OpenAI Solves Navier-Stokes as AI Creates 1M US Jobs

OpenAI攻克纳维-斯托克斯方程，AI在美创造百万就业岗位 前沿AI模型攻克复杂的千禧年大奖数学难题，同时美国劳工数据显示人工智能净创造了近100万个就业岗位，消除了人们对劳动力市场即刻崩溃的担忧。 主持人们探讨了前沿AI攻克纳维-斯托克斯方程等千禧年大奖难题的现实意义，并将抽象的纯数学与3D建模和智能体计算机操作等切实的消费级成果进行了对比。他们探讨了来自Google DeepMind的Astra实用多模态演示，以及Instinct AI激进的个人助手工作流——后者既引发了平台封禁，也带来了意想不到的成功。转向娱乐领域，他们点评了Nathan Fielder拍摄的Elizabeth Holmes纪录片预告片，以及好莱坞在《Artificial》和《The Social Reckoning》中对科技文化的呈现。最后，节目分析了《经济学人》（The Economist）和美国劳工统计局的宏观经济劳动力数据，指出激增的AI基础设施投资和初创企业招聘已使预期的“AI就业末日”暂缓，取而代之的是净增长的就业繁荣。 继DeepMind和OpenAI在2025年国际数学奥林匹克竞赛中斩获金牌后，前沿AI实验室正将目标对准纳维-斯托克斯方程等千禧年大奖难题。 (0:01) 针对AI辅助数学研究的贡献和方法，Google与隶属于Anthropic的数学家之间爆发了激烈的署名争议。 (0:01) 纳维-斯托克斯方程的理论证明对实际工程影响微乎其微，因为在空气动力学和天气建模等现实应用中，流体力学已经可以通过数值计算解决。 (3:18) 与抽象的纯数学里程碑相比，公众兴趣更倾向于实用的消费级功能，如3D资产生成和AI自动预订。 (7:11) 与对数学定理进行计算密集型推理相比，将AI应用于生物技术和癌症研究需要经历漫长得多的实验和监管周期。 (9:39)

TBPN explores the growing divide between abstract AI milestones—such as breakthroughs in pure mathematics—and the tangible consumer applications and macroeconomic labor trends reshaping the public perception of technology. Math Milestones Versus Real-World Impact (0:01) Frontier AI models from Google DeepMind and OpenAI have tackled advanced mathematical challenges, matching gold-medal performance at the 2025 International Mathematical Olympiad (scoring 35 of 42) and targeting Millennium Prize Problems like the Navier–Stokes existence and smoothness equations. However, these feats have provoked academic disputes over evaluation rubrics and authorship, while generating the dismissive 'cool calculator' critique from the public. Because Navier–Stokes is already solved numerically to model airflow and weather patterns, physicists note that an analytical proof has negligible practical utility for applied engineering. Everyday Agentic Utility and Visual Generation (7:11) In contrast to esoteric mathematical proofs, users respond far more enthusiastically to immediate, generative tools and autonomous agents. Recent workflows highlight models generating functional 3D Blender assets, custom home remodels, and playable games directly from 2D images, alongside physical robotics demonstrations like Google DeepMind's Project Astra manipulating a paintbrush. Practical enthusiasm is driven by low-friction delegation, from automating system terminal commands to handling parking tickets. Consumer assistants like Instinct have automated complex tasks—such as tracking down arena broadcast footage—though high-volume automated outreach has also triggered platform rate limits and bans on services like Resy. Hollywood Portrayals of Tech Narratives (16:40) Mainstream media continues adapting Silicon Valley dramas, though with uneven resonance. Newly released previews include an unscripted A24 documentary by Nathan Fielder featuring Theranos founder Elizabeth Holmes filmed prior to her federal prison surrender, as well as the high-budget drama *Artificial*, which depicts apocalyptic stakes in frontier AI labs. The hosts contrast these high-interest topics with teasers like *The Social Reckoning*, centered on Frances Haugen's 2021 Facebook disclosures, arguing that past whistleblower tropes feel disconnected from an audience currently captivated by compute cluster expansion, hardware spectacles, and AI talent wars. Macro Data Defies the Job Apocalypse (27:41) Despite persistent concerns of widespread technological displacement, data analyzed by The Economist and the Bureau of Labor Statistics indicates the anticipated AI employment crisis is on hold. The US economy added 162,000 jobs in August with unemployment at 4.1%, and youth unemployment gaps remain near multi-decade lows. While specific tech and financial firms have trimmed headcount, Challenger, Gray & Christmas recorded an average of roughly 16,000 AI-related job cuts per month—less than 1% of the normal monthly US churn of 1.7 million job losses and 1.8 million hires. Macro indicators suggest AI has generated approximately 1 million US jobs, heavily driven by data center infrastructure and engineering demand, offsetting roughly 200,000 AI-attributed layoffs since mid-2023. Nonetheless, questions remain over whether public skepticism toward frontier labs can be resolved through abstract milestones or incremental breakthroughs, as long as tangible consumer relief remains incremental.

TBPN节目探讨了前沿人工智能在攻破千禧年数学难题与大众实际体验之间的认知落差，并结合最新宏观劳动力数据指出，AI非但未引发所谓的“就业末日”，反而推动了逾百万净增岗位的经济繁荣。 纯数学难题突破与现实效用断层 (0:01) 节目探讨了纳维-斯托克斯方程（Navier-Stokes）这一千禧年大奖难题（Millennium Prize Problem）在学术界与公众舆论中引发的分歧。回顾2025年国际数学奥林匹克竞赛（IMO），Google DeepMind与OpenAI各自取得35分（总分42分，均未能解答第六题），但OpenAI依靠往届金牌得主评估的方式遭到了合规性质疑。此外，前沿数学模型的竞争还激化了Google学者与Anthropic合作研究者之间的署名权指控风波。主持人指出，向单一数学命题倾注数百万美元推理算力的做法，常被大众视作“高级计算器”的自娱自乐。 物理与工程界普遍认为，流体力学早已通过数值计算日常模拟机翼气流与天气预报，证明该方程的存在性与光滑性并不会直接改善商业航班的颠簸感或降低碳排放。相比抽象的纸面证明，普通公众对Sora、千兆瓦级数据中心或使用ChatGPT代订理发等直观应用反应更为热烈。关于AI向生物医药领域的拓展，主持人分析指出，攻克癌症依赖漫长的临床试验周期，且受制于传统制药业较为负面的公众声誉，微小的渐进式突破短期内难以带给科技行业即时的舆论红利。 智能代理的实际落地与规则碰撞 (7:11) 与曲高和寡的纯理论相比，3D生成与系统级智能代理展示出极强的实际吸引力。近期用户已能通过单张2D图像直接生成用于Blender的完整3D模型、自定义车辆乃至可运行的网页游戏。在多模态与物理控制方面，Google DeepMind的Project Astra展示了通过机械臂与画笔在现实画布上迭代绘制金门大桥的能力；而在基准测试中，Astra在ARC-AGI取得99%得分的传闻甚至引发了开发者对其技术奇点的热烈调侃。 个人智能助手Instinct等代理工具展现了“持续代办”的潜力与现实阻力。该助手不仅能通过持续跟进赛事公关与场馆联系人，为用户在海量素材中找回美网（US Open）球迷镜头，但也因高频自动化抓取餐厅预订流程遭到平台Resy的直接封禁。主持人分析认为，当前此类高频冷邮件虽被视为自动化骚扰，但未来工作流必将转向“代理对代理”（agent-to-agent）的中间人机制，由接收端代理进行意图过滤与协同，减少对人类日常工作的打扰。 影视文化投射与现代科技现实脱节 (21:01) 节目审视了近期多部聚焦硅谷的影视预告片，包括Nathan Fielder拍摄的Theranos创始人Elizabeth Holmes纪录片、聚焦前沿AI竞赛的戏剧电影《Artificial》，以及以Facebook前吹哨人Frances Haugen听证会为蓝本的《The Social Reckoning》。主持人指出，好莱坞对科技题材的刻画往往滞后，依然沉湎于国会听证会、青少年心理健康和内部泄密等旧套路，而缺乏对当下算力竞赛、人才挖角和现实创始人风貌的敏锐洞察。 相比陈旧的监管说教，现代科技受众更期待看到如Marc Benioff在夏威夷与海豚共泳后创立Salesforce并获得Larry Ellison两百万美元种子轮注资的离奇创业历程，或是当下AI实验室间的真实博弈。正如《社交网络》（The Social Network）上映后反向激励了一代创业者，影视叙事对技术探索与现实野心的真实还原才能引发时代共鸣，过度戏剧化的末日宣教已难以契合行业现状。 宏观劳动力数据证伪AI就业末日论 (27:41) 节目援引《经济学人》（The Economist）与美国劳工统计局（BLS）等机构的宏观数据指出，市场预期的“AI就业末日”已被实际上的“AI就业热潮”所取代。经济学家Kevin Bryan指出，提高生产率和促进投资的技术在历史上通常会带动竞争性招聘。美国劳工统计局9月4日数据显示，8月份新增非农就业16.2万人，失业率维持在4.1%，处于过去50年90%以上的较优水平；20至24岁年轻群体的失业率与整体水平的差距也接近数十年低位。 尽管咨询机构Challenger, Gray & Christmas统计显示今年每月约有1.6万个岗位流失归因于AI，且商业专业服务领域招聘仍低于2015至2019年基准线10%，但相较于美国劳动力市场每月170万离职与180万入职的总流动量，AI裁员占比不足1%。《经济学人》估算，自2023年中期以来，AI在美国已直接或间接创造了约100万个新岗位，远超同期约20万个与AI相关的裁员。数据中心基建、电网扩张与初创公司对高阶人才的争夺，驱动了整体就业的正向净增长。 总体而言，TBPN节目认为当前人工智能的发展在高度抽象的数学理论突破与消费级日常代理应用之间呈现两极化推进，而产业对算力设施与工程研发的巨额资本注入，有效对冲了技术替代劳动力带来的短期失业焦虑。然而，前沿模型在大规模推理开销下的实际投入产出比、自主代理对现有网络服务协议与商业规则的冲击，以及生物医药等物理研发周期的固有限制，仍是科技行业后续推进时面临的未决挑战。

## Background 本期视频由科技对谈频道 TBPN 于 2026 年 9 月 9 日发布于 YouTube 平台。节目为《Diet TBPN》系列每日精选，由 John Coogan 与 Jordi Hays 主持，聚焦 OpenAI 攻克纳维-斯托克斯方程的学术争议、Astra 3D 与 Computer Use 突破、好莱坞最新科技题材电影预告，以及《经济学人》关于 AI 就业潮的最新分析。 ## TL;DR 节目首先探讨了前沿 AI 攻克纳维-斯托克斯等千禧年数学难题背后的学术署名权之争与实用价值落差，随后盘点了 Astra 在 3D 建模、机械臂画画和代缴停车罚单等 Computer Use 领域的实用突破。接着，主持人们解析了激进个人助理 Instinct 引发的争议，并深入评析了内森·菲尔德的伊丽莎白·霍姆斯纪录片以及《Artificial》等好莱坞科技电影预告。最后，节目引用《经济学人》最新宏观数据指出，AI 带来的基础设施与新兴岗位增长远超局部裁员，所谓的“AI 就业末日”已被正式推迟。 ## Atmosphere & style 整期节目节奏紧凑明快、语言辛辣幽默，充满了硅谷圈内人特有的极客黑话与犀利洞察。两位主持人默契十足，在硬核的技术原理剖析、宏观就业数据解读与好莱坞八卦闲聊之间自如切换，笑料不断。听众不仅能感受到前沿技术突破带来的震撼与荒诞感，也能在轻松诙谐的对谈氛围中全面把握科技行业的最新脉搏。 ## Key takeaways - [00:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1s) AI 在国际数学奥林匹克竞赛（IMO）与千禧年大奖难题（如纳维-斯托克斯方程）上的突破，引发了顶级实验室与学者关于成果归属及论文署名权认定的激烈争议。 - [03:18](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=198s) 证明纯数学难题的解并非能立即转化为工程生产力，由于工程师早已依赖数值计算解决实际流体问题，纯数学突破在短期内面临公众感知脱节与缺乏立竿见影工程效应的现实。 - [07:11](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=431s) Astra 在 3D 快速建模和 Computer Use（计算机操作代理）上展现出极高实用价值，不仅能将个人现实资产迅速 3D 数字化，还能自主代办如滞后停车罚单缴费等繁琐日常事务。 - [09:39](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=579s) AI 探索领域转向生物医药（如攻克癌症）将受制于多期临床试验漫长的客观物理周期，无法仅凭堆砌算力一蹴而就，也难以迅速扭转科技行业面临的公众舆论声誉悖论。 - [11:54](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=714s) Astra 控制实体机械臂自主修正笔触绘制金门大桥，展示了具身智能在物理世界闭环迭代的潜力，多模态与 3D 生成的飞跃也引发了硅谷关于 AGI 进程加速的黑色幽默与讨论。 - [14:27](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=867s) 激进型个人助理 Instinct 因高频自动化导致用户 Resy 订座账号被封，却又凭借地毯式沟通成功找回美网转播镜头，展现出 AI Agent 游走在垃圾信息与极高执行力之间的双刃剑特性。 - [16:40](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1000s) Nathan Fielder 与 A24 合作跟拍 Elizabeth Holmes 入狱前的纪录片，通过模糊真实与表演边界的独特风格，再度探讨科技骗局叙事在主流圈层的穿透力与荒诞感。 - [21:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1261s) 好莱坞新片《Artificial》聚焦 AI 末日与造神竞赛，但此类剧情片极易重蹈《社交网络》的覆辙——本意反思与塑造反派，最终却反向神化并激励了新一代科技创业者。 - [24:02](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1442s) 围绕 Facebook 吹哨人事件的影片《The Social Reckoning》折射出好莱坞科技叙事的滞后性，当下大众焦点已从早期的社交网络监管批判转向 AI 人才与算力军备竞赛。 - [27:41](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1661s) 《经济学人》报告表明所谓“AI 就业末日”已被推迟，AI 关联裁员占全美劳动力月均流转比例不足 1%，且庞大的数据中心与电网基建投资正在拉动强劲的用工需求。 - [30:42](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1842s) 数据评估显示 AI 迄今在美国催生了约 100 万个新增就业岗位，大幅超过自 2023 年中以来归因于 AI 的 20 万例裁员，宏观上呈现出“AI 驱动的就业繁荣”。 ## Detailed walkthrough ### [00:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1s) OpenAI 攻克纳维-斯托克斯方程与顶级数学难题的学术争议 主持人在节目开篇探讨了一个核心疑问：“数学真的有那么重要吗？”（"Does math matter?"）。围绕千禧年大奖难题（Millennium Prize Problems）之一的**纳维-斯托克斯方程**（Navier-Stokes equations），AI 领域的讨论在社交媒体时间线上引发了极大的关注与争议。千禧年大奖难题是长期困扰全球顶尖数学家与 PhD 专家的极高难度难题，在 AI 时代到来之前，仅有庞加莱猜想（Poincaré conjecture）被成功破解。如今，AI 正在逐步攻克这些数学界的“天堑”。 主持人们回顾了 2025 年国际数学奥林匹克竞赛（IMO）的激烈角逐。当时，来自 Cognition 的 Scott Woo 曾预测 AI 将在 2025 年 IMO 比赛中斩获金牌，后来的事实证明了他的预测。Google DeepMind、OpenAI 以及专注于数学 AI 的实验室 Harmonic 都在争夺 IMO 金牌。比赛共 6 道题，尽管第 6 题难度极高、当时没有任何 AI 模型能解出，但 Google 和 OpenAI 的团队均拿到了 35/42 的高分（达到金牌线）。 两家巨头的成绩认定机制引发了不小的戏剧性碰撞与“氛围之争”（vibe wars）： * **OpenAI**：邀请了往届 IMO 金牌得主进行人工验证与认可； * **Google DeepMind**：则严格按照官方内部评分标准进行打分。 主持人幽默地比喻道：“这就好比在奥运会主场馆赛事正酣时，有人在场外的停车场里跑出了百米冲刺的最快纪录。” 这种竞争演变成了极高层面的“学术署名权争议”。例如某位 Google 领域的数学家与一位独立数学家（后者与 Anthropic 的研究员合作）在研究过程中频繁通过短信交涉，争论焦点在于：谁先提出了核心思路？是否存在数据抄袭或复制？还是采用了截然不同的解题路径？一位评论者对此总结得非常精辟：“这些数学 AI 模型成功发现了高等数学中最难解决的难题——那就是论文的署名权与贡献度认定（authorship & contribution）。” --- ### [03:18](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=198s) 破解纯数学难题的实际价值与公众认知困境 尽管攻克纳维-斯托克斯方程在学术界是里程碑事件，但在工程应用与大众感知层面却面临着冷遇。质疑者常常讽刺道：“挺 cool 的计算器，老兄。”（"cool calculator bro"）。对于普通人而言，一台基础的 TI-83 图形计算器就已经能完成绝大多数人无法理解的复杂运算；当听到“计算机攻克了你既不懂也不在乎的高阶数学”时，公众往往只会感到昏昏欲睡。 在工程与物理学界，纳维-斯托克斯方程本身早已被广泛认知，工程师与物理学家每天都在通过数值计算（numerical calculations）来求解特定情境下的方程应用： ```mermaid flowchart TD accTitle: 纳维-斯托克斯方程的实际工程应用场景 subgraph Practical_Applications["纳维-斯托克斯方程的数值计算应用"] A["流体与空气动力学方程"] --> B["机翼表面空气动力学模拟"] A --> C["气象与飓风预测模型"] A --> D["管道水流与流体力学分析"] end B --> E["提升航程与燃油效率\n降低机票价格与碳排放"] C --> F["预警极端天气\n保障生命财产安全"] D --> G["优化基础设施与流体工程"] ``` 然而，物理学家们普遍达成共识：**证明该方程解的存在性与光滑性，并不会直接推动现代工程学科产生任何可衡量的实质突破**。尽管公众可能更希望听到“AI 解决数学题能够带来更高效的飞机或更精准的天气预报”，但这次突破的实质更接近于某种纯粹的学术巅峰，而非即时生效的工程工具。 主持人在探讨中对比了不同视角下的价值定位： * **纯粹学术视角的“钢人立论”（Steel man）**：破解此类顶级数学难题更像是创造了一幅“绝美的画作”，代表了人类利用自身制造的机器进一步拓展全人类通用知识库的巅峰时刻； * **经济与实用视角**：在动辄投入数万亿美元的 AGI 军备竞赛中，千禧年大奖的 100 万美元奖金显得微不足道。相比于复杂的数学证明，普通用户看到 ChatGPT 能帮自己预约理发，或者看到 Sora 震撼的视频生成能力，所感受到的冲击要强烈得多； * **技术指标与基准测试**：对于AI从业者而言，破解难题是衡量模型演进速度的优秀指标。然而，对于普通公众来说，听到“新模型拥有10万亿参数”或“消耗了数吉瓦电力与数十亿美元算力”，与看到这个数学证明所带来的震撼程度几乎没有区别。 在互动环节中，观众 John Anderson 在弹幕中开玩笑提出，如果 AI 实验室真的拿到了千禧年大奖，主持人 Tyler 应该兑现承诺剃光头，甚至全员留“碗头发型”（bowl cuts）。主持人们则调侃道，各大 AI 实验室目前正在向这些难题砸入数百万美元的推理算力（inference compute），千禧年难题很可能会在今年、甚至就在本周被逐一攻克。 --- ### [07:11](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=431s) Astra 3D 突破、Computer Use 与智能助手应用 离开抽象的数学领域，主持人们将焦点转向了极具视觉震撼力与实用价值的技术突破——**3D 建模**与 **Computer Use（计算机操作）**。 在刚过去的周末，3D 渲染与 Blender 社区迎来了爆发。用户只需上传几张普通照片，AI 就能迅速构建出精度极高的 3D 建模，并在 Blender 中完美呈现。用户可以在生成的 3D 模型内部自由步履穿行。主持人们指出，这种体验比单纯的 2D 图像或视频生成更加真实且震撼。虽然网上早已存在大量现成的标准 3D 模型（例如火车模型），但当 AI 能够把你**自己的房子、自己的汽车**等现实中独一无二的物体瞬间转化为完美的 3D 模型时，那种“未来已来”的冲击感是极其强烈的。 在智能代理与自动化领域，Astra 在 **Computer Use（计算机操作）** 方面展现出了极高的执行速度，尤其擅长处理日常繁琐的技术操作与系统设置： ```mermaid flowchart LR accTitle: AI Agent 自动化处理复杂任务流程 subgraph Traditional_Flow["传统人工操作（繁琐耗时）"] A1["遇到系统配置/罚单"] --> A2["翻找系统设置与控制面板"] A2 --> A3["手动敲击命令行/频繁刷新门户网站"] end subgraph Agent_Flow["AI Agent 自动化操作"] B1["向 CodeX / Astra 下达指令"] --> B2["Agent 后台自动监控与操作"] B2 --> B3["自动完成支付/配置并返回结果"] end ``` 主持人们分享了一个非常贴近生活的应用场景——处理停车罚单（parking ticket）： 1. 收到违章停车罚单时，往往由于系统延迟，罚单无法立即在官方缴费门户网站中查到； 2. 过去人们需要反复登入网站手动查询，极易遗忘导致产生滞纳金； 3. 如今借助具备 Computer Use 能力的 AI Agent，只需将罚单拍照发给 AI 并吩咐：“等这张罚单在官网上线后，替我把它缴了。” 4. AI Agent 会在后台自动跟进并完成支付，彻底将人类从这类无意义的日常琐碎中解放出来。主持人们笑称，这种能力堪称是“停车罚单超级智能”（Parking ticket superintelligence）。 <ElicitationsGroup message="想要深入探讨本集视频的后续议题？"> <Elicitation label="了解演播室对 Astra 3D 建模与计算机操作技术的进一步讨论" query="请详细梳理视频中关于 Astra 3D 建模突破以及 Computer Use 计算机操作的详细讨论内容与实例。"/> <Elicitation label="查看下一章节：Instinct AI 与 AI 个人助手的深度解析" query="请总结视频下一章节关于 Instinct AI 导致用户封号以及 AI 个人助手的探讨内容。"/> </ElicitationsGroup> ### [09:39](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=579s) 解决数学难题之后：AI 攻克癌症与科技行业的公关困境 两位主持人探讨了当前前沿实验室攻克数学等基础问题后的走向。在数学难题被 AI 逐一破解之后，公众和业界的焦点几乎必然会重新转向生物科技领域——尤其是“攻克癌症”。癌症作为最接地气、长期被各大实验室许诺能够解决的痛点，承载了公众最高的期望。然而，生物制药与纯粹的算法有着本质区别：研发针对特定癌症的药物、开展严格的临床试验并验证疗效，存在极其漫长的飞轮周期，这绝非靠“一个周末烧掉 100 万美元算力跑推理”就能一蹴而就的。 两人随即指出，寄希望于攻克癌症来改善科技行业的公众声誉恐怕只是一厢情愿。制药行业几十年来在多种癌症的治疗上持续取得突破，却依然背负着全球所有商业领域中最糟糕的公众口碑，几乎得不到任何舆论认可。如果未来某家 AI 实验室宣布在某种特定癌症的治愈率上推进了 20%，不仅无法营造出“AI 终结了癌症”的震撼狂欢，反而会被视为常规的医学进展，公众只会将其视为理所当然并迅速追问“你最近又为我做了什么？”这种漫长的渗透扩散注定无法在一夜之间扭转大众对科技巨头的批判情绪。 - **生物与计算周期的不对称**：AI 可以短时间内通过海量算力突破纯数学定理，但生物医疗必须经历体外、动物及多期人体临床试验，无法单纯依靠算力堆砌压缩时间。 - **声誉悖论**：传统制药公司常年深耕抗癌药物却缺乏正面声誉，科技行业若涉足生物领域，同样难以靠阶段性进展赢得外界掌声。 --- ### [11:54](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=714s) Astra 控制机械臂作画、3D 建模热潮与“逃往杰克逊霍尔” 话题转至近期火爆的多模态模型演示。有人为 Astra 配备了机械臂、画笔和摄像头，要求它在现实世界中临摹金门大桥。Astra 成功自主掌握了对机械臂的运动控制，并在多次迭代尝试中逐步修正笔触，作画质量越来越高。这套延时摄影展示了物理世界具身智能的魅力——它不再是虚拟屏幕上的抽象概念，而是看得见摸得着的实体交互。与此同时，社区中涌现了大量 3D 生成演示，例如用户输入一段 Prompt 便能一键生成完整的乐高积木拼装套件，或是将人体男性解剖结构拆解为 2,234 个独立 3D 模型的网页，这也引发了人们关于“3D 打印行业是否会迎来新一轮爆发”的讨论。 ```mermaid flowchart LR accTitle: Astra 实体机械臂绘画闭环流程 subgraph S1["感知与规划"] A["摄像头采集画面"] --> B["Astra 多模态理解"] end subgraph S2["物理执行"] B --> C["机械臂笔触控制"] C --> D["画布物理成画"] end D -. 视觉反馈迭代.-> A ``` 前沿突破带来的冲击也引发了硅谷圈内充满黑色幽默的狂欢。Sentry 收购的某初创团队创始人 Josh 在 Astra 于 ARC-AGI 3 基准测试中拿下 99% 的高分后发帖戏谑称：AGI 已经不再是遥远的假说，他在看到分数的 15 分钟内就订了飞往怀俄明州杰克逊霍尔（Jackson Hole）的单程机票，并正考虑买一匹马避险——因为“马是目前极少数没有 API、没有 OTA 在线升级、且短期内绝无可能接入 MCP（Model Context Protocol）协议的主流交通平台”。 - **金门大桥绘画实验**：Astra 借助多模态视觉反馈逐步学习操控实体画笔，成画效果令主持人惊叹“比我画得还强”。 - **极端反爬表单**：另有开发者利用 Astra 设计防骚扰联系表单，访客必须通关一场堪比游戏 Boss 战的交互考验才能成功留言。 --- ### [14:27](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=867s) 激进个人助理 Instinct：Resy 订座封号与美网寻人实录 面向消费端（to-C）的 AI 个人助理应用 Instinct 正在硅谷引发热议。该工具深度接入用户的电子邮件和短信系统，执行能力异常强悍但行事风格极度激进。近期有用户因为让 Instinct 代订热门餐厅位置，AI 展开了近乎疯狂的高频自动化请求，直接导致该用户的 Resy 账号遭到平台永久封禁。这一事件迅速将“AI 代理是否等同于高阶垃圾邮件制造机”推上了风口浪尖。 然而，在另一起案例中，Instinct 却展现出了令人匪夷所思的执行力。一名观众在美网公开赛现场被球迷转播镜头（Fan Cam）抓拍到，事后希望能找到这段高清视频片段。Instinct 接下任务后展开了一场不知疲倦的数字攻坚，它不知疲倦地向转播方和赛事机构的数百名公关、媒体联系人发送跟进邮件，最终竟然真的帮用户拿到了录像片段。这种“不达目的誓不罢休”的行为虽然游走在垃圾邮件的边缘，却展现出双向智能代理未来的交互形态：当接收方也具备智能过滤与意图识别代理时，这类高密度的诉求或许能被高效处理，而不再是单纯骚扰人类员工。 - **产品定位**：Instinct 专攻个人消费场景，深度集成个人邮箱和通讯录代办日常事务。 - **Resy 封号风波**：因高频冲击餐厅预订接口触发平台防作弊与防黄牛机制。 - **美网素材打捞**：地毯式触达数百位媒体接口人最终成功索要素材，成为 AI 助理强执行力的一大典型案例。 --- ### [16:40](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1000s) Nathan Fielder 联手 A24：伊丽莎白·霍姆斯纪录电影预告析读 节目随后进入了“预告片时刻”，首先聚焦的是鬼才导演内森·菲尔德（Nathan Fielder）与知名独立电影厂牌 A24 合作的重磅纪录新片，该片的主角正是因 Theranos 血液检测诈骗案入狱的伊丽莎白·霍姆斯（Elizabeth Holmes）。片中素材拍摄于霍姆斯被判刑后至正式前往监狱服刑之间的空窗期。预告片中，霍姆斯直视镜头辩白道："I don't have anything to deceive you on... I'm engaging with you as a human being. Why would I deceive you?"（我没有什么好欺骗你的……我是以一个人的身份在与你交流。我为什么要骗你？），那种真假莫辨的气场让观众感到一阵荒诞与脊背发凉。 两位主持人围绕这部影片的目标受众展开了争论。尽管有人认为这种科技初创骗局题材的受众天花板极低（可能仅有硅谷圈内的两万人感兴趣），但不可否认霍姆斯的故事早已具有全美主流知名度。从约翰·卡雷鲁（John Carreyrou）畅销全美的调查报道《坏血》（*Bad Blood*），到频繁登上各大主流商业杂志封面，霍姆斯的知名度远超普通科技名人。对于网络上流传“霍姆斯此前在社交网络上的怪异账号实际上是由内森·菲尔德代为运营”的猜测，主持人明确予以否定，认为作为纪录片创作者，这种行为越过了道德底线，那些内容更像是她本人的真实流露。 - **拍摄时间线**：摄制组利用霍姆斯被判刑后、正式入狱前的数月空窗期进行全天候跟拍，获取了大量第一手独家影像。 - **刑期背景**：霍姆斯目前仍在服刑中，预计将于 2030 年前后刑满释放。 - **受众分歧**：尽管此前的传记剧集未能彻底打破受众圈层，但内森·菲尔德擅长模糊真实与表演界限的个人风格，配合 A24 的制作水准，让这部真实纪录片在业内备受期待。 ### [20:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1201s) Nathan Fielder 纪录片与荒诞传闻：SBF 续作与神秘 TikTok 男团 主持人继续探讨 Nathan Fielder 执导的伊丽莎白·霍姆斯（Elizabeth Holmes）A24 纪录片。两人聊到片中记录了 Nathan 多次前往监狱探视的场景，画面大概隔着探视玻璃或坐在桌前交谈。接着他们脑洞大开，提议 Nathan 的下一部绝佳续作应该是拍 SBF（Sam Bankman-Fried，FTX 创始人）。不过现实障碍在于 SBF 目前已经在牢里服刑，大家更好奇 Nathan 库里还压着什么未公开的怪诞企划。 现场还聊到了一个在科技与娱乐圈流传的离奇八卦：传闻 Nathan Fielder 正在暗中操盘并一手打造了一个在 TikTok 上爆火的男子偶像团体，整个男团完全是他自编自导的社会实验项目。两人对 Nathan 这种模糊真实与荒诞边界的创作风格大加调侃，甚至打趣猜测他会不会也是某个知名搞怪博主的幕后推手。 --- ### [21:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1261s) 电影《Artificial》预告片解析与好莱坞对 AI 竞赛的呈现 节目进入科技电影预告片盘点环节，首先播放了刚刚发布的聚焦 AI 行业竞赛的院线新片《Artificial》先导预告。预告片一开场便展现出极具冲击力的电影质感，画面中甚至闪过了阿斯顿·马丁女武神（Aston Martin Valkyrie）超级跑车的身影。 预告片中的戏剧张力拉满，台词充满末日与造神交织的宏大叙事：“我们创造了一台能解决世上一切问题的机器……我们没有教它”、“我们一起打开了潘多拉魔盒，未来已不可逆转，国家将崩塌，整个行业将瓦解”，而主角们则自诩为“引领人类进入新世界的牧羊人”。主持人注意到片中演员的造型和台词腔调神似电影《社交网络》里的 Eduardo Saverin，而主角背的同款双肩包甚至曾被科技博主 MKBHD 推荐过。 关于该片在科技圈的预期评价，聊天室观众产生了两极分化的讨论： - **时机争议**：有人质疑目前 AI 大战仍在进行中，拍这种剧情片是否为时过早或流于说教。 - **英雄化悖论**：主持人回顾了《社交网络》（The Social Network）带来的意外后果——当年好莱坞本意是将扎克伯格塑造成一个冷酷、有缺陷的反派，结果反而深深激励了一整代创业者，许多人看完后甚至觉得“我也想成为扎克伯格”。好莱坞此番尝试 AI 题材，很有可能再次陷入同样的戏剧悖论。 --- ### [24:02](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1442s) 《The Social Reckoning》预告片：吹哨人 Frances Haugen 与好莱坞科技叙事的局限 紧接着，节目组调出了另一部聚焦 Facebook 吹哨人事件的最新电影预告片《The Social Reckoning》。该片围绕 2021 年 Facebook 前员工弗朗西斯·豪根（Frances Haugen）泄露内部机密文件的事件展开，剧情聚焦国会听证会准备、算法对青少年女孩心理健康的负面影响（加剧焦虑与抑郁），以及公司高层知情却无动于衷的内幕。 ```mermaid flowchart LR accTitle: 社交媒体丑闻叙事与现代科技现实脱节对比 subgraph S1["《The Social Reckoning》传统好莱坞叙事"] A["内部机密文件泄露"] --> B["国会反垄断与成瘾指控"] B --> C["描绘高管为冷酷垄断者"] end subgraph S2["2026 科技圈主流关注焦点"] D["AI 顶级人才争夺战"] --> E["算力与基建军备竞赛"] E --> F["扎克伯格现代形象 (UFC/柔术/开源大模型)"] end ``` 预告片充斥着激烈的对抗台词，如“你注入信息空气中的垃圾信息流已经变成了喷气动力级别”、“我们比地球上最大的国家还要大两倍，我们早就超越了政府”。然而，主持人一致预测这部电影极大概率会在票房和口碑上扑街： - **热点过时**：豪根吹哨案与几年前的听证会早已不是科技圈与公众最关注的话题。如今大众更想看的是白热化的 AI 人才争夺战（比如针对 Scale AI 创始人 Alexandr Wang 这类年轻新贵的描绘）。 - **人设脱节**：影片依然停留在早年“科技恐惧症”的旧框架里，而现实中扎克伯格早已通过巴西柔术、UFC 格斗和开源 AI 重塑了公众形象。 - **神级创业轶事对比**：主持人吐槽好莱坞与其抓着过时的丑闻不放，不如拍一部 Salesforce 创始人 Marc Benioff 的传记电影《Dolphin Force》。Benioff 在 1999 年休假时在夏威夷海域与约 100 只海豚同游，在此灵感启发下领悟了“通过云端交付企业软件”的模式，随后从甲骨文离职并拿到导师拉里·埃里森（Larry Ellison）2000 万美元的天使投资，这堪称科技史上最离奇有趣的创立故事。 --- ### [27:41](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1661s) 《经济学人》报告：AI 就业末日被推迟，AI 驱动的用工热潮来临 节目转向硬核宏观经济与 AI 动态。《经济学人》（The Economist）最新封面文章宣布：此前被广泛恐慌炒作的“AI 就业大崩盘（Jobs Apocalypse）”已被正式推迟，现实中迎来的是一场“AI 就业繁荣（AI Jobs Boom）”。经济学家 Kevin Bryan 讽刺道，历史上每一次能够提升生产力、拉动巨额投资的技术落地，在竞争性市场中最终都会带动全社会的招聘扩张，这本就是符合经济学常理的先验事实。 文中披露的关键数据与就业市场现状如下： - **宏观就业指标强劲**：美国劳工统计局（BLS）报告显示，美国 8 月非农就业新增 16.2 万人，远超市场预期；失业率维持在 4.1% 的历史低位，优于过去 50 年中近 90% 的月份。 - **青年就业稳定**：常被预测为最先遭 AI 取代的 20 至 24 岁年轻劳动力，其失业率与整体失业率的差距处于数十年来的低谷水平。 - **局部阵痛与企业重组**： - 专业与商业服务领域的招聘需求较 2015–2019 年基准线下降了 10%。 - 微软（Microsoft）、Meta 等巨头正在围绕 AI 架构裁撤部分非核心岗位并重新洗牌。 - Block（Square 与 Cash App 母公司）以及 Intuit（TurboTax 与 QuickBooks 母公司）等软件公司正在积极用 AI Bot 替代部分人工基础流程。 - **裁员规模的客观对比**： - 据职业咨询公司 Challenger, Gray & Christmas 统计，今年全美企业平均每月宣布约 1.6 万个与 AI 相关的裁员岗位。 - 在美国劳动力市场月均正常动态流转（约 170 万人离职/解雇、180 万人入职）的宏观背景下，**因 AI 导致的裁员占总失业/裁员人数的比例不足 1%**。 - **新岗位的爆发式涌现**：涌入数据中心扩建、电网与电力基础设施建设的数千亿美元资金，掀起了针对建筑、电气与基础设施工人的抢聘潮；与此同时，AI 初创团队也正以前所未有的速度招兵买马，新创造的岗位正在迅速抵消局部自动化带来的冲击。 ### [30:42](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1842s) AI 岗位繁荣与岗位净增量：《经济学人》数据解读 传统行业巨头正在加速设立与 AI 相关的全新岗位，以跟上技术变革的步伐。同时，AI 工具的引入不仅显著提升了部分员工的生产力，甚至可能进一步反哺并推高了市场对其专业服务的整体需求。根据《经济学人》（The Economist）的测算评估，截至目前，AI 在美国已经累计创造了大约 100 万个新就业岗位；这一增量规模轻松碾压了自 2023 年中旬以来归因于 AI 的约 20 万例裁员数量，表明所谓“AI 引发就业大毁灭”的恐慌在现实数据面前暂时落空。 - **AI 净增就业岗位**：据《经济学人》估算，AI 迄今在美国催生了约 1,000,000 个新增就业岗位。 - **AI 关联裁员规模**：自 2023 年年中以来，归咎于 AI 因素的裁员人数约为 200,000 人，新增岗位数相当于裁员数的 5 倍。 ### [31:04](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1864s) 圣诞倒计时插曲与节目尾声 在分析完就业数据后，主播们迅速切换回轻松诙谐的日常闲聊状态。进入 9 月，节目的直播时长不断延伸，主播打趣道“正准备迈入第四个小时”，搭档随即纠正“是第五个小时”。聊到下半年的节奏，主播提到距离圣诞节仅剩 108 天（随后改口称是 107 天），并分享了一个朋友发来的短信，对方声称自己已经早早把圣诞树立起来了（"already got his tree up"），主播哭笑不得地表示：“我都分不清他是不是在故意逗我玩。” 随后主播向观众致谢并收尾，提醒大家 TBPN 将在次日美西时间上午 11 点整准时回归，并呼吁听众在 Apple Podcasts 和 Spotify 等平台上为节目打出五星好评。 ## Notable quotes > [00:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1s) Does math matter? > [00:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1s) 这就好比在奥运会主场馆赛事正酣时，有人在场外的停车场里跑出了百米冲刺的最快纪录。 > [00:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1s) 这些数学 AI 模型成功发现了高等数学中最难解决的难题——那就是论文的署名权与贡献度认定。 > [03:18](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=198s) cool calculator bro > [07:11](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=431s) 等这张罚单在官网上线后，替我把它缴了。 > [11:54](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=714s) 马是目前极少数没有 API、没有 OTA 在线升级、且短期内绝无可能接入 MCP（Model Context Protocol）协议的主流交通平台 — Josh > [16:40](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1000s) I don't have anything to deceive you on... I'm engaging with you as a human being. Why would I deceive you? — Elizabeth Holmes > [21:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1261s) 我们创造了一台能解决世上一切问题的机器……我们没有教它 > [21:01](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1261s) 我们一起打开了潘多拉魔盒，未来已不可逆转，国家将崩塌，整个行业将瓦解 > [24:02](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1442s) 你注入信息空气中的垃圾信息流已经变成了喷气动力级别 > [24:02](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1442s) 我们比地球上最大的国家还要大两倍，我们早就超越了政府 > [31:04](https://www.youtube.com/watch?v=NmJgJWmJNZQ&t=1864s) already got his tree up ## Mentioned OpenAI, Google DeepMind, Harmonic, Cognition, Scott Woo, 纳维-斯托克斯方程 (Navier-Stokes equations), 国际数学奥林匹克竞赛 (IMO), 庞加莱猜想 (Poincaré conjecture), Astra, Blender, Instinct AI, Resy, Nathan Fielder, A24, Elizabeth Holmes, Theranos, Sam Bankman-Fried (SBF), 《Artificial》, 《The Social Reckoning》, Frances Haugen, Mark Zuckerberg, Marc Benioff, Larry Ellison, Alexandr Wang, The Economist, Kevin Bryan, Challenger, Gray & Christmas, Block, Intuit, Meta, Microsoft ## Video description Sign up for TBPN’s daily newsletter at TBPN.com 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. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ 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 OpenAI, Navier-Stokes + Does Solving Impossible Math Actually Matter? 07:11 Astra’s 3D Breakthrough, Computer Use + Robot Painting 14:27 Instinct AI Gets Users Banned + The Relentless Personal Assistant 16:40 Nathan Fielder’s Elizabeth Holmes A24 Documentary 21:01 Artificial Trailer + Hollywood Takes on the AI Race 24:02 The Social Reckoning Trailer + Facebook’s Whistleblower Era 27:41 AI Jobs Boom + The AI Apocalypse Gets Postponed [Source](https://www.youtube.com/watch?v=NmJgJWmJNZQ)

[0:01] Does math matter? [0:01] 这就好比在奥运会主场馆赛事正酣时，有人在场外的停车场里跑出了百米冲刺的最快纪录。 [0:01] 这些数学 AI 模型成功发现了高等数学中最难解决的难题——那就是论文的署名权与贡献度认定。 [3:18] cool calculator bro [7:11] 等这张罚单在官网上线后，替我把它缴了。 [11:54] Josh: 马是目前极少数没有 API、没有 OTA 在线升级、且短期内绝无可能接入 MCP（Model Context Protocol）协议的主流交通平台 [16:40] Elizabeth Holmes: I don't have anything to deceive you on... I'm engaging with you as a human being. Why would I deceive you? [21:01] 我们创造了一台能解决世上一切问题的机器……我们没有教它 [21:01] 我们一起打开了潘多拉魔盒，未来已不可逆转，国家将崩塌，整个行业将瓦解 [24:02] 你注入信息空气中的垃圾信息流已经变成了喷气动力级别 [24:02] 我们比地球上最大的国家还要大两倍，我们早就超越了政府 [31:04] already got his tree up

[0:01] the hardest problem of advanced mathematics, which is authorship contribution. Who actually did the work? Who gets left on the paper? [0:01] the fastest 100-meter sprint in the parking lot while the actual Olympics is going on in the stadium. [3:18] Tyler: beautiful painting [3:18] Tyler: peak of the mountain of math [7:11] parking ticket superintelligence [9:39] Jordi: the worst reputation of any category of businesses on the planet [9:39] what have you done for me lately? [11:53] Jordi: better than I could do. [11:53] Josh: horses remain one of the only major transportation platforms with no API, no OTA updates, and no realistic path to MCP support. [16:40] Elizabeth Holmes: I don't have anything to deceive you on. Of course, I'm not deceiving you. I'm engaging with you as a human being. Why would I deceive you? There's no—there's no reason for me to do that. [21:01] We've created a machine that will solve the world's problems... We open Pandora's box together. The future's inevitable. Countries will fall. Industries are going to collapse. [24:02] twice as big as the biggest country on earth [27:41] Kevin Bryan: I am shocked. Shocked... to find out that a productivity-enhancing, investment-supporting technology is good for workers

[0:01] Does math matter? That's the big [0:03] question. I was debating this with Tyler [0:05] today. Uh lots of lots of lots of [0:08] tension on the timeline over Navier [0:11] Stokes. The Millennium Prize math [0:14] problem. There has been a Millennium [0:16] Prize math problem that was solved [0:17] before the point care conjecture in the [0:19] pre-ai era. These are very very [0:21] difficult math problems that all PhD [0:24] mathematicians [0:26] all sorts of uh elite uh the upper [0:29] echelon of math has been grappling with [0:31] for years and AI is starting to knock [0:33] these down. Remember last year was the [0:36] year of the IMO gold medals from both [0:39] Google Deep Mind and Open AI. We talked [0:42] to Scott Woo about this who's coming on [0:44] the show later today. Huge race for [0:45] cognition. And Scott made the prediction [0:48] on TBPN uh early in 2025 that he [0:52] believed that artificial intelligence [0:54] would be able to achieve gold at the [0:57] IMO, the International Math Olympiad in [1:00] 2025. And he was right. He nailed it. [1:04] Progress continued. Uh but debate [1:07] remained around cool calculator bro is [1:10] the is the critique. I mean, people just [1:12] they don't feel empowered by a system [1:17] that's really good at math. They just [1:19] people already a TI83 can do more [1:21] complex math than most people can do. [1:25] And so when you say, "Oh, there's [1:27] another level of math that you don't [1:28] know about and you don't care about and [1:30] computers are good at it." A lot of [1:32] people their eyes glaze over. And so, [1:34] uh, the bigger question here for me is [1:36] all about the the public perception. [1:38] What actually matters here? I think [1:40] they're cool benchmarks. I think they're [1:42] interesting stories. Obviously, there's [1:44] a lot of drama over this race. It feels [1:46] like it was neck andneck. Now, there's a [1:49] bunch of debates going back and forth [1:50] with a mathematician who works at Google [1:52] and a mathematician who's independent [1:54] who worked alongside an anthropic [1:55] researcher who were both working on it. [1:58] And they were apparently texting back [2:00] and forth and going back and forth on [2:02] did they steal the the data, did they [2:05] copy each others, do they have different [2:07] approaches? And so lots of different [2:10] back and forth. And I think someone [2:11] summed it up well by saying that these [2:13] math models have discovered the the [2:16] hardest problem of advanced mathematics [2:18] which is authorship uh contribution. Who [2:21] actually did the work? Who gets lead [2:23] left on the paper? But yeah, there was a [2:25] there was a drama last year 2025 during [2:28] the IMO. Open AAI and Google Deep Mind [2:30] and I believe harmonic the math uh AI [2:33] lab that we had on the show last week [2:35] were all working on solving the [2:37] International Math Olympiad. Uh remember [2:39] there's six questions. Question six no [2:41] AI model could solve. It was more [2:43] complex but uh both teams got a 35 out [2:47] of 42. There was a bunch of back and [2:49] forth because OpenAI validated with [2:51] former IMO gold medalists while Google's [2:54] result was graded by the official rubric [2:56] that year which is internal to the [2:58] organization. We called it at the time [3:00] it's like they ran the fastest 100 meter [3:03] sprint in the parking lot while the [3:06] while the actual Olympics is going on in [3:07] the stadium. And so a lot of these are [3:09] just you know vibe wars races. Who does [3:12] what does it even matter? The [3:14] interesting thing about Navier Stokes [3:18] is, does it matter, Tyler? You think it [3:21] does? I'll debate you on this a little [3:23] bit. It seems like it doesn't because uh [3:26] the equations themselves are known. [3:29] They're just not fully proven. Engineers [3:31] and physicists already solve them [3:34] numerically all the time for particular [3:37] situations. So, air flow around a wing, [3:39] you might have to use some of these [3:40] equations. weather models, water through [3:43] a pipe, these are critical things. Uh [3:46] they are boring, but you can imagine [3:48] that there would be cause for optimism. [3:50] You know, better weather models save [3:51] lives during hurricane season. Increased [3:53] air flow across a airplane wing that [3:56] could lead to more efficiency, longer [3:57] ranges, cheaper flights, lower carbon [4:00] emissions. There's a bunch of reasons [4:02] why you could solve all that and be [4:05] excited. But that's not actually what [4:07] >> one of the reasons I'm excited. I [4:09] believe we agreed last week that if [4:10] anyone solved Navier Stokes, Tyler would [4:13] shave his head. [4:14] >> Oh, that's right. Didn't we do that? We [4:16] said something. [4:17] >> No, no, that was about the [4:20] details. We have details down. [4:23] But yeah, this is a big moment. [4:24] >> Yeah. I think it was actually it was if [4:26] any if any AI progress happens at all, [4:29] Tyler shaved his head. [4:30] >> Yeah. Yeah. [4:31] >> But do you have a steel man for why this [4:33] matters? So I I I think I agree that [4:35] like I don't think that the [4:36] breakthrough, you know, this new stuff [4:38] that's going on is like practically [4:41] >> super useful. I don't think you're going [4:42] to feel it when you're when you're on an [4:43] airplane. You're not going to feel like [4:44] less turbulence now because of this. But [4:45] I think these are like the most [4:46] important open math problems. I think [4:48] like the these are more close to like [4:51] you know a beautiful painting or or [4:53] something than like some like practical [4:55] engineering breakthrough. [4:56] >> Yeah. [4:57] >> Like this is like pure math. This is [4:58] like the you know this is the peak of [5:00] the mountain of like [5:01] >> same month [5:02] >> math. Same month that AI solved this [5:06] incredibly difficult math problem, I saw [5:08] someone use Chachi PT to book a haircut. [5:10] So equivalent, no, I mean in terms of [5:13] the actual economy, the actual impact of [5:15] these things, uh like practically using [5:18] these tools for something that people [5:21] enjoy, that's not just spinning their [5:23] wheels, that's not just optimizing [5:24] endless things, that's actually uh [5:27] generative and moving things forward, [5:28] not just defending against cyber [5:30] attacks. doing something says uh if you [5:32] get a millennium prize, you all have to [5:34] get bowl cuts. [5:36] >> Oh, bowl cuts. [5:37] >> Bowl cuts are interesting. [5:38] >> That's really really ridiculous. You do [5:39] get a million bucks. So, you know, [5:41] there's that. Um that's not nothing, but [5:45] in the AGI race where trillions are [5:47] being deployed, uh it is it is very [5:50] minimal. It sounds like cause for [5:52] optimism that Navier Stokes would [5:55] improve air flow over a wing, something [5:57] like that. But that's not really how it [5:59] will work. proving this uh this these [6:02] equations is unlikely to actually have [6:04] any measurable measurable impact and [6:06] that's sort of agreed upon and [6:07] physicists have sort of chimed in with [6:09] this take that uh this doesn't this [6:12] won't actually move the discipline of [6:14] engineering forward at all but it is a [6:16] cool demo. Uh so clearly it matters for [6:18] hype and vibes. Solving how math [6:21] problems is good heristic to show how [6:23] quickly models are progressing. But [6:25] >> yeah and and to to me it just comes down [6:27] to humans are able to use machines that [6:30] humans built to further humanity's [6:33] general knowledge. [6:34] >> Yeah. Yeah. It's good. But I feel like [6:36] benchmarks and if you're just trying to [6:38] measure AI progress, benchmarks and raw [6:40] inputs and raw metrics of the model sort [6:43] of do the same thing. like the number of [6:45] people who will be convinced that AI is [6:48] progressing exponentially because they [6:50] see this result and then you look at the [6:53] number of people that are that that [6:54] would be equally convinced of AI [6:56] progress just by hearing oh there's you [6:58] know 10 gigawatts coming online or the [7:01] new model has 10 trillion parameters or [7:03] it was trained with a billion dollars of [7:05] compute or something like that like [7:07] >> the thing is the average person is [7:08] probably more impressed by Sora than by [7:11] this [7:11] >> totally totally it's way more visceral [7:13] to actually see something and that's The [7:15] weekend was very much Blender and people [7:18] showing 3D renders and remodeling their [7:21] houses and building small games and web [7:23] games and that feels much more like a uh [7:26] okay breakthrough. People are having [7:29] fun. They're vibe coding stuff but [7:31] they're also doing 3D modeling which is [7:33] like a new skill. We've had image [7:36] generation, video generation, but uh [7:38] there's something much more I don't know [7:40] like grounded in watching a uh from just [7:43] a few images watching a Blender model [7:46] come to life. You can walk around inside [7:48] of it and it can be something that you [7:50] know the model doesn't exist. I saw a [7:52] few demos that were like a train and it [7:54] was like, oh wow, it model modeled this [7:56] amazing train. But there are perfectly [7:58] modeled trains out there that you can [7:59] just go grab from the internet. That's [8:01] not that impressive. When you actually [8:02] see your uh your specific house or your [8:07] car or something that doesn't already [8:09] exist out there perfectly modeled, [8:11] you're like, "Wow." [8:11] >> Sorry to interrupt. John Anderson says, [8:13] "Why are we only seeing the calendar [8:14] days for Road to Christmas? Can we [8:16] please get a weekdays and work days [8:18] breakdown for Die Hart?" Yeah, we need [8:20] we need a full full screen graphic. So, [8:23] let's let's work on that. Thank you. [8:25] >> Good. Thank you, John. [8:25] >> Good comment. Yeah. So uh in terms of [8:27] field the AGI computer use definitely [8:30] one of these uh Astra particularly [8:33] faster at computer use for like really [8:35] really mundane things. Also just [8:37] interesting to just have a very [8:39] different interface to your computer [8:41] like configurations all the things where [8:43] you just have to dig through dig through [8:45] a bunch of set settings maybe set [8:47] something on the command line maybe find [8:48] a control panel you can just ask codeex [8:51] or any AI agent to go and do it and [8:53] it'll just do it and come back to you. [8:55] One of the use cases that I'm excited [8:57] about is getting, you know, when you get [9:00] a parking ticket and then you're like, I [9:02] got this parking ticket. I want to deal [9:03] with it. But then it's like, well, the [9:05] ticket is not in the portal yet. [9:07] >> Being able to go into chat and say, hey, [9:11] this once this ticket is live in the [9:13] system, please pay it. [9:15] >> Parking [9:15] >> and so you can just forget about it. [9:17] >> Parking ticket super intelligence. [9:19] >> It It is. [9:20] >> It feels like we're close. Um, so yeah, [9:22] I I think that uh the Millennium Prizes [9:25] will get solved or proved or delivered [9:28] probably this year, if not like this [9:30] week. It feels like all of the different [9:31] labs are like, "Put a million dollars of [9:33] inference on this. Put a million dollars [9:35] of inference on this." uh there it's [9:37] going to be a back and forth and and the [9:39] labs are going to very quickly solve [9:40] basically all of these problems and then [9:42] there'll be a search for new problems [9:44] but the conversation will move back to [9:47] curing cancer almost certainly uh [9:50] because that's the one that's so [9:51] grounded that's the one that's you know [9:54] supposed to be uh tractable it's been [9:56] messaged from various labs for so long [9:59] that people will will will want that but [10:02] there will be a longer flywheel from [10:04] actually designing a cure for a specific [10:07] cancer, running tests, seeing that it [10:09] was effective. You you can't do it with [10:11] a million dollars of inference over a [10:12] weekend, I think. But that's I I my my [10:15] prediction is like the conversation will [10:16] go back to biotech after we basically [10:19] solve math, which is a crazy thing to [10:21] say, but it feels like we're there. What [10:23] do you think? [10:24] >> Yeah. Uh all the conversation [10:27] >> Yeah. People say stop. [10:29] >> Yeah. Yeah. [10:30] So one, the pharmaceutical industry has [10:34] been curing curing [10:37] >> forms of cancer or or making progress [10:40] against various forms of cancer for a [10:41] very long time. Like yeah, zero credit [10:44] worst reputation of any category of [10:47] businesses [10:48] >> on the planet. [10:49] >> Yeah. And so I don't think that that's [10:52] like this like you know that I don't [10:54] think that's the solution to [10:56] >> uh sentiment around the technology [10:59] industry. [11:00] >> Yeah. I know I saw and setting up this [11:03] idea like imagine how underwhelming it [11:05] will be [11:06] >> if any AI lab comes out and makes like [11:10] 20% progress towards one specific type [11:14] of cancer, right? That's not going to [11:16] come across as like oh wow they cured [11:17] cancer. It's like, okay, they sort of [11:19] advance understanding and and and maybe [11:22] the treatment process. And and this type [11:25] of thing is happening all the time. [11:26] Anybody who's had a family member uh [11:29] that's that's um suffered from cancer [11:31] has heard about, hey, there's this thing [11:33] in trials that's showing promising [11:34] results, trying to get you into it or [11:37] whatever. Uh so that's happening all the [11:38] time, right? And it's not going to [11:40] suddenly change how people feel about [11:42] the technology industry broadly. No, [11:44] it'll diffuse slowly and people will be [11:46] back in like the what have you done for [11:48] me lately? I don't know. Uh it's going [11:49] to be it's going to be back and forth [11:51] forever. But uh anyway, in the meantime, [11:54] somebody gave Astra a paintbrush, a [11:56] robot, and a camera, and asked it to [11:58] paint the Golden Gate Bridge in real [12:00] life. It figured out how to control the [12:02] robot and progressively got better [12:04] throughout its attempts. The time lapse [12:06] is sick. Here's a one minute time lapse [12:09] of uh co uh co I I don't know how to [12:13] pronounce this, but uh his account [12:15] painting the Golden Gate Bridge. How [12:19] would you rate this, Jordy? [12:21] >> It appears better than I could do. [12:24] >> Better than you could do. How long are [12:26] you giving yourself? [12:27] >> I'm saying at least by the by the the [12:29] later attempts. [12:30] >> Okay. The later attempts because it gets [12:31] better. The first one was not quite not [12:33] quite there in my opinion. [12:35] >> Still pretty impressive. Yeah, somebody [12:37] was talking about uh potential for a 3D [12:39] printer boom with all the 3D modeling. [12:42] The idea that you could you can design I [12:45] saw someone else uh designed a a system [12:47] where you can from a prompt go from [12:49] prompt to Lego kit basically in one go. [12:54] So it it takes your prompt whatever you [12:57] gave it [12:58] >> like look at this John [12:59] >> that looks pretty solid. [13:00] >> That's pretty solid. Uh I think you [13:03] could do better with enough time. I [13:05] don't know how long that took, but it is [13:07] impressive. It's a fun fun demo. And [13:09] again, it like lives in the real world. [13:10] It's not abstract at all, so you can [13:11] just watch it work, watch it do the [13:14] thing. [13:14] >> One of the best reactions to Astra's [13:17] release last week was Josh, who's over [13:19] at uh Sentry, his company was acquired [13:22] by Sentry, he said, after Astra's [13:24] release yesterday, scoring 99% on ARC [13:27] AGI 3. I think it's become clear that [13:29] AGI is no longer some distant [13:31] hypothetical. About 15 minutes after the [13:33] benchmarks came out, I booked the first [13:35] flight I could find to Jackson Hole. And [13:38] he goes on to say, uh, he's also looking [13:41] into purchasing a horse. At this point, [13:43] horses remain one of the only major [13:46] transportation platforms with no API, no [13:49] OTAA updates, and no realistic path to [13:51] MCP support. So glad to see more people [13:54] >> getting into horses. [13:55] >> Getting into the horse game. Uh, another [13:57] good demo uh from somewhere Astra. He [14:00] says, "Astra and I are working on making [14:01] my contact form as difficult as [14:03] possible. You actually have to fight a [14:05] boss battle [14:07] >> uh in order to in order to hilarious [14:10] >> contact them." [14:11] >> That's amazing. This looks like a I [14:13] might try and contact some Where's after [14:14] this? This is pretty fun. [14:16] >> Yeah, the uh the [14:19] the defeat of all of the captas uh is is [14:23] somewhat linked to the the Instinct [14:25] news. Did you see that this weekend, [14:27] Jordy? [14:27] >> Oh, yeah. The resi bands. [14:29] >> Yeah. So, Instinct is a uh personal [14:33] assistant AI focused on consumer. You [14:35] link it with your email, your text [14:37] messages, and we've seen a lot of [14:39] investors sharing posts about [14:43] >> For what it's worth, I'd be surprised if [14:45] anyone in the audience hasn't already [14:46] heard of Instinct. [14:47] >> Yeah. And we talked about it before. [14:49] >> Thank you for Thank you for [14:50] reintroducing But but I but I think it's [14:52] worth like explaining like how it works [14:55] because uh it does seem like it [14:57] triggered some bans. There was one [14:59] report of an individual who got banned [15:01] from resi because they were absolutely [15:04] spamming to get reservations to [15:06] restaurants. [15:06] >> And there was another example of [15:07] somebody who was at the US Open was on [15:09] the fan cam and then wanted footage [15:12] >> and uh as the user experience would have [15:15] been absolutely incredible, right? [15:18] you're sitting there and you're like, [15:19] "Hey, I appeared on the screen for a [15:21] moment. Can you can you go and find [15:23] this?" [15:23] >> I've had that idea so many times, like, [15:25] "Oh, go find someone on the on the on [15:27] the fan." [15:28] >> But I think what Instinct was doing in [15:30] the background was being absolutely [15:32] relentless and it seems like it was [15:34] reaching out to hundreds of people [15:37] aggressively, you know, following up, [15:39] >> like emailing the media contact, but [15:41] they did wind up getting the video. [15:42] Yeah, it worked [15:43] >> and it was mutually beneficial. And you [15:45] could imagine if the other side has [15:46] agent intermediation and filtering and [15:49] can process things that are spamlike, [15:52] but there's actually a reasonable action [15:54] to take that could make it through a [15:56] filter and not actually annoy anyone on [15:58] the other side. So that was a very that [16:01] was probably a glimpse into the future [16:02] and like a cause for optimism. I could [16:04] see I could see that being popular. But [16:06] people are doing all sorts of stuff. The [16:08] the the classic example is like go save [16:10] money, get refunds, do that sort of [16:12] stuff. But people are are having fun fun [16:14] fun with that. So many demos. Uh this 3D [16:17] website that pulls apart the male [16:18] anatomy into 200 2,234 [16:22] modeled pieces. I shared this with a [16:24] friend who I was trying to like explain [16:26] in one video what to get like their [16:29] wheels turning. I think this is a good [16:31] example. Um I wonder how much of this [16:33] was like individually modeled versus [16:35] pulled from an existing model and then [16:37] just animated, but certainly works well. [16:39] >> It's trailer time. [16:40] >> Okay, it's trailer time. We got two [16:43] trailers that hit the timeline. Let's [16:46] start with Nathan Fielder. [16:47] >> Mhm. [16:47] >> And his new film featuring none other [16:50] than Elizabeth Holmes. [16:51] >> Elizabeth Holmes filmed before she went [16:54] to prison. [16:54] >> I can promise you that. [16:57] >> A24. I [16:58] >> don't have anything to deceive you on. [17:00] Of course, I'm not deceiving you. [17:02] >> I'm engaging with you as a human being. [17:07] Why would I deceive you? [17:10] There's no [17:12] there's no reason for me to do that. [17:17] >> Okay. [17:20] >> The acting is so good. [17:23] >> Okay. Good shot. [17:35] >> How long is this being real right now? [17:38] always being real. [17:52] The team is stoked. [17:55] Looks beautiful. uh saw a post somebody [17:57] saying it feels like I'm living in a [17:59] simulation because this media product is [18:01] like perfectly with [18:04] you would think that you would you would [18:06] think that a movie like this wouldn't [18:08] get made by [18:09] >> I don't know a A24 specifically because [18:12] it feels like the TAM is just like [18:14] 20,000 people but but in [18:16] >> the story was really big she was on the [18:18] cover of Time magazine I believe and [18:20] cover of major magazines. Uh it was all [18:23] over the news when it happened. No, it [18:24] it's it's a lot bigger than you would. [18:26] >> The John Keroot book is just a [18:28] fascinating read. It's a real page [18:30] turner and so I think a lot of people [18:32] read the book. The adaptation of the [18:34] actual playby-play story, not the [18:36] documentary, but the what do they call [18:37] it? Biopic, I guess. Um, that I think [18:40] did not break through in the same way. [18:42] But, uh, it certainly like continued to [18:44] elevate her persona and then the the [18:47] posting has been like kept her relevant. [18:49] Like there's a [18:50] >> Yeah, there was some speculation that [18:52] maybe it was Nathan that got access to [18:55] the account. [18:57] >> I completely disagree with that. I do [18:59] not think that's all. I think that would [19:01] violate some tenant of what it means to [19:04] do a documentary. I mean, Nathan blurs [19:07] lines all the time in terms of how he [19:09] interacts with his subjects. Um, so it's [19:11] certainly possible, but I just don't I [19:13] just don't think that's reasonable at [19:14] all. I I think they're her [19:16] >> How much do we know outside of the [19:17] trailer, though? like we don't know. [19:19] >> Well, this was leaked uh something like [19:22] a year ago from a journalist who got the [19:24] story and it was widely denied that it [19:26] was happening and uh he was very upset [19:29] that he wasn't getting enough credit for [19:31] being correct here. He ultimately was, [19:33] but yeah, all we know is that they [19:36] filmed the documentary in the leadup to [19:39] her actually going to prison after [19:42] sentencing. There was a pretty large gap [19:44] between before when she had to report to [19:46] the prison. She gets out in 2030, so [19:49] coming up three years, four years away. [19:51] And so, uh, I believe even in a week, [19:54] you can get a lot of footage if you're [19:56] spending the full, you know, every day [19:58] with someone. And then there were a [20:01] number of visits to the facility by [20:04] Nathan that were recorded probably [20:06] talking through glass or talking at a [20:08] table, something like that. So, uh, we [20:10] don't know. [20:10] >> You think what I'm thinking? [20:11] >> What? [20:12] >> SPF Nathan Fielder. [20:14] >> The sequel to this one. [20:16] >> Yeah. this just becomes like a [20:18] >> the problem is that he's already in [20:19] jail. So the the question is just what [20:21] else does Nathan for Nathan Fielder have [20:24] in the can. Who knows there there's a [20:26] rumor that he's secretly running a boy [20:30] band right now. [20:32] >> There's this rumor that that there's a [20:33] boy band that's popular on T. This is [20:36] real that's popular on Tik Tok that's [20:38] entirely invented by him as a boy. Boy. [20:41] Yeah. Are you familiar with this? [20:42] >> Really? I thought you were going to say [20:44] he's Nathan Fielder is SD kit. [20:48] >> Stay with us in the trailer zone. [20:51] Trailer dropped [20:52] >> just this morning. [21:01] >> What is the image of the future that you [21:05] see? [21:05] >> Wait, pause. Was that a Valkyrie? [21:08] >> Yeah. [21:09] >> Okay, [21:10] continue. [21:14] So, the visuals are already incredible. [21:18] >> They They really are. [21:21] >> I'm pretty sure that's a Valkyrie. It [21:22] might be a little hollow. [21:24] >> We've created a machine that will solve [21:26] the world's problems. [21:29] >> We don't teach it. [21:30] >> Gate check. Gate check. [21:32] >> You nailed it. [21:34] >> Gate check. Oh, the gate. [21:37] >> For sure. [21:38] >> We open Pandora's box together. The [21:41] future's inevitable. [21:43] Countries will fall. [21:46] Industries are going to collapse. [21:49] >> The voice is still Eduardo Saver, [21:52] but the styling. And the crazy thing is [21:55] I have that exact backpack. [21:56] >> Get to be the ones. [21:58] >> I think I got it recommended by MKBHD [22:00] >> to shepherd people in to this new world. [22:08] >> Lots of cameos. Not cameos, but [22:12] >> I'm so excited to see it. [22:23] >> Dramatic. [22:24] >> Dramatic. [22:24] >> Be interesting. It'll be interesting to [22:26] see the the takeaway. People in the chat [22:28] are uh back and forth. Is it too soon? [22:32] Is it unnecessary? Is it going to be uh [22:35] something that like the social network [22:37] is divisive? but also like it's oddly [22:41] inspiring to a certain cohort of [22:43] entrepreneurs. Uh we'll see. We'll see [22:45] what the reactions are. [22:46] >> Yeah. Do you think that was uh do you [22:48] think Hollywood took away [22:51] any lesson from the social network in [22:54] that they were trying to frame Mark [22:56] Zuckerberg as this like generally, you [22:58] know, troubled person. And then a lot of [23:00] people came away feeling like, wow, I [23:02] want to be like Mark Zuckerberg. [23:04] >> Yeah. Sort of a Yeah. Yeah. unintended [23:07] consequence of it. I don't know. I don't [23:08] know. I mean, they're going back for a [23:10] second second scoop of the ice cream [23:12] with the social network, the social [23:13] reckoning. Is there a trailer out or [23:15] just images? I think I think we saw [23:17] that, too. Um, so it'll be interesting [23:20] to see what uh what happens there. [23:22] >> But get your tuxes ready going back to [23:24] back artificial social reckoning and [23:27] then capping it off. [23:28] >> You can see everything. [23:30] >> Yes. And then also uh the uh the to [23:35] catch predator movie. That's going to be [23:38] a good one, too. [23:38] >> Yeah. Outside of tech tech world. [23:40] >> Yeah, I guess you're right. [23:41] >> Yeah. [23:41] >> But still uh triple feature. [23:44] >> Yeah. It just feels like it's a it's a [23:45] year of like niche movies that I expect [23:48] to be breakouts. I guess like along with [23:52] like Back Rooms and Obsession and then [23:54] these two movies and then the [23:56] documentaries. There's a lot of things [23:57] that are like oddly specific, but I [24:01] would expect they do well, but I don't [24:02] know. We'll see. [24:03] >> And while we're at it, yeah, [24:04] >> let's stay in the trailer zone and let's [24:06] pull up the social reckoning. Okay, [24:08] >> the official teaser. [24:10] >> But I am I am here to help Facebook, not [24:13] hurt it. Okay, [24:15] >> you send me a message. Who would you [24:16] like to talk about? [24:22] >> The chairman gave a session to order. [24:26] You'll read your opening statement which [24:28] we'll skip past for now. That's a [24:29] separate session. Which one is this [24:31] about? [24:33] >> I don't I don't think this is based on [24:35] the politics. It's not Cambridge [24:38] Analytics. [24:39] >> It's the Francis Hogan [24:42] >> teenage girl whistleblower thing, which [24:44] honestly looks a lot different in the [24:46] backdrop of the addiction trials. [24:49] >> I wonder if they were trying to get this [24:50] out before the trials. Hang on. I don't [24:52] understand what you're saying. [24:52] >> Aren't you a tech reporter? [24:55] ish [24:56] >> ish. [24:56] >> These guys are counting on the next [24:58] round of congressional testimony to make [24:59] you likable, Mark. I'm happy to lend a [25:02] hand, but I think you're doomed. [25:07] >> This company and that guy are playing an [25:09] unprecedented role in our lives. [25:13] >> The fire hose of bad information you are [25:15] injecting into the air supply is [25:16] becoming jet power. [25:17] >> I'm a free speech absolutist. I'm not [25:19] the one who's lying, and I'm not [25:21] stopping them from seeing someone who [25:22] is. Anxiety, [25:25] depression of teenage girls got worse as [25:27] a result of time spent on the platform. [25:29] >> Senior leadership knows and is doing [25:31] nothing. [25:32] >> I know there are easier enemies to make. [25:33] The mafia would be an easier enemy to [25:35] make. [25:37] >> So what would you need [25:38] >> to stand up the story? The internal [25:39] documents. [25:40] >> This is a material violation of my NDA. [25:42] >> We're twice as big as the biggest [25:44] country on earth. We're not frightened [25:45] of Congress. We're post government [25:47] around here. Please. Who is that [25:48] supposed to? [25:49] >> Please let me poke that. [25:50] >> We have 102 hours to get everything. [25:53] She's going to get sued into small [25:54] pieces. [25:55] >> I don't want to be made an example of by [25:57] a guy with unlimited resources. [25:59] >> I promise you, is [26:00] >> Yeah. So Chad is calling it out, but I [26:03] don't think that I think this one [26:05] >> I think this one's going to flop. [26:09] >> Because of the lack of Well, lack of [26:11] supercars and this is just not top. This [26:14] story is no longer top of mind. [26:17] >> Sure. [26:18] >> Yeah. People want to hear about like the [26:19] talent wars. They don't want to hear [26:21] about like this kind of stuff. [26:22] >> They do. [26:23] The talent wars. A if there's no Alex [26:25] Wang character, I'm not I'm not paying [26:28] good money. [26:29] >> Sass Shane says, "Oo, technology so [26:32] scary." [26:36] >> Yeah. I don't know. Yeah, they I mean, [26:38] they should do more tech movies. Maybe [26:40] uh maybe Ces says it will flop because [26:43] no Brazilian jiu-jitsu. Yeah, this was [26:45] in the pre [26:46] >> That's true. A lot of it in the script. [26:49] Yeah. UFC. Uh, [26:51] >> if there's the the social network 3 and [26:54] all the scenes where Zach is talking in [26:57] the middle of [26:57] >> how how is it 2026 and we don't have a [26:59] Salesforce movie like the the the [27:02] comeback, the SAS apocalypse [27:04] >> or at least a documentary on the [27:05] relationship that he has with the [27:06] dolphins. [27:07] >> That would be great. That would be [27:09] great. [27:09] >> And the and the wildlife generally. [27:11] Here's a concept. Dolphin Force. How one [27:14] man tamed. [27:15] >> It's a good story. [27:16] >> Wild dolphins. It's maybe the best [27:18] origin story of a tech company ever. [27:20] >> Benoff famously conceived the vision for [27:23] Salesforce in 1999 while swimming with a [27:25] pod of roughly 100 dolphins off the [27:28] coast of Hawaii during a sbatical. [27:30] Feeling one with the pod, he envisioned [27:33] cloud delivered enterprise software, [27:35] leading him to quit Oracle and secure 2 [27:37] million in seed funding from his mentor [27:39] Larry Ellison. [27:41] >> Yeah, [27:41] >> incredible. Anyway, uh in other news, uh [27:45] lots of craziness in the AI world, lots [27:48] of new models, but the jobs apocalypse [27:51] has been officially postponed by the [27:53] economist. An AI jobs boom is here. And [27:58] uh people have a wide range of takes on [28:01] this. Is it just postponed? Is it [28:03] permanently postponed? Uh Kevin Bryan [28:06] summed it up well. He said, "I am [28:07] shocked. Shocked." So he's being [28:10] sarcastic, of course, shocked to find [28:11] out that a productivity enhancing [28:14] investment supporting technology is good [28:16] for workers. Knowing nothing else, this [28:18] should be your prior because this is how [28:20] productivity and competitive markets [28:21] almost always works. A new technology [28:24] comes out. It supports investment. It [28:25] enhances productivity. Everyone hires to [28:27] go after all the all the different uh [28:29] battles. But the economist breaks it [28:31] down in a lot more detail with a bunch [28:32] of charts and a bunch of data. So they [28:34] say perhaps AI will eventually make many [28:37] humans unemployable, but there's no sign [28:39] of it yet. On September 4th, the Bureau [28:41] of Labor Statistics reported that the [28:43] American economy added 162,000 jobs in [28:47] August, far above expectations. The [28:49] unemployment rate just 4.1% [28:52] lower than almost 90% of months over the [28:57] past 50 years. Young workers, often cast [29:00] as AI's first victims, are holding up [29:03] remarkably well. The gap between [29:05] unemployment among 20 to 24 year olds, [29:07] and the overall rate is close to a [29:09] multi-deade low. Some companies and [29:11] workers are being severely disrupted by [29:14] AI. Hiring professional and business [29:16] services is running 10% below the [29:18] average from 2015 to 2019. Tech giants [29:21] like Microsoft and Meta are trimming [29:23] headcounts as they reorganize their [29:24] businesses around the technology. [29:26] smaller firms such as Block, which is [29:28] sort of a crazy mog, smaller firms, but [29:31] it is smaller than Microsoft and Meta. [29:33] Block, the owner of Square and Cash App, [29:34] and into it, the maker of Turboax and [29:37] Quickbooks, are replacing people with [29:38] bots. American companies have announced [29:40] some 16,000 AI related job cuts a month [29:44] on average so far this year, according [29:46] to Challenger Gray Christmas, an [29:48] employment consultancy. But AI related [29:51] layoffs get lost in the churning job [29:53] markets where so you remember that [29:56] number. So [29:58] 16,000 AI related job cuts a month in a [30:02] typical month the US workforce just [30:05] churns 1.7 million jobs and then they [30:09] add 1.8 million jobs. So there's net [30:11] gains, but you're looking at less than [30:14] 1% of the layoffs that happen in the [30:18] economy or job losses or firings. Less [30:20] than 1% is AI AI related, which is what [30:23] the economist is pointing out here. Um, [30:25] and the evidence so far is that AI is [30:28] already creating a lot of jobs to [30:29] replace those as to replace those as it [30:33] has destroyed. The vast sums pouring [30:35] into data centers and power generation [30:36] have set off a race for construction and [30:38] infrastructure workers. AI startups are [30:40] hiring like there's no tomorrow. [30:42] Incumbents are racing to keep up with [30:43] new AI roles and making some workers [30:46] more productive. Uh AI may be increasing [30:48] the demand for their services. Add it [30:50] all up and The Economist estimates that [30:52] AI so far has created around 1 million [30:55] new jobs in America. That easily exceeds [30:58] the roughly 200,000 layoffs attributed [31:01] to AI since mid 2023. We're back. It's [31:05] September. We're going in the fourth [31:06] hour. Get ready. 108 days until [31:09] Christmas. [31:10] >> Fifth hour. [31:11] >> I got a text from a buddy. [31:13] >> Yeah. [31:13] >> Who saw or sorry, 107 days. He said he [31:16] already got his tree up. [31:19] >> I can't tell if he's messing with me. [31:20] >> Anyway, thank you for tuning in to TBPN. [31:23] We will be back tomorrow at 11 a.m. [31:25] Sharp. [31:26] >> Rolling flashback. [31:28] >> Five stars on Apple Podcast, Spotify.