论文精选 80
Towards Computational Provenance: Carrying Causal-State Evidence in Generated Text· 迈向计算谱系：生成文本携带因果状态证据
A language model's output does not by itself provide verifiable evidence about the interna…
领域：cs.CL作者：Benjamin Belay
📰 AI 研究简报 · 8月19日
今日头条：支付宝“晓政”赋能民政服务数字化升级，七夕婚姻登记智能服务正式上线 —— 七夕婚姻登记高峰将至，支付宝政务 AI 助手 “晓政”联合民政部民政通推出婚姻登记专项便民服务，依托 MCP 能力连接十省官方服务通道，为用户提供结婚登记预约、政策咨询、婚姻登记网点查询等服务。用户无需在多个页面反复查找，只需说出“预约结婚登记”等需求，即可快速获取对应服务及办事指引，让婚姻登记从“找服务”进一步走向“对话办”。 此次婚姻登记智能服务通过MCP接入民政通官方服务能力。针对登记条件、办理材料、办理流程、异地登记规则等高频问题，晓政还可提供AI问答和办事指引，并支持婚姻登记网点查询，帮助用户提前了解办理要求、减少重复查找。通过连接真实政务服务，晓政正推动AI从信息咨询进一步进入实际办事流程。 作为支付宝深耕数字政务领域的标准化解决方案，晓政自2025年9月正式上线以来，依托多融合大模型能力与成熟的智能服务架构，实现从“人找服务”到“服务找人、对话即
共 59 条预计阅读 35 分钟Token 消耗 106,200
arXiv 最新论文
论文精选 65已解读
What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models· 合规检测器读取什么？激活探针和防护模型的审计
Regulatory compliance monitoring in deployed language models is increasingly implemented a…
领域：cs.AI作者：Saisab Sadhu、Aadit Sengupta、Vinay Kumar Sankarapu
论文精选 85
Model Hypnosis: Strong control of AI via additive subliminal effects· 模型催眠：通过潜意识效果强控制AI
We demonstrate that AI models are broadly susceptible to a phenomenon we call model hypnos…
领域：cs.CL作者：Enric Boix-Adsera、Benedict Tessler
论文精选 65已解读
Policy Iteration with Human Feedback: Bringing Post-Training RL to In-context Learning· 带有人类反馈的策略迭代
Generative pretraining established reusable task representations; later work on language-b…
领域：cs.AI作者：Minh-Ha Nguyen、Cathy Shyr
论文精选 65已解读
Quipu: A Governed Bitemporal Knowledge Graph Store· Quipu：治理的双时间知识图谱存储
Agents now write knowledge graphs, but knowledge-graph stores still carry defaults set whe…
领域：cs.AI作者：Steve Brown
论文精选 65已解读
Cross-Sign Language Transfer Learning Using Domain Adaptation with Multi-scale Temporal Alignment· 跨手语的多尺度时间对齐域适应
Sign language serves as a vital means of communication for individuals with hearing impair…
领域：cs.AI作者：Keren Artiaga、Yang Li、Ercan Engin Kuruoglu
论文精选 65已解读
When Agents Coordinate: Measuring Coordination in Multi-Agent AI Coding· 多智能体 AI 编程时的协作测量
We study how teams of AI coding agents coordinate while solving programming tasks. Current…
领域：cs.AI作者：Giuseppe Destefanis、Tomaso Aste
论文精选 83
ClawGym II: Exploring Black-Box RL on Agent Harness· ClawGym II：黑盒RL在代理束上的探索
Agent harnesses have substantially improved performance on long-horizon tasks by coordinat…
领域：cs.CL作者：Huatong Song、Fei Bai、Ming Yang
论文精选 65已解读
GRIP: Grounded Reasoning via Information-Restricted Premises· GRIP：基于信息限制前提的推理
High-capacity encoders in retrieval-augmented generation (RAG) can let the query dominate …
领域：cs.AI作者：Lirui Teng
论文精选 85
LAVA: Logic-Aware Validation and Augmentation Framework for Large-Scale Financial Document Auditing· LAVA：用于大规模金融文档审计的逻辑感知验证和增强框架
Financial document validation in production, such as payroll auditing, tax compliance, and…
领域：cs.AI作者：Ruoqi Shu、Xuhui Wang、Isaac Wang
论文精选 82
Semantic Bandits: In-Context Exploration-Exploitation is Biased by Semantic Priors
Large language models (LLMs) are increasingly deployed as decision-making agents in settin…
领域：cs.CL作者：David Eric Austin、Kaheer Suleman、Jackie Chi Kit Cheung
论文精选 65已解读
FabriMAE I Trust Myself? Self-Evaluating VLA Action Generation with Markov Attention Entropy· FabriMAE：我信任自己？
Vision-Language-Action models (VLAs) integrate visual perception, language instruction, an…
领域：cs.AI作者：Aniri、Chen Yilin、Jinhe Bi
论文精选 82
Does the LM Head Create a Harmful Gradient Bottleneck? A Causal Test· 大模型语言头是否造成有害梯度瓶颈？因果测试
The language-model head maps a hidden state of width D to a vocabulary of size V, so its t…
领域：cs.CL作者：Anand Murugan
论文精选 65已解读
Chronocooked: A Benchmark for Implicit Interval Timing in Reinforcement Learning Agents· Chronocooked：RL代理的隐式时间间隔计时基准
This paper presents Chronocooked, a reinforcement learning (RL) benchmark suite for studyi…
领域：cs.AI作者：Amrapali Pednekar、Alvaro Garrido-Perez、Yara Khaluf
论文精选 65已解读
Reconstruction: A Blind Benchmark for Recovering Research Ideas from Pre-Publication Bibliographies· 重建：从预出版参考文献恢复研究思想的盲基准
Can a language model recover the true research idea of a published paper when given only t…
领域：cs.AI作者：Shaolong Chen、Yanlin Fei、Nazhou Liu
论文精选 65已解读
PDDLCoder: Agentic PDDL Generation for LLM-Assisted Symbolic Planning· PDDLCoder: 用于 LLM 辅助符号规划的 PDDL 生成
LLMs remain unreliable for long-horizon planning, often generating logically inconsistent …
领域：cs.AI作者：Veit Laule、Jiangtao Shuai、Manfred Hauswirth
论文精选 65已解读
Hypergraph-based Multimodal Retrieval-Augmented Generation with Incremental Refinement· 基于超图的多模态检索与生成框架
Modern Multimodal Retrieval-Augmented Generation (M-RAG) systems are fundamentally limited…
领域：cs.AI作者：Shenao Chen、Yidan Xu、Xiangmin Han
论文精选 83
Cost Scales with Change, Not Corpus Size: Incrementally Maintaining an Evolving Semantic Substrate· 增量维护语义底座的成本分析
Retrieval-augmented and agentic question-answering systems increasingly re-derive the mean…
领域：cs.AI作者：Yusuke Takahashi、Kyle Wild、Asako Uraki
论文精选 82
CACSurv: Concordance-Aligned Comparative Learning with Large Language Models for Cancer Survival Prediction· CACSurv: 大语言模型用于癌症生存预测
Cancer survival prediction supports treatment planning, risk stratification, and follow-up…
领域：cs.AI作者：Tianqi Xiang、Qixiang Zhang、Xinpeng Ding
论文精选 82
Physics of Agents: Statistical Mechanics Predicts Collective Behavior of AI Agents· AI代理的物理：统计力学预测集体行为
AI agents increasingly operate as part of interacting systems rather than in isolation. As…
领域：cs.AI作者：Batu El、Jinhee Paeng、Fatih Dinc
论文精选 82
Probabilistic Circuits as Reasoning Machines in Artificial Intelligence (Part I)· 人工智能中的概率电路（上）
This cumulative habilitation thesis studies probabilistic circuits (PCs) as a powerful and…
领域：cs.AI作者：Robert Peharz
论文精选 83
CUBICS: Situation-aware performance estimation for safety-relevant ML components· CUBICS: 安全相关 ML 组件的情境感知性能评估
Machine learning (ML) is a key technology driving innovation today, but ensuring ML safety…
领域：cs.AI作者：Benjamin Herd、Jessica Kelly、Mario Trapp
论文精选 82
DeepInsight II: One Trace from Benchmark to Robot· 从基准测试到机器人的深度洞察
Across a Physical AI stack, evaluation maturity is inversely aligned with deployment risk:…
领域：cs.AI作者：Siyi Li、Yuchen Kang、Wuliang Wang
论文精选 83
Large language models as synthetic clinical experts to inform longitudinal rare-disease modeling· 大模型作为合成临床专家
Due to the limited amount of information, modeling longitudinal rare-disease data can bene…
领域：cs.AI作者：Clemens Schächter、Astrid Pechmann、Janbernd Kirschner
GitHub 热门开源项目
开源项目精选 82
fromleda/text-humanizer· 文本人化器
text-humanizer is an open-source project designed to convert text generated by AI to its "…
⭐ 132 stars语言 Python
精选 AI 行业资讯
行业资讯精选 83
AI isn’t close to curing cancer. This startup says it knows what it will take.· AI 能否治愈癌症？这家初创公司有答案
It's the data, stupid.
行业资讯精选 82
章鱼动力亮相WRC 2026，携“脑-手-数据”技术体系探索具身智能未来范式
章鱼动力展示新型具身智能技术
行业资讯精选 82
挑战 1 比特！ETH Zürich 秦浩桐：如何把大模型「塞进」小设备？| IJCAI 2026
<img data-src="https://static.leiphone.com/uploads/new/images/20260819/6a8584cde95e4.jpg" …
行业资讯精选 65
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
行业资讯精选 85
支付宝“晓政”赋能民政服务数字化升级，七夕婚姻登记智能服务正式上线
七夕婚姻登记高峰将至，支付宝政务 AI 助手 “晓政”联合民政部民政通推出婚姻登记专项便民服务，依托 MCP 能力连接十省官方服务通道，为用户提供结婚登记预约、政策咨询、婚姻登记网…
行业资讯精选 81
百度袁佛玉：从模型进化、数据回流到应用接管，具身智能进入场景落地关键阶段
具身智能产业关注的重点正从“能力涌现”进入“生产力兑现”阶段，拼的不只是单一模型能力。8 月 19 日下午，在北京举行的 2026 世界机器人大会上，百度集团副总裁袁佛玉表示，机器…
行业资讯精选 75
IDC发布2026中国AI50强：360以“智能体+安全”双轮驱动入选
凭借企业级智能体与AI安全的全栈布局，360成为中国人工智能产业发展的代表企业之一。
行业资讯精选 65
MiniMax核心工程负责人阿岛离职
从技术研发到开发者沟通，长期活跃在开发者一线
行业资讯精选 82
具身数据底座开卖，首发5100元：机器人训练数据有了新解法
构建全栈物理AI基础设施
行业资讯精选 75
用DeepSeek网页版就能瓜分鹅厂600万？？！· DeepSeek网页版助力夺冠
冠军姿势长这样
行业资讯精选 82
华尔街实测8款全球主流Agent：千问办公综合排名第一，超Claude Cowork、Codex
8月19日消息，华尔街投行杰富瑞分析师近日对八款全球主流AI Agent 进行了真实办公任务实测，结果显示，通过模型和Harness的协同，阿里千问办公综合得分排名第一，超过美国的…
行业资讯精选 82
写2000字提示词，不如先生成3D白模！AI视频创作进入“预演时代”
AI终于能听严格执行运镜需求
行业资讯精选 85
AI for Science开始“动手”了：机器人正式走进国家级实验室
源络科技，推动AI for Science走进实验室3.0
行业资讯精选 82
英伟达 Cosmos 3 深度拆解：它想做具身智能时代的「安卓系统」| RSS 2026
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 82
RSS 2026 实录：当 AGI 撞上「物理之墙」，想象力成为具身智能的第一生产力· RSS 2026 实录：AGI 与具身智能
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…
行业资讯精选 85
RSS 2026现场直击｜从双臂协同到水下灵巧，机器人操作迈入「少样本时代」
<img class="rich_pages wxw-img" data-aistatus="1" data-backh="300" data-backw="546" data-c…
行业资讯精选 85
李飞飞高徒黄文龙：具身智能需要一次「脑内搜索」革命 | RSS 2026
<img data-src="https://static.leiphone.com/uploads/new/images/20260819/6a8548ba8266f.jpg" …
行业资讯精选 85
纳须弥，于芥子：大模型之后，中国厂商正在把大模型塞进复杂系统 | IJCAI 2026 前瞻
<img data-src="https://static.leiphone.com/uploads/new/images/20260819/6a852aa732da2.jpg" …
行业资讯精选 80
Momenta、XHEART与QNX宣布达成合作，打造全球安全标准自动驾驶平台
近日，Momenta（香港联交所： 6880.HK ）与新芯航途（XHEART）、BlackBerry有限公司（纽交所：BB，多伦多证交所：BB）旗下QNX业务部门联合宣布，将共同…
行业资讯精选 75
支付宝“阿宝”跨端服务落地16家车企 新增60余家定点合作
8月17日，国内首届智能体商业生态合作伙伴大会于杭州正式举办。支付宝携手比亚迪、吉利汽车、理想、蔚来、东风汽车、深蓝汽车、上汽乘用车、一汽奔腾等主流整车企业，联动千问、华为、OPP…
行业资讯精选 75
Cursor capitalizes on GitHub frustration, launches rival hosting platform· Cursor 推出新代码托管平台挑战 GitHub
Cursor, known for its AI Code Editor, is launching a new code-hosting platform to rival de…
行业资讯精选 65
ChatGPT Ads expands across Europe· ChatGPT 广告扩展至欧洲
ChatGPT Ads is expanding to 31 European markets. Learn how advertisers can reach people as…
行业资讯精选 75
Strengthening Democratic Oversight in National Security· 加强国家安全中的民主监督
OpenAI launches an initiative to strengthen democratic oversight of AI in national securit…
行业资讯精选 82
OpenAI institutes new safeguards after Hugging Face breach· OpenAI 新增监控和安全措施
The new safeguards include more detailed monitoring of models during the development proce…
行业资讯精选 80
Etched’s valuation doubles to $21B in a month· Etched 估值一个月翻倍至 210 亿美元
Jane Street has installed Etched's first shipped AI cluster system, and was so impressed, …
行业资讯精选 62
Why Apple’s camera-equipped AirPods may not be the ‘pervert pods’ consumers fear· 苹果带摄像头的 AirPods 不会成为'偷窥豆'
Apple’s leaked camera-equipped AirPods might avoid the privacy pitfalls of other AI wearab…
行业资讯精选 75
百度二季度AI业务收入占比再过半 华尔街两大投资基金大幅加仓
8月18日，百度发布最新季度财报，显示总营收313亿元；百度一般性业务收入252亿元，其中AI业务收入占比50%，连续两季度过半。同期，华尔街两大投资基金大幅建仓、增持百度。 据1…
行业资讯精选 85
Pacing model development in an era of cyber-critical capabilities· 加强 AI 模型的开发节奏
OpenAI is strengthening monitoring, alignment, and security for frontier AI models. See ho…
行业资讯精选 83
Asana cleared 5 years of engineering work in 2 weeks with Codex· Asana用Codex两周完成五年工程任务
Asana used OpenAI Codex to replace an outdated testing system in two weeks, completing wor…
行业资讯精选 80
How NVIDIA scales expertise with ChatGPT Work· 英伟达如何用ChatGPT提升专业能力
NVIDIA teams use ChatGPT Work to reduce manual tasks, connect fast-moving signals, and sca…
YouTube 热门 AI 视频
AI 视频精选 85
RTX Spark· RTX Spark
YouTube · NVIDIA
📺 NVIDIA▶ 635 次观看
AI 视频精选 75
How to Turn a Business Question Into a Strategy Deck With ChatGPT Work | Tutorial· 用 ChatGPT 制作商业策略幻灯片教程
YouTube · OpenAI
📺 OpenAI▶ 490 次观看
AI 视频精选 70
How to Create a Solid Blog Draft With ChatGPT Work | Tutorial· 使用 ChatGPT 创作优质博客草稿
YouTube · OpenAI
📺 OpenAI▶ 987 次观看
AI 视频精选 75
How to Prepare for Customer Meetings With ChatGPT Work | Tutorial· 使用 ChatGPT 准备客户会议
YouTube · OpenAI
📺 OpenAI▶ 735 次观看