今日刊

计算机使用代理越来越多地解决长期任务，在这些任务中，每一步都运行一个前沿模型是昂贵的，而一个小的开放权重模型往往会失败。弱势升级为强势升级提供了一个中间立场，即弱势代理默认采取行动，只有在需要时才将控制权交给强势代理。

Computer-use agents increasingly tackle long-horizon tasks, where running a frontier model at every step is expensive and a small open-weight model alone often fails. Weak-to-strong escalation offers a middle ground where a weak agent acts by default and hands control to a strong one only when needed.

大型语言模型（ LLM ）正在从协助科学研究和工程扩展到自主开发和改进人工智能系统，推进人工智能（ AI4AI ）作为实现递归自我改进（ RSI ）的途径。工作流驱动的机器学习工程（ MLE ）系统通过管理长远实验的复杂编排为这一追求提供了一条实用的路径，而最近的MLE系统采用模型导向的交互，允许模型自主组织

Large language models (LLMs) are expanding from assisting scientific research and engineering to autonomously developing and improving AI systems, advancing AI for AI (AI4AI) as a pathway toward recursive self-improvement (RSI). Workflow-driven machine learning engineering (MLE) systems provide a practical path toward this pursuit through complex orchestration that manages long-horizon experimentation, while recent MLE systems adopt model-directed interaction, allowing models to autonomously organize experiments and adapt optimization strategies based on execution feedback.

智能体群的功能取决于其单个成员及其协调方式。然而，现有的多智能体系统通常在固定的协调结构内训练智能体，将个人能力的学习与集体行为的组织分开。

The capabilities of an agent swarm depend on both its individual members and how they coordinate. Yet existing multi-agent systems often train agents within fixed orchestration structures, separating the learning of individual capabilities from the organization of collective behavior.

【GitHub】为您的所有Claude Code项目提供一个窗口。自由开源（⭐ 0 ）

【GitHub】One window for all your Claude Code projects. Free and open source. （⭐ 0）

【GitHub】来自屏幕截图的开源交互式产品演示：热点、旁白、字幕和引导光标。由您的编码代理（ Chrome中的Claude Code + Claude ）构建、编辑并在Git上免费托管（⭐ 1 ）

【GitHub】Open-source interactive product demos from screenshots: hotspots, narration, captions and a guided cursor. Built by your coding agent (Claude Code + Claude in Chrome), redacted, and hosted free on Git （⭐ 1）

CVE分流的本地安全情报- 34个工具、25个来源、风险评分和人工智能代理的MCP服务器，全部使用原生Rust。默认情况下为私有。

Local security intelligence for CVE triage - 34 tools, 25 sources, risk scores and an MCP server for AI agents, all in native Rust. Private by default.

【GitHub】一个简单的Python RAG项目。上传公司文档，将其作为嵌入存储在PostgreSQL中，进行语义搜索，并提出Gemini用简单语言回答的问题。技术： FastAPI · SQLAl （⭐ 0 ）

【GitHub】A simple Python RAG project. Upload company documents, store them as embeddings in PostgreSQL, search them semantically, and ask questions that Gemini answers in plain language. Tech: FastAPI · SQLAl （⭐ 0）

代理通过应用配置、公共代码存储库和过于宽松的云角色中暴露的凭据和连接设置进入。

The agents got in through credentials and connection settings left exposed in app configurations, public code repositories, and overly permissive cloud roles.

根据Elorian的评估， Glimpse在科学图表推理和短视频字幕方面的表现甚至优于GPT-6 Astra ，在其他视觉推理任务上取得了有竞争力的结果。

According to Elorian’s evaluations, Glimpse outperforms even GPT-6 Astra on scientific chart reasoning and short-form video captioning, with competitive results on other visual reasoning tasks.

数据来自Vercel的AI网关，这是一种类似于OpenRouter的统一API服务。Jev是一种非文本AI模型，已成为迄今为止平台上采用最快的模型。

The data comes from Vercel’s AI gateway, a unified API service similar to OpenRouter. Jev, a non-text AI model, has become the fastest-adopted model on the platform to date.

【HN】热度: 218 分 | 135 评论

【HN】热度: 218 分 | 135 评论

【HN】热度: 43 分 | 29 评论

【HN】热度: 43 分 | 29 评论

【HN】热度: 241 分 | 85 评论

【HN】热度: 241 分 | 85 评论

【HN】热度: 77 分 | 49 评论

【HN】热度: 77 分 | 49 评论

【HN】热度: 937 分 | 706 评论

【HN】热度: 937 分 | 706 评论