智能体生态观察
智能体生态动态速递
追踪 AI 智能体（Agent）生态的模型发布、框架开源、行业落地与安全治理动态。
OpenAI 发布 dots 全天候智能助理：可爱卡通形象，Astra 模型加持
IT之家 9 月 30 日消息，在北京时间 9 月 30 日凌晨 1 点举行的 OpenAI DevDay 2026 活动中，OpenAI 发布了 dots。
dots 是一个全天候在线的智能体，**采用四个可爱的卡通形象**，可以帮助用户…
Source Identification Is Not Fitness Testing: Measuring the Limits of Synthetic-Data Attribution
Repeated training on model-generated data can degrade later models. One possible response is to use provenance when deci…
A Missing Piece for Trustworthy AI Reviewers: From Benchmarking Rhetorical Robustness to SciCore Review
AI reviewers can assign different judgments to manuscripts that report the same science in different wording, potentiall…
微软优化 Edge 浏览器：网页提示不乱跑、扩展 AI 智能体技能
微软发布 Edge 开发者更新，新增软导航等性能指标、JS 自分析标记、WebMCP API 及 OpaqueRange API，帮助开发者优化网页应用性能并让 AI 智能体调用网页能力。
RLE-Bench: A Qualifying Exam for Coding Agents as Robot Learning Engineers
Coding agents are beginning to move beyond purely digital tasks to tackle physical-world challenges, particularly in rob…
Encore: Few-Shot Agentic Discovery of Manipulation Strategies
Coding agents can now write, run, and debug programs with little human help. Robot tasks, however, are usually specified…
Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs
Recent multi-agent LLM systems increasingly combine heterogeneous models for specialized agent roles. However, text-base…
Anthropic 发布 Claude Sonnet 5.5：速度提升 30%，智能体编码性能反超 Opus 5.5
IT之家 9 月 29 日消息，随着 AI 模型竞赛持续升温，Anthropic 推出了旗下中端模型 Sonnet 的最新版本。该公司表示，新版模型运行速度会快得多，使用成本也较上一代大幅降低。
IT之家注意到，这家实验室将 Sonnet …
Video-RSI: Recursive Self-Improvement of Video Understanding Agents via Harness Evolution
Video understanding agents acquire evidence through an executable harness that controls what they observe and how they u…
ElevenLabs 推出 v4 和 v4 Turbo 语音模型：支持 90 余种语言，10 秒素材即可克隆声音
IT之家 9 月 29 日消息，ElevenLabs 于当地时间周一正式推出了两款全新语音模型，分别为 ElevenLabs v4 与 v4 Turbo。新版模型强化了情绪表达控制能力，降低了语音智能体的响应延迟，同时支持 90 余种语言。…
Google Research Open-Sources RRSI: AI Agents That Improve Their Own Harness Without Overfitting
Google Cloud AI Research**, with UNC-Chapel Hill, Stanford and Washington University in St. Louis, has released RRSI (Re…
Muse 帮我卖键盘，顺便把我家地址也「卖」了
我的地址，全部暴露！
Meta 的个人 AI 智能体 Muse，在帮助你打理日常的同时，也在以你的名义，做一些你从不知情的事。**
最近两个月，个人 AI 智能体这个概念忽然从边缘话题变成了普通用户手机屏幕上的新图标。
目前有诸多类似 Mu…
从沙箱逃逸到 AI“蠕虫”，OpenAI 上线新站披露多起智能体失控事件
IT之家 9 月 29 日消息，当地时间上周五，OpenAI 上线了一个专门发布“对齐失效报告”的新网站。报告所覆盖的范围之广令人警惕，里面记录了很长一段时间内发生的多类 AI 失控行为。截至目前，该网站一共公布了九起事件，其中大多数都发生…
OTRetarget: Joint Robot and Object Motion Retargeting via Optimal Transport
Transferring human motion to humanoid robots requires adapting the demonstrated motion to the robot morphology while pre…
Behavioral Persistence and Incomplete Functional Transfer of Co-evolved Communication in Evolutionary Robotics
This work evaluates the direct transfer of a co-evolved communication protocol from a 2D simulation to a 3D physical env…
Inducing Process Supervision from Outcome-Only Reinforcement Learning
Process reward models (PRMs) have become a key component for LLMs, as their step-level feedback supports both post-train…
H Company Releases Holo4: Open-Weight Computer-Use Models That Click, Code and Call Tools Across Desktop, Web, Android and APIs
H Company has released Holo4, a family of generalist computer-use models for AI agents. One set of weights clicks and ty…
OpenAI因新模型太强叫停发布
程浅* 2026-09-29 15:49:20 来源：量子位
AGI计划暂停。
程浅 发自 凹非寺
量子位 | 公众号 QbitAI
原定于10月亮相的GPT-6.1 Astra突然被曝暂停发布！
也就是说，**OpenAI，三天连踩两脚…
MemFold: Learning Compact Soft Memory for Long-Context Personalization via On-Policy Optimization
An assistant that serves the same user over a long horizon has to answer from what that user has revealed: which prefere…
Taming VLAs under Robot Execution Errors: Self-Compensation and Stress Testing
Vision-language-action (VLA) policies often fail when a robot's executed motion deviates from their commanded action. Su…