Goldman Lands SpaceX IPO as Founder Funds Eye $60 Billion Gains

高盛拿下SpaceX IPO，创始人基金瞄准600亿美元回报 SpaceX即将由主承销商高盛牵头启动公开上市，预示着史无前例的风险投资回报；与此同时，Google I/O的产品发布表明，固定的企业活动日程与真正的AI模型突破之间的脱节正进一步加剧。 主持人John Coogan和Jordi Hays在节目开场讨论了一段网络疯传的人形机器人翻车视频，随后深入解析了Google I/O大会的多项发布，包括与Warby Parker和Gentle Monster合作推出的智能眼镜、Gemini模型更新以及Genie 3世界模拟。他们详细探讨了开发者对Google渐进式AI更新的失望情绪，并提及一份显示Google员工正在使用OpenAI旗下Codex的内部爆料。节目最后盘点了SpaceX选定高盛牵头其庞大IPO的消息、Nvidia破纪录的财报业绩，以及Steve Wozniak对AI的风趣见解。 Google 宣布与三星、Warby Parker 及 Gentle Monster 合作推出 AI 智能眼镜；尽管 Warby Parker 拥有成熟的线下分销网络，其股价在发布后仍下挫 14%。 (2:13) Google 推出基于街景海量空间数据模拟交互式 3D 环境的世界模型 Genie 3，并计划利用 Workspace 的办公生态赋能智能眼镜，建立不同于 Meta 社交取向的生产力优势。 (11:35) 开发者社区对 Google I/O 的 AI 模型更新反应冷淡，在 Cursor 基准测试中，新 Gemini 模型的表现不及 Composer 2，且运行成本约为后者的四倍。 (15:15) SpaceX 推进 IPO 进程并出人意料地选择高盛作为主承销商，早期投资机构 Founders Fund 与 Valor Equity Partners 预计将斩获超 600 亿美元的创纪录回报。 (22:22) 英伟达单季营收同比激增 85% 达到 816.2 亿美元，净利润攀升至 429.6 亿美元，CEO 黄仁勋将全球“AI 工厂”的兴建视为史上最大规模的基础设施扩张。 (24:35)

TBPN hosts John Coogan and Jordi Hays analyze Google I/O's latest hardware and AI updates, arguing that Google's core productivity ecosystem remains its strongest asset despite developer frustration with incremental model improvements and competing capital market milestones. Smart Glasses Partnerships and Workspace Integration (2:13) Google unveiled AI-infused smart glasses developed in partnership with Samsung, Gentle Monster, and Warby Parker for a planned fall release. Design approaches differed noticeably: Gentle Monster concealed its internal components cleanly, whereas Warby Parker included a visible camera bump. Despite Warby Parker's stock falling 14% after the announcement, its market capitalization stands at roughly $3 billion, down from $6 billion in 2021 but up 43% over the prior six months. Coogan and Hays argue that Google's core structural advantage over Meta's Ray-Ban line is enterprise and personal productivity integration. While Meta pairs glasses with social messaging platforms like WhatsApp and Instagram, Google can link Gemini directly to central personal workflows across Gmail, Docs, and Drive. However, the hosts note this utility is continually obscured by convoluted branding across overlapping labels such as Gemini Advanced, Workspace, AI Pro, and AI Ultra. Developer Disappointment and Gemini Benchmark Lags (15:15) Developer sentiment following Google I/O was largely underwhelmed. Rather than launching a new pre-trained base model like Gemini 2.0, Google showcased incremental iterations like Gemini 1.5 Flash. Coogan and Hays observed that rigid corporate event calendars force tech giants to ship whatever is ready, unlike independent research labs that coordinate events around finished breakthroughs. Further compounding developer skepticism, commentator Ben Thompson remarked that Google appeared to be 'shipping their org chart.' Performance metrics also faced scrutiny. On the Cursor benchmark, the announced Gemini model trailed Cursor's Composer 2 while costing roughly four times as much to run—reversing Google's traditional strategy of offering near-frontier models at aggressive pricing. Additionally, tech commentator Gergely Orosz noted that Google's Anti-Gravity team inadvertently exposed a folder named 'Codex' in their launch video, reviving debates regarding whether internal Google teams rely on OpenAI or Anthropic models for their own workflows. Spatial AI Datasets and Video Generation Limits (11:35) On world-generation technologies, Google highlighted Genie 3, which simulates interactive 3D environments grounded in Google Street View imagery. Coogan emphasized that while YouTube is widely recognized as a training ground for models like Veo, Street View functions as an untapped mother lode of physical spatial data for interactive simulations. In video generation, Google’s Omni Flash was contrasted with China's SeaDance 2.0, with the hosts pointing out that Chinese models operate under looser intellectual property constraints. While Omni Flash showed high technical capability despite minor physical errors—such as omitting cylinders on a rendered V8 engine—the hosts emphasized that transitioning from short 10-second clips to coherent 10-minute explanatory videos remains fundamentally limited by heavy computational costs and long-horizon logical consistency. SpaceX IPO Mandate and Nvidia Earnings Surge (22:22) Beyond Google, the hosts reviewed major corporate and financial developments. SpaceX is reportedly preparing for an initial public offering with a prospectus possible as early as late May. In an unexpected development, Goldman Sachs was selected as the lead left underwriter over Morgan Stanley's Michael Grimes, who has handled past Elon Musk transactions. Scoop reporter Katie Roof projected unprecedented venture windfalls: Founders Fund and Valor Equity Partners are poised for over $60 billion in combined gains, while Sequoia Capital and D1 Capital Partners each stand to realize over $20 billion. Simultaneously, Nvidia demonstrated massive hardware demand in its financial reporting. Quarterly revenue surged 85% year-over-year to $81.62 billion, while net income climbed to $42.96 billion from $18.8 billion in the prior-year quarter. CEO Jensen Huang framed the global buildout of AI factories as the largest infrastructure undertaking in history. Google maintains formidable structural moats through Google Workspace and exclusive spatial datasets like Street View, yet its calendar-driven model releases continue to generate friction among developers expecting major technical leaps, even as wider AI infrastructure spending scales to record highs.

TBPN 主播 John Coogan 与 Jordi Hays 审视了 Google I/O 大会展示的智能眼镜生态与模型迭代，指出其硬件生态虽具工作流整合优势，但底层模型进展相对温和；同时节目探讨了高盛牵头 SpaceX 潜在 IPO 带来的历史性风投回报，以及 Nvidia 财报所反映的全球 AI 基础设施激增需求。 智能眼镜硬件合作与工作流生态优势 (2:13) Google 宣布与三星、Gentle Monster 及 Warby Parker 合作，计划于秋季推出结合 AI 的智能眼镜。在外观工艺上，Gentle Monster 较好地隐藏了内部硬件，而 Warby Parker 则带有较为明显的侧边摄像头凸起。尽管发布后 Warby Parker 股价出现 14% 的下滑，其市值仍维持在约 30 亿美元。与 Meta 依赖 Luxottica 并在 WhatsApp、Instagram 等社交渠道整合不同，Google 拥有 Gmail、Google Docs 和 Google Drive 等核心生产力入口。主播认为，这种与个人及办公中枢的深度绑定构成了 Google 的结构性分发优势，尽管外界同时批评 Spark、Gemini Advanced 等重叠产品命名过于繁杂。 世界模型探索与开发者对模型的平淡反响 (11:35) 在数据探索层面，Google 推出了基于街景数据的 Genie 3 世界模型，展示了构建可交互 3D 环境的潜力。然而在核心大模型领域，开发者整体情绪偏向平淡甚至失望。相较于市场预期的全新 Gemini 2.0 基座模型，大会推出的 Gemini 1.5 Flash 更多被视作架构上的微调。主播分析指出，大型科技公司受制于提前锁定的发布会时间表，容易在模型未成熟时仓促交卷，而 OpenAI、Anthropic 等独立实验室则倾向于等待突破完成后再发布。Cursor 基准测试显示新模型落后于其自身的 Composer 2，且运行成本高出近四倍；此外，分析师 Ben Thompson 也指出 Google 存在按组织架构而非整合逻辑推产品的现象。 内部工具风波与长视频生成的技术瓶颈 (18:39) Google 官方宣传视频中意外露出名为 Codex 的文件夹，引发了 Gergely Orosz 等科技评论者关于 Google 工程师是否在内部依赖竞品工具的讨论，呼应了此前业内对 DeepMind 内部工具使用效率的争议。而在视频生成领域，针对 Google Omni Flash 与中国 SeaDance 2.0 等模型的讨论显示，国内平台在过往可能受益于较为宽松的版权环境限制。主播指出，虽然当前的 10 秒短片段生成视觉效果显著，但要生成 6 到 10 分钟具有严密内在逻辑且可替代 YouTube 内容的深度讲解视频，在算力开销和逻辑连贯性上仍面临巨大挑战。 SpaceX 潜在上市回报与芯片需求激增 (22:22) 资本市场方面，高盛意外击败长期协助 Elon Musk 的摩根士丹利，被选为 SpaceX 潜在 IPO 的主承销商。据 Katie Roof 报道，早期风投机构将迎来历史性变现，其中 Founders Fund 与 Valor Equity Partners 预计总收益超过 600 亿美元，Sequoia Capital 与 D1 Capital Partners 各自的账面回报也将超过 200 亿美元。与此同时，Nvidia 公布的单季营收同比增长 85% 至 816.2 亿美元，净利润达 429.6 亿美元，CEO 黄仁勋将这一轮 AI 工厂建设定义为人类历史上最大规模的基础设施扩张。在技术热潮之余，Apple 联合创始人 Steve Wozniak 则在高校演讲中以幽默方式提醒公众关注真实的人类智能（Actual Intelligence）。 本期节目突显了大型科技公司在应对定期产品发布与颠覆性科研突破脱节时的压力。随着 AI 从短视频演示走向长逻辑生成，大厂在生态护城河与基座模型性价比之间的权衡仍悬而未决；而资本市场对顶尖算力硬件与龙头航天资产的极端溢价，则反映出市场对技术基础设施变现确定性的高度依赖。

## Background 本期内容由科技对谈频道 TBPN 于 2026 年 5 月 23 日发布在 YouTube 平台。作为其 30 分钟精简版栏目 Diet TBPN，节目由 John Coogan 和 Jordi Hays 共同主持（被《纽约时报》誉为“硅谷的新痴迷”），聚焦复盘 Google I/O 首日发布反响、高盛领衔 SpaceX IPO 的重磅传闻，以及英伟达最新财报等硅谷前沿动态。 ## TL;DR 节目首先深入剖析了 Google I/O 首日看点，对比了 Google 联手 Gentle Monster 与 Warby Parker 打造的智能眼镜与 Meta Ray-Ban 的生态及审美差异，并犀利指出 Gemini Flash 3.5 在 Cursor Bench 上的性价比倒退与 Google 混乱的产品命名体系。随后，主持人们梳理了 Genie 3 街景世界模型、Anti-gravity 演示穿帮出现 Codex 文件夹的趣闻，以及 Omni Flash 与字节 SeaDance 2.0 在视频生成细节与版权壁垒上的博弈。节目最后重点解读了高盛爆冷挂帅 SpaceX IPO 主承销商为 Founders Fund 等机构带来的数百亿美元历史级风投回报，并复盘了英伟达单季近 430 亿美元净利润的夸张财报与沃兹尼亚克的幽默演讲。 ## Atmosphere & style 全片洋溢着浓厚且快节奏的“科技界 ESPN”体育解说风格，两位身着同款西装的主持人默契十足、接梗自如，在毒舌吐槽与深度行话之间无缝切换。从开场拿机器人失控摔倒开涮，到毫不留情地讽刺科技巨头脱离群众的“AI 演示疲惫症”与内部混乱，谈话充满硅谷圈内人的辛辣幽默。而在面对 SpaceX 惊人的百倍早期回报与英伟达堪比印钞机的财务数字时，两人的情绪又迅速拉满，展现出对顶级商业与科技叙事的纯粹兴奋，整场对谈既硬核透彻又极具娱乐观赏性。 ## Bulletin - [22:20](https://www.youtube.com/watch?v=BYITT17BJjI&t=1340s) SpaceX 预计最快提交 IPO 招股书，高盛爆冷击败摩根士丹利拿下主承销商（Lead Left）席位；早期投资机构 Founders Fund 与 Valor 预计分别斩获超过 600 亿美元的惊人账面回报。 - [07:24](https://www.youtube.com/watch?v=BYITT17BJjI&t=444s) Google 宣布联手 Gentle Monster 与 Warby Parker 进军智能眼镜，但由于缺乏即时催化剂且新品需等到秋季，Warby Parker 股价在消息公布后暴跌 14%。 - [16:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=966s) 在第三方评测 Cursor Bench 中，新发布的 Gemini Flash 3.5 评分不敌 Composer 2 且调用成本贵了近四倍，Google 长期依赖的极致性价比优势遭遇剧烈倒退。 - [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) 英伟达最新季度净利润同比翻倍以上飙升至 429.6 亿美元，季度营收跃升至 816.2 亿美元，黄仁勋宣称智能体 AI 时代的基础设施扩张正在全球加速落地。 ## Key takeaways - [07:24](https://www.youtube.com/watch?v=BYITT17BJjI&t=444s) 智能眼镜的早期 PMF（产品市场契合度）将优先在视力矫正刚需人群中建立，Warby Parker 庞大的线下验光渠道与客群是科技巨头梦寐以求的分发资产。 - [05:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=306s) 在工业设计取向上，Gentle Monster 采用隐藏式摄像头更具时尚隐蔽性，而 Warby Parker 的明显凸起遭遇审美考验，二者与 Meta Ray-Ban 的经典 Wayfarer 框型展开正面竞争。 - [09:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=546s) 虽然 Google 内部存在 Gemini、AI Pro、AI Ultra 及 Spark 等混乱的产品命名矩阵，但其 Gmail 与 Docs 生产力资产构成了比 Meta 纯社交关系链更坚固的智能体壁垒。 - [11:34](https://www.youtube.com/watch?v=BYITT17BJjI&t=694s) Google 正将数十年累积的街景（Street View）多维真实世界图像转化为构建 Genie 3 等物理世界模型的核心稀缺燃料，开辟了超越单纯视频训练集的新维度。 - [13:12](https://www.youtube.com/watch?v=BYITT17BJjI&t=792s) 消费级头显硬件市场进入产品饥渴期，市场更期待保留高质量微显示屏并大幅减重降价的轻量化“Apple Vision Air”，而 Meta Orion 等高成本原型机短期内仍难以大众量产。 - [14:26](https://www.youtube.com/watch?v=BYITT17BJjI&t=866s) Gemini Flash 3.5 反响平平折射出大厂固定日程与前沿大模型收敛周期的内在错位，传统开发者大会难以精准匹配科研突破的实际节奏。 - [16:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=966s) Gemini Flash 3.5 在 Cursor Bench 上的高价低效表现动摇了 Google 的竞争基本盘，这也从侧面印证了 Google 积极对外兜售 TPU 算力集群并重金下注 Anthropic 的底层逻辑。 - [17:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1045s) 现场演示通过眼镜语音在天空中生成卡通飞艇照片引发关于“AI 疲劳症”的反思，科技巨头普遍存在为了做 Demo 而设计功能、脱离大众日常真实痛点的倾向。 - [18:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1105s) Google Anti-gravity 演示视频中惊现 Codex 文件夹，侧面印证了硅谷科技巨头内部工程团队普遍混合使用竞争对手工具与自研模型的现实。 - [19:56](https://www.youtube.com/watch?v=BYITT17BJjI&t=1196s) AI 视频模型（如 Omni Flash 与 SeaDance 2.0）短片段拟真度虽已逼近极限，但在真实物理逻辑（如发动机点火时序）以及长篇复杂视频一致性上仍面临高昂算力与机制瓶颈。 - [22:20](https://www.youtube.com/watch?v=BYITT17BJjI&t=1340s) SpaceX IPO 预计为早期坚定持有的风投机构带来数百亿美元的史诗级回报，用实打实的战绩反击了外界对于超级基金规模过大导致缺乏超额回报的偏见。 - [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) 史蒂夫·沃兹尼亚克在演讲中以“真正的人类智能（Actual Intelligence）”幽默化解极客圈对 AGI 的过度炒作，展现了更具亲和力的科技沟通方式。 ## Detailed walkthrough ### [00:02](https://www.youtube.com/watch?v=BYITT17BJjI&t=2s) 西装造型、ESPN 风格与机器人摔倒插曲 节目一开始，两位主持人 John Coogan 和 Jordi Hays 便就今天的着装互相打趣。两人今天双双穿上了西装，观众常调侃分不清他们两个谁是谁，甚至有人开玩笑建议 Jordi 以后只要不穿西装上节目，就得往“脏话罐”（swear jar）里扔罚款——对此 John 提议把每次罚款设为 20 美元直接捐给 OpenAI 的非营利机构。随后他们提到媒体 UnHerd 记者 Alice Key 近期发表的一篇长文，该文深入分析了 TBPN 等科技直播节目为何正在复刻 ESPN 创立的体育媒体打法，并取得了巨大成功。文中还幽默地吐槽两人“穿着同款西装、留着蓬松的发型，实在让人难以分辨”。他们笑称平时常被当成亲兄弟，通常是 Jordi 穿得更随性一些，但这种“兄弟”标签其实是观众的一种喜爱。 在切入正题讨论 Google I/O 前，两人插播了一段人形机器人的滑稽视频。画面里机器人本来步伐矫健，途中稍有踉跄还成功稳住了重心，显得十分矫健自如。然而下一秒，它突然失控直接面朝地面重重摔倒在地，现场悲伤的背景音乐甚至都没来得及关掉，机器人最终只能被工作人员尴尬地抬出场外。两位主持人被逗得大笑，调侃道：“这奇点来得有点软绵绵啊”，并笑称这个机器人可能是“尴尬致死”，或者是脸着地时硬件被彻底摔坏了。 --- ### [02:18](https://www.youtube.com/watch?v=BYITT17BJjI&t=138s) Google I/O 核心看点与 D2C 品牌沉浮录 收起笑意后，节目正式进入 Google I/O 的重磅复盘。团队成员 Brandon Gral 在 TBPN 的 Newsletter 中将本次大会的核心亮点归纳为四大板块：智能眼镜（Intelligent Eyewear）、Gemini Omni 演示、Gemini LLM 开发者反响以及 Anti-Gravity。讨论首先聚焦于智能眼镜生态，Google 宣布正与三星（Samsung）、Gentle Monster 以及 Warby Parker 联手打造全新智能眼镜。这引出了两人对眼镜品牌与 D2C（直接面向消费者）商业模式的探讨。 提到 Warby Parker，Jordi 坦言自己很久不戴眼镜，但一直非常欣赏该品牌的品牌塑造理念及其商业拓展能力。他们回顾了当年轰轰烈烈的 D2C 浪潮：Warby Parker、Allbirds 和 Everlane 曾是这一赛道的标志性三巨头。然而如今后两者的市值均已跌破 1 亿美元，Allbirds 甚至一度跌至仅剩 2000 万美元市值左右。虽然后续 Allbirds 借由转型所谓的“Neo Cloud”搞 AI 概念迎来一波拉升，但两人犀利吐槽这纯属炒作，甚至调侃英伟达财报电话会上黄仁勋是不是还得提一嘴 Allbirds 到底有没有拿到 GB200 或 Cerebras 算力配额。相比之下，Warby Parker 虽然市值从 2021 年高点的 60 亿美元跌至如今约 30 亿美元，但在独立 D2C 品牌大批凋零或被迫卖身的背景下，依然展现出了极强的韧性。 --- ### [05:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=306s) 巨头眼镜暗战：Warby Parker、Gentle Monster 对标 Meta Ray-Ban Google 此次公开了与 Gentle Monster 和 Warby Parker 合作的两款秋季新品设计，引发了网络上的热烈讨论，科技博主们戏称“现在可以把喜欢的 Mag 7 巨头戴在脸上了”。两位主持人细致对比了两家产品的外观工业设计： ```mermaid flowchart TD accTitle: 智能眼镜生态与设计对比 accDescr: 展示 Google、Meta 与传闻中 Apple 在智能眼镜设计取向与合作伙伴上的不同路径 subgraph GoogleCamp ["Google 阵营"] G["Google Gemini"] --> GM["Gentle Monster<br/>(隐藏式镜头/时尚隐蔽)"] G --> WP["Warby Parker<br/>(镜头凸起明显/经典日常)"] end subgraph MetaCamp ["Meta 阵营"] M["Meta AI"] --> RB["Ray-Ban & Oakley<br/>(开创脸部计算认知/辨识度高)"] end subgraph AppleCamp ["Apple 阵营 (预期)"] A["Apple Intelligence"] --> AP["自主工业设计<br/>(不依赖第三方品牌语言)"] end ``` 从设计上看，Gentle Monster 巧妙地把摄像头完全隐藏在镜架结构中，从远处看丝毫不显眼，完全没有科技可穿戴设备的外露感；而 Warby Parker 的版本在镜框边缘则有一个极其明显的摄像头凸起（camera bump），显得较为突兀。Meta Ray-Bans 虽已在市场上培养了用户对“脸部电脑”厚镜腿的认知，但 Gentle Monster 这类隐蔽设计将带来更高的隐蔽性。 围绕科技巨头的眼镜阵营划分，两人援引了推特上的趣味调侃：Apple 绑定了卡尔·蔡司（Carl Zeiss），Meta 联合了雷朋（Ray-Ban）与欧克利（Oakley），Google 则联手 Gentle Monster 与 Warby Parker。网友更是开玩笑发问：“哪家巨头能大胆一点，把 AI 芯片直接塞进 3M 防护护目镜里？”主持人笑称自己非常愿意支持这种“全员赛博朋克、彻底机甲化”的硬核极客路线。 --- ### [07:24](https://www.youtube.com/watch?v=BYITT17BJjI&t=444s) 资本市场反应与智能眼镜的 PMF 逻辑 尽管达成了与科技巨头的重大合作，Warby Parker 的股价在消息公布后却意外暴跌了 14%，让著名投资人 Sheel Mohnot 等人大为不解。分析认为，下跌原因在于产品要等到秋季才正式推出，市场缺乏即时兑现的催化剂；同时，科技博主 Isaac 则指出配备了摄像头与麦克风、搭载 Gemini 的 Google AI 眼镜是真正冲着 Meta Ray-Ban 的腹地而来。 对于智能眼镜的市场前景，Jordi 提出了清晰的 PMF（产品市场契合度）观点： - **刚需转化优先**：智能眼镜必然先在“本身就需要全天佩戴光学矫正眼镜”的人群中实现 PMF。对于视力正常的人来说，要说服他们每天脸上多戴一个硬件极难，但对眼镜党而言，顺手加上智能功能则是极低门槛的顺水推舟。 - **渠道价值凸显**：Warby Parker 坐拥庞大的线下验光渠道与精准的视力矫正用户群，其 30 亿美元的体量未来极有可能直接成为 Google 等科技巨头为了补齐分发短板而收购的标的。 节目随后提及 Google I/O 上展示的个人 AI 助手 Project Astra / Spark 以及各类复杂命名，感叹 Google 内部永远有着数不清的产品线与庞杂生态。 ### [09:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=546s) 混乱的命名矩阵与生态壁垒：Google、Apple 与 Meta 的智能眼镜暗战 节目切入 Nathan Clark 在社交网络上引发热烈共鸣的一则爆笑热梗，精准嘲讽了 Google 内部混乱不堪的产品命名矩阵：你想用 Gemini，就得去 AI Studio；要是个人账号还行，切换到 Workspace 办公账号就必须订购 Gemini Business；可千万别选 Gemini Advanced，因为现在它已经改名叫 AI Pro 了——除非你真正需要的是 AI Ultra；而如果你想折腾智能体（Agents），实际上该去 Spark，而不是 Gemini API 管理后台。这个令人啼笑皆非的“全家桶迷宫”，恰好映射出科技巨头在生态集成上的根本差异。 主持人们进一步剖析了智能眼镜背后的底层生态掌控力。Meta Ray-Bans 固然深度整合了 WhatsApp、Instagram DMs 以及 Facebook Messenger，但除了部分以 WhatsApp 为全天绝对核心通讯工具的用户外，社交聊天软件根本无法承载大众日常生活的核心工作流。绝大多数人真正依赖的核心枢纽，始终是 Google Docs、Google Drive 以及 Gmail（在苹果生态中则是 Apple Mail、iMessage、Apple Files 和相册）。Meta 的智能眼镜由于缺乏企业级和生产力级套件支撑，其内置的 AI Agent 很容易在跨平台调度时撞上各家的“围墙花园”；相反，运行在 Google 或 Apple 体系内的智能体，却能无缝调度用户的整个数字生活，因而拥有高得多的实用价值。 然而，在外观设计维度上，Google 并没有占到便宜。科技投资人 Nick 甚至毫不留情地公开吐槽：“这款 Google 与 Warby Parker 联名的眼镜相比 Meta Ray-Bans 简直丑得灾难，相关团队恐怕要有人为此丢饭碗。”主持人们笑称虽然未必有那么夸张，但 Ray-Ban 标志性的 Wayfarer 框型历经数十年乃至上百年的品牌文化积淀，其经典地位绝非仅创立十余年的 Warby Parker 所能轻易匹敌。 ```mermaid flowchart TD accTitle: 智能生态与智能体集成能力对比 accDescr: 对比 Meta、Apple 与 Google 核心生态系统及其对 AI 智能体生产力支持的差异 subgraph Meta["Meta 封闭社交圈"] M1["WhatsApp / IG DMs"] --> M2["Meta Ray-Bans / AI"] M2 -.->|"受限于外部围墙花园\n缺乏核心生产力套件"| M3["受限的日常任务"] end subgraph Apple["Apple 硬件系统闭环"] A1["Apple Mail / iMessage / Files / 相册"] --> A2["系统级 AI Agent"] A2 --> A3["深度整合个人设备资产"] end subgraph Google["Google 生产力数据核心"] G1["Gmail / Google Docs / Drive 个人与企业资产"] --> G2["Gemini 生态与智能眼镜"] G2 --> G3["掌控高价值生产力与数字生活枢纽"] end ``` - **Google 命名迷宫梗**：涵盖 Gemini、AI Studio、Gemini Business、AI Pro、AI Ultra、Spark 等令人抓狂的重叠概念与订阅层级。 - **核心数据枢纽论**：相比单纯的即时通讯，用户日常最重要的文件、票据与日程几乎全盘托管在 Gmail、Docs 与本地系统盘中，构成 AI Agent 的真正护城河。 - **品牌沉淀的代差**：Wayfarer 等经典太阳镜轮廓代表了近百年的时尚资产，而 Warby Parker 的工业设计在消费级硬件美感上面临巨大审美考验。 --- ### [11:34](https://www.youtube.com/watch?v=BYITT17BJjI&t=694s) Genie 3 唤醒街景数据宝藏，从过山车呕吐代码到世界模型 谈及 Google I/O 带来的前沿技术，Genie 3 展现出极其惊人的想象空间：它现在能够直接基于 Google 街景（Street View）的实景全景图像作为现实锚点，动态模拟出真实的物理空间体验。主持人们感叹，外界此前普遍只盯着 YouTube 庞大的视频库，将其视作训练多模态模型 Omni 及后续 VO3、VO4 视频生成的绝对母矿，却严重低估了 Google 坐拥数十年全球街景数据这一深不可测的“现实世界数据富矿”。 这种对现实物理数据的极度渴望，折射出整个科技巨头圈的集体转向。DeepMind 掌门人 Demis Hassabis 以及诸如 Mark Zuckerberg（传闻曾记录所有 Meta 员工的屏幕使用习惯与操作日志）等科技领袖，如今都表现得极度“数据上头”（data-pilled）。街景数据这种不可替代的稀缺多维数据，正成为构建世界模型的核心燃料。 主持人们还津津乐道地聊起了 Demis 早年在游戏开发界的一段传奇趣事：Demis 骨子里流淌着模拟经营游戏的设计血液，早年在开发类似《主题公园》（Theme Park）的过山车模拟游戏时，他甚至曾亲自编写过一段用算法逻辑生成“游客在过山车上呕吐”的代码。主持人们探讨道，虽然目前 Genie 3 的街景生成还是令人惊艳的视觉演示，但未来它是否会开放给独立开发者作为底层世界引擎制作全新游戏？毕竟对硬核玩家而言，3A 画面固然抓人眼球，但真正让人废寝忘食的永远是那些精妙绝伦的游戏核心机制（Mechanics）。 --- ### [13:12](https://www.youtube.com/watch?v=BYITT17BJjI&t=792s) 渴望摆脱显示器：从 Meta Orion 原型到更轻的 Vision Air 直播互动区中，观众 Bobby Chipman 留言直呼：“真等不及智能眼镜能够彻底替代我桌上笨重的多台显示器了。”这一讨论随即引发了对 AR/VR 硬件演进瓶颈的复盘。目前的 Meta Ray-Bans 显示版本依然停留在类似《使命召唤》（Call of Duty）游戏 HUD 抬头显示界面的单色小浮层状态，远未达到全功能增强现实的及格线。 而回顾 Meta 此前高调展示的 Orion AR 原型机，虽然两位主持人都曾亲身体验过其震撼的实机演示——尽管视场角（FOV）依然相对狭窄，但它确实能够将一块清晰逼真的虚拟屏幕凭空放置在眼前。然而，业界普遍的声音仍然是 Orion 成本极度高昂、机械结构笨重，根本没有做好面向大众量产的商业化准备。如今一年多过去，整个硅谷的资本开支（CapEx）浪潮几乎全盘砸向了底层 AI 算力集群，AR 硬件项目似乎被悄然降级，连预期的夏季 Meta Connect 大会都鲜有后续爆料。 在头戴硬件层面，主持人们表达了普遍的“产品饥渴”，坦言市场早已急需一款好玩且适宜佩戴的全新爆品。对于 Apple Vision Pro，核心诉求其实极其统一：不需要做太多画蛇添足的改动，只要保留那块惊艳绝伦的高素质微显示屏，同时大幅削减机身配重、降低价格，打造出一款真正轻量化的“Apple Vision Air”，就能挽救当前繁琐佩戴带来的生理疲惫感。 --- ### [14:26](https://www.youtube.com/watch?v=BYITT17BJjI&t=866s) 令人失望的 Gemini Flash 3.5：固定日程与模型训练周期的错配 围绕 Gemini Flash 3.5 的实际表现，社区反馈却迅速陷入冷淡与失望。科技博主 Tennebris 等人评价其虽然响应速度极快，但整体依然只是 Google 一贯展现的那种挤牙膏式的微弱迭代（incremental progress），导致整场 Google I/O 显得颇为索然无味。市场本期待继展现出底层大模型全新底色（big model smell）的 Gemini 3 之后，能够在此次大会上见证下一代飞跃基座 Gemini 4 或者完整的 3.5 Pro，但现实中 Pro 版本依旧缺席。 主持人们深刻指出，这里暴露了顶尖前沿 AI 研究与传统大厂产品发布节奏之间的天然错位： 1. **死板的日历与不可控的收敛**：像 Google I/O 这种规格的全球开发者大会通常需要提前一到两年定下举办日期；然而，一个超大规模前沿基座的预训练周期何时能跑完、是否能按期平稳收敛，在科研上充满变量，根本无法像日历日程那样精准装盒打包。 2. **创业实验室的灵活打法**：OpenAI、Anthropic 以及 xAI 等独立 AI 实验室的做法截然不同——它们通常是模型彻底训练完毕并验证成熟后，才临时宣布上线，并围绕模型即兴补办一场发布会、录制一段演示视频或发布技术报告（Model Card）。 3. **数据截止日期的尴尬争议**：面对发布会上展出的模型，许多开发者死抠其知识截止日期（甚至停留在 2025 年初或更早）。尽管大模型具备实时联网检索修正能力，但这种停滞感依然加剧了外界对其并非真正代际革新模型的质疑。 --- ### [16:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=966s) Cursor Bench 惨败与性价比倒退：Google 为何押注 Anthropic 随着第三方开发者实测数据的密集出炉，Google 的新模型口碑迅速遇冷。知名 AI 代码编辑器 Cursor 在自家的 Cursor Bench 权威评测基准中录入了测试成绩，结果令人大跌眼镜：Gemini Flash 3.5 的综合评分不仅直接被 Cursor 自家的 Composer 2 压制，其调用调用成本反而足足贵了将近四倍之多。 主持人们犀利指出，这一跑分事件击中了 Google 长期以来的竞争基石： - 长期以来，Google 在 AI 模型领域的定位一直是“以最具杀伤力的极致性价比，提供接近或比肩第一梯队前沿模型的能力”。 - 如今性能未能实现越级，价格却暴涨四倍，宣告了其性价比策略的剧烈倒退。 - 这也让科技圈彻底看懂了背后的算盘——为什么 Google 会不惜投入天量资本与硬件基础设施资源，甚至反常地重金押注竞争对手 Anthropic。 与此同时，Perplexity 等信源流传的分析指出，Google DeepMind 当前在新模型研发上的最大制约其实早已不是算力瓶颈，而是受困于高质量数据的匮乏，这反过来解释了 Google 积极对外兜售自研 TPU 算力集群套现的商业动机。科技分析师 Ben Thompson 也一针见血地评价道：当下的 Google 显然又陷入了经典的“把公司内部混乱的组织架构图直接当成产品发布出来”（shipping their org chart）的泥潭。 --- ### [17:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1045s) 满载 AI 的审美疲劳：Gentle Monster 演示与脱离大众的现实痛点 大会引发的另一个深层命题，是对全线产品疯狂强塞 AI 造成的“AI 疲劳症”（AI fatigue）。AI 专家 Allie K. Miller 指出，整场 Google 主题演讲中欢呼声最响亮的时刻之一，竟然是舞台上的一段硬件交互秀：演示者 Nishtha 戴上与 Gentle Monster 联名的 Gemini 智能眼镜，轻敲镜腿唤醒助手并下达了一句复杂的全合一语音指令：“拍张照片，并在天空中加上一个写着‘Google I/O 2026’的卡通飞艇。”短短数秒之内，由后台多模态模型 Nano Banana 瞬间渲染合成的照片预览，便精准同步推送到她的智能手表表盘上。 这段主打纯语音驱动、旨在让用户减少看屏幕时间的联动演示技术确实过硬，但也引发了巨大的两极反响。知名数码博主 Greg's Gadgets 一针见血地讥讽道：“这帮科技巨头压根就不知道普通大众每天真正需要的是什么。” 主持人们对此深表共鸣：这种在天空中P上特定字样飞艇的桥段，是典型的“为了做演示而精心设计的产品演示”（demo of the product）。它在实验室舞台上显得极其丝滑、完美展示了端到端的功能链路，但在真实残酷的日常生活中，没有任何普通消费者会高频使用如此脱离现实的细分场景，巨头们依然缺乏真正打动凡夫俗子的杀手级生活创意。 ### [18:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1105s) Google I/O 花絮：Anti-gravity 演示惊现 Codex 文件夹与内部模型争议 在盘点完可穿戴设备和智能手表的创意亮点后，主播们聊到了 Google I/O 上引发推特热议的一幕戏剧性细节：Google 旗下的 Anti-gravity 团队在其发布视频中，竟然不小心露出了一个名为“Codex”的文件夹。知名科技博主 Gergely Orosz（The Pragmatic Engineer）在 X 上发文调侃道：“在 Anti-gravity 发布视频的第 2 分钟，我不得不倒回去反复看两遍。你能清晰看到 Anti-gravity 团队的人在使用 Codex。难道发布前连最基础的视频内容复核都没人做吗？” 这番曝光引来了主持人对界面和工具的探讨。有人吐槽 Anti-gravity 的界面设计看着就非常像 Codex；另一人则戏谑发问，它难道不更像是在“纯粹重写 Windsurf”吗？不过，大家一致认为这其实算不上真正的冷门新闻，毕竟 Google 内部早已大量采购和使用 Anthropic 的各类模型以及其他第三方的 AI 工具。这也让主持人联想到前阵子 Steve Yegge 与 Google DeepMind 负责人 Demis Hassabis 在社交媒体上的公开论战——Steve Yegge 当时爆料称 Google 员工内部普及和部署 AI 工具的效率参差不齐，而 Demis 则激烈反驳称“这完全是无稽之谈，现在公司上下全员都在用 AI”。 - Gergely Orosz 敏锐指出视频仅播放到第 2 分钟就出现了竞品工具 Codex 文件夹的穿帮镜头。 - 讨论触及硅谷大厂的通用现状：即便自研模型如日中天，工程团队内部依然混合使用包括 Claude、Codex 等竞品模型与 IDE。 --- ### [19:56](https://www.youtube.com/watch?v=BYITT17BJjI&t=1196s) 视频模型对决：Omni Flash 细节硬伤与 SeaDance 2.0 的版权“护城河” 转向 AI 视频生成领域，近期发布的 Omni Flash 演示效果令人惊叹，但社区技术宅们迅速拿起放大镜开始找茬。聊天室有观众尖锐指出，演示中生成的 V8 发动机运转动画在“点火顺序”上根本不对，甚至有眼尖的汽修专家调侃“这大概少了两只汽缸，本质上只是个 V6”。与此相对，大家开始将它与字节跳动系的 SeaDance 2.0 进行横向对比。主播们直言，SeaDance 2.0 在内容生成限制上显得宽松许多。 主持人就此展开了一段关于跨国版权执法的辛辣调侃。他们指出，好莱坞或西方版权方想要在海外发起跨国诉讼极其困难，往往光是立案、证据开示（discovery）和旷日持久的诉讼流程就得拖上数年。另一位主持人打趣说，这种现象早已存在了几十年——从完全未经授权的仿冒耐克专卖店，到各类瑞士名表“联名款”，甚至整辆 1:1 毫米级复刻的假奔驰大 G（G-Wagon）应有尽有，外观几乎以假乱真，开去 4S 店验车才会暴露本质。 撇开这些版权争议，主播们认为 Omni Flash 和 SeaDance 2.0 展现出的生成质量都已达到 99% 的拟真度。但行业的评测基准也必须随之水涨船高：目前生成 8 秒、10 秒或 20 秒的高清切片已不成问题，但它们仍然无法替代一段长达 20 分钟、逻辑严密的 YouTube 深度原理解释视频。生成长篇视频的计算成本呈指数级上升，长期时空逻辑的一致性也极难维持。AI 视频生成真正的“Deep Research”级别杀手级应用，仍有待行业跨越。 --- ### [22:20](https://www.youtube.com/watch?v=BYITT17BJjI&t=1340s) SpaceX 即将提交 IPO 招股书：高盛爆冷挂帅 Lead Left，早期风投狂赚百亿 根据 Zero Hedge 的最新线报，SpaceX 的 IPO 招股书预计最快在 5 月 20 日（即录制当天）正式亮相。最引人注目的消息是，投资银行高盛（Goldman Sachs）将担任这笔史诗级交易的主承销商（Lead Left）。这一选择让不少业内人士感到意外，因为摩根士丹利（Morgan Stanley）的资深银行家 Michael Grimes 与马斯克合作多年，原先外界普遍预测主承销头衔会归属于大摩，甚至可能会采取多家投行平分秋色的联合承销结构。 紧接着，知名风投记者 Katie Roof（被主播们冠以“世纪独家新闻运动员”称号）爆料了风投史上史无前例的账面回报天花板： - **Founders Fund** 与 **Valor Equity Partners**：预计将分别斩获超过 **600 亿美元** 的收益。 - **Sequoia Capital（红杉资本）**：回报预计将超过 **200 亿美元**。 - **D1 Capital Partners**：在此前报道中预计获利约 **200 亿美元**。 主持人们由衷赞叹了早期投资人的非凡眼光。这些头寸很多建仓于 2004 年至 2010 年间，彼时 SpaceX 既没有如今星链（Starlink）的盈利神话，也没有所谓“太空数据中心”的宏大叙事；当时它只是一家火箭接二连三爆炸、屡陷破产边缘的小型初创公司。风投撰稿人 Packy McCormick 还在社交平台上讽刺了“超级基金（Mega funds）规模过大导致毫无回报率、沦为纯收管理费机器”的流行论调，而 SpaceX 这次用实打实的印钞级战绩，狠狠反击了这一偏见，为 Shawn Maguire、Luke Nosek、Antonio Gracias 等坚定持有者赢得了丰厚犒赏。 ```mermaid flowchart TD accTitle: SpaceX IPO 潜在承销与核心风投机构回报分布 accDescr: 展示了 SpaceX IPO 的主承销商安排以及核心风险投资机构的预计巨额回报 subgraph Deal["承销商阵营"] GS["主承销商 (Lead Left): 高盛 (Goldman Sachs)"] MS["核心参团: 摩根士丹利 (Michael Grimes)"] end subgraph VentureReturns["风投机构预计回报规模"] FF["Founders Fund: 600亿+ 美元"] Valor["Valor (Antonio Gracias): 600亿+ 美元"] Sequoia["红杉资本 (Sequoia): 200亿+ 美元"] D1["D1 Capital: 约 200亿 美元"] end SpaceX["SpaceX IPO 招股书"] --> Deal SpaceX --> VentureReturns ``` --- ### [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) 英伟达季报疯狂印钞，沃兹尼亚克现场抛出“人类专属 AI”包袱 随后节目迅速切入英伟达（Nvidia）最新披露的季报。黄仁勋在财报中高调宣称：“AI 工厂的建设是人类历史上最大规模的基础设施扩张，其推进速度令人叹为观止。智能体 AI（Agentic AI）时代已然降临，正在创造切实的生产力价值并在各行各业迅速普及。” 尽管主播们对黄仁勋动辄挂在嘴边的“AI 工厂”这一营销术语略有微词，但英伟达交出的财务数据堪称“印钞机本机”： - **季度净利润**：从去年同期的 188 亿美元暴增至 **429.6 亿美元**（逼近 430 亿美元），翻了一倍以上。 - **季度营收**：同比大幅跃升 85%，从去年同期的 440 亿美元增至 **816.2 亿美元**。 节目的另一个高潮是苹果联合创始人史蒂夫·沃兹尼亚克（Steve Wozniak）在大峡谷州立大学（Grand Valley State University）的精彩演讲片段。与此前在活动上因争议言论遭狂嘘的 Eric Schmidt 截然不同，沃兹尼亚克一上台就掌握了全场节奏。他在台上幽默地对台下学生喊话：“你们大家脑子里都装有 AI——真正的人类智能（Actual Intelligence）！”台下顿时爆发出热烈的掌声与笑声。沃兹接着打趣说道：“我们以前总是绞尽脑汁琢磨怎么造出一个大脑，用软件、用硬件、还是用突触芯片？但我当年待过的那家公司，工程师们早就研究出怎么造大脑了——只需要花上 9 个月！” 主持人点评道，尽管沃兹尼亚克没有像硅谷极客那样狂热鼓吹 AGI 或超智能药丸，但他极其懂得如何与普通公众沟通，用这种亲切风趣的方式架起人与科技的桥梁。节目最后在一段疯狂互击头盔的滑稽网络打斗视频与欢快的铜锣声（Gong）中正式落幕。 ## Notable quotes > [00:02](https://www.youtube.com/watch?v=BYITT17BJjI&t=2s) 这奇点来得有点软绵绵啊 — 主持人 > [05:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=306s) 现在可以把喜欢的 Mag 7 巨头戴在脸上了 > [05:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=306s) 哪家巨头能大胆一点，把 AI 芯片直接塞进 3M 防护护目镜里？ > [09:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=546s) 这款 Google 与 Warby Parker 联名的眼镜相比 Meta Ray-Bans 简直丑得灾难，相关团队恐怕要有人为此丢饭碗 — Nick > [13:12](https://www.youtube.com/watch?v=BYITT17BJjI&t=792s) 真等不及智能眼镜能够彻底替代我桌上笨重的多台显示器了 — Bobby Chipman > [16:06](https://www.youtube.com/watch?v=BYITT17BJjI&t=966s) 把公司内部混乱的组织架构图直接当成产品发布出来 — Ben Thompson > [17:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1045s) 这帮科技巨头压根就不知道普通大众每天真正需要的是什么 — Greg's Gadgets > [18:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1105s) 在 Anti-gravity 发布视频的第 2 分钟，我不得不倒回去反复看两遍。你能清晰看到 Anti-gravity 团队的人在使用 Codex。难道发布前连最基础的视频内容复核都没人做吗？ — Gergely Orosz > [18:25](https://www.youtube.com/watch?v=BYITT17BJjI&t=1105s) 这完全是无稽之谈，现在公司上下全员都在用 AI — Demis Hassabis > [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) AI 工厂的建设是人类历史上最大规模的基础设施扩张，其推进速度令人叹为观止。智能体 AI（Agentic AI）时代已然降临，正在创造切实的生产力价值并在各行各业迅速普及。 — 黄仁勋 > [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) 你们大家脑子里都装有 AI——真正的人类智能（Actual Intelligence）！ — Steve Wozniak > [24:35](https://www.youtube.com/watch?v=BYITT17BJjI&t=1475s) 我们以前总是绞尽脑汁琢磨怎么造出一个大脑，用软件、用硬件、还是用突触芯片？但我当年待过的那家公司，工程师们早就研究出怎么造大脑了——只需要花上 9 个月！ — Steve Wozniak ## Mentioned TBPN, John Coogan, Jordi Hays, Alice Key, UnHerd, OpenAI, Google, Google I/O, Brandon Gral, Gemini, Gemini Omni, Anti-Gravity, Samsung, Gentle Monster, Warby Parker, Allbirds, Everlane, Nvidia, 黄仁勋, GB200, Cerebras, Meta, Ray-Ban, Oakley, Apple, Carl Zeiss, Sheel Mohnot, Isaac, Project Astra, Spark, Nathan Clark, AI Studio, Gemini Business, Gemini Advanced, AI Pro, AI Ultra, WhatsApp, Instagram, Facebook Messenger, Google Docs, Google Drive, Gmail, iMessage, Nick, Genie 3, Google Street View, YouTube, VO3, VO4, Demis Hassabis, Mark Zuckerberg, Theme Park, Bobby Chipman, Call of Duty, Orion, Meta Connect, Apple Vision Pro, Apple Vision Air, Gemini Flash 3.5, Tennebris, Anthropic, xAI, Cursor, Cursor Bench, Composer 2, TPU, Perplexity, Ben Thompson, Allie K. Miller, Nano Banana, Greg's Gadgets, Gergely Orosz, Codex, Windsurf, Steve Yegge, Claude, Omni Flash, SeaDance 2.0, 字节跳动, SpaceX, Zero Hedge, Goldman Sachs, Morgan Stanley, Michael Grimes, Katie Roof, Founders Fund, Valor Equity Partners, Sequoia Capital, D1 Capital Partners, Starlink, Packy McCormick, Shawn Maguire, Luke Nosek, Antonio Gracias, Steve Wozniak, Grand Valley State University, Eric Schmidt ## Video description Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive Chapters: 00:00 Suits, TBPN’s ESPN Vibe & Robot Wipeout 02:13 Google I/O Recap: Smart Glasses, Warby Parker & Meta Ray-Bans 11:41 AI Worlds, Street View Simulations & the Future of AR Glasses 14:25 Gemini Reactions, Anti-Gravity, Codex Cameo & AI Video Models 22:26 SpaceX IPO & the Biggest Venture Returns Ever 25:08 Nvidia Earnings, AI Factories, Woz on AI & Gong Outro [Source](https://www.youtube.com/watch?v=BYITT17BJjI)

[0:02] 主持人: 这奇点来得有点软绵绵啊 [5:06] 现在可以把喜欢的 Mag 7 巨头戴在脸上了 [5:06] 哪家巨头能大胆一点，把 AI 芯片直接塞进 3M 防护护目镜里？ [9:06] Nick: 这款 Google 与 Warby Parker 联名的眼镜相比 Meta Ray-Bans 简直丑得灾难，相关团队恐怕要有人为此丢饭碗 [13:12] Bobby Chipman: 真等不及智能眼镜能够彻底替代我桌上笨重的多台显示器了 [16:06] Ben Thompson: 把公司内部混乱的组织架构图直接当成产品发布出来 [17:25] Greg's Gadgets: 这帮科技巨头压根就不知道普通大众每天真正需要的是什么 [18:25] Gergely Orosz: 在 Anti-gravity 发布视频的第 2 分钟，我不得不倒回去反复看两遍。你能清晰看到 Anti-gravity 团队的人在使用 Codex。难道发布前连最基础的视频内容复核都没人做吗？ [18:25] Demis Hassabis: 这完全是无稽之谈，现在公司上下全员都在用 AI [24:35] 黄仁勋: AI 工厂的建设是人类历史上最大规模的基础设施扩张，其推进速度令人叹为观止。智能体 AI（Agentic AI）时代已然降临，正在创造切实的生产力价值并在各行各业迅速普及。 [24:35] Steve Wozniak: 你们大家脑子里都装有 AI——真正的人类智能（Actual Intelligence）！ [24:35] Steve Wozniak: 我们以前总是绞尽脑汁琢磨怎么造出一个大脑，用软件、用硬件、还是用突触芯片？但我当年待过的那家公司，工程师们早就研究出怎么造大脑了——只需要花上 9 个月！

[0:02] Alice Gribbin: My latest for UnHerd is on why tech shows like TBPN are re-running the sports media playbook invented by ESPN and why it's working. They're hard to tell apart in their matching suits and floppy haircuts. [8:48] John Coogan: In terms of like wiring into your life, there are way more people that see Google Docs, Gmail as like the central node in their personal life. [18:39] When will we be in a paradigm where you ask a question and you actually get an explainer video 6 minutes, 10 minutes like you would on YouTube? Very computationally expensive, very difficult to maintain the logic. [24:35] Steve Wozniak: You all have AI... actual intelligence

[0:02] Uh, we're both in suits today. I like [0:04] it. It looks good. Some We were in the [0:06] >> Uh, people people have been saying [0:08] >> they can't tell us apart. [0:09] >> You Is that what you're saying? [0:10] >> That. [0:10] >> Okay. Anyway. [0:11] >> But, people have been saying I need [0:12] something like a swear jar [0:14] >> Okay. [0:14] >> when I don't wear a suit. [0:15] >> Oh, okay. I'd like that. [0:17] >> And [0:18] uh so so yeah, something something to [0:19] consider. It should probably be a pretty [0:21] big jar. [0:22] >> 20 bucks to the Opening Eye nonprofit. [0:24] That's what you got to do. Um no, we [0:26] were in Unheard. Uh we I mean we can [0:28] pull up the full uh the full post later [0:30] in the show, but uh [0:31] there's some funny I mean it's a good [0:33] good analysis of like uh you know, what [0:35] we've taken from ESPN, what works about [0:37] live streaming. This is from uh Alice [0:39] Key. Uh my latest for Unheard is on why [0:43] tech shows like TV show are re-running [0:45] the sports media playbook invented by [0:47] ESPN and why it's working. They're hard [0:50] to tell apart in their matching suits [0:52] and floppy haircuts. I don't know if [0:54] that's good or bad, but uh we don't [0:56] always have matching suits. Usually [0:57] Joey's the casual one, but yesterday we [0:58] were both casual. Today, we're both in [1:00] suits. [1:01] >> We do we do still get the brother thing [1:03] a lot. [1:04] >> Yeah, I think it's a term of endearment. [1:05] I think it's positive. Anyway, we have [1:08] to react to Google I/O. Of course, [1:10] there's a whole bunch of announcements. [1:12] Uh some really exciting stuff, some [1:14] stuff that people are having mixed [1:15] reactions to. We'll take you through it [1:16] all, but first we need to watch this [1:18] video of this humanoid robot. Ramp? Why [1:22] is ramp? [1:23] >> No, I'm kidding. [1:23] >> No, I'm crying. Watch this video. [1:26] >> Let's get some steps. [1:27] >> to go to the beginning. You're spoiling [1:28] it. [1:29] >> Let's get some steps. [1:30] >> beginning. [1:32] Doing pretty well, moving pretty [1:33] quickly. Little bit of a Oh, catches [1:35] itself. Catches itself. Not bad, not [1:38] bad. Okay. Seems like a full recovery. [1:40] Seems like you're ready to go. [1:41] >> Another one. [1:43] >> I'm liking it. [1:44] >> Yeah. [1:45] >> And then? [1:50] Not good. And then I don't I wonder what [1:53] it's thinking because you would think [1:55] that there would be some program [1:56] >> the music. They got to cut the music. [1:58] Don't let the music keep playing while [2:00] your your boy's down. [2:01] >> Just gets carried off like this. It's so [2:04] crazy. Just carried off like this. [2:06] Anyway, singularity is light. [2:08] >> Yeah, [2:09] it's possible that the robot died from [2:12] embarrassment. [2:13] >> Yeah, or maybe it was damaged. It's [2:14] possible that like as it as it hit the [2:16] ground it was just actually taken out. [2:18] Anyway, Google I/O. [2:20] Bunch of different announcements. [2:23] Brandon Gral on our team posted on the [2:25] TVP and newsletter some reactions. Sort [2:27] of bucketed it into four key areas. [2:29] Intelligent eyewear. This is an [2:31] interesting one. I want to go into this. [2:33] Gemini Omni, we talked about the videos. [2:35] We played a little bit of that [2:36] yesterday. Upgrades to Gemini LLMs, [2:38] those have been mixed reactions from [2:39] developers. We'll go through that. And [2:41] then anti-gravity, which is an [2:43] interesting place with an interesting [2:44] history. So, let's start with [2:46] intelligent eyewear. If you had to pick, [2:48] Warby Parker Gentle Monster. Have you [2:50] heard of Gentle Monster before? I've [2:52] heard of Warby Parker. I know the story. [2:54] I'm a fan of [2:55] >> of it. [2:57] Not super familiar. [2:57] >> I haven't worn glasses in a very long [2:59] time, so I'm not really in the market, [3:00] but the Warby Parkers I've always I've [3:02] always enjoyed the way they thought [3:03] about the brand. And I've also been [3:04] impressed by the way they built that [3:06] business. They were early to the D2C [3:07] boom. And then didn't some of the [3:10] founders move over to Harry's? Is that [3:11] the same team? Or is that a different [3:13] >> Maybe. [3:14] They were certainly when I when I think [3:16] of D2C I think of Warby. I think of [3:19] Allbirds. I think of Everlane. [3:21] >> Yes, but when you think of the last two [3:23] out of those three, the market caps are [3:25] sub 100 million. Allbirds was trading at [3:28] what? 20 million or something and then [3:30] spiked because the AI thing, but [3:32] Allbirds, Everlane, not really [3:35] sustainable businesses. Warby Parker on [3:38] the other hand, [3:39] current market caps [3:40] >> Allbirds has pretty much sitting at 3 [3:42] and 1/2 [3:43] Allbirds has pretty much been down only. [3:46] >> Down only since the pump. The AI pump. [3:49] >> Neo Cloud, no surprises there. [3:52] >> Well, have they given us an update on [3:53] how they are rolling out Kubernetes, how [3:55] it's going, building their Neo cloud? [3:57] Did they get allocation? Are they [3:59] racking Cerebrus or are they racking [4:01] GB200s? What are they racking and how [4:03] fast are they getting power? [4:05] >> be It would be funny if [4:07] Jensen ends up having to talk about [4:09] Allbirds on the earnings call today. [4:11] >> Didn't they There's a new Didn't they [4:13] also like fully rebrand the name? It was [4:14] going to be like Bird AI or All AI. Like [4:17] They were moving to [4:18] >> still exists, but they they basically [4:20] kept the public entity [4:23] and that's what they're building [4:25] the Neo cloud through. [4:26] So. [4:27] >> Fun. [4:27] >> Lots of fun. [4:28] >> Uh Warby Parker, uh resilient. I mean, [4:32] in 2021 it was a $6 billion company. Now [4:35] it's a $3 billion company. Uh not the [4:37] best scenario, but surprisingly [4:39] resilient, I think, in a time when a lot [4:41] of people wrote off a lot of the [4:43] standalone direct-to-consumer. It was [4:44] like either get rolled into a bigger [4:46] company or [4:47] go or or like face the face of the [4:50] public markets. But Warby Parker has a [4:52] deal with Google and Samsung. [4:54] Uh Google says we're partnering with [4:56] Samsung, Gentle Monster, and Warby [4:58] Parker on new intelligent eyewear. [5:01] Here's a sneak peek at two designs from [5:03] this fall's upcoming collections. [5:06] Uh and people are [5:09] This is I like Futurenomics from Sam. [5:11] Kind of crazy that you can wear your [5:12] favorite Mag 7 on your face now. You [5:14] can. [5:15] Uh the Gentle Monster one does a really [5:17] good job of hiding the camera. I imagine [5:19] that it will have a light to tell you if [5:21] it's recording, but if someone wore [5:23] these from a distance, also Meta [5:25] Ray-Bans have done sort of the hard work [5:27] of becoming the first face computer. So [5:30] when you see Ray-Bans and they're a [5:32] little thick, you start immediately [5:33] thinking, "Oh, should I be looking for a [5:35] for a camera lens? Am I being recorded?" [5:38] Uh but the Gentle Monster [5:40] design, the silhouette, doesn't scream [5:43] technology. It doesn't scream uh [5:45] wearable face camera. And so uh these [5:47] are going to be a little bit more [5:49] stealthy. Warby Parker's look nice, but [5:51] >> the camera bump on this is huge compared [5:53] to the Warby Parker one. [5:54] >> you know, it makes a lot of sense that [5:56] that that Google's and the Metas have to [5:58] go and partner on different silhouettes. [6:01] My expectation, my uninformed uh [6:05] expectation is that Apple will just make [6:08] Apple glasses, right? They will probably [6:11] I I it's hard to see them taking the [6:13] route at least early on of partnering [6:15] and allowing another company to [6:17] influence the design language, but it [6:19] makes a lot of sense that uh that uh [6:22] Meta would partner with Luxottica. [6:24] >> Okay. Firstly, look look at the camera [6:26] bump on this. If you zoom in as far as [6:28] you can, I don't know if we can zoom in [6:29] any further, but uh the the camera is [6:32] not actually not flush with the frames. [6:34] It's actually protruding a little bit. [6:35] Yeah, you can see it right there. Uh [6:37] interesting design choice. I wonder how [6:38] that will catch the light, how that will [6:40] reflect in in uh in the real world. But, [6:43] th- this is all from a joke uh from Abdu [6:45] says, "Okay, so Apple has Carl Zeiss, [6:47] Meta has Ray-Ban and Oakley. Google has [6:49] Gentle Monster and Warby Parker. Boring. [6:51] Which company is going to be bold enough [6:54] to slap wearable technology into some 3M [6:56] safety glasses?" Would you rock these, [6:59] Gordy? [6:59] >> I like these. I like these. I like [7:01] these. [7:03] I do know. [7:03] >> dust in your eyes. But, these these look [7:05] cool. These are sporty. I'm much more [7:07] likely to just commit to the bit. [7:09] >> Just go full full face computer. [7:11] >> Full cyberpunk. Full clank. [7:12] >> Full Yeah, full sa- you you'd clank out. [7:15] >> I think that's uh I think that might be [7:17] the move. I don't know. For some [7:18] company, a challenger company could [7:20] potentially do that. Maybe Friend or [7:22] something. Uh anyway, um what else? Uh [7:24] so, Warby Parker traded down on the [7:26] news, which uh Sheel Mohnot was [7:27] surprised by. Why is Warby Parker down [7:29] 14%? They announced a partnership at [7:31] Google I/O that's been in the works for [7:32] a while. Is because they aren't [7:34] available yet and our friend Rat King, [7:36] Mac I- my my guy Isaac says, "Okay, [7:38] Google AI glasses with Warby Parker are [7:40] officially coming for Meta Ray-Ban." [7:41] Google also said it would bring Gemini [7:44] to glasses this fall with Samsung [7:45] Electronics and and the eyewear [7:47] companies Warby Parker and Gentle [7:48] Monster. The glasses, which work [7:50] similarly to Meta Ray-Ban's smart [7:52] glasses, come with a camera, microphone, [7:53] and [7:53] >> Yeah, at what point at what point does [7:55] Google just buy Warby Parker, right? [7:57] It's a $3 billion company. It's actually [7:59] done quite well over the last uh over [8:02] the last 6 months, it's up 43% in the [8:04] last uh 6 months, although it's been um [8:07] almost flat uh this year. Uh [8:10] I would say I expect that uh smart [8:13] glasses are going to have product market [8:15] fit among people that need to wear [8:17] glasses first, right? If you already [8:19] have to wear glasses all day long for [8:21] your vision, uh why not throw some smart [8:23] features in there? It's going to be [8:25] harder to get someone that doesn't need [8:27] glasses to add a new device to their [8:30] rotation, right? [8:31] >> Yeah. [8:32] >> Um and so, uh you know, Warby Parker's, [8:35] you know, done quite well and uh has [8:36] been surprisingly resilient, but they [8:38] have, you know, incredible distribution. [8:41] Uh and I wouldn't be surprised if they [8:43] get sniped at some point. [8:45] >> mean, deeper integration into a [8:48] traditional [8:50] uh I don't know, like workflow. Like [8:51] like a lot of the Google IO will get [8:53] into this, but it was talking about [8:55] Spark, the personal AI assistant. And [8:57] when I think about [8:59] >> It's an 80s cartoon. [8:59] >> I know. I know. [9:01] >> Or just look at Omni. [9:03] >> Uh yeah, there there's a lot of names. [9:04] It's Google. There's a lot of products. [9:06] Uh you're referring to, of course, [9:07] Nathan Clark's post. It's it's in [9:09] Gemini. Just created an AI studio. Oh, [9:11] it's for your personal goo- Google [9:14] account. For workspace, you need Gemini [9:15] business. No, not Gemini advanced. [9:17] That's AI Pro now. Unless you need AI [9:19] Ultra. Oh, agents? You do that in Spark, [9:21] actually. No, not Gemini API managers. [9:23] And it's the typical uh meme. The The [9:25] interesting thing is that I I do think [9:28] uh Meta Ray-Bans, like it was always [9:30] like, okay, you have a deep integration [9:31] with WhatsApp, you have a deep [9:32] integration with Instagram DMs, maybe [9:35] Facebook Messenger. Some people are [9:37] still using that, but in terms of like [9:39] wiring into your life, there are way [9:41] more people that see Google Docs, Gmail [9:44] as like the central node in their [9:47] personal life. Like Like people think of [9:50] like my the all the stuff I have saved [9:53] on my the desktop of my MacBook is like [9:56] my core repository. A lot of people [9:58] think, "Okay, for the important stuff, [10:00] I'll put it in Google Docs or Google [10:03] Drive, and then most things flow through [10:05] Gmail, most things flow flow through [10:07] iMessage." There are some people that [10:09] just are like, "Yeah, WhatsApp is the [10:11] number one screen time for me. That's [10:13] where I really organize things." But [10:15] Meta doesn't really have this like [10:17] knock-on effect of like, "Oh yes, you've [10:20] you you you're you're It's not a [10:22] necessarily an enterprise level [10:23] productivity suite, but there are people [10:25] who are like, 'Yeah, I'm using Apple [10:26] Mail, iMessage. I save my files in Apple [10:29] Files, Apple you know, the the cloud [10:31] storage. I use my my camera roll, super [10:33] important.'" So, an an AI agent running [10:36] through the Apple ecosystem can be [10:38] valuable, and an AI agent running [10:41] through the Google's ecosystem can be [10:43] valuable. The Meta smart glasses, it's a [10:46] little bit trickier to go and do [10:48] anything that because you're just like [10:50] sort of bumping up against the walled [10:52] gardens, right? Yeah. But Investor Nick [10:54] doesn't like them for aesthetic reasons. [10:55] He says, "These Google X Warby Parker [10:58] glasses are horrific looking compared to [11:00] these Meta Ray-Bans. Someone is probably [11:03] going to lose their job over this." I [11:05] don't know that they look that much [11:06] worse. I don't know. Ray-Bans are a very [11:08] uh iconic silhouette, and they do look [11:10] good. So, we'll see. We'll see how the [11:12] response goes. I think uh from a product [11:14] perspective, there's obviously fertile [11:16] ground. On the flip side, the Wayfarer [11:19] is just such an iconic It's more iconic [11:21] than anything Warby Parker has produced, [11:24] and that's just sort of the reality of [11:26] brand building over a decade versus a [11:29] century or something like that, however [11:31] long Ray-Ban's around, long time. Uh [11:33] anyway, [11:34] uh Genie 3, you can now simulate real [11:36] places by grounding Genie 3 experiences [11:39] with Street View imagery. Google is [11:42] sitting on a a mother lode of real-world [11:44] data. I was always thinking YouTube was [11:46] going to be so valuable for Omni and [11:49] VO3, VO4 maybe in the future. I hadn't [11:53] considered Street View as a trove of [11:55] data. Demis seems very data-pilled. He [11:59] seems a lot of the the Mag 7 CEOs seem [12:01] very data-pilled. There's the story [12:02] about Mark Zuckerberg uh screen [12:04] recording or or or [12:06] logging all the computer use from all [12:08] the Meta employees. The these important [12:10] troves of of data are increasing in [12:12] value and Street View certainly seems [12:14] like it's one of them. This is cool. I [12:16] wonder how interactive this will be, how [12:19] how this actually instantiates into a [12:21] game. It's a great demo. What does it [12:23] take to [12:23] >> Yeah, or do they allow people to build [12:24] games on top of this? [12:26] >> Yeah, I just think about I don't know. I [12:28] I mean, Demis has a background in games [12:30] and he was sort of alluding to the fact [12:32] that he might go back into games at some [12:33] point or or at least be able to like [12:35] scratch that itch again. Famously, he uh [12:38] wrote uh [12:39] uh a programmatic code to generate vomit [12:43] in a roller coaster simulator, very fun [12:46] story. Uh but again, when I think about [12:49] uh Roller Coaster Tycoon, which was not [12:50] I think I I don't think he was actually [12:52] working on that game. It was a similar [12:53] uh theme park simulator, but uh we are [12:55] moving back into the simulator world, [12:57] but the mechanic is what is so enticing [13:00] to gamers often. When I think about the [13:02] games that I've spent a long time with, [13:04] some of them have incredible graphics, [13:06] AAA graphics. Some of them have 2D [13:08] graphics. Uh but the mechanic is great [13:10] and so that is what gets me to keep [13:11] coming back. [13:12] >> Bobby Chipman in the X Chats says, [13:14] "Can't wait for smart glasses to fully [13:16] replace my monitors." [13:18] >> Yeah, maybe you'll need augmented [13:20] reality or something. Meta Ray Ban [13:22] display, certainly going that direction, [13:24] the Orion. I've I've been surprised. [13:26] Wasn't the first episode we ever did, we [13:28] were talking about Orion Uh and they [13:29] still haven't shipped it, right? I mean [13:30] they shipped the smaller version of the [13:32] Meta Ray Band displays which have sort [13:34] of the the Call of Duty HUD. It's not [13:36] full augmented reality. I was expecting [13:39] I was expecting we we we've demoed the [13:40] Orion headset and it and in it has a bit [13:43] of a narrow field of view but it really [13:44] can put a screen right in front of you [13:46] and I I assumed that you know everyone [13:49] was saying it's really expensive, it's [13:50] clunky, it's not ready for prime time [13:52] but you know look at how fast things are [13:53] going in a year maybe two we'll get it [13:57] and maybe that's coming at the next Meta [13:59] Connect, maybe this summer we'll see it [14:01] but haven't been that many rumbles on it [14:03] and then obviously the massive pitch [14:05] shift to AI CapEx might have taken a [14:07] backseat. I don't know. I'm certainly [14:08] hopeful. I like AR and VR. I think I [14:12] think we're we're we're overdue for a [14:13] new fun product. I'm still waiting for [14:15] the next Apple Vision Pro, Apple Vision [14:17] Air, something just lighter, that's all [14:19] I want, cheaper maybe but lighter and [14:22] same screen. Screen was great. [14:24] Anyway, [14:25] >> We know John. [14:26] >> Gemini Flash 3.5 looks pretty neat [14:28] according to Tennebris and extremely [14:31] fast but still largely the sort of [14:33] incremental progress we've come to [14:34] expect from Google. Generally a pretty [14:37] disappointing IO. Now what's interesting [14:39] is that Gemini 3 felt like a new base [14:43] pre-trained, felt like it had some of [14:44] that big model smell, felt like it was [14:47] you know really delightful to talk to [14:49] and I think a lot of people were [14:50] expecting Gemini 4 here. We're still [14:52] waiting for the next iteration here. And [14:55] also [14:55] >> Yeah, we're still waiting for Pro. [14:57] >> Yeah, Pro isn't out but people are [14:59] speculating that that 3.5 Pro won't [15:02] necessarily be a new pre-train and so it [15:04] seems like it's a there's a little bit [15:06] of [15:07] research being at odds with like the [15:09] product cadence. Like the Google IO is [15:11] scheduled probably like two years in [15:13] advance and whether or not the training [15:15] run finishes on time is a little bit [15:18] harder to package up and nail on a [15:21] specific time. We see this with uh the [15:23] independent labs or the Open AI, [15:25] Anthropic, the other labs, xAI. Like [15:27] they're launching models very much like [15:29] when they're done and then they will [15:31] like instantiate like something that [15:33] looks like a conference around it or [15:34] maybe a video or a blog post, a model [15:37] card. Uh but if you're if you're [15:39] grinding towards a specific date uh and [15:42] this this specific model isn't quite [15:43] ready, uh you come out with something [15:45] that's looks a little bit more [15:46] incremental. People were really really [15:48] uh honing in on the fact that the [15:49] >> the cutoff date was January of 2025, [15:53] right? It was that the Was that the [15:54] date? That Or was it December of 2025? [15:56] Either way, [15:57] >> yeah, I I don't know. I don't know how [15:58] much cutoff time cutoff dates matter [16:00] because uh you know, all these models, [16:03] you know, can query the web and and get [16:05] up-to-date information, but [16:06] >> overall reactions from uh developers [16:10] across the board were not good. Not good [16:13] at [16:14] >> Perplexity or underwhelming. [16:16] >> Cursor [16:17] ranked it on Cursor bench. Uh it is [16:20] below Composer 2. [16:22] >> Is that Is that a fair thing? I mean, [16:27] I'd like to see you rank another another [16:29] live stream on TVP and bench. It doesn't [16:31] match up, you know? It's like [16:33] >> No, I mean, they have they have [16:35] >> I guess it is a bench. [16:36] >> They have all the other They have all [16:37] the other frontier [16:38] >> And some of them are ahead of Cursor's [16:39] own models on Cursor bench. [16:41] >> It's just one data point. [16:42] >> Yeah, yeah, yeah. [16:43] >> But here's the Here's the other thing. [16:44] It's four times it it costs four ti- it [16:47] underperforms Composer 2 even though [16:49] it's roughly four times more expensive. [16:51] >> Interesting. Interesting. Yeah, I I feel [16:53] like for a long time Google's [16:54] positioning was, you know, frontier or [16:56] near frontier, but best possible [16:58] pricing. And this marks sort of a shift [17:01] in the strategy perhaps. [17:03] >> Yeah. Overall, starting to make more and [17:06] more and more sense why Google has put [17:09] so much capital and resources behind [17:12] Anthropic, right? [17:13] >> Perplexity uh seems to indicate that [17:15] DeepMind is constrained by data rather [17:17] than compute for what they intend to do, [17:19] hence the TPU sales. Uh rest of Google [17:22] now shipping their org chart. Ben [17:23] Thompson talked about that a little bit. [17:25] And there was some context on like, you [17:27] know, we were asking the question like, [17:29] will there be AI fatigue from stuffing [17:31] AI in every product surface area? Ali [17:34] Kay Miller shares one of the loudest [17:36] applauses in the entire Google keynote. [17:39] Uh Nishtha Nishtha put on the Gentle [17:42] Monster plus Gemini glasses, tapped the [17:45] side to summon Gemini. An all-in-one [17:48] prompt said, "Take a photo and put a [17:49] cartoon blimp in the sky that says [17:51] Google I/O 2026." And within seconds, [17:55] the preview of the edited photo from [17:56] Nana Banana appeared on her watch. I [17:58] want to spend less time on screens. AI [18:00] is really coming everywhere. And so much [18:02] is driven by voice AI as the interaction [18:04] mode. Uh very cool demo, uh impressive [18:07] technology, but Greg's Gadgets says, [18:09] "These companies truly have no idea what [18:10] regular people want." Because uh yeah, [18:13] that is a little bit of a niche use [18:15] case. You You need to be more creative [18:17] with it uh for when you would actually [18:18] use that because this is a very It It [18:21] It's a perfect demo of the product and [18:23] the functionality, but it lacks that [18:25] like creative spark of like, "Yes, I did [18:26] want a picture of that on my wrist at [18:28] that key moment in time." Uh if you're [18:30] not doing a demo. His point is that [18:32] regular people would not be excited [18:34] about that particular feature, right? [18:36] SpaceX IPO, we're getting more details [18:38] by the day. [18:39] >> other The other thing The other thing [18:40] that is was going pretty viral, last [18:42] thing on I/O, [18:44] was that [18:45] uh Google anti-gravity team [18:48] uh flashed a Codex folder in their uh in [18:51] their actual demo video. [18:53] Uh Gergely says, "I had to do a double [18:55] take in the second minute of the launch [18:56] video for anti-gravity. You can see [18:58] people use Codex on the anti-gravity [19:00] team. Did no one double-check the launch [19:02] video at the very least?" Not a huge [19:04] surprise. Uh obviously anti-gravity [19:06] looks uh a lot of people were saying it [19:08] looks quite quite like uh [19:10] uh Codex. So, clearly [19:12] >> More than Winston? I feel like it would [19:14] be like a way [19:15] they they would have just rebuilt [19:16] Windsurf. I don't know. We'll have to [19:17] see. [19:18] >> No. So, but but anyways, this isn't a [19:20] huge surprise, right? Google's been [19:23] using bunch of Anthropic models. [19:25] Clearly, they're using a ton of [19:26] different models uh in products [19:28] internally. So [19:29] >> What was that drama with Steve Yegge [19:31] going back and forth with Demis about [19:33] like what what teams are using what [19:36] models and stuff? There was a big back [19:38] and forth, big dust-up on the timeline [19:40] like a month ago about like whether or [19:42] not Google's employees were deploying AI [19:45] efficiently or broadly. Some of them [19:47] aren't and some of them are and Demis [19:49] chimed in and said like this is just [19:51] complete wrong and everyone's using AI. [19:54] I don't know. Goes back and forth. Uh [19:56] there's also people are benchmarking [19:57] Omni Flash, which looked amazing when we [19:59] saw the videos. There was there were a [20:01] few like little quirks. Some people in [20:03] the chat were saying that the firing [20:05] order of the V8 was not correct. Maybe [20:07] it was only a V6. [20:08] >> Yeah, it was missing two cylinders. [20:09] >> missing two cylinders. Uh but it looked [20:11] good to me. I don't know. Uh but now [20:13] people are actually copping it to [20:14] SeaDance 2.0, which obviously has much [20:17] looser content restrictions uh because I [20:20] I I guess just like Hollywood can't file [20:23] a lawsuit in China. I I'm not exactly [20:26] sure how that works because SeaDance [20:27] seems to be available in America. It it [20:30] seems like maybe [20:31] >> Oh, I think I think Chinese businesses [20:33] have been relatively immune to US [20:35] copyright law for a very very very long [20:38] time. [20:38] >> And it also might just take like years [20:40] to file a lawsuit, do discovery, [20:42] actually go through and litigate. [20:44] >> Oh, yeah. You you can just go like [20:46] there's malls in China where you can go [20:48] to a Nike store and Nike has nothing to [20:51] do with it and yet all the products [20:53] >> They've been selling Swatch APs over [20:54] there for [20:55] decades. [20:57] >> Yeah, just ask just ask [20:59] Rolex and Patek how they're how they're [21:01] >> Sure. Sure. Yeah, I I've heard fake [21:03] cars, too. Like you you get a full [21:06] replica of like a G-Wagon that's just [21:09] made in a factory and then you could buy [21:11] it, bring it over here, and you take it [21:12] to a Mercedes dealership and they're [21:14] just like, "This is not a Mercedes." [21:16] But, it looks like one like like, you [21:18] know, [21:19] to the millimeter from the outside. [21:21] But, internally it's just it's just [21:23] frauding. [21:24] Uh anyway, [21:26] C-DANCE 2.0 looks great. Omni Flash [21:28] looks great as well. These are both like [21:30] super useful. We'll see how they [21:31] actually play out and how they get [21:32] implemented, how they get used. The [21:34] interesting thing will be like like at [21:36] what point like it still takes a long [21:38] time to generate videos, very hard to [21:40] get them right. The last 90 like we're [21:41] at 99% fidelity, but when you click in, [21:44] you start noticing little details. When [21:45] will we be in a you know, paradigm where [21:47] you ask a question and you actually get [21:50] an explainer video 6 minutes, 10 minutes [21:52] like you would on YouTube. Uh very [21:55] computationally expensive, very [21:56] difficult to maintain the the logic. [21:58] Like, what is the deep research report [22:00] of Omni Flash? These 8-second, [22:02] 10-second, 20-second videos are [22:04] impressive, uh but not perfectly [22:06] substitutable for a 20-minute YouTube [22:08] video because of the time and the level [22:10] of detail that you can go into. Some [22:12] people that are looking for information [22:14] about a V8 engine, they want a breakdown [22:16] that lasts 20 minutes. And so, uh that's [22:19] the next benchmark. We got to move the [22:20] goalposts. Uh SpaceX IPO, the prospectus [22:23] is income is incoming according to Zero [22:25] Hedge as soon as May 20th. That's today. [22:29] We will see Goldman lead left. This was [22:32] a surprise. Michael Grimes has worked [22:34] with Elon for a long time at Morgan [22:35] Stanley. There was some back and forth. [22:36] He went back to Morgan Stanley. There [22:38] was a question about whether or not uh [22:40] there would even be a lead left because [22:42] there's such a big IPO. Maybe they all [22:44] share equally. Obviously, they're all [22:45] going to make a ton of money off of [22:46] this. So, good news from start to [22:48] finish, but it is interesting that [22:50] Goldman was selected. Do you have a [22:51] soundboard cue you want to play? [22:53] >> I'm always ready, John. [22:55] >> Um Katie Roof has a scoop. [22:57] >> Uh [22:58] the scoop athlete of the century, Katie [23:00] Roof, [23:01] >> has a scoop on the biggest venture [23:03] returns ever. Founders Fund and Valor [23:06] are set to make more than 60 billion in [23:09] gains on the SpaceX IPO. Sequoia have [23:11] more than 20 billion. [23:14] >> Yeah, was this was this somewhat of a [23:15] reaction to [23:17] D1 getting a lot of credit earlier in [23:20] the week, right? They're set to [23:22] to to generate roughly 20 billion [23:25] in in returns [23:27] and maybe some of these other funds [23:29] thought thought to put their hand up and [23:31] say [23:31] >> I don't think I I I I think that at this [23:34] scale, like there are so many LPs in [23:36] these funds that are getting updates and [23:38] they've been they've known the numbers [23:40] for a long time. They've known the [23:42] ownership, the holdings and you do some [23:44] back of the envelope and you get to some [23:46] pretty huge numbers. Will be very [23:48] interesting. Huge for Shawn Maguire, [23:50] huge for Luke Nosek and a lot of other [23:52] folks over at Founders Fund and Antonio [23:54] Gracias and Valor and all the other [23:56] Founders Fund folks. [23:57] >> Yeah, look Sequoia, Founders Fund [23:59] >> They needed a win. [24:00] >> They needed a win. [24:00] >> a win. I mean, you go back like, you [24:02] know, the these investments were made [24:04] like 2004, 2010, like it was not [24:07] obvious. Certainly, there was no [24:08] Starlink narrative when these were made. [24:10] There was no space data center [24:11] narrative. This was a rocket company [24:13] that was blowing up rockets left and [24:15] right and not quite getting to [24:18] you know, massive business. So, you [24:20] really had to believe be a believer and [24:21] they were. [24:22] >> Packy was having some fun on the [24:24] timeline. He said [24:26] >> Mega funds are too big to generate [24:29] returns. They're basically just fee [24:31] collectors. [24:32] >> And of course, they're printing. [24:35] >> Yes, they are printing. Jensen Huang [24:37] talked about the quarter that Nvidia [24:39] just had. He said the buildout of AI [24:42] factories, boo, the largest [24:45] infrastructure expansion in human [24:47] history, yay, is accelerating at [24:49] extraordinary speed, yay. [24:52] Not a fan of the AI factory terminology, [24:55] but good that [24:57] there is progress being made. A genetic [24:59] AI has arrived, he says, doing [25:01] productive work, true, generating real [25:03] value, true, and scaling rapidly across [25:07] companies and industries, also true. [25:09] Lots of good stuff. Nvidia net income it [25:12] rose to 42.96 [25:15] billion. They almost hit 43 billion, not [25:18] too bad. Um a year earlier they were [25:22] doing just 18.8 billion in net income. [25:25] Huge huge increase. [25:27] >> Really wild. [25:28] >> Printing. Definition of printing. [25:30] >> All good news. The stock is sort of up [25:34] and down, basically flat, but Nvidia [25:36] revenue jumped 85% to 81.62 [25:40] billion from 44 billion last year, the [25:43] company said. [25:44] >> Shocking. [25:44] >> Yeah, great stuff. Steve Wozniak, the [25:47] co-founder of Apple at Grand Valley [25:50] State University, he talked about AI and [25:52] unlike Eric Schmidt, he did not get [25:54] booed off booed off stage. He actually [25:56] got cheered for his comments. [26:00] >> this was Who was it, Tyler? [26:02] >> Steve Wozniak? [26:04] >> No, no, no. Shots fired. [26:06] >> I'm a big fan of Steve Wozniak. [26:07] >> I love the Woz. [26:08] >> Shots fired. [26:09] >> Let's play the clip from Grand Valley [26:11] State University on Instagram here. [26:13] >> You all have AI. [26:17] You all have AI, actual intelligence. [26:20] >> Oh, mic drop. [26:22] Knee slapper. [26:25] Hey, playing to the crowd. He knows the [26:27] audience. He knows the audience. [26:28] >> We were trying to figure out how to make [26:30] a brain, software or hardware, synapse [26:32] chips, and I was at a company where the [26:34] engineers figured out how to make a [26:36] brain. Takes 9 months. [26:40] >> Um yeah, knee slapper, but it he knows [26:44] the audience. He's delivering the right [26:45] thing. Is he AGI pill? Is he super [26:47] intelligence pill? Probably not, but [26:50] it is [26:51] regardless. I think it's it's [26:54] potentially the right framing for the [26:55] crowd. It's knowing the audience and [26:58] and and that is a way to bridge to a [27:01] broader conversation about AI, a broader [27:03] conversation about how humans fit into a [27:06] post-AGI world. I don't know. We'll have [27:09] to go watch the full eventually. [27:11] >> close it out with this video from Tyler [27:14] that we can pull up. [27:15] >> Okay. What is this? [27:16] >> This is the video I was referencing with [27:19] with Marcus. I'm very concerned about [27:21] these gentlemen [27:22] >> Okay. [27:23] >> and what they're doing. [27:24] >> Pull it up. What's happening? [27:27] >> Like this is [27:29] This is insane contact. [27:32] I mean the helmet is getting dented. I [27:34] think this is brain damage. [27:36] >> Skill issue. [27:38] They really hitting that, aren't they? [27:42] >> But we'll try it out after we get [27:44] >> We have the gong mallet. [27:46] See you tomorrow. Goodbye. [27:47] >> Cheers.