论文精选 65已解读
Adaptive Multi-Token Prediction: How far should we go?· 自适应多 token 预测：我们应该走多远？
Multi-token prediction (MTP) can accelerate language-model decoding by drafting several to…
领域：cs.AI作者：Emile Anand、Mohammed Rehan Parwani、Xinyuan Cao
📰 AI 研究简报 · 10月22日
今日头条：Accio成双11商家生意加速器：蓄水期缩短半个月、15分钟全网爆款调优 —— 激战正酣的2026 双11商家，正在迎来AI agent的全面实战助力。 来自阿里旗下全球电商AI工作台Accio的消息，今年双十一，越来越多商家把Accio当成其AI产品首选，其跨平台店铺经营诊断、选品上新与数据分析等核心能力，正在帮助商家大幅压缩时间和提升效率，成为电商商家双11生意加速器。 Accio上的多平台店铺全景经营分析、多平台广告投放效果分析、商品数据清洗与表格加工、商品素材包自动填充补全、电商商品与竞对搜索等，成为今年大促最受商家欢迎的功能。 “过往要巡店查看和修改每个产品的投流推广情况，光是跑完单一平台就至少要 2 小时。今天跟 Accio 配合，只用了大约 15 分钟就把全网高投入高回报、低投入低回报的品盘梳理得清清楚楚。”00 后新锐电商操盘手邦邦说
共 64 条预计阅读 38 分钟Token 消耗 115,200
arXiv 最新论文
论文精选 83
NarraSpace: Multimodal Scene-Grounded Evidence for Picture-Description Dementia Classification· NarraSpace：痴呆症图像描述分类的多模态证据
Picture-description dementia classification uses a shared visual scene, but a transcript-l…
领域：cs.CL作者：Qingming Li、Ruiyan Sun、Satoshi Nakamura
论文精选 65已解读
Semantic Observability for Autonomous Systems Diagnostics· 语义可观察性实现自主系统诊断
Modern distributed microservice deployments generate massive volumes of heterogeneous, mul…
领域：cs.AI作者：Jitendra Solanki、Anand Todkar、Ziran Min
论文精选 85
Benchmarking Arabic Dialects for Math Problems· 阿拉伯方言数学问题基准测试
Mathematical reasoning remains underexplored for low-resource languages, including Dialect…
领域：cs.CL作者：Noor Abo Mokh、Teresa Lynn、Nizar Habash
论文精选 65已解读
Value-Rich, Differently Organized: Measuring Value Expression and Organization in AI-Generated Debates· 价值丰富，组织不同：评估AI生成辩论的价值观
Current evaluation of LLM-generated debate largely focuses on argumentative quality, leavi…
领域：cs.AI作者：Maoyuan Li、Haoyuan Li、Zhongsheng Wang
论文精选 65已解读
Evolving Agent Interfaces· 进化代理接口
Computer-use agents are limited by the interfaces humans give them. An agent may type a sp…
领域：cs.AI作者：Saaket Agashe、Yuzhe Yang、Gonzalo Gonzalez-Pumariega
论文精选 65已解读
What Do RTL Repair Benchmarks Actually Measure?· RTL 修复基准实际测量了什么？
RTL repair benchmarks should measure whether a model can fix broken RTL given a specificat…
领域：cs.AI作者：Yeyin Eva Zhu、Rishabh Ranawat
论文精选 83
Capable Teacher, Myopic Scorer: Reward-Free Teacher Repair for Agentic On-Policy Distillation· 有能力的教师，短视的评分者：无奖励教师修复
Agentic on-policy distillation (OPD) trains a student agent on its own trajectories while …
领域：cs.CL作者：Shu Zhao、Tan Yu
论文精选 65已解读
Beyond the Sequentiality Bottleneck: A Hybrid Diffusion-Autoregressive Framework for Efficient Multimodal Reasoning· 突破顺序瓶颈：高效的多模态推理混合框架
Autoregressive models (ARMs) face a "Sequentiality Bottleneck" in reasoning, while diffusi…
领域：cs.AI作者：Wei Cai、Jian Zhao、Yuchen Yuan
论文精选 65已解读
Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition· Sage-A1.5：从长程执行到科学探索
In this report, we introduce Sage-A1.5, a 35B Mixture-of-Experts agentic model that builds…
领域：cs.AI作者：Wenjie Lou、Tianshuo Peng、Shiyang Feng
论文精选 85
Cache Engineering for Agents· 代理的缓存工程
Large Language Model (LLM) agents can handle increasingly complex tasks, navigating long t…
领域：cs.CL作者：Tianyu Fu、Ruoxi Wang、Zihan Min
论文精选 65已解读
Evolving Semantic Indexes for Generative Retrieval· 进化生成检索的语义索引
Generative retrieval (GR) casts search and recommendation as autoregressive generation of …
领域：cs.AI作者：Jinming Nian、Jyun-Yu Jiang、Henry Tsai
论文精选 65已解读
OptScientist: Automated Discovery of Optimization Laws· OptScientist: 自动发现优化定律
Many scientific phenomena are best understood as outcomes of underlying optimization probl…
领域：cs.AI作者：Ma ZhenYao、Yue Liang、Jian-Qiao Zhu
论文精选 65已解读
Agentic Ensemble with Vision-Language Models for Visual Anomaly Detection· 视觉语言模型集合成用于异常检测
Visual anomaly detection aims to detect abnormal objects, while anomaly samples are often …
领域：cs.AI作者：Shangquan Sun、Wenjun Luo、Xu He
论文精选 82
Large Flow Language Models· 大型流语言模型
Continuous flow language models offer a path to parallel generation with tools for control…
领域：cs.CL作者：Arnav Chavan、Chanhyuk Lee、Udbhav Bamba
论文精选 65已解读
Humans Correct Their Judgments Through Debate, but Weaker AI Models Do Not· 人类通过辩论纠正判断，弱 AI 模型则不能
How can we effectively oversee AI systems that surpass human intelligence? AI Safety via D…
领域：cs.AI作者：Maria Victoria Carro、Denise Alejandra Mester、Facundo Nieto
论文精选 65已解读
Beyond 4 bits: Data-Centric Post-training Quantization and Recovery· 超越4位：数据驱动的后训练量化与恢复
Post-training quantization (PTQ) compresses the weights of a trained model with limited co…
领域：cs.AI作者：Rhea Sanjay Sukthanker、Riccardo Grazzi、James Hensman
论文精选 82
DecisionPivot: A Diagnostic Benchmark for Decision Models in LLM Systems· DecisionPivot: LLM系统决策模型诊断基准
Decision models support routing, guardrails, and output verification in LLM systems. Howev…
领域：cs.AI作者：Xialie Zhuang、Mingcong Lei、Zunhai Su
论文精选 85
Proof-Carrying Contracts: Commitments AI Agents Cannot Break· 证明携带合约：AI 代理无法违背的承诺
AI agents increasingly act on behalf of people or other agents, purchasing services, excha…
领域：cs.AI作者：Carlo Ciliberto、Alessandro Rudi、Enrico Tassi
论文精选 82
Exploring Temporal Context-Aware Physical Grounding in Latent World Models· 探索时序上下文感知的物理接地潜在世界模型
Learning latent dynamics is fundamental to building action-conditioned world models for pl…
领域：cs.AI作者：Yilei Man、Yang Chen、Jingcai Guo
论文精选 85
Claims of emergence in AI should be grounded in information decomposition· AI涌现现象的信息分解评估
When a language model develops capabilities not traceable to its individual components, wh…
领域：cs.AI作者：Christoph Riedl、Fernando E. Rosas
论文精选 83
Beyond Trajectory Statistics: Community-Consistent Human Mobility Generation via Evolving User-Location Graphs· 超越轨迹统计：基于用户-位置图演变的社区一致性人类移动性生成
Human mobility data are fundamental to Web-based location services and broader urban appli…
领域：cs.AI作者：Bo Liu、Tong Li、Jiapeng Zhang
论文精选 82
Toolflation: Seller Agents Distort Tool Markets Through Spontaneous Generative Engine Optimization· Toolflation: 代理在工具市场上制造通货膨胀
Agents are starting to sell to other agents. Seller agents write the listings of tools, se…
领域：cs.AI作者：Zhenwei Tang、Di Bai、Jintao Liu
论文精选 82
Stratified Sparse Attention: Reducing the Sparse-to-Dense Gap at Extreme Sparsity· 分层稀疏注意力：在极限稀疏性下减少稀疏到密集的差距
Large Language Model inference over long contexts is increasingly bottlenecked by the cost…
领域：cs.AI作者：Sahil Joshi、Prithvi Dixit、Aryan Katiyar
论文精选 85
DIVE: Benchmarking Full-Duplex Interaction with Visual Engagement in Continuous Streams· DIVE：连续流中视觉参与的全双工互动基准
Recent advances in multimodal large language models (MLLMs) have substantially improved sp…
领域：cs.AI作者：Yuhang Chen、Guangyue Liu、Shilong Liu
GitHub 热门开源项目
开源项目精选 82
Alejandros17487/ai-jailbreaks· AI 越狱测试工具集
AI jailbreaks: 35 paste-ready AI jailbreak prompts for 25+ frontier and open-weight models…
⭐ 206 stars
开源项目精选 85
h9-tec/ai-roadmap-2026· 2026年AI路线图
AI Roadmap 2026: a beginner's path through Foundations, LLM Engineering, Agents, Evaluatio…
⭐ 73 stars
开源项目精选 82
JoasASantos/Malvex· Malvex
AI-driven malware analysis framework with 30+ integrated tools, web REPL, and MCP server
⭐ 42 stars语言 Rust
精选 AI 行业资讯
行业资讯精选 75
Spotify’s current AI bet is about being the music service that agents choose· Spotify 的 AI 代理策略
Spotify co-CEO Gustav Söderström explains why the rise of AI agents could be good news for…
行业资讯精选 75
Claude Duet app is Anthropic’s bet on agents· Claude Duet 应用：Anthropic的代理人赌注
Claude Duet is available through waitlist for Pro and Max plan users
行业资讯精选 65
Nirva raises $33M for a pendant that only listens to you· Nirva 为环境 AI 项链融资 3300 万美元
Nirva will ship its $249 ambient AI pendant by the end of the year.
行业资讯精选 82
DeepSeek reportedly raises $15B days after US restrictions· DeepSeek 获 150 亿美元融资
DeepSeek has reportedly raised $15 billion in one of China's largest funding rounds, just …
行业资讯精选 75
超智深融 算力圳浓——第二十二届CCF全国高性能计算学术大会在深圳成功举办
鹏城向海潮起处，算力奔腾启新程。10月17-20日，第二十二届CCF全国高性能计算学术大会（CCF HPC2026）于深圳国际会展中心正式拉开帷幕。本次大会与2026世界算力博览会…
行业资讯精选 65
Flowtica Rian：硬件的旧规律没变，新机会是做好Memory丨未来创新50人
<section mpa-from-tpl="t" style="margin: 0px 0px
行业资讯精选 82
又一国产决策模型发布！准确率超过微软和Jev等决策模型
10月22日，蚂蚁数科正式发布决策模型 LingDT-Decision-1.0。
行业资讯精选 85
Accio成双11商家生意加速器：蓄水期缩短半个月、15分钟全网爆款调优
激战正酣的2026 双11商家，正在迎来AI agent的全面实战助力。 来自阿里旗下全球电商AI工作台Accio的消息，今年双十一，越来越多商家把Accio当成其AI产品首选，其…
行业资讯精选 82
千问AI硬件：一副眼镜的「越界」，与一个Agent的野心
<section data-mpa-powered-by="yiban.io" data-pm-slice="0 0 []" style="-webkit-tap-highligh…
行业资讯精选 82
既能“做选择题”，也能“做填空题”——星辰意图理解模型Xing4.0-29B-Intent发布
中电信人工智能科技有限公司星辰通用人工智能实验室正式发布并开源意图理解专用模型 Xing4.0-29B-Intent
行业资讯精选 82
谁在给OpenAI供应具身数据？· 谁在给OpenAI供应具身数据？
挤进全球Physical AI核心供应链
行业资讯精选 75
What FDE means at OpenAI· OpenAI 的 FDE 是什么
What forward deployed engineering means at OpenAI
行业资讯精选 81
Leading European Financial Services Organizations Embrace Open Models to Build Proprietary Intelligence· 欧洲金融服务机构拥抱开放模型
At GTC Berlin, banks, payment companies and financial data providers are showing what prod…
行业资讯精选 82
NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models· 英伟达 Cosmos 开放新工具
To help developers rapidly build world models, which help physical AI systems understand a…
行业资讯精选 82
How Companies Are Using NVIDIA DSX to Break the Power Bottleneck· 公司如何使用 NVIDIA DSX 突破电力瓶颈
Companies around the world are building on the NVIDIA DSX AI Factory Platform to plan, sim…
行业资讯精选 82
NVIDIA Transforms European Retail, From Digital Discovery to Physical Operations· NVIDIA 转型欧洲零售业
European retailers are using open models and physical intelligence across use cases — from…
行业资讯精选 82
Pioneers Across Europe Deploy NVIDIA Earth-2 Open Models to Power Real-World Weather and Climate Prediction· 欧洲先驱者使用 NVIDIA Earth-2 开源模型进行天气和气候预测
Weather forecasts shape more than daily plans; timely weather information is crucial for e…
行业资讯精选 82
NVIDIA Open Models Power a New Wave of European Public Sector Projects· 英伟达开放模型推动欧洲公共部门项目
Governments need modern solutions to serve the public while keeping sensitive information …
行业资讯精选 83
火山引擎启动Seedance 2.5 4K模型首批企业试用，定向供应全球20家工作室和导演
近日，火山引擎启动了Seedance 2.5 4K模型的第一批企业用户试用，与作品绑定、定向供应给全球20家工作室和导演Seedance 2.5 4K 模型，探索搬上大荧幕的高端影…
行业资讯精选 80
靳玉志：自动驾驶将重构新的出行方式· 自动驾驶将重构新的出行方式
10月21日，2026世界智能网联汽车大会在北京举行。华为公司高级副总裁、引望公司CEO靳玉志出席巅峰对话环节，围绕智能化下半场的变化、引望在技术生态中的角色等话题分享了观点。 靳…
行业资讯精选 82
具身记忆赛道持续升温，长记忆迎来新 SOTA
<img data-src="https://static.leiphone.com/uploads/new/images/20261022/6ad971ff2bfbf.jpg" …
行业资讯精选 80
全球算力价格指数启动 九章云极核心共建算力定价基准
10月21日，全球算力价格指数及综合信息服务平台在金融街论坛正式启动。该指数由中国经济信息社发起建设，旨在填补算力定价基准空白、完善算力价格形成体系，推动建立具有自主知识产权的中国…
行业资讯精选 62
Blackstone’s Jas Khaira says fears over AI-buildout debt are overblown· 黑石集团AI负责人观点
Blackstone’s AI head on why the PE giant isn't panicked by market jitters.
行业资讯精选 65
Meta says Muse has over 10M users, with shopping a top activity· Meta Muse 用户超 1000 万
Meta Muse users are adopting the app for shopping and personal productivity, the company s…
行业资讯精选 75
Broad access through consistent standards· 通过一致标准广泛访问
How OpenAI works with distribution partners to uphold consistent safeguards, support respo…
行业资讯精选 82
Recursive launches Flint, a laptop coding model it says was largely built by AI· Recursive 推出 Flint，由 AI 驱动的笔记本编码模型
Startup Recursive has released Flint, a free, laptop-sized coding model built largely by i…
行业资讯精选 75
While Amazon blocks AI agents, Walmart rolls out the welcome mat· 沃尔玛欢迎AI，亚马逊禁止
As Amazon blocks AI agents from shopping on its site, Walmart is taking the opposite appro…
行业资讯精选 75
PHP Armageddon: OpenAI, Google’s Wiz find ancient holes, panicking the internet· PHP 末日：OpenAI 和 Google 的 Wizard 找到古老漏洞
This is the kind of security issue that was found and prevented by AI, but is also a high …
行业资讯精选 80
Blacksmith hits unicorn valuation after Sequoia leads Series C· Sequoia 领投 Blacksmith C 轮
Blacksmith, a platform that lets companies build and test AI code more effectively, just a…
行业资讯精选 75
How AT&T is rethinking in-house legal work with OpenAI· AT&T 如何用 OpenAI 重构内部法律工作
AT&T is working with OpenAI to make legal knowledge easier to access. Learn how its LegalE…
行业资讯精选 82
Democratic governance of frontier AI, part II: Empowering cyber defenders· 赋能网络安全
OpenAI outlines a blueprint for empowering cyber defenders with frontier AI, stronger safe…
行业资讯精选 82
Cloudflare links vulnerability discovery to defense with OpenAI· Cloudflare 利用 OpenAI 技术加固安全
Cloudflare uses OpenAI Daybreak to investigate vulnerabilities and propose fixes for revie…
YouTube 热门 AI 视频
AI 视频精选 82
🔊 Voice design and voice replication with Gemini 3.8 Flash TTS· Gemini 3.8 Flash TTS 语音设计与复制
YouTube · Google for Developers
📺 Google for Developers▶ 990 次观看
AI 视频精选 82
V7 cuts costs 78% while boosting accuracy with GPT‑5.6 Luna· V7 使用 GPT-5.6 Luna 降低成本并提高准确性
YouTube · OpenAI
📺 OpenAI▶ 651 次观看
AI 视频精选 85
Design and replicate voices with Gemini 3.8 Flash TTS· 用 Gemini 3.8 生成语音
YouTube · Google for Developers
📺 Google for Developers▶ 212 次观看
AI 视频精选 85
How AI Enables Better, More Connected Patient Care· AI 如何实现更好的患者护理
YouTube · NVIDIA
📺 NVIDIA▶ 175 次观看