论文精选 65已解读
OmniScientist: An Omni-Modal Omni-Discipline AI Scientist· OmniScientist：多模态多学科的AI科学家
Recent advances in foundation models have enabled AI scientists to automate increasingly c…
领域：cs.AI作者：Bobo Li、Hao Fei、Tianjie Ju
📰 AI 研究简报 · 8月15日
今日头条：阿里开源Qwen3.8，千问大模型全球下载超30亿次 —— 8月14日晚，阿里千问正式开源Qwen3.8系列模型，所有开发者、科研机构和企业均可自由下载、部署和使用。全新开源的Qwen3.8-27B，系原生多模态稠密（Dense）模型，仅270亿参数规模，整体水平便超越了Qwen3.7-Plus，在编程及办公的真实场景中表现出色。新模型响应速度快，完成质量高，部署便捷，量化后在“消费级”显卡上即可流畅运行，是最受全球AI社区期待的开源模型。Qwen3.8-27B采用宽松的Apache2.0协议开源，有望成为本地部署的主力模型。 此前，旗舰模型Qwen3.8-Max也已开源模型权重。公开数据显示，阿里已累计向全球AI社区开源460余个模型，Qwen模型在全球下载总量超30亿次，衍生模型数超30万个，稳居全球第一开源模型。
共 45 条预计阅读 27 分钟Token 消耗 81,000
arXiv 最新论文
论文精选 65已解读
QuoteBench: How Matched Scores Can Hide Command-Path Failures· QuoteBench: 匹配分数如何掩盖命令路径失败
LLM coding agents issue Bash commands through interfaces that may serialize, wrap, and rep…
领域：cs.AI作者：Shangao Li、Yao Zhang、Volker Tresp
论文精选 82
LittleLearner: Language Models Under Pedagogically Controlled Knowledge Exposure· LittleLearner: 语言模型在教学控制下的知识暴露
Modern language models are trained on heterogeneous web-scale text corpora. Consequently, …
领域：cs.CL作者：Fanfei Li、Jana Zeller、Manuel Prada-Corral
论文精选 83
SAEVerbalizer: Generating Explanations for Sparse Autoencoder Features via Representation Verbalization· SAEVerbalizer：通过表示口语化生成稀疏自动编码器特征解释
Sparse autoencoders (SAEs) are proposed to extract numerous features from large language m…
领域：cs.CL作者：Weihan Meng、Hongzhu Guo、Yi Jing
论文精选 85
DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training Data· DFM Mimir v1：使用合法数据训练的前沿性能10亿参数模型
Current large language model development relies on massive, often non-permissible datasets…
领域：cs.CL作者：Peter Schneider-Kamp、Jacob Nielsen、Gianluca Barmina
论文精选 83
Measuring Task-Agnostic Training Data Influence Across Language Model Pretraining· 跨语言模型预训练的无任务数据影响力测量
Measuring training data influence consistently across language model pretraining is challe…
领域：cs.CL作者：Yuto Nishida、Hirokazu Kiyomaru、Yusuke Oda
论文精选 65已解读
AlayaWorld: Interactive Long-Horizon World Modeling - Full Technical Report (v1.1)· AlayaWorld：交互式长时世界建模
This report presents an improved version of AlayaWorld. While the backbone architecture, c…
领域：cs.AI作者：AlayaWorld Team、Kaipeng Zhang、Chuanhao Li
论文精选 81
Toward a Gricean Retreat: Probing LLMs for Knowledge Boundaries and Referent Specificity· 探索 LLMs 的知识边界和指称具体性
When asked about entities outside their knowledge boundary, LLMs routinely fabricate plaus…
领域：cs.CL作者：Dananjay Srinivas、Saksham Khatwani、Maria Pacheco
论文精选 65已解读
MARC v1: An Open-Source Multi-Agent Framework for Clinical AI Reasoning and Coordination· MARC v1: 开源多智能体临床 AI 推理与协调框架
We present Multi-Agent Reasoning and Coordination (MARC), an open-source framework that re…
领域：cs.AI作者：Saisha Shetty、Satvik Tripathi、Austin Lin
论文精选 65已解读
A Unifying Perspective on Causal World Models: From Observations to Representations to Structure· 因果世界模型的统一视角
World Models (WM) are increasingly seen as a foundation for intelligent agents that can pr…
领域：cs.AI作者：Avinash Kori、Fabrizio Russo
论文精选 65已解读
Academic League of Artificial Intelligence - An Integrative Perspective of Teaching, Research, and Extension· 人工智能学术联盟——教学、研究与推广的整合视角
Academic leagues have become important mechanisms for promoting extracurricular education …
领域：cs.AI作者：Alison R. Panisson、Maria Eduarda W. M. Vianna、Italo Firmino da Silva
论文精选 65已解读
RAIL: An Automatic Classifier of the Artificial Intelligence Readiness Level· RAIL：自动分类AI技术成熟度
Assessing the maturity of artificial intelligence technologies is essential for investment…
领域：cs.AI作者：Juan Irving Vasquez、Juan Terven、Laura-Ivoone Garay-Jimenez
论文精选 65已解读
Enhancing Virtual Agents through SLMs and Edge-Computing: An Exploratory Evaluation of Think and Memory Processes· 通过SLMs和边缘计算增强虚拟代理
Embodied intelligent virtual agents are expected to operate as persistent, adaptive, and c…
领域：cs.AI作者：Aimilios Hadjiliasi、Louis Nisiotis
论文精选 65已解读
Beyond Final Scores: A Systematic Evaluation of Agents for Long-Horizon AI Research and Development· 超越最终得分：对长期AI研究开发代理的系统评估
Autonomous agents are increasingly capable of improving models, systems, and other technic…
领域：cs.AI作者：Yiwei Li、Wanli Yang、Hexiang Tan
论文精选 65已解读
Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings· 谁发言很重要：权威感知的多视角 RAG
Parliamentary proceedings are a primary record of democratic deliberation, yet their volum…
领域：cs.AI作者：Mirko Tritella、Riccardo Pozzi、Matteo Palmonari
论文精选 65已解读
Jointly Predicting Courses and Grades Using a Transformer-Based Model
Existing predictive models in learning analytics often treat student academic history as a…
领域：cs.AI作者：Paul Savala
论文精选 65已解读
TopoIntent: Compiling Security Intent into Executable, Compliance-Checked Network Topologies· TopoIntent：将安全意图编译为可执行的合规网络拓扑
Enterprise security topology design requires translating business intent, regulatory requi…
领域：cs.AI作者：Xiaokang Qu、Jianliang Ma、Zao Fan
论文精选 81
Rules or Character? Scaling Laws for AI Safety Design· 规则还是性格？AI 安全设计的规模化定律
Artificial Intelligence (AI) safety systems combine character shaping (e.g., Reinforcement…
领域：cs.AI作者：Satoshi Takahashi、Nobuji Kouno、Masaaki Komatsu
论文精选 85
LongEarth-R1: Benchmarking and Aligning Vision-Language Models for Long-Horizon Earth Observation Reasoning· LongEarth-R1：长时地球观测推理的多模态模型基准测试与对齐
Long-horizon Earth observation reasoning requires models to organize multi-stage geographi…
领域：cs.AI作者：Yupan Ding、Jing Xiao、Zhenyuan Zhang
论文精选 80
LLM-Guided Graph Generation for Structure-Based Local Improvement Methods· 大模型引导的图生成
Large neighborhood search normally selects a random subset of decision variables for itera…
领域：cs.AI作者：Hai Xia、Vaidyanathan Peruvemba Ramaswamy、Stefan Szeider
论文精选 83
StateBridge: Training-free Hidden-state Alignment for Latent Communication in LLM Multi-Agent Systems· StateBridge: 无需训练的隐藏状态对齐
Large language model based multi-agent systems usually communicate in text, i.e., using di…
领域：cs.AI作者：Yanwen Peng、Delvin Ce Zhang、Xi Wang
论文精选 83
NAS-Driven Hardware Accelerator Exploration for Edge AI and Quantization Effects on the Pareto Space· NAS驱动的边缘AI硬件加速器探索与量化效应
Edge AI deployment demands neural architectures that are simultaneously accurate, computat…
领域：cs.AI作者：Eleftherios Mylonas、Angelos Kouprizas、Michael Birbas
论文精选 83
Towards Context-Aware Clinical Motion Understanding in Daily Living at Home: Freezing of Gait Detection with Egocentric Vision· 家庭日常中的临床运动理解：第一人称视角检测步态冻结
Understanding motion in daily living requires context beyond kinematics, because similar i…
领域：cs.AI作者：Vayalet Stefanova、Diwas Lamsal、Margot Genbrugge
论文精选 82
Sovereign by necessity? Frontier AI export controls, cyber security, and the limits of national AI capability· 必要主权？前沿 AI 出口控制、网络安全与国家 AI 能力的局限
A small number of firms based in two states produce the most capable frontier AI models. T…
领域：cs.AI作者：Alan Woodward、Andrew Rogoyski
论文精选 83
vToken: Token-Level Virtualization for Reclaimable KV Caches· vToken：面向未来的 KV 缓存虚拟化技术
Large language model serving faces a critical memory bottleneck: the KV cache grows with s…
领域：cs.AI作者：Yuanhang Gao、Xiangrui Yang、Yuanfeng Chen
GitHub 热门开源项目
开源项目精选 65
evan-steinhilb/md2hd· MD2HD：Markdown 可视化工具
Markdown, mapped - the MD2HD CLI. Pull the repo, tell your agent to create an MD2HD map, v…
⭐ 91 stars语言 JavaScript
开源项目精选 75
pulseaiclub/phi· phi：pi 俱乐部的编程代理工具
a coding Agent from pi. ∞ providers, sub-agents, hashline edits, and a permission gate
⭐ 79 stars语言 Go
精选 AI 行业资讯
行业资讯精选 85
至知研究院提出大模型可解释性新路线：拆权重，数据成本不到1%
理解大模型，无需再训练一个替代网络
行业资讯精选 80
根治AI音乐通病！这家国产音乐模型正面挑战SUNO
音潮API限时免费开放
行业资讯精选 85
源神启动！一张消费级显卡跑“Opus级”Agent，Qwen3.8-27B多项榜单反超Claude
推理能力还能自定义
行业资讯精选 75
DeepSeek Harness插件一夜燃爆GitHub：长期记忆、电子宠物、4399小游戏全来了· DeepSeek插件在GitHub引起热潮
大家已经给DeepSeek装上了啥
行业资讯精选 80
对话郎咸朋：用机器人创业重做一次“百万智驾量产”
1.9万字全文实录
行业资讯精选 82
索塔无界：全球首家原生物理世界模型落地商超，具身智能迎来“索塔时刻”
在 具身智能仍 陷于“资本热、落地冷”之际，成立仅 <span style="text-
行业资讯精选 75
鹿明发布MOS2：全球首个双臂负载50kg轮臂式机器人，加速AI Worker进入产业现场
8月14日，鹿明机器人发布全新重载轮臂式具身智能机器人 Lumos MOS2 。 <span style="font-family: 'Arial
行业资讯精选 82
限量手办 + 实景体验，浙江人行NAVIAI2026WRC 福利提前曝光
8 月 19 日至 23 日，2026 世界机器人大会（WRC2026）将在北京亦庄亦创国际会展中心举办。浙江人形机器人创新中心将携旗下 NAVIAI 人形机器人亮相 C-105 …
行业资讯精选 82
Universitas Gadjah Mada, Indosat and NVIDIA Open Indonesia’s First University AI Center to Develop Local AI Talent· 印尼首所大学 AI 中心开幕
Indonesia is taking charge of its AI future. This week, the Ministry of Communication and …
行业资讯精选 80
Google will now allow users to remove visible watermark from its AI generations· 谷歌允许用户去除AI生成内容的可见水印
Turning off this setting won't affect invisible benchmarks used to identify an AI generate…
行业资讯精选 65
Does Mark Zuckerberg really believe AI is ‘for everyone’?· Meta的AI模型Glimmer：扎克伯格的普惠AI理念
Meta released Glimmer this week, an open-weight AI model anyone can download and run on th…
行业资讯精选 85
刚刚，Qwen3.8-27B 开源了！家用显卡也能跑· Qwen3.8-27B 开源，家用显卡也能跑
所有人都可免费下载部署及商用。
行业资讯精选 90
阿里开源Qwen3.8，千问大模型全球下载超30亿次
8月14日晚，阿里千问正式开源Qwen3.8系列模型，所有开发者、科研机构和企业均可自由下载、部署和使用。全新开源的Qwen3.8-27B，系原生多模态稠密（Dense）模型，仅2…
行业资讯精选 80
谷歌开始肢解DeepMind，数个团队被划归总部
布林亲自下场督战
行业资讯精选 65
Kog is going deeper to squeeze more inference out of GPUs· Kog 深挖 GPU 推理潜力
The idea that GPUs are poorly suited for agentic workflows may be a misconception, accordi…
行业资讯精选 75
Hyperscalers might regret embracing natural gas if new forecast proves correct· 大型科技公司可能因采用天然气而后悔
Natural gas prices could triple in some parts of the U.S., which could saddle hyperscalers…
行业资讯精选 83
Meta’s ‘open’ AI, and a $250M deal gone very wrong· Meta的‘开放’AI，和一个2.5亿美元的糟糕交易
Meta released Glimmer this week, an open-weight AI model anyone can download and run on th…
行业资讯精选 82
商汤大装置首提TPW，重新定义AI基础设施效能标尺
如何用有限的能源和成本，产出更多Token？这是Token经济时代，AI基础设施面临的新课题。 Token消耗指数级增长，算力需求持续攀升，随之带动的是AIDC用电量在全社会用电结…
YouTube 热门 AI 视频
本板块今日暂无内容