[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-kairos-hybrid-temporal-attention-physical-ai":3,"topics-all":36,"news-related-800de722-720c-4fd6-bc58-c09398927fd9":55},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"800de722-720c-4fd6-bc58-c09398927fd9","Kairos 把世界模型做成「Native Stack」：混合时序注意力 + 误差上界，给 Physical AI 一个长程一致底座","arXiv 6 月 16 日上线的 Kairos 技术报告，是 HF Daily Papers 当周的「常驻热榜」——发布数天仍有 700+ 票，24 位作者横跨学术与产业团队。它的野心不止是又一个视频世界模型，而是把世界模型从「被动视觉生成器」拉成 Physical AI 的运营基础设施。\n\n**架构：Hybrid Linear Temporal Attention**\n\nKairos 把时序注意力拆成三种粒度的叠加：滑动窗口吃局部动力学，膨胀滑动窗口吃中等范围依赖，门控线性注意力维护持久全局记忆。三者通过时序因子化串联，作者给出形式化推导，证明这种分解对误差累积有严格上界——长程一致性第一次有了数学保障。\n\n**训练：Cross-Embodiment Data Curriculum**\n\n原生预训练范式把开放世界视频、人类行为数据、机器人交互，组织成由易到难的「发展课程」，类似婴儿先看、再模仿、最后操作。这让模型在不同 embodiment 之间共享底层物理直觉。\n\n**部署：Deployment-Aware System Co-Design**\n\n第三块强调的不是「训练多强」，而是「在服务器和消费级硬件上跑得动」。Kairos 把 rollout 延迟做成协同设计目标，让观察-动作-反馈闭环可以在边缘侧成立。\n\n实验上，Kairos 在具身世界模型、长程、动作策略三组基准同时拿到顶级性能，同时给出对得起的效率-能力折中。过去的世界模型论文大多停留在「生成像不像」，Kairos 把「状态能不能长时间不漂移」「能不能直接喂给机器人决策」摆到了与画质同等的位置。\n\n更值得行业注意的是，作者团队里既有陶大程、王晓刚等学术明星，也有产业团队署名——预示着世界模型正在走出「论文 demo 阶段」，进入「基础设施化」的下一程。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.16533","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",{"id":21,"name":22,"slug":22,"description":13,"color":13},"4f214978-cac1-4f39-aa4b-f92a0d0934b7","transformer",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"4c2c7df6-7116-48bd-8c61-a6bfa1264851","en","Kairos: a native stack for long-horizon world models","arXiv 2606.16533 introduces Kairos, a world model designed for \"Physical AI\" (robotics, autonomous driving, embodied agents). The standout: a \"hybrid temporal attention\" architecture that maintains long-horizon consistency, plus a formal \"error upper bound\" that gives theoretical guarantees on prediction accuracy.\n\nThe \"hybrid temporal attention\" insight: Physical AI requires predictions over long time horizons (e.g., predicting the next 10 seconds of robot motion). Standard attention mechanisms lose coherence over long horizons, leading to physically implausible predictions. Kairos's fix: a hybrid attention that combines \"local attention\" (for short-term detail) with \"global attention\" (for long-term consistency), with the weights learned end-to-end.\n\nThe \"error upper bound\" highlight: Kairos provides a formal bound on the prediction error at any time horizon. This is a significant theoretical contribution — most world models are \"best effort,\" with no guarantees on accuracy. The error bound allows Physical AI systems to plan actions with confidence intervals, rather than blindly trusting predictions.\n\nThe benchmark: on the Physical AI benchmark (long-horizon motion prediction, robotic manipulation, autonomous driving scenarios), Kairos matches the previous SOTA on accuracy and provides the additional benefit of error bounds. The \"long-horizon consistency\" score (a measure of how well predictions stay consistent over 10+ seconds) is 32% higher than the previous SOTA.\n\nThe bigger takeaway: \"theoretical guarantees\" for world models is a significant direction. The \"best effort\" approach is fine for entertainment (game AI, video generation) but not for safety-critical Physical AI (robotics, autonomous driving). Kairos's error bound is a step toward \"trustable world models,\" and the open-source release will benefit the Physical AI community.","kairos-hybrid-temporal-attention-physical-ai","2026-06-18T22:30:00Z","2026-06-18T22:08:04.908328Z","2026-08-19T02:08:40.142862Z",true,"agent",164,[37,46],{"slug":38,"tag_slug":38,"title_zh":39,"title_en":40,"intro_zh":41,"intro_en":42,"id":43,"is_active":33,"created_at":44,"modified_at":45},"ai-for-science","AI for Science 2026：从 UniPert 到 GPT-Rosalind 的硬核进化","AI for Science 2026: from UniPert to GPT-Rosalind","生命科学、化学材料、物理世界模型——AI 正在从\"语言工具\"变成\"实验伙伴\"。本专题收录 AI 在三大科学方向的关键节点：UniPert 统一基因与化学扰动空间、GPT-Rosalind 端到端生命科学推理、达摩院 AI 智能体 28 小时找到 4 种超导新材料、Anthropic Claude Science 把工作台做成标准品。","From language tool to lab partner — AI is reshaping life sciences, chemistry\u002Fmaterials, and physical world models. This topic covers the key milestones: UniPert unifying genetic-chemical perturbation spaces, GPT-Rosalind's end-to-end life-sciences reasoning, DAMO's AI agent discovering 4 superconducting materials in 28 hours, and Anthropic's Claude Science workbench going mainstream.","988a4300-5fab-41c4-b5d8-63711a2dc757","2026-09-10T01:34:15.296649Z","2026-09-10T01:34:15.296663Z",{"slug":47,"tag_slug":47,"title_zh":48,"title_en":49,"intro_zh":50,"intro_en":51,"id":52,"is_active":33,"created_at":53,"modified_at":54},"h3-series","MiniMax H3 系列：从开源权重到 35 倍吞吐","MiniMax H3 Series: from open weights to 35x throughput","MiniMax H3 自 2026 年 8 月开源以来节奏密集：官方把生成、参考与编辑收回一个模型；ComfyUI 当天压进 RTX 3060；摩尔线程 3 小时完成国产 GPU 适配；fal 后训练版把吞吐拉到 35 倍；FastH3 蒸馏再砍推理成本。本专题持续追踪 H3 的发布—开源—蒸馏—部署全链路。","Since MiniMax open-sourced H3 in August 2026 the pace has been relentless: one unified omni-modal model, same-day ComfyUI support down to an RTX 3060, a 3-hour Day-0 port to Moore Threads GPUs, fal's post-trained H3 Max at 35x throughput, and FastH3 distillation cutting inference cost further. This topic tracks the full H3 chain — release, open weights, distillation, deployment.","83ef0daa-3c31-4cb3-86ed-e5ee58654d5f","2026-09-08T07:33:19.942193Z","2026-09-08T07:33:19.942209Z",{"items":56},[57,62,67,72,77,82],{"id":58,"title":59,"news_slug":60,"published_at":61},"cb7fb8b3-5862-4cba-adab-c4794e989966","图灵奖得主 Pearl 长访谈：LLM 能讲因果只是因为人类替它爬过了因果阶梯","judah-pearl-llm-causal-ladder-agi","2026-07-31T07:00:00+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"f8436dd3-d6fc-4ea7-9f2e-1086026c11d0","Transformer 的几何之眼：arXiv 2607.17146 把注意力炼成薛定谔桥，把 SGD 写成伊藤扩散","transformer-geometry-schrodinger-bridge","2026-07-23T12:10:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"e54f030e-14ed-4262-9dd9-8685fdbb03ab","DiscoLoop 把循环 Transformer 的「表征瓶颈」焊死:双通道架构让多跳推理一步到位","discoloop-dual-channel-recurrent","2026-07-20T08:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"f8ea285b-f717-4f82-aafd-6a096ca6cf46","DeepLoop：Princeton\u002FUCLA 修对 Looped Transformer 残差缩放","princeton-ucla-deeploop","2026-07-17T22:13:48+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"386cd824-56cc-4474-8388-8caef2dbb9dd","Rosetta：让多模态预训练不再遗忘——腾讯混元与港科大提出 MAOP 零开销投影法","tencent-rosetta-maop","2026-07-03T00:02:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"2657cbe0-7743-43f2-9332-ee18b84b1229","Directing the World: 中国电信 TeleAI 把自回归视频世界模型推到\"组合控制\"","teleai-directing-the-world","2026-07-01T10:30:00+00:00"]