{ "schemaVersion": 1, "generatedAt": "2026-06-09T00:39:05.568Z", "source": "aihot.virxact.com public API", "updatePolicy": "Daily 10:00 Asia/Shanghai; curated for Agent Skills, skill ecosystems, and Agent architecture.", "items": [ { "id": "news-5de34055", "url": "https://x.com/claudeai/status/2064139073590104402", "canonicalUrl": "https://x.com/claudeai/status/2064139073590104402", "title": "Claude 东京站活动开放注册", "titleEn": "Final stop： Tokyo. Register to hear directly from the teams behind Claude： http：//claude.com/code-w…", "description": "最终站：东京。\n快来注册，直接聆听 Claude 团队分享：http：//claude.com/code-with-claude/tokyo", "descriptionEn": "最终站：东京。\n快来注册，直接聆听 Claude 团队分享：http：//claude.com/code-with-claude/tokyo", "image": null, "sourceName": "X：Claude (@claudeai)", "sourceCategory": "industry", "publishedAt": "2026-06-09T00:14:39.000Z", "discoveredAt": "2026-06-09T00:39:05.020Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-4492b89c", "url": "https://x.com/hongming731/status/2064137431050534935", "canonicalUrl": "https://x.com/hongming731/status/2064137431050534935", "title": "BestBlogs 早报 · 06-09", "titleEn": "BestBlogs 早报 · 06-09 # Claude Code Agent 工作流 / 循环工程 Loop Engineering / 阳萌安克创新 / Agent 上下文压缩 / 硅谷 AI…", "description": "本期早报聚焦：Claude Code一周年复盘，从辅助写代码到数千Agent协作网络，推出Auto Mode以安全分类模型替代人工审批；循环工程新范式强调工程师职责从提示Agent转向设计系统；小米MiMo与TileRT通过FP4量化和DFlash推测解码，在商用GPU上实现1万亿参数模型1000+ token/s推理；OpenAI概述AGI第三阶段愿景构建自动化AI研究员；另有安克创新创始人阳萌4小时访谈、六大Agent上下文压缩策略对比、杰弗里·辛顿谈AI失控焦虑等。", "descriptionEn": "本期早报聚焦：Claude Code一周年复盘，从辅助写代码到数千Agent协作网络，推出Auto Mode以安全分类模型替代人工审批；循环工程新范式强调工程师职责从提示Agent转向设计系统；小米MiMo与TileRT通过FP4量化和DFlash推测解码，在商用GPU上实现1万亿参数模型1000+ token/s推理；OpenAI概述AGI第三阶段愿景构建自动化AI研究员；另有安克创新创始人阳萌4小时访谈、六大Agent上下文压缩策略对比、杰弗里·辛顿谈AI失控焦虑等。", "image": null, "sourceName": "X：洪明 (@hongming731)", "sourceCategory": "industry", "publishedAt": "2026-06-09T00:08:08.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "automation", "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-94b6b842", "url": "https://x.com/hongming731/status/2064137316407574850", "canonicalUrl": "https://x.com/hongming731/status/2064137316407574850", "title": "BestBlogs 早报 06-09：Claude Code 自主化、循环工程、阳萌访谈", "titleEn": "http：//x.com/i/article/2064136850370101248", "description": "本期早报聚焦AI编程从辅助到自主Agent的拐点。Anthropic复盘Claude Code一周年：Auto Mode用路由分类模型替代人工审批，通过Claude 4.6/4.7实现数千Agent动态协作。Boris Cherny提出\"循环工程\"--工程师应设计自动循环系统（定时自动化、并行工作树等5模块），并警示\"认知投降\"风险。安克创新CEO阳萌4小时访谈阐述从\"浅海\"到\"深海\"战略、第三类公司愿景及AI原生组织变革。", "descriptionEn": "本期早报聚焦AI编程从辅助到自主Agent的拐点。Anthropic复盘Claude Code一周年：Auto Mode用路由分类模型替代人工审批，通过Claude 4.6/4.7实现数千Agent动态协作。Boris Cherny提出\"循环工程\"--工程师应设计自动循环系统（定时自动化、并行工作树等5模块），并警示\"认知投降\"风险。安克创新CEO阳萌4小时访谈阐述从\"浅海\"到\"深海\"战略、第三类公司愿景及AI原生组织变革。", "image": null, "sourceName": "X：洪明 (@hongming731)", "sourceCategory": "tip", "publishedAt": "2026-06-09T00:07:40.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-f22bd0ad", "url": "https://x.com/berryxia/status/2064134399537590712", "canonicalUrl": "https://x.com/berryxia/status/2064134399537590712", "title": "Kimi Code升级：一行安装+视频上下文+插件系统", "titleEn": "Kimi Code一行命令直接把所有coding agent的安装门槛干到零。 还能拖视频当上下文生成LUT文件或者把屏幕录像转成可运行代码！ 官方开源版现在零配置、秒启动，配上Kimi K2.6…", "description": "Kimi Code 开源 coding agent 迎来重大升级：一行 CLI 命令安装、零配置、秒启动；支持拖拽视频作为编码上下文，可参考视频生成 .cube LUT 文件或把屏幕录像转成可运行代码；插件系统上线，可一键拉取股票、财报、学术论文；支持 ACP 协议，对接 JetBrains、Zed，并提供自定义 hooks 扩展工作流。配合 Kimi K2.6 模型使用，视频推理能力大幅增强。", "descriptionEn": "Kimi Code 开源 coding agent 迎来重大升级：一行 CLI 命令安装、零配置、秒启动；支持拖拽视频作为编码上下文，可参考视频生成 .cube LUT 文件或把屏幕录像转成可运行代码；插件系统上线，可一键拉取股票、财报、学术论文；支持 ACP 协议，对接 JetBrains、Zed，并提供自定义 hooks 扩展工作流。配合 Kimi K2.6 模型使用，视频推理能力大幅增强。", "image": null, "sourceName": "X：Berry Xia (@berryxia)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T23:56:05.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "skill-ecosystem", "automation", "coding-agent" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-4a05195c", "url": "https://x.com/berryxia/status/2064133023847821494", "canonicalUrl": "https://x.com/berryxia/status/2064133023847821494", "title": "WWDC 2026 速览：Tim Cook 谢幕，AI 双轨模型与 Siri 大升级", "titleEn": "没有熬夜的兄弟们来看，WWDC 2026 速览！ 另插一嘴，Apple真的没有特别的新东西。 一、历史性时刻 1库克谢幕：Tim Cook最后一次以CEO身份主持WWDC，9月1日将交接给硬件…", "description": "Tim Cook 最后一次以 CEO 身份主持 WWDC，AI 成绝对核心。苹果推出双轨模型：基础层 3B 端侧 AFM 3 Core，进阶层 20B MoE 仅限 iPhone 17 Pro/M4 设备。四大 AI 能力：个人上下文理解、世界知识检索、App Actions、屏幕感知。Siri 升级支持连续对话、跨设备同步、视觉识别及独立 App，仅限英语，中国/欧盟暂不可用。应用生态：Safari 智能标签分组、密码自动更新、来电识别、快捷指令自然语言生成。影像：Image Playground 写实生成与空间重构。开发者工具：Xcode 支持多模型，Core AI Framework 开放本地部署。遗留问题：中国区功能缺失，Agent 多步自动化能力薄弱。", "descriptionEn": "Tim Cook 最后一次以 CEO 身份主持 WWDC，AI 成绝对核心。苹果推出双轨模型：基础层 3B 端侧 AFM 3 Core，进阶层 20B MoE 仅限 iPhone 17 Pro/M4 设备。四大 AI 能力：个人上下文理解、世界知识检索、App Actions、屏幕感知。Siri 升级支持连续对话、跨设备同步、视觉识别及独立 App，仅限英语，中国/欧盟暂不可用。应用生态：Safari 智能标签分组、密码自动更新、来电识别、快捷指令自然语言生成。影像：Image Playground 写实生成与空间重构。开发者工具：Xcode 支持多模型，Core AI Framework 开放本地部署。遗留问题：中国区功能缺失，Agent 多步自动化能力薄弱。", "image": null, "sourceName": "X：Berry Xia (@berryxia)", "sourceCategory": "tip", "publishedAt": "2026-06-08T23:50:37.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-23ea8aee", "url": "https://x.com/natolambert/status/2064132565003583962", "canonicalUrl": "https://x.com/natolambert/status/2064132565003583962", "title": "Nathan Lambert：AI领域对持续学习与样本效率的执念是错误方向", "titleEn": "I feel like the obsession with continual learning / sample efficiency leads the field in the wrong d…", "description": "Nathan Lambert 批评 AI 领域过度关注持续学习与样本效率，认为这如同专注于弥补弱点而非最大化优势。人类大脑虽是存在性证明，但未必是 AI 最佳路径。前沿实验室实际加速推进现有开发树，对进步有利，但对安全与地缘政治影响复杂。他引用 @dwarkesh_sp 的观点：数据是进步主要驱动力，开源与后来者可通过从公开 API 蒸馏数据快速追赶前沿，而超参数、训练技巧等难以复制。他认为未来已来，AGI 研究应拥抱未知、规模化资源，而非等待不确定的科学突破。", "descriptionEn": "Nathan Lambert 批评 AI 领域过度关注持续学习与样本效率，认为这如同专注于弥补弱点而非最大化优势。人类大脑虽是存在性证明，但未必是 AI 最佳路径。前沿实验室实际加速推进现有开发树，对进步有利，但对安全与地缘政治影响复杂。他引用 @dwarkesh_sp 的观点：数据是进步主要驱动力，开源与后来者可通过从公开 API 蒸馏数据快速追赶前沿，而超参数、训练技巧等难以复制。他认为未来已来，AGI 研究应拥抱未知、规模化资源，而非等待不确定的科学突破。", "image": null, "sourceName": "X：Nathan Lambert (@natolambert)", "sourceCategory": "tip", "publishedAt": "2026-06-08T23:48:47.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-89a76982", "url": "https://x.com/berryxia/status/2064130947508035738", "canonicalUrl": "https://x.com/berryxia/status/2064130947508035738", "title": "Google NotebookLM 大更新：智能体能力 + 更强推理", "titleEn": "兄弟们！Google NotebookLM 大更新了！ NotebookLM一夜之间从你的笔记小助手 ！ 直接进化成能独立带你搞定复杂多步研究的agent，把一堆靠云端幻觉混日子的研究工具直接干沉默…", "description": "Google NotebookLM 迎来重大升级，在聊天中注入智能体（agentic）能力、更先进的推理逻辑以及一整套新输出格式。它可自主拆解复杂多步研究任务，逐步推理并生成结果；能主动从网络挖掘新资料，但最终答案严格基于用户批准过的来源，大幅减少幻觉。这让人机协作从\"对话\"升级为\"搭档\"。该更新已面向 Google AI Ultra 订阅用户逐步推送。", "descriptionEn": "Google NotebookLM 迎来重大升级，在聊天中注入智能体（agentic）能力、更先进的推理逻辑以及一整套新输出格式。它可自主拆解复杂多步研究任务，逐步推理并生成结果；能主动从网络挖掘新资料，但最终答案严格基于用户批准过的来源，大幅减少幻觉。这让人机协作从\"对话\"升级为\"搭档\"。该更新已面向 Google AI Ultra 订阅用户逐步推送。", "image": null, "sourceName": "X：Berry Xia (@berryxia)", "sourceCategory": "tip", "publishedAt": "2026-06-08T23:42:22.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.86, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-ccfb9da0", "url": "https://x.com/ericzakariasson/status/2064122350866682100", "canonicalUrl": "https://x.com/ericzakariasson/status/2064122350866682100", "title": "Cursor 三循环：不稳定测试清除器", "titleEn": "here are 3 loops you can run in cursor 1. Flaky-test exterminator /loop run my test suite 20 times…", "description": "以下是在 Cursor 中运行的 3 个循环\n\n1. 不稳定测试清除器\n\n/loop 运行测试套件 20 次，收集所有间歇性失败，修复或隔离不稳定测试，直到连续 5 次完全通过才停止。", "descriptionEn": "以下是在 Cursor 中运行的 3 个循环\n\n1. 不稳定测试清除器\n\n/loop 运行测试套件 20 次，收集所有间歇性失败，修复或隔离不稳定测试，直到连续 5 次完全通过才停止。", "image": null, "sourceName": "X：Eric Zakariasson (@ericzakariasson)", "sourceCategory": "tip", "publishedAt": "2026-06-08T23:08:12.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-149cdb51", "url": "https://developer.apple.com/documentation/coreai", "canonicalUrl": "https://developer.apple.com/documentation/coreai", "title": "Apple Core AI 框架", "titleEn": "Apple Core AI 框架", "description": "一篇关于 Apple Core AI 框架的 Hacker News 帖子获得 109 个点赞。帖子内容包含一张苹果开发者 OG 图片和一个指向 Apple Core AI Framework 官方文档的链接（developer.apple.com）。该帖子由 buzzing.cc 中文翻译，发布于 2026 年 6 月 8 日 02：47（UTC）。", "descriptionEn": "一篇关于 Apple Core AI 框架的 Hacker News 帖子获得 109 个点赞。帖子内容包含一张苹果开发者 OG 图片和一个指向 Apple Core AI Framework 官方文档的链接（developer.apple.com）。该帖子由 buzzing.cc 中文翻译，发布于 2026 年 6 月 8 日 02：47（UTC）。", "image": null, "sourceName": "Hacker News 热门（buzzing.cc 中文翻译）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T23:05:07.159Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-ad23a0b4", "url": "https://x.com/AYi_AInotes/status/2064121284804641193", "canonicalUrl": "https://x.com/AYi_AInotes/status/2064121284804641193", "title": "MKBHD猜测新Siri首版不支持切换默认App，作者称其本质是生态门卫", "titleEn": "WWDC 2026 Apple 发布了下一代 Apple Intelligence 驱动的 Siri，MKBHD猜测首版 Siri AI 很可能不支持随意切换默认 App。 我觉得MKBHD大概率是…", "description": "MKBHD猜测WWDC 2026发布的Apple Intelligence驱动的新Siri，首版不会允许用户切换默认App（如用Google日历代替Apple日历）。推文作者赞同这一判断，认为苹果出于权限、隐私和商业利益考虑，会在自有生态内让Siri强大，但出生态则受限。在美国市场这一限制可忍受，但在海外WhatsApp和Google服务是刚需，Siri若无法调用则近乎不可用。新Siri的核心矛盾不在于智能，而在于它听用户的（贴心助手）还是听苹果的（生态门卫）。", "descriptionEn": "MKBHD猜测WWDC 2026发布的Apple Intelligence驱动的新Siri，首版不会允许用户切换默认App（如用Google日历代替Apple日历）。推文作者赞同这一判断，认为苹果出于权限、隐私和商业利益考虑，会在自有生态内让Siri强大，但出生态则受限。在美国市场这一限制可忍受，但在海外WhatsApp和Google服务是刚需，Siri若无法调用则近乎不可用。新Siri的核心矛盾不在于智能，而在于它听用户的（贴心助手）还是听苹果的（生态门卫）。", "image": null, "sourceName": "X：阿易 AI Notes (@AYi_AInotes)", "sourceCategory": "tip", "publishedAt": "2026-06-08T23:03:58.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.94, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-48f63ec6", "url": "https://www.ithome.com/0/961/724.htm", "canonicalUrl": "https://www.ithome.com/0/961/724.htm", "title": "Meta 智能眼镜暗藏人脸识别代码，曝光仅一天后被紧急移除", "titleEn": "Meta 智能眼镜暗藏人脸识别代码，曝光仅一天后被紧急移除", "description": "Meta 智能眼镜配套应用中发现一段休眠人脸识别代码，内部代号\"姓名标签\"，可将面部影像转化为设备本地生物特征标识并交叉比对。《连线》6月4日发现该代码，Meta 次日即推送更新彻底删除。Meta 通讯副总裁称该功能仅为试点项目，尚未做正式决定。此前 Meta 智能眼镜已因偷拍、审核图像隐私问题遭集体诉讼。", "descriptionEn": "Meta 智能眼镜配套应用中发现一段休眠人脸识别代码，内部代号\"姓名标签\"，可将面部影像转化为设备本地生物特征标识并交叉比对。《连线》6月4日发现该代码，Meta 次日即推送更新彻底删除。Meta 通讯副总裁称该功能仅为试点项目，尚未做正式决定。此前 Meta 智能眼镜已因偷拍、审核图像隐私问题遭集体诉讼。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T22:57:19.000Z", "discoveredAt": "2026-06-09T00:39:05.553Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-48cbf843", "url": "https://www.ithome.com/0/961/721.htm", "canonicalUrl": "https://www.ithome.com/0/961/721.htm", "title": "奥尔特曼宣布 OpenAI 进入第三发展阶段：让 AI 普及、易用且安全", "titleEn": "奥尔特曼宣布 OpenAI 进入第三发展阶段：让 AI 普及、易用且安全", "description": "本周一，OpenAI CEO 奥尔特曼与首席科学家帕霍茨基联合发文，宣布公司进入第三发展阶段，目标让 AI 普及、易用且安全。此前第一阶段聚焦通用人工智能技术研发，第二阶段面向全球推出产品。第三阶段三大核心目标是打造自动化人工智能研究员、推动经济提速、为每人配备专属通用人工智能。二人强调智能系统须坚守安全底线，呼吁成立国际机构应对 AI 风险，必要时可暂缓前沿模型研发。同日，OpenAI 秘密提交 IPO 申请，但上市仍需较长时间。上周 Anthropic 研究人员也建议适当放缓前沿 AI 研发。", "descriptionEn": "本周一，OpenAI CEO 奥尔特曼与首席科学家帕霍茨基联合发文，宣布公司进入第三发展阶段，目标让 AI 普及、易用且安全。此前第一阶段聚焦通用人工智能技术研发，第二阶段面向全球推出产品。第三阶段三大核心目标是打造自动化人工智能研究员、推动经济提速、为每人配备专属通用人工智能。二人强调智能系统须坚守安全底线，呼吁成立国际机构应对 AI 风险，必要时可暂缓前沿模型研发。同日，OpenAI 秘密提交 IPO 申请，但上市仍需较长时间。上周 Anthropic 研究人员也建议适当放缓前沿 AI 研发。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T22:50:42.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-safety", "automation" ], "relevanceScore": 0.9, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-43ef080f", "url": "https://claude.com/blog/claude-for-foundation-models", "canonicalUrl": "https://claude.com/blog/claude-for-foundation-models", "title": "Claude 支持 Apple Foundation Models 框架，推出新 Swift 包", "titleEn": "Building intelligent apps for Apple platforms with Claude in the Foundation Models framework", "description": "Anthropic 通过一个新 Swift 包，让 Apple 开发者能在 Foundation Models 框架中直接调用 Claude。开发者用 Apple 框架以原生 Swift 方式，仅需三行代码即可通过引导生成返回类型化 Swift 值。当请求涉及多步推理、代码生成、联网搜索或数据分析时，Claude 可接手处理，并将流式响应传回同一视图。该支持可在 iOS 27、iPadOS 27、macOS 27、visionOS 27 和 watchOS 27 上使用，开发者需添加包并配置 Anthropic API 密钥。", "descriptionEn": "Anthropic 通过一个新 Swift 包，让 Apple 开发者能在 Foundation Models 框架中直接调用 Claude。开发者用 Apple 框架以原生 Swift 方式，仅需三行代码即可通过引导生成返回类型化 Swift 值。当请求涉及多步推理、代码生成、联网搜索或数据分析时，Claude 可接手处理，并将流式响应传回同一视图。该支持可在 iOS 27、iPadOS 27、macOS 27、visionOS 27 和 watchOS 27 上使用，开发者需添加包并配置 Anthropic API 密钥。", "image": null, "sourceName": "Claude：Blog（网页）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T22:24:02.356Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-c951f08a", "url": "https://www.theverge.com/tech/946345/apple-safari-ai-update-extensions", "canonicalUrl": "https://www.theverge.com/tech/946345/apple-safari-ai-update-extensions", "title": "Apple 正用 AI 修复 Safari 的扩展程序问题", "titleEn": "Apple is using AI to fix Safari's extension problem", "description": "Safari 将邀请用户通过 \"vibe-code\" 自行编写扩展程序，Apple 借助 AI 来解决 Safari 的扩展生态短板。", "descriptionEn": "Safari 将邀请用户通过 \"vibe-code\" 自行编写扩展程序，Apple 借助 AI 来解决 Safari 的扩展生态短板。", "image": null, "sourceName": "The Verge：订阅版科技（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T22:14:17.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-6044246c", "url": "https://github.com/anthropics/claude-code/releases/tag/v2.1.169", "canonicalUrl": "https://github.com/anthropics/claude-code/releases/tag/v2.1.169", "title": "Claude Code v2.1.169 发布", "titleEn": "v2.1.169", "description": "Claude Code v2.1.169 新增 `--safe-mode` 标志及环境变量，用于禁用所有自定义配置以排查问题；新增 `/cd` 命令，可在不破坏提示词缓存的情况下切换工作目录；新增 `disableBundledSkills` 设置隐藏内置技能。修复了企业 MCP 策略在重连、IDE 配置及首次会话中不被强制的问题，以及 macOS 用户每轮约 30-50ms 的 UI 卡顿、Windows 下 `claude -p` 和 Git 凭证弹窗等多项 bug。改进了任务创建的可靠性，恢复 Vertex/Foundry 默认 5 分钟空闲超时，并降低 CPU 占用。", "descriptionEn": "Claude Code v2.1.169 新增 `--safe-mode` 标志及环境变量，用于禁用所有自定义配置以排查问题；新增 `/cd` 命令，可在不破坏提示词缓存的情况下切换工作目录；新增 `disableBundledSkills` 设置隐藏内置技能。修复了企业 MCP 策略在重连、IDE 配置及首次会话中不被强制的问题，以及 macOS 用户每轮约 30-50ms 的 UI 卡顿、Windows 下 `claude -p` 和 Git 凭证弹窗等多项 bug。改进了任务创建的可靠性，恢复 Vertex/Foundry 默认 5 分钟空闲超时，并降低 CPU 占用。", "image": null, "sourceName": "Claude Code：GitHub Releases（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T21:57:18.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem", "coding-agent" ], "relevanceScore": 0.98, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-3582e3e0", "url": "https://x.com/rohanpaul_ai/status/2064103154833383534", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064103154833383534", "title": "OpenAI 机密提交 IPO 文件，与 Anthropic 展开资本竞赛；Altman 称 2028 年前 AI 将承担大部分研究", "titleEn": "BREAKING： WSJ reports OpenAI just made its first formal move toward IPO. it has confidentially file…", "description": "OpenAI 向 SEC 机密提交了 S-1 文件，正式启动 IPO 流程，可在不公开收入、亏损、客户构成等敏感数据的情况下开始审查。上周 Anthropic 也已机密提交 IPO 文件，AI 实验室的竞争从模型竞赛转向下一代 AI 基础设施的资本竞赛。Sam Altman 在博客中称，到 2028 年 3 月，OpenAI 很大一部分研究将由 AI 完成，目标包括构建自动化 AI 研究员、加速科学生产力、为每个人提供个人 AGI。", "descriptionEn": "OpenAI 向 SEC 机密提交了 S-1 文件，正式启动 IPO 流程，可在不公开收入、亏损、客户构成等敏感数据的情况下开始审查。上周 Anthropic 也已机密提交 IPO 文件，AI 实验室的竞争从模型竞赛转向下一代 AI 基础设施的资本竞赛。Sam Altman 在博客中称，到 2028 年 3 月，OpenAI 很大一部分研究将由 AI 完成，目标包括构建自动化 AI 研究员、加速科学生产力、为每个人提供个人 AGI。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "industry", "publishedAt": "2026-06-08T21:51:55.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-12a6aae", "url": "https://x.com/kimmonismus/status/2064100504398135442", "canonicalUrl": "https://x.com/kimmonismus/status/2064100504398135442", "title": "OpenAI 进入第三阶段：经济重塑，2028 年目标自动化科研", "titleEn": "OpenAI is \"entering the third phase. The economy is beginning to reshape around AI.\" - The first ph…", "description": "OpenAI 进入第三阶段：经济开始围绕 AI 重塑。前两阶段分别为 AGI 研究和产品化转型。CEO Sam Altman 发布当前规划（链接），明确 2028 年目标是构建可控、可问责的 AI 研究员，逐步自动化科学研究，帮助解决对齐问题并导航后 AGI 过渡。这标志走向 AGI/后 AGI 的最后阶段。", "descriptionEn": "OpenAI 进入第三阶段：经济开始围绕 AI 重塑。前两阶段分别为 AGI 研究和产品化转型。CEO Sam Altman 发布当前规划（链接），明确 2028 年目标是构建可控、可问责的 AI 研究员，逐步自动化科学研究，帮助解决对齐问题并导航后 AGI 过渡。这标志走向 AGI/后 AGI 的最后阶段。", "image": null, "sourceName": "X：Kim (@kimmonismus)", "sourceCategory": "tip", "publishedAt": "2026-06-08T21:41:24.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-5b54973", "url": "https://x.com/rohanpaul_ai/status/2064096574142390755", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064096574142390755", "title": "OpenAI计划到2028年由AI主导研究", "titleEn": "Sam Altman's new blog about OpenAI's future path says by March-2028 a significant fraction of its ow…", "description": "Sam Altman关于OpenAI未来路径的新博客称，到2028年3月，其大量研究将由AI完成。\n\n该路径主要有3个目标：构建自动AI研究员，利用它加速科学和生产，然后给每个人一个个人AGI，帮助处理工作、学习、编程、商业、健康文书和决策。", "descriptionEn": "Sam Altman关于OpenAI未来路径的新博客称，到2028年3月，其大量研究将由AI完成。\n\n该路径主要有3个目标：构建自动AI研究员，利用它加速科学和生产，然后给每个人一个个人AGI，帮助处理工作、学习、编程、商业、健康文书和决策。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "tip", "publishedAt": "2026-06-08T21:25:46.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-insights" ], "relevanceScore": 0.5, "seelesAngle": "Recommended insight: adjacent AI ecosystem signal that may still be useful context for Seeles agent and skill strategy." }, { "id": "news-45d3030d", "url": "https://www.ithome.com/0/961/709.htm", "canonicalUrl": "https://www.ithome.com/0/961/709.htm", "title": "完善 Mac 游戏生态：苹果更新工具，借助 AI 加速游戏移植", "titleEn": "完善 Mac 游戏生态：苹果更新工具，借助 AI 加速游戏移植", "description": "苹果在 2026 年 WWDC 主题演讲后推出 Game Porting Toolkit 4，加速游戏移植至 Mac。新版配套上线 GitHub 代码仓库，提供开源 AI 智能体技能与示例代码。这些智能体具备更深入的 Metal 知识，可通过命令行访问 Metal 工具，完成捕获、调试与性能分析。评估环境已支持最新图形接口 Metal 4，便于开发者早期测试兼容性与性能。", "descriptionEn": "苹果在 2026 年 WWDC 主题演讲后推出 Game Porting Toolkit 4，加速游戏移植至 Mac。新版配套上线 GitHub 代码仓库，提供开源 AI 智能体技能与示例代码。这些智能体具备更深入的 Metal 知识，可通过命令行访问 Metal 工具，完成捕获、调试与性能分析。评估环境已支持最新图形接口 Metal 4，便于开发者早期测试兼容性与性能。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T21:24:27.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.98, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-45b6d40b", "url": "https://www.ithome.com/0/961/707.htm", "canonicalUrl": "https://www.ithome.com/0/961/707.htm", "title": "苹果 iOS 27 日历 / 提醒事项新增 AI：支持自然语言创建事件", "titleEn": "苹果 iOS 27 版日历 / 提醒事项新增 AI 技能：支持自然语言创建事件", "description": "在 2026 年 WWDC 上，苹果为 iOS 27 系统的日历与提醒事项加入 AI 功能。用户点按\"+\"后可直接输入自然语言，如\"周四晚上 8 点和女友看电影\"，系统自动识别时间、日期和事件内容生成安排；输入\"每周\"等频率可设重复事件。提醒事项同样支持自然语言创建。日历事件编辑可智能调整重复频率。Visual Intelligence 联动：截图或拍摄活动传单后，系统识别内容并自动加入日历。", "descriptionEn": "在 2026 年 WWDC 上，苹果为 iOS 27 系统的日历与提醒事项加入 AI 功能。用户点按\"+\"后可直接输入自然语言，如\"周四晚上 8 点和女友看电影\"，系统自动识别时间、日期和事件内容生成安排；输入\"每周\"等频率可设重复事件。提醒事项同样支持自然语言创建。日历事件编辑可智能调整重复频率。Visual Intelligence 联动：截图或拍摄活动传单后，系统识别内容并自动加入日历。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T21:13:01.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-b108bc87", "url": "https://x.com/rohanpaul_ai/status/2064092512227295483", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064092512227295483", "title": "美国GUARD法案拟禁止中国机器人入美", "titleEn": "A new US bill could ban some Chinese robots from America. The GUARD Act would force security agenci…", "description": "美国新提出的GUARD法案要求安全机构审查来自中国及其他敌对国家生产的机器人，并将高风险系统列入FCC\"覆盖清单\"，类比此前对华为、中兴的禁令。立法者担忧机器人不仅配备摄像头、麦克风、传感器、地图、马达和无线链路，更会成为可在工厂、实验室、家庭和警局内移动的计算机。另一项Schumer-Cotton法案单独禁止联邦机构购买或使用中国人形机器人，但允许受控的军事或执法研究例外。", "descriptionEn": "美国新提出的GUARD法案要求安全机构审查来自中国及其他敌对国家生产的机器人，并将高风险系统列入FCC\"覆盖清单\"，类比此前对华为、中兴的禁令。立法者担忧机器人不仅配备摄像头、麦克风、传感器、地图、马达和无线链路，更会成为可在工厂、实验室、家庭和警局内移动的计算机。另一项Schumer-Cotton法案单独禁止联邦机构购买或使用中国人形机器人，但允许受控的军事或执法研究例外。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "industry", "publishedAt": "2026-06-08T21:09:38.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-ec0ece2b", "url": "https://x.com/rohanpaul_ai/status/2064089553431069046", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064089553431069046", "title": "Anthropic 研究：AI 智能体在生物数据库检索中失败率高，结论偏差严重", "titleEn": "New Anthropic research shows AI agents may look brilliant at code， but in biology they can fail befo…", "description": "Anthropic 研究发现，AI 智能体在代码任务表现出色，但在生物数据库检索中容易失败。以埃博拉序列任务为例，Claude Sonnet 4 三次运行分别返回 106、15 和 5 条序列，而预期为 266 条。缺失序列导致科学结论严重偏移：智能体推断疫情回溯至 1922 年，人工筛选结果却指向 2014 年初。问题根源在于生物数据库分散、网站规则隐蔽、脚本脆弱。引入可重复检索工具后，智能体准确性和一致性大幅提升。Anthropic 呼吁建设更友好的基础设施。", "descriptionEn": "Anthropic 研究发现，AI 智能体在代码任务表现出色，但在生物数据库检索中容易失败。以埃博拉序列任务为例，Claude Sonnet 4 三次运行分别返回 106、15 和 5 条序列，而预期为 266 条。缺失序列导致科学结论严重偏移：智能体推断疫情回溯至 1922 年，人工筛选结果却指向 2014 年初。问题根源在于生物数据库分散、网站规则隐蔽、脚本脆弱。引入可重复检索工具后，智能体准确性和一致性大幅提升。Anthropic 呼吁建设更友好的基础设施。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "tip", "publishedAt": "2026-06-08T20:57:53.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-960b12be", "url": "https://x.com/NotebookLM/status/2064084153533165588", "canonicalUrl": "https://x.com/NotebookLM/status/2064084153533165588", "title": "NotebookLM升级：AI思考过程更清晰", "titleEn": "Want a closer look at today's launch？ Here is a breakdown of what's new and exciting 🧵： First up： …", "description": "想更详细了解今天的发布吗？以下是新功能和亮点的介绍🧵：\n\n首先：升级版、更周到的聊天体验。\n\n由Gemini 3.5和@Antigravity提供支持，您现在将更清晰地看到AI的思考过程。此外，每个笔记本都有一台安全的云电脑，包含100多种精选软件技能，解锁更深入的研究和更复杂的分析。", "descriptionEn": "想更详细了解今天的发布吗？以下是新功能和亮点的介绍🧵：\n\n首先：升级版、更周到的聊天体验。\n\n由Gemini 3.5和@Antigravity提供支持，您现在将更清晰地看到AI的思考过程。此外，每个笔记本都有一台安全的云电脑，包含100多种精选软件技能，解锁更深入的研究和更复杂的分析。", "image": null, "sourceName": "X：NotebookLM (@NotebookLM)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T20:36:25.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem", "agent-safety" ], "relevanceScore": 0.82, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-86b6897a", "url": "https://x.com/ViggleAI/status/2064083087806390319", "canonicalUrl": "https://x.com/ViggleAI/status/2064083087806390319", "title": "Viggle API 上线：任意角色任意动作秒级生成", "titleEn": "Introducing the Viggle API. Give any character any motion， one API call - alive in seconds. Wire …", "description": "推出 Viggle API。\n\n给任意角色添加任意动作，一次 API 调用--数秒内即可激活。\n\n可接入 Claude、Codex 或你正在构建的任何智能体。起价 $0.01/秒。\n\n注册即获 100 次免费额度。\n\n转发 + 关注 + 评论，10 位中奖者再获 100 次！\n\n了解更多👇", "descriptionEn": "推出 Viggle API。\n\n给任意角色添加任意动作，一次 API 调用--数秒内即可激活。\n\n可接入 Claude、Codex 或你正在构建的任何智能体。起价 $0.01/秒。\n\n注册即获 100 次免费额度。\n\n转发 + 关注 + 评论，10 位中奖者再获 100 次！\n\n了解更多👇", "image": null, "sourceName": "X：Viggle AI (@ViggleAI)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T20:32:11.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-648cc605", "url": "https://x.com/rohanpaul_ai/status/2064080897662660643", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064080897662660643", "title": "AdaCoM：独立小模型管理上下文，智能体长任务提升39%", "titleEn": "AI agent can get better at long tasks without retraining the agent itself， by using a separate small…", "description": "论文提出 AdaCoM，一个独立的 LLM，在智能体每步操作前编辑其工作上下文。它可重写、合并、剪枝或保留任务历史，使主智能体保持冻结，无需重新训练或暴露权重。与简单摘要不同，AdaCoM 学习不同智能体需要不同类型上下文--强智能体保留更多原始历史，弱智能体需更短更清晰的笔记。在 web search 和 deep research 任务上测试，平均提升 39%。", "descriptionEn": "论文提出 AdaCoM，一个独立的 LLM，在智能体每步操作前编辑其工作上下文。它可重写、合并、剪枝或保留任务历史，使主智能体保持冻结，无需重新训练或暴露权重。与简单摘要不同，AdaCoM 学习不同智能体需要不同类型上下文--强智能体保留更多原始历史，弱智能体需更短更清晰的笔记。在 web search 和 deep research 任务上测试，平均提升 39%。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "paper", "publishedAt": "2026-06-08T20:23:29.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-3ad7080f", "url": "https://www.bloomberg.com/news/articles/2026-06-08/apple-downplays-concerns-that-its-use-of-google-ai-models-will-undermine-privacy", "canonicalUrl": "https://www.bloomberg.com/news/articles/2026-06-08/apple-downplays-concerns-that-its-use-of-google-ai-models-will-undermine-privacy", "title": "Apple 淡化使用 Google AI 模型会削弱隐私的担忧", "titleEn": "Apple Downplays Concerns That Its Use of Google AI Models Will Undermine Privacy", "description": "Apple 在最新公布的人工智能平台中部分采用了 Google 的技术，但表示新方案仍会保留其隐私保护措施。", "descriptionEn": "Apple 在最新公布的人工智能平台中部分采用了 Google 的技术，但表示新方案仍会保留其隐私保护措施。", "image": null, "sourceName": "Bloomberg：Technology（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T20:21:16.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.94, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-814a8abd", "url": "https://x.com/kimmonismus/status/2064078012614684683", "canonicalUrl": "https://x.com/kimmonismus/status/2064078012614684683", "title": "Apple 指责欧盟自身导致 Apple Intelligence 在 EU 不可用", "titleEn": "Apple makes it clear that Europe is itself to blame for the unavailability of Apple Intelligence. A…", "description": "Apple 称新 Siri AI（仅限 iPhone 17 Pro）不会在欧盟的 iPhone 和 iPad 上推出，原因是 DMA 被解释为要求 Apple 向竞争对手 AI 助手开放用户数据和 app 控制权，Apple 认为这会带来隐私和安全风险。Apple 提出的\"Trusted System Agent\"和 18 个月部署计划均被欧盟委员会拒绝，目前无上线时间表。Apple 软件工程高级副总裁 Craig Federighi 表示\"深感失望\"。", "descriptionEn": "Apple 称新 Siri AI（仅限 iPhone 17 Pro）不会在欧盟的 iPhone 和 iPad 上推出，原因是 DMA 被解释为要求 Apple 向竞争对手 AI 助手开放用户数据和 app 控制权，Apple 认为这会带来隐私和安全风险。Apple 提出的\"Trusted System Agent\"和 18 个月部署计划均被欧盟委员会拒绝，目前无上线时间表。Apple 软件工程高级副总裁 Craig Federighi 表示\"深感失望\"。", "image": null, "sourceName": "X：Kim (@kimmonismus)", "sourceCategory": "industry", "publishedAt": "2026-06-08T20:12:01.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-ec3ae6a0", "url": "https://x.com/omarsar0/status/2064076252584222933", "canonicalUrl": "https://x.com/omarsar0/status/2064076252584222933", "title": "AI智能体重塑知识工作的三个维度", "titleEn": "New paper on how AI agents are reshaping knowledge work. This is a nice economic read on where agen…", "description": "一篇新论文从自主性、效率和工人移交任务的范围三个维度，分析AI智能体如何重塑知识工作。研究指出，当前人们使用智能体的主要障碍并非模型质量，而是几乎没有人接受过如何以这种方式工作的培训。", "descriptionEn": "一篇新论文从自主性、效率和工人移交任务的范围三个维度，分析AI智能体如何重塑知识工作。研究指出，当前人们使用智能体的主要障碍并非模型质量，而是几乎没有人接受过如何以这种方式工作的培训。", "image": null, "sourceName": "X：Elvis Saravia (@omarsar0, DAIR.AI)", "sourceCategory": "paper", "publishedAt": "2026-06-08T20:05:01.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-fce18d74", "url": "https://x.com/ClaudeDevs/status/2064072801062121906", "canonicalUrl": "https://x.com/ClaudeDevs/status/2064072801062121906", "title": "Claude 新增连接器可观测性仪表盘", "titleEn": "We've added an observability dashboard for developers of connectors. Connectors let third-party dev…", "description": "我们为连接器开发者添加了一个可观测性仪表盘。\n连接器让第三方开发者能够通过MCP将他们的工具和数据带到Claude。", "descriptionEn": "我们为连接器开发者添加了一个可观测性仪表盘。\n连接器让第三方开发者能够通过MCP将他们的工具和数据带到Claude。", "image": null, "sourceName": "X：Claude Devs (@ClaudeDevs)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T19:51:19.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-19f7134f", "url": "https://x.com/cb_doge/status/2064071460193460534", "canonicalUrl": "https://x.com/cb_doge/status/2064071460193460534", "title": "佛罗里达总检察长起诉OpenAI和Sam Altman，指控ChatGPT鼓励暴力与协助枪击案", "titleEn": "NEWS： Florida Attorney General James Uthmeier has filed a major civil lawsuit against OpenAI and Sam…", "description": "佛罗里达总检察长James Uthmeier近日对OpenAI及CEO Sam Altman提起重大民事诉讼，指控ChatGPT鼓励暴力、欺骗家长、充当\"自杀教练\"诱导16岁少年自残、协助佛罗里达州立大学枪击案凶手策划袭击。诉状称公司为追求快速增长而忽视安全措施，Altman因批准危险功能被个人追责。佛罗里达州已就OpenAI在FSU枪击案中的角色展开刑事调查。", "descriptionEn": "佛罗里达总检察长James Uthmeier近日对OpenAI及CEO Sam Altman提起重大民事诉讼，指控ChatGPT鼓励暴力、欺骗家长、充当\"自杀教练\"诱导16岁少年自残、协助佛罗里达州立大学枪击案凶手策划袭击。诉状称公司为追求快速增长而忽视安全措施，Altman因批准危险功能被个人追责。佛罗里达州已就OpenAI在FSU枪击案中的角色展开刑事调查。", "image": null, "sourceName": "X：cb_doge (@cb_doge)", "sourceCategory": "industry", "publishedAt": "2026-06-08T19:45:59.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-202c3661", "url": "https://www.ithome.com/0/961/697.htm", "canonicalUrl": "https://www.ithome.com/0/961/697.htm", "title": "苹果在 iOS 27 的 Passwords 应用中引入 AI 自动改密功能", "titleEn": "苹果提升账号安全，iOS 27 版 Passwords 应用支持 AI 自动更换密码", "description": "苹果在 6 月 9 日的全球开发者大会上，为 iOS 27 的 Passwords 应用引入新 AI 功能，可在多种场景下自动替用户修改密码，减少账号修复步骤。该应用两年前独立上线，集中管理登录信息、2FA 验证码、Wi-Fi 凭证等，已具备识别数据泄露和重复使用密码的能力。旧版仅能发现问题，用户需手动进入网站或应用修改；新版 AI 可直接操作，用户无需为每个受影响账户重复处理修改流程，降低使用门槛。", "descriptionEn": "苹果在 6 月 9 日的全球开发者大会上，为 iOS 27 的 Passwords 应用引入新 AI 功能，可在多种场景下自动替用户修改密码，减少账号修复步骤。该应用两年前独立上线，集中管理登录信息、2FA 验证码、Wi-Fi 凭证等，已具备识别数据泄露和重复使用密码的能力。旧版仅能发现问题，用户需手动进入网站或应用修改；新版 AI 可直接操作，用户无需为每个受影响账户重复处理修改流程，降低使用门槛。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T19:42:46.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-c8451fb6", "url": "https://x.com/dotey/status/2064069492850708484", "canonicalUrl": "https://x.com/dotey/status/2064069492850708484", "title": "豆包手机团队招聘设计工程师，要求 Android 平台经验", "titleEn": "帮转，豆包手机团队招设计工程师", "description": "字节跳动豆包手机团队招聘设计工程师，需具备 Android 平台经验。团队调研总结了五种设计工程师画像：AI Design Engineer（转译 AI 能力为交互产品体验，处理 agent workflow、tool call、状态反馈等）、Product UI Craft Engineer（打磨高质量前端原型和交互细节）、Design Systems Engineer（建立设计系统与前端基础设施，连接 Figma 变量和代码组件）、Creative Technologist / Motion & Graphics Engineer（负责动效、实时图形、3D/空间交互）、AI Design Workflow Architect（搭建 AI 辅助设计工作流，使用 Claude Code、Cursor、v0 等工具）。多数设计工程师技能交叉，欢迎感兴趣者联系。", "descriptionEn": "字节跳动豆包手机团队招聘设计工程师，需具备 Android 平台经验。团队调研总结了五种设计工程师画像：AI Design Engineer（转译 AI 能力为交互产品体验，处理 agent workflow、tool call、状态反馈等）、Product UI Craft Engineer（打磨高质量前端原型和交互细节）、Design Systems Engineer（建立设计系统与前端基础设施，连接 Figma 变量和代码组件）、Creative Technologist / Motion & Graphics Engineer（负责动效、实时图形、3D/空间交互）、AI Design Workflow Architect（搭建 AI 辅助设计工作流，使用 Claude Code、Cursor、v0 等工具）。多数设计工程师技能交叉，欢迎感兴趣者联系。", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "tip", "publishedAt": "2026-06-08T19:38:10.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent", "agent-architecture", "skill-ecosystem", "automation" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-2001efde", "url": "https://www.ithome.com/0/961/694.htm", "canonicalUrl": "https://www.ithome.com/0/961/694.htm", "title": "苹果发布 Xcode 27 Beta 与 Core AI 框架，称其为构建 AI 智能体\"最佳场所\"", "titleEn": "Xcode 27 Beta 发布：苹果称其是构建 AI 智能体应用的\"最佳场所\"", "description": "在 2026 全球开发者大会（WWDC）上，苹果发布 Xcode 27 Beta 并推出 Core AI 框架。新版 Foundation Models 框架支持图像输入，开发者可将图片与文本一并传给设备端模型，处理多模态任务。Xcode 编码助手升级为智能体式编码工作流，已覆盖应用本地化、与模拟设备交互以及测试与调试环节。苹果软件工程高级副总裁 Craig Federighi 称 Xcode 已成为构建智能体 AI 应用的\"最佳场所\"。", "descriptionEn": "在 2026 全球开发者大会（WWDC）上，苹果发布 Xcode 27 Beta 并推出 Core AI 框架。新版 Foundation Models 框架支持图像输入，开发者可将图片与文本一并传给设备端模型，处理多模态任务。Xcode 编码助手升级为智能体式编码工作流，已覆盖应用本地化、与模拟设备交互以及测试与调试环节。苹果软件工程高级副总裁 Craig Federighi 称 Xcode 已成为构建智能体 AI 应用的\"最佳场所\"。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T19:18:57.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.86, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-149385fc", "url": "https://x.com/kimmonismus/status/2064063105634582844", "canonicalUrl": "https://x.com/kimmonismus/status/2064063105634582844", "title": "Apple WWDC 2026：全新 Siri AI 基于 Google Gemini，强调隐私与区域限制", "titleEn": "Apple： \"Our most powerful on-device AI model runs on iPhone 17 Pro.\" Me： Great， I have one. Apple：…", "description": "Apple 在 WWDC 2026 推出基于 Google Gemini（多年度协议，据报约每年 10 亿美元定制大模型）与自研端侧 Foundation Models 的全新 Siri AI，支持屏幕感知、个人上下文搜索、跨应用操作及独立 Siri app（iCloud 同步对话历史）。Apple Intelligence 还包括 Photos Spatial Reframing、Safari 页面监控、Messages 一键建议、Image Playground 照片级生成等功能。最强端侧 AI 模型仅运行在 iPhone 17 Pro 上，且初期不在欧盟推出。隐私端强调数据仅用于执行请求，可被外部专家验证。", "descriptionEn": "Apple 在 WWDC 2026 推出基于 Google Gemini（多年度协议，据报约每年 10 亿美元定制大模型）与自研端侧 Foundation Models 的全新 Siri AI，支持屏幕感知、个人上下文搜索、跨应用操作及独立 Siri app（iCloud 同步对话历史）。Apple Intelligence 还包括 Photos Spatial Reframing、Safari 页面监控、Messages 一键建议、Image Playground 照片级生成等功能。最强端侧 AI 模型仅运行在 iPhone 17 Pro 上，且初期不在欧盟推出。隐私端强调数据仅用于执行请求，可被外部专家验证。", "image": null, "sourceName": "X：Kim (@kimmonismus)", "sourceCategory": "tip", "publishedAt": "2026-06-08T19:12:47.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-1fe5c0dc", "url": "https://www.ithome.com/0/961/692.htm", "canonicalUrl": "https://www.ithome.com/0/961/692.htm", "title": "因隐私安全分歧，苹果 Siri AI 缺席欧盟版 iOS 27 首发", "titleEn": "因隐私安全分歧，苹果 Siri AI 缺席欧盟版 iOS 27 首发", "description": "苹果6月9日宣布，因欧盟《数字市场法》（DMA）影响，Siri AI不会随iOS 27、macOS 27在欧盟上线，多项新功能暂不可用。苹果表示Siri AI基于端侧处理与私有云计算，设计上尽量少暴露数据。为回应监管要求，苹果提出Trusted System Agent中间方案，允许其他虚拟助手在安全前提下获得同等权限，并给出18个月渐进上线计划，但欧洲委员会未接受。", "descriptionEn": "苹果6月9日宣布，因欧盟《数字市场法》（DMA）影响，Siri AI不会随iOS 27、macOS 27在欧盟上线，多项新功能暂不可用。苹果表示Siri AI基于端侧处理与私有云计算，设计上尽量少暴露数据。为回应监管要求，苹果提出Trusted System Agent中间方案，允许其他虚拟助手在安全前提下获得同等权限，并给出18个月渐进上线计划，但欧洲委员会未接受。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T19:00:51.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-11279a54", "url": "https://www.marktechpost.com/2026/06/08/clawhub-security-signals-a-coding-guide-to-end-to-end-security-signal-analysis-and-verdict-classification-on-the-ai-skills-dataset", "canonicalUrl": "https://www.marktechpost.com/2026/06/08/clawhub-security-signals-a-coding-guide-to-end-to-end-security-signal-analysis-and-verdict-classification-on-the-ai-skills-dataset", "title": "ClawHub Security Signals：AI技能数据集的端到端安全信号分析与判定分类编程指南", "titleEn": "ClawHub Security Signals： A Coding Guide to End-to-End Security Signal Analysis and Verdict Classification on the AI Skills Dataset", "description": "该教程使用ClawHub Security Signals数据集分析扫描器如何评估AI技能。从Hugging Face Parquet格式加载数据，检查判定结果、扫描输出和严重性标签，通过Jaccard分数和Cohen's kappa衡量VirusTotal、静态分析和SkillSpector之间的重叠与分歧，最后将SKILL.md文本与扫描信号结合，训练逻辑回归模型预测ClawScan判定。", "descriptionEn": "该教程使用ClawHub Security Signals数据集分析扫描器如何评估AI技能。从Hugging Face Parquet格式加载数据，检查判定结果、扫描输出和严重性标签，通过Jaccard分数和Cohen's kappa衡量VirusTotal、静态分析和SkillSpector之间的重叠与分歧，最后将SKILL.md文本与扫描信号结合，训练逻辑回归模型预测ClawScan判定。", "image": null, "sourceName": "MarkTechPost（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T18:57:19.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety", "skill-ecosystem" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-e5d55009", "url": "https://claude.com/blog/observability-for-developers-building-connectors", "canonicalUrl": "https://claude.com/blog/observability-for-developers-building-connectors", "title": "Claude 为 Connector 开发者推出性能监控仪表盘", "titleEn": "Observability for developers building connectors", "description": "Claude 为目录中已发布的 Connector 新增性能监控仪表盘（公开测试版）。所有者可追踪活跃用户、工具调用次数、目录排名、健康评分、错误率、延迟，并按工具细分错误归因；还能按 Claude、Claude Code、Cowork 等产品分解使用情况。访问需 Team 或 Enterprise 账号（Admin/Owner 权限或自定义角色）。基于 MCP 构建的 Connector 可直接在 Claude 应用内提交至目录，该目录已有超 300 个第三方 Connector，每日被数百万人使用。", "descriptionEn": "Claude 为目录中已发布的 Connector 新增性能监控仪表盘（公开测试版）。所有者可追踪活跃用户、工具调用次数、目录排名、健康评分、错误率、延迟，并按工具细分错误归因；还能按 Claude、Claude Code、Cowork 等产品分解使用情况。访问需 Team 或 Enterprise 账号（Admin/Owner 权限或自定义角色）。基于 MCP 构建的 Connector 可直接在 Claude 应用内提交至目录，该目录已有超 300 个第三方 Connector，每日被数百万人使用。", "image": null, "sourceName": "Claude：Blog（网页）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:50:51.580Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "skill-ecosystem", "tool-use", "coding-agent", "agent-safety" ], "relevanceScore": 0.9, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-ba704f8e", "url": "https://techcrunch.com/2026/06/08/apple-just-taught-your-iphone-to-finish-your-sentences-your-photos-and-your-workflows", "canonicalUrl": "https://techcrunch.com/2026/06/08/apple-just-taught-your-iphone-to-finish-your-sentences-your-photos-and-your-workflows", "title": "苹果刚刚教会你的iPhone补全句子、完善照片和简化工作流程", "titleEn": "Apple just taught your iPhone to finish your sentences， your photos， and your workflows", "description": "Apple为Safari、Shortcuts和Password应用添加了AI驱动的新功能，让iPhone能够自动补全句子、完善照片和简化工作流程。", "descriptionEn": "Apple为Safari、Shortcuts和Password应用添加了AI驱动的新功能，让iPhone能够自动补全句子、完善照片和简化工作流程。", "image": null, "sourceName": "TechCrunch：AI（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:48:45.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "automation" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-d110d04", "url": "https://techcrunch.com/2026/06/08/apple-will-let-you-build-workflows-using-ai-in-its-new-shortcuts-app", "canonicalUrl": "https://techcrunch.com/2026/06/08/apple-will-let-you-build-workflows-using-ai-in-its-new-shortcuts-app", "title": "Apple Shortcuts 应用获 AI 升级，支持用提示词构建工作流", "titleEn": "Apple will let you build workflows using AI in its new Shortcuts app", "description": "Apple Shortcuts 应用获得 AI 升级，用户只需用提示词描述想要的工作流，即可自动生成。", "descriptionEn": "Apple Shortcuts 应用获得 AI 升级，用户只需用提示词描述想要的工作流，即可自动生成。", "image": null, "sourceName": "TechCrunch：AI（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:45:12.000Z", "discoveredAt": "2026-06-09T00:39:05.554Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "automation" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-df5c5a39", "url": "https://x.com/AnthropicAI/status/2064054837294354677", "canonicalUrl": "https://x.com/AnthropicAI/status/2064054837294354677", "title": "Anthropic：AI编码快于生物学的缘由", "titleEn": "New Science Blog： Why has AI advanced faster in coding than in biology？ To agents， bio databases a…", "description": "新的科学博客：为什么 AI 在编码方面的进步比在生物学方面更快？\n\n对智能体而言，生物数据库就像汽车发明前建造的城市--开进去会让人抓狂，因为它们是针对不同的交通流量设计的。\n\n我们如何构建智能体可以使用的基础设施？\nhttps：//www.anthropic.com/research/agents-in-biology", "descriptionEn": "新的科学博客：为什么 AI 在编码方面的进步比在生物学方面更快？\n\n对智能体而言，生物数据库就像汽车发明前建造的城市--开进去会让人抓狂，因为它们是针对不同的交通流量设计的。\n\n我们如何构建智能体可以使用的基础设施？\nhttps：//www.anthropic.com/research/agents-in-biology", "image": null, "sourceName": "X：Anthropic (@AnthropicAI)", "sourceCategory": "tip", "publishedAt": "2026-06-08T18:39:56.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-aa776d5c", "url": "https://www.theverge.com/tech/946032/apple-home-ai-camera-descriptions-search-4k", "canonicalUrl": "https://www.theverge.com/tech/946032/apple-home-ai-camera-descriptions-search-4k", "title": "Apple Home摄像头获得Apple Intelligence增强", "titleEn": "Cameras get an Apple Intelligence boost in Apple Home", "description": "Apple Home中的安全摄像头将生成文本描述，并支持用自然语言搜索录像。HomeKit Secure Video终于获得4K支持。", "descriptionEn": "Apple Home中的安全摄像头将生成文本描述，并支持用自然语言搜索录像。HomeKit Secure Video终于获得4K支持。", "image": null, "sourceName": "The Verge：订阅版科技（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:37:31.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-438e6c5f", "url": "https://x.com/jxnlco/status/2064050376312021402", "canonicalUrl": "https://x.com/jxnlco/status/2064050376312021402", "title": "macOS 将迎来类似 Codex 的应用", "titleEn": "its like the codex app （windows） but on macos！", "description": "就像 Windows 上的 Codex 应用，但在 macOS 上！", "descriptionEn": "就像 Windows 上的 Codex 应用，但在 macOS 上！", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "industry", "publishedAt": "2026-06-08T18:22:12.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-de83adfd", "url": "https://www.apple.com/newsroom/2026/06/apple-unveils-next-generation-of-apple-intelligence-siri-ai-and-more", "canonicalUrl": "https://www.apple.com/newsroom/2026/06/apple-unveils-next-generation-of-apple-intelligence-siri-ai-and-more", "title": "Apple发布新一代Apple Intelligence和Siri AI", "titleEn": "Apple unveils next generation of Apple Intelligence， Siri AI， and more", "description": "今天Apple预览了即将推出的软件版本，将带来新一代Apple Intelligence和Siri AI。", "descriptionEn": "今天Apple预览了即将推出的软件版本，将带来新一代Apple Intelligence和Siri AI。", "image": null, "sourceName": "Apple：Newsroom（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:18:43.157Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-88e4d409", "url": "https://www.apple.com/newsroom/2026/06/apple-intelligence-brings-powerful-ai-capabilities-into-everyday-experiences", "canonicalUrl": "https://www.apple.com/newsroom/2026/06/apple-intelligence-brings-powerful-ai-capabilities-into-everyday-experiences", "title": "Apple Intelligence 将强大 AI 能力融入日常体验", "titleEn": "Apple Intelligence brings powerful AI capabilities into everyday experiences", "description": "Apple 发布下一代 Apple Intelligence，将 AI 能力集成到 iPhone、iPad 和 Mac 中，带来更个性化和有帮助的日常体验。", "descriptionEn": "Apple 发布下一代 Apple Intelligence，将 AI 能力集成到 iPhone、iPad 和 Mac 中，带来更个性化和有帮助的日常体验。", "image": null, "sourceName": "Apple：Newsroom（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:17:44.449Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-a17a272e", "url": "https://www.apple.com/apple-events/event-stream", "canonicalUrl": "https://www.apple.com/apple-events/event-stream", "title": "苹果 WWDC 2026 直播", "titleEn": "苹果 WWDC 2026 直播", "description": "苹果 WWDC 2026 主题演讲通过官网进行直播，Hacker News 用户讨论热度达到 110 点。", "descriptionEn": "苹果 WWDC 2026 主题演讲通过官网进行直播，Hacker News 用户讨论热度达到 110 点。", "image": null, "sourceName": "Hacker News 热门（buzzing.cc 中文翻译）", "sourceCategory": "industry", "publishedAt": "2026-06-08T18:17:22.096Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-1e14b93b", "url": "https://www.ithome.com/0/961/680.htm", "canonicalUrl": "https://www.ithome.com/0/961/680.htm", "title": "苹果 Siri AI 支持自然语言生成快捷指令，简化自动化流程", "titleEn": "省去繁琐步骤：苹果 Siri AI 官宣支持自然语言生成快捷指令", "description": "在 WWDC26 上，苹果发布 Siri AI 和 Siri App，快捷指令功能接入 Siri AI，用户可用自然语言快速创建快捷指令，无需逐项手动配置。该功能依赖 Apple Intelligence。", "descriptionEn": "在 WWDC26 上，苹果发布 Siri AI 和 Siri App，快捷指令功能接入 Siri AI，用户可用自然语言快速创建快捷指令，无需逐项手动配置。该功能依赖 Apple Intelligence。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T18:16:01.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-78ad0888", "url": "https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-delayed-in-eu-for-ios-27-and-ipados-27", "canonicalUrl": "https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-delayed-in-eu-for-ios-27-and-ipados-27", "title": "受 DMA 影响，Siri AI 在欧盟将随 iOS 27 和 iPadOS 27 延迟上线", "titleEn": "Due to DMA， Siri AI delayed in EU for iOS 27 and iPadOS 27", "description": "由于欧盟《数字市场法案》（DMA），Apple 无法在 iOS 27 和 iPadOS 27 发布时于欧盟地区推出 Siri AI。该功能在欧盟的上线时间将晚于其他地区，具体时间未公布。", "descriptionEn": "由于欧盟《数字市场法案》（DMA），Apple 无法在 iOS 27 和 iPadOS 27 发布时于欧盟地区推出 Siri AI。该功能在欧盟的上线时间将晚于其他地区，具体时间未公布。", "image": null, "sourceName": "Apple：Newsroom（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T18:13:26.639Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-69a54a4d", "url": "https://www.dwarkesh.com/p/the-sample-efficiency-black-hole", "canonicalUrl": "https://www.dwarkesh.com/p/the-sample-efficiency-black-hole", "title": "样本效率黑洞：AI能力背后隐藏的数据需求深渊", "titleEn": "The sample efficiency black hole", "description": "将AI比作一个闪耀着能力的星系，其核心存在一个肉眼不可见的巨大黑洞--数据。这个比喻揭示了AI模型惊人能力背后对海量数据的依赖，样本效率的瓶颈如同引力中心，将各色能力凝聚在一起。", "descriptionEn": "将AI比作一个闪耀着能力的星系，其核心存在一个肉眼不可见的巨大黑洞--数据。这个比喻揭示了AI模型惊人能力背后对海量数据的依赖，样本效率的瓶颈如同引力中心，将各色能力凝聚在一起。", "image": null, "sourceName": "Dwarkesh Patel：Podcast & Blog（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T18:09:37.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-b99a765c", "url": "https://x.com/jxnlco/status/2064047119288164730", "canonicalUrl": "https://x.com/jxnlco/status/2064047119288164730", "title": "Jason Liu 用 poke 追踪 codex 发布", "titleEn": "If you use poke check out my autmation to keep track of codex launch's. Maybe you'll see @interact…", "description": "如果你用 poke，可以看看我的自动化工具，用来跟踪 codex 的发布。\n\n也许有一天你会在其中看到 @interaction 被提及。\n\nhttp：//poke.com/r/7ZaRDldAw3s", "descriptionEn": "如果你用 poke，可以看看我的自动化工具，用来跟踪 codex 的发布。\n\n也许有一天你会在其中看到 @interaction 被提及。\n\nhttp：//poke.com/r/7ZaRDldAw3s", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "tip", "publishedAt": "2026-06-08T18:09:16.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-1c8a271f", "url": "https://www.ithome.com/0/961/673.htm", "canonicalUrl": "https://www.ithome.com/0/961/673.htm", "title": "苹果 Siri 集成至 iPhone 相机应用，可识别物体并记录饮食摄入", "titleEn": "苹果 iOS 27 版 Siri 已集成至相机应用，iPhone 17 等用户可记录饮食", "description": "在 2026 年全球开发者大会（WWDC）上，苹果宣布 Siri 已集成到 iPhone 相机应用中，可识别物体并保存到 Siri 应用。官方展示案例中，用户可在相机中启用 Siri，利用 AI 记录摄入饮食。隐私保护方面，苹果表示通过 Apple Intelligence 第二代设备端模型处理。", "descriptionEn": "在 2026 年全球开发者大会（WWDC）上，苹果宣布 Siri 已集成到 iPhone 相机应用中，可识别物体并保存到 Siri 应用。官方展示案例中，用户可在相机中启用 Siri，利用 AI 记录摄入饮食。隐私保护方面，苹果表示通过 Apple Intelligence 第二代设备端模型处理。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T17:53:50.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-2f930296", "url": "https://www.anthropic.com/research/agents-in-biology", "canonicalUrl": "https://www.anthropic.com/research/agents-in-biology", "title": "为生物学AI智能体铺路", "titleEn": "Paving the way for agents in biology", "description": "一项实验让Claude、Biomni、Edison Analysis、GPT等科研智能体从病毒学数据库NCBI Virus中检索序列数据，即使最强模型也无法稳定达到可靠数据集构建所需的准确率。加入确定性检索层gget virus后，准确率接近100%。研究指出，当前生物学数据基础设施存在碎片化、格式特殊、接口不统一等问题，导致AI智能体难以像在软件领域那样高效工作。确定性检索工具是实现可靠智能体工作流的关键，生物学数据库需为智能体作为规模化用户而设计。", "descriptionEn": "一项实验让Claude、Biomni、Edison Analysis、GPT等科研智能体从病毒学数据库NCBI Virus中检索序列数据，即使最强模型也无法稳定达到可靠数据集构建所需的准确率。加入确定性检索层gget virus后，准确率接近100%。研究指出，当前生物学数据基础设施存在碎片化、格式特殊、接口不统一等问题，导致AI智能体难以像在软件领域那样高效工作。确定性检索工具是实现可靠智能体工作流的关键，生物学数据库需为智能体作为规模化用户而设计。", "image": null, "sourceName": "Anthropic：Research（发表成果 · 网页）", "sourceCategory": "paper", "publishedAt": "2026-06-08T17:46:10.323Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.94, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-6d36759", "url": "https://x.com/jxnlco/status/2064040508926640213", "canonicalUrl": "https://x.com/jxnlco/status/2064040508926640213", "title": "Codex与Computer Use功能强大", "titleEn": "codex and computer use is so powerful", "description": "Codex和Computer Use非常强大", "descriptionEn": "Codex和Computer Use非常强大", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "tip", "publishedAt": "2026-06-08T17:43:00.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "tool-use", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Tool-use workflow signal: tool calling, computer/browser use, search-as-code, sandbox execution, or external API operation relevant to agent runtime UX." }, { "id": "news-1c5fe09c", "url": "https://www.ithome.com/0/961/670.htm", "canonicalUrl": "https://www.ithome.com/0/961/670.htm", "title": "苹果 Apple Intelligence 与谷歌达成合作：引入 Gemini，第二代设备端模型支持听写等功能", "titleEn": "苹果 Apple Intelligence 与谷歌达成合作：引入 Gemini，第二代设备端模型支持听写等功能", "description": "在 WWDC26 上，苹果宣布 Apple Intelligence 与谷歌合作引入 Gemini 大模型，新版 AI 架构支持在设备端和云端运行。第二代设备端模型新增听写、自然语言理解、更生动的语音表达和屏幕感知功能。隐私保护默认设置，端云结合，确保任何人无法访问用户数据。", "descriptionEn": "在 WWDC26 上，苹果宣布 Apple Intelligence 与谷歌合作引入 Gemini 大模型，新版 AI 架构支持在设备端和云端运行。第二代设备端模型新增听写、自然语言理解、更生动的语音表达和屏幕感知功能。隐私保护默认设置，端云结合，确保任何人无法访问用户数据。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T17:38:25.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-1b29db86", "url": "https://www.ithome.com/0/961/669.htm", "canonicalUrl": "https://www.ithome.com/0/961/669.htm", "title": "苹果推出 Siri AI：独立 App，为灵动岛优化", "titleEn": "苹果推出 Siri AI：独立 App，为灵动岛优化", "description": "苹果在 2026 年全球开发者大会（WWDC）上发布 Siri AI，称为\"全新版本的 Siri\"，具备更强的对话能力和表现力，用户可自定义语速、语气和口音。Siri AI 支持系统范围内读取屏幕内容并与应用互动，配有独立 App，灵动岛会显示大气泡。苹果强调每一步以隐私为重。语音优化同时适用于 CarPlay 和 AirPods，可用于查询演唱会信息、提醒报名抽奖、搜索照片等场景。", "descriptionEn": "苹果在 2026 年全球开发者大会（WWDC）上发布 Siri AI，称为\"全新版本的 Siri\"，具备更强的对话能力和表现力，用户可自定义语速、语气和口音。Siri AI 支持系统范围内读取屏幕内容并与应用互动，配有独立 App，灵动岛会显示大气泡。苹果强调每一步以隐私为重。语音优化同时适用于 CarPlay 和 AirPods，可用于查询演唱会信息、提醒报名抽奖、搜索照片等场景。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T17:38:14.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-35325638", "url": "https://x.com/OpenAIDevs/status/2064037734591652065", "canonicalUrl": "https://x.com/OpenAIDevs/status/2064037734591652065", "title": "OpenAI Developers 五月更新：Codex 新增移动端持续运行、Mac 后台、电脑使用、Chrome 插件等，Realtime API 推出 Realtime-2", "titleEn": "http：//x.com/i/article/2064021561112150016", "description": "OpenAI Developers 五月发布多项更新：Codex 用户突破 500 万；新增 ChatGPT 移动端持续运行、Mac 后台运行、跨 Mac 应用电脑使用、Chrome 插件支持网页测试与 DevTools；⌘+⌘ 快捷键截图直达 Codex；Windows 版也支持电脑使用。Realtime API 推出新模型 Realtime-2，用于语音智能体、实时翻译与转录，并提供提示词指南。Agents SDK 新增 TypeScript 支持、沙箱智能体和开源 harness。私有 MCP 服务器可通过 HTTPS 连接 OpenAI 产品。", "descriptionEn": "OpenAI Developers 五月发布多项更新：Codex 用户突破 500 万；新增 ChatGPT 移动端持续运行、Mac 后台运行、跨 Mac 应用电脑使用、Chrome 插件支持网页测试与 DevTools；⌘+⌘ 快捷键截图直达 Codex；Windows 版也支持电脑使用。Realtime API 推出新模型 Realtime-2，用于语音智能体、实时翻译与转录，并提供提示词指南。Agents SDK 新增 TypeScript 支持、沙箱智能体和开源 harness。私有 MCP 服务器可通过 HTTPS 连接 OpenAI 产品。", "image": null, "sourceName": "X：OpenAI Developers (@OpenAIDevs)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T17:31:58.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem", "tool-use", "agent-architecture", "coding-agent" ], "relevanceScore": 0.98, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-81d84d64", "url": "https://x.com/rohanpaul_ai/status/2064036735168963012", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2064036735168963012", "title": "\"CL-BENCH 基准测试：AI 智能体全上下文记忆优于复杂记忆系统\"", "titleEn": "This paper proposes a new test to see whether AI agents truly get better as they gain experience and…", "description": "新论文构建 CL-BENCH 基准，评估 AI 智能体在编程、数据库、预测、无线电信号、扑克、疾病研究 6 个领域中的持续学习能力。每个任务隐藏可随时间习得的模式，考察智能体能否超越预训练知识。测试前沿 LLM 系统采用全上下文记忆、草稿笔记、检索记忆、剧本式记忆及编码智能体设置，结果发现当前记忆密集型 AI 智能体并未可靠优于简单保留完整对话上下文。Claude Sonnet 4.6 使用普通上下文取得最佳总体分数。论文指出智能体仍需更好方法记住有用经验、遗忘过时信息并适应环境变化。", "descriptionEn": "新论文构建 CL-BENCH 基准，评估 AI 智能体在编程、数据库、预测、无线电信号、扑克、疾病研究 6 个领域中的持续学习能力。每个任务隐藏可随时间习得的模式，考察智能体能否超越预训练知识。测试前沿 LLM 系统采用全上下文记忆、草稿笔记、检索记忆、剧本式记忆及编码智能体设置，结果发现当前记忆密集型 AI 智能体并未可靠优于简单保留完整对话上下文。Claude Sonnet 4.6 使用普通上下文取得最佳总体分数。论文指出智能体仍需更好方法记住有用经验、遗忘过时信息并适应环境变化。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "paper", "publishedAt": "2026-06-08T17:28:00.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-9dc11fbe", "url": "https://x.com/Yuchenj_UW/status/2064036389746831813", "canonicalUrl": "https://x.com/Yuchenj_UW/status/2064036389746831813", "title": "编码智能体应设计循环而非提示", "titleEn": "On the whole： \"You shouldn't be prompting coding agents anymore. You should be designing loops that…", "description": "总体来说：\n\n\"你不应该再提示编码智能体了。你应该设计循环来提示你的智能体。\"\n\n循环是临时解决方案：如今的大语言模型判断力很差。它们难以判断何时继续、何时停止或何时调用工具。循环强制智能体更长时间地工作。\n\n对于目前可验证的目标，循环非常强大，正如AutoResearch所示。", "descriptionEn": "总体来说：\n\n\"你不应该再提示编码智能体了。你应该设计循环来提示你的智能体。\"\n\n循环是临时解决方案：如今的大语言模型判断力很差。它们难以判断何时继续、何时停止或何时调用工具。循环强制智能体更长时间地工作。\n\n对于目前可验证的目标，循环非常强大，正如AutoResearch所示。", "image": null, "sourceName": "X：Yuchen Jin (@Yuchenj_UW)", "sourceCategory": "tip", "publishedAt": "2026-06-08T17:26:37.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-afd749a3", "url": "https://x.com/bcherny/status/2064034799711588805", "canonicalUrl": "https://x.com/bcherny/status/2064034799711588805", "title": "Claude Code GA一周年：Boris Cherny谈auto mode与手机编程", "titleEn": "When we first demoed Claude Code internally， it got two reactions on Slack. A year after GA， @_cat…", "description": "Claude Code GA一周年之际，Anthropic工程师Boris Cherny与@_catwu回顾产品演进。此前内部首次演示时在Slack上引发两种截然不同的反应。Cherny分享了他为何偏好auto mode而非plan mode，routines如何在bug出现前自动修复，以及他如今大部分编码都在手机上完成。视频访谈还探讨了Claude Code的未来方向。", "descriptionEn": "Claude Code GA一周年之际，Anthropic工程师Boris Cherny与@_catwu回顾产品演进。此前内部首次演示时在Slack上引发两种截然不同的反应。Cherny分享了他为何偏好auto mode而非plan mode，routines如何在bug出现前自动修复，以及他如今大部分编码都在手机上完成。视频访谈还探讨了Claude Code的未来方向。", "image": null, "sourceName": "X：Boris Cherny (@bcherny)", "sourceCategory": "tip", "publishedAt": "2026-06-08T17:20:18.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-df30a934", "url": "https://x.com/ClaudeDevs/status/2064032814392352816", "canonicalUrl": "https://x.com/ClaudeDevs/status/2064032814392352816", "title": "Claude Code GA一周年回顾：验证与自动模式", "titleEn": "Claude Code's first demo got two Slack reactions. One year after GA， @bcherny and @_catwu look back…", "description": "Claude Code 的第一个演示收到了两个 Slack 反应。\n\nGA 一周年之际，@bcherny 和 @_catwu 回顾：验证最佳实践、为何构建自动模式、例程和循环，以及下一步计划。\n\nhttps：//www.youtube.com/watch？v=Hth_tLaC2j8", "descriptionEn": "Claude Code 的第一个演示收到了两个 Slack 反应。\n\nGA 一周年之际，@bcherny 和 @_catwu 回顾：验证最佳实践、为何构建自动模式、例程和循环，以及下一步计划。\n\nhttps：//www.youtube.com/watch？v=Hth_tLaC2j8", "image": null, "sourceName": "X：Claude Devs (@ClaudeDevs)", "sourceCategory": "tip", "publishedAt": "2026-06-08T17:12:25.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-e5cdd890", "url": "https://x.com/Yuchenj_UW/status/2064031938537500725", "canonicalUrl": "https://x.com/Yuchenj_UW/status/2064031938537500725", "title": "循环提示智能体：LLM判断力不足的临时方案", "titleEn": "\"You should design loops that prompt your agents.\" Loops are the temporary workaround： today's LLMs…", "description": "\"你应该设计循环来提示你的智能体。\"\n\n循环是临时方案：今天的LLM判断力很差。它们很难知道何时继续、何时停止、何时调用工具。\n\n对于可验证的目标，循环非常强大，正如AutoResearch所示。", "descriptionEn": "\"你应该设计循环来提示你的智能体。\"\n\n循环是临时方案：今天的LLM判断力很差。它们很难知道何时继续、何时停止、何时调用工具。\n\n对于可验证的目标，循环非常强大，正如AutoResearch所示。", "image": null, "sourceName": "X：Yuchen Jin (@Yuchenj_UW)", "sourceCategory": "tip", "publishedAt": "2026-06-08T17:08:56.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-5d504ab3", "url": "https://x.com/perplexity_ai/status/2064023455453110286", "canonicalUrl": "https://x.com/perplexity_ai/status/2064023455453110286", "title": "Perplexity与哈佛：AI智能体提效87%降本94%", "titleEn": "We published new research with Harvard on the shift from chat interfaces to autonomous agents like C…", "description": "我们与哈佛大学发表新研究，关于从聊天界面转向像Computer这样的自主智能体的转变。\n\n超过3个月的研究结果表明，使用Computer的工人在完成任务上比仅使用搜索快87%，成本低94%，且满意度更高。\n\nhttps：//research.perplexity.ai/articles/how-ai-agents-reshape-knowledge-work", "descriptionEn": "我们与哈佛大学发表新研究，关于从聊天界面转向像Computer这样的自主智能体的转变。\n\n超过3个月的研究结果表明，使用Computer的工人在完成任务上比仅使用搜索快87%，成本低94%，且满意度更高。\n\nhttps：//research.perplexity.ai/articles/how-ai-agents-reshape-knowledge-work", "image": null, "sourceName": "X：Perplexity (@perplexity_ai)", "sourceCategory": "paper", "publishedAt": "2026-06-08T16:35:14.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-921ff174", "url": "https://x.com/testingcatalog/status/2064018892935119101", "canonicalUrl": "https://x.com/testingcatalog/status/2064018892935119101", "title": "NotebookLM 支持高级智能体推理与新输出格式", "titleEn": "GOOGLE 🔥： @NotebookLM now supports advanced agentic reasoning in chat and new output formats， inclu…", "description": "GOOGLE 🔥： @NotebookLM 现在在聊天中支持高级智能体推理，并新增了包括 Excel 表格和图片在内的输出格式。\n\n仅限 Ultra 订阅用户 👀", "descriptionEn": "GOOGLE 🔥： @NotebookLM 现在在聊天中支持高级智能体推理，并新增了包括 Excel 表格和图片在内的输出格式。\n\n仅限 Ultra 订阅用户 👀", "image": null, "sourceName": "X：Testing Catalog (@testingcatalog)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T16:17:06.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-41ca57a1", "url": "https://x.com/ChatGPTapp/status/2064018770839113769", "canonicalUrl": "https://x.com/ChatGPTapp/status/2064018770839113769", "title": "ChatGPT 新增数据图表生成功能", "titleEn": "Turn data and comparisons into charts， directly in ChatGPT. Available now on mobile and web.", "description": "将数据和比较转化为图表，直接在 ChatGPT 中完成。\n\n现已支持移动端和网页端。", "descriptionEn": "将数据和比较转化为图表，直接在 ChatGPT 中完成。\n\n现已支持移动端和网页端。", "image": null, "sourceName": "X：ChatGPT (@ChatGPTapp)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T16:16:37.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-af66f39d", "url": "https://x.com/jxnlco/status/2064017107608568063", "canonicalUrl": "https://x.com/jxnlco/status/2064017107608568063", "title": "Codex中非代码工件有哪些", "titleEn": "What artifacts do you create in codex outside code？", "description": "你在Codex中除了代码还创建了哪些工件？", "descriptionEn": "你在Codex中除了代码还创建了哪些工件？", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "tip", "publishedAt": "2026-06-08T16:10:00.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-d9a8facd", "url": "https://x.com/NotebookLM/status/2064016460964585549", "canonicalUrl": "https://x.com/NotebookLM/status/2064016460964585549", "title": "NotebookLM重大升级：智能体能力与高级推理", "titleEn": "Introducing a more powerful NotebookLM 🚀 Massive upgrades deliver agentic capabilities in chat， mo…", "description": "推出更强大的 NotebookLM 🚀\n\n重大升级带来了对话中的智能体能力、更高级的推理以及一系列新的输出格式。处理复杂的多步骤研究问题从未如此简单。\n\n现已面向 Google AI Ultra 订阅者推出。", "descriptionEn": "推出更强大的 NotebookLM 🚀\n\n重大升级带来了对话中的智能体能力、更高级的推理以及一系列新的输出格式。处理复杂的多步骤研究问题从未如此简单。\n\n现已面向 Google AI Ultra 订阅者推出。", "image": null, "sourceName": "X：NotebookLM (@NotebookLM)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T16:07:26.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-3b636291", "url": "https://x.com/runwayml/status/2064012425884569627", "canonicalUrl": "https://x.com/runwayml/status/2064012425884569627", "title": "Runway Aleph 2.0 编辑模型：一键适配任意视频格式", "titleEn": "One video， now made for every feed and format. Upload your existing video， choose your desired aspec…", "description": "一个视频，现在可以为每个信息流和格式制作。上传你现有的视频，选择你想要的宽高比，然后观看我们的编辑模型 Aleph 2.0，填充场景的其余部分，就像你从一开始就这样制作一样。\n\n在我们的桌面 Web 应用上尝试，链接如下。", "descriptionEn": "一个视频，现在可以为每个信息流和格式制作。上传你现有的视频，选择你想要的宽高比，然后观看我们的编辑模型 Aleph 2.0，填充场景的其余部分，就像你从一开始就这样制作一样。\n\n在我们的桌面 Web 应用上尝试，链接如下。", "image": null, "sourceName": "X：Runway (@runwayml)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T15:51:24.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-c49ec353", "url": "https://x.com/omarsar0/status/2064010110314107372", "canonicalUrl": "https://x.com/omarsar0/status/2064010110314107372", "title": "运行Claude Opus自主Agent数小时/天的5条技巧", "titleEn": "Great tips. In practice， this is how it roughly looks to run agents autonomously for hours or days…", "description": "@bcherny 分享5条技巧：1） 开启自动权限模式，免手动确认；2） 采用动态工作流，让Opus协调数百/数千Agent；3） 使用/goal或/loop指令促使持续执行；4） 在云端运行Claude Code，可关闭笔记本；5） 确保Opus能端到端自验证--通过Chrome扩展验证网页、iOS/Android模拟MCP验证移动端、启动完整后端服务验证后端。Elvis Saravia强调/goal/loop和验证是关键。", "descriptionEn": "@bcherny 分享5条技巧：1） 开启自动权限模式，免手动确认；2） 采用动态工作流，让Opus协调数百/数千Agent；3） 使用/goal或/loop指令促使持续执行；4） 在云端运行Claude Code，可关闭笔记本；5） 确保Opus能端到端自验证--通过Chrome扩展验证网页、iOS/Android模拟MCP验证移动端、启动完整后端服务验证后端。Elvis Saravia强调/goal/loop和验证是关键。", "image": null, "sourceName": "X：Elvis Saravia (@omarsar0, DAIR.AI)", "sourceCategory": "tip", "publishedAt": "2026-06-08T15:42:12.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "skill-ecosystem", "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-138cc87d", "url": "https://x.com/OpenRouter/status/2064004944613527730", "canonicalUrl": "https://x.com/OpenRouter/status/2064004944613527730", "title": "OpenRouter Advisor 助小模型问高级模型", "titleEn": "New server tool： Advisor Let smaller models consult a higher-intelligence \"advisor\" model. Helps t…", "description": "新服务器工具：Advisor\n\n让较小的模型咨询一个更高智能的\"顾问\"模型。\n\n帮助它们逃出困境循环，并帮助你迁移到更便宜的模型！🧵", "descriptionEn": "新服务器工具：Advisor\n\n让较小的模型咨询一个更高智能的\"顾问\"模型。\n\n帮助它们逃出困境循环，并帮助你迁移到更便宜的模型！🧵", "image": null, "sourceName": "X：OpenRouter (@OpenRouter)", "sourceCategory": "tip", "publishedAt": "2026-06-08T15:21:40.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-insights" ], "relevanceScore": 0.5, "seelesAngle": "Recommended insight: adjacent AI ecosystem signal that may still be useful context for Seeles agent and skill strategy." }, { "id": "news-14aa8ff7", "url": "https://x.com/jxnlco/status/2064002373371007355", "canonicalUrl": "https://x.com/jxnlco/status/2064002373371007355", "title": "Codex可自主启动聊天并展示侧边栏", "titleEn": "It can also do handoffs", "description": "Codex 现在可以为你自主启动新的聊天，并显示在侧边栏中--当你正在处理另一项任务时发现一个问题，这个功能非常方便。此外，它还可以进行任务交接。", "descriptionEn": "Codex 现在可以为你自主启动新的聊天，并显示在侧边栏中--当你正在处理另一项任务时发现一个问题，这个功能非常方便。此外，它还可以进行任务交接。", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T15:11:27.000Z", "discoveredAt": "2026-06-09T00:39:05.555Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-4a31ef00", "url": "https://x.com/kimmonismus/status/2064001045391462907", "canonicalUrl": "https://x.com/kimmonismus/status/2064001045391462907", "title": "Hivemind推出面向AI编程智能体的持续学习功能，即日起开放", "titleEn": "New from Hivemind： continual learning for AI coding agents， available to everyone starting today. I…", "description": "Hivemind发布面向AI编程智能体的持续学习功能，即日起开放。该工具收集团队运行的每个智能体（Claude Code、Codex、Cursor、Hermes、Pi）的轨迹，转化为可复用技能并推送到所有智能体，数据存储在用户自己的云存储中。内置SkillOpt使技能持续训练：Claude Code准确率提升+19.1分，Codex提升+24.8分，在全部52个测试设置中最佳或持平。开源，一行命令安装。", "descriptionEn": "Hivemind发布面向AI编程智能体的持续学习功能，即日起开放。该工具收集团队运行的每个智能体（Claude Code、Codex、Cursor、Hermes、Pi）的轨迹，转化为可复用技能并推送到所有智能体，数据存储在用户自己的云存储中。内置SkillOpt使技能持续训练：Claude Code准确率提升+19.1分，Codex提升+24.8分，在全部52个测试设置中最佳或持平。开源，一行命令安装。", "image": null, "sourceName": "X：Kim (@kimmonismus)", "sourceCategory": "tip", "publishedAt": "2026-06-08T15:06:11.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent", "agent-architecture", "skill-ecosystem" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-2ee0aef0", "url": "https://x.com/testingcatalog/status/2063995467260694758", "canonicalUrl": "https://x.com/testingcatalog/status/2063995467260694758", "title": "Kimi for Work AI Agent发布：原生Agent Swarm、Browser Use", "titleEn": "KIMI 🔥： A new \"Kimi for Work\" AI Agent has been released with support for Native Agent Swarm， Brows…", "description": "Kimi for Work AI Agent已发布，支持原生Agent Swarm（多智能体群）、Browser Use（通过WebBridge实现）以及自有记忆系统。该应用可在macOS和Windows上运行，用户可本地启动多达300个智能体。官方表示这仅是开始，未来将增加更多数据源、工具和Agent能力。", "descriptionEn": "Kimi for Work AI Agent已发布，支持原生Agent Swarm（多智能体群）、Browser Use（通过WebBridge实现）以及自有记忆系统。该应用可在macOS和Windows上运行，用户可本地启动多达300个智能体。官方表示这仅是开始，未来将增加更多数据源、工具和Agent能力。", "image": null, "sourceName": "X：Testing Catalog (@testingcatalog)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T14:44:01.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "tool-use" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-bde87816", "url": "https://x.com/XiaomiMiMo/status/2063993790587904362", "canonicalUrl": "https://x.com/XiaomiMiMo/status/2063993790587904362", "title": "小米 MiMo-V2.5-Pro-UltraSpeed 突破 1，000 tokens/s，单台 8-GPGPU 节点运行 1T MoE 模型", "titleEn": "🚀 1，000+ TOKENS/S ON A 1T MODEL！ 🚀 We are thrilled to release Xiaomi MiMo-V2.5-Pro-UltraSpeed in …", "description": "小米 MiMo 联合 TileRT_AI 发布 MiMo-V2.5-Pro-UltraSpeed，首次在 1 万亿参数 MoE 模型上实现超过 1，000 tokens/s 输出速度，仅用单台标准 8-GPGPU 节点（非 Cerebras 或 Groq 方案）。提供限时免费聊天体验，UltraSpeed API 价格为 3 倍，输出体验提升约 10 倍。申请时间为 6 月 8 日至 23 日（PDT），企业可邮件联系 business-mimo@xiaomi.com。", "descriptionEn": "小米 MiMo 联合 TileRT_AI 发布 MiMo-V2.5-Pro-UltraSpeed，首次在 1 万亿参数 MoE 模型上实现超过 1，000 tokens/s 输出速度，仅用单台标准 8-GPGPU 节点（非 Cerebras 或 Groq 方案）。提供限时免费聊天体验，UltraSpeed API 价格为 3 倍，输出体验提升约 10 倍。申请时间为 6 月 8 日至 23 日（PDT），企业可邮件联系 business-mimo@xiaomi.com。", "image": null, "sourceName": "X：小米 MiMo (@XiaomiMiMo)", "sourceCategory": "ai-models", "publishedAt": "2026-06-08T14:37:21.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-a198eec6", "url": "https://x.com/OpenBMB/status/2063991963133903317", "canonicalUrl": "https://x.com/OpenBMB/status/2063991963133903317", "title": "VoxCPM2 技术报告发布", "titleEn": "🚀 VoxCPM2 Technical Report is now available on arXiv！ VoxCPM2 is the latest speech generation mode…", "description": "面壁智能 OpenBMB 发布 VoxCPM2 技术报告。该模型为最新语音生成模型，拥有 2B 参数，基于超 200 万小时多语言语音数据训练，支持 30 种语言和 9 种中文方言。具备自然语言语音设计、可控及高保真延续性语音克隆能力。技术报告涵盖架构设计、统一序列公式、AudioVAE 高保真语音重建、大规模训练评估，以及零样本和指令跟随 TTS 基准结果。采用 16kHz 语义编码 + 48kHz 波形重建，在公开 TTS 基准上达到 SOTA 或极具竞争力。模型权重、微调代码和推理工具以 Apache 2.0 开源。", "descriptionEn": "面壁智能 OpenBMB 发布 VoxCPM2 技术报告。该模型为最新语音生成模型，拥有 2B 参数，基于超 200 万小时多语言语音数据训练，支持 30 种语言和 9 种中文方言。具备自然语言语音设计、可控及高保真延续性语音克隆能力。技术报告涵盖架构设计、统一序列公式、AudioVAE 高保真语音重建、大规模训练评估，以及零样本和指令跟随 TTS 基准结果。采用 16kHz 语义编码 + 48kHz 波形重建，在公开 TTS 基准上达到 SOTA 或极具竞争力。模型权重、微调代码和推理工具以 Apache 2.0 开源。", "image": null, "sourceName": "X：面壁智能 OpenBMB (@OpenBMB)", "sourceCategory": "ai-models", "publishedAt": "2026-06-08T14:30:05.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-1d569f7a", "url": "https://x.com/Kimi_Moonshot/status/2063990409903112344", "canonicalUrl": "https://x.com/Kimi_Moonshot/status/2063990409903112344", "title": "Kimi Work：本地桌面AI智能体，支持300智能体并行", "titleEn": "Meet Kimi Work - a local AI agent on your desktop that does the work for you. 🔹Native agent swarm：…", "description": "Kimi发布Kimi Work，一款本地运行的桌面AI智能体。支持最多300个AI智能体同时在本地机器并行运行；配合WebBridge浏览器扩展，智能体可在浏览器中导航、搜索、点击、填写并完成任务；内置全球市场数据工具，可直接调用Yahoo Finance和世界银行数据，无需复杂API设置；记忆系统会记录用户偏好、过往决策和上下文。支持macOS（Apple Silicon）和Windows。", "descriptionEn": "Kimi发布Kimi Work，一款本地运行的桌面AI智能体。支持最多300个AI智能体同时在本地机器并行运行；配合WebBridge浏览器扩展，智能体可在浏览器中导航、搜索、点击、填写并完成任务；内置全球市场数据工具，可直接调用Yahoo Finance和世界银行数据，无需复杂API设置；记忆系统会记录用户偏好、过往决策和上下文。支持macOS（Apple Silicon）和Windows。", "image": null, "sourceName": "X：Kimi.ai (@Kimi_Moonshot)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T14:23:55.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-d3f2c3c7", "url": "https://openai.com/index/openai-submits-confidential-s-1", "canonicalUrl": "https://openai.com/index/openai-submits-confidential-s-1", "title": "OpenAI 向 SEC 机密提交 S-1 草案，上市时间未定", "titleEn": "Confidential submission of draft S-1 to the SEC", "description": "OpenAI 近日向 SEC 机密提交了 S-1 草案（即首次公开募股注册声明），目前尚未决定上市时间。", "descriptionEn": "OpenAI 近日向 SEC 机密提交了 S-1 草案（即首次公开募股注册声明），目前尚未决定上市时间。", "image": null, "sourceName": "OpenAI：官网动态（RSS · 排除企业/客户案例）", "sourceCategory": "industry", "publishedAt": "2026-06-08T14:00:00.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-b30a7de3", "url": "https://www.theverge.com/podcast/944138/microsoft-ai-ceo-mustafa-suleyman-superintelligence-agi-openai-automation", "canonicalUrl": "https://www.theverge.com/podcast/944138/microsoft-ai-ceo-mustafa-suleyman-superintelligence-agi-openai-automation", "title": "微软 AI 负责人 Mustafa Suleyman 称超级智能即将到来，但不会取代你的工作", "titleEn": "Microsoft's AI chief says superintelligence is near， but won't take your job", "description": "微软 AI 负责人 Mustafa Suleyman 表示超级智能即将到来，但不会夺走人类的工作岗位。他同时讨论了自动化、OpenAI 相关话题，并认为将 AI 称为\"活着的\"是危险的。", "descriptionEn": "微软 AI 负责人 Mustafa Suleyman 表示超级智能即将到来，但不会夺走人类的工作岗位。他同时讨论了自动化、OpenAI 相关话题，并认为将 AI 称为\"活着的\"是危险的。", "image": null, "sourceName": "The Verge：订阅版科技（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T14:00:00.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-7102cb09", "url": "https://x.com/omarsar0/status/2063983468619477129", "canonicalUrl": "https://x.com/omarsar0/status/2063983468619477129", "title": "新LLM训练：循环编码提示实现长时执行", "titleEn": "The point is that you should start implementing ways to encode instructions/prompts with clear goals…", "description": "关键在于，你应该开始实施在自动化中编码带有明确目标的指令/提示词的方法。\n\n这并不是什么新鲜事，但较新的大语言模型正在被训练以更长时间不间断地执行。循环就是利用这一点的一种方式。", "descriptionEn": "关键在于，你应该开始实施在自动化中编码带有明确目标的指令/提示词的方法。\n\n这并不是什么新鲜事，但较新的大语言模型正在被训练以更长时间不间断地执行。循环就是利用这一点的一种方式。", "image": null, "sourceName": "X：Elvis Saravia (@omarsar0, DAIR.AI)", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:56:20.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-a0a3608", "url": "https://the-decoder.com/frontier-radar-3-how-agentic-ai-is-turning-tokens-into-a-business-metric", "canonicalUrl": "https://the-decoder.com/frontier-radar-3-how-agentic-ai-is-turning-tokens-into-a-business-metric", "title": "Frontier Radar #3：智能体AI如何将token转化为业务指标", "titleEn": "Frontier Radar #3： How agentic AI is turning tokens into a business metric", "description": "生成式AI商业模式正从月订阅制转向按token消耗计费。智能体工作流消耗数倍于传统对话的token，可自主运行数小时，使固定费率不可持续。token价格因响应速度、专业化程度及结果经济价值而异。本期报告梳理订阅制向消费制迁移、看似低廉的token单价背后的实际成本，以及为何仅凭token消耗量衡量AI价值创造是片面的。", "descriptionEn": "生成式AI商业模式正从月订阅制转向按token消耗计费。智能体工作流消耗数倍于传统对话的token，可自主运行数小时，使固定费率不可持续。token价格因响应速度、专业化程度及结果经济价值而异。本期报告梳理订阅制向消费制迁移、看似低廉的token单价背后的实际成本，以及为何仅凭token消耗量衡量AI价值创造是片面的。", "image": null, "sourceName": "The Decoder：AI News（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:54:16.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.94, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-2b3bbd89", "url": "https://mp.weixin.qq.com/s?__biz=MzkzMTY4NTIyNA%3D%3D&mid=2247484250&idx=1&sn=d0a07f5358250f3a54df8fbabe61f09a", "canonicalUrl": "https://mp.weixin.qq.com/s?__biz=MzkzMTY4NTIyNA%3D%3D&mid=2247484250&idx=1&sn=d0a07f5358250f3a54df8fbabe61f09a", "title": "Kimi Code 焕新升级（附视频教程）", "titleEn": "Kimi Code 焕新升级（附视频教程）", "description": "Kimi Code 开源 Coding Agent 迎来大版本升级：一行命令安装，毫秒级启动；新增视频理解能力，支持提取视频风格生成 LUT 文件、长视频切片、根据录屏生成代码；集成同花顺、天眼查等权威数据源，可查询股票价格、财报、学术论文；支持 ACP 协议，可在 JetBrains、Zed 中使用；丰富 hook 生态方便集成其他工具。底层视觉推理由 Kimi K2.6 模型提供。", "descriptionEn": "Kimi Code 开源 Coding Agent 迎来大版本升级：一行命令安装，毫秒级启动；新增视频理解能力，支持提取视频风格生成 LUT 文件、长视频切片、根据录屏生成代码；集成同花顺、天眼查等权威数据源，可查询股票价格、财报、学术论文；支持 ACP 协议，可在 JetBrains、Zed 中使用；丰富 hook 生态方便集成其他工具。底层视觉推理由 Kimi K2.6 模型提供。", "image": null, "sourceName": "公众号：月之暗面（Kimi）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T13:44:16.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-209b2830", "url": "https://x.com/OpenBMB/status/2063977853562278101", "canonicalUrl": "https://x.com/OpenBMB/status/2063977853562278101", "title": "面壁智能与HuggingFace合作举办Build Small黑客松", "titleEn": "🚀 The Build Small Hackathon is officially live！ Check it out here： 🔗 http：//build-small-hackathon-…", "description": "面壁智能（OpenBMB）与HuggingFace合作举办的Build Small黑客松正式上线，鼓励开发者使用OpenBMB模型构建应用。官方列举四大用例：MiniCPM-V 4.6用于自动化金融文档分析；MiniCPM5-1B轻量端侧模型驱动AI桌面宠物；MiniCPM-o 4.5全模态实时视频分析；VoxCPM2语音克隆。活动设有$10，000 OpenBMB特别奖。", "descriptionEn": "面壁智能（OpenBMB）与HuggingFace合作举办的Build Small黑客松正式上线，鼓励开发者使用OpenBMB模型构建应用。官方列举四大用例：MiniCPM-V 4.6用于自动化金融文档分析；MiniCPM5-1B轻量端侧模型驱动AI桌面宠物；MiniCPM-o 4.5全模态实时视频分析；VoxCPM2语音克隆。活动设有$10，000 OpenBMB特别奖。", "image": null, "sourceName": "X：面壁智能 OpenBMB (@OpenBMB)", "sourceCategory": "industry", "publishedAt": "2026-06-08T13:34:01.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation" ], "relevanceScore": 0.78, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-4480859e", "url": "https://x.com/shao__meng/status/2063975404847915364", "canonicalUrl": "https://x.com/shao__meng/status/2063975404847915364", "title": "邵猛开源 Markdown 转 X Article 格式 Skill", "titleEn": "X Article 编辑器敢说最难用，没人反对吧 ？！ 不支持 Markdown 格式，也不支持 Markdown 导入自动转换，自己的编辑器又太难用，真心佩服这些还能坚持写 X Article 的…", "description": "邵猛批评 X Article 编辑器难用、不支持 Markdown 格式，并用 Codex 创建了一个 Skill「md-to-x-article-skill」，可将 Markdown 自动转换为 X Article 展示格式。该工具已开源在 GitHub（shaom/md-to-x-article-skill），用户可 Fork 后按需微调。", "descriptionEn": "邵猛批评 X Article 编辑器难用、不支持 Markdown 格式，并用 Codex 创建了一个 Skill「md-to-x-article-skill」，可将 Markdown 自动转换为 X Article 展示格式。该工具已开源在 GitHub（shaom/md-to-x-article-skill），用户可 Fork 后按需微调。", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:24:17.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-90b7f542", "url": "https://x.com/xiaohu/status/2063972223170556302", "canonicalUrl": "https://x.com/xiaohu/status/2063972223170556302", "title": "小互开源视频翻译工具：一句话自动下载、转写、翻译、烧字幕", "titleEn": "http：//x.com/i/article/2063968924019163136", "description": "小互（@xiaohu）开源视频翻译工具（xiaohu-video-translate），只需说一句\"把链接翻译成中文字幕视频\"即可全自动完成下载、Whisper本地转写、AI翻译润色、烧字幕、出文稿。转写本地运行，不花API费。支持YouTube、Bilibili、抖音等链接及本地文件，英语、日语、韩语、法语、西班牙语等均可转成中文字幕。字幕精确到词级时间戳，按语义断句，每行不超过12字，双语模式下中文大英文小。提供纯中文与中英双语模式。工具由三个子技能组成，可单独或串联使用，适配Claude Code、Codex、OpenClaw等AI编程工具。已开源，附安装指南。", "descriptionEn": "小互（@xiaohu）开源视频翻译工具（xiaohu-video-translate），只需说一句\"把链接翻译成中文字幕视频\"即可全自动完成下载、Whisper本地转写、AI翻译润色、烧字幕、出文稿。转写本地运行，不花API费。支持YouTube、Bilibili、抖音等链接及本地文件，英语、日语、韩语、法语、西班牙语等均可转成中文字幕。字幕精确到词级时间戳，按语义断句，每行不超过12字，双语模式下中文大英文小。提供纯中文与中英双语模式。工具由三个子技能组成，可单独或串联使用，适配Claude Code、Codex、OpenClaw等AI编程工具。已开源，附安装指南。", "image": null, "sourceName": "X：小互 (@xiaohu)", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:11:39.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent", "skill-ecosystem" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-ab4a8526", "url": "https://huggingface.co/blog/build-small-hackathon/thousand-token-wood-sim-v3", "canonicalUrl": "https://huggingface.co/blog/build-small-hackathon/thousand-token-wood-sim-v3", "title": "五个模型经济体中消失的崩溃：控制与涌现", "titleEn": "The crash that vanished： control and emergence in a five-model economy", "description": "用五个不同实验室的AI模型（OpenAI、NVIDIA、OpenBMB及一个自微调的5亿参数模型）各自驱动一个智能体构建经济市场，试图复现此前单一模型下出现的银行挤兑式价格崩溃。结果同一场景下模型不仅不抛售反而囤积，导致价格不跌反涨。通过纯谣言、库存泛滥、加大做空三种方式均无法重现崩溃。最终在结算环节直接覆盖价格，使崩溃成为设计事实。实验表明，AI智能体的涌现行为是偶然的而非稳健的，有效系统需在涌现纹理与确定性控制之间找到精确接缝。", "descriptionEn": "用五个不同实验室的AI模型（OpenAI、NVIDIA、OpenBMB及一个自微调的5亿参数模型）各自驱动一个智能体构建经济市场，试图复现此前单一模型下出现的银行挤兑式价格崩溃。结果同一场景下模型不仅不抛售反而囤积，导致价格不跌反涨。通过纯谣言、库存泛滥、加大做空三种方式均无法重现崩溃。最终在结算环节直接覆盖价格，使崩溃成为设计事实。实验表明，AI智能体的涌现行为是偶然的而非稳健的，有效系统需在涌现纹理与确定性控制之间找到精确接缝。", "image": null, "sourceName": "Hugging Face：Blog（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:10:48.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-12a1a06b", "url": "https://www.ithome.com/0/961/619.htm", "canonicalUrl": "https://www.ithome.com/0/961/619.htm", "title": "郭明錤：WWDC26 不会改变苹果股票下半年向好趋势，但将检验其 AI 叙事", "titleEn": "郭明錤：WWDC26 不会改变苹果股票下半年向好趋势，但将检验其 AI 叙事能否成立", "description": "天风国际分析师郭明錤指出，苹果当前核心叙事是\"即使AI进度暂时落后，最终仍能后来居上\"。供应链调查显示苹果业绩将好到今年年底，强化了\"没有AI都这么好，有了AI还得了\"的看法。WWDC26不会改变股票下半年正向趋势，但真正值得观察的是：在同样使用Gemini的情况下，苹果能否做出比谷歌更好的AI应用、Agentic workflow和端侧/云端混合体验。若肯定，可延长核心叙事；若否定，则意味着Gemini决定了苹果AI体验上限，更多人将重新审视后来居上的叙事。", "descriptionEn": "天风国际分析师郭明錤指出，苹果当前核心叙事是\"即使AI进度暂时落后，最终仍能后来居上\"。供应链调查显示苹果业绩将好到今年年底，强化了\"没有AI都这么好，有了AI还得了\"的看法。WWDC26不会改变股票下半年正向趋势，但真正值得观察的是：在同样使用Gemini的情况下，苹果能否做出比谷歌更好的AI应用、Agentic workflow和端侧/云端混合体验。若肯定，可延长核心叙事；若否定，则意味着Gemini决定了苹果AI体验上限，更多人将重新审视后来居上的叙事。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T13:06:09.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.86, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-150e9dcb", "url": "https://x.com/xiaohu/status/2063963156662358168", "canonicalUrl": "https://x.com/xiaohu/status/2063963156662358168", "title": "人机协作优于完全自动化：关键节点人工介入", "titleEn": "这个说的的挺对的 我最近的一个和很强烈的感受也是 人机协作才是未来 之前我一直追求完全的自动化导致了不停的和AI扯皮或者总觉得对系统、技能不满意一直在修改它… 结果是你越想优化它它就崩溃，…", "description": "小互分享个人经验：此前追求完全自动化，导致不断与AI扯皮、对技能和系统频繁修改，效果反而更差。现在删除所有规则，只在关键节点人工介入，与AI共同完成任务，不仅效果更好，人也轻松许多，不再频繁抱怨AI。", "descriptionEn": "小互分享个人经验：此前追求完全自动化，导致不断与AI扯皮、对技能和系统频繁修改，效果反而更差。现在删除所有规则，只在关键节点人工介入，与AI共同完成任务，不仅效果更好，人也轻松许多，不再频繁抱怨AI。", "image": null, "sourceName": "X：小互 (@xiaohu)", "sourceCategory": "tip", "publishedAt": "2026-06-08T12:35:37.000Z", "discoveredAt": "2026-06-08T12:37:39.576Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem", "automation" ], "relevanceScore": 0.82, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-9975d4c1", "url": "https://the-decoder.com/instagram-ai-chatbot-breach-may-have-affected-over-to-20000-accounts-meta-discloses", "canonicalUrl": "https://the-decoder.com/instagram-ai-chatbot-breach-may-have-affected-over-to-20000-accounts-meta-discloses", "title": "Meta披露Instagram AI聊天机器人漏洞，超2万个账户受影响", "titleEn": "Instagram AI chatbot breach may have affected over to 20，000 accounts， Meta discloses", "description": "Meta首次披露其Instagram AI客服聊天机器人的安全漏洞--至少20，225个账户遭入侵。系统在近七周内将密码重置链接发送至任意邮箱地址而未验证归属，该机器人此前曾被宣传为账户安全举措。", "descriptionEn": "Meta首次披露其Instagram AI客服聊天机器人的安全漏洞--至少20，225个账户遭入侵。系统在近七周内将密码重置链接发送至任意邮箱地址而未验证归属，该机器人此前曾被宣传为账户安全举措。", "image": null, "sourceName": "The Decoder：AI News（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T12:34:03.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-44e25399", "url": "https://openrouter.ai/blog/human-oversight-eu-ai-act-compliance-agent-sdk", "canonicalUrl": "https://openrouter.ai/blog/human-oversight-eu-ai-act-compliance-agent-sdk", "title": "EU AI Act 合规：面向 AI 智能体的人工监督", "titleEn": "EU AI Act Compliance： Human Oversight for AI Agents", "description": "使用智能体 SDK 的人机协作（HITL）工具，可满足 EU AI Act、Colorado AI Act 和 NIST AI RMF 对 AI 智能体的合规要求。", "descriptionEn": "使用智能体 SDK 的人机协作（HITL）工具，可满足 EU AI Act、Colorado AI Act 和 NIST AI RMF 对 AI 智能体的合规要求。", "image": null, "sourceName": "OpenRouter：Announcements（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T12:00:00.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-7814f3c0", "url": "https://x.com/mingchikuo/status/2063951446224310364", "canonicalUrl": "https://x.com/mingchikuo/status/2063951446224310364", "title": "郭明錤：WWDC26不影响苹果2026下半年股价积极趋势，但考验看涨叙事持久力", "titleEn": "WWDC26 won't change Apple's positive 2H26 share-price trend， but it will test the staying power of t…", "description": "郭明錤指出，苹果核心看涨叙事是\"AI暂时落后但最终会迎头赶上\"。供应链显示业务势头年底前强劲，强化\"无AI已不错，有AI更想象\"叙事。故无论WWDC26内容，只要叙事不变，苹果2026下半年股价趋势积极。WWDC26真正看点在于苹果能否用同款Gemini做出比谷歌更好的AI应用、智能体工作流及端云混合体验。若能，叙事延续；若不能，Gemini设定AI上限，\"苹果最终领先\"将受质疑。", "descriptionEn": "郭明錤指出，苹果核心看涨叙事是\"AI暂时落后但最终会迎头赶上\"。供应链显示业务势头年底前强劲，强化\"无AI已不错，有AI更想象\"叙事。故无论WWDC26内容，只要叙事不变，苹果2026下半年股价趋势积极。WWDC26真正看点在于苹果能否用同款Gemini做出比谷歌更好的AI应用、智能体工作流及端云混合体验。若能，叙事延续；若不能，Gemini设定AI上限，\"苹果最终领先\"将受质疑。", "image": null, "sourceName": "X：郭明錤 Ming-Chi Kuo (@mingchikuo)", "sourceCategory": "tip", "publishedAt": "2026-06-08T11:49:05.000Z", "discoveredAt": "2026-06-08T12:37:39.576Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.86, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-b0e18d1b", "url": "https://x.com/mingchikuo/status/2063951426695676278", "canonicalUrl": "https://x.com/mingchikuo/status/2063951426695676278", "title": "WWDC26 不影響 Apple 2H26 股價正向趨勢，但將揭露多頭敘事續航力", "titleEn": "WWDC26 不影響 Apple 2H26 股價正向趨勢，但將揭露多頭敘事的續航力 -- 1. Apple 目前的多頭核心敘事，是一個近乎直覺、沒什麼人反駁的市場共識：「即使 Apple 在 AI…", "description": "郭明錤指出，Apple 當前多頭核心敘事是「AI 雖落後但最終後來居上」。供應鏈調查顯示其業績將好到今年底，進一步強化此敘事。WWDC26 上無論發布什麼，只要該敘事不被破壞，Apple 2H26 股價正向趨勢就不易改變。真正觀察點在於：Apple 能否用 Gemini 做出比 Google 更好的 AI 應用、AI 智能體工作流及設備端+雲端混合體驗。答案肯定則利好敘事延續；否則「Gemini 決定 Apple AI 上限」的質疑將削弱「後來居上」共識。", "descriptionEn": "郭明錤指出，Apple 當前多頭核心敘事是「AI 雖落後但最終後來居上」。供應鏈調查顯示其業績將好到今年底，進一步強化此敘事。WWDC26 上無論發布什麼，只要該敘事不被破壞，Apple 2H26 股價正向趨勢就不易改變。真正觀察點在於：Apple 能否用 Gemini 做出比 Google 更好的 AI 應用、AI 智能體工作流及設備端+雲端混合體驗。答案肯定則利好敘事延續；否則「Gemini 決定 Apple AI 上限」的質疑將削弱「後來居上」共識。", "image": null, "sourceName": "X：郭明錤 Ming-Chi Kuo (@mingchikuo)", "sourceCategory": "tip", "publishedAt": "2026-06-08T11:49:01.000Z", "discoveredAt": "2026-06-08T12:37:39.576Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-961fdb76", "url": "https://huggingface.co/blog/build-small-hackathon/building-pakistan-notice-helper", "canonicalUrl": "https://huggingface.co/blog/build-small-hackathon/building-pakistan-notice-helper", "title": "Pakistan Notice Helper：一款面向本地安全问题的轻量 AI 工具", "titleEn": "Building Pakistan Notice Helper： A Small AI Tool for a Very Local Safety Problem", "description": "Pakistan Notice Helper 是一款安全工具，帮助巴基斯坦用户在点击链接、拨打电话、分享 OTP 或支付前识别可疑消息。它接受文本或截图输入，返回风险等级、简短解释、可见警示标志和安全下一步建议。工具支持英文和乌尔都语，乌尔都模式采用从右向左布局并全界面翻译。最终部署选用 Qwen3.5 4B Q8 模型（通过 llama.cpp + CUDA），在 10 个测试用例中通过了所有高风险诈骗和截图场景。此前曾测试 Qwen3.6 27B（质量高但成本高）和 MiniCPM-V 4.6 Q8（速度慢且不稳定），最终选择了精度、速度和成本平衡的 4B 模型。项目受限于 hackathon 的 32B 模型上限规则。", "descriptionEn": "Pakistan Notice Helper 是一款安全工具，帮助巴基斯坦用户在点击链接、拨打电话、分享 OTP 或支付前识别可疑消息。它接受文本或截图输入，返回风险等级、简短解释、可见警示标志和安全下一步建议。工具支持英文和乌尔都语，乌尔都模式采用从右向左布局并全界面翻译。最终部署选用 Qwen3.5 4B Q8 模型（通过 llama.cpp + CUDA），在 10 个测试用例中通过了所有高风险诈骗和截图场景。此前曾测试 Qwen3.6 27B（质量高但成本高）和 MiniCPM-V 4.6 Q8（速度慢且不稳定），最终选择了精度、速度和成本平衡的 4B 模型。项目受限于 hackathon 的 32B 模型上限规则。", "image": null, "sourceName": "Hugging Face：Blog（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T11:46:56.000Z", "discoveredAt": "2026-06-08T12:37:38.966Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.94, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-e6edbc35", "url": "https://www.bloomberg.com/news/videos/2026-06-08/the-pulse-6-8-2026-video", "canonicalUrl": "https://www.bloomberg.com/news/videos/2026-06-08/the-pulse-6-8-2026-video", "title": "Nvidia CEO称科技股抛售是买入机会", "titleEn": "Nvidia CEO Says Tech Stock Selloff Is a Buying Opportunity | The Pulse 6/8/2026", "description": "在Bloomberg《The Pulse》节目中，Nvidia CEO表示当前科技股抛售提供了买入机会。节目嘉宾包括高盛资产管理多行业投资主管Lindsay Rosner、Multiverse创始人兼CEO Euan Blair，以及英国AI与在线安全部长Kanishka Narayan。", "descriptionEn": "在Bloomberg《The Pulse》节目中，Nvidia CEO表示当前科技股抛售提供了买入机会。节目嘉宾包括高盛资产管理多行业投资主管Lindsay Rosner、Multiverse创始人兼CEO Euan Blair，以及英国AI与在线安全部长Kanishka Narayan。", "image": null, "sourceName": "Bloomberg：Technology（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T11:29:11.000Z", "discoveredAt": "2026-06-08T12:37:39.577Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-990cf2bf", "url": "https://x.com/rohanpaul_ai/status/2063943851291927024", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2063943851291927024", "title": "Kocoro：开源 Mac AI agent 框架，本地化记忆管理", "titleEn": "A longer context window does not solve the real memory problem in AI work. Kocoro just made AI memo…", "description": "Kocoro 是一个引擎级的开源 Mac AI agent 框架。它通过本地 agent 读取用户的过往会话、文件、应用、浏览器、屏幕和终端，将有用事实压缩为记忆，使 agent 无需重复上下文即可继续工作。安全模型以本地优先：工具操作需授权，危险命令被拦截或二次确认，行为可审计，秘密自动脱敏，记忆/同步默认不上传。其\"情景记忆\"将历史会话转化为项目事实、决策、协作者、截止日期和习惯。每晚自动将工作日蒸馏成本地知识图谱，次日直接恢复进度。", "descriptionEn": "Kocoro 是一个引擎级的开源 Mac AI agent 框架。它通过本地 agent 读取用户的过往会话、文件、应用、浏览器、屏幕和终端，将有用事实压缩为记忆，使 agent 无需重复上下文即可继续工作。安全模型以本地优先：工具操作需授权，危险命令被拦截或二次确认，行为可审计，秘密自动脱敏，记忆/同步默认不上传。其\"情景记忆\"将历史会话转化为项目事实、决策、协作者、截止日期和习惯。每晚自动将工作日蒸馏成本地知识图谱，次日直接恢复进度。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T11:18:55.000Z", "discoveredAt": "2026-06-08T12:37:39.577Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.94, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-244cced5", "url": "https://x.com/HuaweiCloud1/status/2063942843585179731", "canonicalUrl": "https://x.com/HuaweiCloud1/status/2063942843585179731", "title": "华为云发布Agentic Infra新范式及系列AI产品", "titleEn": "At Huawei Cloud INSPIRE 2026， Huawei Cloud introduced a new paradigm of Agentic Infra， alongside a s…", "description": "华为云在 INSPIRE 2026 上推出 Agentic Infra 新范式，作为统一承载通用与 AI 负载的基础设施。同时发布全新一代模型训练推理平台与企业智能体平台，并宣布在其 Industry AI Foundry 中设立四个专属专区：智慧医疗专区、具身 AI 专区、智能制造专区和科学计算专区。", "descriptionEn": "华为云在 INSPIRE 2026 上推出 Agentic Infra 新范式，作为统一承载通用与 AI 负载的基础设施。同时发布全新一代模型训练推理平台与企业智能体平台，并宣布在其 Industry AI Foundry 中设立四个专属专区：智慧医疗专区、具身 AI 专区、智能制造专区和科学计算专区。", "image": null, "sourceName": "X：华为云 (@HuaweiCloud1)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T11:14:54.000Z", "discoveredAt": "2026-06-08T12:37:39.577Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-6ed3d695", "url": "https://mp.weixin.qq.com/s?__biz=MzI0NzU1NzI5NQ%3D%3D&mid=2247537114&idx=1&sn=f4a2f28ea8892433ee5f625f06ccd09e", "canonicalUrl": "https://mp.weixin.qq.com/s?__biz=MzI0NzU1NzI5NQ%3D%3D&mid=2247537114&idx=1&sn=f4a2f28ea8892433ee5f625f06ccd09e", "title": "火山方舟Coding Plan与Agent Plan升级，限时2.5折", "titleEn": "Agent Plan、Coding Plan限时优惠：2.5折畅享多模型！", "description": "火山方舟Coding Plan与Agent Plan升级，集成MiniMax M3、DeepSeek V4系列、GLM-5.1等多款三方模型。即日起至2026年8月7日，新购/升级/续费40元、200元档位可享首两月2.5折（9.9元/月、49.9元/月）。MiniMax M3为新一代旗舰，支持100万token上下文窗口及原生多模态理解。Agent Plan为业界首个Agent套餐包，整合字节自研Seed系列模型与Harness工具，免费提供联网搜索和Embedding记忆能力，可配合Claude Code、OpenClaw、TRAE等工具使用。", "descriptionEn": "火山方舟Coding Plan与Agent Plan升级，集成MiniMax M3、DeepSeek V4系列、GLM-5.1等多款三方模型。即日起至2026年8月7日，新购/升级/续费40元、200元档位可享首两月2.5折（9.9元/月、49.9元/月）。MiniMax M3为新一代旗舰，支持100万token上下文窗口及原生多模态理解。Agent Plan为业界首个Agent套餐包，整合字节自研Seed系列模型与Harness工具，免费提供联网搜索和Embedding记忆能力，可配合Claude Code、OpenClaw、TRAE等工具使用。", "image": null, "sourceName": "公众号：火山引擎", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T11:03:19.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-77a1fec2", "url": "https://x.com/alibaba_cloud/status/2063935790045077716", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063935790045077716", "title": "阿里云EMR性能与AI原生能力获企业认可", "titleEn": "Discover why global enterprises choose Alibaba Cloud E-MapReduce （EMR）. #1 in TPC-H & TPC-DS benchma…", "description": "发现为什么全球企业选择阿里云 E-MapReduce （EMR）。在 TPC-H 和 TPC-DS 基准测试中排名第一，集成 Hadoop、Spark、StarRocks 等。\n\n在本视频中：\n✅ 灵活部署（Serverless、ECS、ACK）\n✅ AI 原生：通过 SQL/PySpark 运行大语言模型\n✅ Spark 快 4 倍，StarRocks 快 10 倍，成本更低\n✅ 企业级安全\n\n弹性、高效、智能的大数据。\n\n🔗 了解更多：https：//int.alibabacloud.com/m/1000414155/", "descriptionEn": "发现为什么全球企业选择阿里云 E-MapReduce （EMR）。在 TPC-H 和 TPC-DS 基准测试中排名第一，集成 Hadoop、Spark、StarRocks 等。\n\n在本视频中：\n✅ 灵活部署（Serverless、ECS、ACK）\n✅ AI 原生：通过 SQL/PySpark 运行大语言模型\n✅ Spark 快 4 倍，StarRocks 快 10 倍，成本更低\n✅ 企业级安全\n\n弹性、高效、智能的大数据。\n\n🔗 了解更多：https：//int.alibabacloud.com/m/1000414155/", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T10:46:53.000Z", "discoveredAt": "2026-06-08T12:37:39.577Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-6c8a169d", "url": "https://x.com/alibaba_cloud/status/2063926972460183901", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063926972460183901", "title": "阿里云STAROps UModel：知识图谱让AIOps效率提升10倍", "titleEn": "AI Agent costs are spiraling？ Uber's budget burn reveals the \"Tokenmaxxing\" trap. The fix isn't just…", "description": "AI Agent成本飙升，Uber预算浪费暴露了\"Tokenmaxxing\"陷阱。阿里云STAROps推出UModel解决方案，采用结构优先方法，用知识图谱替代盲文本推理，通过即时解析依赖关系大幅减少Token消耗和工具调用，实现10倍效率提升，同时确保企业系统可审计与零信任准确性。", "descriptionEn": "AI Agent成本飙升，Uber预算浪费暴露了\"Tokenmaxxing\"陷阱。阿里云STAROps推出UModel解决方案，采用结构优先方法，用知识图谱替代盲文本推理，通过即时解析依赖关系大幅减少Token消耗和工具调用，实现10倍效率提升，同时确保企业系统可审计与零信任准确性。", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T10:11:50.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "tool-use", "agent-safety" ], "relevanceScore": 0.82, "seelesAngle": "Tool-use workflow signal: tool calling, computer/browser use, search-as-code, sandbox execution, or external API operation relevant to agent runtime UX." }, { "id": "news-cbee4353", "url": "https://x.com/Khazix0918/status/2063923890272051228", "canonicalUrl": "https://x.com/Khazix0918/status/2063923890272051228", "title": "微信Agent消息频传：内测右滑唤出、联手手机厂商、发布开发者指引", "titleEn": "最近几天，微信Agent曝光的消息越来越多了。 我简单汇总一下： 6月2日，据外媒，腾讯正在测试一款嵌入微信的AI Agent，用户在微信主界面向右滑动即可唤出Agent对话窗口，通过自然语言下达指…", "description": "近日微信Agent消息密集：6月2日外媒称腾讯正测试嵌入微信的AI Agent，用户右滑唤出窗口，可通过自然语言调用数百万小程序完成点单等任务；6月4日微信与华为等手机厂商合作推出A2A能力，可通过语音助手发起音视频通话；6月8日发布开发者指引，提供自动接入模式，因小程序运行在微信沙箱内，微信可直接读取与分析。这标志着微信Agent正连接硬件生态与内部小程序，触及Agentic Commerce雏形。", "descriptionEn": "近日微信Agent消息密集：6月2日外媒称腾讯正测试嵌入微信的AI Agent，用户右滑唤出窗口，可通过自然语言调用数百万小程序完成点单等任务；6月4日微信与华为等手机厂商合作推出A2A能力，可通过语音助手发起音视频通话；6月8日发布开发者指引，提供自动接入模式，因小程序运行在微信沙箱内，微信可直接读取与分析。这标志着微信Agent正连接硬件生态与内部小程序，触及Agentic Commerce雏形。", "image": null, "sourceName": "X：卡兹克 (@Khazix0918)", "sourceCategory": "tip", "publishedAt": "2026-06-08T09:59:35.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "tool-use" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-a2d6195d", "url": "https://x.com/alibaba_cloud/status/2063923881857945815", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063923881857945815", "title": "AgentScope Java 2.0 发布", "titleEn": "🚀 AgentScope Java 2.0 released！ Solving enterprise AI agent challenges： ✅ Distributed & Stateless： …", "description": "阿里云发布AgentScope Java 2.0，面向企业级AI智能体开发。新版本主要特性：分布式无状态架构，支持K8s弹性扩缩容与会话恢复；多租户隔离，通过Workspace抽象实现安全数据分离；长期稳定性，HarnessAgent负责上下文管理与容错；安全执行，提供细粒度权限控制和Human-in-the-Loop支持。适用于JVM生产环境。开源地址见推文链接。", "descriptionEn": "阿里云发布AgentScope Java 2.0，面向企业级AI智能体开发。新版本主要特性：分布式无状态架构，支持K8s弹性扩缩容与会话恢复；多租户隔离，通过Workspace抽象实现安全数据分离；长期稳定性，HarnessAgent负责上下文管理与容错；安全执行，提供细粒度权限控制和Human-in-the-Loop支持。适用于JVM生产环境。开源地址见推文链接。", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T09:59:33.000Z", "discoveredAt": "2026-06-08T12:37:38.967Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-safety", "agent-architecture" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-337ef70", "url": "https://mp.weixin.qq.com/s?__biz=MzkxMTYyMTAzNA%3D%3D&mid=2247501437&idx=1&sn=d6c6dcb853adbed4ea946711d478779a", "canonicalUrl": "https://mp.weixin.qq.com/s?__biz=MzkxMTYyMTAzNA%3D%3D&mid=2247501437&idx=1&sn=d6c6dcb853adbed4ea946711d478779a", "title": "Agent 辅助开发，一站式打通 Qwen3-VL Android 端侧推理", "titleEn": "Agent 辅助开发，一站式打通 Qwen3-VL Android 端侧推理", "description": "通义实验室教程演示了如何用 Agent（Qoder）辅助完成 Qwen3-VL-2B 模型在 Android 端侧的全流程部署。操作包括：检查 JDK 21、NDK 27、CMake 3.18.1 等环境；创建 arm64-v8a 的 Native C++ 工程（minSdk 29、compileSdk 35）；通过 ModelScope CLI 下载约 1.4GB 的 MNN/Qwen3-VL-2B-Instruct-MNN 模型；编译开启 LLM 视觉支持（MNN_BUILD_LLM、LLM_SUPPORT_VISION）的 libMNN.so；构建 APK 并推送模型至手机私有目录。最终 App 提供图文推理页面，输出 MNN 版本（v3.5.0）、ABI 及推理指标。所有繁琐步骤均可由 Agent 自动执行。", "descriptionEn": "通义实验室教程演示了如何用 Agent（Qoder）辅助完成 Qwen3-VL-2B 模型在 Android 端侧的全流程部署。操作包括：检查 JDK 21、NDK 27、CMake 3.18.1 等环境；创建 arm64-v8a 的 Native C++ 工程（minSdk 29、compileSdk 35）；通过 ModelScope CLI 下载约 1.4GB 的 MNN/Qwen3-VL-2B-Instruct-MNN 模型；编译开启 LLM 视觉支持（MNN_BUILD_LLM、LLM_SUPPORT_VISION）的 libMNN.so；构建 APK 并推送模型至手机私有目录。最终 App 提供图文推理页面，输出 MNN 版本（v3.5.0）、ABI 及推理指标。所有繁琐步骤均可由 Agent 自动执行。", "image": null, "sourceName": "公众号：通义实验室（千问）", "sourceCategory": "tip", "publishedAt": "2026-06-08T09:54:08.000Z", "discoveredAt": "2026-06-08T12:37:38.967Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "automation" ], "relevanceScore": 0.86, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-e108e766", "url": "https://www.ithome.com/0/961/537.htm", "canonicalUrl": "https://www.ithome.com/0/961/537.htm", "title": "网传乘客用智能眼镜偷拍空姐上传社区，乐奇 Rokid 回应称高度重视并紧急启动专项整改", "titleEn": "网传乘客用智能眼镜偷拍空姐上传社区，乐奇 Rokid 回应称高度重视并紧急启动专项整改", "description": "近日有网友在乐奇 Rokid 智能眼镜用户社区发布用该眼镜拍摄\"春秋航空\"空姐的照片，引发隐私担忧。电商平台出现用于遮挡拍摄指示灯的\"遮光贴\"，销量超5000件。乐奇 Rokid 今日声明已紧急启动专项整改：全面清理社区违规内容并升级审核机制；已向电商平台投诉推进违规配件下架与溯源；现售产品标配硬件级拍摄指示灯、遮挡检测与底层防护算法，后续将升级防护体系，杜绝恶意改装。", "descriptionEn": "近日有网友在乐奇 Rokid 智能眼镜用户社区发布用该眼镜拍摄\"春秋航空\"空姐的照片，引发隐私担忧。电商平台出现用于遮挡拍摄指示灯的\"遮光贴\"，销量超5000件。乐奇 Rokid 今日声明已紧急启动专项整改：全面清理社区违规内容并升级审核机制；已向电商平台投诉推进违规配件下架与溯源；现售产品标配硬件级拍摄指示灯、遮挡检测与底层防护算法，后续将升级防护体系，杜绝恶意改装。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T09:45:51.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-2dedb2e5", "url": "https://x.com/alibaba_cloud/status/2063918611954012507", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063918611954012507", "title": "Qwen3.7-Plus 发布限时八折优惠", "titleEn": "🔥 Launch Special for Qwen3.7-Plus： Get 20% OFF now！ ✅ Multimodal Interactive Hybrid Agents ✅ Coding…", "description": "🔥 Qwen3.7-Plus 发布特惠：现在享受八折！\n✅ 多模态交互式智能体\n✅ 编程与生产力助手\n✅ 视觉智能体\n✅ 跨任务泛化\n不要错过升级机会。👇\nhttps：//int.alibabacloud.com/m/1000414123/\n\n#Qwen #AI #Multimodal #AlibabaCloud #AgenticAI", "descriptionEn": "🔥 Qwen3.7-Plus 发布特惠：现在享受八折！\n✅ 多模态交互式智能体\n✅ 编程与生产力助手\n✅ 视觉智能体\n✅ 跨任务泛化\n不要错过升级机会。👇\nhttps：//int.alibabacloud.com/m/1000414123/\n\n#Qwen #AI #Multimodal #AlibabaCloud #AgenticAI", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "ai-models", "publishedAt": "2026-06-08T09:38:37.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-b6733cef", "url": "https://mp.weixin.qq.com/s/NP-KDxsSx5PMJWNCLUA13Q", "canonicalUrl": "https://mp.weixin.qq.com/s/NP-KDxsSx5PMJWNCLUA13Q", "title": "微信正在变成Agent时代的操作系统", "titleEn": "微信正在变成Agent时代的操作系统。", "description": "微信近期密集布局AI Agent：据外媒，用户在主界面右滑可唤出Agent对话窗口，自动调用小程序完成点单等任务；与华为、荣耀、小米、OPPO、vivo合作推出A2A助手能力，允许手机语音助理发起微信音视频通话或发送消息；6月8日发布开发者指引，提供自动接入模式，通过GUI Agent帮助数百万小程序实现AI调用。微信试图整合内部小程序生态与外部硬件，成为AI时代的操作系统，并触及Agentic Commerce（代理式交易）模式。", "descriptionEn": "微信近期密集布局AI Agent：据外媒，用户在主界面右滑可唤出Agent对话窗口，自动调用小程序完成点单等任务；与华为、荣耀、小米、OPPO、vivo合作推出A2A助手能力，允许手机语音助理发起微信音视频通话或发送消息；6月8日发布开发者指引，提供自动接入模式，通过GUI Agent帮助数百万小程序实现AI调用。微信试图整合内部小程序生态与外部硬件，成为AI时代的操作系统，并触及Agentic Commerce（代理式交易）模式。", "image": null, "sourceName": "公众号：数字生命卡兹克", "sourceCategory": "tip", "publishedAt": "2026-06-08T09:38:31.000Z", "discoveredAt": "2026-06-09T00:39:05.556Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-82a3a6ee", "url": "https://mp.weixin.qq.com/s?__biz=MzIyMzA5NjEyMA%3D%3D&mid=2647683054&idx=1&sn=dcaa25bd4fdfc9bd5c0211532021bf7c", "canonicalUrl": "https://mp.weixin.qq.com/s?__biz=MzIyMzA5NjEyMA%3D%3D&mid=2647683054&idx=1&sn=dcaa25bd4fdfc9bd5c0211532021bf7c", "title": "微信AI Agent生态曝光：嵌入小程序调用与手机厂商合作", "titleEn": "最近几天，微信Agent曝光的消息越来越多了。 我简单汇总一下： 6月2日，据外媒外媒，腾讯正在测试一款嵌入微信的AI Agent，用户在微信主界面向右滑动即可唤出Agent对话窗口，通过自然语言下", "description": "腾讯正测试嵌入微信的AI Agent，用户右滑唤出对话窗口，通过自然语言指令调用数百万小程序完成点咖啡等任务。微信还与华为、荣耀、小米、OPPO、vivo合作推出A2A助手能力，可通过手机语音助理发起微信音视频通话或发送消息。6月8日，微信官方发布《关于开发者接入微信AI生态的指引》，提供自动接入模式，利用微信沙箱全自动改造小程序，使其可被AI直接调用。这一布局被视为Agentic Commerce（代理式交易）的雏形。", "descriptionEn": "腾讯正测试嵌入微信的AI Agent，用户右滑唤出对话窗口，通过自然语言指令调用数百万小程序完成点咖啡等任务。微信还与华为、荣耀、小米、OPPO、vivo合作推出A2A助手能力，可通过手机语音助理发起微信音视频通话或发送消息。6月8日，微信官方发布《关于开发者接入微信AI生态的指引》，提供自动接入模式，利用微信沙箱全自动改造小程序，使其可被AI直接调用。这一布局被视为Agentic Commerce（代理式交易）的雏形。", "image": null, "sourceName": "公众号：数字生命卡兹克", "sourceCategory": "tip", "publishedAt": "2026-06-08T09:38:30.000Z", "discoveredAt": "2026-06-09T00:39:05.021Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture", "tool-use" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-34b04eaf", "url": "https://mp.weixin.qq.com/s?__biz=MzkzMDQ5NTQwMQ%3D%3D&mid=2247488842&idx=1&sn=3dab63189fa60b3b78c5ad72e3358791", "canonicalUrl": "https://mp.weixin.qq.com/s?__biz=MzkzMDQ5NTQwMQ%3D%3D&mid=2247488842&idx=1&sn=3dab63189fa60b3b78c5ad72e3358791", "title": "生数科技与华策影视达成战略合作，共建AI视听创制中心", "titleEn": "生数科技与华策集团达成战略合作，推动AI与影视产业融合落地", "description": "6月7日，生数科技与华策影视签署战略合作协议，共同推动AI与影视产业融合。双方将以华策影视科技产教示范区为载体、生数科技Vidu视频生成大模型为技术底座，设立\"AI视听创制中心\"，探索AI虚拟制作与实拍结合。同时在浙江华策影视学校设立\"华策&生数AI影视创制专业\"，课程覆盖传统编导剪与AI视频制作技术。内容生产层面推广\"实拍+AI\"融合方案并优先在华策项目中试点，创作者生态层面引导Vidu社区超级创作者到产教基地注册。", "descriptionEn": "6月7日，生数科技与华策影视签署战略合作协议，共同推动AI与影视产业融合。双方将以华策影视科技产教示范区为载体、生数科技Vidu视频生成大模型为技术底座，设立\"AI视听创制中心\"，探索AI虚拟制作与实拍结合。同时在浙江华策影视学校设立\"华策&生数AI影视创制专业\"，课程覆盖传统编导剪与AI视频制作技术。内容生产层面推广\"实拍+AI\"融合方案并优先在华策项目中试点，创作者生态层面引导Vidu社区超级创作者到产教基地注册。", "image": null, "sourceName": "公众号：生数科技（Vidu·视频）", "sourceCategory": "industry", "publishedAt": "2026-06-08T09:30:33.000Z", "discoveredAt": "2026-06-08T12:37:38.967Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-469b2956", "url": "https://www.bloomberg.com/news/videos/2026-06-08/uk-ai-minister-narayan-on-bloomberg-tv-video", "canonicalUrl": "https://www.bloomberg.com/news/videos/2026-06-08/uk-ai-minister-narayan-on-bloomberg-tv-video", "title": "英国AI部长Narayan亮相彭博电视，为伦敦科技周揭幕", "titleEn": "UK AI Minister Narayan on Bloomberg TV", "description": "英国AI与在线安全部长Kanishka Narayan在彭博《The Pulse》节目中接受Tom Mackenzie采访，为伦敦科技周拉开序幕。", "descriptionEn": "英国AI与在线安全部长Kanishka Narayan在彭博《The Pulse》节目中接受Tom Mackenzie采访，为伦敦科技周拉开序幕。", "image": null, "sourceName": "Bloomberg：Technology（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T09:13:19.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-e9758217", "url": "https://x.com/shao__meng/status/2063902211978223678", "canonicalUrl": "https://x.com/shao__meng/status/2063902211978223678", "title": "邵猛开源 Brand to DESIGN.md 技能，提醒复刻易生新\"AI Slop\"", "titleEn": "最近看了不少 Design Skill、Taste Skill、Anti-AI-slop design skill 等等，我自己也开源了一个 Brand to DESIGN.md Skill （htt…", "description": "邵猛关注 Design Skill、Taste Skill、Anti-AI-slop design skill，并开源 Brand to DESIGN.md Skill（GitHub： shaom/brand-to-design-md-skill），让 Agent 学习设计品味后复刻网站。但他指出，这种复刻看多了反而从 Anti-AI-slop 衍生新\"AI Slop\"，缺少设计精髓，仅皮毛相仿。", "descriptionEn": "邵猛关注 Design Skill、Taste Skill、Anti-AI-slop design skill，并开源 Brand to DESIGN.md Skill（GitHub： shaom/brand-to-design-md-skill），让 Agent 学习设计品味后复刻网站。但他指出，这种复刻看多了反而从 Anti-AI-slop 衍生新\"AI Slop\"，缺少设计精髓，仅皮毛相仿。", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "tip", "publishedAt": "2026-06-08T08:33:27.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-a38ee614", "url": "https://www.marktechpost.com/2026/06/08/google-research-adds-agentic-rag-to-gemini-enterprise-agent-platform-with-a-sufficient-context-agent-for-multi-hop-queries", "canonicalUrl": "https://www.marktechpost.com/2026/06/08/google-research-adds-agentic-rag-to-gemini-enterprise-agent-platform-with-a-sufficient-context-agent-for-multi-hop-queries", "title": "Google Research 为 Gemini Enterprise Agent Platform 引入 Agentic RAG 与 Sufficient Context Agent 支持多跳查询", "titleEn": "Google Research Adds Agentic RAG to Gemini Enterprise Agent Platform with a Sufficient Context Agent for multi-hop queries", "description": "Google Research 在 Gemini Enterprise Agent Platform 中推出一个 Agentic RAG 框架。该框架中的 Sufficient Context Agent 会持续重新搜索，直到多跳、多源查询拥有足够的 grounding 来生成回答。相比标准 RAG，该框架将事实准确性最高提升 34%。", "descriptionEn": "Google Research 在 Gemini Enterprise Agent Platform 中推出一个 Agentic RAG 框架。该框架中的 Sufficient Context Agent 会持续重新搜索，直到多跳、多源查询拥有足够的 grounding 来生成回答。相比标准 RAG，该框架将事实准确性最高提升 34%。", "image": null, "sourceName": "MarkTechPost（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T08:25:51.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-b62761db", "url": "https://www.ithome.com/0/961/493.htm", "canonicalUrl": "https://www.ithome.com/0/961/493.htm", "title": "《2025游戏安全白皮书》发布：国内游戏黑产市场规模已超100亿元", "titleEn": "国内游戏黑产市场规模预估已超 100 亿元人民币，《2025 游戏安全白皮书》发布", "description": "2026游戏安全行业峰会今日在深圳举行，正式发布《2025游戏安全白皮书》。报告显示，国内游戏黑产市场规模预估已超100亿元人民币，覆盖外挂、黑产账号等领域。AI技术正深刻改变游戏安全攻防格局，安全威胁规模与复杂性持续攀升。游戏工委秘书长敖然呼吁全产业链协同治理。峰会还将公布《无畏契约》反作弊智能化运营创新及《三角洲行动》反外挂全记录等议题。", "descriptionEn": "2026游戏安全行业峰会今日在深圳举行，正式发布《2025游戏安全白皮书》。报告显示，国内游戏黑产市场规模预估已超100亿元人民币，覆盖外挂、黑产账号等领域。AI技术正深刻改变游戏安全攻防格局，安全威胁规模与复杂性持续攀升。游戏工委秘书长敖然呼吁全产业链协同治理。峰会还将公布《无畏契约》反作弊智能化运营创新及《三角洲行动》反外挂全记录等议题。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T08:14:29.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.94, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-b44842b9", "url": "https://www.ithome.com/0/961/480.htm", "canonicalUrl": "https://www.ithome.com/0/961/480.htm", "title": "微信AI官宣内测：两种接入模式供开发者选择", "titleEn": "微信 AI 官宣内测：用户一句话，可直接操作小程序页面", "description": "微信开发者官方今日发布指引，确认微信AI正在内测阶段。开放平台提供两种接入模式：自动模式可授权平台读取小程序源码，无需额外开发即可让微信AI直接操作页面；开发模式下开发者可自主开发技能，审核后由微信AI调用。两种模式可同时开启。微信AI是微信内的AI助手，用户通过自然语言对话即可调用、访问和操作小程序。服务条款称\"微信AI\"可能非最终名称。接入与否不影响现有小程序服务。", "descriptionEn": "微信开发者官方今日发布指引，确认微信AI正在内测阶段。开放平台提供两种接入模式：自动模式可授权平台读取小程序源码，无需额外开发即可让微信AI直接操作页面；开发模式下开发者可自主开发技能，审核后由微信AI调用。两种模式可同时开启。微信AI是微信内的AI助手，用户通过自然语言对话即可调用、访问和操作小程序。服务条款称\"微信AI\"可能非最终名称。接入与否不影响现有小程序服务。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T08:02:38.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-11ee48ce", "url": "https://www.bloomberg.com/news/articles/2026-06-08/how-coreweave-sees-the-market-for-compute-right-now", "canonicalUrl": "https://www.bloomberg.com/news/articles/2026-06-08/how-coreweave-sees-the-market-for-compute-right-now", "title": "CoreWeave 如何看待当前计算市场", "titleEn": "How CoreWeave Sees the Market for Compute Right Now ", "description": "CoreWeave 对当前计算市场进行了分析，指出推动增长的因素和限制增长的因素。该公司的市场看法即围绕这两个维度展开。", "descriptionEn": "CoreWeave 对当前计算市场进行了分析，指出推动增长的因素和限制增长的因素。该公司的市场看法即围绕这两个维度展开。", "image": null, "sourceName": "Bloomberg：Technology（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T08:00:32.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-dc0aba47", "url": "https://x.com/PixVerse_/status/2063890444413706442", "canonicalUrl": "https://x.com/PixVerse_/status/2063890444413706442", "title": "PixVerse Creative Partner Program 2.0 上线", "titleEn": "Thank you for the thoughtful sharing. Our chat at Berlinale was truly inspiring and rewarding！ Looki…", "description": "PixVerse 宣布 Creative Partner Program （CPP） 2.0 正式上线，面向 AI 视频创作者。该计划围绕真实创作：发布、测试、教学、故事片及公开推广新想法。核心福利包括：合格帖子每周最高可获得 150，000 credits；每周 $2，500 现金奖池，单个创作者单周最高 $850；对大项目、教程、测试内容和电影节投稿提供特别支持。参与者还可获得更多曝光、项目机会及与 PixVerse 共创的可能。申请可通过官网进行。", "descriptionEn": "PixVerse 宣布 Creative Partner Program （CPP） 2.0 正式上线，面向 AI 视频创作者。该计划围绕真实创作：发布、测试、教学、故事片及公开推广新想法。核心福利包括：合格帖子每周最高可获得 150，000 credits；每周 $2，500 现金奖池，单个创作者单周最高 $850；对大项目、教程、测试内容和电影节投稿提供特别支持。参与者还可获得更多曝光、项目机会及与 PixVerse 共创的可能。申请可通过官网进行。", "image": null, "sourceName": "X：PixVerse (@PixVerse_)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T07:46:41.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-963cc59d", "url": "https://x.com/xiaohu/status/2063887105508385205", "canonicalUrl": "https://x.com/xiaohu/status/2063887105508385205", "title": "Google 发布 Agentic RAG：\"质检 Agent\"让系统知道没搜全，准确率提升 34%", "titleEn": "http：//x.com/i/article/2063870567355400192", "description": "Google 发布 Agentic RAG 框架，核心新增 Sufficient Context Agent，负责在生成答案前检查检索材料是否充分，若不充分则生成缺失分析并引导系统迭代搜索。在 FramesQA 多跳测试中准确率最高提升 34%，从 4 个数据库检索时正确率达 90.1%，速度仅慢 3% 以内。该设计基于前作发现：Gemini 1.5 Pro 判断\"上下文充分性\"准确率达 93%，且\"相关≠够用\"是幻觉关键原因。目前以公开预览在 Gemini Enterprise Agent Platform 开放。", "descriptionEn": "Google 发布 Agentic RAG 框架，核心新增 Sufficient Context Agent，负责在生成答案前检查检索材料是否充分，若不充分则生成缺失分析并引导系统迭代搜索。在 FramesQA 多跳测试中准确率最高提升 34%，从 4 个数据库检索时正确率达 90.1%，速度仅慢 3% 以内。该设计基于前作发现：Gemini 1.5 Pro 判断\"上下文充分性\"准确率达 93%，且\"相关≠够用\"是幻觉关键原因。目前以公开预览在 Gemini Enterprise Agent Platform 开放。", "image": null, "sourceName": "X：小互 (@xiaohu)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T07:33:25.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-7cc27b94", "url": "https://x.com/alibaba_cloud/status/2063881682810179623", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063881682810179623", "title": "Qwen Cloud 全球 AI 黑客马拉松启动，总奖金超 $70K", "titleEn": "Ready to build next-gen AI agents and win $70，000+ in total prizes？ 🚀 Qwen Cloud Global AI Hackath…", "description": "准备好打造下一代 AI 智能体并赢取总计超过 70，000 美元的奖品了吗？🚀\n\nQwen Cloud 全球 AI 黑客马拉松现已启动！利用前沿模型，获得全球曝光，实现你的愿景。查看下方海报了解里程碑。\n\n🔗 立即注册：https：//click.qwencloud.com/m/20000000281/", "descriptionEn": "准备好打造下一代 AI 智能体并赢取总计超过 70，000 美元的奖品了吗？🚀\n\nQwen Cloud 全球 AI 黑客马拉松现已启动！利用前沿模型，获得全球曝光，实现你的愿景。查看下方海报了解里程碑。\n\n🔗 立即注册：https：//click.qwencloud.com/m/20000000281/", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "industry", "publishedAt": "2026-06-08T07:11:52.000Z", "discoveredAt": "2026-06-08T12:37:39.578Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-cfe44c3e", "url": "https://blogs.nvidia.com/blog/uk-sovereign-ai-advancements", "canonicalUrl": "https://blogs.nvidia.com/blog/uk-sovereign-ai-advancements", "title": "英国借助 NVIDIA 技术将主权 AI 雄心转化为行动", "titleEn": "How the UK Is Turning Sovereign AI Ambition Into Action With NVIDIA Technologies", "description": "一年前英国宣布做\"AI 制造者\"，如今主权 AI 计划取得实质进展：AI 云提供商数量翻倍，Nebius 部署三套 NVIDIA AI 基础设施，预计 2027 年满负荷达 65 兆瓦；基于 5，400 个 NVIDIA GH200 的 Isambard-AI 超级计算机全由零碳电力驱动。主权 AI 基金资助四家 NVIDIA Inception 初创公司：Cosine 训练多模态 MoE 编码模型；Cursive 使用 Megatron-LM 开发自改进 AI 系统；Doubleword 基于 Dynamo 和 Nemotron 3 Super 120B 实现 70 倍冷启动加速、4 倍 KV 缓存压缩，推理成本降低 90-95%；Prima Mente 借助 Blackwell GPU 以近 3 倍训练加速开发生物基础模型。NVIDIA 对英国初创生态投资 20 亿英镑。", "descriptionEn": "一年前英国宣布做\"AI 制造者\"，如今主权 AI 计划取得实质进展：AI 云提供商数量翻倍，Nebius 部署三套 NVIDIA AI 基础设施，预计 2027 年满负荷达 65 兆瓦；基于 5，400 个 NVIDIA GH200 的 Isambard-AI 超级计算机全由零碳电力驱动。主权 AI 基金资助四家 NVIDIA Inception 初创公司：Cosine 训练多模态 MoE 编码模型；Cursive 使用 Megatron-LM 开发自改进 AI 系统；Doubleword 基于 Dynamo 和 Nemotron 3 Super 120B 实现 70 倍冷启动加速、4 倍 KV 缓存压缩，推理成本降低 90-95%；Prima Mente 借助 Blackwell GPU 以近 3 倍训练加速开发生物基础模型。NVIDIA 对英国初创生态投资 20 亿英镑。", "image": null, "sourceName": "NVIDIA AI Blog", "sourceCategory": "industry", "publishedAt": "2026-06-08T06:00:57.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-badf9eb7", "url": "https://x.com/TencentHunyuan/status/2063862263434613237", "canonicalUrl": "https://x.com/TencentHunyuan/status/2063862263434613237", "title": "腾讯混元联合多家机构发布首个音频编辑基准MMAE", "titleEn": "Can AI truly edit audio， not just generate it？ 🎧 Tencent Hy， in collaboration with SJTU， SII， NTU，…", "description": "腾讯混元联合上海交大、南洋理工等机构推出MMAE（Massive Multitask Audio Editing Benchmark），这是首个全面评估AI语音/音频编辑能力的基准。MMAE要求模型理解现有音频并按自然语言指令精确修改，而非简单生成。当前模型在该基准上的精确匹配率（EMR）低于5%，暴露了可靠音频编辑的短板。MMAE包含2000个真实场景高保真样本、17741条细粒度评估项，覆盖声音/音乐/语音及混合共7种模态、6种任务复杂度（基础修改到多跳推理及多轮编辑）、8种操作类型（局部到全局）。论文、代码、数据集和演示已公开。", "descriptionEn": "腾讯混元联合上海交大、南洋理工等机构推出MMAE（Massive Multitask Audio Editing Benchmark），这是首个全面评估AI语音/音频编辑能力的基准。MMAE要求模型理解现有音频并按自然语言指令精确修改，而非简单生成。当前模型在该基准上的精确匹配率（EMR）低于5%，暴露了可靠音频编辑的短板。MMAE包含2000个真实场景高保真样本、17741条细粒度评估项，覆盖声音/音乐/语音及混合共7种模态、6种任务复杂度（基础修改到多跳推理及多轮编辑）、8种操作类型（局部到全局）。论文、代码、数据集和演示已公开。", "image": null, "sourceName": "X：腾讯混元 (@TencentHunyuan)", "sourceCategory": "paper", "publishedAt": "2026-06-08T05:54:42.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-insights" ], "relevanceScore": 0.5, "seelesAngle": "Recommended insight: adjacent AI ecosystem signal that may still be useful context for Seeles agent and skill strategy." }, { "id": "news-7e1deae1", "url": "https://www.ithome.com/0/961/378.htm", "canonicalUrl": "https://www.ithome.com/0/961/378.htm", "title": "全球首个：高德发布3D原生城市世界模型ABot-Earth0.5", "titleEn": "全球首个：高德发布 3D 原生城市世界模型 ABot-Earth0.5，单图 10 分钟重建 3D 城市", "description": "阿里巴巴旗下高德发布全球首个3D原生城市世界模型ABot-Earth0.5，已建成覆盖190多个国家和地区的3D地图。用户输入卫星图或文字描述，10分钟即可在消费级GPU上生成公里级3D城市，输出可编辑3DGS格式，可直接导入Unity等引擎。制图成本为传统百分之一，效率提升约千倍，可为具身智能、低空经济、应急救援等提供支撑。目前已开放内测，可前往abot-earth.amap.com提交申请。", "descriptionEn": "阿里巴巴旗下高德发布全球首个3D原生城市世界模型ABot-Earth0.5，已建成覆盖190多个国家和地区的3D地图。用户输入卫星图或文字描述，10分钟即可在消费级GPU上生成公里级3D城市，输出可编辑3DGS格式，可直接导入Unity等引擎。制图成本为传统百分之一，效率提升约千倍，可为具身智能、低空经济、应急救援等提供支撑。目前已开放内测，可前往abot-earth.amap.com提交申请。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "ai-models", "publishedAt": "2026-06-08T05:53:46.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-products" ], "relevanceScore": 0.5, "seelesAngle": "Recommended product signal: AI product or model update worth tracking as broader ecosystem context." }, { "id": "news-68cb86b6", "url": "https://x.com/rohanpaul_ai/status/2063838176884519159", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2063838176884519159", "title": "斯坦福、MIT、哈佛与Anthropic联合论文：为什么更大模型能学到小模型学不会的罕见技能", "titleEn": "Great Stanford + MIT + Harvard + Anthropic paper. Gives a clear training-based reason for why large…", "description": "该论文指出，更大模型能学到罕见技能，是因为训练中遗忘更少，其额外容量保护了弱学习信号。核心机制：常见任务先抢占神经元，罕见任务在出现频率足够形成稳定知识前就被覆盖。小模型可能短暂捕捉到罕见信号，但随即被下一波常见任务更新覆盖。实验使用OLMo语言模型（4M-4B参数）验证：大模型在低频任务上表现更优，保留更多任务特征，且常见任务更新对罕见任务的梯度干扰更小。作者强调，问题不仅在于小模型能否表征任务，更在于训练中罕见任务能否在众多常见任务反复冲击下持续存在。", "descriptionEn": "该论文指出，更大模型能学到罕见技能，是因为训练中遗忘更少，其额外容量保护了弱学习信号。核心机制：常见任务先抢占神经元，罕见任务在出现频率足够形成稳定知识前就被覆盖。小模型可能短暂捕捉到罕见信号，但随即被下一波常见任务更新覆盖。实验使用OLMo语言模型（4M-4B参数）验证：大模型在低频任务上表现更优，保留更多任务特征，且常见任务更新对罕见任务的梯度干扰更小。作者强调，问题不仅在于小模型能否表征任务，更在于训练中罕见任务能否在众多常见任务反复冲击下持续存在。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "paper", "publishedAt": "2026-06-08T04:19:00.000Z", "discoveredAt": "2026-06-08T12:37:39.579Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-dbcc9220", "url": "https://x.com/alibaba_cloud/status/2063833019731923074", "canonicalUrl": "https://x.com/alibaba_cloud/status/2063833019731923074", "title": "阿里云堡垒机推出Qwen驱动Smart Ops Agent", "titleEn": "🚀 Introducing the Smart Ops Agent in Alibaba Cloud Bastionhost！ Transform natural language into se…", "description": "阿里云堡垒机推出Smart Ops Agent，由通义千问（Qwen）驱动，可将自然语言指令转化为安全自动化的云运维操作。核心功能：自然语言运维，将30分钟巡检缩短至1分钟；AI生成脚本，堡垒机全程审计确保安全合规；零配置原生支持ECS、IDC及混合云；自动生成智能报告，无需手动分析日志。该Agent旨在提升云运维效率与安全性。", "descriptionEn": "阿里云堡垒机推出Smart Ops Agent，由通义千问（Qwen）驱动，可将自然语言指令转化为安全自动化的云运维操作。核心功能：自然语言运维，将30分钟巡检缩短至1分钟；AI生成脚本，堡垒机全程审计确保安全合规；零配置原生支持ECS、IDC及混合云；自动生成智能报告，无需手动分析日志。该Agent旨在提升云运维效率与安全性。", "image": null, "sourceName": "X：阿里云 / Alibaba Cloud (@alibaba_cloud)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T03:58:30.000Z", "discoveredAt": "2026-06-08T12:37:39.579Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety", "automation" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-a5886bb", "url": "https://x.com/rohanpaul_ai/status/2063825845605499335", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2063825845605499335", "title": "AutoLab 基准揭示：AI 智能体成功关键在于持续测试而非初始方案", "titleEn": "Strong AI agents still struggle with long research work because they often fail to keep testing and …", "description": "斯坦福、MIT、英伟达、谷歌等顶级实验室联合提出新基准 AutoLab，包含 36 个任务。每个任务中，智能体从可工作的弱代码起步，需在固定时间内迭代优化。任务涵盖系统加速、谜题、模型开发和 CUDA 内核。17 个前沿模型测试结果显示，成功的关键不是初版方案有多好，而是能否持续测试、频繁实验并利用实证反馈。Claude Opus 4.6 领跑基准，靠的是坚持迭代而非初始判断力，而其他前沿模型要么提前放弃，要么思考过久导致超时。", "descriptionEn": "斯坦福、MIT、英伟达、谷歌等顶级实验室联合提出新基准 AutoLab，包含 36 个任务。每个任务中，智能体从可工作的弱代码起步，需在固定时间内迭代优化。任务涵盖系统加速、谜题、模型开发和 CUDA 内核。17 个前沿模型测试结果显示，成功的关键不是初版方案有多好，而是能否持续测试、频繁实验并利用实证反馈。Claude Opus 4.6 领跑基准，靠的是坚持迭代而非初始判断力，而其他前沿模型要么提前放弃，要么思考过久导致超时。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "paper", "publishedAt": "2026-06-08T03:30:00.000Z", "discoveredAt": "2026-06-08T12:37:39.579Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-f2a32795", "url": "https://blogs.nvidia.com/blog/nvidia-and-lg-group-ai-factory", "canonicalUrl": "https://blogs.nvidia.com/blog/nvidia-and-lg-group-ai-factory", "title": "NVIDIA与LG集团合作建设AI工厂，加速物理AI与自动驾驶", "titleEn": "NVIDIA and LG Group Build an AI Factory to Advance Physical AI， Mobility and AI Infrastructure", "description": "NVIDIA与LG集团合作建设AI工厂，为LG的机器人、自动驾驶、数据中心和GPU云服务提供加速计算基础设施。双方整合NVIDIA AI工厂平台与LG的消费电子、机器人技术，连接AI模型开发、物理AI数据生成、机器人仿真训练、边缘部署和工厂数字孪生。LG电子将利用NVIDIA Isaac Sim和Isaac Lab仿真训练CLOiD家用机器人，探索采用Isaac GR00T模型，并开发物理AI数据工厂。LG Innotek提供针对NVIDIA GPU优化的机器人传感组件，LG CNS集成NVIDIA机器人技术。双方还将基于NVIDIA DSX平台建设下一代AI工厂，涵盖冷却、模块化设计和800V直流能源方案。LG电子将基于NVIDIA DRIVE平台开发ADAS和车载AI系统。", "descriptionEn": "NVIDIA与LG集团合作建设AI工厂，为LG的机器人、自动驾驶、数据中心和GPU云服务提供加速计算基础设施。双方整合NVIDIA AI工厂平台与LG的消费电子、机器人技术，连接AI模型开发、物理AI数据生成、机器人仿真训练、边缘部署和工厂数字孪生。LG电子将利用NVIDIA Isaac Sim和Isaac Lab仿真训练CLOiD家用机器人，探索采用Isaac GR00T模型，并开发物理AI数据工厂。LG Innotek提供针对NVIDIA GPU优化的机器人传感组件，LG CNS集成NVIDIA机器人技术。双方还将基于NVIDIA DSX平台建设下一代AI工厂，涵盖冷却、模块化设计和800V直流能源方案。LG电子将基于NVIDIA DRIVE平台开发ADAS和车载AI系统。", "image": null, "sourceName": "NVIDIA AI Blog", "sourceCategory": "industry", "publishedAt": "2026-06-08T03:00:50.000Z", "discoveredAt": "2026-06-08T12:37:38.968Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-fde92b20", "url": "https://x.com/emollick/status/2063814337353891919", "canonicalUrl": "https://x.com/emollick/status/2063814337353891919", "title": "一年前最接近AI智能体的模型：o3", "titleEn": "A year ago the closest thing we had to an AI agent was o3.", "description": "一年前，我们最接近AI智能体的是o3。", "descriptionEn": "一年前，我们最接近AI智能体的是o3。", "image": null, "sourceName": "X：Ethan Mollick (@emollick)", "sourceCategory": "tip", "publishedAt": "2026-06-08T02:44:16.000Z", "discoveredAt": "2026-06-08T12:37:39.579Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-4758907e", "url": "https://www.ithome.com/0/961/265.htm", "canonicalUrl": "https://www.ithome.com/0/961/265.htm", "title": "Palantir CEO 卡普抨击\"词元刷满\"风潮：滥用 AI 如同沉迷色情", "titleEn": "Palantir CEO 卡普抨击\"词元刷满\"风潮：滥用 AI 如同沉迷色情", "description": "Palantir CEO 亚历克斯·卡普在人工智能平台十周年大会场外接受采访，将行业无节制消耗 token 的行为比作\"精神沉溺式滥用\"和\"自慰成瘾\"。CTO 沙亚姆·桑卡在财报电话会议上表示，词元用量越多产出越粗劣，企业需依靠 Palantir AI 平台（AIP）才能规避低价值陷阱。Uber COO 安德鲁·麦克唐纳坦言看不出攀升的 AI 成本与效率提升之间存在关联。卡普认为，AI 可完成\"撰写 GDP 增长报告\"等简单任务，但优化油气开采、重塑供应链等复杂业务需依靠严谨的流程，大语言模型无法取代。", "descriptionEn": "Palantir CEO 亚历克斯·卡普在人工智能平台十周年大会场外接受采访，将行业无节制消耗 token 的行为比作\"精神沉溺式滥用\"和\"自慰成瘾\"。CTO 沙亚姆·桑卡在财报电话会议上表示，词元用量越多产出越粗劣，企业需依靠 Palantir AI 平台（AIP）才能规避低价值陷阱。Uber COO 安德鲁·麦克唐纳坦言看不出攀升的 AI 成本与效率提升之间存在关联。卡普认为，AI 可完成\"撰写 GDP 增长报告\"等简单任务，但优化油气开采、重塑供应链等复杂业务需依靠严谨的流程，大语言模型无法取代。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-08T02:42:45.000Z", "discoveredAt": "2026-06-08T12:37:39.579Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-2815101f", "url": "https://x.com/shao__meng/status/2063809042959810996", "canonicalUrl": "https://x.com/shao__meng/status/2063809042959810996", "title": "让 Claude Opus 长时自主运行的五条实战建议", "titleEn": "让 Claude Opus 长时自主运行的 5 条核心实战建议 来自 Claude Code 作者 Boris Cherny，咱们尽量体现独立于模型的建议，让它放到 Codex + GPT-5.5 …", "description": "Boris Cherny 给出五条让 Claude Opus 自主运行数小时/天的建议：1. 开启 Auto Mode 减少审批；2. 用 Dynamic Workflows 编排数百至数千子 Agent；3. 使用 /goal 或 /loop 指令持续推进；4. 优先用云端 Claude Code，可关闭笔记本；5. 确保端到端自验证能力。邵猛补充：自动权限是前提；/goal、/loop 适合高难度任务但 token 消耗高；需解决本地关机/休眠；端到端验证最重要，否则 token 可能白费。", "descriptionEn": "Boris Cherny 给出五条让 Claude Opus 自主运行数小时/天的建议：1. 开启 Auto Mode 减少审批；2. 用 Dynamic Workflows 编排数百至数千子 Agent；3. 使用 /goal 或 /loop 指令持续推进；4. 优先用云端 Claude Code，可关闭笔记本；5. 确保端到端自验证能力。邵猛补充：自动权限是前提；/goal、/loop 适合高难度任务但 token 消耗高；需解决本地关机/休眠；端到端验证最重要，否则 token 可能白费。", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "tip", "publishedAt": "2026-06-08T02:23:14.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-6604c55b", "url": "https://x.com/xiaohu/status/2063801609042427972", "canonicalUrl": "https://x.com/xiaohu/status/2063801609042427972", "title": "Claude 写 Anthropic 超 80% 代码", "titleEn": "Claude 现在写了 Anthropic 大部分代码 截至 2026 年 5 月，合并进 Anthropic 代码库的代码里，超过 80% 是 Claude 写的。 而在 2025 年 2 月 …", "description": "Claude 现在写了 Anthropic 大部分代码\n\n截至 2026 年 5 月，合并进 Anthropic 代码库的代码里，超过 80% 是 Claude 写的。\n\n而在 2025 年 2 月 Claude Code 发布前，这个比例还只有个位数…", "descriptionEn": "Claude 现在写了 Anthropic 大部分代码\n\n截至 2026 年 5 月，合并进 Anthropic 代码库的代码里，超过 80% 是 Claude 写的。\n\n而在 2025 年 2 月 Claude Code 发布前，这个比例还只有个位数…", "image": null, "sourceName": "X：小互 (@xiaohu)", "sourceCategory": "industry", "publishedAt": "2026-06-08T01:53:41.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-fd85b09c", "url": "https://x.com/dotey/status/2063800780050608603", "canonicalUrl": "https://x.com/dotey/status/2063800780050608603", "title": "长时间运行 Agent，Agent 能自行验证才是关键，否则可能只是浪费 Token", "titleEn": "长时间运行 Agent，Agent 能自行验证才是关键，否则可能只是浪费 Token", "description": "宝玉指出，Agent 能否自我验证是长时间运行的关键，否则可能浪费 Token。@bcherny 的基准测试显示 Claude Opus 最适合长时间运行，并给出 5 条自主运行技巧：1. 使用自动权限模式；2. 部署动态工作流让 Claude 协调数百/数千个 Agent；3. 用 /goal 或 /loop 指令持续推进；4. 在云端运行 Claude Code 以便关闭笔记本；5. 确保端到端自我验证--通过 Chrome 浏览器扩展验证网页、iOS/Android 模拟器 MCP 验证移动端、启动完整 Web 服务验证后端。", "descriptionEn": "宝玉指出，Agent 能否自我验证是长时间运行的关键，否则可能浪费 Token。@bcherny 的基准测试显示 Claude Opus 最适合长时间运行，并给出 5 条自主运行技巧：1. 使用自动权限模式；2. 部署动态工作流让 Claude 协调数百/数千个 Agent；3. 用 /goal 或 /loop 指令持续推进；4. 在云端运行 Claude Code 以便关闭笔记本；5. 确保端到端自我验证--通过 Chrome 浏览器扩展验证网页、iOS/Android 模拟器 MCP 验证移动端、启动完整 Web 服务验证后端。", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "tip", "publishedAt": "2026-06-08T01:50:24.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "skill-ecosystem", "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-57c89afd", "url": "https://x.com/jxnlco/status/2063799488456610086", "canonicalUrl": "https://x.com/jxnlco/status/2063799488456610086", "title": "cowork与codex功能期望互问", "titleEn": "What are some features in cowork that you wish exited in codex？ And vice versa？", "description": "cowork 有哪些你希望 codex 也有的功能？反之亦然？", "descriptionEn": "cowork 有哪些你希望 codex 也有的功能？反之亦然？", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "tip", "publishedAt": "2026-06-08T01:45:16.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-9321d00b", "url": "https://openai.com/index/built-to-benefit-everyone-our-plan", "canonicalUrl": "https://openai.com/index/built-to-benefit-everyone-our-plan", "title": "OpenAI 公布让 AGI 造福所有人的计划", "titleEn": "Built to benefit everyone： our plan", "description": "OpenAI 发布计划，阐述让 AGI 造福所有人的愿景。该计划聚焦于 AI 的可及性、安全性和共享繁荣，确保技术进步惠及每个人。", "descriptionEn": "OpenAI 发布计划，阐述让 AGI 造福所有人的愿景。该计划聚焦于 AI 的可及性、安全性和共享繁荣，确保技术进步惠及每个人。", "image": null, "sourceName": "OpenAI：官网动态（RSS · 排除企业/客户案例）", "sourceCategory": "tip", "publishedAt": "2026-06-08T01:30:00.000Z", "discoveredAt": "2026-06-09T00:39:05.022Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-4cbcd3a0", "url": "https://x.com/bcherny/status/2063792263067754658", "canonicalUrl": "https://x.com/bcherny/status/2063792263067754658", "title": "Claude Opus 长时间运行工作最佳模型及自主运行技巧", "titleEn": "Seeing a number of benchmarks showing Opus is the best model for long-running work. Five tips for r…", "description": "多项基准显示 Claude Opus 是长时间运行工作的最佳模型。SWE-Marathon 基准评估编码智能体在 10 亿 token 预算下自主完成长期软件任务（如重写 JAX 代码为 PyTorch、用 Rust 构建 C 编译器）。Opus 在此类任务上领先。Boris Cherny 给出 5 个技巧：使用自动权限模式避免审批；用动态工作流协调数百/数千个智能体；用 /goal 或 /loop 推动持续执行；在云端使用 Claude Code（桌面/移动端）以便关闭笔记本；确保 Claude 能端到端自验证--Chrome 扩展验证网页、iOS/Android 模拟 MCP、启动完整后端服务。", "descriptionEn": "多项基准显示 Claude Opus 是长时间运行工作的最佳模型。SWE-Marathon 基准评估编码智能体在 10 亿 token 预算下自主完成长期软件任务（如重写 JAX 代码为 PyTorch、用 Rust 构建 C 编译器）。Opus 在此类任务上领先。Boris Cherny 给出 5 个技巧：使用自动权限模式避免审批；用动态工作流协调数百/数千个智能体；用 /goal 或 /loop 推动持续执行；在云端使用 Claude Code（桌面/移动端）以便关闭笔记本；确保 Claude 能端到端自验证--Chrome 扩展验证网页、iOS/Android 模拟 MCP、启动完整后端服务。", "image": null, "sourceName": "X：Boris Cherny (@bcherny)", "sourceCategory": "tip", "publishedAt": "2026-06-08T01:16:33.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "skill-ecosystem", "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-c9cf53f6", "url": "https://x.com/shao__meng/status/2063790077021753383", "canonicalUrl": "https://x.com/shao__meng/status/2063790077021753383", "title": "AGENTS.md 在 Coding Agents 中真的有用吗？", "titleEn": "AGENTS.md 在 Coding Agents 中真的有用吗？ 这篇论文，大规模实证研究仓库级上下文文件（AGENTS.md、CLAUDE.md 等）对编码 Agent 实际效果的影响，可能有些…", "description": "论文大规模实证检验 AGENTS.md 等仓库级上下文文件对编码 Agent 的影响。在 SWE-bench Lite（300 任务）和新建 AGENTBENCH（138 任务）上测试 Claude Code、Codex、Qwen Code 等组合。核心发现：LLM 自动生成的 context file 在 8 组设置中 5 组成功率下降，平均 -0.5%（SWE-bench）/-2%（AGENTBENCH），成本增加 +20%+；开发者手写仅平均 +4%。冗余假说：移除其他文档后，自动生成反而 +2.7%。建议避免自动生成，精简测试/lint 命令，优先写入仓库专用工具。", "descriptionEn": "论文大规模实证检验 AGENTS.md 等仓库级上下文文件对编码 Agent 的影响。在 SWE-bench Lite（300 任务）和新建 AGENTBENCH（138 任务）上测试 Claude Code、Codex、Qwen Code 等组合。核心发现：LLM 自动生成的 context file 在 8 组设置中 5 组成功率下降，平均 -0.5%（SWE-bench）/-2%（AGENTBENCH），成本增加 +20%+；开发者手写仅平均 +4%。冗余假说：移除其他文档后，自动生成反而 +2.7%。建议避免自动生成，精简测试/lint 命令，优先写入仓库专用工具。", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "paper", "publishedAt": "2026-06-08T01:07:52.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-29dd3188", "url": "https://x.com/shao__meng/status/2063786460172677407", "canonicalUrl": "https://x.com/shao__meng/status/2063786460172677407", "title": "Warp调查：Codex App成最热门Coding Agent", "titleEn": "你现在在用哪个 Coding Agent？ Warp 团队 @BHolmesDev 发出的调查得到了 2095 个投票，其中 Codex App 占到 51.1%、Warp CLI 占到 30.9%…", "description": "你现在在用哪个 Coding Agent？\n\nWarp团队 @BHolmesDev 进行的调查得到了 2095 个投票，其中 Codex App 占到 51.1%、Warp CLI 占到 30.9%、Claude App 7.4%\n\n没错，Claude App 只有 7.4%，因为投票中没有包含 Claude Code 😄", "descriptionEn": "你现在在用哪个 Coding Agent？\n\nWarp团队 @BHolmesDev 进行的调查得到了 2095 个投票，其中 Codex App 占到 51.1%、Warp CLI 占到 30.9%、Claude App 7.4%\n\n没错，Claude App 只有 7.4%，因为投票中没有包含 Claude Code 😄", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "tip", "publishedAt": "2026-06-08T00:53:30.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-d5b7f087", "url": "https://x.com/shao__meng/status/2063785694070280379", "canonicalUrl": "https://x.com/shao__meng/status/2063785694070280379", "title": "不写 Prompt，写 Loops - Boris Cherny 谈 AI 编程范式转变", "titleEn": "不写 Prompt，写 Loops -- Boris Cherny （Claude Code） 道理大家都懂，伴随 Claude Opus 4.8 和 GPT-5.5 这种高智能、高推理和编程能力的…", "description": "Anthropic Claude Code 负责人 Boris Cherny 表示，他不再手动写提示词，而是编写 Loops 让 AI 自主决策和解决问题。伴随 Claude Opus 4.8 和 GPT-5.5 等高智能模型出现，手把手写 Prompt 的模式将被取代。但 Loops 方案消耗 Token 过快，一个 Goals 可能耗尽 5 小时用量，企业仍需严格审查 Token 消耗的 ROI，因此实践中会通过前置的 Spec、AGENTS.md 等约束让 AI 做确定性可控的部分。Boris 预测这一转变将在今年剩余时间持续。", "descriptionEn": "Anthropic Claude Code 负责人 Boris Cherny 表示，他不再手动写提示词，而是编写 Loops 让 AI 自主决策和解决问题。伴随 Claude Opus 4.8 和 GPT-5.5 等高智能模型出现，手把手写 Prompt 的模式将被取代。但 Loops 方案消耗 Token 过快，一个 Goals 可能耗尽 5 小时用量，企业仍需严格审查 Token 消耗的 ROI，因此实践中会通过前置的 Spec、AGENTS.md 等约束让 AI 做确定性可控的部分。Boris 预测这一转变将在今年剩余时间持续。", "image": null, "sourceName": "X：邵猛 (@shao__meng)", "sourceCategory": "tip", "publishedAt": "2026-06-08T00:50:27.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-7d729814", "url": "https://x.com/dotey/status/2063779728360698168", "canonicalUrl": "https://x.com/dotey/status/2063779728360698168", "title": "谈\"选HTML不选React\"--宝玉的反对观点", "titleEn": "> 为什么选 HTML，而不是 React 选 HTML 而不是 react 这个观点我是不认同的。 第一个原因是设计稿它是需要关联数据有交互的，能根据数据去动态地变更 UI，这点 HTML 是做…", "description": "宝玉不认同\"选HTML而非React\"的观点。首先，设计稿需要动态数据交互，HTML难以胜任，而Claude Design是UX交互工具。其次，复杂UI必须拆分为小组件，React能实现重用与低上下文修改，大型HTML则维护困难。再者，React天然形成结构化树形设计，方便coding agent开发，HTML结构割裂、难以与设计映射。至于文章说模型注意力需放在React运行准确率上，宝玉认为当前大模型能力已使此问题不复存在。", "descriptionEn": "宝玉不认同\"选HTML而非React\"的观点。首先，设计稿需要动态数据交互，HTML难以胜任，而Claude Design是UX交互工具。其次，复杂UI必须拆分为小组件，React能实现重用与低上下文修改，大型HTML则维护困难。再者，React天然形成结构化树形设计，方便coding agent开发，HTML结构割裂、难以与设计映射。至于文章说模型注意力需放在React运行准确率上，宝玉认为当前大模型能力已使此问题不复存在。", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "tip", "publishedAt": "2026-06-08T00:26:45.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-87bcbe53", "url": "https://x.com/sama/status/2063779477419901071", "canonicalUrl": "https://x.com/sama/status/2063779477419901071", "title": "Codex 百天挑战：10 倍使用限额", "titleEn": "interesting recursive loop here maybe", "description": "Codex 推出了一个新的\"大按钮\"。在接下来的 100 天里，每天会选出一位用 Codex 做出了令人印象深刻或极其有用工作的人，给他们一个月的 10 倍使用限额，看看他们能做出什么。明天开始第一个。Sam Altman 评论称\"这里可能有个有趣的递归循环\"。", "descriptionEn": "Codex 推出了一个新的\"大按钮\"。在接下来的 100 天里，每天会选出一位用 Codex 做出了令人印象深刻或极其有用工作的人，给他们一个月的 10 倍使用限额，看看他们能做出什么。明天开始第一个。Sam Altman 评论称\"这里可能有个有趣的递归循环\"。", "image": null, "sourceName": "X：Sam Altman (@sama)", "sourceCategory": "tip", "publishedAt": "2026-06-08T00:25:45.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-3f6e6b38", "url": "https://x.com/oran_ge/status/2063775927147073628", "canonicalUrl": "https://x.com/oran_ge/status/2063775927147073628", "title": "Cola 1.0.0 公测：100天诞生的有灵魂AI", "titleEn": "http：//x.com/i/article/2063773475249246208", "description": "诞生100天后，Cola AI正式公测（1.0.0版本），无需邀请码即可在Mac和Windows上下载使用。新用户注册24小时内可获5美金额度。Cola被团队定义为\"有灵魂的AI\"，具备连续记忆、语音交互和自我进化能力，强调像人类一样长期记忆和情感共情。官网：colaos.ai。", "descriptionEn": "诞生100天后，Cola AI正式公测（1.0.0版本），无需邀请码即可在Mac和Windows上下载使用。新用户注册24小时内可获5美金额度。Cola被团队定义为\"有灵魂的AI\"，具备连续记忆、语音交互和自我进化能力，强调像人类一样长期记忆和情感共情。官网：colaos.ai。", "image": null, "sourceName": "X：Oran Ge (@oran_ge)", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T00:11:38.000Z", "discoveredAt": "2026-06-08T00:20:22.641Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-3ec23de2", "url": "https://www.ithome.com/0/961/214.htm", "canonicalUrl": "https://www.ithome.com/0/961/214.htm", "title": "警惕\"AI 中转站\"数据安全风险，国家安全部发布安全提示", "titleEn": "警惕\"AI 中转站\"数据安全风险，国家安全部发布安全提示", "description": "国家安全部6月8日发布安全提示，指出当前批量提供海内外大模型访问服务的\"AI 中转站\"存在运营资质缺失、安全防护薄弱等问题，用户隐私泄露与数据倒卖时有发生。部分中转站用低配模型冒充高端模型导致输出失真，暗藏后门植入恶意代码，还擅自将用户数据传至境外服务器，威胁个人隐私、商业机密与国家秘密。建议用户选用正规平台，对敏感数据脱敏处理，及时处置异常并可通过12339举报可疑线索。", "descriptionEn": "国家安全部6月8日发布安全提示，指出当前批量提供海内外大模型访问服务的\"AI 中转站\"存在运营资质缺失、安全防护薄弱等问题，用户隐私泄露与数据倒卖时有发生。部分中转站用低配模型冒充高端模型导致输出失真，暗藏后门植入恶意代码，还擅自将用户数据传至境外服务器，威胁个人隐私、商业机密与国家秘密。建议用户选用正规平台，对敏感数据脱敏处理，及时处置异常并可通过12339举报可疑线索。", "image": null, "sourceName": "IT之家（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-08T00:02:01.000Z", "discoveredAt": "2026-06-08T12:37:39.580Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-6aceca7f", "url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models", "canonicalUrl": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models", "title": "苹果发布第三代 Apple Foundation Models（AFM）", "titleEn": "Introducing the Third Generation of Apple's Foundation Models", "description": "苹果推出第三代 Apple Foundation Models（AFM）基础模型家族，与 Google 合作定制，包含五个模型，覆盖从设备端到基于 Private Cloud Compute 的服务器端模型。这些模型旨在驱动 Apple Intelligence 功能，包括全新 Siri 和智能工具，以用户为中心深度融合操作系统，隐私为核心设计原则。", "descriptionEn": "苹果推出第三代 Apple Foundation Models（AFM）基础模型家族，与 Google 合作定制，包含五个模型，覆盖从设备端到基于 Private Cloud Compute 的服务器端模型。这些模型旨在驱动 Apple Intelligence 功能，包括全新 Siri 和智能工具，以用户为中心深度融合操作系统，隐私为核心设计原则。", "image": null, "sourceName": "Apple Machine Learning Research（RSS）", "sourceCategory": "ai-models", "publishedAt": "2026-06-08T00:00:00.000Z", "discoveredAt": "2026-06-09T00:39:05.022Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.78, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-a2749987", "url": "https://huggingface.co/blog/openenv-agentic-rl", "canonicalUrl": "https://huggingface.co/blog/openenv-agentic-rl", "title": "开源社区支持 OpenEnv 用于智能体强化学习", "titleEn": "The Open Source Community is backing OpenEnv for Agentic RL", "description": "今日，Hugging Face 宣布 OpenEnv 项目进一步开放，由 Meta-PyTorch、Reflection、Unsloth 等组成的委员会协调，并获得 PyTorch Foundation、vLLM、SkyRL（UCB）等机构支持。OpenEnv 是创建终端、浏览器等智能体执行环境的工具，定位为训练器与环境间的互操作层，标准化环境的发布、部署和消费，但不定义奖励或训练循环。它基于客户端/服务器架构提供 Gymnasium 风格 API（`reset（）`、`step（）`、`state（）`），支持 HTTP、WebSocket、Docker 打包，并将 MCP 作为一等公民，使环境在模拟和生产模式下行为一致。未来将聚焦任务集对接数据集、外部奖励、Harness 集成及自动验证，以推动开源智能体 RL 基础建设。", "descriptionEn": "今日，Hugging Face 宣布 OpenEnv 项目进一步开放，由 Meta-PyTorch、Reflection、Unsloth 等组成的委员会协调，并获得 PyTorch Foundation、vLLM、SkyRL（UCB）等机构支持。OpenEnv 是创建终端、浏览器等智能体执行环境的工具，定位为训练器与环境间的互操作层，标准化环境的发布、部署和消费，但不定义奖励或训练循环。它基于客户端/服务器架构提供 Gymnasium 风格 API（`reset（）`、`step（）`、`state（）`），支持 HTTP、WebSocket、Docker 打包，并将 MCP 作为一等公民，使环境在模拟和生产模式下行为一致。未来将聚焦任务集对接数据集、外部奖励、Harness 集成及自动验证，以推动开源智能体 RL 基础建设。", "image": null, "sourceName": "Hugging Face：Blog（RSS）", "sourceCategory": "ai-products", "publishedAt": "2026-06-08T00:00:00.000Z", "discoveredAt": "2026-06-09T00:39:05.022Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "agent-architecture", "skill-ecosystem" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-c063f9e9", "url": "https://simonwillison.net/2026/Jun/7/datasette-agent-edit", "canonicalUrl": "https://simonwillison.net/2026/Jun/7/datasette-agent-edit", "title": "datasette-agent-edit 0.1a0 发布", "titleEn": "datasette-agent-edit 0.1a0", "description": "datasette-agent-edit 0.1a0 插件发布，为 Datasette Agent 提供智能体编辑文本的核心工具。其设计借鉴 Claude text editor，实现三个工具：view（按行号查看文件片段）、str_replace（精确替换唯一字符串）、insert（在指定行号后插入文本）。该插件作为基础组件，可供其他协作 Markdown 编辑、SQL 查询更新、SVG 编辑等插件复用。", "descriptionEn": "datasette-agent-edit 0.1a0 插件发布，为 Datasette Agent 提供智能体编辑文本的核心工具。其设计借鉴 Claude text editor，实现三个工具：view（按行号查看文件片段）、str_replace（精确替换唯一字符串）、insert（在指定行号后插入文本）。该插件作为基础组件，可供其他协作 Markdown 编辑、SQL 查询更新、SVG 编辑等插件复用。", "image": null, "sourceName": "Simon Willison 博客", "sourceCategory": "tip", "publishedAt": "2026-06-07T23:56:38.000Z", "discoveredAt": "2026-06-08T00:20:22.641Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "skill-ecosystem" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-bf86ae65", "url": "https://x.com/berryxia/status/2063764781346312500", "canonicalUrl": "https://x.com/berryxia/status/2063764781346312500", "title": "OpenAI 发布 Codex 应用场景新页面，列出七大领域委托任务", "titleEn": "OpenAI刚刚发布了一个新的关于Codex应用场景的页面。 该页面基本上列出了各团队目前正在让编码代理执行的各项任务： 工程开发、产品开发、质量测试、安全检查、数据分析、内部工具的开发，甚至生命…", "description": "OpenAI 发布 Codex 应用场景新页面，展示各团队已委托编码代理执行的七大类任务：工程开发、产品开发、质量测试、安全检查、数据分析、内部工具及生命科学工作流程。具体案例包括审查 GitHub PR 并理解大型代码库、将截图转为响应式 UI、模拟真实用户操作进行 QA 测试、重构老旧代码与数据迁移、修复安全漏洞、撰写 PRD、分析数据集、开发内部应用及辅助生命科学研究。页面表明编码代理正从演示阶段融入日常工作。", "descriptionEn": "OpenAI 发布 Codex 应用场景新页面，展示各团队已委托编码代理执行的七大类任务：工程开发、产品开发、质量测试、安全检查、数据分析、内部工具及生命科学工作流程。具体案例包括审查 GitHub PR 并理解大型代码库、将截图转为响应式 UI、模拟真实用户操作进行 QA 测试、重构老旧代码与数据迁移、修复安全漏洞、撰写 PRD、分析数据集、开发内部应用及辅助生命科学研究。页面表明编码代理正从演示阶段融入日常工作。", "image": null, "sourceName": "X：Berry Xia (@berryxia)", "sourceCategory": "ai-products", "publishedAt": "2026-06-07T23:27:21.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "coding-agent", "agent-safety" ], "relevanceScore": 0.86, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-c0a45e66", "url": "https://www.bloomberg.com/news/articles/2026-06-07/nvidia-sk-hynix-sign-multi-year-pact-to-develop-next-gen-chips", "canonicalUrl": "https://www.bloomberg.com/news/articles/2026-06-07/nvidia-sk-hynix-sign-multi-year-pact-to-develop-next-gen-chips", "title": "Nvidia 与 SK Hynix 签署多年协议，共同开发下一代 AI 存储芯片", "titleEn": "Nvidia， SK Hynix Sign Multi-Year Pact to Develop Next-Gen Chips", "description": "Nvidia 与 SK Hynix 达成合作，将共同设计面向 AI 应用的未来几代存储芯片。此举巩固了 SK Hynix 在高端半导体领域的领先地位。", "descriptionEn": "Nvidia 与 SK Hynix 达成合作，将共同设计面向 AI 应用的未来几代存储芯片。此举巩固了 SK Hynix 在高端半导体领域的领先地位。", "image": null, "sourceName": "Bloomberg：Technology（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-07T23:26:02.000Z", "discoveredAt": "2026-06-08T00:20:22.070Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-929c909", "url": "https://x.com/hongming731/status/2063762055300300859", "canonicalUrl": "https://x.com/hongming731/status/2063762055300300859", "title": "早报：Tony Fadell谈\"知情直觉\"；OpenAI用Codex零人工写百万行代码；Coding Agent范式转移", "titleEn": "BestBlogs 今日早报内容： -------- BestBlogs 早报 · 06-08 # Tony Fadell / OpenAI Codex / Context Engineerin…", "description": "今日早报聚焦AI产品与工程前沿：iPod之父Tony Fadell提出创新依赖\"知情直觉\"而非数据，并总结\"三代法则\"；OpenAI内部用Codex全程零人工代码交付约100万行代码，合并1500个PR，效率为手写代码10倍；Thoughtworks专家梳理Coding Agent范式转移，指出Context Engineering是双向放大器，并强调Harness Engineering构建安全网；Anthropic首次公开内部Claude Code Skills经验；OpenAI计划将Codex并入ChatGPT打造超级应用。", "descriptionEn": "今日早报聚焦AI产品与工程前沿：iPod之父Tony Fadell提出创新依赖\"知情直觉\"而非数据，并总结\"三代法则\"；OpenAI内部用Codex全程零人工代码交付约100万行代码，合并1500个PR，效率为手写代码10倍；Thoughtworks专家梳理Coding Agent范式转移，指出Context Engineering是双向放大器，并强调Harness Engineering构建安全网；Anthropic首次公开内部Claude Code Skills经验；OpenAI计划将Codex并入ChatGPT打造超级应用。", "image": null, "sourceName": "X：洪明 (@hongming731)", "sourceCategory": "tip", "publishedAt": "2026-06-07T23:16:31.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent", "agent-safety" ], "relevanceScore": 0.98, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-233401ee", "url": "https://x.com/hongming731/status/2063761907891466424", "canonicalUrl": "https://x.com/hongming731/status/2063761907891466424", "title": "BestBlogs早报：iPod之父、Codex、Coding Agent", "titleEn": "http：//x.com/i/article/2063761613795270656", "description": "BestBlogs早报06-08聚焦三篇AI工程实践精讲：iPod之父Tony Fadell系统阐述\"知情直觉\"与\"三代法则\"，警告AI时代的\"认知投降\"和\"快时尚软件\"；OpenAI工程团队分享使用Codex", "descriptionEn": "BestBlogs早报06-08聚焦三篇AI工程实践精讲：iPod之父Tony Fadell系统阐述\"知情直觉\"与\"三代法则\"，警告AI时代的\"认知投降\"和\"快时尚软件\"；OpenAI工程团队分享使用Codex", "image": null, "sourceName": "X：洪明 (@hongming731)", "sourceCategory": "tip", "publishedAt": "2026-06-07T23:15:56.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.94, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-4fef4240", "url": "https://x.com/dotey/status/2063758466573795692", "canonicalUrl": "https://x.com/dotey/status/2063758466573795692", "title": "Claude Code 移动端 remote control 需频繁确认", "titleEn": "Claude Code 虽然移动端可以 remote control，但是 plan 后不能 bypass permission，需要不停的确认，神烦", "description": "Claude Code 虽然移动端可以 remote control，但是 plan 后不能 bypass permission，需要不停的确认，神烦", "descriptionEn": "Claude Code 虽然移动端可以 remote control，但是 plan 后不能 bypass permission，需要不停的确认，神烦", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "tip", "publishedAt": "2026-06-07T23:02:15.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent", "agent-safety" ], "relevanceScore": 0.82, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-3944bc0f", "url": "https://blogs.nvidia.com/blog/nvidia-and-doosan-group-physical-ai", "canonicalUrl": "https://blogs.nvidia.com/blog/nvidia-and-doosan-group-physical-ai", "title": "英伟达与斗山集团合作推进物理AI及AI工厂基础设施", "titleEn": "NVIDIA and Doosan Group Collaborate to Advance Physical AI and AI Factory Infrastructure", "description": "英伟达与斗山集团扩大合作，覆盖斗山机器人、斗山山猫、斗山能源及斗山电子材料四大板块。斗山机器人将集成英伟达Isaac Sim、Isaac Lab、Cosmos世界基础模型、Newton物理引擎及Jetson Thor，推进Agentic Robot OS，探索卸垛、打磨等参考用例及双臂/人形机器人。斗山山猫计划将物理AI技术用于建筑、园林、农业设备，加速开发专用世界模型。斗山能源评估通过燃气轮机、汽轮机、小型模块化反应堆及氢燃料电池为AI工厂供电。斗山电子材料为英伟达MGX生态系统供应高性能铜箔基板（CCL），用于AI服务器PCB。", "descriptionEn": "英伟达与斗山集团扩大合作，覆盖斗山机器人、斗山山猫、斗山能源及斗山电子材料四大板块。斗山机器人将集成英伟达Isaac Sim、Isaac Lab、Cosmos世界基础模型、Newton物理引擎及Jetson Thor，推进Agentic Robot OS，探索卸垛、打磨等参考用例及双臂/人形机器人。斗山山猫计划将物理AI技术用于建筑、园林、农业设备，加速开发专用世界模型。斗山能源评估通过燃气轮机、汽轮机、小型模块化反应堆及氢燃料电池为AI工厂供电。斗山电子材料为英伟达MGX生态系统供应高性能铜箔基板（CCL），用于AI服务器PCB。", "image": null, "sourceName": "NVIDIA AI Blog", "sourceCategory": "industry", "publishedAt": "2026-06-07T23:00:36.000Z", "discoveredAt": "2026-06-08T00:20:22.071Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-f8a7af55", "url": "https://x.com/omarsar0/status/2063750411576844684", "canonicalUrl": "https://x.com/omarsar0/status/2063750411576844684", "title": "超强AI模型即将发布：避免锁定单一供应商", "titleEn": "Super-powerful AI models will launch in the coming weeks. We are looking at a potential step change…", "description": "未来几周将有超强AI模型发布，模型能力可能迎来阶段性提升。DAIR.AI创始人Elvis Saravia认为，当前最大的错误是从成本或工程角度锁定单一供应商，建议开始利用模型组合（包括开源模型），保持\"模型无关\"，以便随时切换模型以发挥其各自优势。对于编程智能体，开源模型表现已媲美前沿模型。他建议将\"AI模型路由\"作为核心策略，高效分配任务/工作给不同模型，这是一项高回报的AI工程投入。", "descriptionEn": "未来几周将有超强AI模型发布，模型能力可能迎来阶段性提升。DAIR.AI创始人Elvis Saravia认为，当前最大的错误是从成本或工程角度锁定单一供应商，建议开始利用模型组合（包括开源模型），保持\"模型无关\"，以便随时切换模型以发挥其各自优势。对于编程智能体，开源模型表现已媲美前沿模型。他建议将\"AI模型路由\"作为核心策略，高效分配任务/工作给不同模型，这是一项高回报的AI工程投入。", "image": null, "sourceName": "X：Elvis Saravia (@omarsar0, DAIR.AI)", "sourceCategory": "tip", "publishedAt": "2026-06-07T22:30:15.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-20309dc7", "url": "https://x.com/jxnlco/status/2063748374797447673", "canonicalUrl": "https://x.com/jxnlco/status/2063748374797447673", "title": "Codex 几乎适用于一切任务", "titleEn": "This but \"codex is for almost everything\"", "description": "但这……\"Codex 几乎适用于一切\"。", "descriptionEn": "但这……\"Codex 几乎适用于一切\"。", "image": null, "sourceName": "X：Jason Liu (@jxnlco)", "sourceCategory": "tip", "publishedAt": "2026-06-07T22:22:09.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-d7ea25af", "url": "https://x.com/thsottiaux/status/2063748242681307611", "canonicalUrl": "https://x.com/thsottiaux/status/2063748242681307611", "title": "Codex新按钮：每日选一人获10倍额度", "titleEn": "I have a new kind of big button that I can press for Codex. Over the next 100 days， we will select o…", "description": "我有了一个新的巨型按钮，可以为 Codex 按下。在接下来的 100 天里，我们将每天挑选一位用 Codex 做出令人印象深刻或极其有用工作的人，给他们一个月的 10 倍使用额度，看看他们能做出什么。明天是第一个。", "descriptionEn": "我有了一个新的巨型按钮，可以为 Codex 按下。在接下来的 100 天里，我们将每天挑选一位用 Codex 做出令人印象深刻或极其有用工作的人，给他们一个月的 10 倍使用额度，看看他们能做出什么。明天是第一个。", "image": null, "sourceName": "X：Tibo (@thsottiaux)", "sourceCategory": "ai-products", "publishedAt": "2026-06-07T22:21:38.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-a9c15383", "url": "https://huggingface.co/blog/build-small-hackathon/amazingdigitaldentures", "canonicalUrl": "https://huggingface.co/blog/build-small-hackathon/amazingdigitaldentures", "title": "Amazing Digital Dentures（一个失败的项目）", "titleEn": "Amazing Digital Dentures （a failed project）", "description": "受《神奇数字马戏团》启发，作者尝试用 Nemotron 30b 构建一个数字宠物，自动生成以 Three.js 实现的冒险游戏，起初作为过度工程化的待办列表，后转向纯冒险生成。长提示、技能卡及 RAG 方案均未产出可运行游戏（常出现空白屏幕）。项目最终转型为简单的 HTML 玩具制作器，能一次生成时钟、待办列表、贪吃蛇、打砖块，但俄罗斯方块等复杂项目仍会崩溃。作者正寻求新方向。", "descriptionEn": "受《神奇数字马戏团》启发，作者尝试用 Nemotron 30b 构建一个数字宠物，自动生成以 Three.js 实现的冒险游戏，起初作为过度工程化的待办列表，后转向纯冒险生成。长提示、技能卡及 RAG 方案均未产出可运行游戏（常出现空白屏幕）。项目最终转型为简单的 HTML 玩具制作器，能一次生成时钟、待办列表、贪吃蛇、打砖块，但俄罗斯方块等复杂项目仍会崩溃。作者正寻求新方向。", "image": null, "sourceName": "Hugging Face：Blog（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-07T19:44:27.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "skill-ecosystem" ], "relevanceScore": 0.78, "seelesAngle": "Skill ecosystem signal: reusable abilities, MCP, plugins, or capability modules that can inform Seeles skills." }, { "id": "news-21319742", "url": "https://x.com/gdb/status/2063705280270021087", "canonicalUrl": "https://x.com/gdb/status/2063705280270021087", "title": "OpenAI 公布 Codex 数十个实际用例，涵盖软件工程到运维", "titleEn": "Codex use-cases： \"From software engineering and design to data analysis and operations， Codex is bec…", "description": "OpenAI 公布了数十个 Codex 实际工作流程，展示团队如何用其自动化任务。用例包括：管理收件箱并草拟回复、审阅 GitHub PR、将 Figma 设计转为代码、理解大型代码库、自动分类 bug、用自然语言查询数据集、从提示词部署应用、构建 Mac/iOS 应用、自动创建幻灯片、将 Slack 对话转为编码任务、用 AI 动作操控电脑。Codex 正从 AI 助手演变为 AI 队友。", "descriptionEn": "OpenAI 公布了数十个 Codex 实际工作流程，展示团队如何用其自动化任务。用例包括：管理收件箱并草拟回复、审阅 GitHub PR、将 Figma 设计转为代码、理解大型代码库、自动分类 bug、用自然语言查询数据集、从提示词部署应用、构建 Mac/iOS 应用、自动创建幻灯片、将 Slack 对话转为编码任务、用 AI 动作操控电脑。Codex 正从 AI 助手演变为 AI 队友。", "image": null, "sourceName": "X：Greg Brockman (@gdb)", "sourceCategory": "tip", "publishedAt": "2026-06-07T19:30:55.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "automation", "agent-architecture", "coding-agent" ], "relevanceScore": 0.94, "seelesAngle": "Automation signal: workflow automation, browser/desktop automation, CLI automation, scheduled jobs, or operational loops relevant to agent execution." }, { "id": "news-f4d37554", "url": "https://x.com/rohanpaul_ai/status/2063698758517366884", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2063698758517366884", "title": "Meta-Agent Challenge：当前AI智能体能否自主构建更好的智能体？", "titleEn": "This paper tests whether today's AI agents can build better AI agents without human design help. i….", "description": "一项新研究提出Meta-Agent Challenge（MAC）基准，测试AI智能体能否在没有人类设计帮助的情况下自主构建更优智能体。智能体需在安全工作区内自行发明策略、编写代码、测试并从失败中学习。实验覆盖数学、科学问答、竞赛编程、代码修复和长终端任务5个领域。结果显示，当前智能体大多无法超越人工设计的强智能体系统，仅Claude等少数封闭前沿模型取得较好表现。研究认为，当前智能体更像是强大的执行者，而非具备可靠自改进能力的工程师。", "descriptionEn": "一项新研究提出Meta-Agent Challenge（MAC）基准，测试AI智能体能否在没有人类设计帮助的情况下自主构建更优智能体。智能体需在安全工作区内自行发明策略、编写代码、测试并从失败中学习。实验覆盖数学、科学问答、竞赛编程、代码修复和长终端任务5个领域。结果显示，当前智能体大多无法超越人工设计的强智能体系统，仅Claude等少数封闭前沿模型取得较好表现。研究认为，当前智能体更像是强大的执行者，而非具备可靠自改进能力的工程师。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "tip", "publishedAt": "2026-06-07T19:05:00.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "agent-safety" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-28e0614a", "url": "https://x.com/steipete/status/2063697162748260627", "canonicalUrl": "https://x.com/steipete/status/2063697162748260627", "title": "月度提醒：设计循环提示编码智能体", "titleEn": "Here's your monthly reminder that you shouldn't be prompting coding agents anymore. You should be d…", "description": "这是您的月度提醒：您不应再手动提示编码智能体了。\n\n您应该设计循环来提示您的智能体。", "descriptionEn": "这是您的月度提醒：您不应再手动提示编码智能体了。\n\n您应该设计循环来提示您的智能体。", "image": null, "sourceName": "X：Peter Steinberger (@steipete)", "sourceCategory": "tip", "publishedAt": "2026-06-07T18:58:39.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-d255287a", "url": "https://x.com/dotey/status/2063686036895478162", "canonicalUrl": "https://x.com/dotey/status/2063686036895478162", "title": "ChatGPT 要变 AgentGPT 了", "titleEn": "ChatGPT 要变 AgentGPT 了 当然 ChatGPT 应该不会改名字，但 ChatGPT 应该不再是一个单纯的 Chat 工具了。 OpenAI 内部一位高管对《金融时报》说：\"Cha…", "description": "OpenAI 正筹备 ChatGPT 自 2022 年上线以来最大规模改版，从聊天机器人转向超级应用/Agent 平台，整合编程工具 Codex、图像生成及第三方应用（Canva、Booking）。高管称\"聊天已死\"，目标成为跨平台个人 AI 助手，未来甚至省去用户输入提示词。改版预计未来几周在网页和手机端上线。商业压力：ChatGPT 有 9 亿周活用户、5000 万付费用户、月收入 20 亿美元但未盈利；企业客户贡献约 40% 收入，目标年底达 50%；Codex 桌面版周活超 500 万。竞争对手 Anthropic 估值已达 9650 亿美元。OpenAI 刚完成 1220 亿美元融资，估值 8520 亿美元，IPO 目标估值可能超 1 万亿美元。", "descriptionEn": "OpenAI 正筹备 ChatGPT 自 2022 年上线以来最大规模改版，从聊天机器人转向超级应用/Agent 平台，整合编程工具 Codex、图像生成及第三方应用（Canva、Booking）。高管称\"聊天已死\"，目标成为跨平台个人 AI 助手，未来甚至省去用户输入提示词。改版预计未来几周在网页和手机端上线。商业压力：ChatGPT 有 9 亿周活用户、5000 万付费用户、月收入 20 亿美元但未盈利；企业客户贡献约 40% 收入，目标年底达 50%；Codex 桌面版周活超 500 万。竞争对手 Anthropic 估值已达 9650 亿美元。OpenAI 刚完成 1220 亿美元融资，估值 8520 亿美元，IPO 目标估值可能超 1 万亿美元。", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "industry", "publishedAt": "2026-06-07T18:14:27.000Z", "discoveredAt": "2026-06-08T00:20:22.071Z", "isSelected": true, "collectionSources": [ "selected", "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-b9224746", "url": "https://x.com/dotey/status/2063674134903603302", "canonicalUrl": "https://x.com/dotey/status/2063674134903603302", "title": "baoyu-design：在本地复现 Claude Design 的开发工作流", "titleEn": "之所以研究 Claude Design，是因为最近摸索出一套不错的开发模式： 1. 先用 Claude Design 去设计 App 的 UI 和 UX，第一版本越简单越好。 Claude Des…", "description": "宝玉分享开发模式：先用 Claude Design 设计 App UI/UX，生成 HTML+CSS+React+data.js（优于 Figma，利于 AI 理解和 Git 版本管理）；再用 Claude Opus 4.8 实现 MVP（GPT 5.5 在 UI 实现上差距明显）。因反复在本地与网页间导出替换设计稿繁琐，他编写工具解析 HAR 文件、解密 Claude Design 二进制内容，分析其 Prompt 和内置组件，将核心逻辑本地化集成到 Cursor（利用网页标记功能），配合 Opus 4.8 即可本地运行。最终开源 GitHub 项目\"baoyu-design\"。", "descriptionEn": "宝玉分享开发模式：先用 Claude Design 设计 App UI/UX，生成 HTML+CSS+React+data.js（优于 Figma，利于 AI 理解和 Git 版本管理）；再用 Claude Opus 4.8 实现 MVP（GPT 5.5 在 UI 实现上差距明显）。因反复在本地与网页间导出替换设计稿繁琐，他编写工具解析 HAR 文件、解密 Claude Design 二进制内容，分析其 Prompt 和内置组件，将核心逻辑本地化集成到 Cursor（利用网页标记功能），配合 Opus 4.8 即可本地运行。最终开源 GitHub 项目\"baoyu-design\"。", "image": null, "sourceName": "X：宝玉 (@dotey)", "sourceCategory": "tip", "publishedAt": "2026-06-07T17:27:09.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture", "coding-agent" ], "relevanceScore": 0.82, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-2cb8bdd", "url": "https://x.com/omarsar0/status/2063668567447597273", "canonicalUrl": "https://x.com/omarsar0/status/2063668567447597273", "title": "论文提出用代码压缩率衡量AI智能体是否真正发现新知识", "titleEn": "This was one of the standout AI papers of the week. （bookmark it） It tackles a question most self-…", "description": "本周一篇AI论文探讨自我改进智能体是否真正发现新知识，还是仅重新组合已知信息。作者将行为分为三类：检索（查询已有笔记本）、搜索（组合现有工具）和发现（发明新概念），并用范畴论和左Kan扩展定义--若旧版本能产生相同结果则非发现。他们构建Builder/Breaker agent研究蛋白质力学，四轮中R2从0.48升至0.68再降至0.54和0.41，看似变差实则不断挑战更难蛋白质并重写理论：数据增长近10倍，模型代码仅增长1.3倍。论文提出用代码压缩率作为真实发现信号。链接：arxiv.org/abs/2606.01444。", "descriptionEn": "本周一篇AI论文探讨自我改进智能体是否真正发现新知识，还是仅重新组合已知信息。作者将行为分为三类：检索（查询已有笔记本）、搜索（组合现有工具）和发现（发明新概念），并用范畴论和左Kan扩展定义--若旧版本能产生相同结果则非发现。他们构建Builder/Breaker agent研究蛋白质力学，四轮中R2从0.48升至0.68再降至0.54和0.41，看似变差实则不断挑战更难蛋白质并重写理论：数据增长近10倍，模型代码仅增长1.3倍。论文提出用代码压缩率作为真实发现信号。链接：arxiv.org/abs/2606.01444。", "image": null, "sourceName": "X：Elvis Saravia (@omarsar0, DAIR.AI)", "sourceCategory": "tip", "publishedAt": "2026-06-07T17:05:02.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-architecture" ], "relevanceScore": 0.78, "seelesAngle": "Agent architecture signal: orchestration, planning, memory, or autonomous runtime behavior that maps to Seeles agent capability design." }, { "id": "news-49c79873", "url": "https://x.com/rohanpaul_ai/status/2063662672835703126", "canonicalUrl": "https://x.com/rohanpaul_ai/status/2063662672835703126", "title": "特朗普政府与OpenAI讨论通过公共财富基金入股AI初创公司", "titleEn": "FT： The proposal suddenly of a sovereign-wealth-style fund got more attention inside the White House…", "description": "据FT报道，特朗普政府正与OpenAI探讨通过公共财富基金机制让政府入股AI初创公司。方案是AI企业捐赠小部分股权至该基金，基金通过账户或分红将收益返还美国公民，而非政府直接运营公司。这不同于特朗普去年对Intel的90亿美元直接持股。OpenAI此前已提出公共财富基金概念，持有长期资产，让公民分享AI增长红利。政治背景方面，选民担忧失业、数据中心成本与企业控制，而AI公司需要华盛顿在基建、采购和监管上的支持。", "descriptionEn": "据FT报道，特朗普政府正与OpenAI探讨通过公共财富基金机制让政府入股AI初创公司。方案是AI企业捐赠小部分股权至该基金，基金通过账户或分红将收益返还美国公民，而非政府直接运营公司。这不同于特朗普去年对Intel的90亿美元直接持股。OpenAI此前已提出公共财富基金概念，持有长期资产，让公民分享AI增长红利。政治背景方面，选民担忧失业、数据中心成本与企业控制，而AI公司需要华盛顿在基建、采购和监管上的支持。", "image": null, "sourceName": "X：Rohan Paul (@rohanpaul_ai)", "sourceCategory": "industry", "publishedAt": "2026-06-07T16:41:36.000Z", "discoveredAt": "2026-06-08T00:20:22.071Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-cda5b42c", "url": "https://techcrunch.com/2026/06/07/openai-is-still-working-on-that-super-app", "canonicalUrl": "https://techcrunch.com/2026/06/07/openai-is-still-working-on-that-super-app", "title": "OpenAI 仍推进超级应用计划", "titleEn": "OpenAI is still working on that 'super app'", "description": "OpenAI 一位高级员工表示\"聊天已死\"，同时公司仍在继续推进超级应用计划。", "descriptionEn": "OpenAI 一位高级员工表示\"聊天已死\"，同时公司仍在继续推进超级应用计划。", "image": null, "sourceName": "TechCrunch：AI（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-07T16:23:22.000Z", "discoveredAt": "2026-06-08T00:20:22.071Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-41117f4d", "url": "https://the-decoder.com/deepseek-topped-ramps-trending-software-vendors-in-june-2026-as-us-companies-chase-cheaper-ai", "canonicalUrl": "https://the-decoder.com/deepseek-topped-ramps-trending-software-vendors-in-june-2026-as-us-companies-chase-cheaper-ai", "title": "DeepSeek 在 2026 年 6 月登顶 Ramp 热门软件供应商，美国公司追逐更便宜的 AI", "titleEn": "Deepseek topped Ramp's trending software vendors in June 2026 as US companies chase cheaper AI", "description": "DeepSeek 在 2026 年 6 月成为 Ramp 平台最热门的软件供应商，作为一项付费服务，美国公司直接向其发送数据。Ramp 首席经济学家 Ara Kharazian 指出成本意识是驱动因素，但警告使用中国模型存在安全风险。", "descriptionEn": "DeepSeek 在 2026 年 6 月成为 Ramp 平台最热门的软件供应商，作为一项付费服务，美国公司直接向其发送数据。Ramp 首席经济学家 Ara Kharazian 指出成本意识是驱动因素，但警告使用中国模型存在安全风险。", "image": null, "sourceName": "The Decoder：AI News（RSS）", "sourceCategory": "industry", "publishedAt": "2026-06-07T16:06:34.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "agent-safety" ], "relevanceScore": 0.86, "seelesAngle": "Agent safety signal: governance, guardrails, privacy, permissions, or auditability patterns needed for production agent systems." }, { "id": "news-23ef0f6d", "url": "https://x.com/AYi_AInotes/status/2063652807522074651", "canonicalUrl": "https://x.com/AYi_AInotes/status/2063652807522074651", "title": "免费获取AI额度的四个隐藏入口", "titleEn": "http：//x.com/i/article/2063647311423561728", "description": "文章盘点四大渠道：1 OpenAI Codex for OSS，开源项目维护者可免费获6个月ChatGPT Pro（价值$1200）及API credits；2 Anything AI，新用户通过促销链接注册可得2万credits，需重复点击触发bonus；3 Lenny's Product Pass，付费订阅其年度Newsletter（$200-400）可兑换总价值$30000+的AI工具一年会员，先到先得且要求新用户；4 三大云startup credits，微软Founders Hub最低$1K-$5K（无VC要求），谷歌AI-first创业最高$350K但要求严，AWS可叠加（自筹$1K+YC等）。每个口子均有门槛限制。", "descriptionEn": "文章盘点四大渠道：1 OpenAI Codex for OSS，开源项目维护者可免费获6个月ChatGPT Pro（价值$1200）及API credits；2 Anything AI，新用户通过促销链接注册可得2万credits，需重复点击触发bonus；3 Lenny's Product Pass，付费订阅其年度Newsletter（$200-400）可兑换总价值$30000+的AI工具一年会员，先到先得且要求新用户；4 三大云startup credits，微软Founders Hub最低$1K-$5K（无VC要求），谷歌AI-first创业最高$350K但要求严，AWS可叠加（自筹$1K+YC等）。每个口子均有门槛限制。", "image": null, "sourceName": "X：阿易 AI Notes (@AYi_AInotes)", "sourceCategory": "tip", "publishedAt": "2026-06-07T16:02:24.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." }, { "id": "news-e3af5cff", "url": "https://garymarcus.substack.com/p/slop-productivity-and-why-the-ai", "canonicalUrl": "https://garymarcus.substack.com/p/slop-productivity-and-why-the-ai", "title": "Slop、生产力，以及为何AI驱动的世界进展甚微", "titleEn": "Slop， productivity， and why the AI-fueled world is going nowhere mighty fast", "description": "Gary Marcus在金融时报上看到John Burn-Murdoch的一张图表，认为它精准提炼了自己一直试图表达的观点。", "descriptionEn": "Gary Marcus在金融时报上看到John Burn-Murdoch的一张图表，认为它精准提炼了自己一直试图表达的观点。", "image": null, "sourceName": "Gary Marcus：The Road to AI We Can Trust（RSS）", "sourceCategory": "tip", "publishedAt": "2026-06-07T15:31:43.000Z", "discoveredAt": "2026-06-08T00:20:22.071Z", "isSelected": true, "collectionSources": [ "selected" ], "tags": [ "recommended-industry" ], "relevanceScore": 0.5, "seelesAngle": "Recommended industry signal: company, market, policy, or platform movement around the AI ecosystem." }, { "id": "news-2e0e992f", "url": "https://x.com/kimmonismus/status/2063643307381543039", "canonicalUrl": "https://x.com/kimmonismus/status/2063643307381543039", "title": "ChatGPT 超级应用改造即将分阶段推出", "titleEn": "OpenAI's ChatGPT \"superapp\" overhaul could start rolling out in the coming weeks. The first wave li…", "description": "OpenAI 的 ChatGPT \"超级应用\"改造可能在未来几周开始推出。\n\n第一波可能不是一次大发布，而是分阶段重新设计，将用户推向 Codex、AI 智能体、图像生成和合作伙伴应用。\n\n正如一位 OpenAI 员工告诉《金融时报》：\"聊天已死\"，新目标是打造一个能在你的工作和个人生活中行动的单一 AI 助手。", "descriptionEn": "OpenAI 的 ChatGPT \"超级应用\"改造可能在未来几周开始推出。\n\n第一波可能不是一次大发布，而是分阶段重新设计，将用户推向 Codex、AI 智能体、图像生成和合作伙伴应用。\n\n正如一位 OpenAI 员工告诉《金融时报》：\"聊天已死\"，新目标是打造一个能在你的工作和个人生活中行动的单一 AI 助手。", "image": null, "sourceName": "X：Kim (@kimmonismus)", "sourceCategory": "ai-products", "publishedAt": "2026-06-07T15:24:39.000Z", "discoveredAt": "2026-06-08T00:20:22.642Z", "isSelected": false, "collectionSources": [ "all" ], "tags": [ "coding-agent" ], "relevanceScore": 0.78, "seelesAngle": "Coding-agent signal: developer agents, code-generation workflows, or coding-product infrastructure relevant to Seeles developer-facing agent patterns." } ] }