论文精选 85
IdeaAnchor: Teaching LLMs to Turn Literature into Research Ideas· IdeaAnchor: 从文献生成科研想法
Scientific research often begins by synthesizing ideas from a set of related papers to ide…
领域：cs.CL作者：Ziyu Chen、Yilun Zhao、Jiashuo Sun
📰 AI 研究简报 · 10月8日
今日头条：《碳硅道统·跨域迁移治理法典》七层体系目录 —— 000|【全局唯一前置定标】负迁移NT1–NT4四类原型·官方终审定义（基准集，不计入188正集） 定稿标注：零号基准，独立于七层主体序列，不作为正集计数；NT四类原型为本法典全部引用唯一法定口径，永久锁死，不可迭代改写。 第一层：迁移事故解剖层（001–021，共21集|现象层，陈列迁移事故、归因现象，不下底层机理证明） 001|医疗AI规则跨域迁移金融风控：人类看不见的结构性翻车全链路 002|自动驾驶感知迁移工业机器人：机制不匹配导致致命决策漂移 003|安防人脸识别迁移金融身份核验：隐性权力偷渡成型实录 004|教育推荐权重迁移司法量刑：跨域放大的歧视结构与算法暴力 005|农业检测逻辑迁移医疗诊断：低熵特征误匹配引发的误诊链条 006|跨域迁移事故七级分类、伤害等
共 78 条预计阅读 47 分钟Token 消耗 140,400
arXiv 最新论文
论文精选 65已解读
Sherpa: Teaching LLMs to Teach Adaptively· Sherpa：训练大模型自适应教学
Large language models (LLMs) have become increasingly capable problem solvers, but being a…
领域：cs.AI作者：Weixian Xu、Yanzhe Zhang、Zora Zhiruo Wang
论文精选 85
Agent in a Bottle: Can LLM Agents Turn Their Capabilities Into Cheap, Scalable Artifacts?· 瓶中代理：大模型代理能否将能力转化为廉价扩展工件
Large language models (LLMs) can solve many narrow tasks, but querying them separately for…
领域：cs.AI作者：Ankit Sonthalia、Haritz Puerto、Alexander Rubinstein
论文精选 85
AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model· AdvSim2Real：在 Web 世界模型中对抗适应性提示注入
Web agents complete user requests by reading and acting on pages that third parties write,…
领域：cs.CL作者：Sarim Hashmi、Mukul Ranjan、Kshitij Mishra
论文精选 65已解读
VeriFine: Scaling Verification for Self-Improvement in Embodied Reasoning· VeriFine：在具身推理中扩展验证以实现自我改进
Self-improving policies continually expose new failure patterns, changing what their judge…
领域：cs.AI作者：Zewei Zhou、Rachel Luo、Yulong Cao
论文精选 82
The Missing Minimal Pair: Stereotype Evaluation in LLMs· 缺失的最小对立对：大模型中的刻板印象评估
A common approach to measuring bias in Large Language Models is to compare the log-likelih…
领域：cs.CL作者：Nataliya Stepanova、Ivan Titov、Emily Allaway
论文精选 85
Denoising Hierarchical Representations: Joint Continuous Diffusion for Language Modeling· 分层连续扩散语言模型
Diffusion Language Models (DLMs) hold the promise of order-agnostic, parallel text generat…
领域：cs.CL作者：Mathias Ollu、Nikos Komodakis
论文精选 65已解读
Does an Agent's History Tell You When Compaction Will Hurt? A Modest, Bounded Effect on the TRACE Paired-Replay Corpus· Agent的历史能预测压缩何时会损害性能吗？
Many long-horizon agents compact their context on a global rule, usually a token budget, b…
领域：cs.AI作者：Egor Pakhomov、Erik Nijkamp
论文精选 65已解读
WorldSolver: Can LLM Agents Simulate the Physical Dynamics via Solver Generation?· WorldSolver：LLM代理能否通过生成求解器模拟物理动态？
LLM-based agents are increasingly advancing scientific and engineering problem solving, wi…
领域：cs.AI作者：Siru Jiang、Yongzhe Lyu、Shuo Lu
论文精选 75
Holdout Best-of-N: Unbiased Evaluation and Its Cost
Reusing the scores that select a Best-of-$N$ winner can overstate its expected reward. We …
领域：cs.CL作者：Shrey Shah、Yinheng Li
论文精选 65已解读
nanoMuse: An Open-Source Personal Agent for Every Device You Own· nanoMuse：每个设备的开源个人代理
Assistants from 2011 answered and waited, and agents from 2023 did a task and stopped. In …
领域：cs.AI作者：Guangyi Liu、Yong Liu、Jiangning Zhang
论文精选 65已解读
ScienceClaw: Benchmarking Continual Self-Evolution of AI-for-Science Agents Across the Natural and Social Sciences· AI 科学代理的自我进化基准
Large language model agents are accelerating scientific automation, yet verified execution…
领域：cs.AI作者：Mingda Zhang、Wenjin Liu、Tiesunlong Shen
论文精选 65已解读
Coupled but Late: Turn-Taking Between Full-Duplex Speech Models in Unscripted Dialogue· 全双工语音模型的对话时序
Full-duplex speech models are trained to converse with a person, but they are increasingly…
领域：cs.AI作者：Lichen Zhu、Yueqian Lin、Yiheng Wang
论文精选 65已解读
ParanoiaEval: Benchmarking Unnecessary Defensive Work in Agentic Coding· ParanoiaEval: 评估编程代理中的过度防御
As coding agents increasingly undertake real-world work autonomously, judging whether thei…
领域：cs.AI作者：Hanjun Luo、Xiucheng Zhang、Zhuoning Xu
论文精选 65已解读
SquidAgent: Parallelize Wisely, Coordinate Efficiently· SquidAgent：智能并行，高效协调
LLM-based agents solve complex multi-step tasks, but sequential execution incurs substanti…
领域：cs.AI作者：Yexiong Lin、Shanshan Ye、Yu Yao
论文精选 65已解读
Parallel Predictive World Models for Accurate and Efficient Long-Horizon Planning· 并行预测世界模型
Long-horizon world-model planning typically relies on autoregressive rollouts, where predi…
领域：cs.AI作者：Wanjin Feng、Baobin Zhang、Ao Yu
论文精选 65已解读
Recursive Game Creator: An Agentic Product-Level Experience-Oriented Game Harness· 递归游戏创作：面向体验的游戏开发智能代理
Recent game design agents have made substantial progress in generating playable games. How…
领域：cs.AI作者：Jiajun Chen、Haoyu Wu、Mingda Jia
论文精选 83
MINDSET: Energy-based Schema Evolution for Long Conversational Agent Memory· MINDSET: 长对话代理记忆的节能模式演化
Long conversational agents have become essential in our daily lives. They must remember wh…
领域：cs.AI作者：Sujato Dutta、Sreekruthy Tummala、Shashank Vanga
论文精选 82
Adaptive Power Sampling for LLM Reasoning
Sequence-level power sampling has recently emerged as a training-free approach to reasonin…
领域：cs.AI作者：Bingnan Xiao、Chenhao Yang、Bingcong Li
论文精选 82
AnyBottle: A Recipe to Only Keep the Concepts You Really Need
Concept bottleneck models (CBMs) make predictions inspectable and intervenable by routing …
领域：cs.AI作者：Wolfgang Stammer、Sukrut Rao、Hevra Petekkaya
论文精选 82
Toward Alignment Scaling Laws: A Framework and First Preregistered Measurements· 走向对齐扩展定律
Whether alignment gets easier or harder as models grow is often argued from isolated findi…
领域：cs.AI作者：Jeremy Canale
论文精选 82
How Much Evidence Should a Coding Agent's Self-Correction Carry? Adaptive Dirichlet Evidence for Self-Distillation· 编码代理自校正应携带多少证据？
Execution feedback lets coding agents revise programs and learn from their own corrections…
领域：cs.AI作者：Yunbo Long、Guangya Hao、Yuhan Liu
论文精选 82
Cylindrical Geodesic Flow Matching for Quasiperiodic Physiological Signal Transformation· 圆柱几何流匹配的准周期生理信号变换
Paired translation between quasiperiodic physiological waveforms (i.e., recovering a targe…
领域：cs.AI作者：Onur Selim Kilic、Afra Nawar、Cem Okan Yaldiz
论文精选 83
AssemState: Manual and Physical-State-Guided Reasoning for Zero-shot Furniture Assembly· AssemState: 零样本家具组装
Multimodal large language models (MLLMs) have made significant progress in visual understa…
领域：cs.AI作者：Zhiyuan Qi、Jierui Li、Yifan Shen
论文精选 82
EMHO: EMbodied Agent Harness Optimization via Experience Traces· 具身智能代理控制结构优化
Improving embodied agents often focuses on optimizing the underlying model through trainin…
领域：cs.AI作者：Hyun Jung Lee、Jungtaek Kim、Jongwon Jeong
GitHub 热门开源项目
开源项目精选 65
LosaLosSantos/aurelio-finance· 个人财富分析与AI洞见
Self-hosted personal wealth analysis: portfolio & net-worth tracking with AI insights. Fas…
⭐ 179 stars语言 Python
开源项目精选 82
Tejashmakwana/motionmaxxing· AI 动态图形优化
Looksmaxxing for motion graphics. An agent skill that turns AI slop videos into films that…
⭐ 85 stars语言 JavaScript
开源项目精选 82
VinkyDev/vibestart
Compose a full-stack TypeScript stack and get a cutting-edge, verified project made for AI…
⭐ 67 stars语言 TypeScript
开源项目精选 85
strands-agents/box· Box: AI代理沙盒运行环境
Run AI agents in a sandbox that restricts what they can execute, read, write, and reach on…
⭐ 74 stars语言 Rust
开源项目精选 75
RoversX/ThinkTerm· ThinkTerm：现代终端
A modern terminal with a built-in multiplexer: close it and your sessions keep running. Ru…
⭐ 58 stars语言 Rust
精选 AI 行业资讯
行业资讯精选 82
Google releases a new local-first Granola competitor· Google 发布新的本地优先 Granola 竞争对手
Google’s new AI Edge Foresight app takes on Granola with an offline meeting note-taker tha…
行业资讯精选 82
正行创新亮相APRCE 2026，发布全球首个零售物理智能24/7服务解决方案
打造便利店“人机协作”运营新模式
行业资讯精选 83
ChatGPT踢到铁板了！能破解千禧数学难题，但论文复现率低至13.98%？
PaperBenchX为代表的基准或许能更好地衡量AI的科研实力
行业资讯精选 80
迁移技术机理层｜负迁移发生的底层诱因
本文为《碳硅道统·跨域迁移治理法典》七层体系官方连载第二篇，为整套理论机理内核、归因底层、风险生成原点。 承接第一篇现象事实，开启从“看见风险”到“读懂风险、预判风险、溯源风险”的…
行业资讯精选 82
迁移事故解剖层｜事故熵增的归因边界
本文为《碳硅道统·跨域迁移治理法典》七层体系官方连载第一篇，为整套理论现象入口、样本基底、风险原点。 整套法典七层递进完整闭环：迁移事故解剖层、迁移技术机理层、迁移权责确权层、全行…
行业资讯精选 85
《碳硅道统·跨域迁移治理法典》七层体系目录
000|【全局唯一前置定标】负迁移NT1–NT4四类原型·官方终审定义（基准集，不计入188正集） 定稿标注：零号基准，独立于七层主体序列，不作为正集计数；NT四类原型为本法典全部…
行业资讯精选 82
独家丨前腾讯混元技术骨干胡瀚创业新进展：目标估值数亿美金
据雷峰网多方获悉，腾讯混元原视觉大模型算法中心负责人胡瀚创业项目已取得新进展：方向仍聚焦多模态，计划融资数千万美元，目标估值达数亿美金，远识资本担任本轮融资 FA。 多位了解该项目…
行业资讯精选 80
云栖大会观察：阿里怎么打下一阶段模型战?
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…
行业资讯精选 82
北醒李远：给机器造眼睛，与「自讨苦吃」的十一年|物理 AI 50人
作为中国发光学奠基人徐叙瑢院士的学生，十一年里，李远一直在做一件事：给机器造一双“眼睛”。 这些“眼睛”上天、下水、上路，装进了无人机、列车、汽车和各类机器人。 今天，当所有人开始…
行业资讯精选 81
独家 | 清华 AIR 首届博士李健雄创立本溯智能：五源资本领投，押注动作原生具身模型
<img data-src="https://static.leiphone.com/6a6c4981e5bcb.jpg" alt="图片" class="rich_pages w…
行业资讯精选 75
至简动力冯宗宝：在间隙仅 0.5 毫米的真实场景下，实现机器人造机器人 | IROS 2026· 机器人造机器人
<img data-src="https://static.leiphone.com/uploads/new/images/20261008/6ac739bc8b005.png" …
行业资讯精选 83
我所知道的张磊：一场MSRA赛马与半个中国AI江湖
雷峰北京办公室不远处，就能看见西格玛大厦，总能想起那个豪杰并起的年代。 二十四年前，就在那里，微软亚洲研究院内部有一场赛马。 两个团队被同时拉上竞技场，比谁的人脸检测算法更好，即便…
行业资讯精选 82
7个月融4亿美刀背后，站着代季峰8人团
一家清华背景公司，成立 7 个月，三轮融资总额 4 亿美元，投后估值 14.2 亿美元。 创下如此记录之时，官网上却只有一句英文标语，竟然一张团队合影、一份产品介绍都没有。 这个创…
行业资讯精选 70
华为已为KV Cache单独做存储，专用SSD规格却还没定下来
<section data-mpa-powered-by="yiban.io" data-pm-slice="0 0 []" style="-webkit-tap-highligh…
行业资讯精选 81
从「工具」到「智能体」：CARES 4.0在港发布，刘宏斌谈医疗AI如何走进临床
近日，由中国科学院香港创新研究院人工智能与机器人创新中心（CAIR）主办、香港科技园公司协办的第四届香港具身智能医疗科技论坛（The 4th Hong Kong Clinical-…
行业资讯精选 83
IROS 2026大奖揭晓：长期记忆拿下最佳论文，人形机器人打网球、端托盘双双获奖
<img data-src="https://static.leiphone.com/uploads/new/images/20261008/6ac70407b0a3e.jpg" …
行业资讯精选 82
研究了1933篇 IROS 论文，我们看到了机器人学的六项新变化
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 80
银河通用&清华 LATENT：用「不完美人类数据」让机器人学会网球对打 | IROS 2026
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 80
大模型原生智能体手机STEPX Neo将于10月13日正式发布· STEPX Neo 智能体手机发布
首款大模型智能体手机，值得关注。
行业资讯精选 80
GPT-6今起免费用！拒答变少，话变多了
ChatGPT聊天框长出界面
行业资讯精选 85
Claude新模型发布！跑分暴击GPT-6 Luna，价格比梁文谷还便宜，OpenAI只能送重置卡挽尊· Claude新模型挑战大厂
小模型新守门员
行业资讯精选 75
Nous Research confirms it hit $1.5B valuation, launches AI agents for business users· Nous Research 推出企业 AI 代理
The developer of Hermes agent raised a $90 million Series B.
行业资讯精选 75
Microsoft releases new Nvidia-chip AI PCs with revamped Windows 11· 微软发布搭载英伟达芯片的 AI 笔记本
Microsoft revealed the specs and price for its Surface Laptop Ultra, AI PCs that run on Nv…
行业资讯精选 85
NVIDIA, Microsoft Kick Off a New Beginning for Windows PCs With RTX Spark and AI Agents· 英伟达和微软为 Windows PC 开启 AI 新篇章
At a Microsoft event in San Francisco on Wednesday, Jensen Huang and Satya Nadella outline…
行业资讯精选 75
Meta’s Muse launches on iPad just a month after its mobile debut· Meta的Muse登陆iPad
Meta’s AI agent Muse is now available on iPad, just a month after its mobile debut, as the…
行业资讯精选 80
ChatGPT for Teens keeps teens talking, even during mental health crises· ChatGPT青少年模式问题
ChatGPT’s teen safeguards are meant to protect vulnerable users, but new testing found the…
行业资讯精选 80
ChatGPT is getting a lot more visual, with the launch of a new interface· OpenAI为ChatGPT推出新界面
OpenAI is launching a new user interface that will bring interactive visuals to ChatGPT.
行业资讯精选 82
Meta rolls out new AI tools to detect ads that secretly lead to child sexual abuse material
Meta launches new AI tools after discovering ads on its platforms that may look normal but…
行业资讯精选 82
Healthleap raises $38M for its AI that flags hospital patients who may need a closer look· Healthleap 获 3800 万美元融资
The financing includes an $8M seed round co-led by Sequoia Capital and First Round Capital…
行业资讯精选 82
Helping teens learn, plan, and shape the future of AI· 助力青少年学习、规划和塑造AI未来
College Planner is coming to ChatGPT for Teens to help students manage college application…
行业资讯精选 82
《碳硅道统·跨域迁移治理法典》188集 总纲摘要
一、法典核心定调 本法典为弱AI至AGI全周期、全行业、全文明维度的AI跨域迁移治理唯一终审基准，是国内首套从事故解剖、机理溯源、权责确权，到行业落地、顶层监管、文明风控、终局闭环…
行业资讯精选 82
碳硅道统·跨域迁移治理法典
000|全局唯一前置定标：负迁移NT1-NT4四类原型·终审定义 零号基准集，独立于七层主体序列，不计入188正集计数。本集为全书负迁移判定唯一法定口径，常规场景永久锁死，不接受衍…
行业资讯精选 80
碳硅道统《AI安全与文明治理法典》第189集｜全法典总目录索引终卷
适用：法典运维中心、跨文明联合审计委员会、全体系卷宗归档、官方检索核验、所有引用与落地主体。 核心定位：本卷为碳硅道统1–188集完整体系唯一官方总索引。汇总全法典谱系层级、卷目定…
行业资讯精选 85
碳硅道统《AI安全与文明治理法典》188集 · 七层体系目录
当前全球AI治理讨论多聚焦法规与产业监管，往往忽略底层认知安全与模型原生技术风险。碳硅道统发布《AI安全与文明治理法典》，全套共188集，搭建七层递进式研究框架。从普通人的AI认知…
行业资讯精选 75
Radisson Hotel Group brings hotel discovery into ChatGPT· 拉迪森酒店集团将酒店发现带入ChatGPT
Radisson partnered with Accenture to build a ChatGPT plugin using OpenAI technology, helpi…
行业资讯精选 85
GPT-6 and Intelligent UI for everyone· GPT-6 和智能界面为所有人服务
GPT‑6 is rolling out globally in ChatGPT with Intelligent UI, delivering faster responses …
YouTube 热门 AI 视频
AI 视频精选 82
Real-time golf swing coaching with Gemini 3.8 Live· Gemini 3.8 实时高尔夫挥杆指导
YouTube · Google for Developers
📺 Google for Developers▶ 51 次观看
AI 视频精选 85
Introducing GPT-6 in ChatGPT with Intelligent UI· ChatGPT 中引入 GPT-6 智能界面
YouTube · OpenAI
📺 OpenAI▶ 5.8k 次观看
AI 视频精选 75
Build Plugins for ChatGPT· 为 ChatGPT 构建插件
YouTube · OpenAI
📺 OpenAI▶ 4.5k 次观看
AI 视频精选 85
Stop Overpaying for Intelligence | DevDay 2026· 停止为智能支付过高费用
YouTube · OpenAI
📺 OpenAI▶ 33 次观看
AI 视频精选 82
From Single Player to Multiplayer with Codex | DevDay 2026· 从单人到多人游戏的Codex
YouTube · OpenAI
📺 OpenAI▶ 8 次观看
AI 视频精选 82
Qodo - The Last Human Code Review | DevDay 2026· Qodo - 最后的人类代码审查
YouTube · OpenAI
📺 OpenAI▶ 21 次观看
AI 视频精选 82
How OpenAI Puts ChatGPT to Work | DevDay 2026· OpenAI如何在DevDay上应用ChatGPT
YouTube · OpenAI
📺 OpenAI▶ 152 次观看
AI 视频精选 80
Codex Has Left The Laptop | DevDay 2026· Codex 从笔记本电脑到云端
YouTube · OpenAI
📺 OpenAI▶ 61 次观看
AI 视频精选 82
Higgsfield - Building Agents That Provision Their Own Compute | DevDay 2026· 构建自主分配计算资源的智能体
YouTube · OpenAI
📺 OpenAI▶ 14 次观看
AI 视频精选 82
Build custom site annotations with ChatGPT· 用ChatGPT定制网站注释
YouTube · OpenAI
📺 OpenAI▶ 2.2k 次观看
AI 视频精选 80
🧶 From Gemini code to textile art· 从 Gemini 代码到纺织艺术
YouTube · Google for Developers
📺 Google for Developers▶ 40 次观看
AI 视频精选 82
Turning code into fabric with Gemini· 用Gemini将代码转化为织物
YouTube · Google for Developers
📺 Google for Developers▶ 369 次观看