{"no":99,"date":"2026.11.13","weekday":"星期五","cn_date":"2026年11月13日 星期五","pub_date":"2026.11.14","pub_weekday":"星期六","pub_cn_date":"2026年11月14日 星期六","total":34,"rest":16,"one_line":"Gemini 4 转向日常编码与智能体，IBM 以标记语言入局；AI 落地提速之际，玛雅误报提醒我们仍需校准边界。","cover":null,"items":[{"n":1,"type":"新模型","src":"DeepMind","title":"Google DeepMind 发布 Gemini 4 Hydrogen：面向日常编码与智能体工作流","sum":"Google DeepMind 发布 Gemini 4 Hydrogen，主打日常编码与智能体工作流的快速执行，并同步登陆 Google AI Studio、Gemini API、Gemini 应用与搜索等全线产品。其默认高思考档在 Artificial Analysis 智能指数上以 <b>51 比 50</b> 超过 Argon，最高 xhigh 档达 <b>54</b>；100 万 token 上下文内输入 <b>1.50 美元</b>、输出 <b>7.50 美元</b>每百万 token。","img":"media/2026.11.13/07.jpg","links":[{"url":"https://deepmind.google/blog/introducing-gemini-4-hydrogen/"},{"url":"https://antigravity.google/blog/gemini-4-hydrogen-in-google-antigravity/"},{"url":"https://www.ithome.com/1/026/446.htm"}],"section":"models","prior":{"date":"2026.09.03","title":"Google DeepMind发布WeatherNext 3气象大模型，每小时刷新与5公里高精度全球预报","pub_date":"2026.09.04"},"ts":"2026.11.13 02:00","imageMeta":{"type":"article","label":"原文配图","url":"https://deepmind.google/blog/introducing-gemini-4-hydrogen/","source":"Google DeepMind"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":449}},{"n":2,"type":"新模型","src":"HuggingFace","title":"IBM 发布 LLMON 标记语言，并推出 Granite 4.1 3B/8B 预览模型","sum":"IBM 团队提出 LLMON（LLM Object Notation），一种用保留特殊 token 标记结构与语义的 LLM 原生标记语言，用于区分指令、数据与要求。基于 Granite 4.1 基座，他们发布 <b>granite-4.1-3b-LLMON-preview</b> 和 <b>granite-4.1-8b-LLMON-preview</b> 两个实验性预览模型，并开源配套工具包与示例。","img":"media/2026.11.13/08.jpg","links":[{"url":"https://huggingface.co/blog/ibm-granite/llmon-an-llm-native-markup-language-now-in-granite"}],"section":"models","prior":{"date":"2026.10.16","title":"Granite Switch 4.1 发布：一个端点同时处理对话、决策与语音","pub_date":"2026.10.17"},"ts":"2026.11.13 03:07","imageMeta":{"type":"article","label":"原文配图","url":"https://huggingface.co/blog/ibm-granite/llmon-an-llm-native-markup-language-now-in-granite","source":"Hugging Face"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":701}},{"n":3,"type":"行业动态","src":"OpenAI","title":"Amplitude 用 OpenAI 模型打造 Wave，产品改进从一个月缩到 1–2 天","sum":"Amplitude 推出主动式产品智能体 Wave，把产品数据、会话回放、客户反馈和代码整合起来，识别机会、制定方案并对接 Codex 等编码智能体落地。<b>Wave 约 80% 由 OpenAI 模型驱动</b>，包括 GPT‑6.1 Sol、GPT‑6 Luna 和 GPT‑Image‑2.5 Flare。Amplitude 估计它能把产品改进的交付时间从约一个月缩短到 <b>1–2 天</b>。","img":"media/2026.11.13/news-auto-92dc9ca522431b417dae.png","links":[{"url":"https://openai.com/index/amplitude"}],"section":"models","prior":null,"ts":"2026.11.13 04:00","imageMeta":{"type":"article","kind":"unknown","label":"原文配图","url":"https://images.ctfassets.net/kftzwdyauwt9/7xyAtQVA8C5CZGHW7aQidr/a71aedc6e0ed7395f081f75411af6d03/amplitude-wave-seo.png?w=1600&h=900&fit=fill","sourcePage":"https://openai.com/index/amplitude","caption":"来源文章封面","sourceFileUnchanged":true,"source":"OpenAI"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":1600,"height":900}},{"n":4,"type":"新产品","src":"新产品","title":"Aircad：浏览器里的专业 2D 制图与 3D 建模，带 AI 智能体和多人协作","sum":"Aircad 是一款浏览器端制图与建模工具，支持 <b>2D 制图与 3D 实体建模</b>、命令行操作、NURBS 建模和内置图块库，可直接打开 DWG 文件。它内置 AI 智能体，可用自然语言提问或描述修改，所有改动先以草稿形式高亮呈现，由用户决定应用或丢弃。协作方面支持多人同时编辑同一文件、评论和权限控制，每次修改都记录修改人和时间。","img":"media/2026.11.13/01.jpg","links":[{"url":"https://aircad.com"}],"section":"creative","prior":null,"ts":"2026.11.10 23:01","imageMeta":{"type":"article","label":"原文配图","url":"https://aircad.com","source":"aircad.com"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":420}},{"n":5,"type":"产品发布","src":"TechCrunch","title":"Nectar 推出 Creator Agent，自动搜寻并联系网红创作者","sum":"AI 营销操作系统 Nectar 推出 <b>Creator Agent</b>，可自动搜索创作者、检查内容与品牌合作、发送个性化消息并跟踪投放效果。它可检索超 <b>3 亿</b>个创作者档案，覆盖 Instagram 和 TikTok，今日全球上线。","img":"media/2026.11.13/09.jpg","links":[{"url":"https://techcrunch.com/2026/11/12/nectar-launches-an-ai-agent-to-automate-creator-discovery-and-outreach-for-brands/"}],"section":"agents","prior":null,"ts":"2026.11.13 03:11","imageMeta":{"type":"article","label":"原文配图","url":"https://techcrunch.com/2026/11/12/nectar-launches-an-ai-agent-to-automate-creator-discovery-and-outreach-for-brands/","source":"TechCrunch"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":450}},{"n":6,"type":"AI事件","src":"The Verge","title":"AI智能体并未发现未知玛雅遗址，考古学家批评其误报","sum":"谷歌研究员Pietro Miotti用<b>48个AI智能体</b>分析NASA激光雷达数据，宣称在墨西哥发现一处未登记的玛雅遗址。但该遗址<b>一年多前已在学术期刊发表</b>，考古学家批评其缺乏专业核查、传播错误信息。Miotti承认自己不是考古学家，已更新博客致谢。","img":"media/2026.11.13/11.jpg","links":[{"url":"https://www.theverge.com/ai-artificial-intelligence/1032125/ai-agents-maya-site-archeology-discovery"}],"section":"agents","prior":null,"ts":"2026.11.13 03:42","imageMeta":{"type":"article","label":"原文配图","url":"https://www.theverge.com/ai-artificial-intelligence/1032125/ai-agents-maya-site-archeology-discovery","source":"The Verge"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":418}},{"n":7,"type":"行业动态","src":"微软研究院","title":"微软研究院焦点第98期：AI通用翻译、31年数学猜想与编程智能体提效","sum":"微软研究院发布《Research Focus》第98期，主持人Anson Ho介绍多项最新研究：AI推动通用翻译更接近现实，<b>31年未解的KLS猜想被证明</b>，并推出<b>COMPACTIONBENCH</b>基准与<b>APEC</b>方法优化编程智能体的上下文压缩，成本最多降低25.4%。","img":"media/2026.11.13/14.jpg","links":[{"url":"https://www.microsoft.com/en-us/research/podcast/microsoft-research-focus-edition-98/"}],"section":"agents","prior":null,"ts":"2026.11.13 05:31","imageMeta":{"type":"article","label":"原文配图","url":"https://www.microsoft.com/en-us/research/podcast/microsoft-research-focus-edition-98/","source":"Microsoft"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":450}},{"n":8,"type":"产品发布","src":"TechCrunch","title":"Phota Labs 推出 iOS 应用：用 AI 把日常照片变成专业级作品","sum":"Phota Labs 推出一款 iOS 应用，提供个人 AI 照片助手，把日常快照变成专业摄影师水准的照片，同时保留画面真实感。它不做传统修图工具，而是用 AI 自动化编辑流程，重点修复拍摄本身的问题并保持人物身份特征。公司由 Cecilia Zhang 和 Zach Xia 创立，二人曾在 Adobe 参与 Project Indigo、Lightroom 和 Photoshop 等计算摄影产品。","img":"media/2026.11.13/03.jpg","links":[{"url":"https://techcrunch.com/2026/11/12/phota-labs-uses-ai-to-turn-everyday-photos-into-professional-looking-shots/"}],"section":"products","prior":null,"ts":"2026.11.13 01:00","imageMeta":{"type":"article","label":"原文配图","url":"https://techcrunch.com/2026/11/12/phota-labs-uses-ai-to-turn-everyday-photos-into-professional-looking-shots/","source":"TechCrunch"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":578}},{"n":9,"type":"行业动态","src":"The Verge","title":"GDC 调查：40% 游戏开发者已在工作中使用生成式 AI","sum":"GDC Festival of Gaming 最新年度行业调查显示，<b>40% 的受访开发者表示在工作中使用生成式 AI</b>，高于 2025 年的 36%。免费游戏开发者使用率达 58%，仅做移动端的工作室员工达 70%。调查同时显示，<b>34% 的受访者称过去两年内曾被裁员</b>。","img":"media/2026.11.13/05.jpg","links":[{"url":"https://www.theverge.com/games/1032085/gdc-festival-of-gaming-annual-survey-ai-layoffs"}],"section":"products","prior":null,"ts":"2026.11.13 01:00","imageMeta":{"type":"article","label":"原文配图","url":"https://www.theverge.com/games/1032085/gdc-festival-of-gaming-annual-survey-ai-layoffs","source":"The Verge"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":416}},{"n":10,"type":"行业动态","src":"新智元","title":"前OpenAI研究员改口：超级AI研究员可能明年就来！","sum":"AI Futures Project在11月10日发布新预测，认为若前沿AI研发没有显著主动放缓，<b>超级AI研究员很可能在2027年出现</b>。主要依据是METR对Anthropic的11周调研：从2025年10月起，其内部模型基准分增速达到此前的<b>1.9倍</b>。团队据此推算，AI带来的研发提速可能比《AI 2027》推演提前约五个月。","img":"media/2026.11.13/20.jpg","links":[{"url":"https://aiera.com.cn/2026/11/13/other/admin/123651/%e5%89%8dopenai%e7%a0%94%e7%a9%b6%e5%91%98%e6%94%b9%e5%8f%a3%ef%bc%9a%e8%b6%85%e7%ba%a7ai%e7%a0%94%e7%a9%b6%e5%91%98%e5%8f%af%e8%83%bd%e6%98%8e%e5%b9%b4%e5%b0%b1%e6%9d%a5%ef%bc%81/"}],"section":"research","prior":null,"ts":"2026.11.13 08:06","imageMeta":{"type":"article","label":"原文配图","url":"https://aiera.com.cn/2026/11/13/other/admin/123651/%e5%89%8dopenai%e7%a0%94%e7%a9%b6%e5%91%98%e6%94%b9%e5%8f%a3%ef%bc%9a%e8%b6%85%e7%ba%a7ai%e7%a0%94%e7%a9%b6%e5%91%98%e5%8f%af%e8%83%bd%e6%98%8e%e5%b9%b4%e5%b0%b1%e6%9d%a5%ef%bc%81/","source":"新智元"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":420}},{"n":11,"type":"AI 事件","src":"IT之家","title":"Anthropic、OpenAI 遭专利诉讼，被指侵犯五项神经网络技术专利","sum":"美国公司 Flinn AI 于当地时间 11 月 12 日向特拉华州地区法院提起两起诉讼，将 <b>Anthropic PBC</b> 和 <b>OpenAI OpCo</b> 共同列为被告，指控其侵犯五项神经网络相关专利。涉案专利涉及<b>迭代式注意力神经网络训练与处理</b>，Flinn AI 称其可追溯至 2015 年末至 2016 年初的申请。针对 Anthropic 的指控主要涉及 Claude 系列大语言模型。","img":"media/2026.11.13/18.jpg","links":[{"url":"https://www.ithome.com/1/026/464.htm"}],"section":"governance","prior":null,"ts":"2026.11.13 07:33","imageMeta":{"type":"article","label":"原文配图","url":"https://www.ithome.com/1/026/464.htm","source":"IT之家"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":893}},{"n":12,"type":"AI医疗","src":"新智元","title":"250名三甲医生输给中国AI！引发中美医疗权威热议","sum":"未来医生联合国内顶级三甲医院胸外科开展多中心研究，用40例真实疑难肺癌病例测试AI与医生。未来医生MedGPT平均得分<b>85.29</b>，250名三甲医生平均<b>53.26</b>分，一对一比较中AI以<b>88:12</b>胜出，且答卷平均短四成以上。","img":"media/2026.11.13/23.jpg","links":[{"url":"https://aiera.com.cn/2026/11/13/other/admin/123606/250%e5%90%8d%e4%b8%89%e7%94%b2%e5%8c%bb%e7%94%9f%e8%be%93%e7%bb%99%e4%b8%ad%e5%9b%bdai%ef%bc%81%e5%bc%95%e5%8f%91%e4%b8%ad%e7%be%8e%e5%8c%bb%e7%96%97%e6%9d%83%e5%a8%81%e7%83%ad%e8%ae%ae/"}],"section":"models","prior":null,"ts":"2026.11.13 08:03","imageMeta":{"type":"article","label":"原文配图","url":"https://aiera.com.cn/2026/11/13/other/admin/123606/250%e5%90%8d%e4%b8%89%e7%94%b2%e5%8c%bb%e7%94%9f%e8%be%93%e7%bb%99%e4%b8%ad%e5%9b%bdai%ef%bc%81%e5%bc%95%e5%8f%91%e4%b8%ad%e7%be%8e%e5%8c%bb%e7%96%97%e6%9d%83%e5%a8%81%e7%83%ad%e8%ae%ae/","source":"新智元"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":420}},{"n":13,"type":"论文","src":"新智元","title":"Coding agent写测试集体失败！AI代码合格率被高估一倍","sum":"南京大学刘佳恒团队提出测试生成基准TestPrism，为每个任务配备1份参考实现、4份正确替代实现和5份错误实现，并以联合成功函数（JSF）为核心指标。<b>14种Coding Agent配置中，最强配置JSF仅28.00%，而其在单一参考评测下成功率达59.67%</b>，超过一半通过参考答案的测试换一批实现就不可靠。团队还归纳出3大类7种错误机制，并提出TestHelix框架将JSF提升11.67至12.00个百分点。","img":"media/2026.11.13/22.jpg","links":[{"url":"https://aiera.com.cn/2026/11/13/other/admin/123621/coding-agent%e5%86%99%e6%b5%8b%e8%af%95%e9%9b%86%e4%bd%93%e5%a4%b1%e8%b4%a5%ef%bc%81ai%e4%bb%a3%e7%a0%81%e5%90%88%e6%a0%bc%e7%8e%87%e8%a2%ab%e9%ab%98%e4%bc%b0%e4%b8%80%e5%80%8d/"},{"url":"https://jigou.jiqizhixin.com/articles/2026-11-13-4"}],"section":"agents","prior":{"date":"2026.08.11","title":"开发 Coding Agents 应该选择哪种编程语言？","pub_date":"2026.08.12"},"ts":"2026.11.13 08:04","imageMeta":{"type":"article","label":"原文配图","url":"https://aiera.com.cn/2026/11/13/other/admin/123621/coding-agent%e5%86%99%e6%b5%8b%e8%af%95%e9%9b%86%e4%bd%93%e5%a4%b1%e8%b4%a5%ef%bc%81ai%e4%bb%a3%e7%a0%81%e5%90%88%e6%a0%bc%e7%8e%87%e8%a2%ab%e9%ab%98%e4%bc%b0%e4%b8%80%e5%80%8d/","source":"新智元"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":420}},{"n":14,"type":"数据集发布","src":"机器之心","title":"后GPT-6时代，具身数据怎么做？Ropedia发布Xperience 2，给出下一代数据基石金标准","sum":"Physical AI初创公司Ropedia发布<b>Xperience 2</b>，将真实场景扫描与人的第一视角活动记录在同一度量坐标系中对齐，把人的动作转换为机器人轨迹、扫描场景转换为可交互仿真环境。本次发布<b>1,589个场景</b>、<b>4,230段录像</b>，覆盖12类真实空间，并围绕空间能力构建三项Benchmark。","img":"media/2026.11.13/37.jpg","links":[{"url":"https://jigou.jiqizhixin.com/articles/2026-11-13-6"}],"section":"research","prior":null,"ts":"2026.11.13 14:55","imageMeta":{"type":"source","label":"来源配图","url":"https://image.jiqizhixin.com/uploads/article/cover_image/079f612d-20fd-45d9-93fb-229998ebb127/0__22_(2).jpg","source":"机器之心"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":622,"height":622}},{"n":15,"type":"AI事件","src":"IT之家","title":"边警告 AI 风险边开发先进模型：OpenAI、Anthropic、谷歌等五家 AI 巨头遭起诉，被要求叫停 ChatGPT、Gemini、Claude","sum":"四家非营利组织向旧金山高等法院起诉<b>OpenAI、Anthropic、谷歌、Meta及SpaceXAI</b>，要求叫停ChatGPT、Gemini、Claude等先进AI产品的开发。原告大量援引被告高管自己关于AI可能造成文明毁灭的公开警告，并请求法院在被告证明技术受人类控制前暂停下一代模型训练与部署。","img":"media/2026.11.13/25.jpg","links":[{"url":"https://www.ithome.com/1/026/466.htm"}],"section":"models","prior":null,"ts":"2026.11.13 07:46","imageMeta":{"type":"article","label":"原文配图","url":"https://www.ithome.com/1/026/466.htm","source":"IT之家"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":808}},{"n":16,"type":"行业动态","src":"新智元","title":"百度押中AI下一幕！百万智能体组团上岗，AI科学家昼夜狂奔","sum":"在百度世界2026大会上，李彦宏提出AI发展的下一幕是<b>智能组织</b>，并给出打破岗位边界、沉淀隐性知识、设计流程闭环三条原则。百度智能云随后发布<b>「芯云模体」一体化解决方案</b>，由昆仑芯与百舸供算力、数据平台管数据、三类智能体干活，面向AI4S科研一线。","img":"media/2026.11.13/24.jpg","links":[{"url":"https://aiera.com.cn/2026/11/13/other/admin/123577/%e7%99%be%e5%ba%a6%e6%8a%bc%e4%b8%adai%e4%b8%8b%e4%b8%80%e5%b9%95%ef%bc%81%e7%99%be%e4%b8%87%e6%99%ba%e8%83%bd%e4%bd%93%e7%bb%84%e5%9b%a2%e4%b8%8a%e5%b2%97%ef%bc%8cai%e7%a7%91%e5%ad%a6%e5%ae%b6/"},{"url":"https://www.ifanr.com/1689182?utm_source=rss&utm_medium=rss&utm_campaign="}],"section":"agents","prior":null,"ts":"2026.11.13 08:00","imageMeta":{"type":"article","label":"原文配图","url":"https://aiera.com.cn/2026/11/13/other/admin/123577/%e7%99%be%e5%ba%a6%e6%8a%bc%e4%b8%adai%e4%b8%8b%e4%b8%80%e5%b9%95%ef%bc%81%e7%99%be%e4%b8%87%e6%99%ba%e8%83%bd%e4%bd%93%e7%bb%84%e5%9b%a2%e4%b8%8a%e5%b2%97%ef%bc%8cai%e7%a7%91%e5%ad%a6%e5%ae%b6/","source":"新智元"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":420}},{"n":17,"type":"产品发布","src":"量子位","title":"大模型都能控机器人，Human Data还得怎么卷？","sum":"在2026 CoRL上，简智机器人发布DAS 2.0旗下两款新品：第一视角头戴式多模态采集设备<b>DAS Ego 2.0</b>和灵巧手数据采集手套<b>DAS Dex 2.0</b>。Ego 2.0搭载6颗全局快门摄像头、水平视野340°、主相机500万像素；Dex 2.0单手16块触觉传感器、980个触觉感知单元、触觉分辨率1mm。两者可实现约1ms级头手时间对齐。","img":"media/2026.11.13/46.jpg","links":[{"url":"https://www.qbitai.com/2026/11/509468.html"}],"section":"research","prior":null,"ts":"2026.11.13 11:31","imageMeta":{"type":"article","label":"原文配图","url":"https://www.qbitai.com/2026/11/509468.html","source":"量子位"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":720,"height":720}},{"n":18,"type":"论文","src":"机器之心","title":"从数学理论「解」出语言模型——AGI理论核心原语的首次工程验证：BriLLM-MetaPred","sum":"上海交通大学赵海教授团队上传 arXiv 论文 BriLLM-MetaPred，把 AGI 数学理论推导出的机器学习原语 MetaPred 首次做成可运行的语言模型。它用单一动力学状态同时承载表示与预测，约 <b>13.2M 参数</b>下性能与同规模 GPT 基线相当，最大验证尺寸已达 <b>1.5B</b>。","img":"media/2026.11.13/31.jpg","links":[{"url":"https://jigou.jiqizhixin.com/articles/2026-11-13-2"}],"section":"models","prior":null,"ts":"2026.11.13 11:12","imageMeta":{"type":"source","label":"来源配图","url":"https://image.jiqizhixin.com/uploads/article/cover_image/8367830e-9c0f-4d6c-b5a2-8fa5dc7b32c9/Screenshot_2026-11-13_at_11(2).12.49.png","source":"机器之心"},"coverPolicy":{"version":1,"status":"image","kind":"unknown","fit":"contain","focal":"50% 50%","width":800,"height":676}}]}