论文精选 65已解读
ReWorld: An Interactive World Model with Long-Horizon Memory· 具有长时记忆的交互式世界模型
An interactive world model must follow the user's actions, remember the places it has show…
领域：cs.AI作者：Zhifei Chen、Luozhou Wang、Guibao Shen
📰 AI 研究简报 · 8月26日
今日头条：DeepSeek’s AI Just Learned To Upgrade Itself —— YouTube · Two Minute Papers
共 69 条预计阅读 41 分钟Token 消耗 124,200
arXiv 最新论文
论文精选 82
SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?· SWE Refactor Bench：代码代理能否完成全库迁移？
Modern software systems accumulate technical debt over decades of development, which makes…
领域：cs.CL作者：Deyao Hong、Yizhe Chi、Wenyi Li
论文精选 65已解读
Prime Agent: A Self-Improving RLM Harness· Prime Agent：自我改进的 RLM 框架
Language models are sequential processors, but long-horizon agency requires external infor…
领域：cs.AI作者：Seth Karten、Alex L. Zhang、Kevin Thomas
论文精选 83
ConvergeFlow: Language Flow with Provable Convergence to Token Embeddings· ConvergeFlow: 词嵌入收敛的语言流
Recent advances in continuous diffusion and flow-based language models (LMs) have achieved…
领域：cs.CL作者：Na Li、Yuchen Jiao、Changxiao Cai
论文精选 65已解读
How AI Assistance Affects Human Skill Development: A Study of Learning with Logic Puzzles· AI 辅助对人类技能发展的影响
While AI assistance can improve human task performance in the short term, it may also unde…
领域：cs.AI作者：Shang Wu、Catarina G Belem、Shuyuan Fu
论文精选 65已解读
Correcting a learned physical invariant improves world-model rollouts
World models can predict video without learning dynamics that they reliably preserve. We t…
领域：cs.AI作者：Richard Bao
论文精选 65已解读
EarthVerse: Benchmarking Scientific Agents Across Dynamic Earth Systems and Natural Hazards· EarthVerse：跨动态地球系统和自然灾害的科学智能体基准测试
Earth-system analysis reconstructs changing physical processes from observations that diff…
领域：cs.AI作者：Zhiqing Cui、Xinxiang Yin、Yihong Tang
论文精选 80
When Names Cross Scripts: A Source-Grounded Benchmark for Historical Entity Reconciliation in the Mongol World· 多语言历史人物身份匹配基准测试
Historical people may appear under different languages, scripts, and transcription traditi…
领域：cs.CL作者：Xiang Chen、Zeyu Zhang
论文精选 65已解读
Mitigating Reasoning-Induced Misalignment via Safety-Direction Penalty· 通过安全方向惩罚缓解推理引起的对齐问题
Reasoning-Induced Misalignment, where fine-tuning on reasoning data containing no harmful …
领域：cs.AI作者：Yipeng Zhao、Qishun Yang、Shenzhe Zhu
论文精选 65已解读
SRPO: Self-Reflective Policy Optimization for Long-Horizon Reasoning· 自省策略优化
Self-reflection is a powerful mechanism for credit assignment in human learning, convertin…
领域：cs.AI作者：Jialong Liu、Yuling Shi、Ning Yang
论文精选 65已解读
Multi-Modal Semantic Expansion with Constrained LLM Reranking for Conversational Music Recommendation· 多模态约束LLM重排序在对话音乐推荐中的应用
We present Team Semiintelligencn's solution for the ACM RecSys 2026 TalkPlayData Challenge…
领域：cs.AI作者：Naman Garg、Sarika Jain、George Fazekas
论文精选 82
On the Threat Model of Weird Generalization and Emergent Misalignment· 微调数据对奇怪泛化和突发偏差的影响
Narrow fine-tuning on small, domain-specific datasets can produce broad and surprising cha…
领域：cs.CL作者：Miriam Wanner、Mark Dredze、William Walden
论文精选 65已解读
StrategyBench: Evaluating Explicit Strategy Induction in Large Language Models· 策略基准：评估大模型的显式策略归纳
As large language models are increasingly used in data-scarce and evolving task scenarios,…
领域：cs.AI作者：Jinghan Tan、Yuanzheng Wang、Lu Chen
论文精选 82
What's the Catch? Evaluating Temporal Consistency in Vision-Language Models· 时间一致性评估：视觉-语言模型的挑战
Vision-language models (VLMs) achieve strong performance on video and image-sequence bench…
领域：cs.CL作者：Marek Hradil、Danae Sánchez Villegas
论文精选 65已解读
Characterizing Necessary Losers to Explain Tournaments Losers· 解释比赛失败者
We study the problem of formally explaining why a candidate was not selected by a given to…
领域：cs.AI作者：Contet Clément、Umberto Grandi、Jérôme Mengin
论文精选 65已解读
SkillAlchemy: Open-World Agent Skill Creation· 开放世界代理技能创建
Agent skills are reusable procedural artifacts that extend language agents with specialize…
领域：cs.AI作者：Hengjun Wang、Shuyue Wei、Boyi Liu
论文精选 65已解读
MediSkill-Evo: Process-Constrained Self-Evolution for Evidence-Grounded Clinical Interaction· MediSkill-Evo：证据支持的临床交互自进化
Interactive clinical agents must gather decisive evidence and convert it into grounded act…
领域：cs.AI作者：Ruoyu Wu、Shenfu Xie、Yinqian Sun
论文精选 82
Modalities Should Talk to Each Other: Dual-Stream Multimodal Learning for Long-Horizon Influenza Forecasting· 模态应互动：用于长期流感预测的双流多模态学习
Forecasting long-range influenza-like illness (ILI) matters for public health readiness. P…
领域：cs.AI作者：Seyed Mohammad Hossein Hashemi、Mohsen Hooshmand、Parvin Razzaghi
论文精选 85
Walking on the DARKSIDE· 漫步黑暗面：逻辑增强的LLM生成方法
Large Language Models (LLMs) recognise patterns but do not natively track the path of excl…
领域：cs.AI作者：Aldo Gangemi、Emanuele Bottazzi
论文精选 85
Agent-G$^2$: Gaussian Guidance for Agentic Reinforcement Learning· Agent-G$^2$：高斯引导的代理强化学习
Hint-based reinforcement learning addresses reward sparsity in long-horizon agentic tasks …
领域：cs.AI作者：Zixuan Wang、Yanrui Miao、Zhengxi Lu
论文精选 85
EviSafe: Evidence-Grounded Safety Evaluation for Vision-Language Models
Vision-language model safety benchmarks typically evaluate only final responses: whether a…
领域：cs.AI作者：Xuetong Li、Gaofeng Liu
论文精选 82
Apodex 1.1: Scaling Agentic Intelligence for Complex Work· Apodex 1.1: 扩展代理智能以应对复杂工作
General-purpose language models can reason and synthesize knowledge, but complex work also…
领域：cs.AI作者：Apodex Team、B. An、B. Li
论文精选 82
Hidden in the Request: Explaining Unethical LLM Compliance through Token Relevance· 请求中的隐藏：通过令牌相关性解释不道德的LLM合规性
Although Large Language Models (LLMs) are aligned to optimize for both helpfulness and har…
领域：cs.AI作者：Or Biton、Tomer Krichli、Itai Allouche
论文精选 75
Automated Construction of FAIR Digital Object Knowledge Graphs from Flat Cultural Heritage Records· 从平面记录构建FAIR知识图谱
The FAIR Digital Object (FDO) framework mandates that metadata attribute values be express…
领域：cs.AI作者：Zeyd Boukhers、Lingxiao Kong、Xenophon Zabulis
论文精选 82
Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data· RL真的优于SFT吗？重新审视训练策略
Recent work proposes next-chunk reasoning RL for leveraging no-CoT data---corpora such as …
领域：cs.AI作者：Yinhao Tang、Youqing Fang、Yanan Sun
GitHub 热门开源项目
开源项目精选 80
stems-arraign-48/neurocursor-ai
AI-powered, camera-based mouse cursor control written in C++. Turn your webcam into a hand…
⭐ 41 stars语言 C++
开源项目精选 75
dshplugin/dsh-plugin-hub
DeepSeek Harness 社区内置插件市场（dsh-plugin）— 搜索插件、下载并安装 4000+ 人工精选社区插件，每日更新、完全免费。内置在 Harness「设置 …
⭐ 47 stars语言 TypeScript
开源项目精选 83
yanght27/GPU-Perf-Playground
GPU 性能与 AI Infra 学习项目：CUDA/Triton 算子、NCU/NSYS、vLLM/SGLang/TRT-LLM/ms-swift、PyTorch/DeepSpe…
⭐ 55 stars语言 Python
开源项目精选 82
Dpro-at/Tel-Agent· Dpro-at/Tel-Agent: 开源 AI 电话助手
AI phone assistant | open-source
⭐ 45 stars语言 TypeScript
开源项目精选 80
zimodzh/dsh-plugin-dev-skills· 开发 DSH 插件的 Agent Skill
An Agent Skills skill for developing DeepSeek Harness (DSH) plugins（开发 DSH 插件的 Agent Skill…
⭐ 39 stars
精选 AI 行业资讯
行业资讯精选 82
Robot brain builders are pushing out of their GPT-2 era· 机器人脑开发步入后GPT-2时代
Robot bodies are waiting for their AI brains to catch up.
行业资讯精选 82
QueryStory wants you to believe what AI is telling you· QueryStory 让你相信AI的建议
The startup came out of stealth with $6 million in seed funding and a plan to use LLMs and…
行业资讯精选 65
Arga Labs is building a better way to train enterprise AI agents· Arga 为训练企业 AI 代理筹集 1000 万美元
Arga has raised $10 million in a seed funding round that was led by General Catalyst, with…
行业资讯精选 82
Hearing tech startup Legato emerges from stealth with $12M and a peek at its AI hearing glasses· Legato推出AI助听眼镜
The glasses, called Legato Frames, integrate the company’s patented hearing-assistance tec…
行业资讯精选 82
Runable hits $21M to bet AI agents can go from building businesses to growing them· Runable 获 2100 万美元押注 AI 代理
Runable says 60%–70% of its 1 trillion-plus token usage in the last 90 days came from payi…
行业资讯精选 65
AI视频应用井喷，美图打开新的增长空间
从图像到视频，美图找到了一门更赚钱的生意
行业资讯精选 85
小宇宙推出《AI趋势报告》：AI创作、AI办公、协作型AI等成讨论新趋势
AI内容创作者增长187%，相关节目数量增长239%
行业资讯精选 82
硅谷今日最热具身模型！不用后训练，看一遍就学会
具身智能迈向GPT时刻
行业资讯精选 82
ECCV 2026｜4步去噪实现12.66 FPS，清华、港科大提出实时流式视频编辑框架LiveEdit
来源：机器之心 原文链接：https://mp.weixin.qq.com/s/HT6AMtSUoKPSpEKI2TLy5Q 文本指令驱动的视频编辑技术正在从离线内容生产走向直播特…
行业资讯精选 75
达卯科技与福建智算方舟达成合作，全栈式算电协同服务在闽落子
长乐机场综保区智算中心全栈算电协同服务
行业资讯精选 75
2026世界机器人大会落幕，谁在物理AI赛道抢跑？
<img data-src="https://static.leiphone.com/uploads/new/images/20260826/6a8ebc40e3115.jpg" …
行业资讯精选 82
即梦 AI 推出内容厂牌“即梦片场”，探索 AI 影像发行新路径
8月26日，即梦AI正式推出影视内容厂牌“即梦片场”，面向电影、剧集项目开放征集，入选项目将获得算力、技术、资金、行业资源等多维支持。依托 “即梦片场” 的分级合作机制，即梦 AI…
行业资讯精选 82
英伟达Vera Rubin的Agent吞吐暴涨最高提升30倍，却不只靠GPU | Hot Chips 2026
在Agent推理上，英伟达又把性能往前推了一大步。 Hot Chips 2026期间，英伟达披露了Vera Rubin NVL72最新Agent工作负载测试结果：在DeepSeek…
行业资讯精选 70
独家丨王云鹤创业估值数亿美元，押注多模型 Harness和 Agent-Native Model
<section data-mpa-powered-by="yiban.io" data-pm-slice="0 0 []" style="-webkit-tap-highligh…
行业资讯精选 83
Agent 走向具身世界：从语言智能到机器人「大脑指挥官」
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 82
IJCAI 2026 专访：动作指挥计算，VLA 加速不降准 | GAIR Paper 125
<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/m…
行业资讯精选 80
刚刚，Top级的视频AI免费了！我反手做了个精致版《牛来》· Top级视频AI免费，创作《牛来》示范
Agnes Video 2.5 Flash
行业资讯精选 80
宇树智元共用一个大脑！神秘模型Demo炸场，10分钟一镜到底
展示了多机器人共享AI模型的能力
行业资讯精选 83
大厂AI To B大战，这回战场为什么是办公？
大厂AI To B商业化的战火，已经烧到AI办公。 内部赛马、业务整合、资源重配，过去一个多月，腾讯、阿里、字节三家大厂几乎同时将AI资源砸向办公场景。 “AI作为生产力工具，最终…
行业资讯精选 80
具身智能走进千行百业，浙江人形 WRC2026 诠释 "具身在场"
2026 WRC 现场，三台机器人默契配合，正完成工业制造拆垛、分拣、搬运的全流程工作；几步之外，观众在零售舱前下单，等机器人递来饮料；厨房里，一顿饭从食材处理一路做到最后的清洁；…
行业资讯精选 65
India’s Ringg gets backing from Peak XV as it pushes voice AI past the phone call· 印度语音AI公司Ringg获 Peak XV 投资
Ringg has raised $10 million from Peak XV as a part of its Series A extension.
行业资讯精选 82
Falcon TST 2.0获世界权威测评第一名，推动时间序列基础模型从通用预测走向金融应用
蚂蚁国际日前正式发布自研时序AI预测大模型“鹰序TST”2.0版。
行业资讯精选 83
深度实测「豆包工作」+飞书：目前最接近企业Agent终局的答案· 深度实测「豆包工作」+飞书
shui jiang shuai xian sha
行业资讯精选 65
OpenAI「辣椒芯」干翻英伟达！老黄股价不跌反涨· OpenAI「辣椒芯」干翻英伟达！
不和老黄商量
行业资讯精选 75
Robotics startup Generalist reaches $3B valuation, sources say· 机器人初创公司 Generalist 估值达30亿美元
The $200 million extension comes just months after the physical AI startup reached a $2 bi…
行业资讯精选 70
OpenAI loses a top data center exec, as stream of high-profile departures continues· OpenAI 高层离职潮继续
Before Malone left, OpenAI had already reshuffled its infrastructure org, shifting his rep…
行业资讯精选 75
How loveholidays is making everyone a builder with Codex· loveholidays 用 Codex 让所有人成为构建者
Discover how loveholidays uses OpenAI Codex to make software development accessible across…
行业资讯精选 65
Stability AI, maker of image generator Stable Diffusion, raises $76 million in fresh funding· Stability AI 获新融资
The company's new fundraising total now stands at $232 million.
行业资讯精选 75
Claude Cowork finally remembers what you told the app in chat· Claude终于记得你在聊天中告诉应用的内容
Anthropic is giving Claude a shared memory across chat and Cowork, so users no longer have…
行业资讯精选 82
基于阿里千问，路透社推出首个自研大模型
8月25日，据外媒报道，国际知名新闻机构汤森路透（路透社母公司）近日推出了其自研大模型Thomson-1。该模型基于阿里Qwen大模型开发，旨在降低对Claude等闭源模型的依赖并…
行业资讯精选 85
The full stack behind abundant intelligence· AI 技术栈的全面解析
OpenAI CFO Sarah Friar explains how advances across chips, compute, models, and products c…
行业资讯精选 82
Introducing the Admin plugin for ChatGPT Work and Codex· ChatGPT Work 和 Codex 的管理员插件
Use the Admin plugin for ChatGPT Work and Codex to analyze workspace usage, manage members…
YouTube 热门 AI 视频
AI 视频精选 85
DeepSeek’s AI Just Learned To Upgrade Itself· DeepSeek 的 AI 学会了自我升级
YouTube · Two Minute Papers
📺 Two Minute Papers▶ 6.4k 次观看
AI 视频精选 75
Your coding agent keeps solving the same problem· 你的代码代理不断解决相同问题
YouTube · DeepLearningAI
📺 DeepLearningAI▶ 134 次观看
AI 视频精选 82
What Codex Unlocks for loveholidays· Codex为loveholidays解锁什么
YouTube · OpenAI
📺 OpenAI▶ 652 次观看
AI 视频精选 82
How to Manage Your Workspace With ChatGPT Work· 用ChatGPT优化工作空间
YouTube · OpenAI
📺 OpenAI▶ 1.2k 次观看
AI 视频精选 65
NVIDIA Vera Rubin NVL72 production racks are here.· NVIDIA 推出 Vera Rubin NVL72 生产级机架
YouTube · NVIDIA
📺 NVIDIA▶ 685 次观看
AI 视频精选 65
Codex, but make it weird. Hahahacks.· 趣味 Codex hack
YouTube · OpenAI
📺 OpenAI▶ 1.2k 次观看
AI 视频精选 75
No Limits on Text Chats with ChatGPT· ChatGPT 文本聊天无限制
YouTube · OpenAI
📺 OpenAI▶ 267 次观看