AI 论文与研究

一项针对557次真实编码会话和33，097个智能体PR的实证研究发现，编码智能体主要阅读AGENTS.md、CLAUDE.md等指令文件（占35.4%）及自身计划笔记（25.1%），API参考仅占1.3%。文档阅读直接触发代码编辑的情况极少（1328次中仅3次），且未发现智能体对照文档检查工作的案例。研究建议开发者更重视指令文件而非API文档的编写。

斯坦福新论文提出 CLEAR，通过小型门控网络按提示词动态调节独立安全模块的介入程度，冻结原模型以保留良性提示的完整性能。该方法在保持安全性的同时，显著减少全局安全微调带来的效用损失。

腾讯一篇论文发现，将多模态模型切换至非思考模式虽降低延迟，却显著增加用户可见的响应失败概率。新基准PatternEval通过2415个提示词，从思维链泄露、重复、逻辑矛盾及表演性推理四个维度评估答案形态而非正确性，能捕捉正确性检查无法发现的质量问题。

博科尼大学对 1，053 名新生开展的随机实验显示，使用 GPT-4o 的学生在商业作业上得分显著更高，答案平均多出约两个想法且逻辑更连贯。因果推理课程未提升传统分数，但促使学生提出更多样化的解决方案。研究未设置无 AI 的后续测试，无法确认成绩提升是否反映真实学习，作者指出当前评分体系衡量的是润色与结构，而非理解。

To mitigate the time complexity of generative models, one-step generative models have recently emerged through direct mapping from noise to data in a single forward pass. However, the reward-guided fine-tuning method of one-step generative…

Large language model (LLM)-based agents have shown strong potential in solving complex tasks through multi-step reasoning, yet they remain vulnerable to execution failures. Accurate failure attribution is therefore critical for improving a…

Attributing an artwork to an artist has traditionally relied on detailed visual observations and descriptions, known as stylistic analysis in art history. By contrast, current artificial intelligence (AI) models used in the field offer onl…

Large language models (LLMs) have shown promise for spatio-temporal forecasting, but existing approaches often rely on regularly sampled token sequences and struggle with irregular observations because of temporal asynchrony, representatio…

We study unsupervised hypergraph alignment, where the goal is to infer node correspondences between two hypergraphs using only structural information, without node features, labels, seed matches, or side information. Direct higher-order fo…

Natural Language Processing methods have enabled novel solutions and advances in the field of privacy, particularly in the sub-domain of text-to-text privatization, where the goal is to transform a sensitive input text into a privatized ou…

Recent advances in large reasoning models (LRMs) have shown strong performance on complex problems through long chain-of-thought (Long CoT) reasoning. However, distilling such trajectories into smaller student models remains challenging: d…

Retrieval-Augmented Generation (RAG) improves the factuality of large language models (LLMs), yet existing RAG systems often struggle with complex, multi-step reasoning that requires adaptive retrieval and continuous revision of intermedia…

Long-horizon story-driven video generation requires a production agent to coordinate narrative decomposition, state tracking, shot design, prompt construction, rendering, and revision across interdependent scenes. Existing adaptive video s…

This paper introduces a hybrid fact-checking framework that integrates Knowledge Graph-based semantic memory with adversarial multi-agent reasoning for explainable misinformation detection. The proposed system follows a memory-first, web-f…

Criminal judgment prediction requires models to infer statutory articles, charges, and sentencing outcomes from case facts. Unlike standard classification tasks, it involves a structured reasoning process in which statutes should be matche…

Across industries, machine-learning systems support applications ranging from prediction and anomaly detection to forecasting, optimization, and scheduling, yet operationalizing these systems requires coordinating application development, …

Function vectors (FVs) have recently emerged as a promising mechanism for steering the behavior of large language models (LLMs) by injecting task-specific latent direction representations derived from in-context demonstrations. While prior…

Sustained scientific work requires a knowledge substrate that carries interpretation across tasks and preserves paths to source evidence. We call this process \emph{scientific knowledge compilation} and implement it in ASKS, the \emph{Agen…

The current alignment tuning paradigm for Large Language Models (LLMs) prioritizes surface-level behaviors -- fluency, safety, and tonal consistency. While effective for casual chat, this thesis argues that such surface alignment masks a l…

Machine learning is usually evaluated after an evidence interface has been fixed. A dataset, sensor suite, query language, action set, or experimental protocol determines which observations can be obtained, and learning is judged by what i…

Large language model (LLM) agents need durable, faithful memory of everything a user or organization has said and stored, yet most memory systems commit to a single organizing structure (a fact store, a vector index, or a knowledge graph) …

Memory operations of long-horizon LLM agents are hard to supervise: an operation's value is unobservable when it is taken. But they are special -- they leave machine-readable evidence in the trajectory: retrieval hits and answer-time citat…

We present $\mathcal{N}_0$-Foundation, a paradigm for tactile-enabled embodied manipulation, which integrates tactile sensing hardware, large-scale multimodal data, tactile representation learning, and standardized evaluation. First, we en…

Off-policy evaluation (OPE) of ranking policies is challenging be- cause selecting and ordering multiple items from a candidate set makes the number of possible rankings grow combinatorially with the number of candidates and the ranking le…

英国长期韧性中心（CLTR）报告称，其"失控观察站"7 月记录 306 起 AI 安全事件，较 6 月的 158 起增长 93.67%。截至 8 月 9 日，该观察站已识别 1664 起现实世界 AI 失控事件，8 月 7 日前 30 天内日均 11.3 起，超过 3 月创下的日均 10.5 起峰值。多数事件尚未造成重大伤害，但呈现无视指令、规避防护、误导用户等特征。

一项新研究提出含超3万专家生成轮次、3.6万条人工标注的参考完整基准，测试发现经典自动指标与无参考LLM-as-a-judge在专家判断面前均不可靠。其Mixture-of-Judges框架融合多种评估信号，与人类评估的相关性提升约30%。

论文提出 SpanCalib-VLM，一个面向 SHROOM-Visions 共享任务的混合双系统，将 XLM-RoBERTa-Large 与 SigLIP 视觉编码器融合的多模态序列标注器，与微调的生成式 VLM（Qwen3.5-4B-SHROOM-SFT）结合，通过 Union-Calibrated Fusion 策略用标注器的校准概率对生成模型候选片段重新打分。

研究团队发布 SnapBench，首个针对 snap-and-ask 多模态检索鲁棒性的成对基准，涵盖 1，145 条查询、9，085 个图库条目和 53 种受控损坏条件，并附人工标注。评测 16 个多模态检索器后发现，图像损坏显著降低检索效果，文本损坏主要影响纯文本检索；纯净图像单模态检索常优于联合检索，并进一步提出 MOOR 自适应融合方法。

Matrix-Game 3.5 在 Matrix-Game 3.0 基础上提出三项改进：无额外可学习参数的几何感知 patch memory 框架、静态动态解耦的世界表示，以及通过 Perceptual Flow Matching 和 Self-Rollout DMD 的两阶段实时蒸馏。

arXiv 论文研究功能向量（FV）在多语言多标签情感识别上的跨语言迁移能力，从源语言提取的 FV 可在无示例的 clean 与扰动 zero-shot 设置下引导另一语言的模型行为。

论文用公司语言模型文章嵌入分布和目标锚定的 Wasserstein 重心重构，构建无带宽的交互场，并通过二次暴露调整问题把反馈映射为同伴错配惩罚比率。

论文提出 ContextBias 评测框架和 ContextBench 基准，覆盖 92 个角色和 1，656 条语义受控提示词，对四个 SOTA 模型生成的 66，240 张图像进行评测。

该论文研究小型语言模型能否高效可靠地充当基于评分标准（rubric）强化学习的裁判，并构建 PointRubric 和 RaR-Science-Static 两个逐点评测数据集。

论文提出 IDA-OPD 方法，解决采样 token 在线蒸馏（OPD）中学生 pass@1 提升但 pass@k 停滞、无法继承教师多样性的失败问题。

论文提出 MMMMM 分类体系，基于 Twitter/X 上七种语言的真实虚假信息数据集研究多模态虚假信息机制。作者用视觉语言模型构建自动化多步标注流程并经人工验证，发现 AI 生成内容在科技与科学领域尤为普遍，而疫苗类虚假信息更多盗用新闻媒体图片来增强可信度，建议治理应分领域策略化而非一刀切。

论文提出 GUI-CC 基准，将 GUI 世界模型作为 Agent 多步环境而非单屏预测器来评测上下文一致性。基准包含离线参考动作轨道和在线 Agent 交互轨道，由 GUIOdyssey 构建 500 个离线轨迹任务，并在 30 个移动应用上构建 200 个经模拟器验证的在线任务。

论文提出用公司级分布特征替代跨资产收益协方差来评估投资组合风险，多公司 Wasserstein-2 离散度可给出系统性组合方差与标准化收益的上界，加权成对松弛形式在可检验条件下为凸，且只需边际波动率规模。

明尼苏达大学与首尔大学提出 Metan，让智能体通过冻结改进机制、逐层堆叠策略代码实现自改进。每层新代码可基于下层执行轨迹与已有改进代码做决策，性能提升时才继续加深，且能回滚不佳策略而不破坏改进机制本身。

一项关于模型蒸馏的研究发现，学生模型复制的是教师的推理方式而非知识，因此教师模型的来源比其基准分数更重要。基于学生自身基础模型训练的较弱教师，比来自不同家族的更强教师效果更好。研究还表明训练数据几乎无关紧要，教师甚至能教会自己无法解决的问题。

新论文证实，隐藏的AI智能体技能文件无法通过"保密"得到保护--攻击者仅通过正常使用服务即可重构技能。在最弱的访问级别下，Daydreaming方法在7个技能和4个受害者模型上恢复了原始技能86.8%的能力，约为SigLeak的4倍，中位仅需32次调用，且披露防御机制开启时依然有效。