Anthropic Hits $17B Revenue as Enterprise AI Bottlenecks Threaten ROI

Anthropic营收达170亿美元，企业AI落地瓶颈恐危及投资回报率 企业软件采用正面临投资僵局，因现有AI智能体除常规软件编码外，尚无法处理脱节且依赖上下文的工作流。 在纽约证券交易所与彭博社的Joe Weisenthal连线直播中，John Coogan和Jordi Hays探讨了中期选举结果及预测市场如何与数据中心基础设施及产业政策交织。他们分析了华尔街在AI交易上的两难处境，将《财富》500强企业在采用落地上面临的阻碍，与Anthropic高达170亿美元的季度营收数据以及行业对真正自主多智能体软件的持续探索进行了对比。节目最后还快速报道了欧洲超市冰柜遭遇的未索要赎金式黑客攻击、Mark Walter考虑以30亿美元估值出售Cadillac F1车队、Northwood Space完成3.75亿美元C轮融资，以及Andrej Karpathy定制的HackerTube客户端。 Anthropic公布第三季度营收为170亿美元，折合年化营收运行率达680亿美元，同比增长1,127%，环比连续增长47%。 (17:03) 黑客针对西欧各地的商用冰箱和超市冷柜发起攻击，在未提出赎金要求的情况下推高温度（在瑞典甚至高达122°F），导致库存被迫报废。 (17:55) Bridgit Mendler创立的Northwood Space以20亿美元估值完成3.75亿美元C轮融资，由a16z和DFJ Growth联合领投，用于建设地面卫星通信网络。 (20:16) 据报道，Mark Walter正在探讨以30亿美元估值出售Cadillac F1车队，这突显出一级方程式赛车资产的交易看重的是切实的国际现金流，而非传统体育声望。 (20:16) SemiAnalysis发布了用于评估AI智能体处理现实经济任务能力的GDPx基准，目前领先的前沿模型得分仅为12%。 (23:32)

Broadcasting from the New York Stock Exchange, TBPN and Bloomberg's Joe Weisenthal examined how political resistance, enterprise adoption hurdles, and physical infrastructure constraints challenge the broader artificial intelligence and tech trade. Political Pushback Against Texas Data Centers (1:10) Texas has long been viewed as the epicenter of AI data center expansion, but local political support is eroding under voter concerns regarding massive power and land requirements. Bloomberg's Joe Weisenthal observed that Texas Governor Greg Abbott, historically a vocal backer of industrial development, has recently moderated his stance amid local pushback and talk of environmental moratoriums. Abbott's vulnerability is reflected in recent polling showing Democratic challenger Gina Hinojosa ahead by one point, lowering Abbott's betting market odds to near 60%. Weisenthal argued that prediction markets assigning a 40% probability to a Democratic win indicate substantial electoral uncertainty that tech planners must factor in. This political friction feeds into corporate maneuvering: some market observers suggest existing hyperscalers might quietly welcome local data center bans and tighter regulations. Constricting future compute supply could protect existing players' margins, prevent oversupply, and establish durable economic moats. Enterprise AI ROI and the Productivity Gap (12:07) Beyond software-native tech giants that achieve immediate software development gains, broader enterprise adoption of AI is stalling. Citing observations from Apollo Global Management's Torsten Sløk, the hosts noted that traditional industrial companies—such as tire makers, forestry businesses, and energy producers—struggle to turn generative AI into measurable revenue growth or workforce reductions. Demonstrating enterprise caution over operational risks and data management, Goldman Sachs has clamped down on non-technical employees "vibe coding" custom internal dashboards. The broader macroeconomic impact also remains unproven. Economists project potential productivity increases may peak around 3.5%, echoing prior technological waves where macroeconomic gains took years to appear. Autonomous multi-agent systems face practical limitations as well; SemiAnalysis's newly launched GDPx benchmark found that leading frontier models achieved only a 12% success rate on practical, real-world bug-patching and economic workplace tasks. Anthropic Q3 Financials and Growth Deceleration (17:03) Anthropic reported Q3 revenue of $17 billion during its investor day, representing an annualized run rate between $68 billion and $69 billion. While this marked a 1,127% year-over-year increase, sequential quarter-over-quarter expansion slowed to 47%, prompting discussion among investors regarding the durability of generative AI growth curves. Sustaining high infrastructure spending will depend heavily on whether upcoming model iterations, such as rumored video models like Fable 5.5, can successfully progress from reactive chatbots to fully autonomous enterprise systems. Ransom-Free Cyberattacks on European Cold Chains (17:55) Commercial refrigeration systems in supermarkets and food retailers across Northern and Western Europe—including Sweden, Belgium, Ireland, Luxembourg, Estonia, the Netherlands, and Finland—were hit by remote cyberattacks that deliberately spiked temperatures. Victims included IKEA, Swedish grocer ICA (where freezers reached 122°F), 13 SuperValu stores in Ireland, and Belgian butcher shops, resulting in substantial food spoilage. Belgian cybersecurity officials confirmed that no ransom demands were made, suggesting the campaign was motivated entirely by supply chain disruption and property destruction. The absence of financial extortion led security commentators to suggest that critical businesses may need to reconsider deploying internet-connected appliances for baseline operations. Strategic Capital Deployment in Space and Racing (20:16) Capital markets directed significant funding toward specialized physical networks, led by Northwood Space closing a $375 million Series C round at a $2 billion valuation. Co-led by Andreessen Horowitz and DFJ Growth under nine months after its Series B, the startup is expanding independent satellite teleport ground stations to serve commercial operators and defense programs alongside SpaceX. Concurrently, sports investment is shifting from prestige trophy assets toward cash-flow generation. Mark Walter explored a sale of Cadillac F1 at a $3 billion valuation, reflecting how Formula 1 team economics are increasingly evaluated on international commercial footprints and cash flow rather than traditional domestic team valuations. Even as frontier AI labs report rapid revenue expansion and specialized infrastructure commands premium valuations, the sector faces unresolved headwinds: uncertain enterprise ROI outside the tech sector, emerging physical vulnerabilities in internet-connected infrastructure, and mounting political resistance to data center power demands.

本期TBPN节目在纽约证券交易所现场探讨指出，尽管生成式AI领域仍录得惊人的资本投入与营收规模，但其前沿商业化正面临宏观政治阻力、企业投资回报率迟滞以及现实物理系统复杂性的严峻考验。 得州政治风向与数据中心供给博弈 (1:10) 彭博社评论员Joe Weisenthal指出，科技界与AI产业应密切关注得克萨斯州的选情演变。长期支持工业与技术扩张的得州州长Greg Abbott，在地方选民反对数据中心占用庞大电网与土地资源的压力下，言辞已出现软化迹象并开始顾虑环保暂缓令。相关民调显示其支持率收紧，预测市场上Abbott的胜选概率已滑落至约60%。Weisenthal认为预测市场能有效提炼并量化模糊的共识，不能将民主党挑战者在该传统红州40%的胜算视为小概率噪音。此外，讨论提出一种高风险企业战略猜想：已建立庞大算力的基础设施巨头或许暗中乐见监管从严或数据中心禁令出台，因为人为抑制未来供给能够形成护城河、保护其现有资本的定价权与利润率。而在华尔街层面，针对“AI交易”的利润究竟会被芯片硬件厂商、前沿模型实验室、企业软件集成商还是通用终端所捕获，投资者至今未达成共识。 企业AI落地瓶颈与Anthropic财报 (12:07) 阿波罗全球管理公司的Torsten Sløk观点显示，除美股“七巨头”等软件原生公司能迅速借智能体优化代码并实现财务回报外，多数传统标普500企业（如林业、轮胎制造、油气公司）正面临“AI消化不良”，难以依靠现有模型提升营收或削减人力。高盛集团已对非技术人员及交易员私自搭建“氛围式编程”（vibe coding）工具予以限制，以规避操作风险与数据合规隐患。结合“繁杂工作假说”，大多数现实职能并非重复代码，而由高度情境化的碎片任务构成，当前的AI系统仍属被动执行指令的附庸，缺乏自主管理预算与跨部门协调的能动性。SemiAnalysis新发布的GDPx基准测试同样显示，顶级前沿模型在实际工作与修障任务中的得分仅为12%。与应用端的转化迟缓相对照，Anthropic在Q3投资者日公布单季营收达170亿美元，年化运行率达680亿至690亿美元，虽录得1127%的同比增幅，但环比增速放缓至47%，引发市场对技术扩散曲线可持续性的辩论。 欧洲超市冷柜遭黑客攻击致温升损毁 (17:55) 节目报道了近期席卷欧洲的异常网络攻击事件：黑客远程入侵了瑞典、比利时、卢森堡、爱沙尼亚、爱尔兰、荷兰及芬兰等多国的商用制冷设施，蓄意调高冷藏与冷冻温度，导致数以千计美元的生鲜商品变质报废。受害者涵盖宜家（IKEA）、瑞典ICA超市、爱尔兰13家SuperValu门店以及比利时独立肉铺等不同规模的商户，其中ICA超市的冷冻柜温度一度被调升至122华氏度（50摄氏度）。该事件的核心异常在于黑客未提出任何赎金索求。比利时网络安全中心发言人Katrien Eggers证实，调查人员尚未确定攻击者身份，由于缺乏勒索信，可能永远难以锁定真凶，初步研判其动机纯粹为制造经济破坏与资产损毁。安全分析人士借此指出，商户或许需要审慎评估过度接入物联网的运营风险，并在关键基础设施中保留非联网的传统机械备份。 Northwood空间融资与F1资产重估 (20:16) 在商业投融资领域，由前迪士尼演艺人员Bridgit Mendler创立的卫星地面站网络初创企业Northwood Space宣布完成3.75亿美元C轮融资，投后估值达20亿美元，由Andreessen Horowitz（a16z）与DFJ Growth联合领投。该轮融资距其B轮不足9个月，筹集资金将用于履行企业及防务合同、引进工程团队并构建共享型全球卫星地面站网络，其定位是为商业运营方、美国政府及盟国防务项目提供独立于SpaceX星链（Starlink）以外的关键通信骨干。同时，针对体育商业媒体Sportico披露Mark Walter正探索以30亿美元估值出售Cadillac F1车队股权的传闻，主持人分析指出，F1车队有别于北美传统三大球类主要依靠城市归属感的“纯荣誉型资产”（trophy assets），在预算帽制度与全球巡回运作下，其估值逻辑正日益锚定于真实的国际赞助现金流与商业合作网络。 从地方选区对数据中心能源消耗的反弹、传统产业在AI回报率上的停滞，到无赎金物联网网络破坏，表明前沿科技在向实体经济渗透时正遭遇现实物理资产、电网负荷与基础设施安全层面的多维摩擦。

## Background 本期节目由科技播客频道 TBPN 于 2026 年 11 月 3 日发布在 YouTube 平台上。这一名为“Diet TBPN”的精简栏目由 John Coogan 和 Jordi Hays 主持，聚焦当天科技与财经领域的核心议题，主要涵盖了彭博社嘉宾对中期选举与预测市场的分析、AI 算力与投资回报率博弈、Anthropic 财报、欧洲超市冷柜遭黑客攻击以及 Northwood Space 融资等资讯。 ## TL;DR 节目在纽约证券交易所现场展开，邀请彭博社记者 Joe Weisenthal 探讨中期选举背景下得州数据中心面临的政治阻力，以及预测市场对主流共识的量化价值。随后，主持人和嘉宾深入剖析了科技巨头与传统企业之间显著的 AI 投资回报率鸿沟、当前多智能体系统的协作瓶颈，并解读了 Anthropic 狂揽 170 亿美元但环比增速放缓的第三季度财报。节目后半段还梳理了欧洲多国超市冷柜离奇被黑、Cadillac F1 股权出售传闻、Northwood Space 狂揽 3.75 亿美元融资，以及 Andrej Karpathy 借助 AI 极简构建定制工具 HackerTube 等行业前沿动态。 ## Atmosphere & style 现场氛围节奏紧凑、风趣犀利且充满华尔街与硅谷交织的行业行话。两位主持人与嘉宾之间配合默契，既能在高维层面上推演 AI 算力供需背后的政治博弈，又能随时用“把超市改成桑拿房”或“洗手间该不该摆圣诞树”的幽默吐槽调剂严肃话题。整体叙事兼具宏观财经的深度思辨与科技圈吃瓜的轻松感，听起来既有临场直播的生动活力，又不失专业内行的敏锐洞察。 ## Bulletin - [17:03](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1023s) Anthropic 第三季度财报公布，季度营收达 170 亿美元，年化运行率增至 680 亿至 690 亿美元，同比激增 1127%，但环比增速放缓至 47%。 - [20:16](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1216s) Northwood Space 在完成 B 轮融资不到 9 个月后，获 a16z 与 DFJ Growth 联合领投的 3.75 亿美元新融资，估值攀升至 20 亿美元，用于搭建全球地面站网络。 - [18:06](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1086s) 欧洲多国超市与冷藏设备遭远程网络黑客入侵，温度被恶意调高导致大量生鲜变质，由于攻击者未勒索赎金，事件被怀疑为纯粹恶意破坏。 - [25:59](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1559s) SemiAnalysis 推出聚焦实际工程 Bug 修复能力的全新基准测试 GDPx，当前顶尖 AI 模型得分仅约 12%，并指出 Google Gemini Carbon 版本在关键领域表现强劲。 ## Key takeaways - [00:02](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=2s) 在得州州长选战中，因选民对数据中心环保与电网负荷的反弹，原本亲科技的阿博特州长态度动摇，在预测市场中的胜率降至 60% 左右。 - [03:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=180s) Joe Weisenthal 认为预测市场的真正核心价值在于能够量化并提炼主流共识，其统计赔率远比主观猜测更能客观反映概率走向。 - [07:07](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=427s) 市场上出现“算力过剩阴谋论”，猜想科技巨头或暗中支持监管政策以人为压缩未来新增数据中心算力供给，借此守住出租利润率。 - [07:07](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=427s) 华尔街对“AI 交易”的终局落点毫无共识，对于模型层、软件巨擘、芯片厂商还是下游实际应用能截留超额利润仍存巨大分歧。 - [08:46](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=526s) 美国劳动生产力账面数据的短期暴涨，很大程度上源于企业激进裁员引发的人力成本骤降，而非技术本身带来了清晰的赋能拐点。 - [10:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=600s) 美国消费市场持续强劲，高净值家庭得益于资产升值、低收入群体依赖借贷，加之企业 AI 基建与政府巨额支出共同支撑了繁荣局面。 - [12:08](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=728s) 传统财富 500 强企业深陷 AI 投资回报率（ROI）困境；受制于“杂务工假设”，单点任务出色的 AI 无法直接替换复杂的真实岗位，高盛等机构已限制非技术员工盲目编写 AI 脚本。 - [15:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=900s) 当前自主智能体研发面临同质化瓶颈，业内指出亟须打破“克隆博士”式的孤立调用，转向引入竞争与对抗机制的自主多智能体编排架构。 - [17:03](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1023s) Anthropic Q3 营收虽同比飙升 1127%，但环比增速出现放缓迹象，资本市场正密切观望下一代前沿大模型能否带来新的生产力突破。 - [18:06](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1086s) 欧洲多国零售冷柜遭遇恶意温控篡改且无赎金勒索，凸显出基础设施物联网面临新型破坏性安全威胁，也反衬出非联网设备的优势。 - [20:16](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1216s) Northwood Space 逆势大举融资 3.75 亿美元，精准锚定 SpaceX 等火箭发射之外日益扩张的全球地面基站通信与卫星对接需求。 - [24:06](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1446s) Andrej Karpathy 利用 Claude 仅用数句极简提示词便搭建出过滤 AI 垃圾内容的定制版 HackerTube，印证了个人定制化软件时代的降临。 - [25:59](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1559s) SemiAnalysis 推出的 GDPx 测试专注于排查业务 Bug 等直接影响商业 ROI 的实战任务，目前顶尖模型整体通过率仅有 12%，揭示出落地能力的真实短板。 ## Detailed walkthrough ### [00:02](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=2s) 纽交所现场：中年期选举与得州数据中心政治风波 节目在纽约证券交易所（NYSE）现场进行直播。主持人 John Coogan 和 Jordi Hays 谈及 TBPN 最近被《New York Magazine》报道，自嘲道其实只是被顺带提了一句，更像是一个脚注。随后，他们迎来了彭博社资深财经媒体人、《Odd Lots》播客主持人 Joe Weisenthal。双方开玩笑探讨政治话题，Joe 直言虽然非常庆幸平时不必频繁谈论政治，但在金融市场与科技行业交织的今天，某些选举节点已避无可避。 讨论迅速切入即将揭晓的中期选举，外界普遍预计民主党人可能会迎来一个大胜之夜。在科技界尤为关注的得克萨斯州州长选战中，州长格雷格·阿博特（Greg Abbott）过去一向被视作数据中心建设的最后盟友，但近几个月来态度也开始产生动摇，政策上甚至出现带有环保诉求的暂停令倾向。政客们在面对选民对基础设施扩张的反弹时，不得不小心翼翼地保持中立或两面下注。 ```mermaid flowchart LR accTitle: Data Center Policy Dilemma accDescr: Political candidates balance tech expansion with voter environmental concerns. subgraph Politics["选举政治考量"] A["选民环保/电网反弹"] --> B["政治候选人态度动摇"] B --> C["推行临时禁令/环境审查"] end subgraph Industry["科技与资本诉求"] D["得州制造业/AI基建"] --> E["算力与数据中心扩张"] end C -.->|"政策不确定性"| E ``` 在得州政坛格局上，长期稳坐职位的阿博特虽然内心笃信科技发展与制造回流，但随着选情收紧，最新民调显示挑战者已领先一个百分点，阿博特在预测市场的胜率已滑落至 60% 左右。Joe Weisenthal 评价道：“He's down to about 60% in the betting market. So that rounds to 50% in my book.”（他在赌盘上的胜率降到了大约 60%，在我看来这基本相当于五五开。） - 得克萨斯州被科技圈与制造业视为未来中心，但如今没有任何候选人敢在选战中公然打出激进支持数据中心的旗号。 - 预测市场中阿博特的连任概率下挫至 60% 边缘，在竞争激烈的得州政局中已不再稳操胜券。 --- ### [03:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=180s) 预测市场的“常识提炼”功能与全国选举观察 对于预测市场的波动，Joe 阐述了他对市场的核心看法：他基本信任预测市场，因为大众集体智慧的定价往往超越个人的主观猜测。当看到赔率显示某件事有 20% 对 80% 的逆转时，就该明白这种小概率事件在统计上本就应该每五次发生一次。 面对批评者认为预测市场不过是在反映“主流共识”（conventional wisdom），Joe 提出了鲜明的反驳。他认为能够量化主流共识本身就极具价值——例如某候选人是占有 55% 还是 80% 的优势，能以明确的数字体现出来，从而帮助市场看清概率底牌。正如 Joe 所强调的：“Markets are really good conventional wisdom distillers and that's actually very valuable.”（市场是非常出色的主流共识提炼器，这其实非常有价值。）因此，当预测市场显示民主党在得州有 40% 的获胜概率时，任何人都必须严肃对待。 在其他地区的选情上： - **纽约州**：整体政治走向没有重大悬念，外界关心的总检察长选战难以掀起全国波澜；即使民主党全国表现强势，北方部分现任民主党议员可能因选民的“反现任情绪”表现略逊，但纽约绝非主战场。 - **选务混乱与意外因素**：亚利桑那州马里科帕县（Maricopa County）传出局部电力故障，同时一股强风暴正席卷美国腹地，外界关注恶劣天气是否会压低投票率。 - **社交网络推波助澜**：大家半开玩笑地戏谑这场风暴背后是不是数据中心或那些曾上节目的“天气控制创业公司”在捣鬼。同时，如果密歇根州左翼候选人阿卜杜勒·埃尔-赛义德（Abdul El-Sayed）成为极少数意外胜选的民主党人，推特上的争吵与舆论风暴必将彻底失控。 --- ### [07:07](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=427s) AI 算力过剩与背后的“五维棋”阴谋论 随着政治话题延伸至硅谷底层逻辑，主持人提出了一个引人入胜的“四维国际象棋”假说：科技巨头是否在暗中促成“蓝色浪潮”（民主党获胜）？其背后的动机是，目前各大科技公司过度建设了 AI 算力，随着大批芯片与算力集群上线，供给过剩隐现；如果促成民主党执政并出台针对数据中心与算力建设的严格禁令，就能人为削减未来新增供给、推高算力租赁价格，从而在短期内捍卫甚至抬高巨头的利润率。 ```mermaid flowchart TD accTitle: The AI Compute Bottleneck Hypothesis accDescr: Hypothesis of tech giants using regulatory bans to artificially constrain compute supply and protect profit margins. A["AI算力建设过剩\n(大量芯片与集群上线)"] --> B["政治游说/支持监管政策"] B --> C["出台数据中心建设禁令\n(人为制造供给瓶颈)"] C --> D["算力市场供应受限"] D --> E["维持/推高算力出租价格与利润率"] ``` Joe 称赞这一观点甚至已经达到了“五维空间”的思考深度，并借此切入华尔街当前最大的迷思：究竟谁能在“AI 交易”（AI trade）中真正赚到钱？目前市场对于 AI 商业落地的最终赢家毫无共识： 1. **模型实验室**：如 OpenAI、Anthropic 等前沿模型开发商是否能守住高估值与商业护城河； 2. **科技巨擘与企业联盟**：如微软加 Palantir 的商业落地方案对抗其他前沿闭源或开源阵营； 3. **硬件与芯片制造商**：英伟达等基础设施提供商是否会通吃整个行业的超额利润； 4. **实体经济与下游应用**：底层技术是否会迅速商品化（commoditized），使得所有利润最终流入传统行业的实际应用端。 在这场围绕 AI 终局的博弈中，Joe 指出市场对前景存在巨大的不确定性：“People talk about the AI trade, right, on Wall Street. And the thing that I don't totally understand is like, well, there's no consensus on who yet on who actually makes money.”（华尔街都在谈论‘AI 交易’，但我完全无法理解的是，目前对于究竟谁能在其中真正赚到钱，其实根本没有任何共识。）对某些极度乐观者而言，只要 AI 能够驱动 GDP 实现 20% 的惊人增长就算是成功，但对资本市场而言，寻找真正的价值沉淀点依然扑朔迷离。 ### [08:46](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=526s) 生产力悖论与美国经济的虚假繁荣 讨论从大众对 AI 带来生产力爆发的高预期切入。无论是 Anthropic 的员工还是普通的美国大众，对未来生产力跃升的预期都高得惊人，反倒是经济学家表现得格外冷静，仅保守预测生产力可能增长到 3.5% 左右。主持人与嘉宾调侃道，每当讨论 AI 对生产力的实际影响时，情况其实与当年的互联网或智能手机如出一辙——人们很难在宏观经济图表上直观捕捉到技术带来的确切生产力拐点。《金融时报》近期虽然刊登了一张显示美国劳动生产力正在加速并大幅超越其他国家的图表，但这背后的驱动因素并非单纯的技术赋能。 大家分析认为，这一波生产力数据的跃升很大程度上源于企业的大规模裁员。美国企业迅速裁撤了大量员工并用遣散支票替代，这种人力成本的骤降在短期内美化了人均产出和生产力账面数据。叠加过去几年美国宏观经济整体表现显著强于欧洲，两者之间的生产力差距便被进一步放大，很难将其直接归功于 AI 带来的技术红利。 * **经济学家的保守态度**：普通大众与科技圈预期极为乐观，但经济学家预测的生产力增速仅在 3.5% 左右。 * **数据美化效应**：裁员引发的人工成本骤减直接推高了短期人均产出指标，形成了生产力暴涨的表象。 --- ### [10:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=600s) “不可战胜”的美国消费者与全方位支出热潮 围绕美国消费市场的真实底色，演播室展开了生动的探讨。即便市场上充斥着各种黑天鹅传言与脆弱性担忧，但在纽约这类大都市的真实街景中，丝毫感受不到任何消费放缓的迹象。大街上随处可见排着长队去买 30 到 40 美元商品的人群，餐厅一位难求，洛杉矶周一上午十点唱片店门口甚至也会排起长队。尽管企业高管们从 2022 年起就年年老调重弹，宣称“消费者在购买时开始变得更加精打细算、开始消费降级”，但实际的数据和体感却依然火爆。 嘉宾将当下的经济划分为三个相互交织的支出部门：家庭部门、企业部门和政府部门。家庭部门呈现出鲜明的 K 型消费，高净值人群依靠股权资产升值持续套现消费，底层群体则通过扩大消费信贷强行支撑开支；企业部门正全力将资金押注在 AI 基础设施与技术采购上；政府部门则在老年人福利保障上持续巨额开销。众人还幽默地插话补充了眼下不可忽视的“第四大部门”——博彩与预测市场，并提到街头铺天盖地的广告词已经从“豪赌一把（Bet Big）”变成了“大举预测（Predict Big）”。 > “If you're walking around New York City, you could never be convinced that there is a consumer slowdown.” —— Joe Weisenthal --- ### [12:08](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=728s) 财富500强的 AI 投资回报困境与“杂务工假设” 华尔街投行与大型企业正面临 AI 支出回报率（ROI）的严重分化。阿波罗全球管理公司（Apollo）的首席经济学家托斯滕·斯洛克（Torsten Sløk）指出，除“美股七巨头”（Magnificent Seven）以外的标准普尔 500 指数企业普遍出现了 AI 支出的“消化不良”。科技巨头能够直接将 AI 智能体嵌入现有软件生态并迅速实现商业变现，但对传统行业——例如轮胎制造、林业开采或石油天然气企业来说，部署 AI 既无法直接拉动营收，也无法有效替代现有员工。 为了控制风险与挽救实际工时，各大机构开始紧急踩刹车。彭博社记者托德·吉莱斯皮（Todd Gillespie）披露，高盛集团（Goldman Sachs）正在严厉限制包括交易员在内的非技术员工自行使用 AI 搞“氛围感写代码”（vibe coding）搭建内部看板，理由是此举不仅存在数据安全与合规风险，还在严重损耗实际生产力。演播室由此提到了经济学界著名的“杂务工假设（messy jobs thesis）”：现实中的绝大多数岗位包含数十种细碎繁杂的任务，目前的 AI 充其量只能在其中一两项任务上表现出色，完全无法替代一个由多样化技能构成的整体岗位。 > “His latest is that basically Fortune 500 having a little bit of indigestion around AI spend outside of the Mag 7.” —— John Coogan ```mermaid flowchart TD accTitle: AI ROI Divergence Across Corporate Sectors accDescr: Comparison of AI investment adoption and ROI between Big Tech and traditional Fortune 500 companies subgraph BigTech["科技巨头（Mag 7）"] A["自有软件与技术底座"] --> B["快速集成 AI 智能体"] B --> C["立竿见影的正向 ROI"] end subgraph Traditional["传统 500 强（制造/能源/林业）"] D["复杂且碎片化的工作任务<br>(杂务工假设)"] --> E["AI 仅能完成单点任务"] E --> F["无法直接替代岗位 / 难增收"] F --> G["支出消化不良 / 叫停盲目编码"] end ``` --- ### [15:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=900s) 突破瓶颈：我们需要真正具备自主性的多智能体 演播室指出，眼下这一轮 AI 交易逻辑若要持续，就必须迎来一次核心能力上的跃升。目前企业所拥有的 AI 并非真正的智能代理，而更像是一百个坐在会议室里一言不发、彼此孤立且毫无主动性的“克隆博士”。当前的智能体在本质上都是受指令驱使的“子智能体（sub-agents）”，缺乏宏观目标导向下的自主权。理想的模式应该类似于 Meta 的自动化广告投放：管理者只需设定一个预算上限，智能体便能自主探索、规划并调配资源，以最具性价比的方式达成商业增长。 针对多智能体协作陷入瓶颈的现状，Google 的 Alex Dimakis 曾阐述过其背后的本质难题：当前的智能体彼此同质化严重，就像让一百个受过一模一样训练的双胞胎博士协同工作，根本无法产生真正的团队博弈。业界必须引入类似“角斗场（Thunderdome）”般的竞争机制，赋予不同智能体鲜明的观点碰撞与正交的决策框架。主持人和嘉宾还打趣地回忆起过去人们单纯依靠调高模型“温度（temperature）”来引入随机性与幻觉的做法，但真正的突破必然仰赖多智能体编排架构的底层进化。 > “Part of the reason people haven't really cracked autonomous multi-agent teams is that it's actually kind of all the same agent.” —— Jordi Hays --- ### [17:03](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1023s) Anthropic Q3 财报披露与下一代前沿模型的期盼 在节目的最后，主播们同步了当天科技圈最重磅的商业动态：Anthropic 正式公布了第三季度的财务数据。尽管此前投资者日上已有部分数据流出，但官方披露的业绩依然极为惊艳——第三季度营收达到 170 亿美元，折合年化运行率（run rate）约为 680 亿至 690 亿美元，实现了 1127% 的惊人同比增幅。 然而，强劲的数字背后也透露出隐忧：其环比增速放缓至 47%，相较第二季度的狂飙突进出现了减速迹象，这引发了二级市场的部分焦虑。与财报交织出现的，则是关于代号“Fable 5.5”等下一代前沿大模型的泄露与模糊传闻，据称其已能实现直接从零生成高清视频等全新突破。整个行业再次将焦点凝聚在同一个核心议题上：下一个能撑起万亿资本期待的模型跃升与杀手级应用，究竟何时才会真正降临。 * **季度营收**：Q3 营收达 170 亿美元，折算年化运行率达 680 亿至 690 亿美元。 * **增长趋势**：同比增长达 1127%，但环比增长降至 47%，呈现出让市场警惕的减速信号。 * **下一代前沿进展**：市场流传着关于 Fable 5.5 等新架构从头生成视频能力的传闻，资本正密切观望下一个生产力拐点。 ### [18:06](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1086s) 欧洲超市冷柜遭黑客集体入侵：纯粹的破坏行为还是新型威胁？ 节目话题转向了一起离奇的网络安全事件：欧洲多地的超市冷柜与冷藏设备遭到黑客远程入侵。黑客恶意调高冷藏温度，导致大量生鲜食品变质，迫使商家销毁价值数千美元的存货。主持人提到此前曾在一处军事基地也发生过类似的冷柜入侵案例，并对这种攻击的动机产生怀疑——这究竟是一套到处乱扫漏洞的自动脚本，还是有预谋的针对性破坏？因为攻击者并没有立即跳出来勒索赎金，食物变质后也无法再挽回，让人完全摸不着头脑。 受影响的范围涵盖了瑞典、比利时、卢森堡、爱沙尼亚、爱尔兰、荷兰和芬兰等西方国家，受害者包括宜家（IKEA）等知名零售巨头。在瑞典的一家 ICA 超市，冷柜温度甚至被黑客调高到了 122 华氏度（约 50 摄氏度）。主持人打趣称，冷柜的设计理论上根本不该产出这么高的热量，难道是黑客把风扇逆转了，“老超市以后都能直接改造成桑拿房了”。在爱尔兰，共有 13 家 SuperValu 门店中招；而在比利时，一家肉铺的冷藏设备被调至 64 华氏度（约 17.8 摄氏度）。比利时网络安全中心发言人 Katrien Eggers 表示，目前无人知晓幕后黑手是谁，鉴于对方从未提出过任何赎金要求，真相可能永远无法查明。 - **纯粹破坏的恶意**：外界越来越怀疑这次攻击的目的是搞破坏而非谋利，正如现场引述的观点：“There's growing suspicion that the goal is simply to cause damage rather than make money.” - **不联网的家电优势**：Melissa Chen 指出，这类攻击事件的增加正是保留非联网老式家电设备的好理由。 ### [20:16](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1216s) 凯迪拉克 F1 估值传闻与 Northwood Space 狂揽 3.75 亿美元 在谈及苹果（Apple）和 Palantir 即将发布财报的市场预期之余，主持人梳理了两条重磅商业资讯。首先是 Sportico 报道称，Mark Walter 正在探索以 30 亿美元的估值出售凯迪拉克 F1 车队（Cadillac F1）的相关权益。尽管就在不到两个月前，TWG 集团还表态无意退股。主持人对比了梅赛德斯 F1 车队领队 Toto Wolff 此前在播客中的谈话，探讨了赛车生意的本质：与 NBA、NFL 等具有极强本地归属感和虚荣溢价的“奖杯资产（trophy assets）”不同，F1 是一项全球巡回的商业游戏，其估值往往更看重现金流而非单纯的炒作。 随后，主持人高度赞扬了 Northwood Space 的爆发式增长。创始人 Bridgit Mendler 此前多次做客节目，如今公司在 B 轮融资不到 9 个月后，再次完成由 Andreessen Horowitz（a16z）与 DFJ Growth 联合领投的 3.75 亿美元融资，估值飙升至 20 亿美元。这笔资金将用于加速现有合同交付、招募顶级人才，以及为商业客户、美国政府与盟国航天任务搭建全球地面网络。主持人评价这并非单纯做空马斯克，而是一记极具远见的“反共识操作（contrarian move）”，精准捕捉到了 SpaceX 之外庞大的地面基站通信需求。 ```mermaid flowchart TD accTitle: Northwood Space 融资与业务扩展路径 accDescr: 展示 Northwood Space 最新一轮 3.75 亿美元融资的资金用途及业务面向。 subgraph Capital ["融资信息 (估值 20 亿美元)"] Lead["a16z & DFJ Growth 领投"] Fund["3.75 亿美元资金到账"] Lead --> Fund end subgraph Expansion ["资金核心用途"] Deliver["加速现有合同交付"] Hiring["关键技术人才引进"] GroundNet["搭建全球地面站网络"] Fund --> Deliver Fund --> Hiring Fund --> GroundNet end subgraph Clients ["服务面向领域"] Commercial["商业航天任务"] USGov["美国政府专属通道"] Allied["盟国空间任务"] GroundNet --> Commercial GroundNet --> USGov GroundNet --> Allied end ``` - **悼念创业者**：节目沉痛提及德克萨斯州发生的一起坠机事故，电商营销平台 retention.com 的创始人、44 岁的 Adam Robinson 及其家人在空难中不幸遇难，令奥斯汀科技圈陷入悲痛。 ### [24:06](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1446s) Andrej Karpathy 搓出 HackerTube：自定义 AI 软件时代的来临 为了转换情绪，主持人聊到了著名 AI 研究员 Andrej Karpathy 的最新黑客作品——“HackerTube”。Karpathy 利用 Claude 构建了一个完全贴合个人意向的定制版 YouTube 前端应用。整个工具的运行逻辑是：抓取 Hacker News 上的评论与提交内容，通过 Claude 进行智能筛选过滤，从而完全按照“真正程序员的品味”来推荐视频。 主持人感叹如今个人软件定制的门槛已经被彻底打破。Karpathy 透露的提示词其实极其简单，只有短短两句话，核心要求十分鲜明：按我的算法来、拒绝任何 AI 生成的伪音配音、拒绝创作者本人不露声出镜讲解的视频，并且严格剔除纯 AI 生成的营销脚本。大家还半开玩笑地提议，过滤算法最好加上“频道成立必须超过 4 年”这一规则，这样可以一举干掉批量生产的水军号。主持人对这款工具赞不绝口，唯一的小遗憾是自己很少在桌面端刷视频，隔空喊话 Karpathy 赶快上架一个原生 iPhone App。 - **极简提示词开发**：针对这套“氛围编程（vibecoded）”系统，主持人感叹道：“It was like two sentences. It was like, here's the idea, make it.” ### [25:59](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1559s) SemiAnalysis 推出 GDPx 基准测试与圣诞树趣味争论 分析机构 SemiAnalysis 正式推出了名为 GDPx 的全新前沿 AI 基准测试，该测试重点评估 AI Agent 在最高价值行业领域中解决实际问题的能力。SemiAnalysis 创始人 Dylan Patel 甚至为此公开向 Google 致歉，承认 Google 在 Gemini 模型的 Carbon 版本中展现了惊人实力。Patel 评价道：“He says they cooked with carbon. It's spikier in terms of domains and capabilities, but it's really good in a few key areas.” 与脱离现实的纯数学推理评估不同，GDPx 侧重于补丁修复、排查业务系统 Bug 等能够直接带来商业投资回报率（ROI）的具体工作。目前即使是最顶尖的前沿模型，在 GDPx 上的得分也仅有 12%，各大顶尖实验室正处于激烈的爬坡竞逐阶段。节目的最后，两位主持人由科技记者 Natasha Mascarenhas 提出的一个生活趣题引发了欢快的拌嘴：11 月 1 日立起圣诞树究竟早不早？两人的共识是只要提前安排好备用树的轮换订阅，无论多早摆放都不算过分。 ### [27:55](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=1675s) 节日圣诞树摆放的“换算率”趣谈与节目收尾 在节目接近尾声时，两位主持人展开了一段轻松幽默的节庆闲聊，讨论起家里到底需要放多少棵圣诞树。他们打趣地提出一种策略：可以在家里先摆放较小的树，然后随着节日临近“逐渐换成越来越大的树”，因为每棵真树实际上只需要维持一个月左右，完全可以通过轮换的方式一直过渡到最终的主圣诞树登场。紧接着，两人讨论起一栋房子里圣诞树的“标配数量”到底是多少。其中一位主持人认为基准线是“每个房间一棵”，甚至连洗手间都不能放过——洗手间里完全可以放一棵小型的圣诞树，毕竟洗手间在定义上也算作一个独立的房间。 随后他们算了一笔夸张的账：如果是一套五室五卫（5 bed, 5 bath）的住宅，按照每个房间一棵的标准，实际上就需要整整十棵圣诞树。不过他们也笑称，装饰品之间其实存在一种奇妙的“折算汇率”——比如在屋里摆上一个雪人玩偶，就能顶替不少装饰配额。最后，主持人们向听众告别，呼吁大家在 Apple Podcasts 上留下五星好评，并订阅官方网站 tbpn.com 的日常通讯（newsletter），提醒观众明天太平洋时间上午 11 点继续收看直播。 - 主持人调侃道：“It is technically a room. A bathroom. I think you got to put a small tree in there.”（洗手间在技术上也算一个房间。我觉得你必须在里面放一棵小圣诞树。） - 针对大户型的装饰算式：一套典型的五卧五卫住宅，按每室一树的标准需要配备 10 棵圣诞树，但大型雪人饰品可以适度折抵部分装饰需求。 - 节目在结尾呼吁听友前往 Apple Podcasts 留下五星评价并订阅 tbpn.com 简报，直播时间为太平洋时间每个工作日上午 11 点。 ## Notable quotes > [15:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=900s) Part of the reason people haven't really cracked autonomous multi-agent teams is that it's actually kind of all the same agent. — Jordi Hays > [03:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=180s) Markets are really good conventional wisdom distillers and that's actually very valuable. — Joe Weisenthal > [10:00](https://www.youtube.com/watch?v=Xo4bnJzMblU&t=600s) If you're walking around New York City, you could never be convinced that there is a consumer slowdown. — Joe Weisenthal ## Mentioned John Coogan, Jordi Hays, Joe Weisenthal, Greg Abbott, Abdul El-Sayed, Torsten Sløk, Todd Gillespie, Alex Dimakis, Katrien Eggers, Melissa Chen, Mark Walter, Toto Wolff, Bridgit Mendler, Adam Robinson, Andrej Karpathy, Dylan Patel, Natasha Mascarenhas, TBPN, 彭博社 (Bloomberg), Odd Lots, 纽约证券交易所 (NYSE), OpenAI, Anthropic, 微软 (Microsoft), Palantir, 英伟达 (Nvidia), 阿波罗全球管理 (Apollo Global Management), 高盛 (Goldman Sachs), Meta, 宜家 (IKEA), SuperValu, ICA, Cadillac F1, TWG 集团, Northwood Space, SpaceX, Andreessen Horowitz (a16z), DFJ Growth, retention.com, HackerTube, Hacker News, Claude, SemiAnalysis, GDPx, Google, Gemini Carbon, Apple, Fable 5.5 ## 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 Live From the NYSE: Joe Weisenthal on the Midterms + Prediction Markets 07:07 Who Actually Wins the AI Trade? + The Economic Growth Debate 12:07 Fortune 500’s AI ROI Problem + Why AI Agents Need a Breakthrough 17:03 Anthropic’s $17B Quarter + The Next Frontier Model Leap 18:01 Hackers Hijack Supermarket Freezers Across Europe 20:16 Cadillac F1’s $3B Sale Talks + Northwood Space’s $375M Raise 24:06 Andrej Karpathy Builds HackerTube + The Custom AI Software Era 25:59 SemiAnalysis Launches GDPx + Google’s Gemini Carbon Breakthrough [Source](https://www.youtube.com/watch?v=Xo4bnJzMblU)

[15:00] Jordi Hays: Part of the reason people haven't really cracked autonomous multi-agent teams is that it's actually kind of all the same agent. [3:00] Joe Weisenthal: Markets are really good conventional wisdom distillers and that's actually very valuable. [10:00] Joe Weisenthal: If you're walking around New York City, you could never be convinced that there is a consumer slowdown.

[3:00] Joe Weisenthal: Markets are really good conventional wisdom distillers and that's actually very valuable. [8:46] If you’re walking around New York City, you could never be convinced that there is a consumer slowdown.

[0:02] We are live from the New York Stock [0:03] Exchange here in New York. We're in New [0:06] York Mag. We were featured. Not really [0:08] featured. [0:09] >> One sentence, [0:09] >> more of a footnote, but we're [0:10] technically in the New York Magazine. [0:12] We're here with Joe Eisenthal. [0:14] >> Uh, thank you so much for joining us. [0:15] How you doing? [0:16] >> Love joining the third brother. [0:18] >> Thank you. Thank you for coming to town. [0:19] >> Yes. Always, [0:21] >> Joe. [0:21] >> Yeah. [0:22] >> Explain the midterms to me. What What [0:24] are the midterms? [0:26] Well, you see, in the US Constitution, [0:28] you know, they elect the president every [0:30] four years. No, here's the thing. You [0:32] guys are probably like me in the sense [0:34] where it's like I don't know. I feel [0:35] like [0:36] >> very lucky to have a job where I don't [0:37] have to talk about politics very much. [0:39] It's like, oh my god, thank God. Like, I [0:41] get crazy because you you end up talking [0:43] about politics maybe like 10 or 100 [0:45] times more than I do. [0:46] >> I know. Which is still not very. But [0:48] unfortunately, like sometimes it matters [0:50] and sometimes you can't talk about [0:52] markets, you can't talk about tech. [0:53] >> Sure. Yeah. Yeah, there's a lot [0:55] eventually sort of like, all right, [0:56] probably should learn a little bit about [0:57] this midterm thing like you guys. [1:00] >> This is my takeaway, you know, is that [1:03] seems the expectation from a lot of [1:05] people is the Democrats are going to [1:06] have a huge night. [1:07] >> Yeah, that's [1:08] >> the AI world. [1:10] >> You guys should just be paying attention [1:11] to uh the Texas governors, right? [1:12] >> Explain what's going on. [1:14] >> There's basically nobody who is data [1:16] centerp anymore, right? But Abbott is [1:19] the closest thing to the last man [1:20] standing. [1:21] >> Yeah, basically. And even he even he has [1:24] he flinching a little bit. [1:25] >> He's been flinching for months now. So [1:27] it's like these like weird like [1:28] moratoria that are like we should do a [1:30] little more environment. [1:31] >> Yeah. You can't it wouldn't make sense [1:32] for him to go into the election like [1:34] super protest. Like I was even I was [1:37] talking with a a city council candidate [1:40] um who got endorsed by this like crazy [1:42] group in my town, right? [1:44] >> And we're like that's a little [1:45] suspicious. Like why' they do that? He's [1:47] like, "Ah, like we're just, you know, [1:48] like, you know, I'm not committing to [1:50] anything one way or another." Like you [1:52] kind of have to like, [1:52] >> so they play both sides a little bit. [1:54] >> So no one is going into the election [1:56] saying I'm into data, but some people [1:58] might actually believe in things like [1:59] progress and development and technology [2:01] and stuff like that. [2:03] >> You know, obviously so much of the tech [2:04] world, so much of the manufacturing [2:06] world, etc. sort of sees Texas as the [2:09] future. Probably deep down Abbott does [2:11] himself, but he's been playing it down. [2:13] But, you know, up until a few weeks ago, [2:14] the assumption was that it was a lock. [2:16] Texas maybe, you know, the Senate race [2:19] has been um competitive for a while now. [2:22] The Democrat, James Terico, heavily [2:23] favored, but no, but the but the [2:26] Republican Senate um nominee, most [2:29] people perceive him to be like a deeply [2:31] flawed candidate, like probably the [2:33] least competitive Republican they could [2:35] have picked. Abbott is Abbott, though. [2:37] He's just like a regular Texas [2:39] Republican for a long time. [2:40] >> But it's going to like now everyone [2:42] thinks it's going to be close. There was [2:43] actually a poll last night um that came [2:46] out that had hinosa his competitor up [2:48] one. He's down to about 60% in the [2:50] betting market. So that rounds to 50% in [2:53] my book. [2:54] >> Yeah. Yeah. [2:55] I thought they all the in Brazil it was [2:58] 80% and then it traded. [3:00] >> What's going on in the prediction [3:01] markets? [3:02] >> Here's my take. [3:03] >> Yeah. [3:05] >> I still basically trust markets. Like [3:07] I'm at cor. [3:08] >> Yeah. Including prediction markets. Like [3:10] I doubt any of us know more than what [3:12] people bet like the collective wisdom [3:14] that whole thing like the collective [3:16] wisdom of the market. [3:18] >> I basically believe in that. [3:19] >> Okay. So, so when you see Wait, when you [3:21] see a 20% to 80% upset, you think, man, [3:25] this should happen one every five times. [3:27] >> Yeah. I mean, I was pretty surprised [3:29] like this is what I think. This is what [3:32] I think. [3:32] >> Yeah. [3:33] >> Markets are really good conventional [3:35] wisdom distillers and that's actually [3:37] very valuable. In other words, people [3:39] will say that as a sneer, they're like, [3:40] "Oh, this is just [3:41] >> this is just conventional wisdom." And I [3:43] like, "No, that's actually really good [3:44] because if we say like [3:46] >> Yeah. Because this is a useful thing to [3:48] have. There's like a meaningful [3:49] difference between like 55. So we say, [3:51] oh, this candidate A is favored over [3:53] candidate B, right? [3:55] >> What is it 55% or is it 80%." [3:57] >> Yeah. Yeah. [3:58] >> Now we actually have a number that we [3:59] could put on it. And so it's like, oh, [4:01] what is that conventional wisdom? So I [4:02] think it's pretty useful. I by and large [4:04] think it's like hard for most people to [4:07] confidently look at any market and say [4:09] like no this is wrong. So when a market [4:11] has a Democrat winning um at a 40% [4:14] chance of winning Texas my view is like [4:16] you got to take that seriously. [4:17] >> You got to take that seriously. [4:18] >> Like I don't know. That's all I'm [4:19] saying. That's all I think is like it's [4:21] very unlikely. [4:23] >> Yeah. [4:23] >> Yeah. A lot of people [4:24] >> What about here in New York? Is this an [4:26] important election? [4:28] >> No, I don't think so. [4:28] >> It's less important. I don't think there [4:30] were some people a couple weeks ago who [4:32] are like trying to make the the uh the [4:35] attorney general's race a thing [4:36] >> thing but no [4:37] >> I don't think I don't [4:38] >> but it's not the major swing. Okay. It [4:40] would might be interesting to see [4:42] there's there's been this talk cuz you [4:43] know it's a heavily Democratic state. It [4:45] will remain so maybe like one [4:47] possibility is that Democrats have a [4:49] huge year nationally or sorry a huge [4:52] night nationally but that like some of [4:54] the incumbent Democrats of the North do [4:56] a little bit worse than their brethren [4:58] like some sort of like anti-incumbent [5:00] bias that hurts them. I would be very [5:02] shocked if anyone's talking about New [5:03] York. [5:04] >> What's your PK? Apparently there's like [5:06] uh electrical outages in some in [5:08] Maricopa County and Arizona. I didn't [5:10] have I didn't have I didn't have a [5:12] chance to to fact my opinion, but that [5:15] that feels like a [5:17] >> and we were on the flight yesterday. [5:20] >> We spoke uh with a buddy who's in the [5:22] the legacy media world and he was coming [5:24] out for the election and he's like [5:26] expecting this to be like all week [5:28] basically like he doesn't have his his [5:29] he doesn't have his return flight yet. [5:31] >> Interesting. So if it's like all right [5:34] like there's two kinds of P chaos I [5:36] guess. One is apparently there's some [5:37] big storm coming up the other heart of [5:39] the country. Have you guys seen the [5:40] weather? Like so like maybe that'll like [5:42] in a depressed election day. There's [5:44] people people know are like looking at [5:46] the weather and it's like oh Democrats. [5:48] >> Who do you think is behind the storm [5:55] data center? [5:55] >> Which side? So what sides? [5:57] >> You guys must have some of these [6:00] ask. You guys must have had some like [6:01] weather startup guy like those cloud [6:03] guys. [6:05] We've had about a bunch. Yeah, [6:06] >> I know. It's funny because it's like no [6:08] one really controls the weather. It's [6:09] like well what about those guys you had [6:10] on your [6:12] >> that's his entire pitch. [6:13] >> You know what I think could be just like [6:15] a very crazy outcome tomorrow. So like [6:17] you know the um there's some questions [6:20] about the Senate and who's going to take [6:21] the Senate. Number of competitive races [6:23] could swing um Democratic. The one [6:26] interesting one is that in Michigan um [6:29] you know there's the far-left candidate [6:31] or the left candidate Abdul El Say. [6:33] Yeah. [6:34] >> I think it would be a very funny outcome [6:36] from a sort of like chaos [6:39] um like people are just going absolutely [6:41] insane online if he ends up being the [6:44] only competitive Democrat that wins. So [6:46] it's like if Republicans pulled out in [6:48] Kansas or Iowa and all these others but [6:51] then they lose the one that [6:54] like I'm just thinking what would get [6:55] like people like the most insane cuz you [6:57] know we live for fighting on Twitter [7:00] this is like we wake up for to see [7:01] people arguing online if that were the [7:03] final outcome like you would have people [7:05] just absolutely losing% [7:07] >> what what do you think of the 4D chess [7:09] that the tech community is secretly [7:11] behind the blue wave because they [7:13] overbuilt AI compute and they want to [7:16] create bans on compute developments so [7:19] that that reduces supply that can drive [7:21] up prices, earn more margin in the short [7:23] term. [7:24] >> I like this take. [7:25] >> Yeah, because there's a lot of supply [7:26] coming online. [7:27] >> This is the this is this is a good take [7:30] because this No, I like this because it [7:32] gets to one of the things I've been [7:34] thinking about which is like [7:35] >> okay let and because [7:36] >> this actually might be 5D [7:39] >> they might be in the fifth dimension. [7:41] People talk about the AI trade, right, [7:43] on Wall Street. And the thing that I [7:45] don't totally understand is like, well, [7:48] there's no consensus on who yet on who [7:50] actually makes money. Let's just say [7:52] it's like everything goes well. It all [7:54] goes well. It's great for the economy. I [7:56] would still say there's not obvious [7:57] consensus on like who the money makers [7:59] are in that scenario whether it's the [8:01] lab companies [8:04] right now there's even yeah diffusion [8:06] like the Microsoft plus palunteer versus [8:08] openi plus or just like or is it just [8:10] going to be like the chip makers is it [8:12] going to be everyone else the rest of [8:13] the economy doesn't commodize the top [8:16] there's ambiguity on this question [8:18] >> so like it certainly seems to me at le I [8:21] love the bottleneck they want to create [8:23] bottlenecks so they're funing data [8:25] center bands to create more bottlenecks [8:27] so they can trade them. [8:28] >> Yeah. [8:30] >> Because it's like no no like we want to [8:31] be the winners again. [8:32] >> Exactly. But but I think but I think AI [8:35] going well to some people just means [8:36] like 20% GDP growth, [8:39] >> right? Like that's actually No, but but [8:41] look, there was there was a survey at at [8:43] a company that that I won't I won't [8:45] >> was [8:46] >> I think it was it was anthropic [8:49] employees were really high, but then the [8:51] average American was really high too. [8:52] >> Yeah. Yeah. Yeah. Everyone is expecting [8:54] like [8:54] >> except economist [8:57] the economists are like oh yeah I could [8:59] really drive a huge productivity boom we [9:01] might get up to three and a half% [9:03] >> but this is but you know we joke about [9:05] it [9:06] >> but we all should have I mean anytime [9:08] this comes up what will be the [9:09] productivity effects of AI [9:12] >> it's like what were the productivity [9:13] effects of the internet they're hard to [9:14] see you don't see them in the charts [9:16] even though what were the productivity [9:17] effects of the mobile phone [9:18] >> I I thought I did just see a chart in [9:20] the financial times that showed worker [9:22] productivity in the United States was [9:24] growing and it was outpacing other [9:26] countries. Did you dig into that at all? [9:28] >> I know. I haven't [9:30] >> you could attribute it. You could [9:31] attribute it to layoffs. Right. Right. [9:33] And there Yeah. So there's like a Yeah. [9:35] Like that's probably what it is because [9:37] ultimately like [9:38] >> the US allowed a bunch of people to get [9:40] laid off and substituted them with [9:43] checks. So that flatters your [9:44] productivity for a while. And also the [9:48] US has so that probably started the [9:50] divergence and then the US economy has [9:52] just been much stronger than Europe for [9:54] the last several years. [9:57] >> It's hard for me to imagine it's a big [9:58] AI difference, but [10:00] >> Sure. Sure. Sure. Yeah. I mean, in [10:02] general, are you still feeling uh that [10:04] the American consumer is undefeated? It [10:07] feels like there's all this chaos in the [10:08] market. There's there's a lot of [10:10] fragility where it's like, oh, if this [10:11] doesn't go perfectly, could be bad. But [10:13] all of that is basically just froth [10:16] while I go back and forth all the time. [10:18] I'm like I'm in this bubble in New York [10:20] City. And if you walk around New York [10:21] City, you cannot fathom right now that [10:23] like there's some slowdown in there [10:25] people waiting in line down the block [10:26] for [10:28] 30 for [10:30] $4. [10:31] >> New York feels absolutely everyone's [10:33] like I can't get a dinner table. Oh, I [10:35] see we see this in LA where people [10:37] you'll have it'll be like Monday at [10:39] 10:00 a.m. and people will be lines [10:41] around the block for a record store. [10:43] >> You know, I'm like if you're walking [10:45] around New York City, you could never be [10:47] convinced that there is a consumer [10:48] slowdown. I do think that for the and [10:51] even like when you listen to corporate [10:52] executives speak, it's a little bit [10:54] ambiguous because I've been hearing well [10:56] we've noticed the consumer is starting [10:58] to get a little bit more discriminating [10:59] in their purchasing. I've been hearing [11:01] that since 2022. So it's like for every [11:03] year. So we notice that the consumer is [11:05] starting to trade down a little bit. [11:08] >> I don't know. I've seen some back and [11:10] forth that it still feels like there's a [11:12] lot of spending in this economy from [11:13] every sector. This is the most important [11:15] thing right now actually which is that [11:17] you know you can can divide the economy [11:19] into roughly three sectors. There's the [11:21] household sector, the business sector [11:23] and the government sector. [11:24] >> And where's the gambling sector? [11:26] >> The gambling [11:27] >> that's a huge part of [11:30] the business. I did I did see a a ad on [11:33] on the way here that just said predict [11:35] big [11:35] >> predict big just predict predict bro [11:40] >> I just predict what do you do [11:42] >> on black [11:43] >> you don't really see bet big [11:45] >> predict big [11:46] >> but every one of them is spending this [11:48] is households are spending they're going [11:50] they're either like [11:51] >> selling based if they're on the top part [11:53] of the K they're like selling based on [11:55] their equity accumulation if they're on [11:57] the lower side of the K they're probably [11:58] like going deeper into consumer credit [12:00] Businesses are spending either from the [12:03] AI buildout or just like spending on AI [12:06] and the government is spending on old [12:07] people. [12:08] >> Although did you see Torston Slock over [12:09] at Apollo saying what's his latest? [12:11] >> His latest is that basically uh Fortune [12:14] 500 having a little bit of indigestion [12:17] around AI spend outside of the Mag 7. [12:21] Mag 7 is seeing positive ROI. If you're [12:23] a tech company, you can immediately [12:25] deploy the AI agents, improve your [12:27] software, see an ROI. But if you're the, [12:30] you know, you're in the S&P 500, but you [12:32] make tires or you do forestry or you do [12:34] oil and gas, a lot harder to actually [12:36] improve revenue or actually replace a [12:38] job. Many of the jobs are clearly [12:40] >> my my stance on this is like for this [12:43] for the overall trade to work, [12:46] >> you have to have a massive capabilities [12:49] jump [12:49] >> like soon. Like in Q4 of last year, [12:52] everyone everyone everyone was like, [12:54] "Whoa, it doesn't seem like chat bots [12:56] are going to be like [12:57] >> 4.5 [13:00] and then you had agents." And it's like [13:02] the trade right now [13:04] >> requires that like Elon Elon saying [13:06] like, "Yeah, like you have some [13:07] deceleration, [13:09] you know, like anthropic Q3, you see a [13:11] little bit of deceleration. It's still [13:12] growing tremendously." Um and and Elon [13:15] is saying like you know BA basically all [13:18] the SpaceX projections are like uh just [13:20] focus on supply right which is like [13:22] we're going to bring on insane amount [13:24] don't worry about demand we're bringing [13:25] on insane amount of supply right um and [13:28] and that could prove to be [13:31] >> an amazing decision right [13:33] >> but we need that capabilities jump [13:36] >> like fast [13:37] >> yeah right now there's no one outside of [13:39] an engineer that can go to their boss [13:40] and say I want to spend $10,000 a month [13:42] on AI Yeah, pretty hard. So, this is in [13:45] marketing. [13:46] >> My colleague at Bloomberg today, Todd [13:48] Gillespie, has a had a good thing. [13:49] Goldman is clamping down on its [13:51] non-technical employees like vibe [13:54] coding, including its traders, which I [13:56] thought was pretty telling and and it [13:58] said it was for both um risk and [14:01] productivity. So, it was just risk, you [14:03] know, I know there's some issues with [14:04] people there's still anxiety about data [14:06] sharing and stuff like that or data [14:08] retention or whatever, but it was also [14:09] about the productivity. So it's like [14:11] maybe not every trailer needs to be vive [14:13] coding up their own dashboards [14:16] and stuff like that. [14:17] >> I thought that was a telling story that [14:19] speaks to the [14:20] >> tour. Yeah. There's something about an [14:22] engineer who's actually seen what a [14:24] month of engineering work and how hard [14:27] that is and they're maybe a little bit [14:28] more judicious with the [14:29] >> children. The thing is like as we know [14:31] like I'm a very big believer in what [14:33] some economists and I wish I could cite [14:35] the person who like first either wrote [14:37] the name but like they call like the [14:39] messy jobs thesis. We all have like, you [14:41] know, within a job we have like 30 [14:43] tasks, right? And so forth. And so you [14:47] could say, well, like AI is really good [14:48] at this task, but it's not really a [14:50] replacement for someone's job. [14:52] >> Yeah. Totally. [14:52] >> Except maybe if you're just producing [14:55] lots of code, then maybe your job isn't [14:57] so messy. [14:57] >> There are some jobs that look a lot like [14:59] repeat tasks. [15:00] >> Yeah. And so I think I think what we [15:02] need to see is like real agency. And I [15:05] don't mean like a I don't mean agents of [15:06] like you're telling an agent to go and [15:08] do something. you need like you know [15:10] right now we have the equivalent of like [15:12] every company has access to like a [15:14] hundred PhDs and they just like sit in [15:16] the room and they don't even talk or [15:17] they don't even look around right [15:19] >> and so the ne I do think the next leg up [15:21] can come from like hey I'm setting a [15:23] budget cap [15:25] >> now use that most effectively to improve [15:27] my business right you know I've played [15:28] around [15:29] >> and that's effectively what you do on [15:30] meta which is like you're like here's [15:31] how much I'm willing to spend get me as [15:33] many customers as you can right [15:35] >> um and I do feel like what what's [15:38] missing is still true agency, right? And [15:41] so agents are like almost like um [15:44] >> No, that's true. [15:45] >> The agents that we have today are like [15:47] really all like sub agents in some ways, [15:49] right? That's right. They're only [15:50] they'll only do things that they're told [15:51] to do. And so [15:53] >> well, they're also they're also clones [15:54] of each other. Like this is like another [15:56] issue which is like you think like oh if [15:58] you gathered 100 PhDs in a room and told [16:00] them to go [16:02] >> accumulate customer do whatever it might [16:04] actually work. But if it was a hundred [16:06] of the exact same PhD who had all done [16:09] the exact Alex Emos who's over at [16:10] Google. [16:12] >> Yeah. He's a cool guy. He's talked about [16:13] this like part of the reason this hasn't [16:16] really like people haven't really [16:17] cracked like sort of like autonomous [16:19] multi- aent teams is that it's actually [16:21] kind of all the same agent. In many [16:24] cases you sort of need to get a [16:25] Thunderdome like battle make them battle [16:27] out with like get them like genuinely [16:30] different points of view. And I don't [16:31] know if anyone cracked the right [16:33] orchestration for that. I mean, people [16:34] used to just like turn up the [16:35] temperature on these. Do you remember [16:36] this whole thing? But like [16:38] >> I forgot about temperature. [16:40] >> Some of these things [16:42] going back and forth for just getting [16:44] like weirdness up and get it to let's [16:47] get some more hallucinations. Yeah. I [16:48] mean, people were saying like [16:49] hallucinations are uh sometimes like a [16:51] feature more than a bug. Like not you [16:53] don't always want to hammer it out [16:54] because you want some creativity. Uh the [16:56] other big the other big news I there's a [16:58] few big news stories coming across the [17:01] timeline today. Anthropics Q3 numbers [17:04] are out officially. These have been [17:07] rumored or leaked from the investor day. [17:10] It didn't seem like a huge surprise, but [17:12] uh Q3 revenue came in at $17 billion. I [17:15] think that's a 69 billion68 billion run [17:20] rate. Uh but based on that quarter, it's [17:23] 1127% [17:25] year-over-year growth, 47% [17:28] quarter-on-quarter growth. Uh that's a [17:30] deceleration from Q2 which has people a [17:34] little nervous but overall uh some this [17:38] is there's already a balancing act going [17:40] on with some uh some like leaked rumored [17:44] vague posting around Fable 5.5 [17:46] generating videos from scratch and doing [17:48] a bunch of cool stuff. So uh again the [17:50] big question is yeah what is the next [17:52] leg up? What is the the next application [17:55] of the models? uh if they get better, if [17:57] the straight line on the log graph [17:59] continues, uh what is the unlock? What [18:02] is the next thing that that everyone uh [18:04] can't get enough of? Um the other news, [18:06] fridges are getting hacked. Hackers are [18:08] remotely remotely taking control of [18:11] supermarket refrigerators and freezers [18:12] across Europe, turning up their [18:14] temperatures and forcing stores to throw [18:15] away thousands of dollars worth of food. [18:17] We talked about this in the context of [18:20] uh of a military base that saw this hack [18:23] happen with the fridges. And I wonder [18:25] how deliberate this is. Is this just a a [18:28] script that goes out and tries to just [18:31] wreak havoc? Is it is there any [18:33] targeting to it? It says, [18:34] >> "Yeah, it doesn't feel like it feels [18:35] like they just come in, ruin all the [18:37] food, and then what what is the next [18:40] ask?" Like, "Hey, [18:41] >> we'll stop ruining your food if you pay [18:42] a bounty, [18:43] >> but it feels like heard about those." [18:45] >> At that point, all the food is ruined, [18:47] so you could just try to [18:49] >> do it again. I don't know. Yeah, it is [18:51] odd. And the countries, it doesn't tie [18:53] together. together. I mean, it's just [18:54] western countries right now. Sweden, [18:55] Belgium, Luxembourg, Estonia, Ireland, [18:58] Netherlands, Finland, IKEA is affected, [19:01] uh, a bunch of other stores. And so, um, [19:04] one, uh, ICA supermarket in Sweden [19:08] reported that its freezers reached 122 [19:11] degrees Fahrenheit. [19:12] >> That's that's really really hot. How do [19:14] they go? How do they go that way? [19:16] >> If they're designing a freezer, it [19:17] should shouldn't be able to pump up the [19:19] heat that much. But I guess if you [19:20] reverse it or something, reverse the [19:22] fan. [19:23] >> Dude, people are going to realize that [19:25] old supermarkets can be turned into [19:28] saunas. [19:29] >> Okay. [19:30] >> And start the conversion process [19:32] >> potentially. Potentially. I think it's [19:34] time to get an ice box. Get the get the [19:35] ice ice man coming in delivering with [19:37] the big tongs. Time for that. In [19:39] Ireland, 13 superv value stores were [19:42] affected. In Belgium, a butcher reported [19:44] that hackers raised his refrigerators to [19:46] 64 degrees. Again, very odd that uh the [19:49] the the goal seems to be just to cause [19:52] damage. Uh Ketrian Edgars, spokesperson [19:55] for Belgium's Center for Cyber Security, [19:57] says no one knows who is behind these [19:58] attacks and that we may never know since [20:01] no one has demanded a ransom. There's [20:03] growing suspicion that the goal is [20:04] simply to cause damage rather than make [20:07] money. Strategy risks. Melissa Chen [20:09] thinks the rise of these kinds of [20:10] attacks is good reason to hold on to [20:13] older appliances that aren't connected [20:14] to the internet. Interesting. [20:16] >> In other news, Apple and Palunteer [20:20] reporting earnings today. [20:22] >> Uh [20:24] we really want Apple to uh [20:26] >> we really want Apple to crush otherwise [20:29] I think it's we're going to be seeing a [20:31] lot of red out here after that. Um yeah, [20:34] and Palunteer will be an interesting [20:36] signal as well, just from a diffusion [20:38] standpoint. [20:39] >> Yeah. [20:39] >> Um more news. Uh Mark Walter is [20:42] exploring selling Cadillac F1 at a $3 [20:46] billion valuation. [20:47] >> Interesting. [20:47] >> Um less than two months ago, TWWG said [20:50] it was not considering exiting its stake [20:52] in a team. This is from Sportico. [20:55] >> Um [20:56] >> I wonder how uh I wonder how the [20:58] business is doing. We heard Toto Wolf on [21:00] Cheeky Pint talking to John Collison [21:02] about uh the economics of the Mercedes [21:05] F1. [21:05] >> Yeah, it seems like that he just has [21:06] like an amazing business. Yeah. [21:08] >> But he was like very realistic about [21:09] like he he was basically saying like I'm [21:11] being valued on my cash flows more than [21:14] hype. [21:15] >> Yes. It's not a trophy asset like an NBA [21:17] team, like an NFL team, like some of the [21:19] other teams that trade where they're not [21:21] really generating a lot of cash flow. [21:23] Yeah, it's interesting because with a [21:24] with a team there's so much [21:26] >> there's so much demand from somebody [21:28] who's like I live in LA and I want to [21:30] own the Lakers, right? Or like I live in [21:33] New York and I want to own XYZ team. [21:35] With F1, it's like your team is just [21:37] going all over the world all the time. [21:39] And so you really have to one want to go [21:42] to a lot of races [21:43] >> or have business all over the world [21:45] potentially and be more uh more global. [21:49] Uh yeah, it almost feels like uh if you [21:52] want to go really big in DAFF one, like [21:56] >> you know, partnering with Las Vegas in [21:58] some way or Miami and and and if you're [22:00] a Miami uh you know, business person, [22:02] you you would just want to make sure [22:04] that you're you want to buy the heat and [22:05] then also participate when [22:08] >> do what George Kurts did, which is, you [22:11] know, become become a big part of uh AMG [22:15] and uh and then, you know, deliver the [22:18] world's youngest this world champion. [22:20] >> Yes. [22:20] >> Um, incredible [22:21] >> yesterday. Really, really amazing. [22:23] >> So good. [22:24] >> And early too or still early in the [22:26] season, so great to see. [22:28] >> Um, [22:29] >> Northwood. Northwood is on an absolute [22:31] tear. Northwood Space. We've had Bridget [22:33] on the show multiple times at this [22:35] point. Uh, she's back raising $375 [22:38] million at a $2 billion valuation, [22:40] co-led by a A16Z and DFJ growth just [22:44] less less than nine months after series [22:46] B. Northwood says this round of funding [22:47] will be used for accelerating delivery [22:49] on existing contracts, talent [22:51] acquisitions, and the buildout of [22:52] Northwood's ground global ground network [22:54] for commercial, US government, and [22:55] allied space missions. And we've seen uh [22:59] just yeah, just incredible uh it was a [23:04] very contrarian move. It's not exactly [23:05] like a bet against Elon, but it's it's [23:07] it's standing it's expecting that the [23:11] other players will be taken seriously [23:13] that there will be uh you know a way to [23:15] either work with SpaceX ultimately or [23:18] work with all the other partners who [23:20] want to have a separate backbone. So, uh [23:22] it was just like the right idea at the [23:24] right time, great execution and uh yeah, [23:27] what what an absolute run. Um what else [23:30] is in the top? [23:32] >> Uh really sad news. Um, the Texas family [23:35] that was on the uh plane crash yesterday [23:39] uh was a 44 year old tech entrepreneur [23:41] named Adam Robinson who had a company uh [23:44] called retention.com. I never knew Adam, [23:46] but um I know that a lot of people in [23:49] ecom uh were were customers and um yeah, [23:53] just a really tragic event all around. [23:56] Um [23:58] and uh a rough year for the Austin tech [24:02] community [24:03] >> for sure. Yeah. There was another plan. [24:04] >> Prayers with with their family. [24:06] >> Yes. Yes. Is a pallet cleanser. I mean [24:09] let's uh go over to Andre Carpathy [24:11] because he hacked together hacker tube [24:14] which is he built his own YouTube. You [24:16] saw this with a a content [24:17] recommendation. So he has he has all the [24:19] different comments and uh hackernew [24:23] submissions. He filters those through [24:25] Claude and then creates a new YouTube [24:28] fully vibecoded uh just clone of [24:31] YouTube. Uh so he can uh just watch [24:34] exactly the videos that he wants [24:36] recommended by real programmers. uh sort [24:39] of an interesting moment of um of you [24:42] know these like in instantiating like [24:44] custom software these like uh it's like [24:46] a lot [24:46] >> and I like I like how people were like [24:48] what what's your prompt and it was just [24:49] like this most simple you shared and it [24:52] was like it was like two sentences [24:53] >> it was like here's the idea make it yeah [24:57] I want I want I want this I basically [24:59] want my YouTube my exact YouTube [25:01] algorithm no AI voice [25:04] >> I don't want any I don't want any videos [25:06] where the creator [25:08] >> isn't talking through it themselves. [25:09] That's a good start. Yeah. And then I [25:11] think secondary is like detecting like [25:13] pure AI scripts. [25:14] >> You can also you can easily go through [25:16] the channel and go backwards to see when [25:18] the channel was created and that helps [25:20] to know is it some new [25:22] >> channels older than than four years old [25:25] >> basically. Um and this is interesting. [25:29] The the trick is I don't watch a lot of [25:31] YouTube on desktop. I need an iPhone [25:33] app, Andre. So let's get it in the app [25:35] store. Uh, I don't watch YouTube on TV [25:38] >> really. [25:39] >> Yeah, usually background audio on my [25:42] phone for most of these things. Like [25:44] anything here I would either be uh [25:45] watching on my phone, although of course [25:47] you could do it on the in Safari, but uh [25:49] I like I like an app. I like an iPhone [25:51] app. Um anyway, uh the other the other [25:54] big news, we're going to talk to the [25:55] semi- analysis folks um uh tomorrow or [25:58] the next day, but uh they have a big [26:00] benchmark out GDPX. [26:02] uh they're getting in the benchmark game [26:04] and um really really interesting to see [26:06] Dylan Patel issuing an official apology [26:10] to Google. He says they cooked with [26:12] carbon. It's spikier in terms of domains [26:14] and capabilities, but it's really good [26:16] in a few key areas. And so, uh GDPX is a [26:19] benchmark for measuring agentic [26:21] capabilities across the most valuable [26:22] domains. And they highlight the fact [26:24] that the best model scores just 12%. [26:27] It's neck andneck between a lot of the [26:29] frontier labs and uh of course takes [26:32] into account the you know semi analysis [26:35] a lot about ROI and the actual value of [26:38] these tasks. So it'll be very [26:40] interesting to see the labs hill climb [26:42] all of this and maybe this is the answer [26:44] to your question about the next leg up. [26:46] Um because these are very practical [26:49] things like so many so many jobs so much [26:52] work doesn't look like a you know like a [26:55] mathy eval on the day-to-day but a lot [26:58] of these GDP X benchmarks are specific [27:02] tasks around uh things that businesses [27:05] would like to do more of like go and [27:07] patch bugs go and uh you know fix things [27:10] in the in the in the actual business. So [27:13] congrats to the semi analysis team on [27:15] the launch. Uh, Natasha Moscaranus asked [27:17] an important question. We got to get to [27:19] it. [27:20] >> Jordy, did you see this? She was a big [27:22] fan of our She was a big fan of our uh, [27:26] Elf on the Shelf interview. But Natasha [27:28] asks on November 1st, 2026, "Is it too [27:32] early to put my Christmas tree up?" What [27:34] do you think, Jordy? [27:35] >> Absolutely not. [27:36] >> Absolutely not. In fact, [27:38] as long as you have a backup tree. [27:40] >> Back a backup tree. [27:41] >> As long as you basically slate a [27:43] delivery of a second tree. Sure. [27:46] >> Probably like December 1st. Yeah. [27:47] >> Well, you're a fan of the Christmas tree [27:49] subscription that starts in uh late July [27:52] and then you're getting a monthly tree [27:54] leading up and they just get [27:55] progressively bigger and bigger in your [27:56] house. So, you can only you you really [27:58] only need the tree to last a month. Swap [28:00] it out till the final tree arrives. [28:02] >> What's the reg [28:05] >> in in one house? [28:06] >> Yeah. [28:06] >> One per room would be the baseline. But [28:09] then also you can say outside. I'm [28:10] talking about like bathrooms. She wants [28:13] >> a smaller tree in there. [28:15] >> It is technically a room. A bathroom. I [28:16] think you got to put a small tree in [28:17] there. [28:18] >> Yeah. Yeah. So, if somebody's saying [28:19] like, [28:20] >> you know, five bed, five bath, it's [28:22] really like 10 trees. [28:23] >> 10 trees. But, I mean, what there is a [28:25] little bit of like an exchange rate [28:26] because one snowman goes pretty far. [28:28] >> Leave us five stars on Apple Podcast. [28:30] Sign up for our newsletter, tbp.com, and [28:33] we will see you tomorrow at 11:00 a.m. [28:35] Pacific. Bye.