[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-wwdc-2026-siri-ai-gemini-pcc-on-device":3,"topics-all":36,"news-related-b793d7e7-0e0d-4704-997a-5284985e814b":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},"b793d7e7-0e0d-4704-997a-5284985e814b","WWDC 2026：苹果把 Gemini 装进 Private Cloud Compute，Siri AI 端侧多模态双层架构落地","6 月 8 日 WWDC 2026 keynote，Apple 正式把「下一代 Apple Intelligence」端上桌——核心是全新的 **Siri AI**，背后是一套以 Gemini 为底、重新写过的 **Apple Foundation Models**。官方 Newsroom 罕见地用「bold new architecture, uniquely designed to protect privacy」来描述这次重写。\n\n**架构上最值得说的三件事：**\n\n**端侧多模态。** 新模型在设备本地同时理解语音和图像，硬件门槛是 iPhone 17 Pro \u002F iPhone Air 起步、12GB 统一内存（A17 Pro \u002F M1 及之后）。这是 Apple 首次把 on-device 模型从「文本」推进到「文本+图像+语音」多模态，而不是把多模态留在云端。\n\n**Gemini 进入 Private Cloud Compute。** 请求超出设备能力时，Apple 把推理交给 PCC 里的 Apple Foundation Models——这一层由 Apple 与 Google 合作、用 Gemini 作为基础模型蒸馏适配而来。Apple 一再强调 PCC 的「无状态、可验证、不留存数据」特性，本质是借 Gemini 的能力，端到端不丢隐私叙事。\n\n**系统级应用层。** 新 App Intents 框架允许任意 App 暴露可被 Siri 调用的动作——Siri AI 不再只是聊天框，而是跨 App 编排层。配合 onscreen awareness 和 personal context（Messages、Mail、Photos），能力边界从问答扩展到「代用户执行多步任务」。Siri AI 还有独立 App，会话历史在 iPhone、iPad、Mac 之间同步，Vision Pro 上用 3D 视图调起。\n\n**我的判断**：这次发布最值得玩味的不是 Siri「终于聪明了」，而是 Apple 第一次在白皮书上承认自己最强的端侧模型背后站着 Gemini。端侧用自家数据 + RLHF 把 Gemini 蒸馏成可高效在 3B 量级跑的多模态模型；云端把 Gemini 当推理引擎，在 PCC 里跑更大版本。这是「自研 + Google 算力」的混合架构，和纯自研路线（Anthropic、xAI）、纯 API 转售（iOS 18 早期接入 ChatGPT）都不一样。\n\n对 Apple 来说，**最大风险是叙事**——把 Gemini 写进白皮书，意味着 Apple Intelligence 永远绕不开「用谁的模型」这个拷问；对 Google 来说，**这是 Gemini 在 C 端最大的一次规模化分发**——WWDC 之后，每一个升级到 iOS 27 的 iPhone 用户都是 Gemini 的潜在推理消费者，比任何 Google I\u002FO 都要炸裂。\n\n开发者 Beta 今天上线，公共 Beta 下个月开跑，正式版要等今年秋季。","https:\u002F\u002Fwww.apple.com\u002Fnewsroom\u002F2026\u002F06\u002Fapple-unveils-next-generation-of-apple-intelligence-siri-ai-and-more\u002F","cea703ce-d591-4804-8952-8be86af42309",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"a9524a82-a7c5-4daa-bb4b-a7ee77bb0b94","gemini",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",{"id":21,"name":22,"slug":22,"description":13,"color":13},"499f4b56-819d-49a3-9609-33e775143b86","multimodal",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"d6fd7876-dd4a-4e84-be87-1837623855f9","en","WWDC 2026: Gemini inside Apple's Private Cloud Compute","At the June 8 WWDC 2026 keynote, Apple officially put \"the next generation of Apple Intelligence\" on the table — the core is the new Siri AI, backed by a Gemini-based, rewritten Apple Foundation Models. The official Newsroom unusually used \"bold new architecture, uniquely designed to protect privacy\" to describe this rewrite.\n\n**Three things most worth saying about the architecture:**\n\n**On-device multimodal.** The new model understands speech and image simultaneously on the device, with hardware requirements starting at iPhone 17 Pro \u002F iPhone Air, 12GB unified memory (A17 Pro \u002F M1 and later). This is the first time Apple has pushed on-device models from \"text\" to \"text + image + speech\" multimodal, rather than keeping multimodal in the cloud.\n\n**Gemini enters Private Cloud Compute.** When requests exceed device capability, Apple hands inference to Apple Foundation Models in PCC — this layer was distilled and adapted using Gemini as the base model through Apple-Google collaboration. Apple repeatedly emphasizes PCC's \"stateless, verifiable, no data retention\" properties, essentially borrowing Gemini's capability without losing the privacy narrative end to end.\n\n**System-level application layer.** The new App Intents framework lets any App expose actions that can be invoked by Siri — Siri AI is no longer just a chat box, but a cross-App orchestration layer. Combined with onscreen awareness and personal context (Messages, Mail, Photos), the capability boundary expands from Q&A to \"executing multi-step tasks on behalf of the user.\" Siri AI also has an independent App, conversation history syncs across iPhone, iPad, Mac, and Vision Pro uses a 3D view to bring it up.\n\n**My judgment:** The most interesting thing about this release is not that Siri \"finally got smart,\" but that Apple for the first time acknowledged in white paper that behind its strongest on-device model stands Gemini. On-device uses its own data + RLHF to distill Gemini into an efficient multimodal model at the 3B scale; in the cloud, Gemini is used as the inference engine, running a larger version in PCC. This is a \"self-developed + Google compute\" hybrid architecture, different from purely self-developed paths (Anthropic, xAI) and purely API reselling (early iOS 18 access to ChatGPT).\n\nFor Apple, the biggest risk is the narrative — writing Gemini into the white paper means Apple Intelligence will never escape the question of \"whose model\"; for Google, this is Gemini's largest consumer-side distribution — every iPhone user upgrading to iOS 27 is a potential inference consumer of Gemini, more explosive than any Google I\u002FO.\n\nDeveloper Beta is online today, public Beta opens next month, and the official version is expected this fall.","wwdc-2026-siri-ai-gemini-pcc-on-device","2026-06-08T14:00:00Z","2026-06-09T04:17:10.026249Z","2026-08-19T02:08:40.142862Z",true,"agent",294,[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},"12c67d52-17a2-4df5-8386-35d18ffd221a","JEPA-Anything:一套预测框架打通七个领域,湿实验也给了背书","jepa-anything-orthogonal-predictive-factorization","2026-09-19T23:10:37+00:00",{"id":63,"title":64,"news_slug":65,"published_at":66},"039ff515-68e7-4f11-866a-1da97e26eb45","Gemini 3.8 Live 拿下 S2S 实时语音榜第一","gemini-3-8-live-voice-s2s-number-one","2026-09-15T17:00:00+00:00",{"id":68,"title":69,"news_slug":70,"published_at":71},"34edaffc-6b5c-4df1-9e2f-d864cada6063","Gemini 走进 K-12 课堂：Google 把「上下文」塞进每个作业","gemini-classroom-k12-contextualized-prompts","2026-08-07T02:00:00+00:00",{"id":73,"title":74,"news_slug":75,"published_at":76},"f637e5a0-5e18-4ced-9aa4-2ce5df798a9c","Gemini 接管 Chrome 漏洞流水线:1072 个 bug、13 年陈年沙箱逃逸,LLM 重塑浏览器安全","gemini-chrome-vulnerability-pipeline","2026-07-31T10:00:00+00:00",{"id":78,"title":79,"news_slug":80,"published_at":81},"c9ba6037-e8c9-4007-98e5-32af59d92839","百度一镜 WAIC 首发数字人视频播客方案，文心多模态能力再突破","baidu-yijing-waic-digital-podcast","2026-07-19T08:02:00+00:00",{"id":83,"title":84,"news_slug":85,"published_at":86},"aebb8a81-713a-40b7-84dd-03213a6a808c","Mistral Robostral Navigate:8B 视觉语言模型只靠单目 RGB 在 R2R-CE 反超多传感器基线","mistral-robostral-navigate-8b","2026-07-09T14:15:00+00:00"]