GameStop Bids $55B for eBay as AI Shift Threatens SaaS Models

GameStop出资550亿美元竞购eBay，AI变革威胁SaaS模式 GameStop首席执行官Ryan Cohen提议以550亿美元收购eBay，而新兴的神经计算机范式正威胁着让传统软件应用走向淘汰。 主持人John Coogan和Jordi Hays探讨了向“神经计算机”以及Andrej Karpathy提出的“软件3.0”范式的转变；在这一背景下，前沿AI模型能够动态渲染界面，并让传统SaaS应用逐渐过时。他们分析了GameStop出资550亿美元收购eBay这一备受争议的要约，并指出了Ryan Cohen在接受CNBC采访时暴露出的重大资金缺口。其他探讨的话题还包括“围墙花园互联网”与本地执行智能体、网络上引发热议的售价850美元的汽油动力笔记本电脑、白宫拟议的AI监管政策，以及一款可追踪空场电影放映场次的应用。 Andrej Karpathy提出了“神经计算机”范式，即由AI推理引擎按需动态渲染动态用户界面与输出，而不再依赖传统安装的SaaS应用程序。 (1:14) John演示了前沿AI模型如何仅凭单一提示词就在数秒内执行完整的多步骤财务研究流程和横向对比分析，从而省去了中间的编写代码和电子表格处理工作。 (3:06) 向“软件3.0”的转变意味着神经网络的原始输入和输出正在取代复杂的软件技术栈，导致传统的套壳应用走向淘汰。 (6:41) 非技术背景的创始人面临着新的战略挑战：需要判断一款产品究竟需要成为独立的应用，还是其功能通过单条聊天提示词即可一次性完成。 (8:56) SaaS企业若专注于包装、用户体验、分发护城河、专有数据和网络效应，仍能借助原生AI能力实现蓬勃发展。 (8:56)

Hosts Jordi Hays and John Coogan examine how frontier AI models are evolving into direct execution engines that threaten traditional software models, while analyzing the financial discrepancies in GameStop’s $55 billion bid for eBay. The Neural Computer and Software 3.0 (1:14) Drawing on Andrej Karpathy's remarks at Sequoia's AI Ascent event, the hosts outline an emerging paradigm shift from manual coding toward "neural computers." Rather than relying on installed software and traditional user interfaces, neural computing leverages inference engines to dynamically generate bespoke workflows, interfaces, and outputs on demand directly from raw audio, video, or image inputs. Karpathy illustrated this concept through his project "Menu Gen," an application built on Vercel to translate and generate pictures of restaurant menu items using OCR. Karpathy realized the entire application was obsolete when a single prompt in Gemini could annotate food photos directly onto raw menu images. Under this "Software 3.0" model, heavy deployment frameworks and multi-step coding pipelines collapse into end-to-end multi-modal inference. Value Capture and Defensibility in AI (12:47) John Coogan draws an analogy between current AI developments and Union Square Ventures' 2016 "Fat Protocols" thesis, which predicted that blockchain value would accumulate at the shared protocol layer rather than in application layers. In the AI market, frontier models are functioning as "fat models," progressively absorbing features and logic that previously required independent software products. To demonstrate this shift, Coogan notes how a single AI prompt replaced an entire multi-tool financial research workflow—automatically pulling metrics, calculating valuation multiples, and outputting comparative infographics for eBay and GameStop. Because generalist models can easily "one-shot" standard workflows, the hosts argue that defensible software ventures must now rely on specialized distribution, network effects, liquidity, or proprietary data rather than basic software packaging. Navigating Walled Gardens With Local Agents (15:31) The hosts propose a "walled garden internet theory," arguing that corporate platforms like Substack, YouTube, Facebook, and enterprise vendors like SAP intentionally restrict cross-platform data access. While frontier models are technically capable of indexing open information, corporate policies and legal hurdles create deliberate silos that limit direct API interaction. To bypass these restrictions, users are increasingly turning to local agents like OpenClaw that operate directly on physical machines via screen capture and automated mouse controls. The discussion notes observations from figures like Stripe's John Collison, pointing out that until digital barriers yield, humans often serve as mechanical intermediaries—manually exporting PDFs and feeding data between disconnected platforms on behalf of local AI agents. GameStop’s Unresolved $55 Billion eBay Bid (17:46) The broadcast reviews GameStop CEO Ryan Cohen’s proposal to acquire eBay for an estimated $55 billion to $56 billion, offering $125 per share—a 46% premium over eBay's unaffected February 4, 2026 closing price. GameStop revealed a 5% economic stake in eBay built via common stock and derivatives, structuring the tender offer as 50% cash and 50% GameStop common stock. However, the hosts highlight significant capital questions raised during Cohen’s CNBC interview. While GameStop holds roughly $9 billion in cash on an $11 billion market capitalization, and secured a $20 billion financing letter from TD Bank, the committed capital totals approximately $40 billion. This leaves a roughly $16 billion funding gap that Cohen declined to explain on air, creating widespread market skepticism around the mechanics of the proposed transaction. As foundational AI models absorb conventional software layers and local agents bypass corporate data walls, technical execution is rapidly consolidating, even as capital-intensive corporate dealmaking remains bound by balance sheet constraints and regulatory scrutiny.

TBPN 主持人 Jordi Hays 与 John Coogan 探讨了前沿人工智能向“神经计算机”与“软件 3.0”演进如何颠覆传统 SaaS 及应用层，并深入剖析了 GameStop 首席执行官 Ryan Cohen 提出的 550 亿美元收购 eBay 提案中的财务可行性疑点。 神经计算机与软件 3.0 (1:14) 基于 Andrej Karpathy 在红杉资本 AI Ascent 活动上的讨论，技术工作流正从初级的“氛围编码”（vibe coding）迈向以大语言模型或推理引擎为核心的“神经计算机”范式。该范式不再依赖预先安装的固定软件或传统用户界面，而是根据输入的音视频等原始数据流，通过扩散模型等技术按需即时生成定制化界面与输出。Karpathy 以自身开发的“Menu Gen”应用为例指出，原本通过 OCR 识别菜单文本再调用图像生成器的多层应用架构，已被直接向 Gemini 提示并在像素层面覆盖菜品图像的单次交互完全取代，这表明大量中间层软件可能彻底过时。 胖模型假说与应用护城河 (12:47) John 将当前 AI 价值捕获模式与 Union Square Ventures 在 2016 年提出的区块链“胖协议”理论进行了类比。在 Web 2.0 时代，HTTP 等协议层捕获价值极低，主要利润被 Facebook 等应用层攫取；而当前的“胖模型”则在持续吸收多步骤执行、自动化脚本等功能，导致价值逐步向核心模型层集中，压缩轻量级套壳工具的生存空间。主持人指出，虽然许多初创点子可被前沿大模型单次提示直接解决，但应用若具备分发渠道优势、网络效应、私有数据源或复杂的专业工作流，仍能通过体验封装建立防御壁垒。 围墙花园与本地 AI 代理 (15:31) 针对互联网数据隔离，节目提出了“围墙花园互联网理论”。高质量内容分散于 Substack、Facebook、X、YouTube 等平台，其不互通属于商业与法律层面的壁垒而非技术障碍。主持人指出，AI 模型将不可避免地绕过这些限制，主要受限于推理成本而非绝对访问权限。例如，用户可通过 OpenClaw 等工具在本地运行脚本、捕获屏幕并模拟鼠标操作，以此绕过类似 SAP 等平台对自动化代理的限制；这也契合了 Stripe 的 John Collison 所称的人类正演变为代 AI 执行机械任务以突破平台封锁的“物理载体”。 GameStop 收购 eBay 案疑云 (17:46) 根据《华尔街日报》报道，GameStop 首席执行官 Ryan Cohen 拟以约 550 亿至 560 亿美元全盘收购 eBay，报价为每股 125 美元（较 2026 年 2 月 4 日基准溢价 46%），结构为 50% 现金加 50% GameStop 普通股。John 通过大模型生成的两家公司财务对比显示，eBay 年营收约 500 亿美元且增长 8%，营业利润达 22.8 亿美元；而 GameStop 营收仅约 150 亿美元且下滑 5%，营业利润仅 2.32 亿美元。此外，CNBC 主播 Andrew Ross Sorkin 指出，GameStop 市值约 110 亿美元、账面现金 90 亿美元，即便加上道明银行（TD）提供的 200 亿美元高确定性融资信，总资金约 400 亿美元，仍存在约 160 亿美元缺口，而 Cohen 在专访中并未给出具体的资金补足机制。 无论是基础大模型持续吞噬独立应用功能的软件范式转移，还是 GameStop 杠杆收购成熟电商平台面临的巨额资金缺口，当前技术与资本运作均展现出高估值预期与底层现实执行力之间的显著张力。

## Background 该视频由科技播客频道 TBPN 于 2026 年 5 月 22 日发布在 YouTube 平台。视频是其核心科技直播脱口秀的 30 分钟精炼版（Diet TBPN），由 John Coogan 和 Jordi Hays 主持，内容涵盖前沿科技突破、重磅商业并购动态及硅谷热点趣闻。 ## TL;DR 节目首先从 Andrej Karpathy 的演讲切入，探讨多模态端到端直出 UI 的“神经计算机”与 Software 3.0 范式，分析底层大模型变“胖”并吞噬传统独立 App 的趋势。随后，主持人深度拆解了 GameStop 首席执行官 Ryan Cohen 拟以 550 亿美元收购 eBay 的惊人提案，并尖锐指出其在 CNBC 专访中暴露的 160 亿美元巨额资金漏洞。最后，节目还评述了白宫拟设立的 AI 前置审查机制与特朗普的“婴儿论”，并趣谈了民间手工打造的“汽油动力笔记本”以及能一键检索 AMC 零票房包场影厅的爆款小工具。 ## Atmosphere & style 节目氛围轻松自如、极客范十足且节奏明快，两位主持人宛如在茶水间与听众畅聊，开场便充满办公室打游戏的欢脱打趣。在切入商业与技术议题时，他们展现出敏锐而辛辣的行业洞察，尤其在复盘 GameStop 电视采访的尴尬名场面以及调侃特朗普的“婴儿论”时频频爆笑，讽刺效果拉满。整期内容兼具严肃的商业逻辑推演与荒诞的互联网极客整活，听感生动有趣、信息密度极高。 ## Bulletin - [01:14](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=74s) Andrej Karpathy 提出 Software 3.0 范式与“神经计算机”概念，未来终端或无需传统操作系统与独立 App，由端侧多模态模型接收指令直接端到端渲染交互界面。 - [18:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1111s) GameStop 向 eBay 抛出高达 550 亿至 560 亿美元的半现金半股票收购要约（溢价 46%），但因存在 160 亿美元未明资金缺口，在 CNBC 直播专访中遭遇严峻信任危机。 - [27:03](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1623s) 白宫正研讨建立前沿 AI 模型发布前的官方审查机制（类比 AI 版 FDA），特朗普则公开表态称应将 AI 视作初生婴儿加以呵护、警惕荒谬的政治过度监管。 - [29:01](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1741s) 独立开发者 Riley Walz 推出 Empty Screenings 工具，利用 AMC 影院约 10% 场次零出票的现状，帮助用户一键搜索身边的“私人包场”放映厅。 ## Key takeaways - [01:14](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=74s) 神经计算机（Neural Computer）雏形显现：未来终端或不再需要传统 OS 或本地 App，大模型可根据原始音频/视频输入瞬时生成当下的 UI 交互界面。 - [03:06](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=186s) 端到端生成正淘汰传统中间数据流：以往需要结合 Excel、Python 和 Markdown 排版的财报比对，现在单条 Prompt 即可直接输出专业信息图。 - [06:42](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=402s) Software 3.0 吞噬胶水代码：Karpathy 亲身经历证实，此前繁琐的 OCR+图像生成管线，如今可被前沿大模型原生一键合并在原始像素上完成。 - [09:25](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=565s) 氛围编程（Vibe Coding）的盲区：大量所谓的新 App 构想，实际上直接用 ChatGPT 或 Claude 对话即可单轮解决，无需投入精力部署独立应用。 - [12:47](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=767s) AI 时代的“胖协议”理论重现：不同于 Web 2.0 时代应用层捕获大部分价值，如今底层模型能力越发厚重，正全面虹吸外围应用层的商业价值。 - [15:30](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=930s) “围墙花园互联网理论”成立：各大平台互设壁垒封禁 AI 爬虫，迫使人类逐步沦为帮 AI 截屏、打印 PDF 或点击鼠标的“物理容器”工具人。 - [18:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1111s) GameStop 拟 550 亿美元蛇吞象收购 eBay：提出 125 美元/股（溢价 46%）的收购提议，且已暗中建立 eBay 约 5% 的经济利益仓位。 - [19:54](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1194s) CNBC 专访暴露 160 亿美元资金黑洞：GameStop 自身市值仅 110 亿、现金 90 亿，加上道明银行非约束性意向函 200 亿，仍无法覆盖总对价。 - [22:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1351s) 商业可信度遭遇滑铁卢：如买房只掏八成首付，面对深耕各垂类赛道且现金流极佳的 eBay，Ryan Cohen 的跨界收购大概率流产。 - [25:15](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1515s) 极客硬核整活“汽油动力笔记本”：民间打造搭载酷睿双核与 WinXP 的两冲程燃油笔电，以“离网计算”和“每加仑刷 300 条推文”成为社区笑料。 - [27:03](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1623s) 白宫酝酿 AI 发布前审核制度：考虑签署行政令成立联合工作组，在确保公共安全与防范过度扼杀创新之间寻求政策平衡。 - [29:01](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1741s) AMC 零票房场次成冷门痛点：Riley Walz 开发的小工具精准切入 10% 空座率影厅数据，迅速成为破百万浏览的爆款轻应用。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=0s) 录制前的游戏闲聊与本周行程 节目开场，Jordi 和 John 互相打趣，透露演播室在最后一刻还在手忙脚乱地准备，原因居然是有人把《使命召唤：现代战争2》（Modern Warfare II）装到了演播室的大屏幕上，导致实习生们打游戏打得有些拖延了开播进度。他们调侃另一位实习生 Ben 和 Tyler 之间那场漫长而激烈的对决，笑称 Ben 输得很惨，而且考虑到他正处在黄金玩家年龄段，这显得格外尴尬；不过主持人也打趣说，游戏当年在 Rust 地图流行的时候，Ben 可能才两岁大，确实得重新学习那些经典的老战术。 闲聊过后，主持人谈到本周的工作安排：周三和周四因为要去参加行业峰会，节目大概率会停更两天，但在周一、周二和周五都会照常带来精彩的直播。面对搭档是否去听“音乐会”的口误玩笑，他们重申是正经的科技会议，并迅速切入本期要探讨的核心科技动态。 ### [01:14](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=74s) 神经计算机的雏形与知识工作流演进 节目的第一个核心话题源于 Andrej Karpathy 近期在红杉资本 AI Ascent 峰会上的演讲。Karpathy 在访谈中回顾了自己的开发习惯是如何被“Vibe Coding（氛围编程）”重塑的，而 John 由此联想到自己的知识获取与处理工作流也正在发生翻天覆地的变化。尤其是现在的图像生成模型在制作信息图表（Infographics）和幻灯片设计上的表现突飞猛进，让人开始真正窥见所谓“神经计算机（Neural Computer）”的雏形。 关于神经计算机的构想，自 AI 爆发以来就屡屡被业内提及，马斯克此前构想的 Macrohard 软件项目也有类似愿景。其核心理念在于：未来的终端设备可能根本不需要安装任何传统意义上的操作系统或应用软件，机器内部只有大语言模型、端侧推理能力，或者直接连接云端算力；用户需要什么功能，AI 就在那一瞬间当场生成。正如 Karpathy 所描述的：未来的设备接收原始音频或视频输入，经由神经网络处理，直接通过扩散模型在屏幕上渲染出针对当下场景量身定制的交互界面（UI）。这种按需瞬时生成 UI 的范式，将对传统的 SaaS 软件产业带来颠覆性冲击。 ```mermaid flowchart TD accTitle: 传统工作流与神经计算机工作流对比 accDescr: 展现从多步骤传统检索到端到端神经计算机单提示词出图的架构转变 subgraph Traditional["传统/早期 AI 工作流"] A["多数据源检索 (Yahoo Finance / SEC)"] --> B["电子表格整理 (Excel / Sheets)"] B --> C["Markdown 排版 / 生成比对表格"] end subgraph NeuralComputer["神经计算机 (Software 3.0)"] D["单条 Prompt 指令"] --> E["多模态大模型 (端侧/云端推理)"] E --> F["直接渲染最终信息图 (端到端输出)"] end ``` ### [03:06](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=186s) 案例实操：GameStop 竞购 eBay 的一键可视化比对 为了验证这一概念，John 分享了自己分析商业新闻时的切身经历。近期 GameStop 的 Ryan Cohen 提出了一项引发轰动的提议，意图让 GameStop 收购规模远大于自身的 eBay。为了快速看懂两家公司的基本面差距，在过去，分析师需要打开 Google Finance、Yahoo Finance，甚至调用 Bloomberg 或 Capital IQ 把财报数字拷入 Excel 制作比对表格；即便在 ChatGPT 出现后，通常也是生成一段 Markdown 格式的数据表，阅读体验并不直观。 而现在，整个流程被简化为了一条简单的 Prompt。John 要求 AI 去调研 GameStop 和 eBay 的估值、增长率、营收、营业利润及估值倍数，并直接生成一张并排对比的信息图。生成结果一目了然： - **营收规模**：eBay 年营收约 500 亿美元，约为 GameStop（150 亿美元）的 3 倍。 - **营收增速**：eBay 保持 8% 的正增长，而 GameStop 同比萎缩了 5%。 - **营业利润**：eBay 达到 22.8 亿美元，是 GameStop（2.32 亿美元）的近 10 倍。 - **营业利润率**：eBay 超过 20%，GameStop 仅为 6.4%。 - **估值指标**：市销率（P/S）两者相近（GameStop 约 4 倍，eBay 约 4.5 倍）；但从市盈率（P/E）角度看，GameStop 的估值倍数高达 34 倍，显著高于 eBay 的 25 倍。 尽管这张生成的图片仍有瑕疵（例如把 GameStop 的正向营收数据误标成了代表亏损的红色），但它在几秒钟内便完美解决了用户的核心诉求——以一张可以直接转发给朋友的高清方图，清晰展现了双方的体量优劣。用户完全无需在意模型底层是调用了 CSV、Excel、Python Pandas 还是 Scikit-learn，一切中间软件和技术栈都被彻底抽象掉了。 ### [06:42](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=402s) Software 3.0 时代的到来：Karpathy 的 Menu Gen 反思 这种不再需要编写任何胶水代码或搭建软件系统的趋势，被 Karpathy 定义为“Software 3.0”。Karpathy 在 Sequoia AI Ascent 上分享了一个极其极端的亲身经历：他此前开发过一个名为“Menu Gen”的应用，初衷是因为去餐厅看文字菜单时经常有 30% 到 50% 的菜品让人摸不着头脑，于是他通过 Vibe Coding 搭建了一个完整的全栈应用——用户拍下菜单照片，应用部署在 Vercel 上，先运行 OCR 提取菜名，再调用图像生成 API 为每道菜画出样图，最后重新排版呈现给用户。 然而不久之后，Software 3.0 的解法直接让他“大为震撼（blew my mind）”：用户只需要拍下菜单照片丢给 Gemini，并附带一条指令让模型通过 Nano Banana 技术将菜品插图直接合成在原菜单上。模型直接输出了一张在原始菜单像素上无缝嵌入菜品真实渲染图的新照片。Karpathy 意识到，自己之前辛苦搭建的 Menu Gen 整个系统在一夜之间变成了“多余的存在（spurious）”。在 Software 3.0 范式下，输入是原始图像，输出也是图像，中间所有繁琐的应用程序逻辑完全被端到端的神经网络所吞噬。 ### [09:25](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=565s) 氛围编程热潮的反思：原生模型与独立 App 的边界之争 科技行业之外的人群如今也掀起了一股“氛围编程”（vibe coding）的浪潮，许多非技术背景的人频频向主播推销各种 App 构想。然而令人啼笑皆非的是，这些点子中几乎绝大多数都可以直接用 Claude Code、Codex 甚至 ChatGPT、Gemini 等聊天界面的单轮对话直接搞定，根本不需要单独开发一个独立的 App。人们正在借助氛围编程工具，费尽心机去写一些“本无必要存在”的应用。 这一现象折射出开发者思维方式上的脱节与盲区： - **上架部署的沉没成本**：像 a16z（Andreessen Horowitz）这样的风投机构虽然极力推崇熬夜折腾、把应用一键部署到 Vercel 上的极客精神，但这种熬通宵写代码、配置环境的繁琐环节未来注定会被淘汰。 - **高阶决策闭环（Higher-order Loop）的缺失**：面对一个需求，用户首先应该思考的是：**这真的需要专门写一个 App 吗？还是直接用当前前沿模型的原生功能一次性（one-shot）解决即可？** 比如随手拍一张食物照片让模型开始追踪卡路里，聊天 App 原生就能完成，根本不需要再去写一个健身记账软件。 - **底层能力的全面抽象**：在约 90% 的场景下，前沿模型（Frontier Models）在底层就能够实例化解决问题所需的一切构件，将具体的代码和工具实现完全抽象屏蔽。 不过，将模型原始能力包装成 App 依然有其商业逻辑。即便底层能力是现成的，只要能以极其简易友好的方式交付给用户，依然能够获得付费客户。此外，具备特定流动性池（liquidity pool）、独特壁垒或差异化功能的产品依然有生命力；很多时候，App 甚至只是一个强大的“营销获客载体”——用户并不知道聊天模型能做到这点，而擅长投放买量广告的公司就能借此获取海量下载。 --- ### [12:47](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=767s) “胖协议”理论重现：AI 价值链向底层大模型聚拢 关于模型能力对应用层的蚕食，令人联想到风险投资机构 Union Square Ventures（USV）在 2016 年发表的经典博文《胖协议》（Fat Protocols）。该理论探讨了 Web 互联网与区块链在价值捕获机制上的根本差异： ```mermaid flowchart TD accTitle: Fat Protocols Comparison accDescr: Comparison between the thin protocol layer of Web 2.0 and the fat protocol layer of Crypto/AI subgraph WebInternet["Web 1.0 / 2.0 架构"] direction TB AppWeb["胖应用层 (Facebook, Google 等)<br/>*捕获绝大部分商业价值*"] ProtoWeb["瘦协议层 (TCP/IP, HTTP, SMTP 等)<br/>*创造标准但未捕获经济价值*"] AppWeb --> ProtoWeb end subgraph CryptoAI["区块链 & 现代 AI 价值链"] direction TB AppCrypto["瘦应用层<br/>*轻量化前端、面临被吞噬风险*"] ProtoCrypto["胖协议 / 底层大模型<br/>*Bitcoin/Ethereum 或 GPT/Claude 捕获核心价值*"] AppCrypto --> ProtoCrypto end ``` 在互联网早期，维护 HTTP 或 TCP/IP 的开发者建立了驱动整个网络的通用标准，但经济价值几乎全被构建在其之上的应用（如 Facebook）拿走。而在区块链世界中，价值直接沉淀在底层协议和代币中（如 Bitcoin、Solana、Ethereum）。USV 当年撰文时，比特币市值仅为 100 亿美元（如今已达数千亿至万亿美元级别），以太坊市值仅 10 亿美元，但底层协议的市值远远压倒了其上的单体应用（尽管后来也出现了 Coinbase 这类数百亿美元市值的巨头）。 如今类似的动态正在 AI 价值链中上演：**底层模型每个月都在变得“更胖”（fatter）**。它们正不断侵蚀并吞噬外围应用的能力边界，用户直接从核心模型中获取的产出越来越多，协议/模型层对价值的虹吸效应极其显著。 --- ### [15:30](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=930s) “围墙花园互联网”与 AI Agent 的破壁博弈 针对“死网理论”（Dead Internet Theory），主播提出了一个更切合现状的概念——**“围墙花园互联网理论”（Walled Garden Internet Theory）**。互联网并没有死，Substack、传统主流媒体、Facebook、X 以及 YouTube 上依然充斥着极具价值的信息，但各大科技巨头之间筑起高墙、互相屏蔽，拒绝互通。 各大平台对 AI 的封锁并非技术不可行，而是一场法律与商业利益的拉锯： - **技术突破围墙的方式**：通过本地脚本、在 Mac Mini 上用无头浏览器进行自动化控制（puppeteering），或是直接读取本地 iMessage 数据，模型总能找到绕过限制的途径。 - **企业的防守与阻拦**：企业级软件巨头 SAP 近期公开表态严禁未经授权的 AI 代理（unauthorized agents）进入系统，试图加深护城河、放进“鳄鱼”吓跑 AI Agent。 - **难以防范的本地控制**：如果用户本地运行 SAP，直接对屏幕截图并让模型控制鼠标点击，平台在技术层面上根本无法抵挡。正如 Rune 在推特上的戏谑：**人类如今正在沦为 AI 的“物理容器”（vessels）**——AI 发号施令，人类充当其傀儡并照章执行。 - **“人类是拿高架货物的工具”**：引用 Stripe 联合创始人 John Collison 的比喻，人类在很多流程里只是帮 AI“拿下高处货架上的箱子”。例如由于系统限制模型无法直接读取某些网页 URL 时，人类只能老老实实扮演工具人，手动把网页导出或打印成 PDF，再上传投喂给 ChatGPT。 --- ### [17:46](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1066s) GameStop 拟巨资收购 eBay：Ryan Cohen 的千亿零售野心 节目随后切入重磅商业新闻：《华尔街日报》披露游戏零售商 GameStop 计划向电商先驱 eBay 发起收购要约。这是 GameStop 首席执行官瑞安·科恩（Ryan Cohen）将这家昔日散户抱团概念股打造成千亿美元市值巨无霸的核心计划之一。 围绕这起突发并购的关键内幕包括： - **暗中建仓与突袭**：在正式提出收购前，GameStop 已经悄然在二级市场买入并建立了 eBay 的股份仓位。 - **闪电出手**：原本市场预计要约会在本月晚些时候提交，但 GameStop 动作迅猛，在当天早晨就已经正式出价。 - **诉诸股东投票的后手**：如果 eBay 董事会态度消极或拒绝谈判，Ryan Cohen 准备越过管理层，直接向 eBay 股东发起敌意收购或公众要约。 - **公开信施压**：GameStop 已向 eBay 董事会主席保罗·普莱斯勒（Paul Pressler）发出正式信函；主播随即在节目中半开玩笑地透露，Pressler 碰巧还是自己的朋友。 ### [18:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1111s) GameStop 突袭收购 eBay：550 亿美元要约与 46% 溢价 节目主持人首先聊起了自己与 eBay 董事会主席保罗·普雷斯勒（Paul Pressler）的私交，调侃这位在商界极具声望的领袖如今被卷入了资本风暴的中心。游戏驿站（GameStop）首席执行官瑞安·科恩（Ryan Cohen）向前者递交了一封令人瞠目结舌的收购提议函：GameStop 正式提议以每股 125 美元的价格全资收购 eBay 的全部流通股。GameStop 披露，他们已经通过衍生品及普通股受益所有权积累了 eBay 约 5% 的经济权益，并紧接着向监管机构提交 Schedule 13D 报告以及 HSR 反垄断申报。 这笔总计约 550 亿至 560 亿美元的交易结构被设定为“50% 现金加 50% GameStop 普通股”。按照提议，125 美元的报价较 eBay 在 2026 年 2 月 4 日（即 GameStop 开始建仓之日）不受影响的收盘价，溢价幅度高达 46%。消息一出，整个金融圈为之震动，外界的普遍反应完全是难以置信——一家市值仅约 110 亿美元的实体零售游戏公司，究竟要如何吞下一家体量超过 500 亿美元的电商巨头？面对铺天盖地的质疑，瑞安·科恩在当天早晨直接登上了 CNBC 试图辩解。 - **报价细节**：每股 125 美元，相比 2026 年 2 月 4 日未受影响股价溢价 46%。 - **交易构成**：对半开方案（50% 现金 + 50% GameStop 股票）。 - **持股规模**：通过普通股和衍生品工具持有 eBay 5% 的经济利益。 - **监管申报**：提议发出次日向美国 SEC 提交 Schedule 13D，并递交 HSR 反垄断审查通知。 --- ### [19:54](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1194s) CNBC 专访“车祸现场”：索尔金追问 160 亿美元资金缺口 在 CNBC 旗舰节目《Squawk Box》中，主持人安德鲁·罗斯·索尔金（Andrew Ross Sorkin）与瑞安·科恩展开了一场极具戏剧性的针锋相对。索尔金开门见山地要求科恩厘清这笔天文数字般的交易在财务上到底如何实现闭环。科恩却显得格外含糊，一再避重就轻地重复：“详情都在我们的官网上，是半现金半股票。”当索尔金追问他能否为电视观众解释清楚时，科恩更是尴尬反问：“具体指哪一部分？” 索尔金随即为科恩算了一笔极为刺眼的账：GameStop 自身市值大概只有 110 亿美元，资产负债表上躺着 90 亿美元现金；即便 GameStop 把自身全部股票打包进去再加上所有现金，总计也才 200 亿美元。虽然科恩还拿到了一份来自道明银行（TD Bank）价值 200 亿美元的“高度置信函”（Highly Confident Letter），两者相加也不过 400 亿美元，与约 560 亿美元的总对价相比，仍存在至少 160 亿美元的巨大窟窿。更致命的是，索尔金指出银行的高度置信函并不等同于具有法律约束力的承诺融资。面对“剩下 160 亿美元到底从哪儿来”的核心质疑，科恩只能顾左右而言他，甚至以“我们有权增发股票促成交易”、“答案在官网上”来敷衍，场面一度陷入静止与滑稽。 ```mermaid flowchart TD accTitle: GameStop 对 eBay 收购要约资金缺口示意图 accDescr: 展示 GameStop 当前可用资金、银行融资意向以及收购要约总额之间的巨大资金漏洞 subgraph Capital["现有与意向资金（约 400 亿美元）"] A["现有现金储备<br/>$90 亿"] B["自身全部股权价值<br/>约 $110 亿"] C["道明银行高度置信函<br/>$200 亿（非锁定承诺）"] end subgraph Gap["未解资金漏洞"] D["不明来源的资金缺口<br/>约 $160 亿"] end subgraph Total["eBay 拟议收购总对价"] E["GameStop 报价总额<br/>约 $550 - $560 亿"] end A --> E B --> E C --> E D -.->|"无法自圆其说"| E ``` - **财务拆解**：自身市值约 110 亿美元 + 现金储备 90 亿美元 + 道明银行贷款意向 200 亿美元 = 400 亿美元。 - **资金黑洞**：距离 550 亿至 560 亿美元的总要约，仍凭空缺少约 160 亿美元。 - **融资性质**：道明银行提供的仅为 Highly Confident Letter（高度确信函），并无锁定承销效力。 - **科恩回应**：“我们有权继续发售股票来达成这笔交易。” --- ### [22:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1351s) “买房只掏八成首付”：要约面临信任危机与 eBay 的坚固护城河 节目的两位主持人对科恩在电视上的表现大跌眼镜，形容这一幕就像电影里的“胶卷定格、唱针刮盘”——科恩在心里仿佛自言自语：“大家一定很好奇，我是怎么试图拿 400 亿去买一家 550 亿公司的。”原本市场预期科恩此举是为了在周一制造话题、推高 GameStop 股价，但当日股价反而走跌。主持人们指出，科恩完全可以用更专业的投行话术来化解尴尬，比如坦承这只是第一轮框架协议、后续还会联合其他银行和私募股权财团组建银团，但他却选择了在全美直播中当场卡壳。 这种局面被形象地比作去买房：“你跑去跟房东说自己有 80% 的购房款，其中 50% 来自银行按揭、30% 是自己手头的现金，剩下两成还没着落，房东只会告诉你‘我需要全款’。”eBay 董事会主席保罗·普雷斯勒看完电视采访后，恐怕很难将这份提议视作严肃的商业要约。主持人们进一步探讨了 eBay 这家公司的强大韧性：多年来，eBay 在球鞋、奢侈手表、艺术品到二手车等各个垂类领域，都遭遇了垂直电商对手的猛烈围剿，但由资深老将杰米·扬诺内（Jamie Iannone）领导的管理层业绩执行力极强，公司今年股价表现强劲。科恩甚至没有机会在电视上阐述他能为 eBay 带来何种增值协同，仅仅因为算不清账，这桩收购案很可能还没正式开始就已宣告流产。 - **形象比喻**：手头筹备了八成资金就去提交全款买房要约，商业可信度极低。 - **现任管理层实力**：eBay CEO 杰米·扬诺内自 2020 年掌舵，早年曾于 2001 至 2009 年在 eBay 深耕多年，深谙业务底细。 - **垂直围攻中的耐力**：面对来自各个垂直赛道竞争对手的冲击，eBay 的综合电商业务依然保持了极佳的盈利与现金流。 --- ### [25:15](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1515s) 极客硬核整活：850 美元的“汽油动力笔记本电脑” 在紧张的商业要约之后，节目画风突变，插播了一条献给游戏玩家的奇葩“重大突破”。面对市面上昂贵的游戏笔记本、容易老化的电池以及水涨船高的电费账单，民间极客手工打造了一台独一无二的“汽油驱动笔记本电脑”（Gasoline-Powered Laptop），并在二手网络上以 850 美元的价格公开发售。 这台设备搭载了一台微型两冲程内燃机，号称“极易拉线启动”，加满一箱油即可支持约一个半小时的连续运行，非常适合作为当红概念的“离网计算”（Off-Grid Computing）终端。主持人们煞有介事地调侃起该机器的古董级硬件配置：它搭载了英特尔酷睿 2 双核处理器（Core 2 Duo）、2GB 内存（主持人开玩笑称在内存涨价的今天这是“升值资产”）、120GB 机械硬盘，并运行着经典的 Windows XP 系统。直播聊天室里的观众也纷纷加入狂欢，网友 Colin 留言称这台机器“一加仑油能刷 300 条 X 帖子”，还有人打趣说“要是失误把《GTA V》画质开到最高，电脑可能会直接冒烟爆炸”，展现了极具幽默感的黑客极客精神。 - **售价与动力**：标价 850 美元，配备两冲程汽油发动机，满油续航 1.5 小时。 - **核心配置**：Intel Core 2 Duo 处理器、2GB 内存、120GB 硬盘、预装 Windows XP 操作系统。 - **调侃梗点**：被戏称为离网野外求生专属设备，“每加仑汽油能发 300 条推文”。 --- ### [27:03](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1623s) 白宫拟设立“AI 审查机制”：特朗普的“婴儿论”与监管权衡 节目最后切入严肃的科技政策突发新闻：白宫正在积极研讨在 AI 前沿模型向公众开放之前设立严格的审查与审核机制。这一举措标志着政策风向的显著转变，此前政府大多对 AI 产业采取不干预的自由发展态度，如今这种前置审批机制被外界视作针对人工智能领域的“AI 版 FDA”（类比食品药品监督管理局）。据悉，政府可能通过签署行政令建立一个由科技界高管和政府官员共同组成的 AI 特别工作组，专门研究并确立未来的监管审核程序。 主持人特别提到了唐纳德·特朗普（Donald Trump）近期针对人工智能发表的一段引发热议的原话。特朗普公开表示：“我们要让这个产业彻底成为世界第一，因为现在的它就像刚诞生的漂亮婴儿。我们必须培育这个婴儿，让这个婴儿茁壮成长；我们绝不能扼杀它，绝不能用政治手段、愚蠢甚至荒谬的规则去阻碍它。”主持人一边被特朗普独特的比喻逗得捧腹大笑，一边直言“这根本不是婴儿，而是一个价值 10 万亿美元、牵动全球经济引擎的超级产业”。智库学者迪恩·鲍尔（Dean Ball）在报道中指出，AI 技术演进极快，目前缺乏正式规程，政策制定者正处于防止过度监管与确保公共安全之间的微妙平衡中。 - **政策动向**：白宫考虑发布行政令，召集科技高管与政府官员成立工作组，建立模型发布前审查框架。 - **特朗普经典言论**：将 AI 行业形容为“刚出生的美丽婴儿”，警告不可用“愚蠢的规则与政治干预”阻碍其繁荣。 - **专家观点**：智库研究员迪恩·鲍尔强调，必须在技术狂飙突进、保障模型可靠安全的同时，避免扼杀创新的过度监管。 ### [28:38](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1718s) 总结与推文趣评：“锤子学博士”的思维定势 在探讨完前一个议题关于负外部性与现实权衡的话题后，主持人转向了当天节目的最后一条精选推文。这条推文来自用户 Tommy，内容充满了互联网极客式的幽默：“本人是锤子学博士（PhD in hammerology）。好了，现在摆在我们眼前的正是一颗钉子。” 主持人听后哈哈大笑，赞同这就是最标准的思维模式——正如谚语所说，“当你的手里只有一把锤子时，看什么都像钉子”（When all you have is a hammer, everything looks like a nail）。 --- ### [29:01](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1741s) Riley Walz 爆款新项目：一键查找 AMC 零票房“私人包场”影厅 主持人接着向观众极力推荐独立开发者 Riley Walz 的最新力作。Riley 一直保持着每周发布新作品的极高产出节奏，而他刚刚公布的这个小工具在 X 上迅速引爆，斩获了超过 100 万次浏览和 2.7 万次点赞。该项目的核心洞察来自于一条惊人的影院数据：AMC 影院有整整 10% 的放映场次其实一张电影票都没有卖出去。 为了帮那些渴望独自享受静谧观影、体验无干扰“私人影院”效果的观众，Riley 开发了一个名为 Empty Screenings 的网站（访问地址为 `walzr.com/emptyscreenings`）。用户只要输入所在地区的邮编（ZIP code），网站就会自动筛选出当下完全零购票的空场放映信息，并明确提示最佳订票时机。主持人在节目中现场演示了周边查询：纽约当天中午 12:30 放映的《挽救计划》（Project Hail Mary）赫然在列，主持人打趣道：“要是你今天不用上班，完全可以现在就跑去独享整间放映厅。”此外，他们还在附近搜到了当晚 10:45 放映、同样零上座率的《穿普拉达的女魔头 2》（Devil Wears Prada 2），直呼“好好享受活着的感觉吧”。 除了完全无人的场次，该网站还会抓取仅售出 1 张或 2 张票的场次。主持人调侃称这绝对是“结交新朋友的一种奇葩方式”：你本以为自己抢到了空无一人的包场，结果走进去发现还有另一个怪人坐在那里，对方可能心里正想着“我倒要看看是哪个奇葩跟我一样会挑这种冷清场次”，然后走过来拉着你狂聊个不停。 在节目的最后，两位主持人呼吁听众在 Apple Podcasts 和 Spotify 上为节目留下五星好评，并前往 TBPN.com 订阅专属 Newsletter。伴随着两人默契稍欠、略微错位的温馨告别声与欢笑声，本期节目正式画上句号。 - **Riley Walz 项目数据**：新推文斩获 100 万次浏览量与 2.7 万次点赞； - **AMC 影院行业数据**：约 10% 的 AMC 电影场次处于零出票状态； - **查询工具**：`walzr.com/emptyscreenings` 支持邮编检索空场，现场检索案例包括纽约中午 12:30 的《挽救计划》与晚上 10:45 的《穿普拉达的女魔头 2》。 ## Notable quotes > [06:42](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=402s) 大为震撼（blew my mind） — Andrej Karpathy > [06:42](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=402s) 多余的存在（spurious） — Andrej Karpathy > [09:25](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=565s) 这真的需要专门写一个 App 吗？还是直接用当前前沿模型的原生功能一次性（one-shot）解决即可？ — John Coogan > [15:30](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=930s) 人类如今正在沦为 AI 的“物理容器”（vessels） — Rune > [15:30](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=930s) 拿下高处货架上的箱子 — John Collison > [19:54](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1194s) 我们有权继续发售股票来达成这笔交易。 — Ryan Cohen > [22:31](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1351s) 你跑去跟房东说自己有 80% 的购房款，其中 50% 来自银行按揭、30% 是自己手头的现金，剩下两成还没着落，房东只会告诉你‘我需要全款’。 — Jordi Hays > [25:15](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1515s) 一加仑油能刷 300 条 X 帖子 — Colin > [27:03](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1623s) 我们要让这个产业彻底成为世界第一，因为现在的它就像刚诞生的漂亮婴儿。我们必须培育这个婴儿，让这个婴儿茁壮成长；我们绝不能扼杀它，绝不能用政治手段、愚蠢甚至荒谬的规则去阻碍它。 — Donald Trump > [27:03](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1623s) 这根本不是婴儿，而是一个价值 10 万亿美元、牵动全球经济引擎的超级产业 — John Coogan > [28:38](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1718s) 本人是锤子学博士（PhD in hammerology）。好了，现在摆在我们眼前的正是一颗钉子。 — Tommy > [28:38](https://www.youtube.com/watch?v=6WdmNnuoiS8&t=1718s) 当你的手里只有一把锤子时，看什么都像钉子 — Jordi Hays ## Mentioned Jordi Hays, John Coogan, 使命召唤：现代战争2 (Modern Warfare II), Andrej Karpathy, 红杉资本 (Sequoia Capital), Vibe Coding, 神经计算机 (Neural Computer), Macrohard, SaaS, Software 3.0, GameStop, Ryan Cohen, eBay, Menu Gen, Gemini, Claude Code, Codex, ChatGPT, a16z (Andreessen Horowitz), Union Square Ventures (USV), Fat Protocols, SAP, John Collison, Paul Pressler, CNBC, Andrew Ross Sorkin, 道明银行 (TD Bank), Jamie Iannone, Windows XP, Intel Core 2 Duo, Donald Trump, Dean Ball, Riley Walz, AMC Theatres, Empty Screenings, 《挽救计划》 (Project Hail Mary), 《穿普拉达的女魔头 2》 (Devil Wears Prada 2) ## Video description Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive [Source](https://www.youtube.com/watch?v=6WdmNnuoiS8)

[6:42] Andrej Karpathy: 大为震撼（blew my mind） [6:42] Andrej Karpathy: 多余的存在（spurious） [9:25] John Coogan: 这真的需要专门写一个 App 吗？还是直接用当前前沿模型的原生功能一次性（one-shot）解决即可？ [15:30] Rune: 人类如今正在沦为 AI 的“物理容器”（vessels） [15:30] John Collison: 拿下高处货架上的箱子 [19:54] Ryan Cohen: 我们有权继续发售股票来达成这笔交易。 [22:31] Jordi Hays: 你跑去跟房东说自己有 80% 的购房款，其中 50% 来自银行按揭、30% 是自己手头的现金，剩下两成还没着落，房东只会告诉你‘我需要全款’。 [25:15] Colin: 一加仑油能刷 300 条 X 帖子 [27:03] Donald Trump: 我们要让这个产业彻底成为世界第一，因为现在的它就像刚诞生的漂亮婴儿。我们必须培育这个婴儿，让这个婴儿茁壮成长；我们绝不能扼杀它，绝不能用政治手段、愚蠢甚至荒谬的规则去阻碍它。 [27:03] John Coogan: 这根本不是婴儿，而是一个价值 10 万亿美元、牵动全球经济引擎的超级产业 [28:38] Tommy: 本人是锤子学博士（PhD in hammerology）。好了，现在摆在我们眼前的正是一颗钉子。 [28:38] Jordi Hays: 当你的手里只有一把锤子时，看什么都像钉子

[6:41] Andrej Karpathy: All of my Menu Gen is spurious. It's working in the old paradigm. That app shouldn't exist... The Software 3.0 paradigm is a lot more raw. The neural network is doing more and more of the work, and your prompt or context is just the image and the output is an image. [25:14] Donald Trump: We're going to make this industry absolutely the top because right now it's a beautiful baby that's born. We have to grow that baby and let that baby thrive. We can't stop it. We can't stop it with politics. We can't stop it with foolish rules and even stupid rules.

[0:00] Would you look at that? Would you look [0:01] at what, Jordy? Everything looks fine to [0:03] me. [0:03] >> We're back. We're back. [0:05] That's what I'm looking at. [0:06] Uh we're we're getting ready the last [0:08] second. We may have installed [0:11] uh Modern Warfare [0:12] II on this large screen and gotten a [0:16] little bit behind schedule with some of [0:18] the interns playing [0:19] >> Yeah, Ben Ben and Tyler were [0:22] It was a long drawn out fight. You won [0:24] by a lot, right? [0:25] >> Yeah. Well, we'll have to make up [0:27] throughout the show. [0:28] >> embarrassing for Ben considering that he [0:31] is still in that um not not not uh not [0:35] chief chief producer Ben, but other Ben. [0:37] Considering he's still in like the main [0:38] kind of like prime video game demo. [0:40] >> But at the same time, I mean, you were [0:42] probably 2 years old when Modern Warfare [0:44] II came out, so you got to sort of [0:46] relearn the the old the old tricks the [0:49] old tricks on Rust. [0:51] Anyway. [0:52] >> Well, big show [0:53] >> had a great weekend. [0:54] >> Hope you had a great weekend. [0:55] Uh big week. We'll be uh we are going to [0:58] a conference on Wednesday and Thursday, [0:59] so we might uh be off both those days, [1:02] but we have some great shows for you [1:04] planned Monday Tuesday Friday. [1:06] >> It's a conference. [1:07] >> Oh, I thought you said concert. [1:08] >> No. [1:09] >> [laughter] [1:09] >> Although I think there might be a [1:10] concert. [1:11] >> That would be extremely out of [1:13] character. [1:14] Well, uh I mean, there's a bunch of news [1:16] we're going to go through, but uh first [1:17] I was just uh sort of reflecting on [1:19] There was a great interview with uh [1:21] Andrej Karpathy at a Sequoia AI Ascent I [1:23] think last week. It went up on YouTube [1:25] and uh it was he was sort of reflecting [1:28] on how his workflow is changing around [1:31] vibe coding, and I was sort of [1:33] reflecting on how my knowledge workflows [1:35] are changing s- particularly around [1:38] image generation now that image [1:40] generation's really good at infographics [1:42] and effectively designing slides or [1:45] output. And so we're we're starting to [1:47] see the rumblings of this idea of like [1:49] the neural computer. There was this uh [1:52] people have been talking about this for [1:53] years since the AI boom began, but the [1:55] basic idea [snorts] is like you would [1:56] have a computer that basically has no [1:57] software whatsoever on it. It would just [1:59] have an LLM or just an AI model just [2:02] inference capability or potentially [2:04] [snorts] uh connection to cloud [2:06] inference uh that would generate [2:08] whatever you want whatever you need on [2:10] demand on the fly. And so I think Elon [2:12] talked about this with macro hard a [2:14] little bit. That was a piece of the [2:16] vision. This has always been uh [2:18] theorized, but it's becoming more and [2:20] more real. And so uh Karpathy describes [2:23] this idea of like a neural computer this [2:25] way. And I think it's an interesting [2:27] framing. Obviously it'll have [2:28] implications for uh SaaS products that [2:31] might be used in a headless under the [2:33] hood scene uh [2:36] way or uh might be competed with against [2:39] these neural computers, but Karpathy [2:40] describes this Imagine a device that [2:42] takes raw video or audio into basically [2:46] what's a neural net and uses diffusion [2:48] to render a UI that's unique for that [2:50] moment. And so this sort of like on the [2:52] fly uh instantiation of the exact UI [2:56] that you need for that particular [2:58] question or whatever problem you're [2:59] trying to solve whatever you're trying [3:00] to do is an interesting paradigm shift [3:02] that it feels like we're starting to see [3:04] glimpses of. So I most recently felt it [3:06] when I was trying to understand Ryan [3:08] Cohen's proposal for GameStop to take [3:10] over eBay. This is a big story. We'll go [3:12] through it today. Uh but I haven't [3:14] tracked each either either company [3:15] closely. We've had Ryan on the show uh [3:17] and uh we've talked about eBay and [3:20] GameStop intermittently, but I couldn't [3:22] tell you off the top of my head what's [3:23] the revenue for each company? What's the [3:25] profit like? What are what are the [3:26] different multiples? Uh and so in a pre [3:30] chat GPT world I would have gone to [3:32] Google Finance or Yahoo Finance and [3:35] pulled some data maybe had two tabs up [3:37] maybe used one of their comparison [3:38] tools. If I wanted to be really [3:39] advanced, I would have copy and pasted [3:41] the stats into [3:43] uh a spreadsheet. If you're really [3:45] working on Wall Street, you might have [3:46] like CapIQ or Bloomberg plugging into a [3:48] sheet an Excel sheet that then can uh [3:51] build you a comparison table and do like [3:53] comps. And then once we got into the [3:55] chat GPT world, you might do a deep [3:56] research report, pull all that data, put [3:58] it into a table, which is effectively [3:59] markdown. And sometimes the tables [4:02] renders a little weird and like you can [4:04] kind of bounce around, but now the whole [4:06] process from start to finish is just a [4:08] single prompt and it outputs an image. [4:10] So you can pull up the image that I [4:11] generated. So this was this was one [4:13] prompt. I said, "Do a bunch of research [4:15] on GameStop and eBay's valuation key [4:16] financial metrics things like growth [4:18] rate, top line, earnings, revenue, [4:19] valuation, how the multiples fit [4:21] together, build a nicely designed [4:23] side-by-side comparison of the two [4:24] companies." Uh and you wind up getting [4:27] something that is very digestible. Like [4:30] it just looking at this I mean it's [4:32] obviously a little zoomed out, but you [4:33] can zoom in and see okay, eBay has about [4:35] uh three times the revenue 50 billion [4:38] versus 15 for GameStop uh [4:41] and uh revenue growth eBay's growing [4:43] while GameStop's uh shrunk by 5%. eBay [4:46] grew 8%. Operating income eBay has 10 [4:49] times the operating income at 2. 2.28 [4:52] billion versus 220 232 million for [4:55] GameStop. And so you just get this like [4:57] very easy okay, what's the operating [4:59] margin? eBay's up a 20% GameStop's down [5:01] at 6.4%. And so you can start to see on [5:04] a price-to-sales ratio GameStop's at 4x [5:07] eBay's at 4 and 1/2x, but on [5:10] uh on a market cap to net income [5:12] GameStop's higher has a higher value 34x [5:15] versus net income versus eBay's 25x. So [5:18] you can just sort of see this table and [5:20] this is something that usually would [5:21] have been like three or four steps to [5:23] get here and instead it's just this [5:25] single prompt. And so uh I think that [5:27] there's uh like this is not a perfect [5:29] result. Like in in that image you can [5:31] see that like it chose red as the color [5:33] for all of GameStop's financials, which [5:35] is not what you'd normally do because [5:37] red is usually for negative numbers, but [5:39] those revenue figures are positive. Like [5:41] it could be better. I could probably go [5:43] further and prompt it a couple more [5:44] times to get exactly what I wanted, but [5:45] it solved my problem of having like [5:48] here's the summary of the question that [5:50] you were actually asking, which is like [5:52] how do these companies stack up to each [5:54] other? What's the relative size of the [5:56] business? What are the strengths and [5:57] weaknesses of each of them? And then [5:59] boom you have a square image that you [6:00] can easily text to someone and and it's [6:02] ultimately shareable. More importantly, [6:04] I don't care what it used under the [6:06] hood. It could have puppeteered a a [6:09] spreadsheet and put it all in [6:11] comma-separated values and make a CSV [6:13] and it could have transformed it with [6:15] Excel or Google Sheets under the hood. [6:16] It could have written Python. It could [6:17] have used Pandas or scikit-learn. It [6:19] could have done anything it wanted to, [6:21] but it's all abstracted to me and I [6:22] don't even think about it. And this is [6:24] different from the previous era of like [6:26] okay, well, if I wanted to do some sort [6:28] of stock comparison tool, I could vibe [6:29] code a stock comparison tool with API [6:31] integrations, make sure you have the the [6:33] the data connections, but it's just kind [6:35] of less necessary as the models get [6:37] fatter and they sort of eat more and [6:39] more of the process. And so Karpathy [6:42] describes this concept as software 3.0, [6:45] and we should pull up his example [6:47] because it's very similar. Of course it [6:49] happened like months ago cuz he's ahead [6:51] of me on everything obviously, but he [6:53] gave a good example of shifting from [6:54] like you have a problem that there's no [6:57] solution for, so you're going to vibe [6:58] code an app to just a few weeks or [7:01] months later like the AI tools can just [7:03] do it and you don't need any code. You [7:06] don't need any system to build. Even [7:07] though it's fun to build a system and [7:09] it's interesting and it allows for more [7:11] maybe more speed more reliability. Like [7:13] more and more things are like [7:14] one-shottable by the model. So let's [7:16] pull up Andrej Karpathy uh talking at [7:19] Sequoia AI Ascent about his experience [7:21] with software 3.0. I think one more [7:23] maybe uh example that comes to mind that [7:25] is even more extreme than that is when I [7:27] was building um menu gen. So menu gen is [7:30] this idea where you come to a [7:31] restaurant, they give you a menu, [7:33] there's no pictures usually, so I don't [7:34] know what any of these things are uh [7:36] usually. I like 30% of the things I have [7:38] no idea what they are, 50%. So I wanted [7:40] to take a photo of the restaurant menu [7:42] and to get pictures of what those things [7:44] might look like in a generic sense. [7:46] And so I built I vibe coded this app [7:48] that basically lets you upload a photo [7:50] and it does all this stuff and it runs [7:52] on Vercel and uh it basically re-renders [7:55] the menu and it gives you like all the [7:57] items and it gives you a picture that it [7:59] uses an image um you know, generator uh [8:01] for to basically OCR all the different [8:04] titles uh use the image generator to get [8:06] pictures of them and then shows it to [8:07] you. [8:08] And then I saw the software 3.0 version [8:10] of this, which is which blew my mind, [8:12] which is literally just take your photo, [8:14] give it to Gemini, and say, "Use Nano [8:17] Banana to overlay the the things onto [8:19] the menu." Uh [8:21] and Nano Banana basically returned an [8:22] image that is exactly the picture of the [8:24] menu that I took, but it actually put [8:26] into the pixels it rendered the [8:28] different things in the menu. [8:30] And this blew my mind because [8:32] actually all of my menu gen is spurious. [8:35] It's working in the old paradigm. That [8:36] app shouldn't exist. Uh and uh yeah, the [8:39] software 3.0 paradigm is a lot more kind [8:41] of raw. It just neural network is doing [8:43] more and more of the work and your [8:45] prompt or context is just the image and [8:47] the output is an image and there's no [8:49] need to have any of the app in between. [8:51] Um so I think that people have to kind [8:54] of like reframe [8:56] >> Um yeah, it it it it I mean, it it it's [9:00] it's real and [9:25] like the numbers are big, but it's not [9:28] at 20% of the US population. Like it's [9:31] just not at that level of adoption as [9:33] opposed to chat apps, which are at like [9:36] 70 80% penetration, right? Yeah, the [9:39] other the other thing that's been [9:40] interesting is uh non like people [9:43] outside of tech that have gotten into [9:44] vibe coding that have been pitching me [9:46] their ideas here and there. Almost every [9:48] time they're pitching me the idea, it is [9:50] something that Claude Code and Codex can [9:54] do themselves pretty well today. Like [9:57] just in one chat thread. [9:58] >> Or the app. Like, the apps can do that. [10:00] And that's what I'm like what's blowing [10:02] my mind now is that is that in many ways [10:04] >> that's what I'm saying. So, [10:06] like, they're using vibe coding tools to [10:08] vibe code something that doesn't [10:10] necessarily need to exist because you [10:12] could just use the app itself to do the [10:14] thing. [10:15] And they're already widely available. [10:17] So, it's it's been been interesting. [10:18] >> Yeah, so I think there's I think there's [10:20] two things. Like, one is that if you've [10:21] been like hesitant to jump into vibe [10:23] coding uh because [10:25] like it's just it's a little bit too [10:27] much of a hassle. Like, Andreessen [10:28] Horowitz is like obviously very very [10:30] comfortable being like, "Oh, yeah, let [10:31] me deploy to Vercel and do all this." [10:33] Like, you can figure all that out, but [10:34] that leads to this world where it's [10:36] like, "Oh, I was staying up all night. I [10:38] was I it was really really like burning [10:40] the midnight oil to get this app [10:41] deployed and like do all this stuff." A [10:42] lot of that's going to go away. And [10:43] like, you're not going to need to do [10:45] that. But then there's also the the this [10:47] question what you had, which is like, [10:49] there needs to be this higher order loop [10:51] of thinking around, okay, you have a [10:53] problem. Should you actually vibe code [10:56] an app? Or should you just try and one [10:58] shot it with the current model [11:00] capabilities? Because for a lot of [11:02] things You mean and within Within Yeah, [11:05] within ChatGPT, within Gemini, within [11:07] Claude, the actual apps. Like, you can [11:09] like take a picture of your food and [11:11] say, "Hey, start tracking my calories." [11:13] Like, there's a lot of things that the [11:15] that the apps can just do in one chat [11:16] thread that people are doing, but I [11:18] think that [11:20] there's like this tension between when [11:22] when you actually need to for sure go [11:24] and vibe code something versus when you [11:26] can just do it in like a one-shotted LLM [11:28] context. Frontier models are already [11:30] able to in in basically 90% of [11:32] situations I feel like instantiate [11:33] exactly whatever's required, to solve [11:36] the actual problem [11:38] under the hood, entirely abstracting [11:41] away like code and tools. Like, you will [11:42] just not be aware of what's happening [11:44] and it doesn't matter. And then the [11:46] second [11:47] >> yeah, I would I would add to to my [11:49] previous statement by by saying that [11:51] doesn't mean that there's not [11:52] necessarily a business there because [11:54] sometimes taking a raw capability and [11:56] presenting it to people in a way that's [11:58] very easy for them to digest, you can [11:59] still deliver value. And you can get [12:02] customers and people will pay you money. [12:04] But, it is it has been fascinating to [12:06] see like does this actually need to be [12:08] an app? Yeah, yeah, I mean there's [12:09] there's a ton of apps and software that [12:12] will still be valuable whether it has a [12:14] liquidity pool or some sort of unique [12:17] source of strength or some [12:19] differentiation point that that the [12:22] existing chat apps can't hack it all. [12:24] And then there's also just like like [12:26] marketing arms effectively where it's [12:28] like, "Okay, yes, the any any frontier [12:32] model in a chat app could do this, but [12:34] you weren't aware of it and this company [12:37] was really good at running ads to [12:39] actually get awareness going and then [12:42] drive downloads of the specific thing." [12:44] And so, we see those in the [12:46] in the App Store all the time. The other [12:47] thing that I was [12:49] reminded of was did you ever read Union [12:51] Square Ventures 2016 blog post fat [12:54] protocols? Are you familiar with this? [12:56] So, fat protocols [12:58] was this concept around how in in the [13:02] web like web I guess 1.0 2.0, there were [13:05] protocol layers which are like TCP/IP, [13:08] HTTP, SMTP, like file transfer [13:11] protocols, HTTP. And there and a lot for [13:13] like a couple of years the crypto [13:15] community was like the group that [13:17] developed and maintained HTTP, they [13:20] basically created the standard that the [13:22] web ran on. And yet very little value [13:25] accrued to the creators and the [13:27] maintainers of that protocol. And crypto [13:29] would be different because the Bitcoin [13:31] protocol would had the value capture [13:34] component like baked into it. And so, [13:36] there was this idea of like the [13:38] application layer in blockchain would [13:40] accrue very little value and the [13:42] protocol layer would capture the vast [13:44] majority of value. So, this is on the [13:46] web, the applications on top of HTTP. [13:48] You can think of like Facebook as a [13:51] beneficiary of the protocol of HTTP [13:53] because that's how what how how the [13:55] actual information, the photos, and the [13:57] text gets transferred to you. But, the [13:59] HTTP standard does not accrue the value. [14:01] The value accrues to the to the [14:03] application on top. And if you scroll [14:04] down, you'll see the blockchain example [14:07] which was sort of born out [14:10] that the application layer was pretty [14:11] thin on top and most of the value went [14:13] to like the tokens in the protocol [14:15] below. Yeah, exactly. Ethereum is a good [14:17] example. Solana is a good example. Of [14:19] course, there's there's value in the [14:20] application layer and there's some [14:21] companies that are being built, but this [14:23] was basically this thesis that the [14:25] he says, "We see that very early. We see [14:27] this very clearly in two dominant [14:29] blockchain networks, Bitcoin and [14:30] Ethereum. The Bitcoin network has a 10 [14:32] billion market cap." [14:34] Wow. I think it's like a trillion now, [14:36] right? Isn't it 700 billion? Um [14:39] yet the [music] largest companies built [14:40] on top are worth a few hundred million [14:42] at best. [14:44] Now we have Coinbase which is in the [14:46] tens of billions. So, both sides of the [14:48] protocol application layer did very [14:51] well, but the point is still true. [14:53] Similarly, Ethereum has a 1 billion [14:55] dollar market cap even before the [14:57] emergence of real breakout applications [14:59] on top and only a year after its public [15:01] release. And so, [15:03] that was sort of the core thesis on this [15:05] fat protocols thing. And and I think [15:07] that there's something similar happening [15:08] in the AI value chain. Of course, [15:09] there's like a bunch of other dynamics [15:11] going on in in the AI value chain and [15:14] there's a lot of capture and and [15:16] complicated market dynamics, but the [15:18] models feel like they're getting fatter [15:20] every month and they're sort of eating [15:21] away at the edges of what you can do [15:24] with them. And and so, increasingly you [15:26] can just get more and more out of the [15:27] core model, which is just interesting [15:29] dynamic. Third, there's still like this [15:31] huge question of like walled garden [15:33] jumping. We've talked about this before, [15:34] but it's almost that we need like a [15:36] different term for like the dead [15:37] internet theory. That's like the it's [15:39] like the walled garden internet theory. [15:40] Like, the internet's not dead. There's [15:42] great information in Substack, on [15:46] certain legacy webs legacy media [15:49] websites, and on Facebook, and on X, and [15:52] on YouTube. But like, all of those [15:54] companies don't want to interact with [15:55] each other. And so, that's where you get [15:57] something like, "Oh, well, like if you [15:59] write code, you do get access to it [16:02] loosely." Or if you're puppeteering a [16:04] browser on a Mac Mini, you get access to [16:06] that. Or if you're digging through [16:07] iMessage locally, like that can require [16:10] a different workflow. But that's more of [16:12] like a legal and business discussion [16:13] than a technical one. Like, there's no [16:15] reason technically that a single LLM [16:17] wouldn't be able to just query every [16:19] single web resource except for the fact [16:21] that the big the the various tech [16:23] companies don't want each other to talk [16:25] to each other. [16:26] and so, um the models I think will [16:29] continue to find a way under the walled [16:32] garden, over the walled garden, through [16:34] the walls. Like, they'll they'll seep [16:35] everywhere. Uh and it's more of a [16:37] question of just uh [16:40] inference cost, how how long it takes to [16:42] actually grind through the wall. But, uh [16:45] they're already figuring out a way [16:46] around. And then and open claw is a good [16:48] example of that. A lot of the walled [16:50] gardens were sort of brought down by [16:52] running it. [16:53] >> I think SAP came out and said, "No [16:55] unauthorized agents here." They're [16:57] trying to they're trying to put up the [16:59] walls. They're trying to build a moat. [17:00] They're trying to get some alligators to [17:02] scare off the the agents. [17:04] >> Yes, but I would be very surprised if [17:05] they're able to stop me from if I have [17:08] SAP and I'm running it locally for me to [17:11] take a screenshot of my computer and [17:13] then tell the mouse to go where it [17:15] wants. Like, it it it's very hard to [17:17] fight back against these. [17:18] >> Rune had a good tweet where he was like, [17:19] you know, people are now just like [17:21] vessels for the AI where they just like [17:22] the AI tells them what to do and then [17:23] they just act exactly [17:25] >> like Yeah, like the the model said. [17:26] Wasn't John Collison talking about this? [17:29] >> [laughter] [17:30] >> John Collison was saying like the humans [17:32] get the thing off the high shelf. Every [17:34] time I have to like go and like export a [17:36] PDF and upload it to ChatGPT because I I [17:39] can't get it in there by default. Even [17:41] though I could just give them a web URL, [17:42] I have to export it or print or [17:44] whatever. Let's talk about I think we [17:46] should talk about GameStop. [17:47] >> Let's do it. What's up with GameStop? [17:49] >> on with GameStop? Yesterday, it came out [17:52] through the Wall Street Journal, [17:53] GameStop was preparing to make an offer [17:54] for eBay as part of Ryan uh CEO Ryan [17:56] Cohen's plan to turn the retailer into a [17:58] 100 billion dollar plus juggernaut. [18:00] >> Mhm. GameStop has been quietly building [18:03] a stake in eBay shares ahead of a [18:04] potential offer and could submit an [18:05] offer as soon as later this month, which [18:07] they did [18:09] this morning. If eBay isn't receptive, [18:11] Cohen could decide to take the offer [18:12] directly to eBay's shareholders. Mhm. [18:15] And [18:16] uh they released a letter yesterday to [18:21] uh Paul Pressler, who's the chairman of [18:23] the board over at eBay, happens to be a [18:25] friend of mine. [18:27] Uh yeah. [laughter] [18:29] Uh yeah. [18:31] Uh I I I [18:33] um his daughter and my wife are good [18:35] friends. Um so, I end up we end up [18:38] hanging out a decent amount. Does he do [18:40] Does he do a lot of Does he do a lot of [18:43] like podcasts or press appearances? It [18:46] can it can we can it can be discussed. [18:48] >> I'm I'm just saying like all the [18:50] determinism of Pressler would be like [18:51] somebody who dominates the press. [18:53] >> The press circuit, the podcast circuit. [18:54] circuit all the time. [18:55] >> Um anyways, Paul Paul is fantastic. And [18:58] Ryan wrote this letter to him yesterday [19:01] saying, "GameStop is proposing to [19:02] acquire all common stock of eBay at 125 [19:05] dollars per share. We have accumulated a [19:07] 5% economic stake in eBay through [19:08] derivatives and beneficial ownership of [19:11] common stock and are filing a schedule [19:13] 13D and HSR notification tomorrow. Our [19:15] offer is 125 dollars per share [19:17] comprising 50% cash and 50% GameStop [19:20] common stock, which we will get to in [19:22] just a little bit because Ryan Cohen [19:25] discussed this on CNBC this morning. [19:27] That represents a 46% premium to eBay's [19:30] unaffected closing price on February [19:32] 4th, 2026, the day GameStop started [19:34] accumulating its position in eBay. Uh [19:37] and blah blah blah blah blah blah blah [19:39] blah blah blah blah blah blah. But, [19:40] let's [laughter] go straight to the CNBC [19:42] >> there was a question. So, Biz let said, [19:46] "Can someone please tell me how GameStop [19:48] has 56 billion dollars? It's not a 56 [19:52] billion company. There were questions [19:54] and Ryan Cohen went in the ring with [19:57] Aaron Sorkin over at CNBC or Andrew Ross [20:01] Sorkin. [20:01] >> Uh Andrew Ross Sorkin on CNBC Squawk Box [20:05] to interview about the GameStop eBay [20:07] acquisition. We can play this. through [20:09] how how you could get to that price and [20:11] how it would work. [20:14] It's on our website. It's half uh cash [20:18] half stock. [20:20] Uh but but the details are are on our [20:23] website. [20:26] Can Can you help I I've read them, but [20:28] can you help our audience understand [20:29] them? [20:32] Yeah, what what which part exactly? [20:36] Well, [20:37] >> [laughter] [20:38] >> I think we can start with the idea that [20:39] the market cap of GameStop is call it [20:43] $11 billion. [20:45] Uh you have $9 billion uh on your [20:47] balance sheet. [20:49] Arguably, if you're if you're providing [20:51] uh effectively all of your stock [20:54] and then and then the cash that gets you [20:56] to 20 [20:57] you have this letter from TD. That's [20:59] another 20. Uh we're now at 40, uh but [21:03] we're still off uh by call it uh 16. [21:08] And and the 20, as far as I understand, [21:11] while it's considered a highly confident [21:13] letter, meaning TD's saying they're [21:14] highly confident uh that they would [21:16] provide the financing, it's not locked [21:18] financing. [21:21] Yeah, we'll see what happens. [21:23] >> [laughter] [21:25] >> Um Founder mode. [21:29] Never doubt. I I I hear you. Duh duh [21:32] duh. [21:32] >> [laughter] [21:32] >> I understand that. I'm I'm just trying [21:34] to understand where the the rest of the [21:36] money would come from. [21:39] It's half cash, half stock. [21:43] I I I'm I hear you. I'm just saying that [21:46] that math doesn't get you to the [21:50] to the price that you're offering. [21:55] So [21:55] >> a pretty straightforward question. I [21:57] don't get it. Thank [laughter] you for [21:58] chiming in. Where's the rest of the [22:00] money coming from? [22:02] Andrew laid it out for you clearly. [22:05] I I don't understand your question. [22:06] We're offering half cash, half stock and [22:08] we have the ability to issue stock in [22:11] order to get the deal done. [22:13] But the full details of of the offer on [22:16] our are on our website. But you're on [22:18] our air. We thought we'd get [22:21] Uh [22:23] So but I don't understand your question. [22:26] Where's the money coming from? That's [22:27] the question. [22:31] You're wondering Record scratch. Freeze [22:32] frame. [22:33] >> [music] [22:33] >> You're wondering how I tried to buy a [22:35] $55 billion company with $40 billion, [22:37] he remarked. Do you think he was [22:39] expecting this to go out on Sunday, [22:42] Monday, GameStop stock to pop like crazy [22:46] and it's actually down today? [22:48] >> I guess that's possible. I I I don't [22:49] know. I I also don't understand why he [22:51] can't just say like, "Hey, we're in the [22:53] process. Like we got a highly likely [22:56] letter from a bank for 20. Yeah, we need [22:59] 16 more, but we're going to go get more [23:01] letters from other banks. We're going to [23:03] go get other equity investors. Like this [23:05] is a whole process. We're excited to [23:06] announce this and this is like our first [23:09] close. Like we're not like we're fully [23:11] ready." But maybe So the tough thing is [23:13] he took he took this offer to Paul, [23:15] chairman of the board. Now Paul has to [23:17] look at this and be like, "Okay, is this [23:19] a real offer?" And I imagine Paul will [23:22] watch CNBC and that's not going to give [23:25] anyone a lot of confidence. [23:26] >> Yeah. It's Yeah, I mean you could [23:28] imagine it in the in the context of like [23:30] buying a house, you know, you show up [23:32] and someone says like, "I have [23:34] 80% of the money and like my bank will [23:37] underwrite me for 50% and I have 30% in [23:40] cash and you're just like, Look, I need [23:42] the full amount. Like I [23:43] >> [laughter] [23:44] >> I need the full amount from [23:45] >> is like I'm I'm I'm wondering I'm [23:47] wondering what uh Ryan he you know, eBay [23:50] is an incredible business. It's been [23:51] remarkably durable. They've faced an [23:53] onslaught of competition for every [23:55] single category from sneakers to watches [23:58] to art to name any cars, right? Any [24:02] category, there is like a vertical [24:04] competitor to eBay and yet the business [24:06] has done has been remarkably remarkably [24:09] strong. [24:10] >> Yeah. It's up pretty meaningfully this [24:12] year, right? Like management is [24:14] executing and [24:16] uh it's unfortunate for Ryan and bid [24:19] that the most viral sort of video clip [24:22] out of all of this is just him failing [24:24] to answer like, you know, a pretty [24:26] straightforward question and not having [24:28] an opportunity to talk about, "Okay, why [24:30] do you you know, if presumably if you're [24:31] buying this company, you think it can [24:33] and should be worth a lot more. What's [24:35] your What's your plan? What's your plan [24:36] for the business? Why are you better [24:38] suited to run it than [24:40] uh than Jamie who's been in in the seat [24:42] since 2020, worked at eBay from 2001 to [24:46] 2009. So he's a veteran, knows the [24:48] business very well [24:50] and uh you're coming in $16 billion [24:53] short. At least that's what it looks [24:55] like. Anyways, uh this bid could be [24:58] could be over before it really started. [25:01] Yeah. Or I mean it could attract a bunch [25:03] of investors who want to line up and [25:06] fall in line and wind up producing the [25:08] the 55 or so required, but uh it does [25:12] feel like it's [25:13] it's a ways away. Breaking news for [25:15] gamers out there. What's that? I know a [25:17] lot of gamers out there. They have [25:19] laptops. They're worried about the price [25:21] of these laptops, the batteries, [25:23] electricity's getting expensive. What [25:25] are you going to do? How are you going [25:26] to charge? We got the solution for you. [25:29] It's a gas It's a gasoline-powered [25:31] laptop. So it's pretty simple. It's a [25:34] one-of-a-kind gasoline-powered laptop. [25:36] It's offered for just $850. [25:39] Looks like it's running on Windows XP. [25:41] You might not be able to play the latest [25:42] and greatest games. Will it run Crysis? [25:44] Maybe, maybe not. [25:45] >> full tank, you can get an hour and a [25:47] half of run time out of this thing. It [25:48] runs a two-stroke engine and it's [25:50] perfect for off-grid computing. That's [25:52] hot right now. And so uh the laptop [25:55] specs, let's take you through it. It's [25:56] got an Intel Core 2 Duo, two gigs of [25:59] RAM. RAM's going up in price. This is [26:01] valuable. This is an appreciating asset. [26:03] 120 gigs of hard drive space. That's [26:07] going to hold a lot of games. When [26:08] you're off-grid, you only have a little [26:09] bit of gasoline. You want to game? This [26:12] is the laptop for you. And of course [26:15] it's running Windows XP. And it says [26:17] that it starts easy. The the the [26:20] two-stroke engine on the [26:21] gasoline-powered laptop, yes, it does [26:23] start easy. [26:24] >> Colin in the X chat says it gets 300 [26:26] tweets to the gallon. [26:27] >> [laughter] [26:31] >> Uh Oh no, I accidentally said GTA V to [26:34] max settings. Well, at least it'll serve [26:37] as a benchmark test. [26:39] Uh how many Chrome tabs can it [26:40] uh open before it crashes? People are [26:42] having fun with this. But [26:44] uh the gasoline-powered laptop, this is [26:46] true hacker mindset. Whoever built this [26:49] is an incredible engineer and did [26:50] something They did the impossible. They [26:52] built a gasoline-powered laptop. I've [26:54] seen a couple other these like uh [26:56] gasoline-powered [26:57] uh projects, people making all sorts of [26:59] different things. It's always a funny [27:00] gag. Obviously, Uh the actual breaking [27:03] news is the White House is considering [27:04] vetting AI models before they are [27:06] released. Hm. Oh, I think FDA [27:08] >> took a non-interventionist approach to [27:11] AI, is now discussing imposing oversight [27:12] on AI models before they are made [27:14] publicly available. Well, FDA for AI. [27:16] We'll see. Uh it would be potentially an [27:19] executive order. Create an AI working [27:21] group that would bring together tech [27:22] executives and government officials to [27:23] examine potential oversight procedures. [27:25] Okay, so this would be an executive [27:27] order to create a working group that [27:29] could potentially create an oversight uh [27:33] body. Okay. So we're a couple steps [27:34] away, but [27:36] seems reasonable. I don't know. Depends [27:38] on what the [27:40] what the what the what the benchmarks [27:41] are, but [27:42] uh you certainly don't want And Trump [27:44] says, "We're going to make this industry [27:46] absolutely the top because right now [27:47] it's a beautiful baby that's born." [27:51] Interesting way to put it. We have to [27:52] grow that baby and let that baby thrive. [27:54] We can't stop it. It's a real quote. [27:56] >> [laughter] [27:57] >> Are you messing with me? I It's a real [27:58] quote [28:00] that Trump said about AI. He said, "We [28:02] have to grow that baby and let that baby [28:04] thrive. We can't stop it. We can't stop [28:06] it with politics. We can't stop it with [28:07] foolish rules and even stupid rules." [28:10] >> a baby. It's a $10 trillion industry. [28:13] >> [laughter] [28:14] >> It's like the the engine of the global [28:16] economy. [28:17] Uh anyway, Dean Bauls got a quote in [28:19] here. [28:19] >> he say? The technology is moving [28:20] extremely fast and there are few formal [28:22] procedures, but they don't want to [28:24] overregulate. Said it's a tricky [28:26] balance. I say don't release it unless [28:28] it's acing tax bench. It's got to be [28:30] able to do the taxes before it gets out [28:32] into the wild. No, obviously, you want [28:34] these models to be safe. You want them [28:35] to be reliable. You want them to avoid [28:38] negative externalities and uh anything [28:41] that's gets us in that direction is [28:43] probably good, but everything comes with [28:45] trade-offs. So Final post of the day [28:48] from Tommy. [28:49] I, PhD in hammerology here. [28:53] All right. So what we're looking at is a [28:54] nail. [28:55] >> [laughter] [28:56] >> That is the correct mindset. When all [28:59] you have is a hammer, everything looks [29:00] like a nail. Uh also, go check out Riley [29:02] Walls' new project. He's shipping stuff [29:04] every week. This one got a million [29:06] views. You probably already saw it. [29:07] 27,000 likes. 10% of AMC movie showings [29:11] sell no tickets at all. So if you want [29:13] to go see a movie in a private theater [29:14] with no one else, he made a site that [29:17] finds empty theaters and tells you [29:20] exactly when you should go and book. You [29:23] can go see Project Hail Mary at 12:30 [29:25] p.m. today in New York. [29:29] Uh if you don't have work, you can go [29:30] see Project Hail Mary in your private [29:32] theater. It's available at [29:34] walzer.com/emptyscreenings. [29:37] w a l z r.com emptyscreenings. You can [29:40] search by zip code. Let's see what's [29:42] around us. Is there anything good? [29:44] There's 10:45 p.m. Devil Wears Prada 2. [29:46] Okay. around us. Zero seats. Enjoy it. [29:49] Enjoy being alive. [29:52] This is very funny. Yeah, there's [29:53] he does he does [29:55] surface some that have one seat or two [29:57] seats. Interesting way to make a new [29:59] friend. Me and you. Cuz you think so you [30:02] think oh I I got the zero seat [30:06] theater. You're you're in the one seat [30:07] theater and then somebody's like I want [30:08] to meet the psycho that went to the [30:10] empty theater. And then they're talking [30:12] your ear off, who knows. Leave us five [30:13] stars on Apple Podcast and Spotify, sign [30:15] up for our newsletter at tvn.com and we [30:18] will see you tomorrow. [30:19] >> Love you. Goodbye. [30:20] >> [laughter] [30:21] >> A little bit miss timed there.