{"version":1,"computed_at":"2026-11-05T08:00:45.847Z","window_hours":72,"limit":20,"item_count":20,"items":[{"rank":1,"id":"blog:jiqizhixin:78f318a30d4a77f9","title_zh":"米哈游联合清华等高校开源 GameScaling 系列通用游戏模型","summary_zh":"这套模型想回答一个问题：当每个游戏的画面、物理和操作规则都不同时，加大数据量和参数量还能不能带来跨游戏迁移的能力。","source":"机器之心","url":"https://www.jiqizhixin.com/articles/2026-11-05-5","published_at":"2026-11-05T05:03:00.000Z","score":82.259,"event_source_count":1},{"rank":2,"id":"blog:qbitai:3fd34e47c220eb96","title_zh":"GPT-Image-2.6 全量上线，API 拆分规划与画质参数","summary_zh":"生图 API 第一次把「想多少」和「画多细」分开调，改图成本和等待时间明显下降，同时补上了 GIF 与 SVG 两个长期缺口。","source":"量子位","url":"https://www.qbitai.com/2026/11/507707.html","published_at":"2026-11-05T01:59:37.000Z","score":74.459,"event_source_count":1},{"rank":3,"id":"blog:the-verge:fcc18b15ca769256","title_zh":"Nothing 预告首款无屏幕 AI 设备，将于下月发布","summary_zh":"Nothing 将在 12 月推出首款无屏幕 AI 硬件，CEO 称它不能以 app 形式存在，这意味着又一家手机厂商在赌 AI 硬件的独立形态。","source":"The Verge","url":"https://www.theverge.com/ai-artificial-intelligence/1026586/nothing-screen-free-ai-device-carl-pei","published_at":"2026-11-04T18:26:07.000Z","score":68.983,"event_source_count":2},{"rank":4,"id":"blog:jiqizhixin:c6449506ac764b92","title_zh":"LiberAI 发布世界动作模型 Liber-0 及配套技术报告","summary_zh":"Liber-0 用 60 项未训练任务的真机测试加一份数据 Scaling 研究，把机器人基础模型的能力边界摊开给外界检验。","source":"机器之心","url":"https://www.jiqizhixin.com/articles/2026-11-05-3","published_at":"2026-11-05T02:37:09.000Z","score":65.164,"event_source_count":1},{"rank":5,"id":"blog:techcrunch:bbbf47c695ed3b41","title_zh":"Google 要求学校启用 AI 才能使用搜索，引发反弹","summary_zh":"Google 把搜索与 AI 访问权限捆绑销售给学校，招来政界“垄断勒索”的指控，这是观察 AI 进校园与反垄断交叉争议的一个缩影。","source":"TechCrunch","url":"https://techcrunch.com/2026/11/04/google-faces-backlash-for-plan-that-would-turn-off-search-for-schools-that-restrict-ai","published_at":"2026-11-04T16:17:22.000Z","score":64.643,"event_source_count":1},{"rank":6,"id":"blog:qbitai:d6b1383da92184bb","title_zh":"招商局狮子山实验室发布具身推理模型 Lion-ER-35B-A3B","summary_zh":"它把「造数据」从堆量改成按能力蓝图和评测结果闭环生产，3B 激活参数就在十项具身基准上全面涨分，给机器人理解三维物理世界提供了一条省算力的路子。","source":"量子位","url":"https://www.qbitai.com/2026/11/507586.html","published_at":"2026-11-04T10:08:38.000Z","score":59.575,"event_source_count":1},{"rank":7,"id":"blog:openai:295354ff422b0d4d","title_zh":"OpenAI 发布 Images 2.6：生成更快、支持 GIF、编辑更精确","summary_zh":"Images 2.6 是一次产品迭代：速度、GIF 和编辑精度同时提升，并开放新的 API 参数，方便开发者把它接进自己的流程。","source":"OpenAI","url":"https://openai.com/index/introducing-images-2-6","published_at":"2026-11-04T00:00:00.000Z","score":57.133,"event_source_count":1},{"rank":8,"id":"blog:openai:1be4fdd8e6ca80f0","title_zh":"OpenAI 收购 Pierre Computer，扩展 Codex 代码工具能力","summary_zh":"OpenAI 收购 Pierre Computer，把它的技术并入 Codex，补齐代码审查环节的工具链，是 AI 编程助手竞争中的一次能力收购。","source":"OpenAI","url":"https://openai.com/index/openai-to-acquire-pierre","published_at":"2026-11-04T14:30:00.000Z","score":51.644,"event_source_count":2},{"rank":9,"id":"blog:mit-tech-review:89c457ebe97048d8","title_zh":"MIT 研究：agentic AI 规模化落地需要数据血缘","summary_zh":"这篇讲企业用 agentic AI 后数据会不断被复制、摘要、改写，必须靠数据血缘（追踪数据从哪来、流向哪里）才能管住风险和追责。","source":"MIT Technology Review","url":"https://www.technologyreview.com/2026/11/04/1147596/securing-the-agentic-enterprise-with-data-lineage","published_at":"2026-11-04T13:00:00.000Z","score":50.878,"event_source_count":1},{"rank":10,"id":"blog:the-verge:b6bfab50d66d69a0","title_zh":"AT&T 智能家居安防系统接入 Gemini for Home","summary_zh":"Google 的 Gemini 家居能力首次通过 Home APIs 嵌入运营商自有的安防套餐，AI 从独立设备走向被集成进订阅服务。","source":"The Verge","url":"https://www.theverge.com/tech/1026787/att-gemini-for-home-connected-life-security-system","published_at":"2026-11-04T20:23:24.000Z","score":49.363,"event_source_count":1},{"rank":11,"id":"blog:techcrunch:32e73ddd4d076122","title_zh":"OpenAI 发布政策议程，试图塑造 2028 年美国大选议题","summary_zh":"OpenAI 想在大选前抢先定义 AI 政策议程：借新政叙事争取中间选民，却缺少让公司为 AI 冲击承担成本的硬性约束。","source":"TechCrunch","url":"https://techcrunch.com/2026/11/04/openai-is-already-trying-to-shape-the-2028-election","published_at":"2026-11-04T21:25:44.000Z","score":48.168,"event_source_count":1},{"rank":12,"id":"blog:together:3e89035fdd0e3c0b","title_zh":"Together AI 提出多态推理：用 AI agent 持续编译优化推理","summary_zh":"推理市场从“静态调优”转向模型、流量、硬件都在变的局面，Together AI 的解法是让 AI agent 搜索优化内核，在生产环境持续编译。","source":"Together AI","url":"https://www.together.ai/blog/polymorphic-inference","published_at":"2026-11-04T00:00:00.000Z","score":46.133,"event_source_count":1},{"rank":13,"id":"blog:mit-tech-review:e98e8f1032afef90","title_zh":"得州民主党赢得州选，AI 数据中心建设面临叫停风险","summary_zh":"得州选举变天，让这个 AI 数据中心建设最密集的州可能转向限制新建；叠加银行收紧放贷，AI 基建扩张的不确定性在上升。","source":"MIT Technology Review","url":"https://www.technologyreview.com/2026/11/04/1148043/texas-was-the-new-epicenter-of-the-ai-data-center-buildout-that-might-be-kaput","published_at":"2026-11-04T18:50:55.000Z","score":42.261,"event_source_count":1},{"rank":14,"id":"podcast:msr-podcast:02af5df8f895ba11","title_zh":"微软研究院 Mike Walker 谈黑客思维：安全规则是社会建构","summary_zh":"一位主导过 DARPA 网络挑战赛的安全研究者，讲为什么“懂规则”是黑客思维的核心，以及安全设计为何必须先算信任这笔账。","source":"Microsoft Research","url":"https://www.microsoft.com/en-us/research/podcast/learn-the-rules-the-hacker-mindset","published_at":"2026-11-03T17:12:55.000Z","score":41.517,"event_source_count":1},{"rank":15,"id":"blog:qbitai:ae61fa9082b82a43","title_zh":"百度发布世界模型 ApolloWorld，用于 L4 无人车训练与验证","summary_zh":"百度用可交互的世界模型把路测搬上云端，一天抵三年；更值得看的是它提出的「2+3」评价框架，用因果性、交互性和物理真实性给世界模型划出 L4 训练的能力门槛。","source":"量子位","url":"https://www.qbitai.com/2026/11/507647.html","published_at":"2026-11-04T16:13:23.000Z","score":36.604,"event_source_count":2},{"rank":16,"id":"blog:jiqizhixin:a4e63a3dc09bee24","title_zh":"ChatGPT Images 2.6 上线：支持直接生成 GIF 与参考图局部编辑","summary_zh":"OpenAI 把生图从「出一张图」延伸到「出图之后的改动」：动图、局部替换、可编辑矢量图都能用自然语言完成，但动画跳帧等细节仍是问题。","source":"机器之心","url":"https://www.jiqizhixin.com/articles/2026-11-05-6","published_at":"2026-11-05T05:07:12.000Z","score":34.358,"event_source_count":0},{"rank":17,"id":"blog:together:30689b894f03e4c2","title_zh":"Together AI 推出 Together Compiler，24 小时内最高提升推理吞吐 2.6 倍","summary_zh":"Together AI 用 agent 自动调优推理引擎，把实验吞吐提升 252 倍，前沿模型 TPS 最高涨 2.6 倍，说明编译器式持续优化能跟上流量漂移。","source":"Together AI","url":"https://www.together.ai/blog/together-compiler-continual-compilation-for-ai-inference","published_at":"2026-11-04T00:00:00.000Z","score":34.133,"event_source_count":1},{"rank":18,"id":"blog:techcrunch:22ee7d4fae222b3f","title_zh":"Hugging Face 政策负责人谈开源模型的成本、风险与取舍","summary_zh":"Hugging Face 政策负责人拆解开源 AI 的商业账：谁在为免费模型付费、中美开源模型竞争，以及开放生态怎么兼顾安全。","source":"TechCrunch","url":"https://techcrunch.com/podcast/building-on-open-source-ai-hugging-faces-policy-chief-on-the-costs-risks-and-trade-offs","published_at":"2026-11-04T18:44:57.000Z","score":31.194,"event_source_count":1},{"rank":19,"id":"podcast:latent-space:8dbeb9541e7819d7","title_zh":"研究显示 LLM 可直接读写神经网络权重，RSI 循环或不再依赖自动研究流程","summary_zh":"模型跳过训练流程，直接读写另一个模型的权重；“权重空间编程”若成立，递归自我改进（模型自主训练更强模型）的路径会明显缩短。","source":"Latent Space","url":"https://latent.space/p/ainews-llms-can-read-write-neural-network-weights-weight-space-programming-cuts-autoresearch-out-of","published_at":"2026-11-03T05:35:00.000Z","score":28.504,"event_source_count":1},{"rank":20,"id":"blog:the-verge:5e28c5f93eef89fa","title_zh":"亚马逊 Bee Frontier 体验：当 AI 替你过滤对话与记忆","summary_zh":"长期佩戴持续录音的 AI 挂坠后，作者发现摘要再准也无法替代人对情境的判断，也带出隐私与知情同意的现实问题。","source":"The Verge","url":"https://www.theverge.com/column/1026464/optimizer-amazon-bee-frontier-ambient-listening-wearables","published_at":"2026-11-04T18:00:00.000Z","score":23.697,"event_source_count":0}]}