Blue Origin Raises $10B as Apple Scraps Low-Cost Vision Pro Screen

Blue Origin融资100亿美元，苹果放弃低成本Vision Pro屏幕 Jeff Bezos旗下的太空探索公司以1300亿美元估值斩获首笔外部投资，而苹果因需求疲软及制造良率过低放弃了廉价版Vision Pro显示屏。 主持人深入剖析了多项重大科技与商业动态：首先是Blue Origin在其首轮外部融资中以1300亿美元估值募集了100亿美元；接着探讨了苹果因供应链良率和低迷的市场需求而取消更廉价Vision Pro显示屏计划；随后分析了Getty Images与Shutterstock在英国监管机构干预后宣告破裂的37亿美元合并案。此外，本期节目还涵盖了废弃IBM园区引发网络热潮的城市探险（Urbex）趋势、OpenAI推出GPT Live语音模型，以及GPT-5.6模型即将向公众发布等内容。 Jeff Bezos 创立的蓝色起源（Blue Origin）正以 1300 亿美元估值寻求 100 亿美元的首轮外部融资，其中 Bezos 个人出资 20 亿美元，投资机构 Coatue 出资 40 亿美元。 (0:03) 因面临高昂制造成本及大众消费需求疲软，苹果据报已取消平价版 Vision Pro 显示屏开发计划，并正逐步终止与三星的显示屏合作。 (0:43) 因英国竞争与市场管理局（CMA）要求剥离 Shutterstock 的编辑业务，Getty Images 正式终止了双方价值 37 亿美元的合并交易，而生成式 AI 的崛起已导致两家公司市值双双暴跌至约 3 亿美元。 (0:56) OpenAI 宣布 GPT-5.6 Soul、Terra 和 Luna 将于本周四正式向公众发布，测试者反映该系列模型在交互体验和计算机操控能力上均展现出巨大飞跃。 (23:24)

TBPN explores major shifts across deep tech, consumer hardware, antitrust regulation, and artificial intelligence, highlighted by Blue Origin's $10 billion outside capital raise and OpenAI's expanding model portfolio. Blue Origin's Milestone Outside Funding Round (0:03) Jeff Bezos's Blue Origin is securing $10 billion in its first external funding round across its 25-year history, achieving an estimated $130 billion valuation. The transaction includes a personal $2 billion commitment from Bezos and a $4 billion allocation from investment firm Coatue. While critics note that Blue Origin's valuation comes against an estimated $5 billion burn rate this year and roughly $27 billion burned to date, commentators argue the company is being priced on its rare technological status as the second entity capable of orbiting and landing a reusable rocket, outpacing Chinese rivals and standing alongside public space peers like Rocket Lab and AST SpaceMobile. Hardware Retreats and Mixed-Reality Reality Checks (0:43) In consumer hardware, Apple has reportedly scrapped plans for a lower-cost Vision Pro display and is ending its development partnership with Samsung Display by September. Apple originally subsidized early-stage, single-digit yield displays from Samsung to gain an edge in visual fidelity, but high unit costs, device weight, and limited developer momentum have dampened consumer adoption. Commentators observe that consumer preferences continue to favor interactive mobile devices over isolated VR experiences, nudging Apple's long-term R&D toward everyday smart glasses. Regulatory Intervention and Legacy Media Disruption (0:56) The planned $3.7 billion merger between stock media firms Getty Images and Shutterstock collapsed after the UK's Competition and Markets Authority required Shutterstock to divest its editorial division. Getty's board rejected the condition and abandoned the transaction, despite earlier clearance from the US Department of Justice. The hosts criticize the regulatory block, pointing out that both firms have seen their market caps tumble to around $330 million to $340 million under existential competitive pressure from generative AI image tools. The decision was compared to previous interventions, such as the UK blocking Meta's acquisition of Giphy, while platforms like Canva face criticism for failing to label AI-generated imagery within stock libraries. OpenAI Expands Real-Time and Flagship Models (19:57) OpenAI introduced GPT Live, a full-duplex voice model capable of concurrent listening, speaking, and background reasoning, shifting real-time interaction toward multi-threaded conversations. Simultaneously, OpenAI scheduled the public rollout of GPT-5.6 Soul, Terra, and Luna following preview testing. Early testers and commentators note distinct behavioral personalities across new frontier architectures, driven by specialized distillation pipelines and proprietary data rather than uniform web scraping, with Soul noted for responsive handling and advanced computer-use capabilities. Across tech sectors, massive capital continues to flow into capital-intensive frontiers like orbital spaceflight and advanced AI architectures, while consumer hardware and legacy media platforms grapple with shifting user habits and regulatory headwinds.

本期节目梳理了科技前沿领域的资本分化与路线迭代：蓝色起源与OpenAI等头部力量在航天制造和全双工AI模型上持续获得巨额注资并取得技术突破；与此同时，以苹果头显为代表的高端硬件和以Getty-Shutterstock为代表的传统版权图库，则在消费需求不足、生成式AI冲击与反垄断监管的多重压力下陷入停滞或估值崩塌。 蓝色起源百亿融资与商业航天估值 (0:03) Jeff Bezos创立25年的蓝色起源（Blue Origin）正在推进其首轮外部资本融资。根据《纽约时报》Andrew Ross Sorkin的报道，该轮融资规模达100亿美元，投后估值约1300亿美元。其中Bezos个人注资20亿美元，投资机构Coatue认购40亿美元（Bezos家族办公室为Coatue旗下基金的主要LP）。针对批评者提出的高估值与高现金消耗质疑——公司历史上已累计消耗约270亿美元（年均约10亿美元），且今年预计现金消耗将达50亿美元——节目主播指出，该估值主要建立在其稀缺的工程实力上，即作为全球第二家具备轨道级可重复使用火箭发射与回收能力的企业，其技术进度仍领先于中国同业竞争对手。 在公开市场估值参照上，同赛道的Rocket Lab市值略低于500亿美元，AST SpaceMobile估值在270亿至300亿美元之间。节目回顾了早年Uber估值达到140亿至170亿美元被视作不可思议的历史，认为在当下数万亿美元巨头并存的资本市场环境下，核心硬科技资产的私有市场定价逻辑已发生显著变化。 消费级头显遇阻与苹果显示屏战略转向 (0:43) 在消费硬件领域，苹果公司据报已搁置低成本Vision Pro显示屏的研发计划，并正逐步结束与三星显示的合作项目，该合作预计将于9月正式终止。Vision Pro虽然拥有前沿的硬件配置，但在实际市场中面临定位尴尬，其高昂定价、设备自重以及第三方开发者生态匮乏制约了产品普及。 从供应链经济学来看，苹果此前为在视觉清晰度上超越Meta，选择向三星采购原本距规模量产尚有4年的显示屏幕，并承受了良品率仅有个位数的早期原型制造高成本（而成熟科技制造通常要求良品率达到90%至99.9%）。然而，这种通过巨额补贴换取顶尖显示效果的路径并未随时间推移转化为预期的消费级需求规模。主播认为，当前用户更偏好在智能手机上进行高互动的日常碎片化浏览，而非佩戴头显进行孤立观影，苹果的长期研发重心也因此转向日常智能眼镜品类。 反垄断干预与生成式AI夹击版权图库 (0:56) 图片机构Getty Images与Shutterstock之间价值37亿美元的合并交易正式流产。该交易于2025年1月公布，虽在同年4月获得美国司法部批准，但英国竞争与市场管理局（CMA）以保护英国媒体选择权及防范涨价为由要求剥离Shutterstock的编辑业务，Getty董事会拒绝该条件并于2026年7月正式终止合并协议。主播对此提出批评，指出Meta的Muse、OpenAI工具以及Nano Banana Pro等生成式AI图像模型正在对传统图库造成颠覆性冲击；由于媒体可以直接低成本生成定制图像，图库平台寻求合并以维持定价权属于合理的防御性自救。 随着并购告吹，两家公司的市值自2025年初以来大幅缩水，当前Shutterstock市值降至约3.29亿美元，Getty约为3.40亿美元。节目将此案与2022年英国CMA叫停Meta收购Giphy（而苹果iMessage动图功能实由Bing支持）以及美国叫停Meta收购VR健身创企相提并论，认为反垄断监管在应对已被AI解构的行业时存在脱节。此外，针对观众反映Canva图库目前未对AI生成图像做明确标注的情况，主播也指出了当前版权素材库规范缺位的现状。 全双工语音架构与GPT-5.6模型发布 (19:57) OpenAI近期推出了集成于ChatGPT中的新一代实时语音模型GPT Live。该模型采用全双工（full duplex）底层架构，支持系统同步进行收听与语音输出，实现了自然的即时打断与双向对话；同时，它能够在后台将复杂任务路由至专门的推理模型，而不影响前台正在进行的语音交流。主播将深度测试此类模型的用户比作“白帽黑客”，指出测试群体通过探索原生工具缺失、倒计时设置和上下文窗口维持极限，为语音AI从简单的听写交互走向多线程协作伙伴建立了实质性测试基准。 与此同时，OpenAI宣布GPT-5.6旗下的Soul、Terra和Luna版本在扩大全球预览测试后，将于本周四正式向公众发布。反馈显示，与此前主打极致推理速度的Fable模型相比，GPT-5.6 Soul展现出更具表现力且细腻的模型性格，并具备领先的计算机操作能力（computer-use）。评论认为，顶尖前沿模型之间的个性与性能差距拉大，主要源于各家团队采用的专有训练数据、更严格的知识蒸馏流程以及各异的微调偏好，而不再仅取决于通用网络抓取数据的规模。 早期算力园区废弃与社交媒体探险热潮 (16:59) 《华尔街日报》报道了位于纽约州萨默斯（Somers）、距离曼哈顿约一小时车程的前IBM园区的荒废现状。该园区曾是IBM Watson在《危险边缘》（Jeopardy!）夺冠时期的标志性研发基地，如今已全面空置。其独特的建筑结构和空荡氛围使其成为社交网络上“城市探险”（Urbex）群体的热门打卡点，探险者通过侵入未开放设施拍摄视频并在TikTok和Instagram上获取流量。 根据报道，今年4月，62岁的退休工程师Robert Carlton在园区拦截了逃跑的青少年，纽约州警方随后以非法侵入罪逮捕了7名未成年人。类似的违规入侵行为在多地呈扩散态势，新泽西州利文斯顿警方（针对停业的Livingston购物中心）与佛罗里达州杰克逊维尔海滩警方（针对关闭的Adventure Landing游乐园）均已发布公开警告，对社交平台煽动的废墟探险挑战进行执法警示。 本期节目折射出当前技术周期的深刻演变：资金与研发正加速向掌握稀缺制造能力的商业航天和具备全双工交互能力的底层AI模型聚集；与此相对，缺乏刚需支撑的高成本硬件与依赖传统版权壁垒的数字资产平台，则在生成式AI浪潮与滞后的反垄断监管下迅速承压。从昔日AI算力园区的荒废到新一代模型架构的迭代，技术更替对既有产业格局的重塑仍在持续加速。

## Background 本期内容由科技播客频道 TBPN 于 2026 年 8 月 3 日发布在 YouTube 平台。该视频属于《Diet TBPN》系列，由 John Coogan 和 Jordi Hays 共同主持，旨在用 30 分钟高密度浓缩呈现每日直播科技脱口秀的核心精华与前沿动态。 ## TL;DR 节目首先深入剖析了蓝色起源（Blue Origin）寻求 100 亿美元首轮外部融资、估值达 1300 亿美元的里程碑事件，并探讨了商业航天赛道的估值坐标；接着讨论了苹果叫停低成本 Vision Pro 屏幕研发背后的硬件与注意力困局，以及 Getty Images 与 Shutterstock 37 亿美元合并案因英国反垄断阻击告吹、传统图库被生成式 AI 重创的行业惨状。随后，主持人们闲聊了纽约废弃 IBM 园区变身探险圣地与“邪恶 AI 实验室”的趣闻。最后，节目重点解读了 OpenAI 推出的全双工实时语音模型 GPT Live，以及即将登场的 GPT-5.6 系列大模型在顶级智力分化与 Computer Use（电脑操控）能力上的飞跃。 ## Atmosphere & style 整期节目洋溢着硅谷极客圈特有的轻松诙谐与辛辣讽刺，两位主持人默契十足，在严肃的商业研判与无厘头的脑洞吐槽之间切换自如。从调侃贝索斯长达 25 年的“自力更生”式烧钱，到戏谑英国监管机构对两家被 AI 逼到悬崖边缘的图库“落井下石”，再到构想用 AI 在《反恐精英》里代打上分，对话充满密集的笑点与反讽。听感宛如坐在极客酒吧里旁听两位行业老手畅聊内幕，既有对前沿技术突破的敏锐洞察，又带着消解行业炒作泡沫的清醒与松弛。 ## Bulletin - [00:03](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=3s) 蓝色起源（Blue Origin）寻求首轮 100 亿美元外部融资，估值预计达 1300 亿美元；贝索斯与 Coatue 分别认购 20 亿与 40 亿美元，用于支撑公司今年约 50 亿美元的高额现金消耗。 - [04:32](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=272s) 苹果已叫停与三星显示合作的廉价版 Vision Pro 屏幕项目，在硬件沉重、缺乏 PMF 及开发者生态难以破局的背景下，苹果正将更多注意力转向轻量化智能眼镜。 - [09:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=596s) Getty Images 与 Shutterstock 宣布终止此前估值 37 亿美元的合并案，主因是英国反垄断监管要求剥离编辑业务；受生成式 AI 工具严重分流冲击，两家公司当前市值均已缩水至约 3.3 亿至 3.4 亿美元。 - [19:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1196s) OpenAI 推出具备实时全双工架构的 GPT Live 语音模型，并宣布 GPT-5.6（Soul、Terra、Luna）将于周四发布，大模型竞争迈入数据源差异化与 Computer Use（电脑操控）实用能力驱动的新阶段。 ## Key takeaways - [00:03](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=3s) 蓝色起源成立 25 年来累计烧钱约 270 亿美元，首度开放 100 亿美元外部融资不仅让贝索斯的长期投入获得公开估值坐标，也为其锁定了 12 至 18 个月的运营跑道。 - [03:47](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=227s) 商业航天领域的估值逻辑脱离了传统市销率，更看重稀缺的不可替代战略能力；蓝色起源作为全球仅次于 SpaceX 实现火箭入轨及回收着陆的选手，享有极高的战略定价。 - [04:32](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=272s) 初代 Vision Pro 依赖采购前沿但良率极低的原型屏幕以拉开规格差距，但随着低成本面板合作项目的叫停，通过平价硬件普及大众市场的路径受阻。 - [07:45](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=465s) 重型 VR 观影的受阻反映出当代注意力的结构性转移：被动封闭的长视频体验难以抗衡手机信息流高操作频率（APM）和微互动所带来的掌控感与多任务属性。 - [09:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=596s) 在生成式 AI 生成图像质量逼近真实摄影、导致传统图库业务遭受降维打击的背景下，Getty 与 Shutterstock 抱团取暖的诉求被英国监管机构强行切断。 - [09:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=596s) 反垄断监管在应对前沿科技变革时往往存在滞后性，在新兴技术已颠覆行业基本面时继续实施严苛阻击，容易沦为对困境中传统企业的落井下石。 - [18:53](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1133s) 城市废墟探险（Urbex）在社交平台算法助推下走向泛滥并引发治安隐患，折射出大众网络打卡风潮与传统巨头遗存物理地标之间的荒诞冲突。 - [19:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1196s) GPT Live 的全双工架构支持双向边听边说与实时插话，并支持将复杂推理请求路由到后台异步执行，使人机交互首次具备类人的前后台多线程协作潜力。 - [23:25](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1405s) 大模型行业正告别同质化的全网通用语料爬取，依托差异化垂直数据源与严格的模型蒸馏，顶级模型在综合性能跃升的同时呈现出鲜明的风格与功能分化。 - [25:51](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1551s) Computer Use（电脑操控自动化）正大幅提升大模型在复杂工作流中的依赖度，但大众认知普遍滞后于模型迭代，亟需更具说服力的实际工作范式来消除认知断层。 ## Detailed walkthrough ### [00:03](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=3s) 蓝色起源首轮外部融资：百亿美元注入，估值达1300亿 节目在轻松的调侃中开场，主持人打趣说《Invest Like the Best》的主题曲让人迅速进入极佳的精神状态，而眼下同样心情大好的当属杰夫·贝索斯（Jeff Bezos）——因为他终于能给自掏腰包投入的数百亿美元来一次估值上调了。《纽约时报》的安德鲁·罗斯·索金（Andrew Ross Sorkin）独家爆料（主持人们戏称这很可能是从太阳谷峰会流出的八卦），蓝色起源（Blue Origin）正在寻求首轮外部融资，筹资金额高达 100 亿美元，预计估值将达到 1300 亿美元。贝索斯本人将直接追加 20 亿美元投资，而知名投资机构 Coatue 则获得了双倍的份额，认购 40 亿美元，其余部分将向其他真正的外部投资机构开放。 有趣的是，贝索斯的家族办公室本身就是 Coatue 旗下专注于前沿科技初创企业的“创新策略基金”（Innovative Strategies Fund）的重要出资人（LP），双方早已建立了深厚的合作基础。把一家成立了 25 年、全靠创始人个人财富单打独斗支撑至今的企业定义为“自力更生”（bootstrapped），听起来有些滑稽，但这确实是贝索斯二十五年磨一剑的真实写照。在此之前，贝索斯几乎是凭一己之力资助他的星际探索梦想，如今公司历史累计烧钱已达约 270 亿美元，平均每年消耗 10 亿美元左右；而仅在今年，蓝色起源的预计现金消耗就达到 50 亿美元。因此，这次 100 亿美元的融资正好相当于标准且合理的 12 到 18 个月运营跑道。 对于 1300 亿美元的非公开市场估值，主持人们感叹时代变迁之快。当年加文·贝克（Gavin Baker）在富达（Fidelity）任职时，曾力主将 Uber 估值为 140 亿美元，结果在竞价拍卖中推高至 170 亿美元，当时所有人都震惊于一家非上市公司竟能如此庞大；而放到今天，170 亿甚至 600 亿美元在私募市场上都显得十分“小巧玲珑”，毕竟当下已有数家万亿美元级以及大量破千亿美元估值的独角兽。 - **融资金额与估值**：筹资 100 亿美元，投后估值预计为 1300 亿美元。 - **主要出资方**：贝索斯直接注资 20 亿美元，Coatue 出资 40 亿美元，其余份额向外部机构开放。 - **财务消耗情况**：成立 25 年来累计烧钱约 270 亿美元，今年预计现金消耗（burn rate）为 50 亿美元。 ```mermaid flowchart TD accTitle: 蓝色起源 100 亿美元外部融资结构 accDescr: 展示蓝色起源首轮外部融资中贝索斯、Coatue 及其他外部投资者的出资比例与公司估值关系 subgraph Investors["主要投资方"] JB["杰夫·贝索斯<br>20 亿美元"] Coatue["Coatue 资本<br>40 亿美元"] Others["其他外部投资机构<br>40 亿美元"] end subgraph BlueOrigin["蓝色起源 (Blue Origin)"] BO["公司估值：1300 亿美元<br>累计消耗：约 270 亿美元<br>本年度消耗：50 亿美元"] end JB -->|直接追加| BO Coatue -->|领投分配| BO Others -->|参投首轮外部融资| BO ``` --- ### [03:47](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=227s) 商业航天估值坐标：SpaceX、AST SpaceMobile 与 Rocket Lab 面对蓝色起源的高估值，埃隆·马斯克（Elon Musk）的支持者们随即声称 SpaceX 显然被严重低估了。虽然外界尚不清楚蓝色起源的具体营收数据，但从其实际业务规模来看，营收数字必然不高，这意味着本轮融资对应着极高的市销率倍数。然而，投资人并非基于现金流或营收来对这类企业定价，核心考量在于“不可替代的战略能力”。蓝色起源是全球范围内仅次于 SpaceX、成功将火箭送入轨道并实现回收着陆、证明了可重复使用技术的第二家企业，其研发进度甚至抢在了多年来一直试图复制 SpaceX 路线的中国同行之前。尽管此前遭遇过发射台爆炸的重大挫折，但团队已找到了解决方案，目前正按部就班推进。 在评估这笔交易时，二级市场也提供了清晰的估值锚点： - **AST SpaceMobile (ASTS)**：尽管入轨载荷试验仍处于早期阶段，其市值已达到 270 亿至 300 亿美元左右。 - **Rocket Lab**：作为商业航天领域的成熟选手，公开市场估值同样逼近 500 亿美元。 主持人们指出，太空是一个庞大无比的增量市场，在创始人具备坚定决心、又有顶级资本保驾护航的情况下，投资人完全有理由保持长期乐观。 --- ### [04:32](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=272s) 苹果搁置廉价版 Vision Pro 屏幕：三星供应项目终止 随后，话题转向了一条令人唏嘘的硬件消费电子动态：据 MacRumors 报道，苹果已叫停为低成本版本 Apple Vision Pro 研发更便宜显示屏的计划，供应商三星显示（Samsung Display）预计将在 9 月前正式关停该项目。主持人调侃称，这对“包括我在内的全网仅剩的三位 Vision Pro 粉丝”构成了沉重打击。尽管有某位硬件创始人声称自己每晚都在坚持使用，但不可否认的是，Apple Vision Pro 从未真正找到产品市场契合点（Product-Market Fit, PMF）。 主持人们回顾了在演播室使用该设备看电影的惊艳体验。戴着它看《阿拉伯的劳伦斯》（*Lawrence of Arabia*）、《敦刻尔克》（*Dunkirk*）、《教父2》（*The Godfather Part II*）尤其是《怒海争锋》（*Master and Commander*），体验极具沉浸感，“让人感觉自己真的置身于一艘战舰之上”。然而，整个硬件生态陷入了死结：机身沉重且佩戴繁琐、售价高昂，导致开发者生态始终没有建立起来，形成典型的“先有鸡还是先有蛋”的僵局。 行业原本指望推出廉价版显示模组以降低整体售价和重量，从而推进行业普及。回顾初代 Vision Pro 的技术路径，苹果当年为了在显示规格上彻底超越通过收购 Oculus 稳步迭代的 Meta，走了一条极为激进的捷径： 1. **跨代采买**：苹果直接找到三星，要求采购对方原本规划在 3 至 4 年后才推出的超前原型显示屏。 2. **容忍极低良率**：这类前沿面板尚未接入自动化产线，需要大量手工调试，良品率极低（通常早期仅有百分之几，而成熟量产需要 90% 到 99.9%）。 3. **高定价覆盖成本**：苹果当时的打法是“不惜代价用超高采购成本买下最先进技术，再通过昂贵的售价转嫁”，以此确立顶级规格的壁垒。 原本外界预期经过数年发展，这类高端微型显示面板会逐渐普及并平民化，但苹果此次单方面叫停低成本显示屏的研发，预示着廉价版头显在短期内难以落地。 --- ### [07:45](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=465s) 沉浸式观影 vs. 刷屏成瘾：VR 头显为何陷入受众困境 苹果虽然暂停了廉价版 Vision Pro 屏幕的推进，但并未停止在轻量化智能眼镜（Specs / Smart Glasses）上的研发投入。这一战略取舍引发了关于当代用户注意力变迁的深层讨论：苹果或许敏锐地察觉到，“坐下来戴着头显看完整部电影”的用户习惯在今天已经极度边缘化。 主持人们深入剖析了当下的消费心理与行为习惯： - **沉浸式长视频的没落**：如今很少有人能够静下心完整阅读一本书或从头到尾看完一部两小时的电影；一旦在 VR 中开启全屏观影模式，用户就会被困在单向输入的信息流中，中途不能分心。 - **“刷屏中毒”（Doomscrolling）的机制优势**：普通人在手机上无休止刷短视频虽然被批评为“大脑腐蚀”和自我孤立，但在交互心理学上它实际上具有极强的“多人互动与游戏属性”。 - **操作交互频率（APM）的悬殊**：刷信息流的过程包含密集的微交互——点开评论区互动、转发给好友吐槽、唤醒 AI 求证事实真相、点击跳转至博主主页等。这种高频、自主掌控的互动体验是 VR 观影完全无法提供的，VR 头显也根本无法增强这种体验。 随着手机滑动浏览彻底主导大众的碎片时间，传统主打沉浸式闭门体验的重型 VR 设备在日常场景中的存在感正在迅速萎缩。VR 往往只能吸引少数极端前卫的技术理想主义者，却始终难以跨越鸿沟迈向真正的大众日常消费市场。 ### [09:28](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=568s) 硬件坚守者与 VR 赛道的执着 在科技行业的长周期波动中，能够顶住外界质疑坚持研发的创业者尤为可贵。节目开篇聊到了那些愿意在漫长的炒作周期（hype cycles）中“看起来像个傻子”，却依然不断低头做产品的硬件团队。主持人特别点名了 Bigscreen 及其创始人——对方此前曾带着一款极具未来感的轻量化眼镜原型登台演示。在主持人看来，这位创始人可能是全地球日常使用 AR 和 VR 设备时间最长的人，对这一赛道的胜利抱有极度专注与纯粹的执念。这种在泡沫退去后仍能耐得住寂寞、持续迭代的工程师精神，正是整个消费级硬件生态最稀缺且令人敬佩的特质。 ### [09:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=596s) Getty Images 与 Shutterstock 37 亿美元合并案告吹 另一则引人注目的重磅新闻，是视觉素材行业两大巨头 Getty Images 与 Shutterstock 宣布终止合并。Getty 一向以记录硅谷名流而闻名，尤其是每年正在举行的太阳谷峰会（Sun Valley），吃瓜群众要想第一时间看到乔希·库什纳（Josh Kushner）等投资巨鳄的街拍，往往得依赖 Getty 的订阅。然而在底层业务上，两家传统图库正面临生成式 AI 的毁灭性冲击：从 Meta 推出的 Muse、Nano Banana Pro，到 OpenAI 的图像生成工具，AI 生成的森林、风景等通用素材质量已完全媲美真实摄影，连 Instagram 负责人亚当·莫塞里（Adam Mosseri）发布的高清自拍都让人难辨真伪。面对客户被免费或廉价 AI 工具分流的惨状，两家公司近年来主要靠向 AI 大模型实验室授权训练数据维持增量收入，抱团取暖本是行业大势所趋。 ```mermaid flowchart TD accTitle: Getty Images 与 Shutterstock 合并受阻与市值缩水历程 accDescr: 展现从2025年1月合并宣布到被英国监管否决后市值大幅跌落的全过程 subgraph DealFlow ["交易推进与监管博弈"] A["2025年1月：宣布合并<br/>基于当时股价估值达 37 亿美元"] B["2025年4月：美国司法部批准"] C["英国竞争与市场管理局（CMA）介入<br/>要求剥离 Shutterstock 编辑业务"] D["Getty 董事会拒绝剥离<br/>于周二正式发函终止合并"] A --> B --> C --> D end subgraph MarketImpact ["估值暴跌与市场惨状"] E["传统图库客户被生成式 AI 严重分流"] F["Shutterstock 市值跌至约 3.29 亿美元"] G["Getty Images 市值跌至约 3.40 亿美元"] E --> F E --> G end D -.-> MarketImpact ``` 这笔最初于 2025 年 1 月宣布、估值达 37 亿美元的合并案，在 2025 年 4 月已顺利通过了美国司法部（DOJ）的反垄断审查，却最终在英国遭遇灭顶之灾。英国竞争与市场管理局（CMA）坚持要求交易必须以剥离 Shutterstock 的编辑业务为前提，认定合并会导致英国媒体的选择减少并推高图片采购价格；Getty 董事会断然拒绝该条件，直接发函终止合并。这导致 Getty 盘前股价下跌 1.7%，Shutterstock 微涨 1%。主持人对此感到极其荒谬：如今无论媒体还是个人都在拥抱 AI 生成图，合并赋予的所谓微弱定价权，不过是为了弥补客户流失的生存本能。更具讽刺意味的是，当年 37 亿美元的估值基于 2025 年初的股价，而如今 Shutterstock 和 Getty 的市值已双双腰斩暴跌至仅剩 3.29 亿和 3.40 亿美元。英国监管机构此举非但没有保护市场，反倒像是在对两家奄奄一息的公司“落井下石”。 - 历史监管案例对照：美国监管层曾以“防止垄断 VR 健身行业”为由阻击 Meta 收购 Within（Supernatural 开发商），而事实证明所谓 VR 健身行业根本未成气候；英国 CMA 亦曾在 2022 年强行阻击 Meta 收购动图平台 Giphy。 - 动图生态补充：针对 iMessage 内部集成的动图与贴纸搜索功能，节目澄清其图像搜索接口实际由微软 Bing 驱动，而非完全被 Giphy 垄断，市场本就存在竞争。 - 行业标注乱象：在线设计平台 Canva 目前同样在图库中混用原生与 AI 生成图像，且据观众反馈并不会主动对 AI 生成内容添加标签。 ### [16:59](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1019s) 纽约废弃 IBM 园区成“探险圣地”，适合做 AI 邪恶实验室？ 节目随后将视线转向《华尔街日报》头条报道的一处奇特地标——位于纽约州萨默斯（Somers）、被戏称为“世界上最诡异办公室”的废弃 IBM 园区。这座建筑群远离曼哈顿尘嚣（车程约一小时），空置多年后不仅爬满了涂鸦，还沦为了全美“废墟探险者”（urban explorers）趋之若鹜的圣地。报道讲述了一则荒诞的抓捕插曲：2026 年 4 月的一天，62 岁的退休工程师罗伯特·卡尔顿（Robert Carlton）在家门口取信时，迎面撞见一群从树林里夺路狂奔的少年；老人当即横在路中央张开双臂断喝道：“停下，一切都结束了！”（"Stop. It's over."）。紧接着三名纽约州警从林间冲出，以非法侵入（trespassing）的罪名将这 7 名少年当场逮捕，罪名正是“冒犯了国际商业机器公司（IBM）”。 主持人们对着园区的航拍与内景视频大开脑洞，调侃称这绝对是当下 AI 创业团队梦寐以求的总部圣地。如果有哪家风格硬核、主打末日科技感的新锐实验室（Neo-lab）——名字最好叫作“险恶智能”（Ominous Intelligence）——完全可以低价接盘这处园区，将幽暗空旷的厂房改造成微型数据中心。整座建筑自带强烈的“007 电影大反派巢穴”气场，员工在深夜加班穿行于空旷肃杀的长廊里，既有硬核黑客的闭关锁死氛围，又能完美契合极客对技术统治力的冷酷想象。 ### [18:53](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1133s) 从 IBM Watson 与废墟探险热潮聊起 节目在轻松的闲聊中展开。主持人首先调侃起 IBM 的前办公室以及当年的 IBM Watson。Watson 曾被视为机器智能的先驱，在《危险边缘》（Jeopardy!）智力竞赛中大放异彩，堪称那个时代的前沿科技象征。随后，话题跳转到了当下社交媒体上极为火爆的“Urbex”（城市废墟探险）现象。探险本身并不新鲜，但在 Instagram 和 TikTok 上那些极具艺术感、充满滤镜的视频助推下，这一小众爱好被彻底引爆，吸引了大量盲目的好奇者争相打卡。 这种风潮也带来了实质性的公共安全隐患。新泽西州利文斯顿（Livingston）警方最近发布紧急警告，严禁公众进入已关闭的利文斯顿购物中心（Livingston Mall）；佛罗里达州杰克逊维尔海滩（Jacksonville Beach）警方也在 4 月对闯入废弃的 Adventure Landing 游乐园发出了类似警告。主持人感叹，潜入废弃游乐园确实是顶级的心跳冒险，但 TikTok 上的探险挑战已经变得极其危险，甚至让人怀疑这种煽动违规冒险的内容是否还应被允许传播。 - **IBM Watson**：曾是 AI 领域的门面代表，通过在《Jeopardy!》击败人类冠军而声名大噪。 - **Urbex（城市废墟探险）**：因短视频平台的病毒式传播而升温，导致多地警方向探险者发出治安与安全警告。 --- ### [19:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1196s) OpenAI 推出 GPT Live 语音模型：全双工架构与多线程交互设想 随后节目切入重磅科技新闻：OpenAI 正式发布并向 ChatGPT 用户推送新一代语音模型 GPT Live，旨在提供极其自然的 AI 人机交互体验。针对 Instagram 上那些专门挑旧版语音模型刺、抱怨其体验差的博主，主持人打趣称对方简直是语音模型界的“释放者普林尼”（Pliny the Liberator）或白帽黑客。尽管这类用户总在冷嘲热讽，但他们确实精准指出了旧模型的硬伤——比如缺乏工具调用能力（无法直接计时或数数）、缺乏超长上下文窗口导致对话容易遗忘、以及无法多智能体互相对话。如果新模型能扛住这些人的刁钻测试，产品才算真正成熟。 主持人探讨了实时全双工（Full-Duplex）架构的真正威力。全双工意味着 AI 能够一边聆听一边说话，随时感知语气并恰当切入，同时还能将复杂问题分发给后台更强的推理模型去运算。主持人分享了自己的用车痛点：他在开车时往往不会用旧版语音模式，而是点按录音键口述一段超长的复杂 Prompt，发送后等待后台生成，再让它读出来。全双工不仅能消除这种繁琐的多步操作，更开启了“多线程对话”的可能。在同一个对话流中，用户可以同时交代 AI：“去写一份关于 IBM Watson 的深度研究报告”，在 AI 后台运转的同时，继续同它闲聊最新的早间新闻；随后 AI 适时汇报报告进度，用户再无缝切入其他工作决策讨论，整个过程如同与一位高效的人类助手协作。 ```mermaid flowchart TD accTitle: 全双工多线程语音交互架构 accDescr: 展示用户通过语音与模型同时进行前台闲聊与后台异步任务分发的过程 subgraph UserInteraction ["用户端交互"] U["用户语音输入"] end subgraph FullDuplexEngine ["全双工对话引擎 (GPT Live)"] Listen["实时双向监听与打断"] Router{"任务智能路由"} end subgraph ExecutionPaths ["处理路径"] Foreground["前台实时交流<br/>(新闻资讯、即时闲聊)"] Background["后台强推理模型<br/>(深度研究报告、代码编写)"] end U --> Listen Listen --> Router Router -->|即时对话| Foreground Router -->|复杂委托| Background Background -.->|任务完成通知| Listen ``` - **全双工架构（Full Duplex）**：支持双向并发语音，边听边说，具备自然的插话与打断能力。 - **任务分发与并发**：可将复杂任务路由至后台推理模型，不中断当前前台对话流。 --- ### [23:25](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1405s) GPT-5.6 系列登场与 AI 大模型的“分化时代” 除了 GPT Live，OpenAI 旗下的大模型 GPT-5.6 Soul 以及 Terra、Luna 版本也宣布将于本周四正式公开发布，全球预览测试资格正在加速扩大。社交媒体上早已被内测资格相关的段子刷屏。例如 John Palmer 调侃道：“看来我现在可以聊聊 5.6 了。虽然我不是内测人员，但我听说它非常棒。等我拿到权限，它大概会成为我的主力模型。”Kit Langton 则搞笑讽刺：“我用 5.6 已经好几个月甚至几十年了，回看过去的人生，显而易见的事实就是——你们根本没拿到权限。” 在正经的模型评测方面，测试者们普遍感受到了明显的风格与能力分化。有人评价 Fable 像“横跨星系的超光速跳跃”，而 GPT-5.6 Soul 则像“在纽博格林赛道上驰骋的保时捷”，极具操控感和参与感。AI 研究学者 Ethan 评价道：“Soul 和 Fable 都比上一代模型实现了大幅跃升，并拉开了与第二梯队的巨大差距；如果你的工作极其依赖顶级智力，这两个模型是唯二的选择。”Dean 也指出，AI 史上罕见地出现了两个顶级模型“既断层式领先其他对手，彼此之间又特色迥异”的局面。主持人分析，这种差异源自各家更加严格的蒸馏协议，以及训练数据的多样化——行业已经告别了全网抓取相同语料的阶段，不同公司开始依赖自有的专有数据集、数据经纪人以及团队独特的研究品味。 - **GPT-5.6 系列**：包含 Soul、Terra 和 Luna，周四公开发布。 - **双雄格局（Soul vs Fable）**：两者性能全面领先行业，但在互动性、计算风格与工作流体验上截然不同。 - **训练数据分化**：各机构摆脱单一全网语料，采用精细化蒸馏与差异化垂直数据源。 --- ### [25:51](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1551s) Computer Use 能力的探索与对未来的展望 针对 GPT-5.6 Soul 的实际功能，科技博主 Leo 指出其顶级的“Computer Use”（操作电脑）能力使其使用频率提升了整整 100 倍，甚至在早先因华盛顿特区诉讼行动导致访问中断的几周里，让人产生一种“没有它简直要发疯”的重度依赖。针对部分网友提问“Computer Use 到底有什么用”，主持人们反驳道，答案正是“任何你在电脑上做的事情”。就像 Deep Research（深度研究）功能通过直接输出 20 页 PDF 报告树立了使用范式一样，普及真正有价值的 AI 工作流至关重要。 主持人指出，大众对 AI 的理解存在明显的断层：一小部分高级玩家能瞬间抓住新模型的特点，高效榨取价值而不浪费 token；但大部分普通用户依然受困于陈旧的认知，用一年前早已解决的“幻觉问题”或“知识库截止时间”等网络烂梗来评判当下的前沿模型。因此，行业需要更多生动的 Computer Use 演示来重塑认知。节目的最后，主持人开玩笑称，检验 Computer Use 终极实力的最好标准根本不是什么办公软件，而是看它“能不能在《反恐精英》（Counter-Strike）里帮我打上全球精英（Global Elite），或者帮我故意掉分去炸鱼（smurf）”。在欢笑与击掌声中，本期节目正式收尾。 - **电脑操控（Computer Use）**：成为 GPT-5.6 的核心飞跃之一，大幅提升了自动化生产力。 - **认知滞后现象**：大众常因过时的网络梗（如幻觉、知识截断）忽视了模型的最新技术突破。 - **终极恶搞测试**：用 AI 代打《反恐精英》冲击大地球段位，被调侃为最硬核的电脑操作考验。 ## Notable quotes > [04:32](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=272s) 包括我在内的全网仅剩的三位 Vision Pro 粉丝 — 主持人 > [04:32](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=272s) 让人感觉自己真的置身于一艘战舰之上 — 主持人 > [09:28](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=568s) 看起来像个傻子 — 主持人 > [16:59](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1019s) 停下，一切都结束了！ — 罗伯特·卡尔顿 > [16:59](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1019s) 冒犯了国际商业机器公司（IBM） — 主持人 > [19:56](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1196s) 去写一份关于 IBM Watson 的深度研究报告 — 主持人 > [23:25](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1405s) 看来我现在可以聊聊 5.6 了。虽然我不是内测人员，但我听说它非常棒。等我拿到权限，它大概会成为我的主力模型。 — John Palmer > [23:25](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1405s) 我用 5.6 已经好几个月甚至几十年了，回看过去的人生，显而易见的事实就是——你们根本没拿到权限。 — Kit Langton > [23:25](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1405s) Soul 和 Fable 都比上一代模型实现了大幅跃升，并拉开了与第二梯队的巨大差距；如果你的工作极其依赖顶级智力，这两个模型是唯二的选择。 — Ethan > [25:51](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1551s) 任何你在电脑上做的事情 — 主持人 > [25:51](https://www.youtube.com/watch?v=TeqsmQtDWnE&t=1551s) 能不能在《反恐精英》（Counter-Strike）里帮我打上全球精英（Global Elite），或者帮我故意掉分去炸鱼（smurf） — 主持人 ## Mentioned 杰夫·贝索斯 (Jeff Bezos), 安德鲁·罗斯·索金 (Andrew Ross Sorkin), 加文·贝克 (Gavin Baker), 埃隆·马斯克 (Elon Musk), 乔希·库什纳 (Josh Kushner), 亚当·莫塞里 (Adam Mosseri), 罗伯特·卡尔顿 (Robert Carlton), John Palmer, Kit Langton, Ethan, Dean, Leo, John Coogan, Jordi Hays, 蓝色起源 (Blue Origin), Coatue, SpaceX, AST SpaceMobile, Rocket Lab, 苹果 (Apple), 三星显示 (Samsung Display), Meta, Oculus, Bigscreen, Getty Images, Shutterstock, 英国竞争与市场管理局 (CMA), 美国司法部 (DOJ), Giphy, Canva, 微软 (Microsoft), IBM, OpenAI, Apple Vision Pro, GPT Live, GPT-5.6, GPT-5.6 Soul, Terra, Luna, Fable, IBM Watson, Deep Research, Computer Use, Muse, Nano Banana Pro, 《反恐精英》 (Counter-Strike), 《危险边缘》 (Jeopardy!), 《怒海争锋》 (Master and Commander), 《阿拉伯的劳伦斯》 (Lawrence of Arabia), 《敦刻尔克》 (Dunkirk), 《教父2》 (The Godfather Part II) ## 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 [Source](https://www.youtube.com/watch?v=TeqsmQtDWnE)

[4:32] 主持人: 包括我在内的全网仅剩的三位 Vision Pro 粉丝 [4:32] 主持人: 让人感觉自己真的置身于一艘战舰之上 [9:28] 主持人: 看起来像个傻子 [16:59] 罗伯特·卡尔顿: 停下，一切都结束了！ [16:59] 主持人: 冒犯了国际商业机器公司（IBM） [19:56] 主持人: 去写一份关于 IBM Watson 的深度研究报告 [23:25] John Palmer: 看来我现在可以聊聊 5.6 了。虽然我不是内测人员，但我听说它非常棒。等我拿到权限，它大概会成为我的主力模型。 [23:25] Kit Langton: 我用 5.6 已经好几个月甚至几十年了，回看过去的人生，显而易见的事实就是——你们根本没拿到权限。 [23:25] Ethan: Soul 和 Fable 都比上一代模型实现了大幅跃升，并拉开了与第二梯队的巨大差距；如果你的工作极其依赖顶级智力，这两个模型是唯二的选择。 [25:51] 主持人: 任何你在电脑上做的事情 [25:51] 主持人: 能不能在《反恐精英》（Counter-Strike）里帮我打上全球精英（Global Elite），或者帮我故意掉分去炸鱼（smurf）

[23:24] John Palmer: this will probably be my default model from here on [23:24] Kit Langton: many months, if not decades

[0:03] It's great to be back. [0:04] >> Yes, the invest like the best hype song [0:07] really [0:08] it really has [0:10] I don't know, it just burned its way [0:12] >> Yeah, puts me in the zone in an [0:13] incredible [0:14] >> Yeah. [0:14] >> mental state. [0:15] >> Uh 100% 100%. You know who else is in a [0:18] good mental state? Jeff Bezos cuz he's [0:20] getting a markup on the 25 billion he's [0:24] putting into [0:24] >> himself up. [0:25] >> himself up. [0:26] >> Alongside Coatue. [0:27] >> Yeah, the news is that Blue Origin is [0:30] raising outside capital for the first [0:32] time. He's participating. A fund Coatue [0:35] who he is also invested in through his [0:37] family office is also participating. Um [0:39] but there will be an opportunity for [0:41] some other true outsiders to get their [0:43] first slice of of Blue Origin. Uh Andrew [0:46] Ross Sorkin had the scoop in the New [0:48] York Times today. You have to wonder if [0:51] this is [0:52] gossip coming out of Sun Valley. Uh the [0:54] company's raising $10 billion. This is [0:56] the first outside fund raise. Uh it's [0:59] hilarious to think about this [1:00] 25-year-old company as being previously [1:03] bootstrapped, but that's technically [1:04] what it is basically. Uh one founder's [1:07] capital just uh chipping away at a [1:10] really hard problem for a quarter [1:11] century and now we are here. Bezos has [1:15] been funding this basically solo on his [1:18] uh quest to visit the stars. Uh he's [1:21] doubling down. [1:22] >> People can't really complain about the [1:23] valuation cuz he's a big [1:25] >> Yeah. [1:25] >> part of uh setting the price. [1:27] >> Yeah, yeah. So uh he's doubling down. [1:30] He's putting in $2 billion directly and [1:32] Coatue is getting twice as much [1:34] allocation, $4 billion. Uh there's [1:36] another interesting Coatue tidbit I [1:38] alluded to earlier from the New York [1:40] Times piece. Bezos's family office is a [1:43] major investor in Coatue's innovative [1:45] strategies fund which is focused on [1:47] emerging technology startups. So I'm not [1:49] sure if the money's coming out of that [1:51] fund specifically, but there is a [1:52] connection there. He's been building a [1:54] relationship with Coatue for a long time [1:56] and they are optimistic about Blue [1:58] Origin's prospects. The valuation is [2:00] expected to be a hundred and thirty [2:01] billion dollars. Big for a private [2:04] company. I was listening to Gavin Baker [2:06] and Travis Kalanick talk about how crazy [2:08] it was when Gavin was at Fidelity and [2:12] the he wanted to value Uber at fourteen [2:14] billion and everyone thought he was [2:16] crazy. The valuation turned into an [2:17] auction, came in at seventeen and [2:19] everyone was like this is unprecedented. [2:22] We've never seen a private company this [2:24] big and now uh seventeen billion, even [2:26] sixty billion in the private markets [2:28] looks quaint by modern standards in an [2:30] era where we have uh several trillion [2:33] dollar private companies, uh many up in [2:36] the hundred billion plus range. Uh it's [2:38] a wild wild time. The Elon bulls are [2:41] saying oh SpaceX is underpriced now [2:43] because you have to imagine I don't know [2:45] if there's an exact revenue figure [2:47] that's leaked for Blue Origin, but based [2:49] on the size of that business it has to [2:50] be small and so this has to be a very [2:53] high multiple, but you're not really [2:55] valuing this company on cash flow or [2:57] even revenue. You're valuing it on the [2:59] capability which is it's the second [3:02] company in the world I believe to uh [3:04] bring a rocket to orbit, land it [3:06] successfully, [3:08] you know, prove you reusability. They [3:10] got there before China which has been [3:12] trying to copy SpaceX for years and so [3:14] lots of really solid progress. Of course [3:16] there was that setback with the uh [3:19] launchpad that exploded, but they [3:21] figured out a solution to that and [3:23] things are chopping along just uh just [3:25] fine. The company's estimated to burn [3:27] five billion this year. So, this looks [3:30] like a pretty standard twelve to [3:32] eighteen month fundraise at ten billion [3:34] and it's burned something like [3:35] twenty-seven billion to date which, you [3:38] know, a billion a year on average over [3:39] twenty-five years, not too bad for such [3:42] an ambitious project. Still a lot of [3:43] money, uh but a big opportunity. And uh [3:47] and you have to imagine um [3:49] there are a number of other comps like [3:51] AST space mobile as you like to call [3:53] them. [3:54] >> [laughter] [3:54] >> AST space mobile [3:56] >> the AST [3:57] space mobile. [3:57] >> ASTS market cap [3:59] >> 27, 28 billion dollar market [4:00] >> 30 billion dollar company and I don't [4:02] believe that they've gotten much to [4:03] orbit yet. They've done some [4:05] experiments, some tests. They're still [4:06] early, but it's a huge market and we've [4:08] seen how fast [4:10] >> of mass to orbit. [4:11] >> Yeah, and and and we've seen how fast [4:13] Rocket Lab and various uh it's a big [4:15] really it's a really really big market [4:16] and so when you're going after big [4:18] opportunity and you've been working on [4:19] it a long time and you got Jeff Bezos [4:22] funding it uh lots of reasons to be [4:24] optimistic. [4:25] >> Another comp uh investors would [4:27] certainly look at would be Rocket Lab [4:29] sitting at just under 50 billion. [4:32] >> We have some heartbreaking news. [4:34] Heartbreaking news. Apple has scrapped [4:37] plans for a cheaper Apple Vision Pro [4:39] display and is winding down work at [4:42] Samsung. This is from MacRumors. Really [4:45] hits me and the other three Apple Vision [4:47] Pro fans particularly hard. Uh [4:50] >> Brutal. [4:50] >> I did I Who did I talk to? I talked to [4:52] somebody who said that they they [4:53] actually are a daily user of the Apple [4:56] Vision Pro. I thought I was in the top [4:57] 1% [4:58] >> the the founder that's making [5:00] >> Yes, that's right. Yeah. He says he [5:02] watches it he watches he uses it every [5:04] night or something like that. That's a [5:05] lot of Apple Vision Pro use. Obviously [5:08] the the the product never really found [5:09] product market fit. Uh although we have [5:11] one in the studio, we've been having fun [5:13] with it. Uh I still think it's a it's [5:15] it's an incredible piece of technology, [5:17] but a little bit of a chicken and an egg [5:19] problem. The developers never showed up [5:21] and it was very expensive, very heavy, [5:22] very cumbersome. It requires a lot to [5:25] put that thing on to watch Lawrence of [5:26] Arabia, but when you do, it's worth it. [5:29] It's worth it. What did I watch most [5:31] recently? I watched Dunkirk in it and I [5:33] watched Godfather Part II and Master and [5:36] Commander. Master and Commander, that's [5:38] a great movie especially in Vision Pro. [5:40] Really, you want to be in the cinematic [5:42] feel. Feel like you're on a on a on a [5:45] warship. Uh so Apple has has reportedly [5:48] stopped development on a cheaper display [5:50] for a low lower cost Vision Pro. This [5:52] feels like what the industry needs to [5:55] move forward. The actual display was [5:57] fantastic. It just needed to be cheaper [5:59] and lighter so more people could get [6:00] them. That at least was my thesis, but [6:02] Apple disagrees with me. What do I know? [6:04] So, uh supplier Samsung Display expected [6:07] to formally end the project by [6:08] September. And if you remember, there [6:10] were rumors of the crazy hoops that [6:13] Apple had to jump through to get that [6:15] incredible display because they really [6:17] leapfrogged Meta, who had been working [6:19] with the acquisition of Oculus, and [6:20] they'd been improving the [6:23] uh the the display fidelity. They got [6:25] rid of the screen door effect. They were [6:27] increasing the resolution per eye. [6:29] Everything was getting better about that [6:31] display, but on a very linear trend. And [6:34] allegedly, uh what Apple did was they [6:36] went to Samsung and said, "We know that [6:39] you are working on next year's display. [6:41] You're also working on displays 2 years [6:43] out, 3 years out, 4 years out." Now, [6:46] what's the problem with the one that's 4 [6:47] years out? Well, it's amazing, but it's [6:49] not ready for manufacturing at scale. [6:51] It's not on an automated line. It needs [6:53] a lot of manual work to get it to [6:56] actually produce, and there's some crazy [6:58] low yield. So, in in in chips, in [7:01] displays, yield rules everything. If you [7:05] run the machine a bunch of times, you [7:07] want to get 99.9% [7:10] yield, 90%. In the early days, when [7:12] you're prototyping a new technology, you [7:14] might be getting just a few percent [7:16] yield. And Apple said, "That's fine. [7:18] We'll just jack up the price and pay a [7:20] ton for these cutting-edge displays, [7:22] jump to the front of the category with [7:24] the best display technology. We'll [7:26] charge a fortune for it, but we'll have [7:27] the best product, and that will be our [7:28] market entry strategy." And so, I was [7:30] waiting for that display technology to [7:32] commoditize because it's been a few [7:34] years. Uh I thought it would just [7:36] naturally get cheaper. It probably has, [7:38] but it has not solved the the [7:40] of virtual reality, augmented reality. [7:43] We haven't seen someone pick up [7:45] >> they're just bearish on the near term of [7:48] people watching movies? [7:50] >> Maybe. Yeah. [7:51] >> In in in VR because I don't They're [7:54] They're certainly not stopping [7:55] [laughter] [7:56] R&D on their specs and their and their [8:00] sort of everyday glasses product. And [8:02] so, it might just be that that they [8:04] studied you and found that you're built [8:07] different. You're one of one, the only [8:09] person on Earth. [8:11] >> No, no. I mean, truly, like sitting down [8:12] and watching a full movie is a is a rare [8:14] experience today. People aren't reading, [8:16] they're not watching movies, they're [8:18] just scrolling, and people seem to be [8:20] satisfied with their phone. And also, [8:22] the nature of a doom scroll is is [8:25] somewhat interactive, somewhat [8:26] multiplayer. Although, obviously, people [8:29] see it as brain rotting and very [8:30] isolating, there is a game to [8:34] >> [clears throat] [8:34] >> be played by [8:35] actively scrolling, which is something [8:38] that doesn't require virtual reality. [8:39] Certainly not enhanced by virtual [8:41] reality. But then also, sometimes you [8:43] want to read the comments. Sometimes you [8:44] want to send it to a friend. Sometimes [8:46] you want to fact-check it and ask AI [8:48] about it or spin off and do something [8:50] else and go to someone's profile. So, I [8:53] I think that the number of taps and user [8:55] interaction experiences in a doom scroll [8:58] session are obviously much much higher [9:01] than watching a film, which you just put [9:02] on and hopefully you don't pause it at [9:04] all because you're engrossed the entire [9:06] time from start to finish. [9:07] And so, if that becomes as scrolling [9:10] means more and more, maybe the the VR [9:12] experience is less less and less [9:14] relevant. So, [9:15] >> VR really attracts these [9:18] >> visionary visionary chats, [9:21] >> that are willing to go that that seem [9:23] No, no. I'm actually [9:25] I could make that joke, but I won't. But [9:28] just just people that are willing to [9:30] just continually look silly for a long [9:34] time to ride through the hype cycles and [9:36] just like [9:37] >> True. [9:37] >> keep building. I think Bigscreen is an [9:39] is an example of that. The founder that [9:41] we had on that has a very cool glasses [9:44] uh demo product. He's I from my [9:46] understanding probably using uh AR and [9:49] VR more than anyone else on the planet. [9:51] Like so committed to just like winning. [9:53] >> Yeah. [9:53] >> Um and so I really appreciate that about [9:55] the category. [9:56] >> Yeah. [9:56] >> Here is a crazy story. I mean it's such [9:58] a minor story, but uh it certainly [10:00] grinds my gears. I think we're going to [10:03] dig into it and see what you think, what [10:04] I think once we dig into it. But Getty [10:06] Images, we know and love them from their [10:09] trove of silicon of Silicon Valley [10:12] titans photographed [10:13] >> at Sun Valley. [10:14] >> at Sun Valley, which is happening now. [10:16] If you don't have a Getty Images [10:17] subscription, sign up now because you're [10:19] going to be first to know when the new [10:21] Josh Kushner uh [10:24] paparazzi photo drops. Getty Images was [10:26] planning to merge with Shutterstock. [10:28] These are two essentially stock photo [10:31] sites. In theory, they should be [10:33] absolutely decimated by AI image [10:35] generation. Meta put out a new one, Muse [10:39] Muse image. Uh [10:41] Nano Banana Pro has been fantastic. [10:44] Images too from Open AI has been really [10:46] good. And like they're they're honestly [10:48] getting to the point where a lot of the [10:50] example images are just like, okay, [10:52] that's just a photo. I saw Adam Mosseri [10:54] posted some on Instagram and I was like, [10:56] is this AI or did you just take this [10:58] photo? Like it's just you. And then he [11:00] did a couple with filters and hair and [11:03] crazy hair. And those were cool. And and [11:04] that breeds creativity. So I think [11:06] there's a lot of value in the [11:07] technology. But in terms of just like [11:09] I'm trying to illustrate I'm trying to [11:11] make a point visually, uh I need a [11:13] picture of a forest. AI image generation [11:16] is very, very competitive with the stock [11:18] image industry. And this should have [11:21] this should be a business that's sort of [11:23] affected like the way [11:25] >> Yeah, I imagine a lot of a lot of these [11:27] businesses [11:27] >> cheating website check. [11:28] >> move their net new revenue over the last [11:30] couple years is just working with labs [11:32] that need [11:33] >> licensing, yeah, all sorts of things. [11:35] So, um certainly in the face of growing [11:38] competition, consolidation of an [11:40] industry is very normal. That makes a [11:42] lot of sense. This doesn't This doesn't [11:44] seem like it should raise any regulatory [11:45] flags, and yet it did in the United [11:47] Kingdom. Uh Getty Images has called off [11:50] the Shutterstock deal after the UK uh [11:53] created some uh barriers around it. So, [11:56] the $3.7 billion deal would have created [11:59] a visual content company able to offer a [12:01] more expansive library to users, says [12:03] the Wall Street Journal. On Tuesday, [12:05] Getty delivered a written notice to [12:07] Shutterstock terminating their planned [12:09] tie-up, which was first announced in [12:10] January of 2025, and was and received [12:12] clearance from the Justice Department in [12:14] America in April. So, pretty quick [12:17] turnaround in the United States. Uh the [12:18] $3.7 billion deal would have created a [12:20] visual content company able to deliver a [12:22] more expansive library to users, helping [12:25] meet booming demand for licensed images [12:27] and videos as artificial intelligence [12:29] disrupts the business of content [12:30] creation. Getty stock slipped 1.7% in [12:34] pre-market trading. Uh shares of [12:36] Shutterstock ticked up 1%. So, maybe the [12:39] Shutterstock shareholders are happy [12:40] about remaining an independent company. [12:43] The termination came after Getty last [12:45] week said that its board voted to not [12:48] proceed with the deal if it meant shut [12:50] selling [snorts] Shutterstock's [12:52] editorial business, a condition required [12:55] by the UK Competition and Markets [12:57] Authority. So, the decision to uh a uh [13:01] by Getty to abandon its merger with [13:02] Shutterstock was ultimately a commercial [13:04] choice, said the chair of independent [13:06] inquiry group that led the CMA's [13:08] investigation of the deal. The inquiry [13:10] group found that a loss of competition [13:12] between the two businesses would reduce [13:14] choice for UK media outlets and could [13:16] lead to higher prices. And I would I [13:18] would assume that this would give them [13:20] some sort of pricing [13:22] power in the face of, you know, [13:24] essentially unlimited free editorial [13:27] images in the form of AI generated [13:29] images. If you want to go with the real [13:31] thing, you have to pay a little bit [13:33] more, but that sort of offsets all the [13:36] pressure of the customers that are [13:37] leaving and no longer paying because [13:38] they don't need to pay because they're [13:40] generating them with generative AI. So, [13:42] this felt like a very logical story, um [13:46] and I'm I'm a little bit uncertain that [13:49] this is a good decision for the UK to [13:52] help these businesses. [clears throat] [13:53] It feels like it's just going to wind up [13:55] hurting both of them. [13:55] >> Yeah, are UK media outlets not allowed [13:58] to use AI content? [14:00] >> I don't think that there's any broad [14:01] rule against it, no. So, they [14:06] you would it would be a deliberate [14:07] choice, and I'm sure there's plenty I I [14:09] can imagine that the Financial Times [14:11] does not use AI images, and that's by [14:13] choice, and they make that very clear, [14:15] and there's probably a disclosure. And [14:17] if they Of co- Of course, they will use [14:18] an AI image if they're talking about the [14:20] AI image industry, but in terms of what [14:22] you see on the front page of the [14:23] Financial Times, it's almost always [14:25] editorial [14:26] >> that Yeah, that image. [14:28] >> This is from Reuters. [14:29] >> Okay, so I was trying to figure out why [14:32] that the headline number here was 3.7 [14:34] billion because it was based on the [14:36] January 6th, 2025 share prices, not [14:39] today's prices, which are far, far, far [14:42] lower. [14:43] >> Yeah. [14:43] >> Shutterstock is sitting at $329 million [14:47] market cap. Getty is at $340 million [14:50] market cap. [14:51] >> Woah. [14:51] >> So, these businesses have been [14:53] >> Okay. [14:53] >> really, really beat up. [14:55] >> So, [14:55] >> over the period since they decided, [14:57] "Hey, we should [14:58] >> Yeah. [14:58] >> tie up." [14:59] >> Interesting. So, is there is there a [15:01] potential [15:02] >> UK, I think regulators are actually [15:04] trying to just kick these companies [15:05] while they're down. [15:06] >> Is that what's happening? Or or is it a [15:08] situation where Getty I mean, that that [15:10] is possible. I mean, it could just be [15:12] like a just a complete mistake. I mean, [15:14] as you go back to Meta, and you think [15:15] about the US regulators blocked Meta [15:18] from buying a VR fitness company that [15:22] they said would create a monopoly in the [15:23] VR fitness industry. And of course, like [15:25] the VR fitness industry like never [15:27] happened. Like it's not a it's not a not [15:29] an industry worth monopolizing. And so, [15:31] if you think about you know, the [15:33] business model of Quest, like would it [15:35] have been advantageous? Maybe, but it's [15:38] worth letting Meta take a shot at that [15:41] and see if having a [15:43] VR fitness company in the portfolio will [15:45] actually get that product to you know, [15:48] to scale. Tyler, what do you have [15:50] >> Uh another interesting anecdote is the [15:52] UK blocked Meta from buying Giphy in [15:54] 2022. [15:55] >> Yes. [15:55] >> So, [15:56] to prevent a monopoly in GIFs, I [15:58] presume. [15:59] This could be pretty dangerous. [16:00] >> Yes, which is also odd. What what does [16:03] Giphy power the GIFs that you see in [16:06] iMessage? Do you know what I'm talking [16:07] about? When you go to [16:10] you you go to there's Genmoji, there's [16:13] stickers, and then there's also hashtag [16:16] images. And you can find images [16:19] and you can send you know, different [16:21] pictures and and GIFs. I wonder if these [16:23] are from Giphy or [16:24] >> It's powered by Bing, not Giphy. [16:26] >> Okay, so there's competition in this [16:28] market. [16:28] >> Yeah. [16:29] >> I don't know. It's it it seems [16:31] >> Amanda says Canva doesn't label AI in [16:33] their stock images. That's so annoying. [16:34] [clears throat] [16:35] So, yeah, I didn't even know Canva was a [16:37] player here. [16:37] >> Mhm. [16:38] >> But it would make sense to like [16:40] >> you're in Canva and you want to you [16:42] know, have a picture of a forest on your [16:44] greeting card and you can pick from a [16:47] library. Some of them will be generated, [16:48] some of them won't and they probably [16:49] also have a a generative AI tool where [16:51] you can generate a new image, but if [16:54] you're just got looking for a generic [16:55] stock image that's already been [16:56] described, you can probably pull from [16:58] whatever library is out there. Anyway, [16:59] on the cover of the Wall Street Journal, [17:01] it's the world's creepiest office and [17:03] people won't stay away. This is the the [17:06] New York complex vacated by IBM has [17:08] become a magnet for urban explorers. We [17:11] have a video of this place. If you're a [17:13] scrappy startup looking to expand, this [17:16] might be the key to success, the [17:19] ultimate lock-in factory in sort of [17:22] rural New York, I think. Well, we'll get [17:25] into it. So, this is in Somers, New [17:28] York. I don't know exactly how far away [17:29] that is from Manhattan. [17:30] >> Yeah, so if you were like a neo-lab and [17:33] your name was like ominous intelligence. [17:35] >> Yeah, there we go. [17:37] >> be a great sort of HQ, potentially, you [17:39] know, your first uh little micro data [17:42] center. Very James Bond villain sort of [17:46] vibe. [17:48] Uh and yeah, it looks looks like you [17:50] could make it, you know, really, really [17:51] homey. [17:52] >> It is 1 hour. You can be in the city in [17:54] an hour. That's [17:55] >> And imagine walking these halls [17:57] late at night after a long day at work. [18:00] >> Yes. [18:01] >> It wouldn't be It wouldn't be creepy at [18:03] all. [18:03] >> Yes. Robert Carlton was getting the mail [18:06] one day in April when he saw a group of [18:08] teenage boys sprinting away from nearby [18:11] woods. He cut them off at the road and [18:13] raised his arms. I just said, "Stop. [18:15] It's over." The 62-year-old retired [18:18] engineer recalled. Soon, he said, "Three [18:21] pur- uh pursuing New York State Police [18:23] troopers emerged from the tree line and [18:25] arrested seven juveniles." The criminal [18:28] charge, trespassing on the former IBM [18:31] campus here. Disrespecting International [18:34] Business Machines. Don't do it, folks. [18:36] The long vacant site has become a magnet [18:38] for so-called urban explorers who prowl [18:40] abandoned malls, hospitals, power [18:43] plants, amusement parks, factories, and [18:45] any other unused uh disused structure [18:48] they can breach. [18:49] >> of the graffiti. [18:50] >> Do you think putting a data center here? [18:52] We got to figure out how much power [18:53] >> write all over [18:56] >> [clears throat] [18:56] >> the former [18:57] office of International Business [18:58] >> Yeah, exactly. [18:59] >> What was the What was the IBM AI? [19:02] Watson? Watson, the machine [19:04] intelligence. [19:05] >> Was probably first [19:06] >> Still at the frontier. [19:08] >> It still at the frontier. It was super [19:10] intelligent at [19:11] at Jeopardy. The global Urbex, which is [19:15] short for urban exploration phenomenon, [19:18] Urbex isn't new, but it's been turbo [19:20] charged by artsy videos on Instagram and [19:23] TikTok that spur others to create their [19:25] own posts luring more curiosity seekers. [19:29] Police in Livingston, New Jersey [19:30] recently warned people to stay out of [19:32] the closed Livingston Mall property. [19:34] Jacksonville Beach, Florida police [19:36] issued a similar caution in April about [19:38] the shuttered Adventure Landing [19:39] amusement park. Going into an abandoned [19:42] amusement park, that is thrill seeking [19:43] at its best. We are aware of a TikTok [19:46] challenge to explore to explore the [19:48] property. Who's doing these TikTok [19:50] challenges and should they should they [19:52] be allowed to post those? This is this [19:54] is dangerous for some people. GPT Live, [19:57] a new generation of voice models for [19:59] natural human AI interaction, launched [20:02] today from OpenAI rolling out to ChatGPT [20:04] starting today. Taking shots at that guy [20:07] on Instagram who's unsatisfied with [20:10] legacy voice models. It's a lot smarter. [20:13] >> a feeling he's going to find a way to [20:15] keep using the old models. [20:17] >> Oh, maybe. Or he'll just become Who's [20:19] the Pliny the Liberator? He's sort of [20:22] the Pliny the Liberator of voice models [20:24] in that I know he's taking shots and [20:25] he's like very unhappy with the quality [20:29] of the voice models, but at the same [20:30] time he does a great job of actually [20:33] exposing and finding limitations, not [20:35] necessarily jailbreaks, but he he By [20:37] watching those videos, you can very [20:39] clearly understand, "Okay, I understand [20:41] what's going on here. It's not It [20:43] doesn't have tool use, so it can't [20:44] count. It can't set a timer. It doesn't [20:47] have you know, a long running context [20:49] window that can stay with you for a long [20:51] time, so it gets it gets lost. It can't [20:53] talk to each other." There's a whole [20:54] bunch of different ways. And once you [20:55] have all of those documented by just a [20:58] He's basically like a white hat hacker. [21:00] He's he's he's [21:01] >> a bunch of benchmarks. [21:02] >> Oh, wait, what? [21:03] >> He found a bunch of like good [21:04] benchmarks. [21:04] >> Yeah, yeah, yeah. Basically benchmarks. [21:05] Like if you can satisfy his his videos [21:08] and run those experiments again in a [21:10] satisfying way, you probably have a [21:12] better product at the end of the day. [21:14] So, I don't know. I think it's I think [21:16] it's cool. I'm excited. [21:17] >> Very Yeah, super excited for this one. I [21:20] I wonder if the Yeah, the real-time [21:22] nature unlocks like like almost like a [21:25] new kind of like pod like real-time like [21:28] podcast format where you can just [21:29] actually have a conversation about a [21:31] topic, right? [21:32] >> Yeah, we should have it sitting right [21:33] here. It has a full duplex architecture, [21:36] which means that it can listen and speak [21:39] at the same time. So, it can jump in [21:41] appropriately. And it can also route [21:43] like certain questions to smarter models [21:46] to run in the background. You can still [21:47] have a conversation. It's very cool. [21:49] Which is great because my default flow, [21:52] like if I'm driving, which I don't think [21:54] I should be doing, but I will tap the [21:57] record button, not the voice mode, and I [22:00] will dictate a really long prompt. And [22:04] then I will click send, and then I will [22:06] wait for it to cook, and then I will [22:07] have it read that back to me. And that's [22:09] all just like a couple extra clicks, a [22:11] couple extra steps to get what I really [22:13] want, which is [22:15] a conversation. I was interested about [22:16] the idea of [22:18] where this goes. Like is there a world [22:20] where you are talking to a model and you [22:23] can have sort of multiple threads of [22:25] conversations going within one [22:28] conversation. So, you can say, "Hey, I [22:30] need you to go and do a deep research [22:32] report on [22:35] you know, IBM Watson." And it goes and [22:37] cooks on that. And then while it's doing [22:39] that, you're still talking to it, and [22:40] you're still having a conversation. But [22:42] then you also might say, "Well, catch me [22:44] up on the latest news." And it has that [22:46] at the fingertips. So, it can talk to [22:48] you about the latest news. And then you [22:50] can go and it can say, "Hey, by the way, [22:52] I'm done with the report." Or "I'm done [22:54] with the the code that I wrote, but [22:56] you're talking to me about other [22:58] problems in your organization, other [22:59] decisions, other research projects that [23:01] you have." And it's all done in one [23:03] continuous conversation like you know, [23:06] me interacting with you. Like, you know, [23:07] we might we you might delegate something [23:10] and then continue the conversation and [23:12] then move on. Oh, I got to text this [23:13] person. Okay, I'll text them. They'll do [23:14] that. They'll come back to me in 25 20 [23:17] minutes with a revision to the the the [23:19] thing that we're working on. But, in the [23:20] meantime, let's talk about whatever else [23:22] we got going on today. So, interesting [23:24] to see where this goes. In other OpenAI [23:26] news, GPT-5.6 Soul, along with Terra and [23:29] Luna, will will launch publicly this [23:32] Thursday. Preview access is expanding [23:35] globally. Lots of people having fun with [23:37] this. Some people got early access. [23:38] Other people were joking about clearly [23:41] not getting early access, but wanting [23:43] early access. But, a lot of people are [23:45] having fun. [23:46] >> Yeah, what did what did [23:48] what did John Palmer [clears throat] [23:48] say? [23:49] >> What did he say? [23:50] >> And he says, [23:51] "It seems I'm allowed to talk about 5.6 [23:53] now. I wasn't one of the testers, but I [23:55] heard it's really good. Again, I haven't [23:57] used it yet, but this will probably be [23:58] my default model from here on when I get [24:01] access." [24:02] >> It's a good one. [24:02] >> Uh people are having a lot of fun. [24:04] Uh Kit Langton says, "Uh I can finally [24:07] talk about GPT-5.6. I've been using it [24:09] for many months, if not decades. And [24:11] what is glaringly obvious to me now as I [24:13] look back on my life is that you did not [24:15] have access to it." [24:17] Uh it's very fun. Uh people are very [24:19] happy with it. Interesting, there's a [24:21] there's a little bit of a fork around [24:23] like when how the the vibe around 5.6 is [24:27] different than Fable 5, when how [24:30] interactive it is versus uh somebody [24:33] said like uh Fable is like [24:35] light speed across the galaxy and uh and [24:39] Soul is like a Porsche on the [24:41] Nürburgring, more engaging. I don't [24:42] know. Uh it seems like it we're we're [24:44] entering uh some some fragmentation in [24:47] like the right tool for the job. [24:49] >> Yeah, Ethan says, "My big takeaway is [24:50] that both Soul Ethan had access." So, he [24:52] says, "Both Soul and Fable represents [24:54] represent jumps over previous models and [24:57] have opened a large gap with the next [24:59] best AIs. People will have preferences [25:01] for one of the other, but if you're [25:03] doing any work where better intelligence [25:04] matter, these two those two models are [25:07] your only choices. Dean says, I think [25:09] for me, the main takeaway with Soul and [25:11] Fable is that I can't remember a time [25:13] when the leading models were A, so [25:15] decidedly ahead of everything else, and [25:16] B, so distinct from one another. So, [25:19] yeah, a lot of people are talking about [25:21] >> [clears throat] [25:21] >> how they're using the models together. [25:24] >> It's probably a function of [25:25] just tighter tighter distillation [25:29] protocols. Like there's less pollution. [25:31] Like also, like not everyone's training [25:33] on the exact same data anymore because [25:36] it used to just be everyone just trains [25:38] on the entire web corpus, and now each [25:41] company has different data sets, [25:43] different data brokers, different usage [25:45] patterns that they can train on, and [25:47] then you different sort of research [25:48] tastes. So, go give it a try [25:51] >> Leo says, I forgot to mention this. GPT [25:53] 5.6 Souls world leading in computer use [25:56] have made me use it 100x more. When we [25:58] lost access to 5.6, I quickly started to [26:00] go insane without it. So, yeah, people [26:01] in the early access program obviously [26:03] lost access for a couple weeks [26:05] >> Sure. [26:06] >> due to the DC action. [26:10] But yeah, I saw some people saying like, [26:11] what are some good uses for computer [26:13] use? And it's like, anything on your [26:15] computer. [26:16] What do you use your computer for? [26:18] They're not all going to be great uses, [26:19] but [26:20] I'm curious. [26:22] >> I still think it's worth It's worth it. [26:23] It's always worth like popularizing [26:25] those like [26:27] genuinely useful [26:29] patterns. I mean, like deep research was [26:32] that in many ways. Just a very like it [26:35] gave you a 20-page PDF report on the [26:39] topic that you were asking for, and it [26:40] did a lot of the work for you. And I [26:42] still think the the come up with a good [26:45] prompt idea, find the right reason to [26:49] use AI is a problem for a ton of people. [26:52] There there there are there are plenty [26:54] of people that [26:55] adopt the latest AI tools, immediately [26:57] find what the model's good at, how to [26:59] use it, how to use it profitably, [27:02] effectively, not waste tokens, not waste [27:05] their time. [27:06] Um but other people need ideas. They [27:08] need to know that there's a new [27:10] capability. You see this with the with [27:11] the chat app you see this with the voice [27:13] mode where people are like judging a [27:15] model based on hallucinations that [27:17] disappeared a year ago or judging a [27:18] model based on some oh it has a cutoff [27:21] window and then that hasn't been a thing [27:23] for a year but it's still burned its [27:26] burned its way into like the the memes [27:28] of the online culture and thus [27:31] uh people need to be updated on that. [27:33] And so [27:34] uh I'm I'm in favor of coming up with [27:37] good computer use uh demos. The real one [27:40] is just can it play Counter-Strike [27:43] effectively? That's the real computer [27:45] use test. Can it rank me up to global [27:48] elite? [27:49] Can it down rank me so I can smurf a [27:51] little bit? This is this is the future. [27:53] This is what I want. And we'll see you [27:54] >> Ooh, a wing flashback. [27:55] >> tomorrow. Goodbye. [27:57] >> Cheers. Love you.