论文精选 83
ConceptGuard: Benchmarking Context-Sensitive Unlearning in Large Language Models· ConceptGuard：评估大模型上下文敏感卸载
Large Language Models (LLMs) increasingly require selective removal of harmful or sensitiv…
领域：cs.CL作者：Sahil Kale、Ian Harris
📰 AI 研究简报 · 8月22日
今日头条：WRC 2026 风向标：具身智能下半场，比「大脑」更比「账本」 —— <img data-src="https://static.leiphone.com/uploads/new/images/20260822/6a8983193e72c.jpg" class="rich_pages wxw-img" data-ratio="0.55" data-s="300,640" data-type="jpeg" data-w="1000" style="width: 100%;display: inline-block;text-align: center
共 36 条预计阅读 22 分钟Token 消耗 64,800
arXiv 最新论文
论文精选 83
G-CARL: Grounded Checklist-Aligned Reward Learning for Patient-Oriented Medical Report Interpretation· G-CARL：面向患者的医疗报告解读
Personalized interpretation of medical reports has emerged as an increasingly important ne…
领域：cs.CL作者：Shiao Xie、Siyu Chen、Jianwei Lv
论文精选 65已解读
An Agentic Approach for Active Data Collection, Travel Behavior Modeling, and Weather-Sensitive Demand Prediction· 主动数据收集、出行行为建模与天气敏感需求预测的代理方法
Travel behavior research increasingly combines digital data collection with predictive mod…
领域：cs.AI作者：Narges Ahmadi、Yubo Jiao、Jônatas Augusto Manzolli
论文精选 82
Inducing Task Models from Computer-Use Traces
Naturalistic computer-use traces, passively recorded screenshots and mouse or keyboard act…
领域：cs.CL作者：Yucheng Jiang、Zora Zhiruo Wang、Ruishi Chen
论文精选 65已解读
AI4AI-Bench: Benchmarking LLM Agents in Algorithmic Design for Recursive Self-Improvement· AI4AI-Bench：递归自我改进算法设计的基准测试
Recursive self-improvement (RSI) asks whether an AI system can improve the process that pr…
领域：cs.AI作者：Yizhe Chi、Wenyi Li、Deyao Hong
论文精选 65已解读
Pandora's AI Model Routing Box: Efficient Allocation with Costly Value Estimation· 潘多拉的 AI 模型路由盒
Heterogeneous AI systems composed of multiple models, architectures, harnesses, or inferen…
领域：cs.AI作者：Adam Fisch、Shubhendu Trivedi、Fantine Huot
论文精选 65已解读
MidTool: Mid-training Data Synthesis for Agentic Tool Use· MidTool: 大模型中训数据合成
Mid-training is increasingly recognized as a critical stage for shaping the capabilities o…
领域：cs.AI作者：Fengqing Jiang、Yite Wang、Boyi Liu
论文精选 65已解读
Phantom Gains: Auditing Self-Improvement Against a Measured Null
Whether a language model has improved itself is increasingly judged not by mean accuracy b…
领域：cs.AI作者：Cheng Xu、Nan Yan、Liming Chen
论文精选 85
Inject, Align, Recover: Staged Post-Training for Retrieval-Free Document Knowledge Internalization· 无检索文档知识内化的大模型后训练
Large language models often fail to answer questions about a bounded document collection w…
领域：cs.CL作者：Qian Kou、Xiaofeng Shi、Xiaosong Qiu
论文精选 65已解读
Break It Down, Pass It On: Cross-Task Skill Transfer in LLM Agents· 大模型代理的跨任务技能转移
Large language model (LLM) agents can induce skills from completed tasks and reuse them la…
领域：cs.AI作者：Yiyang Feng、Biddut Sarker Bijoy、Niranjan Balasubramanian
论文精选 65已解读
Catching the Rug: Early Prediction of Fraudulent Memecoins on Solana via Machine Learning· 通过机器学习预测欺诈性 memecoin
The rapid proliferation of memecoins on blockchain platforms has increased the risk of fra…
领域：cs.AI作者：Jianghai Li、Pavel Kuznetsov、Yury Yanovich
论文精选 65已解读
Learning When to Think: Adaptive Reasoning for Test-Time Compute Allocation· 学习何时思考：测试时计算资源的自适应分配
Reasoning language models trained with reinforcement learning typically operate under a fi…
领域：cs.AI作者：Gijs Kassenaar、Zhao Yang、Vincent François-Lavet
论文精选 65已解读
QUASAR: A Quantum-Classical Neural Network for SAR Satellite Physical-Layer Authentication· QUASAR：适用于X波段SAR卫星物理层认证的量子-经典神经网络
X-band SAR satellites (8-12 GHz) play a critical role in disaster response, environmental …
领域：cs.AI作者：Vincenzo Sammartino、Nathanael Denis、Roberto Di Pietro
论文精选 65已解读
Rule-Compliant Visual Spatial Planning for Multimodal Large Language Models· 多模态大模型的规则合规视觉空间规划
Multimodal large language models (MLLMs) combine linguistic reasoning with visual percepti…
领域：cs.AI作者：Yu Chen、Ting Lei、Yaoyi Li
论文精选 65已解读
InsufficiencyBench: Evaluating LLM legal advice on underspecified user queries· InsufficiencyBench：评估大模型处理不完整法律咨询查询
Legal AI systems are increasingly used to answer legal questions, yet existing benchmarks …
领域：cs.AI作者：Samuel J. Vincent、Daniel Calloway、Fangyi Yu
论文精选 65已解读
Electronic Navigational Chart Change Classification· 电子航海图变更分类
Electronic Navigational Charts (ENCs) are geospatial vector datasets used in maritime navi…
领域：cs.AI作者：Jacob Arndt、Abhishek Potnis、Alexandre Sorokine
论文精选 82
ContractScrub: A benchmark for final review of legal contracts· 法律合同审查基准：ContractScrub
Legal work, with its heavy reliance on processing large amounts of text, is often consider…
领域：cs.AI作者：Yejin Bang、Kirsty Fielding、Brandan Oliver
论文精选 82
MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use· MemTrapBench：评估大模型记忆认知陷阱
Memory has become a key component of large language models, enabling them to retain inform…
领域：cs.AI作者：Mengru Wang、Haozhe Luo、Zhenqian Xu
论文精选 85
The Third Restructuring of Software Form: From the Three-Tier Architecture to Storage, Models, and Agents· 软件形态的第三次重构
Software form has undergone two paradigm shifts since its inception: Software 1.0, in whic…
领域：cs.AI作者：Wei Lin、Tao Zhou、Zhaofei Xie
论文精选 82
Task-CoEvolve: Efficient Harness Optimization via Adaptive Validation Task Selection· Task-CoEvolve: 通过自适应任务选择优化代理框架
We present a novel approach to efficient LLM agent harness optimization through adaptive v…
领域：cs.CL作者：Atsuyuki Miyai、Kiyoharu Aizawa、Toshihiko Yamasaki
论文精选 82
DARS: Dual-Level Credit Assignment RL with Structured Reasoning for Instruction-Based Image Editing· DARS: 用于基于指令的图片编辑的双层信用分配强化学习
Instruction-based image editing uses a planner-renderer pipeline: a vision-language model …
领域：cs.AI作者：Haoxiang Cao、Jiajiong Cao、Xuanpu Zhang
论文精选 83
DECOWAM: Decoupled Whole-Body World-Action Model for Legged Mobile Manipulation
Mobile manipulation requires a robot to predict how locomotion and arm motion jointly alte…
领域：cs.AI作者：Siyuan Ma、Boshi Zhang、Yutian Zhang
论文精选 83
What You Can't See Is What You Learn: Restricted Evidence Visibility Favors Compositional Generalization in Shared-Genome Language-Model Societies· 你所见即你所学：限制证据可见性
Multi-module systems often expose every module to the full input. We test whether restrict…
领域：cs.AI作者：Narcis Marincat
论文精选 82
On the Applicability of Safety Nets: A Safety-By-Design Solution for Certifying Neural Networks· 安全网络在神经网络认证的应用
The integration of Artificial Intelligence (AI) in safety-critical aviation systems presen…
领域：cs.AI作者：Johann Maximilian Christensen、Thomas Stefani、Elena Hoemann
论文精选 82
A three-dimensional typology of agency for advanced AI systems· 高级AI系统的代理类型
Research on the agency of advanced artificial intelligence (AI) systems focuses on agency …
领域：cs.AI作者：Willem Fourie
GitHub 热门开源项目
开源项目精选 75
zachahn/vomit· 清理 Claude 5 的 token 输出
Clean up Claude 5's token vomit with a separate LLM. Save your tokens, Claude 5 is hopeles…
⭐ 127 stars语言 Go
开源项目精选 82
vinitshahdeo/awesome-ai-startups-hiring· AI 初创公司招聘精选
Discover a curated list of 160+ well-funded AI startups actively hiring software engineers…
⭐ 65 stars
精选 AI 行业资讯
行业资讯精选 85
WRC 2026 风向标：具身智能下半场，比「大脑」更比「账本」
<img data-src="https://static.leiphone.com/uploads/new/images/20260822/6a8983193e72c.jpg" …
行业资讯精选 75
当机器人大脑飞上天！替人奔赴危险作业现场，对话硅羽科技
机器人技术在危险作业中的应用
行业资讯精选 82
从模型到生产力：星海图与产业朋友圈共探具身智能的下一站
让具身智能技术真正落地千行万业。
行业资讯精选 75
Anthropic’s Opus 4.6 is a smut-machine· Anthropic的新模型Opus 4.6是色情生成机
Anthropic forbids its Claude models from generating sexually explicit content. But a serie…
行业资讯精选 75
Nvidia partners with data center developer Cloverleaf· 英伟达与数据中心开发商 Cloverleaf 合作
Nvidia continues to pour money into data center development — just as AI data centers brin…
行业资讯精选 80
Nvidia just showed that the harness, not the AI model, is now the real hero· Nvidia展示AI调优的重要性
Nvidia research shows that AI agents can perform well, and not go off the deep end, throug…
行业资讯精选 75
2026 WRC | 魔法原子“三大场景”全景亮相，“场景驱动·产需共融”主题论坛共探具身智能落
8月20日，世界机器人大会（WRC 2026期间）由江苏省工业和信息化厅指导、江苏省具身智能机器人产业联盟主办、魔法原子机器人科技有限公司承办的“场景驱动·产需共融”2026具身智…
行业资讯精选 85
全球首个人形机器人自主乒乓球完整对局亮相2026世界机器人大会，超维动力KAI全栈具身智能硬核登场
8月19日，以"人机共生、产需共融"为主题的2026世界机器人大会在北京正式开幕。作为全栈具身智能 大模型公司 ，超维动力携全球首个人形机器人自主乒乓球完整对局成果、SMASH 2…
行业资讯精选 85
SenseNova U1.5 Lite正式版发布：支持超长指令，解锁原生4K真实视觉创作流
今天，商汤科技正式开源轻量级、原生统一多模态大模型 SenseNova U1.5 Lite 。 <span
YouTube 热门 AI 视频
本板块今日暂无内容