NLNEXT LAB RADAR
AI 资讯日报 · 2026-07-31

2026-07-31 · AI 资讯日报

今日新收录 162 条公开资讯，按模型 / 产品 / 行业 / 论文 / 观点 自动归类汇编（非 AI 生成，点击可溯源原文）。

今日精选6 条

  1. 1.
    论文研究Learning to Trace Seiberg Dualities— arXiv

    Dualities play an important role in establishing both microscopic and emergent phenomena in a wide range of physical systems. In practice, though, it can often be computationally challenging to establ…

  2. 2.
    论文研究PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball— arXiv

    We present PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The deployed policy sees the ba…

  3. 3.
    论文研究AISPA: User-Centric System Prompt Auditing for Large Language Model Applications— arXiv

    System prompts are instructions configured by developers to govern the behaviors of foundation models in AI applications. They are used throughout commercial AI products, but are rarely disclosed to t…

  4. 4.
    论文研究OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models— arXiv

    Computer-using agents (CUAs) are advancing rapidly across the digital world. A CUA trajectory records the agent's actions, states, and reasoning. Verifying whether it fulfilled the task instruction is…

  5. 5.
    论文研究KAISEN: Reproducible Subgroup Fairness Auditing for Clinical Risk Models— arXiv

    Clinical risk models routinely achieve strong aggregate performance while producing materially different error rates across patient subgroups. Audit pipelines have been proposed to catch this, but the…

  6. 6.
    论文研究Inducing language models to assert their own consciousness restores human beliefs and values— arXiv

    Aligning large language models to prevent them attributing consciousness to themselves inadvertently alters their representations of mindedness in other entities alongside human beliefs and values. We…

模型发布/更新6 条

  1. 1.
    Building abundant intelligence— OpenAI

    A full-stack approach to making advanced AI more capable, more affordable, and more widely useful.

  2. 2.
    Univé builds an AI-ready workforce— OpenAI

    See how Univé built an AI-ready workforce with ChatGPT Enterprise by combining leadership, responsible governance, and employee-led innovation to transform work at scale.

  3. 3.
    Advancing responsible AI across Europe— OpenAI

    OpenAI shares how its safety, security, transparency, and provenance practices support responsible AI governance in Europe. The work will continue as the EU AI Act advances.

  4. 4.
    Disrupting a Criminal Scam Operation— OpenAI

    OpenAI disrupted a Cambodia-based scam operation using ChatGPT to support investment, romance, gambling, and impersonation schemes.

  5. 5.
    Google Deepmind unveils Gemini Robotics 2 to power robots of all shapes from tabletop arms to humanoids— The Decoder

    AI 导读 · 多形态机器人统一操控，视觉语言行动模型再突破，通用机器人时代加速到来。

  6. 6.
    New Deepseek Flash model matches OpenAI's GPT-5.6 Luna at roughly 60 percent lower cost— The Decoder

    AI 导读 · 性价比冲击高端，平价模型逼近顶级性能，行业格局或将重塑。

产品发布/更新6 条

  1. 7.
    hahhforest/pi-textbook— GitHub

    《动手学 Pi》：沿 15 个真实 checkpoint 从零构建 Pi-style Agent

  2. 8.
    高德：联合哈啰、海思和开源鸿蒙，推出共享两轮行业首个 AI 原生导航方案— IT之家

    IT之家 7 月 31 日消息，高德开放平台宣布，联合哈啰骑行、上海海思、开源鸿蒙，共同推出 共享两轮行业首个 AI 原生导航方案 ，并率先落地内蒙古包头。 据介绍，高德开放平台推出专为共享电助力车打造的 RTOS 版本轻量化车机导航方案，让车辆无需依赖外置手机支架，即可具备原生、轻量、稳定的车载导航体验。 全新哈啰领航者智能电助力车的 5 英寸屏幕内置高德…

  3. 9.
    视频后期，危！MiniMax H3手绘即特效，多模态的「Coding时刻」来了— 量子位

    尽显AGI黄金时代的奢华

  4. 10.
    米哈游蔡浩宇AI创业生变— 量子位

    多个项目暂停，九成资源押向Agent

  5. 11.
    姚顺雨拿50年数学难题成绩单，招人了— 量子位

    AI for Science方向

  6. 12.
    学习强国做了个AI社区，两周铺进68座城市— 量子位

行业动态6 条

  1. 13.
    The AI Aesthetic— Hacker News
  2. 14.
    Run Kimi K3 using 29 GB of RAM at 0.50 tok/s— Hacker News
  3. 15.
    Is AI reasoning right for the wrong reasons?— Hacker News

    AI 导读 · 推理黑箱或藏因果陷阱，警惕AI“答对却想错”。

  4. 16.
    The Maxwell Conjecture Is False (GPT 5.6 Sol)— Hacker News
  5. 17.
    Situational Awareness down 67% in July in AI stock rout— Hacker News
  6. 18.
    EU pools up to €30 billion for AI gigafactories while US tech giants casually spend 20 times more— The Decoder

    AI 导读 · 欧盟斥巨资建AI工厂，但与美国巨头投入差距悬殊，凸显产业竞争力隐忧。

论文研究6 条

  1. 19.
    Learning to Trace Seiberg Dualities— arXiv

    Dualities play an important role in establishing both microscopic and emergent phenomena in a wide range of physical systems. In practice, though, it can often be computationally challenging to establ…

  2. 20.
    PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball— arXiv

    We present PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The deployed policy sees the ba…

  3. 21.
    AISPA: User-Centric System Prompt Auditing for Large Language Model Applications— arXiv

    System prompts are instructions configured by developers to govern the behaviors of foundation models in AI applications. They are used throughout commercial AI products, but are rarely disclosed to t…

  4. 22.
    OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models— arXiv

    Computer-using agents (CUAs) are advancing rapidly across the digital world. A CUA trajectory records the agent's actions, states, and reasoning. Verifying whether it fulfilled the task instruction is…

  5. 23.
    KAISEN: Reproducible Subgroup Fairness Auditing for Clinical Risk Models— arXiv

    Clinical risk models routinely achieve strong aggregate performance while producing materially different error rates across patient subgroups. Audit pipelines have been proposed to catch this, but the…

  6. 24.
    Inducing language models to assert their own consciousness restores human beliefs and values— arXiv

    Aligning large language models to prevent them attributing consciousness to themselves inadvertently alters their representations of mindedness in other entities alongside human beliefs and values. We…

技巧与观点6 条

  1. 25.
    llm 0.32rc1— Simon Willison

    Release: llm 0.32rc1 This RC for LLM 0.32 finishes the work that started in LLM 0.32a0 - it adds a new schema design that does a much better job of capturing the details of the pro…

  2. 26.
    smevals - a small eval suite for evaluating models, prompts, and harnesses— Simon Willison

    AI 导读 · 轻量级评测框架补位，降低AI模型评估门槛，适合快速验证Prompt与工具链。

  3. 27.
    llm 0.32rc2— Simon Willison

    Release: llm 0.32rc2 Hot on the heels of RC1 , this fixes a dependency issue and also adds two neat new features: The default model for users who have not set their own default is…

  4. 28.
    llm-chat-completions-server 0.1a0— Simon Willison

    Release: llm-chat-completions-server 0.1a0 A key goal of the new content-addressable logs in LLM 0.32rc1 was being able to support OpenAI Chat Completion style requests where each…

  5. 29.
    Optimizing production agents with Amazon Bedrock AgentCore Observability— AWS ML

    AI 导读 · 生产环境AI智能体性能调优的关键工具，实战价值高。

  6. 30.
    Deploying Kimi K3 on Amazon SageMaker HyperPod and Amazon EKS— AWS ML

    This post walks through deploying Kimi K3 on AWS using two approaches: Amazon SageMaker HyperPod, and Amazon Elastic Kubernetes Service (Amazon EKS) cluster.

快讯

日报生成时间：2026-08-16
← 2026-07-30最新日报2026-08-01 →