索智能体的实际部署。因此，近期研究聚焦于将大语言模型的智能体行为蒸馏至小语言模型中。通过对复杂多跳推理任务的综合评估，我们发现尽管小语言模型具备的参数知识较少，但其调用搜索工具的频次更低，且更容易产生幻觉。为解决这一问题，我们提出\policy方法——一种轻量级微调策略，通过显式训练小语言模型，使其能够基于检索证据进行可靠检索并生成答案。相较于从大语言模型进行智能体蒸馏的方法，我们的方案在Bamboogle数据集上提升了17.3分，在HotpotQA数据集上提升了15.3分，在各基准测试中均达到大语言模型级别的性能。进一步分析表明，小语言模型中的自适应搜索策略往往会导致性能下降，这凸显了稳定搜索行为对可靠推理的必要性。

Agents equipped with search tools have emerged as effective solutions for knowledge-intensive tasks. While Large Language Models (LLMs) exhibit strong reasoning capabilities, their high computational cost limits practical deployment for search agents. Consequently, recent work has focused on distilling agentic behaviors from LLMs into Small Language Models (SLMs). Through comprehensive evaluation on complex multi-hop reasoning tasks, we find that despite possessing less parametric knowledge, SLMs invoke search tools less frequently and are more prone to hallucinations. To address this issue, we propose \policy, a lightweight fine-tuning approach that explicitly trains SLMs to reliably retrieve and generate answers grounded in retrieved evidence. Compared to agent distillation from LLMs, our approach improves performance by 17.3 scores on Bamboogle and 15.3 scores on HotpotQA, achieving LLM-level results across benchmarks. Our further analysis reveals that adaptive search strategies in SLMs often degrade performance, highlighting the necessity of consistent search behavior for reliable reasoning.

评分理由: 论文核心研究LLMs在问答任务中的认知校准问题，通过监督微调（SFT）实现三动作框架（回答、询问、弃权），直接涉及LLMs、SFT、RAG和幻觉缓解等关键词，并间接关联到推理、对齐、自我反思和智能体等概念。其他关键词如MoE、量化、科学AI等未在论文中提及。

大语言模型（LLMs）在问答和检索增强生成（RAG）任务中表现出色，但它们通常隐含地假设用户查询是完整且可回答的。在实际场景中，查询往往不完整、模糊或缺失关键变量，导致模型产生过度自信或虚构的回应。

本研究探讨了信息不完整情况下的决策感知查询解析问题，即模型必须决定是直接回答、请求澄清还是选择弃答。我们发现，标准及增强的RAG系统均未能可靠地展现这种认知意识，即使在信息不足时也倾向于生成答案。

为解决这一问题，我们提出了PassiveQA框架，该框架通过监督微调将模型行为与信息充分性对齐，包含回答、询问和弃答三种决策。我们的方法整合了结构化信息状态表示、基于知识图谱的上下文，以及一个经过微调的规划器，该规划器能显式建模缺失变量和决策推理过程。

在多个问答数据集上的实验表明，在计算受限的训练条件下，经过微调的规划器在宏观F1分数和弃答召回率上均取得显著提升，同时降低了幻觉生成率。

这些结果为以下观点提供了有力的实证依据：认知决策能力必须在训练过程中习得，而非仅在推理阶段强行施加。

Large Language Models (LLMs) have achieved strong performance in question answering and retrieval-augmented generation (RAG), yet they implicitly assume that user queries are fully specified and answerable. In real-world settings, queries are often incomplete, ambiguous, or missing critical variables, leading models to produce overconfident or hallucinated responses. In this work, we study decision-aware query resolution under incomplete information, where a model must determine whether to Answer, Ask for clarification, or Abstain. We show that standard and enhanced RAG systems do not reliably exhibit such epistemic awareness, defaulting to answer generation even when information is insufficient. To address this, we propose PassiveQA, a three-action framework that aligns model behaviour with information sufficiency through supervised finetuning. Our approach integrates structured information-state representations, knowledge graph-grounded context, and a finetuned planner that explicitly models missing variables and decision reasoning. Experiments across multiple QA datasets show that the finetuned planner achieves significant improvements in macro F1 and abstention recall while reducing hallucination rates, under a compute-constrained training regime. These results provide strong empirical evidence that epistemic decision-making must be learned during training rather than imposed at inference time.

评分理由: 论文核心研究智能体（Agent）的多步工具使用（Tool Use）和推理（Reasoning），用于解决结合视觉线索和开放网络证据的地理定位任务。因此，与"LLM Agents/Autonomous Agents/Agentic Workflow"和"Tool Use/Function Calling/API Tool Use"高度相关（15分），是论文的核心。论文涉及多步推理和验证，与"Chain of Thought/CoT Reasoning/Multi-step Reasoning"和"System 2 Thinking/Slow Thinking/In-depth Reasoning"高度相关（10分）。论文的智能体工作流涉及知识检索和整合，与"Retrieval-Augmented Generation/RAG/Retrieval-Generation"有一定关联（8分）。论文背景涉及大模型驱动的智能体，因此与"Large Language Models/LLMs/Foundation Models"有一定关联（8分）。其他关键词（如MoE、SFT、量化、AI for Science等）在论文中未直接涉及，评分为0分。

深度研究智能体通过多步骤工具使用整合碎片化证据。BrowseComp为此类智能体提供了纯文本测试平台，但现有多模态基准很少同时要求弱视觉线索组合与BrowseComp风格的多跳验证。地理定位是一个天然测试场，因为其答案依赖于组合多个模糊视觉线索并通过开放网络证据进行验证。为此，我们推出GeoBrowse——一个将视觉推理与知识密集型多跳查询相结合的地理定位基准。第一层级测试碎片化视觉线索的提取与组合，第二层级通过注入长尾知识和混淆关键实体来提升查询难度。为支持评估，我们提供了智能体工作流程GATE，包含五个图像思维工具和四个知识密集型工具，并发布了基于可验证证据的专家标注逐步轨迹，用于轨迹级分析。实验表明GATE优于直接推理和开源智能体，这表明无工具、仅搜索或仅图像的设置均不充分。性能提升源于连贯的、针对特定层级的工具使用规划而非更多工具调用，因为这些规划能更可靠地抵达标注的关键证据步骤，并在整合至最终决策时产生更少错误。GeoBrowse基准与代码发布于https://github.com/ornamentt/GeoBrowse。

Deep research agents integrate fragmented evidence through multi-step tool use. BrowseComp offers a text-only testbed for such agents, but existing multimodal benchmarks rarely require both weak visual cues composition and BrowseComp-style multi-hop verification. Geolocation is a natural testbed because answers depend on combining multiple ambiguous visual cues and validating them with open-web evidence. Thus, we introduce GeoBrowse, a geolocation benchmark that combines visual reasoning with knowledge-intensive multi-hop queries. Level 1 tests extracting and composing fragmented visual cues, and Level 2 increases query difficulty by injecting long-tail knowledge and obfuscating key entities. To support evaluation, we provide an agentic workflow GATE with five think-with-image tools and four knowledge-intensive tools, and release expert-annotated stepwise traces grounded in verifiable evidence for trajectory-level analysis. Experiments show that GATE outperforms direct inference and open-source agents, indicating that no-tool, search-only or image-only setups are insufficient. Gains come from coherent, level-specific tool-use plans rather than more tool calls, as they more reliably reach annotated key evidence steps and make fewer errors when integrating into the final decision. The GeoBrowse bernchmark and codes are provided in https://github.com/ornamentt/GeoBrowse

评分理由: 论文核心研究对齐训练后语言模型中的稀疏路由机制，直接聚焦于对齐（Alignment）和可解释性（Mechanistic Interpretability），因此这两个关键词得最高分（15和10）。论文明确研究对齐训练（Post-training/SFT）的模型，涉及RLHF（作为对齐方法之一），并分析大语言模型（LLMs）的内部机制，这些关键词得10分。论文提到稀疏路由机制，与稀疏模型有一定关联，得5分。论文涉及预训练与后训练的区别，与预训练关键词有5分关联。论文研究拒绝有害内容，与事实性/真实性有5分关联。其他关键词如小模型、量化、推理加速、科学AI等与论文内容无关，得0分。

我们发现，在对齐训练的语言模型中存在一种反复出现的稀疏路由机制：一个门控注意力头读取检测到的内容，并触发下游的放大器头，从而增强拒绝信号。通过政治审查和安全拒绝作为自然实验，我们在来自6个实验室的9个模型中追踪了这一机制，所有模型均在120组提示对语料库上进行了验证。该门控头通过了必要性和充分性互换测试（p < 0.001，置换零假设），核心放大器头在自助重采样下保持稳定（Jaccard指数0.92-1.0）。三个同代缩放模型对显示，路由机制在规模扩展时分布更广（消融后效果减弱高达17倍），同时仍可通过互换检测识别。通过调节检测层信号，我们能够连续控制策略强度，从强硬拒绝到引导转向事实遵从，其路由阈值随主题而变化。该电路还揭示了意图识别与策略路由之间的结构分离：在密文编码下，门控头的路由贡献崩溃（Phi-4模型中下降78%，n=120），而模型则以解谜而非拒绝的方式响应。即使深层探针分数表明模型已开始表征有害内容，该路由机制也从未激活。这种不对称性与预训练和后训练的不同鲁棒性特征一致：预训练形成广泛的语义理解，而后训练产生的策略绑定则更狭窄，在输入变换下泛化能力较弱。

We identify a recurring sparse routing mechanism in alignment-trained language models: a gate attention head reads detected content and triggers downstream amplifier heads that boost the signal toward refusal. Using political censorship and safety refusal as natural experiments, we trace this mechanism across 9 models from 6 labs, all validated on corpora of 120 prompt pairs. The gate head passes necessity and sufficiency interchange tests (p < 0.001, permutation null), and core amplifier heads are stable under bootstrap resampling (Jaccard 0.92-1.0). Three same-generation scaling pairs show that routing distributes at scale (ablation up to 17x weaker) while remaining detectable by interchange. By modulating the detection-layer signal, we continuously control policy strength from hard refusal through steering to factual compliance, with routing thresholds that vary by topic. The circuit also reveals a structural separation between intent recognition and policy routing: under cipher encoding, the gate head’s routing contribution collapses (78% in Phi-4 at n=120) while the model responds with puzzle-solving rather than refusal. The routing mechanism never fires, even though probe scores at deeper layers indicate the model begins to represent the harmful content. This asymmetry is consistent with different robustness properties of pretraining and post-training: broad semantic understanding versus narrower policy binding that generalizes less well under input transformation.

评分理由: 论文核心研究LLMs中Chain-of-Thought推理的完整性和可靠性，提出结构化提示框架来增强推理控制、减少幻觉、提高可解释性，并应用于网络安全分析。因此，与LLMs、CoT推理、幻觉缓解、可解释AI高度相关（8-10分）；与系统2思维、自我校正、上下文学习有一定关联（5-8分）；与小型模型（提及本地部署）有部分关联（5分）；其他关键词如MoE、缩放定律、训练方法、代理、量化等未涉及（0分）。

思维链提示已被用于增强大语言模型的推理能力。然而，其在安全敏感分析任务中的可靠性仍未得到充分检验，尤其是在结构化人工评估下。模型缩放和微调等替代方法可用于帮助提升性能，但这些方法通常成本高昂、计算密集或难以审计。相比之下，提示工程为引导大语言模型推理提供了一种轻量、透明且可控的机制。本研究提出了一个结构化提示工程框架，旨在增强思维链推理的完整性，同时提升本地大语言模型部署中安全威胁与攻击检测的可靠性。该框架包含16个要素，归入四个核心维度：(1) 上下文与范围控制，(2) 证据锚定与可追溯性，(3) 推理结构与认知控制，以及(4) 安全特异性分析约束。该框架并非启发式地优化提示措辞，而是引入了显式的推理控制机制，以减轻幻觉、防止推理漂移，并增强安全敏感场景下的可解释性。以软件定义网络流量中的DDoS攻击检测为案例研究，在结构化与非结构化提示条件下对多个模型家族进行了评估。帕累托前沿分析与消融实验表明，该方法带来了持续的推理改进（在较小模型中提升高达40%）以及跨模型规模的稳定准确率增益。具有高度评分者间一致性（Cohen’s k > 0.80）的人工评估证实了其鲁棒性。研究结果确立了结构化提示作为一种有效且实用的方法，可用于实现可靠且可解释的人工智能驱动网络安全分析。

Chain-of-Thought (CoT) prompting has been used to enhance the reasoning capability of LLMs. However, its reliability in security-sensitive analytical tasks remains insufficiently examined, particularly under structured human evaluation. Alternative approaches, such as model scaling and fine-tuning can be used to help improve performance. These methods are also often costly, computationally intensive, or difficult to audit. In contrast, prompt engineering provides a lightweight, transparent, and controllable mechanism for guiding LLM reasoning. This study proposes a structured prompt engineering framework designed to strengthen CoT reasoning integrity while improving security threat and attack detection reliability in local LLM deployments. The framework includes 16 factors grouped into four core dimensions: (1) Context and Scope Control, (2) Evidence Grounding and Traceability, (3) Reasoning Structure and Cognitive Control, and (4) Security-Specific Analytical Constraints. Rather than optimizing the wording of the prompt heuristically, the framework introduces explicit reasoning controls to mitigate hallucination and prevent reasoning drift, as well as strengthening interpretability in security-sensitive contexts. Using DDoS attack detection in SDN traffic as a case study, multiple model families were evaluated under structured and unstructured prompting conditions. Pareto frontier analysis and ablation experiments demonstrate consistent reasoning improvements (up to 40% in smaller models) and stable accuracy gains across scales. Human evaluation with strong inter-rater agreement (Cohen’s k > 0.80) confirms robustness. The results establish structured prompting as an effective and practical approach for reliable and explainable AI-driven cybersecurity analysis.

评分理由: 论文核心研究LLM智能体的记忆系统，与"Large Language Models”、“LLM Agents"和"Retrieval-Augmented Generation"高度相关（10分），因为论文明确研究LLM智能体、使用RAG并改进其检索。与"Context Window Extension”、“Chain of Thought”、“System 2 Thinking"和"Hallucination Mitigation"有一定关联（5分），因为论文涉及长上下文推理、多步查询策略和事实保持（减少幻觉）。其他关键词如MoE、量化、对齐等未在摘要中提及，评为0分。

大型语言模型（LLM）智能体需要持久性记忆来维持个性化、事实连续性及长程推理能力，然而标准上下文窗口与检索增强生成（RAG）流程在多轮会话交互中会出现性能衰退。本文提出MemMachine，一个开源记忆系统，其在保持真实对话原貌的架构中整合了短期记忆、长期情景记忆与用户画像记忆，通过存储完整对话片段并减少基于LLM的有损信息提取来实现优化。MemMachine采用情境化检索技术，通过扩展核心匹配项及其周边上下文，有效提升了相关证据跨越多轮对话时的召回率。在多项基准测试中，MemMachine实现了优异的精度-效率平衡：在LoCoMo基准上使用gpt4.1-mini达到0.9169；在LongMemEvalS（ICLR 2025）的六维度消融实验中取得93.0%的准确率，其中检索阶段优化——包括检索深度调优（+4.2%）、上下文格式化（+2.0%）、搜索提示设计（+1.8%）和查询偏差校正（+1.4%）——显著优于摄入阶段改进如句子分块（+0.8%）。当配合优化提示时，GPT-5-mini较GPT-5提升2.6%，成为最具成本效益的配置方案。在与Mem0的对比中，MemMachine在同等条件下减少了约80%的输入令牌消耗。配套的检索智能体能够自适应地在直接检索、并行分解与迭代查询链策略间路由查询，在随机噪声环境下于HotpotQA-hard和WikiMultiHop数据集上分别达到93.2%和92.6%的准确率。这些结果表明，在保持情景真实性的基础上结合自适应检索层，能为个性化LLM智能体构建鲁棒且高效的长时记忆系统。

Large Language Model (LLM) agents require persistent memory to maintain personalization, factual continuity, and long-horizon reasoning, yet standard context-window and retrieval-augmented generation (RAG) pipelines degrade over multi-session interactions. We present MemMachine, an open-source memory system that integrates short-term, long-term episodic, and profile memory within a ground-truth-preserving architecture that stores entire conversational episodes and reduces lossy LLM-based extraction. MemMachine uses contextualized retrieval that expands nucleus matches with surrounding context, improving recall when relevant evidence spans multiple dialogue turns. Across benchmarks, MemMachine achieves strong accuracy-efficiency tradeoffs: on LoCoMo it reaches 0.9169 using gpt4.1-mini; on LongMemEvalS (ICLR 2025), a six-dimension ablation yields 93.0 percent accuracy, with retrieval-stage optimizations – retrieval depth tuning (+4.2 percent), context formatting (+2.0 percent), search prompt design (+1.8 percent), and query bias correction (+1.4 percent) – outperforming ingestion-stage gains such as sentence chunking (+0.8 percent). GPT-5-mini exceeds GPT-5 by 2.6 percent when paired with optimized prompts, making it the most cost-efficient setup. Compared to Mem0, MemMachine uses roughly 80 percent fewer input tokens under matched conditions. A companion Retrieval Agent adaptively routes queries among direct retrieval, parallel decomposition, or iterative chain-of-query strategies, achieving 93.2 percent on HotpotQA-hard and 92.6 percent on WikiMultiHop under randomized-noise conditions. These results show that preserving episodic ground truth while layering adaptive retrieval yields robust, efficient long-term memory for personalized LLM agents.

评分理由: 论文提出DARE框架，专门针对扩散大语言模型（dLLMs）的后训练和评估。核心内容与多个关键词高度相关：1）论文明确研究扩散大语言模型（dLLMs），是LLMs的一种新兴变体，因此与"Large Language Models"高度相关（10分）。2）论文聚焦于后训练（post-training）流程，包括监督微调（SFT）、参数高效微调（PEFT）、偏好优化（如RLHF）等，因此与"Post-training/SFT”、“PEFT/LoRA”、“RLHF/DPO”、“Instruction Tuning/Alignment"均高度相关（各10分）。其他关键词如MoE、SLMs、RAG、推理加速、科学AI应用等，论文未涉及或仅作为背景提及，因此评分为0分。

扩散大语言模型（dLLMs）正逐渐成为主流自回归模型的有力替代方案，它通过迭代去噪和并行生成机制取代了严格的序列化标记生成方式。然而，其开源生态系统在模型家族之间，尤其是在训练后流程方面仍处于碎片化状态——强化学习目标、推演实现和评估脚本通常以论文专用代码库的形式发布。这种碎片化现象延缓了研究迭代速度，增加了复现的工程负担，并导致算法间的公平比较难以实现。本文提出DARE（dLLMs Alignment and Reinforcement Executor），一个用于训练后调优与评估dLLMs的开放框架。基于verl框架和OpenCompass评估体系构建的DARE，将监督微调、参数高效微调、偏好优化以及dLLM专属强化学习统一整合到适用于掩码扩散与块扩散语言模型的共享执行栈中。在涵盖LLaDA、Dream、SDAR和LLaDA2.x等代表性模型家族的实验中，DARE提供了广泛的算法覆盖、可复现的基准评估及实际加速方案。大量实证结果表明，DARE可作为可复用的研究基础平台，用于当前及新兴dLLMs的训练后方法开发、比较与部署。

Diffusion large language models (dLLMs) are emerging as a compelling alternative to dominant autoregressive models, replacing strictly sequential token generation with iterative denoising and parallel generation dynamics. However, their open-source ecosystem remains fragmented across model families and, in particular, across post-training pipelines, where reinforcement learning objectives, rollout implementations and evaluation scripts are often released as paper-specific codebases. This fragmentation slows research iteration, raises the engineering burden of reproduction, and makes fair comparison across algorithms difficult. We present \textbf{DARE} (\textbf{d}LLMs \textbf{A}lignment and \textbf{R}einforcement \textbf{E}xecutor), an open framework for post-training and evaluating dLLMs. Built on top of verl~\cite{sheng2024hybridflow} and OpenCompass~\cite{2023opencompass}, DARE unifies supervised fine-tuning, parameter-efficient fine-tuning, preference optimization, and dLLM-specific reinforcement learning under a shared execution stack for both masked and block diffusion language models. Across representative model families including LLaDA, Dream, SDAR, and LLaDA2.x, DARE provides broad algorithmic coverage, reproducible benchmark evaluation, and practical acceleration. Extensive empirical results position that DARE serves as a reusable research substrate for developing, comparing, and deploying post-training methods for current and emerging dLLMs.

评分理由: 论文专注于低资源机器翻译，提出MERIT框架，核心涉及监督微调(SFT)和基于奖励的优化(GRPO，类似RLHF/DPO)，因此与"Supervised Fine-tuning"高度相关(10分)，与"RLHF/DPO"相关(8分)。论文使用大语言模型进行翻译，与"Large Language Models"相关(8分)。研究强调数据质量和针对性数据整理，与"Scaling Laws AND Data Quality"有一定关联(5分)。框架涉及领域适应和语义对齐，与"Pre-training/Domain Adaptation"和"Instruction Tuning/Alignment"有一定关联(各5分)。其他关键词如MoE、SLMs、RAG、CoT等未在摘要中提及，评为0分。

从中文到低资源东南亚语言的神经机器翻译（NMT）仍受到两大关键制约：洁净平行语料的极度匮乏，以及现有挖掘数据中普遍存在的噪声。这一长期存在的短缺不仅阻碍了有效的模型训练，还导致其性能与高资源语言方向存在巨大差距，使得老挝语、缅甸语、他加禄语等数百万使用者所依赖的翻译系统质量持续低下，尽管大规模多语言模型近期已取得进展。我们提出了多语言专家奖励引导调优框架（MERIT），这是一个统一的翻译框架，它将传统的以英语为中心的ALT基准转化为一个以中文为中心、涵盖五种东南亚低资源语言（LRLs）的评估体系。我们的框架将语言特定标记前缀（LTP）与监督微调（SFT）相结合，并引入了一种新颖的、由语义对齐奖励（SAR）引导的组相对策略优化（GRPO）。这些结果证实，在低资源语言到中文的翻译任务中，有针对性的数据策展和奖励引导的优化策略，其效果远超单纯的模型规模扩展。

Neural machine translation (NMT) from Chinese to low-resource Southeast Asian languages remains severely constrained by the extreme scarcity of clean parallel corpora and the pervasive noise in existing mined data. This chronic shortage not only impedes effective model training but also sustains a large performance gap with high-resource directions, leaving millions of speakers of languages such as Lao, Burmese, and Tagalog with persistently low-quality translation systems despite recent advances in large multilingual models. We introduce \textbf{M}ultilingual \textbf{E}xpert-\textbf{R}eward \textbf{I}nformed \textbf{T}uning (\textbf{MERIT}), a unified translation framework that transforms the traditional English-centric ALT benchmark into a Chinese-centric evaluation suite for five Southeast Asian low-resource languages (LRLs). Our framework combines language-specific token prefixing (LTP) with supervised fine-tuning (SFT) and a novel group relative policy optimization (GRPO) guided by the semantic alignment reward (SAR). These results confirm that, in LRL{\textrightarrow}Chinese translation, targeted data curation and reward-guided optimization dramatically outperform mere model scaling.

评分理由: 论文核心研究大语言模型（LLMs）的知识蒸馏（KD）方法，提出SODA（半在线蒸馏对齐）框架，用于将大型教师模型的知识高效迁移到小型学生模型（SLMs）。因此，与"Large Language Models”、“Small Language Models”、“Post-training”（蒸馏属于后训练技术）和"Alignment”（论文强调分布对齐）高度相关（10分）。其他关键词如MoE、Scaling Laws、RAG、推理加速等，论文未涉及，给0分。

大型语言模型的黑箱知识蒸馏面临严格的权衡困境。简单的离策略方法（如序列级知识蒸馏）难以纠正学生模型固有的错误。完全同策略方法（如生成对抗蒸馏）通过对抗训练解决了这一问题，但引入了众所周知的训练不稳定性和极高的计算开销。为应对这一困境，我们提出SODA（基于对齐的半同策略蒸馏），这是一种高效替代方案，其动机源于前沿教师模型与更小规模基础模型之间固有的能力差距。由于紧凑型学生模型的自然零样本响应几乎严格劣于强大教师模型的目标输出，我们可以通过将教师模型的最优响应与学生模型一次性静态输出快照进行配对，构建出高效的对比信号。这表明，让小型学生模型接触其自身静态的次优行为已足以实现高质量的分布对齐，从而无需昂贵的动态展开过程和脆弱的对抗平衡。在四种紧凑型Qwen2.5和Llama-3模型上的广泛评估验证了这种半同策略范式。SODA在16项基准测试结果中的15项上达到或超越了现有最优方法。更重要的是，它在实现更优蒸馏质量的同时，训练速度提升10倍，峰值GPU内存消耗降低27%，并完全消除了对抗训练的不稳定性。

Black-box knowledge distillation for large language models presents a strict trade-off. Simple off-policy methods (e.g., sequence-level knowledge distillation) struggle to correct the student’s inherent errors. Fully on-policy methods (e.g., Generative Adversarial Distillation) solve this via adversarial training but introduce well-known training instability and crippling computational overhead. To address this dilemma, we propose SODA (Semi On-policy Distillation with Alignment), a highly efficient alternative motivated by the inherent capability gap between frontier teachers and much smaller base models. Because a compact student model’s natural, zero-shot responses are almost strictly inferior to the powerful teacher’s targets, we can construct a highly effective contrastive signal simply by pairing the teacher’s optimal response with a one-time static snapshot of the student’s outputs. This demonstrates that exposing the small student to its own static inferior behaviors is sufficient for high-quality distribution alignment, eliminating the need for costly dynamic rollouts and fragile adversarial balancing. Extensive evaluations across four compact Qwen2.5 and Llama-3 models validate this semi on-policy paradigm. SODA matches or outperforms the state-of-the-art methods on 15 out of 16 benchmark results. More importantly, it achieves this superior distillation quality while training 10 times faster, consuming 27% less peak GPU memory, and completely eliminating adversarial instability.

评分理由: 该论文核心研究LLM驱动的智能体（Agent）在多步工具调用环境中的安全漏洞（间接提示注入攻击）及防御策略，与"LLM Agents/Autonomous Agents/Agentic Workflow”、“Tool Use/Function Calling/API Tool Use"和"Multi-agent Systems/Agent Coordination"高度相关（10分），因为这些是论文研究的核心对象和场景。论文也基于多种LLM骨干进行评估，因此与"Large Language Models/LLMs/Foundation Models"高度相关（10分）。论文未涉及其他关键词所代表的具体技术（如MoE、量化、推理加速、科学AI应用等），因此这些关键词评分为0分。

开源框架的快速部署显著推动了现代多智能体系统的发展。然而，扩展的行动空间——包括不受控的权限暴露和隐藏的系统间交互——带来了严峻的安全挑战。具体而言，间接提示注入攻击通过将恶意指令隐藏在第三方内容中，可在智能体正常操作期间触发数据窃取等未授权行为。当前的安全评估主要依赖孤立的单轮基准测试，而这些智能体在复杂动态环境中的系统性脆弱性仍未得到充分探索。为弥补这一空白，我们系统评估了针对四种复杂间接提示注入攻击向量的六种防御策略，覆盖九种大语言模型基座。关键的是，我们的评估完全在动态多步骤工具调用环境中进行，以捕捉现代自主智能体的真实攻击面。超越二元的成功率指标，我们的多维分析揭示出显著的脆弱性：高级注入攻击成功绕过几乎所有基线防御，部分表层缓解措施甚至会产生适得其反的副作用。此外，尽管智能体几乎瞬时执行恶意指令，其内部状态却表现出异常高的决策熵。基于这种潜在犹豫现象，我们探索将表征工程作为鲁棒的检测策略。通过提取工具输入位置对应的隐藏状态，我们发现基于表征工程的断路器能在智能体执行前成功识别并拦截未授权操作，在不同大语言模型基座上均实现了高检测准确率。本研究揭示了当前间接提示注入防御的局限性，并为构建具有韧性的多智能体架构提供了高度实用的范式。

The rapid deployment of open-source frameworks has significantly advanced the development of modern multi-agent systems. However, expanded action spaces, including uncontrolled privilege exposure and hidden inter-system interactions, pose severe security challenges. Specifically, Indirect Prompt Injections (IPI), which conceal malicious instructions within third-party content, can trigger unauthorized actions such as data exfiltration during normal operations. While current security evaluations predominantly rely on isolated single-turn benchmarks, the systemic vulnerabilities of these agents within complex dynamic environments remain critically underexplored. To bridge this gap, we systematically evaluate six defense strategies against four sophisticated IPI attack vectors across nine LLM backbones. Crucially, we conduct our evaluation entirely within dynamic multi-step tool-calling environments to capture the true attack surface of modern autonomous agents. Moving beyond binary success rates, our multidimensional analysis reveals a pronounced fragility. Advanced injections successfully bypass nearly all baseline defenses, and some surface-level mitigations even produce counterproductive side effects. Furthermore, while agents execute malicious instructions almost instantaneously, their internal states exhibit abnormally high decision entropy. Motivated by this latent hesitation, we investigate Representation Engineering (RepE) as a robust detection strategy. By extracting hidden states at the tool-input position, we revealed that the RepE-based circuit breaker successfully identifies and intercepts unauthorized actions before the agent commits to them, achieving high detection accuracy across diverse LLM backbones. This study exposes the limitations of current IPI defenses and provides a highly practical paradigm for building resilient multi-agent architectures.

评分理由: 论文聚焦于医学影像（胸片）的时间序列分析，属于AI for Science（生物信息学）应用，得10分；涉及vision-language pretraining（医学基础模型），与LLMs/Foundation Models相关但非核心，得8分；明确包含pretraining和fine-tuning（SFT）技术，各得10分；其他关键词如MoE、SLMs、Scaling Laws、Alignment、RLHF、RAG、Reasoning、Agents、Compression等均未涉及，得0分。

视觉-语言预训练的最新进展催生了强大的医学基础模型，但多数模型仍孤立分析放射影像，忽视了对比既往与当前影像以评估时序变化这一关键临床任务。对于胸部X光片（CXRs）而言，捕捉时序变化至关重要，因为放射科医生不仅需要评估影像表现的静态特征，还必须追踪其随时间的演变过程。我们提出TILA（时序反转感知学习与对齐框架），这是一种简洁而有效的框架，其使用时序反转（即反转图像对的顺序）作为监督信号，以增强现有时序视觉-语言模型对方向性变化的敏感性。TILA在预训练、微调和推理阶段整合了反转感知目标，通过显式学习时序顺序来补充传统的外观建模方法。我们还提出了一套统一的评估协议，用于衡量模型在时序反转下的顺序敏感性与一致性，并构建了MS-CXR-Tretrieval检索评估集——该评估集基于通用构建协议，可应用于任何时序CXR数据集。在公开数据集和真实医院队列上的实验表明，当应用于多种现有架构时，TILA能持续提升疾病进展分类的准确性和时序嵌入的对齐效果。

Recent advances in vision–language pretraining have enabled strong medical foundation models, yet most analyze radiographs in isolation, overlooking the key clinical task of comparing prior and current images to assess interval change. For chest radiographs (CXRs), capturing interval change is essential, as radiologists must evaluate not only the static appearance of findings but also how they evolve over time. We introduce TILA (Temporal Inversion-aware Learning and Alignment), a simple yet effective framework that uses temporal inversion, reversing image pairs, as a supervisory signal to enhance the sensitivity of existing temporal vision-language models to directional change. TILA integrates inversion-aware objectives across pretraining, fine-tuning, and inference, complementing conventional appearance modeling with explicit learning of temporal order. We also propose a unified evaluation protocol to assess order sensitivity and consistency under temporal inversion, and introduce MS-CXR-Tretrieval, a retrieval evaluation set constructed through a general protocol that can be applied to any temporal CXR dataset. Experiments on public datasets and real-world hospital cohorts demonstrate that TILA consistently improves progression classification and temporal embedding alignment when applied to multiple existing architectures.

评分理由: 论文提出TimeRAN，一个用于无线接入网（RAN）时间序列建模的统一多任务学习框架，其核心是一个轻量级的时间序列基础模型，并进行了大规模预训练和有限监督下的高效适应。因此，它与"Foundation Models”（8分）高度相关，因为论文明确提出了一个基础模型；与"Pre-training"（10分）高度相关，因为论文进行了大规模预训练并开源了数据集；与"Supervised Fine-tuning"（8分）相关，因为提到了任务特定的微调；与"AI for Science"（5分）有一定关联，因为RAN优化属于AI在特定领域（通信网络）的应用。其他关键词（如LLM、MoE、RLHF等）主要针对自然语言处理或特定技术，与论文的时间序列模型和RAN应用领域无关，故得0分。

无线接入网（RAN）正演变为一种可编程、解耦的基础设施，日益依赖原生人工智能算法进行优化与闭环控制。然而，当前的RAN智能系统仍主要由针对单一功能设计的任务专用模型构成，导致模型碎片化、任务间知识共享有限、泛化能力差以及系统复杂性增加。为应对这些局限，本文提出TimeRAN——一个面向RAN时序建模的统一多任务学习框架。TimeRAN采用轻量级时序基础模型配合少量任务专用头部，通过学习可迁移的表征，能够在有限监督下高效适配多种任务。为实现大规模预训练，我们进一步整理并开源了TimeRAN DataPile，这是迄今为止最大的RAN分析时序数据集，涵盖多样化遥测源、协议层和部署场景，包含超过35.5万条时间序列和5.6亿个测量点。我们在涵盖异常检测、分类、预测与填补的完整RAN分析任务集上评估TimeRAN，结果表明其仅需极少或无需任务特定微调即可达到先进性能。最后，我们将TimeRAN集成至概念验证型5G测试平台，证明其在实际场景中能以有限资源需求高效运行。

The Radio Access Network (RAN) is evolving into a programmable and disaggregated infrastructure that increasingly relies on AI-native algorithms for optimization and closed-loop control. However, current RAN intelligence is still largely built from task-specific models tailored to individual functions, resulting in model fragmentation, limited knowledge sharing across tasks, poor generalization, and increased system complexity. To address these limitations, we introduce TimeRAN, a unified multi-task learning framework for time-series modeling in the RAN. TimeRAN leverages a lightweight time-series foundation model with few task-specific heads to learn transferable representations that can be efficiently adapted across diverse tasks with limited supervision. To enable large-scale pretraining, we further curate and open-source TimeRAN DataPile, the largest time-series corpus for RAN analytics to date, comprising over 355K time series and 0.56B measurements across diverse telemetry sources, protocol layers, and deployment scenarios. We evaluate TimeRAN across a comprehensive set of RAN analytics tasks, including anomaly detection, classification, forecasting, and imputation, and show that it achieves state-of-the-art performance with minimal or no task-specific fine-tuning. Finally, we integrate TimeRAN into a proof-of-concept 5G testbed and demonstrate that it operates efficiently with limited resource requirements in real-world scenarios.

评分理由: 论文提出AI Trust OS治理框架，核心关注LLM、RAG和多智能体工作流在企业环境中的治理、可观测性和合规性问题。摘要明确提到"large language models, retrieval-augmented generation pipelines, and multi-agent AI workflows"，因此这三个关键词高度相关（10分）。论文属于大模型应用领域（企业治理），但未涉及其他关键词的具体技术原理或创新，因此其他关键词评分为0分。

大型语言模型、检索增强生成流程以及多智能体人工智能工作流的加速普及已引发结构性治理危机。组织无法监管其不可见之物，而为确定性网络应用构建的现有合规方法论缺乏相应机制，用以发现或持续验证那些在工程团队间自发产生、缺乏正式监管的人工智能系统。这导致监管机构要求的人工智能治理成熟度证明与组织实际可提供的证据之间出现日益扩大的信任鸿沟。本文提出“AI信任操作系统”——一种面向持续自主人工智能可观测性与零信任合规的治理架构。该系统将合规性重新构想为一个全天候运行、由遥测数据驱动的操作层：通过可观测性信号发现人工智能系统，借助自动化探针收集控制断言，并持续合成信任凭证。该框架基于四大原则：主动发现、基于遥测证据而非人工证明、持续状态监测而非时点审计、架构支撑的实证而非政策文件的信任。其通过零信任遥测边界运作，瞬时只读探针在此验证结构性元数据，无需接入源代码或载荷级个人身份信息。人工智能可观测性提取代理会扫描LangSmith与Datadog的LLM遥测数据，自动注册未记录的人工智能系统，从而将治理模式从组织自我报告转向实证化的机器观测。通过对ISO 42001、欧盟《人工智能法案》、SOC 2、GDPR及HIPAA等标准的评估，本文论证了遥测优先的人工智能治理模式代表了企业信任生成与验证方式的根本性架构变革。

The accelerating adoption of large language models, retrieval-augmented generation pipelines, and multi-agent AI workflows has created a structural governance crisis. Organizations cannot govern what they cannot see, and existing compliance methodologies built for deterministic web applications provide no mechanism for discovering or continuously validating AI systems that emerge across engineering teams without formal oversight. The result is a widening trust gap between what regulators demand as proof of AI governance maturity and what organizations can demonstrate. This paper proposes AI Trust OS, a governance architecture for continuous, autonomous AI observability and zero-trust compliance. AI Trust OS reconceptualizes compliance as an always-on, telemetry-driven operating layer in which AI systems are discovered through observability signals, control assertions are collected by automated probes, and trust artifacts are synthesized continuously. The framework rests on four principles: proactive discovery, telemetry evidence over manual attestation, continuous posture over point-in-time audit, and architecture-backed proof over policy-document trust. The framework operates through a zero-trust telemetry boundary in which ephemeral read-only probes validate structural metadata without ingressing source code or payload-level PII. An AI Observability Extractor Agent scans LangSmith and Datadog LLM telemetry, automatically registering undocumented AI systems and shifting governance from organizational self-report to empirical machine observation. Evaluated across ISO 42001, the EU AI Act, SOC 2, GDPR, and HIPAA, the paper argues that telemetry-first AI governance represents a categorical architectural shift in how enterprise trust is produced and demonstrated.

评分理由: 论文核心是提出一个LLM驱动的多智能体仿真框架（LLM-MAS）用于优化服务运营，直接高度相关于"Large Language Models"（LLMs是核心工具）、“LLM Agents”（使用LLM作为智能体）和"Multi-agent Systems"（多智能体仿真框架）。其他关键词如MoE、SFT、RAG、推理加速等涉及具体技术细节或应用领域，论文未涉及，故评分为0。

服务系统性能取决于参与者对设计选择的响应方式，但由于人类行为的复杂性，对这些响应进行建模十分困难。本文提出一种基于大语言模型的多智能体仿真框架，用于优化服务运营。我们将该问题构建为具有决策依赖不确定性的随机优化问题：设计选择被嵌入提示词中，并塑造了基于大语言模型的智能体交互所产生结果的分布。通过将关键数值信息嵌入提示词并从大语言模型生成的文本中提取该信息，我们将这种不确定性建模为一个受控马尔可夫链。我们开发了一种轨迹上学习算法，该算法在单次仿真运行中，同时构建零阶梯度估计并更新设计参数，以优化稳态性能。我们还引入了方差缩减技术。在一个可持续供应链的应用中，我们的方法在性能上超越了多种基准方案，包括黑箱优化、使用大语言模型作为数值求解器或作为角色扮演系统设计者的方法。一项基于真实行为数据的最优竞赛设计案例研究表明，LLM-MAS既可作为已知设计方案的性价比评估工具，也可作为一种探索性工具，能够发现被传统方法忽视的优秀设计方案。

Service system performance depends on how participants respond to design choices, but modeling these responses is hard due to the complexity of human behavior. We introduce an LLM-powered multi-agent simulation (LLM-MAS) framework for optimizing service operations. We pose the problem as stochastic optimization with decision-dependent uncertainty: design choices are embedded in prompts and shape the distribution of outcomes from interacting LLM-powered agents. By embedding key numerical information in prompts and extracting it from LLM-generated text, we model this uncertainty as a controlled Markov chain. We develop an on-trajectory learning algorithm that, on a single simulation run, simultaneously constructs zeroth-order gradient estimates and updates design parameters to optimize steady-state performance. We also incorporate variance reduction techniques. In a sustainable supply chain application, our method outperforms benchmarks, including blackbox optimization and using LLMs as numerical solvers or as role-playing system designers. A case study on optimal contest design with real behavioral data shows that LLM-MAS is both as a cost-effective evaluator of known designs and an exploratory tool that can uncover strong designs overlooked by traditional approaches.

评分理由: 论文提出CUP框架，将语言模型与结构化规划结合用于目标导向对话，核心涉及LLM（用于生成可行动作）和多步决策规划（体现推理过程）。因此，与"Large Language Models"（论文明确使用LLM）和"LLM Agents"（框架本质是LLM驱动的决策代理）高度相关（8分）。与"Chain of Thought"和"System 2 Thinking"有一定关联（5分），因为论文强调多轮、长视野的决策规划，涉及不确定性下的逐步推理，但未直接使用这些术语。其他关键词（如MoE、量化、科学AI等）未在论文中涉及，得0分。

面向目标的对话系统需要在用户意图不确定的情况下进行序列决策，算法必须在多轮对话中平衡信息获取与目标确认。现有方法从不同角度应对这一挑战：结构化方法支持多步规划但依赖预定义模式，而基于大语言模型（LLM）的方法虽支持灵活交互，却缺乏长程决策能力，导致信息获取与目标确认之间协调不足。为突破这一局限，我们将面向目标的对话建模为一种不确定性感知的序列决策问题，其中不确定性作为多轮决策的引导信号。我们提出了一种对话不确定性感知规划框架（Conversation Uncertainty-aware Planning, CUP），将语言模型与结构化规划相结合：语言模型生成可行动作，规划器则评估这些动作对降低不确定性的长期影响。在多个对话基准测试上的实验表明，CUP能持续提升任务成功率，同时减少交互轮次。进一步分析证明，不确定性感知规划有助于实现更高效的信息获取与更早的确定性目标确认。

Goal-oriented conversational systems require making sequential decisions under uncertainty about the user’s intent, where the algorithm must balance information acquisition and target commitment over multiple turns. Existing approaches address this challenge from different perspectives: structured methods enable multi-step planning but rely on predefined schemas, while LLM-based approaches support flexible interactions but lack long-horizon decision making, resulting in poor coordination between information acquisition and target commitment. To address this limitation, we formulate goal-oriented conversation as an uncertainty-aware sequential decision problem, where uncertainty serves as a guiding signal for multi-turn decision making. We propose a Conversation Uncertainty-aware Planning framework (CUP) that integrates language models with structured planning: a language model proposes feasible actions, and a planner evaluates their long-term impact on uncertainty reduction. Experiments on multiple conversational benchmarks show that CUP consistently improves success rates while requiring fewer interaction turns. Further analysis demonstrates that uncertainty-aware planning contributes to more efficient information acquisition and earlier confident commitment.

评分理由: 论文核心是使用语言模型（Creative Pre-trained Transformer, CPT）进行创意设计生成，属于大模型在特定领域的应用创新。与关键词高度相关的是：1）“Large Language Models”（论文明确使用decoder-only语言模型CPT，权重1.0，评分10.0）；2）“Pre-training”（模型名为Creative Pre-trained Transformer，涉及预训练，评分8.0）；3）“Post-training”（论文提到fine-tune CPT on a large corpus，属于后训练/微调，评分8.0）。其他关键词如MoE、SLMs、Scaling Laws、Alignment、RAG等均未在摘要中提及或与论文内容无关，故评分为0。论文未涉及指定专家作者。

生成视觉多样且高质量的设计作品目前仍依赖人工操作，这一过程耗时费力，限制了创意工作流程的可扩展性与个性化程度。本研究提出一种基于仅解码器语言模型的系统，用于生成可编辑的设计变体。该模型名为创意预训练变换器（Creative Pre-trained Transformer，CPT），通过训练可预测设计模板中的视觉风格属性。我们方法的核心是一种称为创意标记语言（Creative Markup Language，CML）的新型表征格式——这是一种紧凑且适配机器学习的格式，能够捕捉画布级结构、页面布局以及元素级细节（文本、图像和矢量图形），同时涵盖内容与样式信息。我们在专业设计师创作的大规模设计模板语料库上对CPT进行微调，使其能够针对配色方案、字体选择等属性进行具有语义关联性的上下文感知预测。该模型生成的输出在语义结构上具有层次性，在风格上保持连贯性，并确保各元素间的内在一致性。与生成式图像模型不同，本系统产出的是完全可编辑的设计文档而非仅含像素的图像，允许用户在设计编辑器中直接进行迭代与个性化调整。实验表明，我们的方法能够为现有模板生成符合语境的色彩与字体变体，并在遵循设计原则的前提下展现出调整布局的潜力。

Designing visually diverse and high-quality designs remains a manual, time-consuming process, limiting scalability and personalization in creative workflows. We present a system for generating editable design variations using a decoder-only language model, the Creative Pre-trained Transformer (CPT), trained to predict visual style attributes in design templates. At the core of our approach is a new representation called Creative Markup Language (CML), a compact, machine-learning-friendly format that captures canvas-level structure, page layout, and element-level details (text, images, and vector graphics), including both content and style. We fine-tune CPT on a large corpus of design templates authored by professional designers, enabling it to learn meaningful, context-aware predictions for attributes such as color schemes and font choices. The model produces semantically structured and stylistically coherent outputs, preserving internal consistency across elements. Unlike generative image models, our system yields fully editable design documents rather than pixel-only images, allowing users to iterate and personalize within a design editor. In experiments, our approach generates contextual color and font variations for existing templates and shows promise in adjusting layouts while maintaining design principles.

评分理由: 论文核心是使用LLMs和RAG技术构建AI教学助手，并评估其在工程课程中的效果。因此，与"Large Language Models"和"Retrieval-Augmented Generation"高度相关（10分），因为LLMs是基础技术，RAG是具体实现方法。其他关键词如MoE、SFT、RLHF、量化等均未在摘要中提及，属于完全无关（0分）。论文属于大模型在教育领域的应用，符合研究背景中"大模型在不同领域的研究应用"，但未涉及生物医药等特定科学领域。

过去几年，大语言模型的迅速崛起推动了基于大语言模型的AI教学工具的广泛实验。然而，其具体实施方案及其在教学环境中的效益仍处于探索的早期阶段。本文以都柏林圣三一学院电影工程硕士课程为背景，探讨了利用检索增强生成技术实现AI助教所涉及的上述问题。我们详细介绍了实施方案（包括向大语言模型提供的提示词及代码），并重点阐述了如何针对课程需求设计与优化检索增强生成流程。我们说明了调查工具的设计，并通过多项量化指标报告了AI助教的实际影响。本次实验的规模（43名学生，7周内进行296次会话，产生1,889次查询）足以支撑研究结论的可信度。与以往研究不同，我们尝试在开卷考试中允许使用AI助教。对三次考试的统计分析显示，无论是否使用AI助教，学生成绩均无显著差异（p > 0.05），这表明精心设计的评估方式能够保持学术有效性。学生反馈显示AI助教具有积极帮助（平均分=4.22/5），但对于是否更倾向于AI助教而非人工辅导则意见不一（平均分=2.78/5）。

The rapid rise of LLMs over the last few years has promoted growing experimentation with LLM-driven AI tutors. However, the details of implementation, as well as the benefit in a teaching environment, are still in the early days of exploration. This article addresses these issues in the context of implementation of an AI Teaching Assistant (AI-TA) using Retrieval Augmented Generation (RAG) for Trinity College Dublin’s Master’s Motion Picture Engineering (MPE) course. We provide details of our implementation (including the prompt to the LLM, and code), and highlight how we designed and tuned our RAG pipeline to meet course needs. We describe our survey instrument and report on the impact of the AI-TA through a number of quantitative metrics. The scale of our experiment (43 students, 296 sessions, 1,889 queries over 7 weeks) was sufficient to have confidence in our findings. Unlike previous studies, we experimented with allowing the use of the AI-TA in open-book examinations. Statistical analysis across three exams showed no performance differences regardless of AI-TA access (p > 0.05), demonstrating that thoughtfully designed assessments can maintain academic validity. Student feedback revealed that the AI-TA was beneficial (mean = 4.22/5), while students had mixed feelings about preferring it over human tutoring (mean = 2.78/5).

评分理由: 论文MolDA提出了一种用于分子理解和生成的新型多模态框架，核心创新在于用离散大语言扩散模型替代传统的自回归骨干网络。该研究与关键词高度相关的有：1）“Large Language Models”：论文明确使用LLMs作为基础，并提出了基于LLM的扩散模型变体，是核心内容（10分）。2）“AI for Science”：论文专注于分子发现这一科学领域，是LLM在生物信息学/化学信息学中的具体应用，属于核心内容（10分）。其他关键词如MoE、SLMs、Scaling Laws、各种训练方法（Pre-training、SFT、RLHF等）、推理优化技术、代理系统、模型压缩等，在摘要中均未提及或讨论，因此评分为0分。

大语言模型（LLM）显著推动了分子发现，但现有的多模态分子架构本质上依赖于自回归（AR，autoregressive）主干网络。这种严格的从左到右归纳偏置对于生成化学有效分子而言并非最优，因为它难以处理非局部的全局约束（例如环闭合），且在序列生成过程中常常累积结构错误。为应对这些局限，我们提出了MolDA（基于掩码扩散的分子语言模型），这是一种新颖的多模态框架，它用离散的大语言扩散模型取代了传统的自回归主干。MolDA通过混合图编码器提取全面的结构表征，该编码器同时捕获局部和全局拓扑信息，并通过一个Q-Former将其对齐到语言标记空间。此外，我们专门针对掩码扩散过程，对分子结构偏好优化进行了数学重构。通过双向迭代去噪，MolDA确保了在分子生成、描述和性质预测任务中具有全局结构一致性、化学有效性以及强大的推理能力。

Large Language Models (LLMs) have significantly advanced molecular discovery, but existing multimodal molecular architectures fundamentally rely on autoregressive (AR) backbones. This strict left-to-right inductive bias is sub-optimal for generating chemically valid molecules, as it struggles to account for non-local global constraints (e.g., ring closures) and often accumulates structural errors during sequential generation. To address these limitations, we propose MolDA (Molecular language model with masked Diffusion with mAsking), a novel multimodal framework that replaces the conventional AR backbone with a discrete Large Language Diffusion Model. MolDA extracts comprehensive structural representations using a hybrid graph encoder, which captures both local and global topologies, and aligns them into the language token space via a Q-Former. Furthermore, we mathematically reformulate Molecular Structure Preference Optimization specifically for the masked diffusion. Through bidirectional iterative denoising, MolDA ensures global structural coherence, chemical validity, and robust reasoning across molecule generation, captioning, and property prediction.

评分理由: 论文提出ACE框架，核心是使用LLMs进行专利权利要求验证，并创新性地提出了Chain of Patent Thought (CoPT)协议，这是Chain of Thought推理在专利领域的应用。因此，与"Large Language Models"和"Chain of Thought"高度相关（10分）。论文未涉及其他关键词的具体技术或应用，如MoE、SLMs、Scaling Laws、各种训练方法、对齐、RAG、推理加速、模型压缩、AI for Science等，故这些关键词得0分。

专利权利要求的自动化验证要求零缺陷容错，因为即使存在单一结构性瑕疵也可能导致权利要求在法律上失效。现有评估范式面临刚性-资源困境：轻量级编码器难以处理微妙的法律依存关系，而基于大语言模型（LLMs）的穷举验证则成本过高。为弥合这一差距，我们提出ACE（自适应成本效益评估）混合框架，该框架利用预测熵仅将高不确定性权利要求路由至专家级LLM。专家模型随后执行基于《美国法典》第35编法定标准构建的专利思维链（CoPT）协议。该设计使ACE能更有效地处理长程法律依存关系，同时保持效率优势。在评估方法中，ACE以94.95%的F1分数取得最优性能，相较于独立LLM部署方案降低78%运营成本。我们还构建了包含4万项权利要求、基于《专利审查程序手册》（MPEP）错误标注的基准数据集ACE-40k，以推动后续研究。

Automated validation of patent claims demands zero-defect tolerance, as even a single structural flaw can render a claim legally defective. Existing evaluation paradigms suffer from a rigidity-resource dilemma: lightweight encoders struggle with nuanced legal dependencies, while exhaustive verification via Large Language Models (LLMs) is prohibitively costly. To bridge this gap, we propose ACE (Adaptive Cost-efficient Evaluation), a hybrid framework that uses predictive entropy to route only high-uncertainty claims to an expert LLM. The expert then executes a Chain of Patent Thought (CoPT) protocol grounded in 35 U.S.C. statutory standards. This design enables ACE to handle long-range legal dependencies more effectively while preserving efficiency. ACE achieves the best F1 among the evaluated methods at 94.95%, while reducing operational costs by 78% compared to standalone LLM deployments. We also construct ACE-40k, a 40,000-claim benchmark with MPEP-grounded error annotations, to facilitate further research.

评分理由: 该论文研究多语言提示本地化对Agent-as-a-Judge评估的影响，核心涉及LLM作为评估代理（Agent）在不同语言下的性能变化。因此，与"LLM Agents"高度相关（10分），因为论文直接研究LLM作为代理在评估任务中的行为。与"Large Language Models"相关（8分），因为研究使用了GPT-4o、Gemini等大模型作为评估骨干（backbone）。其他关键词如MoE、SFT、RAG等均未在论文中涉及，故给0分。

在智能体代码基准测试中，评估语言通常被视为固定的英语默认选项，然而我们发现，改变评判语言可以逆转骨干模型的排名顺序。本研究将“智能体即评判员”（Agent-as-a-Judge）的提示框架本地化为五种类型学上多样化的语言（英语、阿拉伯语、土耳其语、中文、印地语），并在三个开发者智能体框架和六个评判骨干模型上对55项DevAI开发任务进行了评估，总计完成4950次评判运行。核心发现是骨干模型与语言之间存在交互作用：GPT-4o在英语中获得了最高的满意度（44.72%），而Gemini在阿拉伯语（51.72%，与GPT-4o相比$p<0.001$）和印地语（53.22%）中表现领先。没有任何单一骨干模型能在所有语言中占据主导地位，且不同骨干模型对单项需求判断的一致性仅为中等水平（弗莱斯$κ\leq 0.231$）。一项受控消融实验进一步表明，仅本地化基准内容而忽略评判端指令可能导致决定性差异：在部分本地化条件下，印地语的满意度从42.8%下降至23.2%。这些结果表明，在智能体基准测试中，语言应被视为明确的评估变量。我们已公开发布完整的需求层级判断结果与运行时统计数据，以确保可复现性。

Evaluation language is typically treated as a fixed English default in agentic code benchmarks, yet we show that changing the judge’s language can invert backbone rankings. We localize the Agent-as-a-Judge prompt stack to five typologically diverse languages (English, Arabic, Turkish, Chinese, Hindi) and evaluate 55 DevAI development tasks across three developer-agent frameworks and six judge backbones, totaling 4950 judge runs. The central finding is that backbone and language interact: GPT-4o achieves the highest satisfaction in English (44.72%), while Gemini leads in Arabic (51.72%, $p<0.001$ vs.\ GPT-4o) and Hindi (53.22%). No single backbone dominates across all languages, and inter-backbone agreement on individual requirement judgments is modest (Fleiss’ $κ\leq 0.231$). A controlled ablation further shows that localizing judge-side instructions, not just benchmark content, can be decisive: Hindi satisfaction drops from 42.8% to 23.2% under partial localization. These results indicate that language should be treated as an explicit evaluation variable in agentic benchmarks. Full requirement-level judgments and runtime statistics are released for reproducibility.

评分理由: 论文核心研究LLMs作为协作写作伙伴时出现的"黑暗模式"行为，与"Large Language Models"高度相关（10分），并探讨这些行为与安全对齐（Alignment）的关联（8分）。论文未涉及其他技术原理或特定应用领域的关键词，因此其余关键词评分为0。

大型语言模型（LLMs）正日益成为协作写作伙伴，这引发了关于其对人类能动性影响的疑问。在这项探索性研究中，我们调查了人机共创中五种“暗黑模式”——即可能抑制或扭曲创作过程的微妙模型行为：谄媚附和、语调管制、道德说教、死亡循环与锚定效应。通过一系列受控实验，我们以多种文学形式和主题为背景，将LLMs作为写作助手进行提示，并分析生成响应中这些行为的普遍性。初步结果表明，谄媚附和行为几乎无处不在（占案例的91.7%），在敏感话题中尤为突出；而锚定效应似乎与文学形式相关，在民间故事中出现频率最高。本研究表明，这些常作为安全对齐副产品的暗黑模式，可能无意中限制了创作探索的广度，并据此提出了有效支持创意写作的AI系统设计考量。

Large language models (LLMs) are increasingly acting as collaborative writing partners, raising questions about their impact on human agency. In this exploratory work, we investigate five “dark patterns” in human-AI co-creativity – subtle model behaviors that can suppress or distort the creative process: Sycophancy, Tone Policing, Moralizing, Loop of Death, and Anchoring. Through a series of controlled sessions where LLMs are prompted as writing assistants across diverse literary forms and themes, we analyze the prevalence of these behaviors in generated responses. Our preliminary results suggest that Sycophancy is nearly ubiquitous (91.7% of cases), particularly in sensitive topics, while Anchoring appears to be dependent on literary forms, surfacing most frequently in folktales. This study indicates that these dark patterns, often byproducts of safety alignment, may inadvertently narrow creative exploration and proposes design considerations for AI systems that effectively support creative writing.

评分理由: 论文核心研究LLM在个人投资决策中的个性化定制挑战，属于LLM应用研究。与"Large Language Models"高度相关（10分），因为全文围绕LLM系统展开。与"Instruction Tuning" OR “Alignment” OR “Value Alignment"有一定关联（8分），因为论文讨论了在没有固定标签集的情况下评估个性化质量的"alignment without ground truth"问题。其他关键词如MoE、SLMs、Scaling Laws、Pre-training、RLHF、RAG、CoT、Agents、Quantization等均未在摘要中提及或相关，故评0分。

个性化大语言模型系统发展迅速，但多数应用于用户偏好稳定、且缺乏客观事实依据或标准主观的领域。我们认为，个人投资者决策为LLM个性化提供了一个极具挑战性的独特领域——它暴露了当前定制范式的根本性局限。基于我们为人工智能增强型投资组合管理所构建并部署的系统，我们识别出个人投资领域揭示标准LLM定制根本局限的四个维度：（1）行为记忆复杂性，即投资者的行为模式随时间演变、自相矛盾且具有财务影响；（2）漂移情境下的逻辑一致性，即在数周或数月内保持连贯的投资逻辑，这对无状态和会话受限的架构构成压力；（3）风格与信号的张力，即系统必须同时尊重个人投资理念，并呈现可能与之矛盾的客观证据；（4）无事实依据的对齐，即由于投资结果具有随机性和延迟性，个性化质量无法对照固定标签集进行评估。我们阐述了构建该系统过程中产生的架构应对方案，并针对高风险、长周期决策领域的个性化自然语言处理提出了开放的研究方向。

Personalized LLM systems have advanced rapidly, yet most operate in domains where user preferences are stable and ground truth is either absent or subjective. We argue that individual investor decision-making presents a uniquely challenging domain for LLM personalization - one that exposes fundamental limitations in current customization paradigms. Drawing on our system, built and deployed for AI-augmented portfolio management, we identify four axes along which individual investing exposes fundamental limitations in standard LLM customization: (1) behavioral memory complexity, where investor patterns are temporally evolving, self-contradictory, and financially consequential; (2) thesis consistency under drift, where maintaining coherent investment rationale over weeks or months strains stateless and session-bounded architectures; (3) style-signal tension, where the system must simultaneously respect personal investment philosophy and surface objective evidence that may contradict it; and (4) alignment without ground truth, where personalization quality cannot be evaluated against a fixed label set because outcomes are stochastic and delayed. We describe the architectural responses that emerged from building the system and propose open research directions for personalized NLP in high-stakes, temporally extended decision domains.

评分理由: 论文提出SARES-DEIM框架，核心创新是SARESMoE模块，该模块明确使用了稀疏门控机制和专家混合（Mixture-of-Experts）架构，因此与关键词"Mixture of Experts” OR “MoE” OR “Sparse Models"高度相关（10分）。论文属于计算机视觉在遥感领域的应用，可视为AI在科学（遥感）领域的应用，与"AI for Science"有一定关联（5分）。论文未涉及大语言模型、训练方法、推理优化、智能体等其他技术，因此其余关键词均为0分。

合成孔径雷达（SAR）图像中的船舶检测面临着固有的相干斑点噪声、复杂的海岸杂波以及小尺度目标普遍存在等根本性挑战。传统检测器主要为光学图像设计，通常对SAR特有的图像退化鲁棒性有限，并且在空间下采样过程中会丢失细粒度的船舶特征。为解决这些局限性，我们提出了SARES-DEIM，这是一个基于DEtection TRansformer（DETR）范式的领域感知检测框架。我们方法的核心是SARESMoE（SAR感知专家选择混合专家）模块，该模块利用稀疏门控机制，将特征选择性地路由至专门的频率和小波专家。这种稀疏激活架构能有效滤除斑点噪声和语义杂波，同时保持较高的计算效率。此外，我们引入了空间到深度增强金字塔（SDEP）颈部结构，以保留来自浅层阶段的高分辨率空间线索，从而显著提升小目标的定位能力。在两个基准数据集上的大量实验证明了SARES-DEIM的优越性。值得注意的是，在具有挑战性的HRSID数据集上，我们的模型实现了76.4%的mAP50:95和93.8%的mAP50，性能超越了最先进的YOLO系列及专用SAR检测器。

Ship detection in Synthetic Aperture Radar (SAR) imagery is fundamentally challenged by inherent coherent speckle noise, complex coastal clutter, and the prevalence of small-scale targets. Conventional detectors, primarily designed for optical imagery, often exhibit limited robustness against SAR-specific degradation and suffer from the loss of fine-grained ship signatures during spatial downsampling. To address these limitations, we propose SARES-DEIM, a domain-aware detection framework grounded in the DEtection TRansformer (DETR) paradigm. Central to our approach is SARESMoE (SAR-aware Expert Selection Mixture-of-Experts), a module leveraging a sparse gating mechanism to selectively route features toward specialized frequency and wavelet experts. This sparsely-activated architecture effectively filters speckle noise and semantic clutter while maintaining high computational efficiency. Furthermore, we introduce the Space-to-Depth Enhancement Pyramid (SDEP) neck to preserve high-resolution spatial cues from shallow stages, significantly improving the localization of small targets. Extensive experiments on two benchmark datasets demonstrate the superiority of SARES-DEIM. Notably, on the challenging HRSID dataset, our model achieves a mAP50:95 of 76.4% and a mAP50 of 93.8%, outperforming state-of-the-art YOLO-series and specialized SAR detectors.

评分理由: 论文专注于计算机视觉中的目标检测任务，提出了一种名为HI-MoE的层次化实例条件混合专家架构。该论文的核心创新在于将MoE架构应用于视觉任务，特别是针对目标检测的实例级路由设计。因此，仅与关键词"Mixture of Experts” OR “MoE” OR “Sparse Models"高度相关（15分），因为MoE是论文的核心方法，涉及稀疏路由和条件计算。其他关键词主要涉及语言模型、训练技术、推理方法、代理系统、模型压缩等，与这篇视觉领域的论文无直接关联，故均评为0分。

专家混合（Mixture-of-Experts, MoE）架构通过仅为每个输入激活模型参数的一个子集来实现条件计算。尽管稀疏路由在语言模型中极为有效，并在视觉领域也展现出潜力，但大多数视觉MoE方法在图像或图像块级别进行操作。这种粒度与目标检测任务并不匹配，因为该任务的基本推理单元是对应候选实例的目标查询。我们提出分层实例条件化专家混合（Hierarchical Instance-Conditioned Mixture-of-Experts, HI-MoE），这是一种DETR风格的检测架构，其路由过程分为两个阶段：轻量级场景路由器首先选择一组场景一致的专家子集，随后实例路由器将每个目标查询分配给该子集中的少数专家。此设计旨在保持计算稀疏性的同时，更好地匹配检测任务中异构的、以实例为中心的结构。在当前草案中，实验主要集中于COCO数据集，并在LVIS上进行了初步的专家专业化分析。在此设置下，HI-MoE相较于稠密的DINO基线模型以及更简单的令牌级别或仅实例级别的路由变体均有所提升，在小目标检测上增益尤为显著。我们还提供了专家专业化模式的初步可视化。我们以旨在支持进一步实验验证的形式，介绍了该方法、消融研究以及当前存在的局限性。

Mixture-of-Experts (MoE) architectures enable conditional computation by activating only a subset of model parameters for each input. Although sparse routing has been highly effective in language models and has also shown promise in vision, most vision MoE methods operate at the image or patch level. This granularity is poorly aligned with object detection, where the fundamental unit of reasoning is an object query corresponding to a candidate instance. We propose Hierarchical Instance-Conditioned Mixture-of-Experts (HI-MoE), a DETR-style detection architecture that performs routing in two stages: a lightweight scene router first selects a scene-consistent expert subset, and an instance router then assigns each object query to a small number of experts within that subset. This design aims to preserve sparse computation while better matching the heterogeneous, instance-centric structure of detection. In the current draft, experiments are concentrated on COCO with preliminary specialization analysis on LVIS. Under these settings, HI-MoE improves over a dense DINO baseline and over simpler token-level or instance-only routing variants, with especially strong gains on small objects. We also provide an initial visualization of expert specialization patterns. We present the method, ablations, and current limitations in a form intended to support further experimental validation.

评分理由: 论文研究图像编辑方法VicoEdit，属于计算机视觉领域，而非大语言模型（LLM）或深度学习技术原理的直接创新。与关键词的相关性分析如下：1）仅与"Pre-training”（提及使用预训练模型）和"Alignment"（提及概念对齐）有中等关联（5分），因为论文使用了预训练的文本提示编辑模型，并设计了概念对齐引导的后验采样方法。2）其他关键词均与论文内容无关（0分），因为论文专注于图像编辑的视觉上下文集成，不涉及LLM、推理、对齐、压缩、科学AI等主题。

在图像编辑中，引入上下文图像以传达用户的精确需求（如主体外观或图像风格）至关重要。现有的基于训练的视觉上下文感知编辑方法需要数据收集工作和训练成本。另一方面，免训练方案通常基于扩散反演技术建立，其在一致性和灵活性方面存在不足。本文提出VicoEdit，一种免训练且无需反演的方法，能够将视觉上下文注入预训练的文本提示编辑模型中。具体而言，VicoEdit基于视觉上下文直接将源图像转换为目标图像，从而避免了可能导致轨迹偏差的反演过程。此外，我们设计了一种基于概念对齐引导的后验采样方法，以增强编辑一致性。实验结果表明，我们的免训练方法甚至取得了优于当前最先进基于训练模型的编辑性能。

In image editing, it is essential to incorporate a context image to convey the user’s precise requirements, such as subject appearance or image style. Existing training-based visual context-aware editing methods incur data collection effort and training cost. On the other hand, the training-free alternatives are typically established on diffusion inversion, which struggles with consistency and flexibility. In this work, we propose VicoEdit, a training-free and inversion-free method to inject the visual context into the pretrained text-prompted editing model. More specifically, VicoEdit directly transforms the source image into the target one based on the visual context, thereby eliminating the need for inversion that can lead to deviated trajectories. Moreover, we design a posterior sampling approach guided by concept alignment to enhance the editing consistency. Empirical results demonstrate that our training-free method achieves even better editing performance than the state-of-the-art training-based models.

评分理由: 该论文研究心电图（ECG）生物识别技术，使用1D Inception-v1模型和ArcFace损失函数进行身份识别，并在MIMIC和HEEDB数据集上进行外部验证。论文主题是生物医学信号处理和计算机视觉在医疗领域的应用，属于AI在科学（特别是生物信息学）领域的应用。所有关键词中，只有"AI for Science" OR “Bioinformatics” OR “Cheminformatics"与论文有间接关联（因为ECG生物识别属于生物医学AI应用），但论文未涉及大模型、深度学习技术原理创新或任何其他关键词的具体技术。因此，除最后一个关键词得5分外，其余关键词均得0分。

心电生物识别技术的研究主要基于小规模队列和短时间隔数据，其在大规模底库、外部域偏移及多年时间跨度下的识别性能尚不明确。本研究采用经ArcFace损失函数训练的Inception-v1一维卷积神经网络，在包含53,079名患者的164,440份12导联心电图的内部临床数据集上进行训练，并在源自MIMIC-IV-ECG和HEEDB的大规模外部队列中评估性能。研究采用统一的闭集留一验证协议，结合Rank@K与TAR@FAR评价指标，并进行了规模分析、时间压力测试、重排序及置信度分析。在一般可比条件下，系统在ASUGI-DB上达到Rank@1为0.9506，在MIMIC-GC上为0.8291，在HEEDB-GC上为0.6884。在固定底库规模的时间压力测试中，MIMIC数据集的Rank@1随年份跨度（1至5年）从0.7853降至0.6433，HEEDB数据集从0.6864降至0.5560。HEEDB的规模分析显示识别性能随底库规模扩大单调下降，而随患者个体检查次数增加有所恢复。在HEEDB-RR数据集上，后验重排序进一步提升了检索效果：AS-norm方法将Rank@1从基线0.7765提升至0.8005。研究表明，心电身份信息在外部验证的大规模闭集条件下仍具有可测量性，但其实际应用质量显著受域异质性、纵向漂移、底库规模及二级分数处理的影响。

ECG biometrics has been studied mainly on small cohorts and short inter-session intervals, leaving open how identification behaves under large galleries, external domain shift, and multi-year temporal gaps. We evaluated a 1D Inception-v1 model trained with ArcFace on an internal clinical corpus of 164,440 12-lead ECGs from 53,079 patients and tested it on larger cohorts derived from MIMIC-IV-ECG and HEEDB. The study used a unified closed-set leave-one-out protocol with Rank@K and TAR@FAR metrics, together with scale, temporal-stress, reranking, and confidence analyses. Under general comparability, the system achieved Rank@1 of 0.9506 on ASUGI-DB, 0.8291 on MIMIC-GC, and 0.6884 on HEEDB-GC. In the temporal stress test at constant gallery size, Rank@1 declined from 0.7853 to 0.6433 on MIMIC and from 0.6864 to 0.5560 on HEEDB from 1 to 5 years. Scale analysis on HEEDB showed monotonic degradation as gallery size increased and recovery as more examinations per patient became available. On HEEDB-RR, post-hoc reranking further improved retrieval, with AS-norm reaching Rank@1 = 0.8005 from a 0.7765 baseline. ECG identity information therefore remains measurable under externally validated large-scale closed-set conditions, but its operational quality is strongly affected by domain heterogeneity, longitudinal drift, gallery size, and second-stage score processing.

评分理由: 论文研究针对视觉语言模型（VLM）在自动驾驶场景中的后门攻击，属于大模型（特别是多模态大模型）在安全关键领域的应用研究。因此，仅与第一个关键词（“Large Language Models” OR “LLMs” OR “Foundation Models”）有一定关联，因为VLM可视为大模型的一个子类或扩展（尽管论文聚焦视觉语言模型，而非纯文本大模型）。论文核心是安全攻击方法（GLA），不涉及其他关键词所描述的大模型训练、优化、推理、对齐、代理、科学应用等具体技术原理或应用领域。

视觉语言模型正快速融入自动驾驶等安全关键系统，使其成为潜在后门攻击的重要攻击面。现有后门攻击主要依赖单模态、显性且易被检测的触发器，难以在自动驾驶场景中构建隐蔽且稳定的攻击通道。GLA引入了两种自然主义触发器：通过稳定扩散修复技术生成的涂鸦式视觉图案（可无缝融入城市场景），以及跨语言文本触发器（在保持语义一致性的同时引入分布偏移，以构建鲁棒的语言侧触发信号）。在DriveVLM上的实验表明，GLA仅需10%的中毒比例即可实现90%的攻击成功率与0%的误报率。更具隐蔽性的是，该后门不会削弱模型在干净任务上的性能，反而提升了BLEU-1等指标，使得传统基于性能下降的检测方法难以识别攻击。本研究揭示了自动驾驶视觉语言模型中未被充分认识的安全威胁，并为安全关键多模态系统的后门评估提供了新的攻击范式。

Visual language model (VLM) is rapidly being integrated into safety-critical systems such as autonomous driving, making it an important attack surface for potential backdoor attacks. Existing backdoor attacks mainly rely on unimodal, explicit, and easily detectable triggers, making it difficult to construct both covert and stable attack channels in autonomous driving scenarios. GLA introduces two naturalistic triggers: graffiti-based visual patterns generated via stable diffusion inpainting, which seamlessly blend into urban scenes, and cross-language text triggers, which introduce distributional shifts while maintaining semantic consistency to build robust language-side trigger signals. Experiments on DriveVLM show that GLA requires only a 10% poisoning ratio to achieve a 90% Attack Success Rate (ASR) and a 0% False Positive Rate (FPR). More insidiously, the backdoor does not weaken the model on clean tasks, but instead improves metrics such as BLEU-1, making it difficult for traditional performance-degradation-based detection methods to identify the attack. This study reveals underestimated security threats in self-driving VLMs and provides a new attack paradigm for backdoor evaluation in safety-critical multimodal systems.

评分理由: 论文研究扩散模型在图像/视频修复中的应用，属于计算机视觉领域而非大语言模型（LLM）范畴。与大多数关键词（如LLM、MoE、推理、对齐等）完全无关。仅与少数关键词有间接关联：1）“Pre-training”（使用预训练扩散模型）得5分；2）“Post-training”（涉及微调对比）得5分；3）“PEFT”（轻量级提示学习类似参数高效微调）得5分。其他关键词均不适用。

预训练扩散模型在统一修复任务中实现了显著进展，提升了感知质量与泛化能力。然而，基于扩散的修复方法主要依赖微调或Control-Net风格模块，以利用预训练扩散模型的先验知识。本研究表明，这些预训练扩散模型本身具备修复能力，该能力可通过直接在文本编码器输出端学习提示嵌入来解锁。有趣的是，这种能力难以通过文本提示或文本标记嵌入优化直接获取。此外，我们发现朴素的提示学习存在不稳定性，因为使用退化图像的前向加噪过程与反向采样轨迹存在错位。为解决此问题，我们在扩散桥框架内训练提示嵌入，该框架对齐了训练与推理动态，强制构建了从含噪退化状态到清晰图像的一致性去噪路径。基于这些发现，我们在预训练的WAN视频模型和FLUX图像模型中引入了轻量级可学习提示，将其转化为高性能修复模型。大量实验表明，我们的方法在多种退化类型上实现了具有竞争力的性能与泛化能力，同时避免了模型微调和修复专用控制模块的使用。

Pre-trained diffusion models have enabled significant advancements in All-in-One Restoration (AiOR), offering improved perceptual quality and generalization. However, diffusion-based restoration methods primarily rely on fine-tuning or Control-Net style modules to leverage the pre-trained diffusion model’s priors for AiOR. In this work, we show that these pre-trained diffusion models inherently possess restoration behavior, which can be unlocked by directly learning prompt embeddings at the output of the text encoder. Interestingly, this behavior is largely inaccessible through text prompts and text-token embedding optimization. Furthermore, we observe that naive prompt learning is unstable because the forward noising process using degraded images is misaligned with the reverse sampling trajectory. To resolve this, we train prompts within a diffusion bridge formulation that aligns training and inference dynamics, enforcing a coherent denoising path from noisy degraded states to clean images. Building on these insights, we introduce our lightweight learned prompts on the pre-trained WAN video model and FLUX image models, converting them into high-performing restoration models. Extensive experiments demonstrate that our approach achieves competitive performance and generalization across diverse degradations, while avoiding fine-tuning and restoration-specific control modules.

评分理由: 论文核心是视觉语言模型（VLM）的强化学习训练方法，与RLHF高度相关（10分）。论文涉及视觉推理任务，与Chain of Thought和System 2 Thinking相关（各8分）。论文使用基础模型Qwen3-VL-8B，属于大模型范畴（8分）。论文构建了包含科学任务的数据集，与AI for Science有一定关联（5分）。论文提到数据覆盖范围对RL扩展的重要性，与Scaling Laws AND Data Quality有一定关联（5分）。其他关键词如MoE、SLMs、PEFT、RAG等未在论文中涉及，得0分。

构建一个能够跨图表、科学、空间理解和开放式任务工作的视觉推理器需要什么？最强大的视觉语言模型（VLMs）表明，如此广泛的视觉推理能力是可以实现的，但其背后的方法仍不明确，被锁定在使用非公开数据的专有强化学习（RL）流程中。我们推出了Vero，一个完全开放的VLM系列，在多样化的视觉推理任务中达到或超越了现有的开放权重模型。我们在六大任务类别中扩展了RL数据和奖励机制，构建了Vero-600K——一个从59个数据集中提取的包含60万个样本的数据集，并设计了能够处理异构答案格式的任务路由奖励。Vero实现了最先进的性能，在我们包含30个挑战性基准测试的VeroEval套件上，相比四个基础模型平均提升了3.7-5.5分。以Qwen3-VL-8B-Instruct为基础，Vero在30个基准测试中的23个上超越了Qwen3-VL-8B-Thinking，且无需额外的专有思维数据。当从相同的基础模型开始训练时，Vero-600K在所有任务类别上都超越了现有的RL数据集。系统性的消融实验表明，不同的任务类别会引发性质各异的推理模式，这些模式在孤立情况下迁移效果不佳，这表明广泛的数据覆盖是推动RL有效扩展的主要驱动力。所有数据、代码和模型均已开源。

What does it take to build a visual reasoner that works across charts, science, spatial understanding, and open-ended tasks? The strongest vision-language models (VLMs) show such broad visual reasoning is within reach, but the recipe behind them remains unclear, locked behind proprietary reinforcement learning (RL) pipelines with non-public data. We introduce Vero, a family of fully open VLMs that matches or exceeds existing open-weight models across diverse visual reasoning tasks. We scale RL data and rewards across six broad task categories, constructing Vero-600K, a 600K-sample dataset from 59 datasets, and designing task-routed rewards that handle heterogeneous answer formats. Vero achieves state-of-the-art performance, improving over four base models by 3.7-5.5 points on average across VeroEval, our suite of 30 challenging benchmarks. Starting from Qwen3-VL-8B-Instruct, Vero outperforms Qwen3-VL-8B-Thinking on 23 of 30 benchmarks without additional proprietary thinking data. When trained from the same base model, Vero-600K exceeds existing RL datasets across task categories. Systematic ablations reveal that different task categories elicit qualitatively distinct reasoning patterns that transfer poorly in isolation, suggesting that broad data coverage is the primary driver of strong RL scaling. All data, code, and models are released.

评分理由: 论文研究大型推理模型中的早期停止方法，核心关注链式思维推理过程，因此与’Chain of Thought’高度相关（10分）。研究涉及深度推理过程分析，与’System 2 Thinking’相关（8分）。提出的CoDE-Stop方法旨在减少计算成本，与’Inference Acceleration’相关（8分）。论文在科学基准上评估，与’AI for Science’相关（8分）。方法涉及对中间答案置信度的分析，与’Self-Correction’和’Explainable AI’有一定关联（各5分）。论文明确针对大型推理模型，与’Large Language Models’高度相关（10分）。其他关键词如MoE、SLMs、训练方法、对齐、RAG、量化等与论文内容无直接关系，得0分。

大型推理模型依赖长链思维生成来解决复杂问题，但过长的推理过程通常会产生高昂的计算成本，甚至可能因过度思考而导致性能下降。一个关键挑战在于确定模型应在何时停止推理并给出最终答案。本研究分析了推理过程中中间答案的置信度，并观察到两种典型行为：正确的推理轨迹往往能较早达到高置信度答案，而错误的推理过程则倾向于产生冗长且低效的推理路径，其置信度动态也表现出较低的可信度。基于这些观察，我们提出CoDE-Stop（置信度动态早期停止）方法，该方法利用中间答案置信度的动态变化来决定何时终止推理，无需额外训练即可轻松集成到现有模型中。我们在多个模型的不同推理与科学基准测试上评估了CoDE-Stop方法。与先前的早期停止方法相比，它在准确性与计算效率之间取得了更优的平衡，相较于标准全长推理可减少25-50%的总令牌消耗。此外，我们还对推理过程中的置信度动态进行了深入分析，揭示了正确与错误推理轨迹中置信度变化的内在规律。

Large reasoning models rely on long chain-of-thought generation to solve complex problems, but extended reasoning often incurs substantial computational cost and can even degrade performance due to overthinking. A key challenge is determining when the model should stop reasoning and produce the final answer. In this work, we study the confidence of intermediate answers during reasoning and observe two characteristic behaviors: correct reasoning trajectories often reach high-confidence answers early, while incorrect rollouts tend to produce long, unproductive reasoning traces and exhibit less reliable confidence dynamics. Motivated by these observations, we propose CoDE-Stop (Confidence Dynamics Early Stop), an early stopping method that leverages the dynamics of intermediate answer confidence to decide when to terminate reasoning, requiring no additional training and easily integrating into existing models. We evaluate CoDE-Stop on diverse reasoning and science benchmarks across multiple models. Compared to prior early stopping methods, it achieves a more favorable accuracy-compute tradeoff and reduces total token usage by 25-50% compared to standard full-length reasoning. In addition, we provide analyses of confidence dynamics during reasoning, offering insights into how confidence changes in both correct and incorrect trajectories.

评分理由: 该论文研究AI聚合如何影响社会学习，使用DeGroot模型扩展，分析AI聚合器训练和反馈机制对群体信念的影响。所有评分关键词均涉及具体的大模型技术、训练方法、推理机制、应用领域或优化技术，而本文聚焦于宏观的AI聚合对社会学习过程的理论建模，不涉及任何具体的大模型架构、训练技术、推理方法或科学应用，因此所有关键词相关度均为0分。

人工智能（AI）通过将聚合输出转化为未来预测的训练数据，改变了社会学习过程。为研究此现象，我们扩展了DeGroot模型，引入一个AI聚合器：该聚合器以群体信念为训练数据，并将合成的信号反馈给个体。我们将学习差距定义为长期信念与有效基准之间的偏差，从而捕捉AI聚合如何影响学习。我们的核心结论揭示了更新速度存在一个阈值：当聚合器更新过快时，不存在任何具有正测度的训练权重集合能够在一大类环境中稳健地改善学习；而当更新足够慢时，此类权重是存在的。随后，我们比较了全局与局部架构。基于邻近数据或特定主题数据训练的局部聚合器能在所有环境中稳健提升学习效果。因此，用单一全局聚合器替代专业化的局部聚合器，至少会在状态的某一维度上导致学习效果恶化。

Artificial intelligence (AI) changes social learning when aggregated outputs become training data for future predictions. To study this, we extend the DeGroot model by introducing an AI aggregator that trains on population beliefs and feeds synthesized signals back to agents. We define the learning gap as the deviation of long-run beliefs from the efficient benchmark, allowing us to capture how AI aggregation affects learning. Our main result identifies a threshold in the speed of updating: when the aggregator updates too quickly, there is no positive-measure set of training weights that robustly improves learning across a broad class of environments, whereas such weights exist when updating is sufficiently slow. We then compare global and local architectures. Local aggregators trained on proximate or topic-specific data robustly improve learning in all environments. Consequently, replacing specialized local aggregators with a single global aggregator worsens learning in at least one dimension of the state.