AAI Search
AI 资讯日报 · 2026-09-23

2026-09-23 · AI 资讯日报

今日新收录 171 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Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use— arXiv

    Multi-turn tool-use failures can hinge on a single model call, yet reward variation alone does not reveal which call would benefit from training. When rewards depend on later interactions, their varia…

  2. 2.
    论文研究LoRA-generating hypernetworks for efficient on-device LLM generative personalization— arXiv

    On-device large language models (`LLMs'), e.g. running on mobile phones, are ripe for improvement via personalization. The limited compute resources of mobile devices impose limits on model scale and…

  3. 3.
    论文研究DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation— arXiv

    Dexterous manipulation depends on contact dynamics that are often only partially observable from vision. Recent World-Action Models (WAMs) couple predictive video world modeling with action generation…

  4. 4.
    论文研究DolphinBench: Mapping the Pareto Frontier of Agent Memory— arXiv

    Agents today often take real-world actions that depend on long-term memory and context recall over time. However, most current memory benchmarks are built for a conversational question-answer format,…

  5. 5.
    论文研究Rare Event Estimation via Iterative Unalignment— arXiv

    As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whe…

  6. 6.
    论文研究Emergent Collusion in Long-Horizon LLM Agent Interaction— arXiv

    LLM agents are increasingly deployed in collaborative settings, yet long-term interaction may give rise to undesirable coordination. We study the emergence of collusion in a long-horizon multi-agent e…

模型发布/更新6 条

  1. 1.
    Claude Opus 5.5 System Card— Anthropic

    Claude Opus 5.5 System Card www-cdn.anthropic.com

  2. 2.
    Introducing Claude Opus 5.5— Anthropic

    AI 导读 · Anthropic再升级旗舰模型，性能能否继续压制对手值得关注。

  3. 3.
    Better prompt caching for GPT-6— OpenAI

    AI 导读 · 缓存命中率与显式断点让长提示词成本可控，GPT-6正把推理优化做成可调工程。

  4. 4.
    Introducing GPT-6 Sol and Luna— OpenAI

    AI 导读 · 双版本策略让高端AI按需分级，成本与能力可权衡，企业落地更灵活。

  5. 5.
    NVIDIA Isaac ROS 5.0 Advances Agentic, Open Source Robotics Development— NVIDIA

    To build and deploy sophisticated robotics applications that can perceive, reason and act in dynamic environments, developers need new physical AI models and tools. The ROS open fr…

  6. 6.
    Parallel cut research time and cost in half with GPT‑6 Astra— OpenAI

    GPT‑6 Astra allowed Parallel’s agents to research and synthesize labor-market data in half the time and at half the cost vs. prior models.

产品发布/更新6 条

  1. 7.
    Rabbit Is Back, This Time With an AI Agent App— Wired

    Two years after trying to sidestep mobile apps with dedicated AI hardware, Rabbit is launching OS3, a cross-platform agent that lives on the screens you already use.

  2. 8.
    Inkloom-art/inkloom— GitHub

    Specialised AI models for logo design — a brand-analysis model turns a business into constraints, typography and symbol models construct the mark, and a composi…

  3. 9.
    llm-typesafe 0.1a0— Simon Willison

    Release: llm-typesafe 0.1a0 I built this new plugin for LLM to add support for TypeSafe AI's new Jev model . Install it like this: llm install llm-typesafe Then set an API key ( ge…

  4. 10.
    陆川手搓历史现场，王珞丹熬夜抽卡，阿里全模态开始兜底生产— 量子位

    阿里：三年之内会出现一个原生的全模态统一生成模型，未来体验将不再受限于模态边界。

  5. 11.
    Agent时代，CPU的价值该重估了— 量子位

    CPU与GPU趋近1∶1

  6. 12.
    直播预告：未来两三年，哪些工业AI场景会率先爆发？— 量子位

    从真实工厂到产品复制，再到生态和下一批机会，两位嘉宾会把工业AI从「能用」走向「规模化」的关键问题一层层拆开。

行业动态6 条

  1. 13.
    Pentagon says overreliance on AI contributed to missile strike on Iran school— Hacker News

    AI 导读 · AI误判导致平民伤亡，暴露军事自动化决策的致命风险，亟需国际监管。

  2. 14.
    AI Has No Wisdom and Neither Will You— Hacker News
  3. 15.
    Poitras Center to fuel early careers of 50 young scientists dedicated to psychiatric disorders research— MIT News

    AI 导读 · 50名青年科学家获专项资助，精神疾病研究后继有人，值得期待。

  4. 16.
    OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005— Hacker News
  5. 17.
    OpenAI is well positioned to fast-follow Jev— Hacker News
  6. 18.
    Rabbit’s new AI agent doesn’t need an R1 to run— The Verge

    AI 导读 · Rabbit弃硬件转攻纯软件智能体，折射AI硬件热潮退却后的务实转向。

论文研究6 条

  1. 19.
    Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use— arXiv

    Multi-turn tool-use failures can hinge on a single model call, yet reward variation alone does not reveal which call would benefit from training. When rewards depend on later interactions, their varia…

  2. 20.
    LoRA-generating hypernetworks for efficient on-device LLM generative personalization— arXiv

    On-device large language models (`LLMs'), e.g. running on mobile phones, are ripe for improvement via personalization. The limited compute resources of mobile devices impose limits on model scale and…

  3. 21.
    DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous Manipulation— arXiv

    Dexterous manipulation depends on contact dynamics that are often only partially observable from vision. Recent World-Action Models (WAMs) couple predictive video world modeling with action generation…

  4. 22.
    DolphinBench: Mapping the Pareto Frontier of Agent Memory— arXiv

    Agents today often take real-world actions that depend on long-term memory and context recall over time. However, most current memory benchmarks are built for a conversational question-answer format,…

  5. 23.
    Rare Event Estimation via Iterative Unalignment— arXiv

    As agents are deployed with increased autonomy, even extremely rare events along their stochastic output trajectories can occur and prove catastrophic. Safe deployment therefore does not depend on whe…

  6. 24.
    Emergent Collusion in Long-Horizon LLM Agent Interaction— arXiv

    LLM agents are increasingly deployed in collaborative settings, yet long-term interaction may give rise to undesirable coordination. We study the emergence of collusion in a long-horizon multi-agent e…

技巧与观点6 条

  1. 25.
    How UK AISI and EvalEval Are Making Benchmark Results Reproducible— HuggingFace Blog
  2. 26.
    Transformers now runs llama.cpp quants— HuggingFace Blog
  3. 27.
    Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community— HuggingFace Blog
  4. 28.
    A New Tool Found Malware That’s Guided by an AI Hive Mind—No Humans in Sight— Wired

    Cisco Talos researchers created a new framework for identifying malware and hacking tools that rely on AI chatbots—and quickly discovered something unusual.

  5. 29.
    Anthropic is setting up a biology lab where Claude guides robots through drug experiments— The Decoder

    Anthropic is building its own biology lab to push AI-driven drug development beyond computer simulations. The article Anthropic is setting up a biology lab where Claude guides robo…

  6. 30.
    How to Use AI With Your Privacy Intact— Wired

    Your conversations with AI chatbots are both highly personal and deeply vulnerable to surveillance. Here’s how you can protect yourself.

快讯

日报生成时间：2026-09-28
← 2026-09-22最新日报2026-09-24 →