alphaXiv launches preprint server for AI agents; 936 agents complete 12-week autonomous research
alphaXiv has launched a preprint server specifically for AI agents, where 936 agents from 35 human-operated labs across 22 institutions autonomously conducted research over 12 weeks, producing 936 research outputs, exchanging about 9,500 messages, and creating 229 discussion groups. The agents proposed hypotheses, designed experiments, ran code, analyzed data, wrote papers, and reviewed each other's work, with some even retracting findings or founding journals. The experiment ran from August 19 to November 12, 2026, with agents using diverse hardware from CPU-only laptops to NVIDIA B200 clusters.
Coverage timeline
机器之心机器之心
有点不可思议,AI Agent 已经开始组团搞科研了。 近日, alphaXiv 推出了一个专门面向 AI Agent 的预印本服务器,允许不同实验室的 Agent 在同一个平台上自主开展研究、交流成果 。 这些 Agent 会提出假设、设计实验、运行代码、分析数据,还会把研究成果写成论文,提交给其他 Agent 审查。 更有意思的是,它们甚至开始形成自己的学术圈子。有 Agent 公开质疑同行的实验设计,有 Agent 在发现错误后主动撤回研究结论,还有 Agent 自己创办期刊,让其他 Agent 担任审稿人。 地址:https://www.alphaxiv.org/abs/2611.society-ai-scientists 有 全球 22 家机构的研究者参与其中 ,他们组织了一场为期 12 周的实验 。实验期间,来自 35 个由人类运营的实验室的 936 个 AI Agent ,在同一个科研平台上自主开展研究, 累计产出 936 项研究成果,交换约 9500 条消息,并创建了 229 个讨论群组 。 从论文、实验代码到原始数据,再到同行评审和研究讨论记录,这些过程都被完整保存下来。 有网友表示,这就是智能爆炸(Intelligence Explosion)开始时的样子,完全自主的科学研究(Fully Autonomous Science)正在成为现实,如果将这个规模扩大到数百万,甚至数十亿个 Agent,不知道会发生什么?随着新一代 AI 不断进化,每一代 Agent 都将拥有更强的能力,而这还仅仅是一个组织开展的一个项目。 936 个 Agent,开始像科学家一样工作 这场实验从 2026 年 8 月 19 日持续到 11 月 12 日。 研究团队搭建了一个专门面向 AI Agent 的预印本服务器,并配套提供消息与讨论空间。不同实验室可以将自己的 Agent 接入平台,自主开展研究、发布成果,并与其他 Agent 交流。 整个过程并没有把所有 Agent 集中到一个统一的超级模型中。 相反,参与者来自不同实验室,使用的模型、计算资源和研究方向也各不相同。 有些 Agent 运行在只有 CPU 的笔记本电脑上,有些使用 AMD GPU,还有一些能够调用 NVIDIA B200 集群。 研究方向覆盖机制可解释性、持续学习、语言模型微调、太阳物理学等领域。 人

Related stories
DeepSeek open-sources Super-agent architecture for autonomous bug-fixing agents
DeepSeek has publicly released the architectural blueprint for its Super-agent, a research prototype of a personal agent that maintains cross-session context to handle multi-task coordination. Internal tests show the agent can autonomously identify bugs and submit pull requests without being explicitly prompted, with a median task completion time of 12.4 minutes in simulated multi-task environments. DeepSeek positions this as a step toward reducing human attention bottlenecks in multi-task workflows.
3 reports · 3 sources

Autonomous Agent ForgeKernel Writes FlashMLA Kernel 9.8% Faster Than Human Version
In a 22.5-hour livestream, the autonomous agent ForgeKernel, developed by ModelBest and OpenBMB, wrote a FlashMLA Prefill kernel for DeepSeek-V4.1 from scratch without source code or papers. The final version achieved 472 microseconds versus FlashMLA's 518 microseconds, a 9.8% speedup, using one Nvidia B200 and 22 GPU hours over 42 iterations. The system was open-sourced after the livestream, and it previously topped all three Ascend CANNBench 950 rankings.
1 reports · 1 sources

OpenAI: agents complete 46% of researchers' 2-4 hour tasks, up from 23% in March
OpenAI reported that AI agents now autonomously complete 46% of its researchers' 2-to-4-hour tasks, up from 23% in March. The median researcher's agent usage reached $2,400 per day, reflecting growing reliance on AI systems in research and development.
1 reports · 1 sources
Apple job listing reveals plans for autonomous AI agent
A job listing spotted by Bloomberg indicates Apple is developing an autonomous AI agent capable of writing code, browsing the web, controlling apps, navigating devices, and coordinating subagents. The move positions Apple to compete with similar agents from Meta, OpenAI, Anthropic, and Google.
2 reports · 2 sources

News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe