🤖 AI Agent 研究Research
形成准确的心智理论（ ToM ）是多智能体协作成功的关键，需要在不确定性下进行推理，以捕捉多个可能的合作伙伴状态。这在大型语言模型（ LLM ）代理的原生文本表示空间中很难做到，因为它们更容易描述单个合作伙伴状态。
Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state.
无论是在完成任务还是过滤合成训练数据时，代理都需要可靠的方法来验证其输出。我们研究验证器生成：从可用于解决任务的指令和环境中，代理编写一个可执行验证器，该可执行验证器必须接受每个有效解决方案并拒绝无效解决方案。
Agents need reliable ways to validate their outputs, both when completing tasks and when filtering synthetic training data. We study verifier generation: from the instruction and environment available for solving a task, an agent writes an executable verifier that must accept every valid solution and reject invalid ones.
轻触“登录”后会出现登录屏幕；轻触“查看订单”后会出现相同的屏幕，这是一种攻击。因此，对于GUI代理来说，安全是过渡的属性，而不是屏幕的属性，而仅检查屏幕的监视器可以通过重复使用合法的监视器来消除。
A login screen that appears after a tap on Sign in is expected; the same screen after a tap on View order is an attack. For GUI agents, safety is therefore a property of the transition rather than of the screen, and a monitor that inspects only screens can be defeated by reusing a legitimate one.
医疗工作流程通常需要多个相互依赖的分析，但现有的专家和通才AI模型并不是为了协调它们而设计的。我们推出了MMedAgent-Plus ，这是一种多模式医疗工具使用代理，它将视觉语言规划器与18种领先的专业模型相结合，涵盖12项任务和8种模式。
Medical workflows often require multiple interdependent analyses, but existing specialist and generalist AI models are not designed to coordinate them. We introduce MMedAgent-Plus, a multimodal medical tool-use agent that combines a vision-language planner with 18 leading specialist models spanning 12 tasks and 8 modalities.
⭐ GitHub 热门项目GitHub Trending
【GitHub】Anything to API (万物皆可2API): put an AI agent, a script, or even yourself behind an OpenAI-compatible API（⭐ 21）
【GitHub】Anything to API (万物皆可2API): put an AI agent, a script, or even yourself behind an OpenAI-compatible API（⭐ 21）
【GitHub】Mac内部的灯光。免费、开源的Mac应用程序，可与您的Claude Code配合使用。（⭐ 1 ）
【GitHub】The light inside your Mac. Free, open-source Mac app that works with your Claude Code. （⭐ 1）
【GitHub】使用AI代理回收磁盘空间：自上而下测量，按风险对发现进行分组，并仅删除您批准的内容。7种语言提示。macOS、Linux、Windows。（⭐ 13 ）
【GitHub】Reclaim disk space with an AI agent: measure top-down, group findings by risk, and delete only what you approve. Prompts in 7 languages. macOS, Linux, Windows.（⭐ 13）
【GitHub】本地首个开源工具，用于跨Codex和Claude Code安全同步和验证AI编码代理指令。（⭐ 1 ）
【GitHub】A local-first open-source tool for safely synchronizing and validating AI coding agent instructions across Codex and Claude Code. （⭐ 1）
【GitHub】MCP服务器，为您的AI 智能体提供投资委员会。多代理股权研究，具有引用的来源、可重复的回溯测试和可审计的跟踪记录。适用于Claude Code、Codex、Cursor和OpenCode。本地优先和开源。（⭐ 1 ）
【GitHub】An MCP server that gives your AI agent an investment committee. Multi-agent equity research with cited sources, reproducible backtests, and an auditable track record. Works with Claude Code, Codex, Cursor, and OpenCode. Local-first and open source.（⭐ 1）
🚀 模型与行业动态Models & Industry
麻省理工学院的研究人员在一个虚拟岛屿上放逐了13个人工智能，他们通过发明游戏，哀悼丢失的戒指，最终在工作中变得更好。
MIT researchers marooned 13 AIs on a virtual island, and they responded by inventing games, mourning a lost ring and, eventually, getting better at their jobs.
新功能允许智能体登录客户的个人账号，同时保护用于保护这些账号的私人凭据。
The new feature allows agents to log into a customer's personal accounts while still protecting the private credentials that secure those accounts.
这家拼车巨头以前为ChatGPT Work部署了一个规模小得多的试点，但现在正在全公司范围内采用企业代理。
The ride-share giant previously deployed a much smaller pilot for ChatGPT Work but is now taking the enterprise agent company-wide.
Spotify联合首席执行官Gustav Söderström解释了为什么人工智能代理的崛起可能是音乐服务的好消息。
Spotify co-CEO Gustav Söderström explains why the rise of AI agents could be good news for the music service.
🔥 社区热议Community
【HN】热度: 246 分 | 310 评论
【HN】热度: 246 分 | 310 评论
【HN】热度: 120 分 | 59 评论
【HN】热度: 120 分 | 59 评论
【HN】热度: 309 分 | 162 评论
【HN】热度: 309 分 | 162 评论