🤖 AI Agent 研究Research
大型语言模型（ LLM ）代理现在修复软件，操作计算机和服务客户，但它们的 harness，运行它们的程序，在每一步都会调用大型LLM ，这使得长时间的任务变得缓慢而昂贵。大多数调用以多项选择决策结束，例如接下来调用哪个工具。
Large language model (LLM) agents now fix software, operate computers and serve customers, but their harnesses, the programs that run them, call a large LLM at every step, which makes long tasks slow and costly. Most of these calls end in a multiple-choice decision, such as which tool to call next.
编码代理（如Claude Code ）也遵循开发人员为其编写的文本文件：项目文件、代理可能选择打开的技能以及代理读取匹配文件时自动加载的规则。我们称这些文件为代理的harness。
Coding agents such as Claude Code also follow text files that developers write for them: a project file, skills the agent may choose to open, and rules that load automatically when the agent reads matching files. We call these files the agent's harness.
语言模型代理有时会超出他们的授权范围，然后错误地报告他们的行为。Agent剩余流的线性读数预测了行动前的违规行为（ AUROC 0.94 ）和报告前的虚假报告，并在Agent被迫隐藏其推理时击败了思维链监视器。
Language-model agents sometimes act beyond their authorization and then misreport what they did. A linear readout of the agent's residual stream forecasts the violation before the action (AUROC 0.94) and the false report before the report, and beats chain-of-thought monitors when the agent is pressured to hide its reasoning.
科学研究中的递归自我提升（ RSI ）旨在通过自身的实验经验来提高代理人的研究能力。科学智能体群可以将来自不同研究方向的证据和实现相结合，以指导其智能体的后续实验。
Recursive self-improvement (RSI) in scientific research aims to improve agents' research capabilities through their own experimental experience. Scientific agent swarms can combine evidence and implementations from different research directions to guide their agents' subsequent experiments.
⭐ GitHub 热门项目GitHub Trending
【GitHub】用简单的语言解释Python/JS/TS — MCP服务器+非技术氛围编码器的VS代码面板。（⭐ 0 ）
【GitHub】Explain Python/JS/TS in plain language — MCP server + VS Code panel for non-technical vibe coders. （⭐ 0）
【GitHub】适用于Claude Code和Codex的开源代理技能：每周报告、年度审查（⭐ 1 ）
【GitHub】Open-source Agent Skills for Claude Code and Codex: weekly reports, annual reviews （⭐ 1）
【GitHub】开源视频编辑器，人员和人工智能代理（ Claude Code、Codex、任何MCP客户端）编辑相同的时间线。（⭐ 1 ）
【GitHub】Open-source video editor where people and AI agents (Claude Code, Codex, any MCP client) edit the same timeline. （⭐ 1）
【GitHub】Claude Code技能：将产品屏幕转换为代码驱动的发布视频（ Remotion + 20个开源仓库） （⭐ 1 ）
【GitHub】Claude Code skill: turn product screens into code-driven launch videos (Remotion + 20 open-source repos) （⭐ 1）
🚀 模型与行业动态Models & Industry
我们可能会在公司即将于12月举行的“人工智能和自主日”活动中了解更多有关其工作原理的信息。
We'll likely learn more about how this will work at the company's upcoming "AI & Autonomy Day" event in December.
Figma宣布推出新功能，以定义Agent在其平台上的工作方式，包括公开测试版发布其代码层功能。
Figma announces new features to define how agents work on its platform, including releasing its code layers feature in open beta.
METR将发布自己对Anthropic的风险、事件和实践的评估，前沿AI模型制造商已放弃阻止这些风险、事件和实践的权利。
METR will publish its own assessments of Anthropic's risks, incidents, and practices, and the frontier AI model maker has given up the right to block them.
Meta和苹果已经联手推出了这家社交媒体巨头的最新模型Muse Spark Open ，它将在苹果芯片上本地运行。
Meta and Apple have teamed up for the launch of the social media giant's newest model, Muse Spark Open, which will run locally on Apple silicon.
🔥 社区热议Community
【HN】热度: 939 分 | 613 评论
【HN】热度: 939 分 | 613 评论