论文精选 60已解读
Handover of In-Context Learning State Across Session Boundaries· 会话间上下文学习状态传递
This study investigates the methodological and theoretical properties of session handover …
领域：cs.AI作者：Masahiro Kato、Taka Kato
📰 AI 研究简报 · 8月18日
今日头条：π0引用的中国团队，又出手了：世界仿真器新作发布 —— 给机器人造一个更接近真实的“第二世界”
共 60 条预计阅读 36 分钟Token 消耗 108,000
arXiv 最新论文
论文精选 60已解读
Participatory Moral AI Is Not Neutral: The Invisible Hand of Developers· 参与式道德AI并非中立
As AI systems make more morally loaded decisions across society, one response has been mor…
领域：cs.AI作者：Taenyun Kim、Edyta Bogucka、Daniele Quercia
论文精选 60已解读
Split the Labor: Separating Evidence Interpretation from Decision Aggregation· 分离劳动：从多个来源中得出结论的系统设计
Systems that ask a language model to reach a conclusion from many sources usually concaten…
领域：cs.AI作者：Zhelun Wu
论文精选 60已解读
Twin: Playing an Unknown Game with a Test-Time Digital Twin· 测试时的数字双胞胎系统
We present a Test-time World-model Inference (Twin) system, in which a frontier coding age…
领域：cs.AI作者：Alexy Skoutnev、Kirill Acharya、Gaston Longhitano
论文精选 82
You Only Pass Once: Answering and Abstaining Together in a Single Forward Pass of a Frozen Language Model
A frozen language model on reasoning tasks has two coupled weaknesses: it under-uses evide…
领域：cs.CL作者：Ziyang Luo、Zhongyao Chu、Xinjie He
论文精选 82
Information Satisfaction: A Reader-Centered Axis for Summarization Evaluation· 信息满足：以读者为中心的总结评估方法
The majority of work on summarization evaluation focuses on general summary quality (e.g.,…
领域：cs.CL作者：Isabel Cachola、William Walden、Reno Kriz
论文精选 60已解读
Shift Aware Transfer Learning with Adaptive Dual-Encoder Fusion for PM Forecasting in Data-Limited Environments· 数据有限环境中 PM2.5 预测的迁移学习
Short-horizon forecasting of fine particulate matter (PM2.5) remains difficult when observ…
领域：cs.AI作者：Shahab Band、Hamed Mohammadi
论文精选 60已解读
SheetCompass: Hierarchical Relation Graphs for Agentic Spreadsheet Reasoning· SheetCompass：用于主动表格推理的层次关系图
Spreadsheets are widely used to organize, analyze, and manipulate semi-structured data, ye…
领域：cs.AI作者：Panjing He、Mingyue Cheng、Yucong Luo
论文精选 60已解读
Wyvern: An Agentic Framework for Generating Grounded Multimodal Reports· Wyvern: 生成有根据的多模态报告的代理框架
In the current artificial intelligence-driven innovation era, the pace of knowledge growth…
领域：cs.AI作者：Beatrice Alessandra Motetti、Emilien Guandalino、Daniele Jahier Pagliari
论文精选 60已解读
PACE-Bench: Benchmarking Physics Adaptation via Code Evolution in Dynamic Environments
Self-evolving agents improve future behavior from interaction experience, yet existing eva…
领域：cs.AI作者：Yuhao Zhan、Bingxiang He、Zecong Tang
论文精选 60已解读
The Dynamics of Intelligence Explosions· 智能爆炸的动力学
AI is increasingly being used to help with AI R&D. Under certain conditions this feedback …
领域：cs.AI作者：Toby Ord
论文精选 60已解读
Knowing When to Stop: Bayesian Optimal Stopping for LLM Evaluations· LLM 评估的自适应停止
LLM evaluations often use fixed sampling budgets, testing every item the same number of ti…
领域：cs.AI作者：Toby D. Pilditch
论文精选 60已解读
The Past and Future of AI Scientists· AI科学家的过去与未来
We present a survey of the past and future of AI Scientists: machines capable of automatin…
领域：cs.AI作者：Ross D. King
论文精选 60已解读
LLMs Don't Pay for the Jump· 大模型无法完成推理跃迁
Zahavy [2026] argues that Large Language Models, despite their capabilities in induction a…
领域：cs.AI作者：Paras Balani、Subhrakanta Panda
论文精选 82
Tripwire: Triggering Aligned Refusal via Statistically Certified Safety Neurons· 触发对齐拒绝：统计认证的安全神经元
Neuron- and path-level interventions offer the finest-grained route to defending large lan…
领域：cs.AI作者：Wei Zhao、Zhe Li、Peixin Zhang
论文精选 82
AgentRewind: Recoverable Execution for Long-Horizon LLM Agents
Many real-world tasks require LLM agents to interact with their environments over long exe…
领域：cs.AI作者：Yu Zhuang、Kefei Chen、Yitong Duan
论文精选 82
A Survey of Large Models in Sports· 体育领域的大模型综述
Sports have witnessed growing global enthusiasm in recent years, serving as a vital force …
领域：cs.CL作者：Yichen Xu、Jianzhe Ma、Chuhan Wang
论文精选 82
Wrong but Useful: Trajectory Value Beyond Answer Correctness in Multi-Agent Messages· 多代理消息中的错误轨迹价值
Multi-agent reasoning systems often use agreement, confidence, or automated scores to deci…
领域：cs.AI作者：Chih-Hsuan Yang、Anjir Ahmed Chowdhury、Cheng-Hau Yang
论文精选 82
Local and Global Regimes of Geometric Complexity in Language Model Representations· 语言模型表示的局部和全局几何复杂性
Intrinsic dimensionality (ID) is widely used to probe the representational complexity of l…
领域：cs.CL作者：Arwa Osman、Marco Baroni、Iuri Macocco
论文精选 85
Designing Sustainable Federated Learning as a Service using Neural Architecture Search· 设计可持续的联邦学习服务
The sustainability constraints of FLaaS consumers pose significant challenges to maintaini…
领域：cs.AI作者：Keya Patel、Sajib Mistry、Sheik Fattah
论文精选 82
Disentangled Shared Representations Improve Morpho-Transcriptomic Integration
Spatial transcriptomics (ST) enables the simultaneous profiling of gene expression and tis…
领域：cs.AI作者：Julian Ostermaier、Swann Ruyter、Reuben Dorent
论文精选 83
ScienceFlow: A long-horizon agent for ML research, scientific discovery and beyond· ScienceFlow：实现长期自主研究的 ML 代理
Enabling LLM agents to sustain productive, stable, and goal-aligned research over extended…
领域：cs.AI作者：Mingming Zhao、Jiqian Dong、Kangping Xu
论文精选 83
Clearing the Fog: Towards Installing and Refining Proactive Exploration Capabilities in LLM Agents· 安装和精炼LLM代理的主动探索能力
We study proactive exploration in LLM agents, i.e., the ability to explore an environment …
领域：cs.AI作者：Zhizhao Guan、Chen Huang、Ziming Liu
论文精选 83
Program-space Diffusion for Morphology-to-Transcriptomics Prediction
Spatial transcriptomics (ST) enables genome-wide gene expression profiling while preservin…
领域：cs.AI作者：Ruyter Swann、Dorent Reuben、Racoceanu Daniel
论文精选 85
Envs-FORGE: Frontier-Optimized Reward-Grounded Environment Synthesis for Agent RL· Envs-FORGE：面向前沿的强化环境合成
Reinforcement learning (RL) for terminal agents needs executable training environments wit…
领域：cs.CL作者：Xiaojun Wu、Cehao Yang、Honghao Liu
GitHub 热门开源项目
开源项目精选 80
bawadou/ai-data-extractor· AI 数据提取器
Free open-source extractor for AI coding assistant chat histories. Supports Claude Code, C…
⭐ 118 stars语言 Python
开源项目精选 85
memoket/memoket-kite· Memoket-kite：开源 AI 长期记忆引擎
An open-source, vector-free long-term memory engine for AI agents, achieving SOTA on LoCoM…
⭐ 80 stars语言 Python
开源项目精选 82
pgrundev/pgbot· pgbot - Postgres 智能库
pgbot - Postgres intelligence for AI agents & apps
⭐ 203 stars语言 Go
精选 AI 行业资讯
行业资讯精选 75
Warp’s new system is an out-of-the-box software factory for AI development· Warp推出AI开发软件工厂
On Tuesday, Warp introduced Warp Factories, a new infrastrructure system designed to make …
行业资讯精选 75
OpenAI launches a safer ChatGPT for teens — years after teens started using it· OpenAI 推出青少年版 ChatGPT
ChatGPT for Teens adds age-appropriate safety measures, parental controls, and learning to…
行业资讯精选 75
Perplexity’s free AI offer left it with millions more users in India· Airtel免费AI服务推动Perplexity印度用户增长
Perplexity's India revenue rose about 60% after the Airtel offer ended for new users, even…
行业资讯精选 82
Introducing ChatGPT for Teens: Built for learning, backed by protections· 引入面向青少年的 ChatGPT
ChatGPT for Teens helps teens learn, think critically, and use AI with confidence, with st…
行业资讯精选 83
Partnering with CodeAI to prepare the first AI generation
OpenAI and CodeAI are partnering to help students build AI literacy, think critically abou…
行业资讯精选 80
IJCAI-ECAI 2026 开幕：吴佳俊斩获「计算机与思想奖」，SMOTE 算法获首届时间检验奖
<img data-src="https://static.leiphone.com/uploads/new/images/20260818/6a84351fe908e.jpg" …
行业资讯精选 82
网易传媒发布”蜜蜂AI” ：从工具到伙伴，让AI更懂人
8月18日，网易传媒举办“蜜蜂AI媒体沟通会”
行业资讯精选 85
π0引用的中国团队，又出手了：世界仿真器新作发布
给机器人造一个更接近真实的“第二世界”
行业资讯精选 80
以GLM-5为例，探究九章智算云强化学习系统如何落地“训推一致”
过去几年，大模型竞争的主线很清晰：更大的模型、更多的数据和更强的GPU集群。 <span style="font-family: '微
行业资讯精选 75
豆包工作任务再升级，支持虚拟桌面操作Windows电脑
近日豆包发布产品更新，“工作任务”模式下，可通过全新的“豆包虚拟桌面”操作Windows电脑。 豆包虚拟桌面基于更通用的GUI（图形用户界面）能力，模型在没有MCP、API、插件和…
行业资讯精选 75
网易传媒发布"蜜蜂AI"：从工具到伙伴，让AI更懂人
8月18日，网易传媒举办“蜜蜂AI媒体沟通会”，首次向外界系统介绍面向人工智能时代打造的统一AI能力底座——“蜜蜂AI”。目前，蜜蜂AI已率先在年轻化内容社区产品网易小蜜蜂中落地，…
行业资讯精选 82
NPU不够，诚恒微为什么还给端侧AI芯片加了GPGPU？
如果只看今天端侧AI芯片的热门叙事，NPU几乎是绕不开的主角。 大模型从云端向手机、PC、机器人等终端下沉，芯片厂商不断刷新TOPS，围绕Transformer、增加专用加速能力。…
行业资讯精选 80
6个Agent组团Vibe Gaming：自己生成、试玩、修Bug
代码能跑≠游戏能玩
行业资讯精选 70
今日起，阿里“千问办公”接入企业微信
8月18日，阿里旗下Agent产品“千问办公”正式接入企业微信。此前，该产品已接入钉钉和飞书。至此，“千问办公”实现国内三大主流协同办公平台的全面支持。 “千问办公”产品界面 用户…
行业资讯精选 82
大厂押注AI办公，文库网盘正在从“通用”走向“专业
从十年前起，互联网大厂就一直想帮打工人提升效率。 在线文档、企业微信、打卡审批——工具不断变换，看似办公室里的工作方式跟着变了，但好像真正被释放的时间并没有减少。 直到十年后，大模…
行业资讯精选 80
650亿美元！IPO前夕，Anthropic营收底牌曝光· 650亿美元！IPO前夕，Anthropic营收底牌曝光
回头看，A社就这样完成了反超
行业资讯精选 85
阿里杀进Agent上下文战场：钉钉聊天、企业文档、工作数据终于要被Agent吃进去了
MyContext补上Agent的数据加工层
行业资讯精选 83
荣耀找一个新身份
荣耀上一次让全行业意外，是 2020 年从华为独立出来。 但这一次，是一部手机。 几天前，荣耀 CEO 李健在广州正式发布 Robot Phone，9.59 毫米厚的机身里，塞着一…
行业资讯精选 82
比起让大模型写 100 万字的小说，AI 大牛们更想用 AI 拿诺贝尔奖
<img data-src="https://static.leiphone.com/uploads/new/images/20260818/6a83d492ceb9f.jpg" …
行业资讯精选 82
阿里云扩建全球基础设施，韩国第三座数据中心上线
8月18日消息，阿里云在韩国的第三座数据中心正式上线，并同步推出Agentic AI相关服务，进一步满足当地市场持续增长的AI及云计算需求。近期，阿里云已先后在法国、日本、马来西亚…
行业资讯精选 85
AI Infra进入自进化时代！清华团队AI优化AI造就国产万亿Token工厂· AI Infra进入自进化时代！
ALL IN国产算力生态
行业资讯精选 75
OPPO唐凯：模型能力已经超过产业链的组织能力
近日， OPPO 软件工程事业部总裁唐凯出席 支付宝AI生态合作伙伴大会 。会上，唐凯表示，当前 <span style="font-family: 宋体; font-s
行业资讯精选 82
DeepSeek 为什么把缓存涨了 11 倍？长上下文需求暴涨，得交点「存储税」了
<img class="rich_pages wxw-img" data-ratio="0.55" data-s="300,640" data-type="jpeg" data-w…
行业资讯精选 80
Anthropic’s annualized revenue surges to $65B· Anthropic 年收入激增至 650 亿美元
The model maker added $18 billion in annualized revenue in two months.
行业资讯精选 75
AI automation startup Relay shuts down, staff joins Google’s Chrome team· AI 自动化初创公司 Relay 关闭，团队加入谷歌 Chrome
"We have some really ambitious plans to help you work with AI in Chrome to get things done…
行业资讯精选 82
Amazon, which started off selling books, is destroying rare texts to train AI
Rare books are incredibly valuable for training LLMs, since these models have already trai…
行业资讯精选 75
Groq raises $350M to fuel its pivot from AI chips to neocloud· Groq 获 3.5 亿美元融资转型新云计算
Groq raised $350 million at a $3.5 billion valuation as the former AI chipmaker pivots to …
行业资讯精选 81
Nvidia investing $1.5B in SoftBank data center developer behind OpenAI project· 英伟达投资 SoftBank 数据中心开发商
Nvidia's investment in SoftBank's data center developer will guarantee its chips power an …
行业资讯精选 65
How to tell if your AI platforms’ accounts have been hacked· 如何检测你的 AI 平台账户是否被黑
A guide on how to check if hackers have broken into your accounts on the most popular AI p…
YouTube 热门 AI 视频
AI 视频精选 82
Build a live translation broadcast app with the Gemini Live API and LiveKit· 使用 Gemini 和 LiveKit 构建实时翻译广播应用
YouTube · Google for Developers
📺 Google for Developers▶ 204 次观看
AI 视频精选 85
Hands on with Gemini 3.7 Flash· Gemini 3.7 Flash 实践
YouTube · Google for Developers
📺 Google for Developers▶ 718 次观看
AI 视频精选 75
How Base44 Uses GPT-5.6 to Build Apps With 20% Fewer Tokens· Base44 如何使用 GPT-5.6 构建应用
YouTube · OpenAI
📺 OpenAI▶ 752 次观看