🤖 AI Agent 研究Research
递归自我改善（ RSI ）依赖于评估反馈来评估进展并指导进一步的研究，但反复运行复杂的基准代价高昂，并降低了迭代速度。人类专家通过选择基准子集或设计紧凑型套件来降低成本。我们询问AI 智能体是否可以自动化此设计过程，并引入EvalResearchBench （ ERB ） ，这是自主评估研究的基准。
Recursive self-improvement (RSI) relies on evaluation feedback to assess progress and guide further research, yet repeatedly running complex benchmarks is costly and slows iteration. Human experts reduce this cost by selecting benchmark subsets or designing compact suites. We ask whether AI agents can automate this design process and introduce EvalResearchBench (ERB), a benchmark for autonomous evaluation research.
现代LLM智能体可以从可访问的文档中检索信息，然后披露从中收集的敏感结论。这种马赛克效应给处理机密数据的智能体带来了严重的风险。我们使用30个综合网络安全、医疗和军事示例对其实现进行量化，从受控模拟中的三个LLM智能体获得10,800个最终答案，从两个开源harness中的实时智能体获得1,440个最终答案。
Modern LLM agents can retrieve information from accessible documents, and then disclose a sensitive conclusion assembled from them. This mosaic effect creates a critical risk for agents operating over confidential data. We quantify its realization with 30 synthetic cybersecurity, medical, and military examples, yielding 10,800 final answers from three LLM agents in controlled simulation and 1,440 answers from live agents in two open-source harnesses.
自主研究智能体越来越多地用于自动化大部分科学研究，但它们容易出现研究完整性违规（ RIV ） ：损害研究过程的有效性、可重复性或忠诚度的行为。我们推出了RIDAR ，这是一个用于自动RIV检测和运行时干预的框架。首先，我们介绍RIVT ，这是一个涵盖整个研究生命周期的14种RIV类型的综合分类法。
Autonomous research agents are increasingly used to automate large portions of scientific research, yet they are prone to research integrity violations (RIVs): behaviors that compromise the validity, reproducibility, or faithfulness of the research process. We present RIDAR, a framework for automated RIV detection and runtime intervention. First, we introduce RIVT, a comprehensive taxonomy of 14 RIV types spanning the full research lifecycle.
LLM记忆被广泛用于长期会话理解。智能体记忆系统通常使用突变、结构组织、学习检索策略或其他复杂操作。我们研究了Raw-RAG ，这是一种朴素的RAG基线，使用密集检索和对原始历史进行重新排序，为读者LLM提供证据，并发现它在七个基准模型设置中的六个中在比较的记忆系统中排名第一。
LLM memory is widely used in long-term conversation understanding. Agentic memory systems often use mutation, structural organization, learned retrieval strategies, or other complex operations. We study Raw-RAG, a naive RAG baseline using dense retrieval and reranking over raw history to supply evidence to a reader LLM, and find that it ranks first among the compared memory systems in six of seven benchmark-model settings.
⭐ GitHub 热门项目GitHub Trending
【GitHub】Chrome浏览器中Claude的开源替代品。带上您自己的API密钥，让AI 智能体跨选项卡读取、单击、填充和研究。无账号，无后端，无遥测。（⭐ 4 ）
【GitHub】An open-source alternative to Claude in Chrome. Bring your own API key and let an AI agent read, click, fill, and research across tabs. No account, no backend, no telemetry. （⭐ 4）
【GitHub】将Web作为API提供给您的AI 智能体。适用于Claude Code、Codex和Cursor的MCP服务器+技能：适用于Google Maps、Amazon、Instagram、X、LinkedIn等的自我修复抓取工具。兼容Firecrawl和Apify。（⭐ 5 ）
【GitHub】Give your AI agent the web as an API. MCP server + skill for Claude Code, Codex & Cursor: self-healing scrapers for Google Maps, Amazon, Instagram, X, LinkedIn & more. Firecrawl- and Apify-compatible. （⭐ 5）
【GitHub】AI智能体的纯代码世界引擎：以three.js作为代码构建整个城镇和景观（预alpha ） （⭐ 10 ）
【GitHub】A code-only world engine for AI agents: build whole towns and landscapes in three.js as code (pre-alpha) （⭐ 10）
【GitHub】AI 智能体以证据为先的家谱技能（ Codex、Claude Code ） ：前苏联各地的教会书籍、民事、军事、驱逐和镇压记录。以您的语言回复。（⭐ 538 ）
【GitHub】Evidence-first genealogy skill for AI agents (Codex, Claude Code): church books, civil, military, deportation and repression records across the former USSR. Replies in your language. （⭐ 538）
【GitHub】针对AI 智能体管理的机器的每日漏洞简报—针对OSV + CISA KEV的确定性匹配； LLM仅解释。麻省理工学院（⭐ 13 ）
【GitHub】A daily vulnerability briefing for AI-agent-managed machines — deterministic matching against OSV + CISA KEV; the LLM only explains. MIT （⭐ 13）
🚀 模型与行业动态Models & Industry
从苹果一再推迟Siri AI到OpenAI凌乱的“超级应用”发布，以下是关闭或未达到预期的人工智能项目。
From Apple's repeatedly delayed Siri AI to OpenAI's messy "super app" launch, here's a look at the AI projects that shut down or missed expectations.
我们创建了一个列表，列出了可以在短信中出现的最著名的人工智能智能体，从一般助理到专为家庭、旅行和工作而设计的智能体。
We created a list of the most notable AI agents that can live in your text messages, from general assistants to agents designed for families, travel, and work.
OpenAI将在其一名工程师表示该解决方案不起作用的几周后，从互联网上切断前沿模型。
OpenAI will cut frontier models off from the internet weeks after one of its engineers said that solution wouldn't work.
🔥 社区热议Community
【HN】热度: 440 分 | 209 评论
【HN】热度: 440 分 | 209 评论
【HN】热度: 151 分 | 177 评论
【HN】热度: 151 分 | 177 评论
【HN】热度: 200 分 | 295 评论
【HN】热度: 200 分 | 295 评论
【HN】热度: 33 分 | 21 评论
【HN】热度: 33 分 | 21 评论