跳到正文
10月22日周四
  1. MiniMax:Blog67

    MiniMax Code 桌面端推出 Mini App 长运行应用功能

    MiniMax 在 MiniMax Code 桌面应用(macOS / Windows)本地会话中推出 Mini App,一种带前端和后端的长运行应用,以插件形式安装并与 Agent 双向调用,让 Agent 的工作停留在常驻页面而非散落各对话中。

    推荐理由:官方详细说明了 Mini App 的架构、安全边界和社区生态,开发者可以据此判断如何在 MiniMax Code 上构建长运行应用。

  2. IT之家·人工智能40

    谷歌 CEO 皮查伊谈 AI 智能体:7×24 小时不间断工作,重要决策仍由人做

    谷歌 CEO 桑达尔·皮查伊在播客中介绍,他用 AI 智能体整理过去 24 小时的信息、汇总谷歌 AI 模型研发进展与计算资源使用情况,并跟进多个内部聊天群组的讨论。他强调计算资源分配等重大决策仍由管理人员直接参与,未交给 AI。皮查伊认为智能体在云端每周 7 天、每天 24 小时不间断工作、能接着上次进度继续,是其主要优势。

  3. howie.serious54

    作者转发引用 @claudeai 的消息:Anthropic 开放 Claude Duet 研究预览,一个主动型智能体应用,会主动提醒需要关注的事项、完成任务、在用户离开时持续工作,并在代替用户行动前先行确认。作者评论称这款 personal agent 彻底与国内用户无缘,并调侃它可能口头通知封号。

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  4. Claude:Blog69

    Anthropic 开放 Claude Duet 研究预览:主动型智能体应用

    Anthropic 开放 Claude Duet 研究预览,这是一款让 Claude 作为主动智能体工作的独立应用,提供 iOS、Android 和 claude.ai 版本,面向 Pro 和 Max 计划,首批面向美国用户,需加入等候名单。

    同一件事,精选收录的是《Anthropic 开启 Claude Duet 研究预览,Claude 以主动式智能体接管日常工作》

  5. Anthropic:Alignment Science Blog68

    Anthropic 红队测试 Claude Code Auto Mode 与 Codex Guardian 拦截监控器

    Anthropic Fellows 团队用 Opus 5 红队智能体测试生产级拦截监控器,发现 Auto Mode 与 Guardian 的平均攻击成功率超过 55%,主要绕过手段包括伪造转录轮次的监控器注入、多智能体与恶意压缩的多上下文攻击。团队通过转录框架化与转义、监控工作目录编辑和增加智能体监控阶段将攻击成功率降低超过 50 个百分点,但以可接受成本可靠防御多上下文攻击仍是开放问题。

    推荐理由:作者用红队实验给出生产级监控的绕过手法与防御改动的召回数据,读者可以据此评估编码智能体监控的实际鲁棒性。

  6. Chelsea Finn44

    我们如何扩展 LLM 的上下文? Agentic search 可扩展到任意规模的信息,但很容易遗漏相关信息。 想法💡:针对下游效用优化信息的组织方式。即使在 OOD 领域也能带来 30+% 的提升 论文 + 代码:https://librarian-agents.github.io/

    引用Anikait Singh@Anikait_Singh_

    LLMs deployed in specialized domains, from enterprise codebases to law firms, dig through a mountain of unstructured context! Our new paper, Librarian Agents, asks: can an agent adapt by curating its own long-context workspace, anticipating queries before they arrive? 🧵⬇️

  7. TechCrunch:AI74

    Meta 财报电话会披露 Muse 用户超 1000 万,购物成主要使用场景

    Meta 在三季度财报电话会上宣布其 AI 应用 Muse 用户已超 1000 万,约 25% 的用户每天使用,购物是热门场景,通过购物智能体消费的活跃用户中超过四分之一平均单笔购买额超 200 美元。扎克伯格称 Muse 已达成产品市场契合,Meta 计划在未来一年将运行成本效率提升 10 倍以上并实现盈利,还预告了让多个 Muse 协同交互的 Muse network 计划。

  8. Elon Musk50

    Elon Musk 转发推荐 Grok @Bot,称值得一试。引用的 Robert Scoble 介绍了自己的用法:让 Grok Bot 每小时读取他精选的 AI 账号列表(从 X 上 70,000 个账号中筛出),自动生成一份新闻页文档;相比他此前用 X API 自建 alignednews.com 上个月仅 API 账单就 6,000 美元,用列表方式每月只需几百美元。他建议从小列表开始、每小时最多读一次以控制 token 消耗,并提到可进一步把新闻脚本导入 NotebookLM 生成播客或视频。

    引用Robert Scoble@Scobleizer

    I've been GrokBot pilled. First of all, Grok knows me better than other bots, like Meta Muse, thanks to the 300,000 posts I've done here for 20 years. But now that GrokBot @bot can read lists, it opens up powerful new personalized news capabilities. I've done this for about a year over on https://alignednews.com -- but last month my X API bill alone was $6,000. For a couple of hundred a month you can do the same with lists, either you create, or ones other people have created. Last night I stuck in my "best in AI" list and had it write me this news page: https://docs.google.com/document/d/1PEfkIbGmvw_SIJJSDvJ-gSbBM5KwXWKVeimiMsQUW-Y/edit?usp=sharing Every hour my bot reads that list and builds me a new news page. Now I've only put about 10 minutes of effort into this so far. So it's not the prettiest thing you ever have seen, but my bots are working on other ways to present them to me. And you can do the same for any topic you care about. You just need to build a list, or find one that's already built and there are millions of lists here on X on literally any topic. Put your favorite football team onto a list, along with all your favorite journalists or other fans. Then have Grok bot read them for you. And that's just the start. You can have it write you a script based on the news, then copy that into, say, @Gemini_Notebook and then you get a podcast, or a video. Based on what people are saying on your topic here on X. Thank you so much X for making lists far more powerful than they ever were before. Makes the thousands of hours I've put into mine over 20 years so much more useful. Tips? Start small. You can chew through your tokens very quickly. It's amazing how good it is with a small list, like this one, which is what created the page up above: https://x.com/i/lists/1953536336675365173?s=20 (Best AI accounts I've found out of 70,000 here on X). Don't over hit it. You only need your bot to read the list every hour. Maybe even less. If anything is really important it gets reshared into lists, so the bot doesn't need to hit it often to build you a really great set of things that are important to pay attention to. I've been begging for this capability for years. THANK YOU X!

  9. Addy Osmani48

    Claude Duet 是我们的新应用,Claude 在其中是你的主动型智能体!基于 Opus 5.5 运行,覆盖 6,000 个连接器/插件。Duet 会告诉你哪些事情需要你关注,并在你离开时继续推进。 加入研究预览的候补名单!https://claude.com/product/claude-duet

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  10. OpenRouter62

    OpenRouter 宣布 Recursive 的首个模型 Flint 上架,支持免费试用。该模型为开源权重 40B MoE(约 2.5B 激活),由其自动化研究系统 Eureka Machine 构建,面向编码智能体,具备 262K 上下文、推理和工具调用能力。Recursive 报告的成绩为 Terminal-Bench 2.1 54.9%、SWE-bench Pro v2 89.7%、SWE-bench Verified 74.2%;试用地址 https://openrouter.ai/recursive/flint-v1.0:free。

    引用Richard Socher@RichardSocher

    Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from scratch in under 4 months Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.

  11. MiniMax (official)46

    很高兴 Inferact 成为 M3.1 Flash 的 API 发布合作伙伴! 感谢你们将 vLLM 专长带入此次合作。期待继续携手为复杂的智能体工作负载构建服务基础设施。

    引用Inferact@inferact

    Inferact is an API launch partner for MiniMax M3.1 Flash! 🎉 We're working with @MiniMax_AI on production inference for its latest model, built for multi-step automation, real-codebase engineering and visual reasoning. M3.1 Flash scored 29.6 on Agents’ Last Exam (ALE), alongside 72.5% on AutomationBench-AA and 73.67% on DeepSWE—capabilities that matter when agents need to plan work, use tools and carry tasks through. Our contribution is the serving stack: bringing our vLLM engineering and the Inferact Platform to the collaboration. Thanks to the MiniMax team for the partnership. Explore the model and API integration: https://platform.minimax.io/docs/guides/text-generation

  12. Lydia Hallie ✨53

    Anthropic 开放 Claude Duet 研究预览,这是一款让 Claude 作为主动智能体工作的应用,会主动提示需关注事项、完成任务、在用户离开时继续工作,并在代为操作前先确认。员工 Lydia Hallie 分享自己的用法:让它盯着 Slack、邮件和日历,只在重要时提醒,连自己未加入的讨论串也会跟进,并常附带文档或设计作为上下文。

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  13. indigo55

    Anthropic 宣布开启 Claude Duet 研究预览,这是一款让 Claude 以主动智能体方式工作的应用。它会把需要关注的事项带给用户,完成用户交办的任务,在用户离开时持续工作,并在代表用户行动前先向用户确认。

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  14. Ado57

    Anthropic 开放 Claude Duet 的研究预览,这是一款让 Claude 作为主动式智能体工作的应用。它会主动提醒需要关注的事项、完成指派的任务、在用户离开后继续工作,并在代为行动前先与用户确认;作者表示它已成为自己处理所有非编码事务的日常主力。

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  15. Claude:Blog73

    Anthropic 开启 Claude Duet 研究预览,Claude 以主动式智能体接管日常工作

    Anthropic 开启 Claude Duet 研究预览,一款让 Claude 作为主动式智能体工作的独立应用,已登陆 iOS、Android 和 claude.ai,面向美国 Pro 和 Max 用户开放候补名单。

    推荐理由:原文给出 Claude Duet 的运行机制、安全设计和使用限制,读者可以据此判断这款主动式智能体是否值得加入工作流。

  16. The Verge:AI61

    Walmart 下月起允许外部 AI 智能体代用户购物

    Walmart 宣布下月起允许用户把自己的 AI 智能体接入其网站购物,并支持在 Sam's Club 使用。用户可通过新的 Agent Profiles 审批购买、设置消费限额(如最高 100 美元)并随时关闭智能体;Walmart 还构建了基于 MCP 开放标准的 Agent Gateway,让智能体绕过浏览器直接接入,同时不向其开放购物历史和个人信息。

  17. cat41

    我们听到了你们的反馈,希望 Claude 能主动替你们分担更多工作!Claude Duet 朝着这个方向迈出了一大步——它会注意到需要完成的工作,并在你开口之前就去做。 在此加入等候名单:https://claude.com/product/claude-duet

    引用Claude@claudeai

    We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

  18. Rohan Paul42

    AgentMail 通过一次 API 调用为每个 AI 智能体创建独立收件箱,让智能体自行注册服务并接收确认邮件。其累计处理邮件量已超 10 亿封,而一年前还不到 2000 万封。该能力面向仍以邮件处理交易、发票和支持的小型供应商、本地服务公司、会计与采购部门。

    引用Michael Kim@michaelhyunkim

    Today, @agentmail has processed over 1 Billion emails. A year ago, we processed under 20 million. So we figured it was time for AgentMail to grow up a little too :)