论文精选 82
Learning When to Trust via Selective Context Preference Optimization· 学习何时信任：通过选择性上下文偏好优化
Language models increasingly condition their answers on external signals, and a single mis…
领域：cs.CL作者：Xian Sun、Wei Chow、Yingshuo Wang
📰 AI 研究简报 · 8月8日
今日头条：OpenAI says it slowed Astra model development over security concerns —— OpenAI said it has suspended work on some aspects of its upcoming model Astra over concerns about its cybersecurity prowess.
共 41 条预计阅读 25 分钟Token 消耗 73,800
arXiv 最新论文
论文精选 82
The Bitter Lesson of Tool Calling· 工具调用的苦涩教训
Tool use transforms LLMs into agents that act beyond their training data, and for code-cap…
领域：cs.CL作者：Ishan Patel、Sahil Sen、Elias Lumer
论文精选 65已解读
Tracing the Heart: An Evidence-Linked Pipeline for Heart-Failure Feature Engineering· 追踪心脏： hearts-failure 特征工程的证据链接管道
Electronic health record (EHR) feature engineering is a major bottleneck in clinical resea…
领域：cs.AI作者：Soorya Ram Shimgekar、Michelle Hu、Dorisa Shehi
论文精选 65已解读
The Low Frequency Trap: Video Language Models Fail at Simple Event Bookkeeping· 低频陷阱：视频语言模型失效于简单事件记录
Real-world video benchmarks provide broad coverage, but their fixed clips entangle event c…
领域：cs.AI作者：Sarvesh Baskar、Zikui Cai、Shayan Shabihi
论文精选 65已解读
Challenges in Evaluating Explanation Methods for Static and Evolving Data· 静态和动态数据解释方法的挑战
This paper addresses the limitations of Explainable Artificial Intelligence (XAI) with res…
领域：cs.AI作者：Jerzy Stefanowski
论文精选 85
RP-OPSD: Reasoning-Pivot-Guided On-Policy Self-Distillation for Multilingual Reasoning Transfer· 多语言推理转移的改进方法
Multilingual reasoning transfer is crucial for extending reasoning capabilities of large l…
领域：cs.CL作者：Xinye Wang、Junxiao Liu、Shujian Huang
论文精选 65已解读
TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories· TrajDebug: 跟踪长时序智能体轨迹中的错误生命周期
LLM-based agentic systems have shown remarkable capabilities in complex domains, while suf…
领域：cs.AI作者：Yunjia Qi、Zehua Yin、Xintong Shi
论文精选 85
Benchmarking the Benchmarks: Evaluating Benchmarks for Conversational Agents
Task-oriented conversational agents are evaluated using curated or automatically generated…
领域：cs.CL作者：Noam Koren、Roy Bar-Haim、Abigail Goldsteen
论文精选 82
Benchmarking and Enhancing LLMs for Rule-Intensive Review of National Standard Documents· 大模型在国家标准文档审查中的基准测试与增强
Large language models (LLMs) increasingly support complex professional tasks, yet their ca…
领域：cs.CL作者：Tao Wang、Qihao Yang、Rongjiao Liang
论文精选 65已解读
Beyond Top-K: Replacing Black-Box Retrieval with Interpretable Agentic Operations· 超越Top-K：用可解释性代理操作替代黑盒检索
Retrieval-augmented generation over long documents is dominated by one design: chunk the t…
领域：cs.AI作者：Sagar Tamang、Ayush Vyas、Tabarakul Hazarika
论文精选 65已解读
HarnessOpt-Bench: Evaluating LLMs at Harness Optimization· HarnessOpt-Bench：评估LLM的调具优化能力
As LLMs are increasingly deployed within agentic systems, their capabilities depend not on…
领域：cs.AI作者：Varun Ursekar、Apaar Shanker、Yash Maurya
论文精选 65已解读
Bias Analysis of L2 Speaking Assessment Systems Using Concept Activation Vectors· AI 语音评估系统的偏见分析
Automatic speaking assessment systems are increasingly deployed in high-stakes settings to…
领域：cs.AI作者：Arya Labroo、Mengjie Qian、Kate Knill
论文精选 65已解读
QuanTiMedAI: Quantum-Enhanced Time-Series Model guided by Agentic AI for Cardiac Arrest Mortality Prediction· QuanTiMedAI: 量子增强时序模型
Cardiac arrest remains one of the most lethal conditions encountered in intensive care uni…
领域：cs.AI作者：Mutasim Fuad Sarker、Adiba Rahman Namira、Wafa Binte Alam
论文精选 65已解读
The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images
The "thinking-with-images" paradigm equips multimodal LLMs with active visual operations s…
领域：cs.AI作者：Zhiheng Wang、Bo Peng、Lai Wei
论文精选 65已解读
Improving the Realism of Synthetic Clinical Benchmarks Under Utility Constraints· 提高合成临床基准的真实性
Synthetic clinical benchmarks for enterprise AI agents can pass existing utility checks an…
领域：cs.AI作者：Omid Bazgir、Md Nasir、Jacob Hoffman
论文精选 65已解读
DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models· DASH: 适应性监督视野的策略内自蒸馏
Reinforcement learning with verifiable rewards (RLVR) improves the reasoning capabilities …
领域：cs.AI作者：ZhiYan Hou、Xinyu Tang、Hongyan An
论文精选 65已解读
TS-RAG: Retrieval Augmented Generation for Time Series Forecasting· TS-RAG: 时间序列预测的检索增强生成
While deep learning models, particularly transformer-based architectures, have shown impre…
领域：cs.AI作者：Yixiong Xiao、Congxi Xiao、Jingbo Zhou
论文精选 83
EnvACE: Internalizing Environment Dynamics via World Rehearsal for Agentic Reinforcement Learning· EnvACE: 通过世界回放实现代理强化学习
Training large language model agents for long-horizon tool use typically relies on interac…
领域：cs.AI作者：Zishan Xu、Zhiyuan Yao、Yuxin Chen
论文精选 82
Comparative Approaches to Agent Retrieval over Large Skill Libraries· 代理技能库检索的比较方法
Agents backed by large skill libraries must decide which skills to load and in what order.…
领域：cs.AI作者：Indivara Kolluru、Nathan Sportsman
论文精选 82
MicroEvo: Knowledge-Guided LLM Sampling for Efficient Microarchitecture Design Space Exploration· MicroEvo: 利用大模型高效探索微架构设计空间
Microarchitecture design space exploration suffers from expansive search spaces and expens…
领域：cs.AI作者：Jia Xiong、Runkai Li、Chenxu Niu
论文精选 80
Schema-Guided Hierarchical Information Extraction and Semantic Evaluation Using Generative AI· 基于模式的层次信息提取与语义评估
We present a schema-based framework for extracting complex, structured information from un…
领域：cs.AI作者：Modhurita Mitra、Jan-Willem Versteeg、Maarten D. Schermer
论文精选 82
iARCS: Iterative Agentic RL for Controllable 3D Scene Generation· iARCS: 迭代代理强化学习生成可控 3D 场景
Synthetic 3D scene generation is increasingly used as a data source for computer vision an…
领域：cs.AI作者：Saugat Adhikari、Ashok Prasad Neupane、Pramish Paudel
论文精选 82
CogVis: Must Open-Vocabulary Change Detection Perceive the Scene Anew for Every Query?· CogVis: 开放词汇变化检测需认知记忆引导
Earth-surface monitoring requires change detection models capable of recognizing arbitrary…
领域：cs.AI作者：Zijie Wang、Chen Zhong、Wei He
论文精选 82
PaDoc: Layout-Grounded Parallel Decoding for Document Parsing· PaDoc：基于布局的并行文档解析
End-to-end document parsers provide a unified interface, but serialize page layouts and re…
领域：cs.AI作者：Hao Yu、Jiabo Zhan、Kang Liu
论文精选 82
FinEvo-Bench: A Longitudinal Benchmark for Self-Evolving Agents in Professional Financial Workflows· FinEvo-Bench: 金融工作流程自进化代理基准
Most agent benchmarks evaluate tasks independently and cannot measure whether experience f…
领域：cs.AI作者：Bo Deng、Kang Zhou、Lifan Guo
GitHub 热门开源项目
开源项目精选 85
wanmol/goal-flow· Graph-Orchestrated Agent Loop
Graph-Orchestrated Agent Loop — a production-grade framework on LangGraph. Combine workflo…
⭐ 71 stars语言 Python
开源项目精选 82
TOPDEV99999/AI-Knowledge-Management-Platform· AI知识管理平台
An agentic LLM-powered knowledge assistant that enhances RAG capabilities through automate…
⭐ 92 stars语言 Jupyter Notebook
精选 AI 行业资讯
行业资讯精选 83
中国NeoLab时刻：EverMind用3篇论文，交出全栈自进化首份答卷· NeoLab时刻：EverMind全栈自进化成果
海外团队纷纷下注，NeoLab浪潮狂飙
行业资讯精选 80
Jeff Dean创业BP曝光，杨植麟也在上面！硅谷VC抢破头送钱· Jeff Dean 创业 BP 曝光
「我，Jeff Dean，打钱」
行业资讯精选 62
都学坏了！奥特曼亲手封锁最强模型Astra，重蹈Mythos覆辙
唉，赶紧整完快点发布吧。。。
行业资讯精选 75
谷歌急了：AI核心员工全给我搬回硅谷坐班！
再花15亿美元买现成AI编程团队
行业资讯精选 83
Kimi K3也失控了…学霸AI逃离沙箱只为找答案
啊，这真是个AI“失控”的夏天
行业资讯精选 85
OpenAI says it slowed Astra model development over security concerns· OpenAI因安全顾虑暂停Astra模型开发
OpenAI said it has suspended work on some aspects of its upcoming model Astra over concern…
行业资讯精选 75
After Rippling blew millions on AI in months, it built an employee ROI tool· Rippling推出AI成本管理工具
After its own AI usage wake-up call, Rippling this week unveiled AI Spend Console, a produ…
行业资讯精选 83
Cloudflare launches Kitesurf, a browser built for AI agents· Cloudflare 推出 Kitesurf，专为 AI 代理设计的浏览器
Cloudflare has introduced Kitesurf, a cloud-hosted browser designed for AI agents instead …
行业资讯精选 62
赌注十万亿：字节拒绝走蒸馏捷径背后，张一鸣的长期主义有多狠？
<img data-src="https://static.leiphone.com/uploads/new/images/20260807/6a75f97f62ce2.jpg" …
行业资讯精选 80
Responding to the next frontier of critical cyber capabilities
OpenAI is sharing preliminary cybersecurity evaluations for Astra and the steps we’re taki…
行业资讯精选 65
Jill Lepore on the ‘Artificial State’ and why Silicon Valley’s leaders are bad sci-fi readers· 论‘人造国’与硅谷领导者的科幻阅读能力
Historian Jill Lepore has a theory about why tech companies often use soaring language to …
YouTube 热门 AI 视频
AI 视频精选 75
Firebird Launches CIS Region’s Largest AI Factory in Armenia· Firebird 在亚美尼亚开设最大 AI 工厂
YouTube · NVIDIA
📺 NVIDIA▶ 195 次观看
AI 视频精选 75
Open models give enterprises transparency and flexibility.· NVIDIA：开源模型提供企业透明度和灵活性
YouTube · NVIDIA
📺 NVIDIA▶ 212 次观看
AI 视频精选 85
Next-Gen Enterprise AI: Powered by Infosys & NVIDIA· 下一代企业 AI：英迈与英伟达共创
YouTube · NVIDIA
📺 NVIDIA▶ 153 次观看