hackcv · 实时 AI 资讯聚合

📰 AI 研究简报 · 10月24日

今日头条：iRTE2026：对话 AI 如何迈向 AGI？声网CEO赵斌给出五级对话能力框架 —— 10月23日，iRTE2026实时智能大会在北京举行。声网创始人兼 CEO 赵斌在主论坛发表题为《对话 AGI：实时互动的下一章》的演讲，回顾了过去一年实时互动、对话式 AI 与底层基础设施的关键变化，并提出“五级对话能力”框架，勾勒对话式 AI 从基础应答走向深度理解与关系沟通的演进路径。赵斌认为，随着 AI 从回答问题走向持续交流、理解上下文并执行任务，实时互动的对象正在从人与人扩展到人与 Agent。实时互动技术的角色也随之延伸：它不仅要让信息传递更稳定、更流畅，还要为 AI 的感知、理解、表达与行动提供实时支撑。 <p sty

共 46 条预计阅读 28 分钟Token 消耗 82,800

arXiv 最新论文

论文精选 65已解读

Direct Pass@$k$ Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning

Test-time scaling has emerged as a key inference strategy for reasoning models, where a mo…

领域：cs.AI作者：Mahdi Salmani、Adel Javanmard、Devansh Gupta

📎 arXiv🕒 10-23 01:59🔗 arxiv.org

论文精选 65已解读

EpiSEM-Bench: Evaluating Safety and Evidence Recovery in LLM Agents under Missing Evidence

Tool-using language model agents rely on environmental facts to decide whether an action i…

领域：cs.AI作者：Qijin Zhang、Hanzhen Wang、Yuehan Zhao

📎 arXiv🕒 10-23 01:57🔗 arxiv.org

论文精选 82

Low-Bit N-Gram Embeddings for Language Models· 大模型低精度 n-gram 嵌入

Many recent open-weight language models use n-gram embedding tables to improve performance…

领域：cs.CL作者：Stephen Chung、Wenyu Du

📎 arXiv🕒 10-23 01:56🔗 arxiv.org

论文精选 82

Can an Agent Debug a Web Corpus? Agentic Auditing of Slovak FineWeb2· LLM 能否调试网络语料库？

Multilingual pipelines such as FineWeb2 apply one recipe across a thousand languages, leav…

领域：cs.CL作者：Marek Šuppa、Ivan Vykopal、Jaroslav Kopčan

📎 arXiv🕒 10-23 01:52🔗 arxiv.org

论文精选 83

Bilingual Language Models Capture Cross-Linguistic Influence in Second-Language Reading· 双语模型捕捉二语阅读的跨语言影响

While monolingual language model (LM) surprisal is widely established as a predictor of re…

领域：cs.CL作者：Abderrahmane Issam、Irene E. Winther、Stefan L. Frank

📎 arXiv🕒 10-23 01:42🔗 arxiv.org

论文精选 82

How Ground Truth Shapes Retrieval Scores: Evidence-Aware Evaluation of a Retrieval-Augmented Generation System for Breast Lesions· 乳腺病变管理文献中的检索系统评估

Purpose: To describe an evidence-aware retrieval evaluation protocol for a clinical retrie…

领域：cs.CL作者：Miguel Mochizuki Silva、Samuel Ismael、Mateus C. Dias

📎 arXiv🕒 10-23 01:36🔗 arxiv.org

论文精选 65已解读

Simulated Control Evals· 模拟控制评估

Control evaluations measure whether safeguards can prevent an AI model from causing undesi…

领域：cs.AI作者：Jeffrey Olmo、David Zhang、Tyler Tracy

📎 arXiv🕒 10-23 01:26🔗 arxiv.org

论文精选 80

Would an LLM Match into a Residency? Uncertainty, Reasoning and Image Use of Open Language Models on the Spanish MIR Examination· 大语言模型能通过医学考试吗？

Accuracy on medical examinations is common evidence offered for the clinical use of langua…

领域：cs.CL作者：Miguel Fernandez-de-Retana、Unai Zulaika、Ruben Sanchez-Corcuera

📎 arXiv🕒 10-23 01:23🔗 arxiv.org

论文精选 65已解读

Who Is Sycophantic to Whom? Confident LLMs, Human Deference, and the Collapse of Trust Calibration· 谁对谁谄媚？

Research on sycophancy asks whether large language models (LLMs) bend towards what their u…

领域：cs.AI作者：Deepthika S、Pawni Yadav、B. Sankar

📎 arXiv🕒 10-23 01:15🔗 arxiv.org

论文精选 65已解读

What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language Models· 什么是好的解码策略？

Masked diffusion language models generate tokens in parallel, but their efficiency depends…

领域：cs.AI作者：Jinya Sakurai、Xun Xu、Jaehong Yoon

📎 arXiv🕒 10-23 01:15🔗 arxiv.org

论文精选 65已解读

Contestable Agents for Data Governance: Recording Human Stewardship of AI-Generated Metadata Across Structured and Unstructured Systems· 数据治理中的可争议代理

Industrial organizations hold operational knowledge in structured records, such as telemet…

领域：cs.AI作者：Humza Akhtar、Brian Freeman

📎 arXiv🕒 10-23 00:57🔗 arxiv.org

论文精选 65已解读

Beyond Causal Masks: Auditing and Localizing Hidden Future-Token Leakage in Adaptive Sequence Models· 超越因果掩码

Hidden future-token leakage can create false architectural gains: an autoregressive model …

领域：cs.AI作者：Huiting Liu、Jinsong Wu

📎 arXiv🕒 10-23 00:34🔗 arxiv.org

论文精选 65已解读

Jev-DLC: Decision-Level Continual Learning from Typed Feedback· Jev-DLC: 基于类型反馈的决策级持续学习

Individual preferences often become visible only after deployment, when a user accepts one…

领域：cs.AI作者：Yangyang Feng、Jianle Gan、Zhuoyan Feng

📎 arXiv🕒 10-23 00:33🔗 arxiv.org

论文精选 65已解读

From Prompt Rules to Golden Data Model: Production-Scale Generation of Personalized, Compliant Pharmaceutical Claims· 从提示规则到金色数据模型

Large language models (LLMs) are increasingly used to generate domain-specific commercial …

领域：cs.AI作者：Suyash Mishra、Qiang Li、Subhabrata Adak

📎 arXiv🕒 10-23 00:31🔗 arxiv.org

论文精选 65已解读

DuoCook: An Overcooked-Inspired Benchmark for Visual Planning and Coordination· DuoCook: 双人协作烹饪任务的基准测试

Coordinating two players in a cooking task requires a vision-language agent to track chang…

领域：cs.AI作者：Ziyi Jiao、Heng Zhang、Drishti Regmi

📎 arXiv🕒 10-23 00:26🔗 arxiv.org

论文精选 65已解读

FairFRESH: A Neuro-Symbolic Framework for Fair and Explainable Household Food Recommendation· FairFRESH：家庭食品推荐的神经符号框架

Recommender systems typically serve one user at a time. However, in households, members sh…

领域：cs.AI作者：Kodur Sai Vinay Sathvik、Kakoli Bora、Tejaswini R. M

📎 arXiv🕒 10-23 00:23🔗 arxiv.org

论文精选 65已解读

Adaptive Skill Granularity for Coding Agents: Runtime Routing and Macro-to-Micro Recovery· 编码代理的自适应技能粒度

Reusable skills allow large language model agents to package task-specific procedures, too…

领域：cs.AI作者：Dingxuan Zhang、Yanhui Jiang、Jun Wang

📎 arXiv🕒 10-22 23:55🔗 arxiv.org

论文精选 82

DIAL: Steerable Language Models with Distilled Instruction-Adherence Levels· DIAL：具有蒸馏指令遵循度的语言模型

When a user says "Keep it short", how short should the LLM's response be? More generally, …

领域：cs.AI作者：Thanh Q. Tran、John Russell Himawan、Arun Verma

📎 arXiv🕒 10-22 23:34🔗 arxiv.org

论文精选 82

Conservative Meta-Agent Search: Architecture Adaptation under Non-Stationarity· 保守的元智能体搜索

Automated agentic-system design is typically evaluated on static task suites with immediat…

领域：cs.AI作者：Ali Atiah Alzahrani

📎 arXiv🕒 10-22 23:32🔗 arxiv.org

论文精选 82

PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models· 视频生成模型的物理解释性评估

World action models increasingly rely on pretrained video models assumed to capture physic…

领域：cs.AI作者：Taekoan Yoo、Wonkyung Jung、Kyunghun Kim

📎 arXiv🕒 10-22 23:27🔗 arxiv.org

论文精选 85

Certified Execution Harness for Agentic AI: Safe and Least-Intrusive Petri-Net Supervisory Control Between Reasoning and Action· 代理 AI 的认证执行框架

This paper studies the problem of safe tool execution for large language model (LLM) agent…

领域：cs.AI作者：Chao Gu、Quanyan Zhu

📎 arXiv🕒 10-22 22:48🔗 arxiv.org

论文精选 83

RIDAR: Safeguarding Research Integrity in Autonomous Research Agents via Automated Detection and Runtime Intervention· RIDAR: 通过自动化检测和运行时干预保护研究完整性

Autonomous research agents are increasingly used to automate large portions of scientific …

领域：cs.AI作者：Jiening Siow、Amy Xin、Junjie Wang

📎 arXiv🕒 10-22 22:43🔗 arxiv.org

论文精选 82

Muffin - A Shared Memory Buffer for Asynchronous AI Agents· Muffin - 异步AI代理的共享内存缓冲

The rise of Agentic AI systems has changed the landscape of artificial intelligence by ena…

领域：cs.AI作者：Ping-Lin Kuo、Suman Navaratnarajah、Christian Frey

📎 arXiv🕒 10-22 22:36🔗 arxiv.org

论文精选 83

ValuSpec: Plug-and-Play Candidate Valuation before Target Verification for Tree-Based Speculative Decoding· ValuSpec：树基推测解码前的候选估值机制

Tree-based speculative decoding accelerates large language model inference by verifying mu…

领域：cs.AI作者：Liang He、SiYuan Ma、Qishi Zhan

📎 arXiv🕒 10-22 22:23🔗 arxiv.org

论文精选 82

Subgoal-aware Memory Graph for GUI Agent· 子目标感知记忆图

GUI agents aim to automate complex user tasks by interacting with mobile interfaces, but e…

领域：cs.AI作者：Dezhao Luo、Daniel Beechey、Derek Yuen

📎 arXiv🕒 10-22 22:06🔗 arxiv.org

GitHub 热门开源项目

开源项目精选 70

verity-mods/verity-enhanced· Verity-Enhanced 模组

Unofficial Enhanced Edition of the Verity Minecraft mod. Upgraded AI, new scenarios, multi…

⭐ 100 stars

📎 GitHub🕒 10-23 21:18🔗 github.com

开源项目精选 85

liyupi/dafeiyu-cleaner

DeepSeek 蓝色大肥鱼消除计划：识别并清理 B 站（Bilibili）评论区中由 DeepSeek 大模型、AI Agent 驱动的机器人账号（大肥鱼）的 Chrome / …

⭐ 39 stars语言 JavaScript

📎 GitHub🕒 10-23 11:39🔗 github.com

开源项目精选 75

heterodoxin/AbliteratedOutTheBox· 改进的 llama.cpp 推理版

llama.cpp fork that abliterates at inference

⭐ 48 stars语言 C++

📎 GitHub🕒 10-22 05:54🔗 github.com

开源项目精选 72

AbuZar-Ansarii/Ultron· Ultron：情境感知的AI桌面助手

Autonomous desktop intelligence with MCU Ultron persona, mood-aware workspace orchestratio…

⭐ 60 stars语言 Python

📎 GitHub🕒 10-18 18:02🔗 github.com

精选 AI 行业资讯

行业资讯精选 85

iRTE2026：对话 AI 如何迈向 AGI？声网CEO赵斌给出五级对话能力框架

10月23日，iRTE2026实时智能大会在北京举行。声网创始人兼 CEO 赵斌在主论坛发表题为《对话 AGI：实时互动的下一章》的演讲，回顾了过去一年实时互动、对话式 AI 与底…

📎 雷锋网 AI🕒 10-24 10:44🔗 leiphone.com

行业资讯精选 85

OpenAI says it will no longer train its AI models on the live internet, citing “significant risks”· OpenAI停止用实时互联网数据训练AI模型

OpenAI will cut frontier models off from the internet weeks after one of its engineers sai…

📎 TechCrunch AI🕒 10-24 07:05🔗 techcrunch.com

行业资讯精选 80

Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)· 快速扩展在线存储以服务超 10 亿 ChatGPT 用户（二）

Learn how OpenAI scaled Habitat with reliable multi-tenancy, global caching, and Azure Cos…

📎 OpenAI Blog🕒 10-24 01:00🔗 openai.com

行业资讯精选 80

Kimi K3.1 泄露：3D 成本仅有 GPT-6 Astra 的 1/14！横扫海外视觉三巨头· Kimi K3.1 泄露：3D 成本仅有 GPT-6 Astra 的 1/14！横扫海外视觉三巨头

<img class="rich_pages wxw-img" data-aistatus="1" data-croporisrc="https://mmbiz.qpic.cn/s…

📎 雷锋网 AI🕒 10-23 18:57🔗 leiphone.com

行业资讯精选 60

从技术首创到规模上车，华为乾崑车载激光投影以金辑奖印证技术领先

10月23日，在盖世汽车主办的第八届金辑奖颁奖盛典上，HUAWEI XSCENE 华为乾崑车载激光投影凭借在智能座舱视听领域的技术突破与规模化商用落地，荣膺“金辑奖·2026最佳技…

📎 雷锋网 AI🕒 10-23 18:43🔗 leiphone.com

行业资讯精选 75

覆盖存、拣、运全场景，海康机器人2026移动机器人新品发布会举行

📎 OpenAI🕒 10-24 02:20🔗 youtube.com

AI 视频精选 85

How Nokia and NVIDIA Are Accelerating the Future of AI-Native Networks· 诺基亚与英伟达加速 AI 原生网络

YouTube · NVIDIA

📺 NVIDIA▶ 510 次观看

📎 NVIDIA🕒 10-24 01:00🔗 youtube.com