Apple Privacy Prompts Push Developers to Linux for AI Agent Workflows

苹果隐私提示促使开发者在AI智能体工作流中转向Linux macOS强制安全对话框破坏了无头自动化，威胁到苹果二十年来在硅谷开发者心智中的主导地位。 主持人探讨了爆火的AI游戏混合制作（如《我的世界》版《使命召唤》和《阿卡姆骑士》版《蜘蛛侠》）背后的工程现实，阐明了底层逆向工程和“九九定律”（90/90法则）如何将吸睛短片与真正可玩的游戏软件区分开来。他们审视了近期的民意调查，显示千禧一代与较年长选民相比，明显对AI末日论更不以为意；风趣地探讨了厚卡纸取代手工彩纸的原因；并讨论了用户为何会重复编写ChatGPT本身即可原生处理的工具。随后，讨论转向英国对Jaecoo 7等平价中国豪华SUV的采纳，最后探讨了苹果macOS的权限封锁，以及一起自主AI智能体在一台遭入侵的Mac mini上检测并清除加密货币挖矿程序的奇特案例。 如《现代战争2》中的《我的世界》或《蝙蝠侠：阿卡姆骑士》中的《蜘蛛侠》等爆火的AI辅助游戏混合作品，体现了工程领域的“九九定律”——炫目的概念验证虽能在30秒片段中正常运行，却需要耗费数年排查修复碰撞和物理引擎漏洞。 (9:10) 民调显示，千禧一代（30至44岁）是一个反常群体，他们对AI危害的担忧正在减少，这与较年长群体（尤其是65岁以上选民）形成了鲜明对比，后者越来越将AI视作生存威胁。 (16:15) 奇瑞等中国车企正凭借Jaecoo 7等售价仅为路虎揽胜一半的替代性豪华SUV迅速抢占英国市场份额，而美国的关税和拟议法案则将其拒于美国市场之外。 (22:47) 苹果不断收紧macOS安全权限，在运行自主开发者智能体的无头Mac mini上强制弹出图形化确认弹窗，危及高级用户的工作流，促使用户转向Linux。 (27:30)

TBPN explores how rapid AI advancements are colliding with real-world execution, from the fragile engineering behind viral game mods to the platform friction between Apple's security policies and autonomous developer workflows. Viral Game Mods and the Engineering Gap (0:02) Recent AI-assisted gaming mashups, such as running Modern Warfare 2 mechanics inside Minecraft or pairing Spider-Man with Batman: Arkham Knight, demonstrate impressive technical feats using custom Rust harnesses and NSA-developed reverse-engineering tools like Ghidra. However, the hosts emphasize that these viral demonstrations obscure deep underlying fragility. In live tests, mechanics break down quickly—such as swimming physics failing over Minecraft geometry—illustrating Tom Cargill's 90/90 rule of software engineering, where the final polish demands the vast majority of development time. Generative Demonstrations Versus Sustained Playability (11:51) The program draws a direct parallel between today's 30-to-60-second social media mod clips and the viral March 2023 'Harry Potter by Balenciaga' video created by DemonFlyingFox. That video relied on rapid cuts under two seconds and dim lighting to mask Midjourney background hallucinations, sparking predictions that automated AI feature films were imminent that ultimately never materialized. While landmark historical games like Counter-Strike and Dota emerged from community modding roots, the hosts argue current AI mashups primarily exploit dual nostalgia rather than delivering stable, 100-hour core gameplay loops. Diverging Generational Perceptions of AI Risk (16:15) Polling from late 2024 reveals a significant demographic divide in artificial intelligence sentiment. While senior voters aged 65 and older are surging toward panic—with roughly 40% viewing AI as a very serious threat—millennials aged 30 to 44 are reported by The Economist to be less worried than ever. Commentators attribute this resilience to millennials coming of age alongside the consumer internet and early mobile computing, which gave them higher tolerance for technological disruption and a more pragmatic, risk-managed outlook compared to older voting blocs. Chinese Electric Vehicles Challenge Western Markets (22:47) Chinese carmakers are expanding rapidly in the UK by offering luxury-tier aesthetics and technology at competitive price points. Models like Chery's Jaecoo 7 sell for approximately £50,000, roughly half the price of a comparable Range Rover, driving Chery's UK sales to more than triple year-over-year through September. Although critics label such vehicles 'Temu Range Rovers,' the hosts note the build quality and engineering are mechanically solid. This UK market acceptance stands in contrast to the United States, where strict tariffs, proposed bans, and domestic automaker anxieties continue to lock Chinese imports out. Apple's Platform Guardrails Versus Autonomous Workflows (27:30) Apple is introducing stricter macOS data controls to protect unencrypted local files, such as iMessage databases, from being accessed by AI agents. However, these recurring graphical permission prompts hinder headless workflows, leading power users like analyst Ben Thompson to consider switching to Linux for autonomous development. A highlighted developer incident demonstrated the trade-off: an open network port allowed an attacker to install an unauthorized crypto miner on a Mac Mini, but the owner's autonomous agent detected the breach, closed the port, and purged the malware within 15 minutes, illustrating how deep system access can double as both a security vulnerability and an active defense mechanism. Across gaming, automotive manufacturing, and developer operating systems, emerging technologies face a critical transition from viral proof-of-concept to institutional adoption, where long-term viability hinges on software stability, user education, and security governance.

## Background 本期视频由科技对谈频道 TBPN 于 2026 年 10 月 7 日发布在 YouTube 平台。该节目为“Diet TBPN”精简栏目，视频简介说明节目将当日直播浓缩为约 30 分钟的高密内容，涵盖 AI 游戏 Mod、中国汽车在英国的渗透热潮以及苹果收紧 Mac 数据控制等焦点科技议题。 ## TL;DR 节目首先从技术底层深入拆解了 AI 混合游戏 Mod 的运作原理，结合“90/90 定律”指出这类病毒式 Demo 距离成型作品仍有巨大工程长尾与 IP 侵权阻碍。随后，主持人们分析了千禧一代因对技术变革脱敏而反常地“最不恐慌 AI”的民调现象，并穿插探讨了电商纸箱对传统彩色手工纸的挤占。后半程节目聚焦商业与生态，探讨了奇瑞“Temu 版路虎”在英国市场的低价狂飙与潜在售后隐患，最后深度评析了 macOS 严格的权限沙盒对 AI Agent 自动化的阻碍，并以 Mac mini 遭挖矿病毒入侵后由本地 AI 在 15 分钟内自治防御的戏剧性案例作结。 ## Atmosphere & style 整体基调轻松诙谐且充斥着浓厚的极客硬核气质，主播约翰与霍尔迪在随性调侃与深度技术剖析之间自如切换。现场互动节奏明快、充满硅谷式幽默——从开场对圣诞节倒计时的执念，到直播中途拿着手电筒猎杀蚊子，再到拆解 NSA 反编译工具与跨引擎数据流，节目在极度接地气的生活插曲与严肃的系统级架构探讨间保持着极高信息密度。听感上既像是在偷听两位懂技术的极客老友在车库里闲聊吐槽，又能从中敏锐捕捉到前沿软硬件与地缘商业博弈的真实切片。 ## Bulletin - [00:41](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=41s) 民间开发者正借助 Rust 轻量引擎、Codex 脚本及 NSA 开源逆向工具 Ghidra，将《现代战争2》与《我的世界》等截然不同的游戏资产及底层物理代码进行跨引擎拼接，但水下判定等跨系统交互脱节暴露了当前的粗糙度。 - [06:05](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=365s) 新型跨游戏 Mod 技术“ArkWeb”实现了双 3A 游戏在后台并行运行与毫秒级实时数据穿透，成功将漫威《蜘蛛侠》物理系统置入 DC《蝙蝠侠：阿卡姆骑士》的哥谭市中，但受限于软件工程“90/90定律”，离稳定商业化游戏仍有巨大鸿沟。 - [16:15](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=975s) 《经济学人》民调显示，30 至 44 岁的千禧一代对 AI 的危害恐慌降至新低，而 45 岁以上群体恐慌率攀升至约 40%；专家分析千禧一代将 AI 视作如香烟般的“已知可控风险”，具备更高的技术免疫力。 - [22:11](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1331s) 奇瑞旗下 SUV 杰酷 7（Jaecoo 7）以约 5 万英镑的半价与高配置迅速占领英国市场并被戏称为“Temu版路虎”，带动奇瑞前三季度英国销量同比翻三倍以上，而美国市场则持续收紧对中国汽车的准入壁垒。 - [27:18](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1638s) 苹果收紧 macOS 系统权限与强制 UI 弹窗确认机制，导致 Ben Thompson 等开发者的“无头 Mac mini”自动化 AI Agent 频繁受阻，面临将核心黑客群体推向 Linux 的生态风险；但真实案例也表明，拥有高权限的本地 Agent 能在 15 分钟内自主防御并清除黑客植入的挖矿木马。 ## Key takeaways - [03:42](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=222s) 游戏混合 Mod 呈现多梯队技术实现，从官方授权皮肤、传统贴图替换，到通过 NSA 开源的 Ghidra 进行 ROM 二进制反编译提取底层物理逻辑（如 libsm64）。 - [06:05](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=365s) 跨游戏实时数据穿透技术（如 ArkWeb）通过动态重映射坐标系与劫持控制器输入，实现了双 3A 引擎在后台并发运行及近 60 FPS 的画面渲染合成。 - [09:25](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=565s) 软件工程“90/90定律”揭示了 AI 混搭演示的虚假繁荣：前 90% 的原型代码耗费 90% 时间，而解决物理冲突、穿模和系统连锁崩溃等剩下的 10% 尾声工作往往还需耗费另外 90% 的时间（可能长达 5 年）。 - [11:08](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=668s) 类比 2023 年《巴黎世家哈利波特》AI 视频，网络上爆火的跨游戏混搭依赖于“双重情怀叠加”和快节奏剪辑掩盖缺陷，属于快餐式迷因，无法替代原版游戏的长期可玩性。 - [15:06](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=906s) 主流硬核玩家对生成式 AI 仍存抵触情绪，且民间资产搬运面临严重的公然侵权风险；要扭转玩家偏见，唯有顶级厂商（如 Rockstar 制作全球级《GTA 7》）利用 AI 实现质的飞跃。 - [16:15](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=975s) 千禧一代（30-44 岁）成为最不害怕 AI 的人群，因其成长于互联网与智能手机变革期，对算法危害与屏幕成瘾有着防御经验，形成了类似于看待“香烟”的理智脱敏反应。 - [19:20](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1160s) 电商快递纸箱（瓦楞纸板）的后匮乏普及，使得结构脆弱的传统彩色手工纸（construction paper）在儿童与创客手工生态中逐渐边缘化。 - [20:38](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1238s) ChatGPT 等大语言模型面临严重的“隐藏功能困境”，诸如定时计划任务等进阶自动化能力未被有效宣传，导致用户甚至花费半小时“重新发明”其原生功能。 - [22:11](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1331s) 中国车企凭借“半价路虎”杰酷 7 等产品通过过硬工业制造与高配置在英国迎来销量井喷，但仍需直面海外售后维保网络脆弱的潜在隐患。 - [26:20](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1580s) 英美对中国汽车态度分化鲜明：英国成为中国汽车渗透率最高的西方工业化国家，而美国则出于传统汽车业保护与国家安全立法拟永久封杀。 - [27:18](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1638s) macOS 针对 iMessage 等本地数据的安全防范与严苛的沙盒弹窗，破坏了无头模式（headless）下 AI Agent 自动编程的流畅体验，动摇了 Mac 作为“黑客首选平台”的传统地位。 - [29:00](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1740s) AI Agent 系统权限是一柄双刃剑：过度限制会扼杀自动化潜能，而放开底层控制虽增加暴露面，却能在 15 分钟内完成检测入侵、封堵端口和清除挖矿木马的自主防御闭环。 ## Detailed walkthrough ### [00:02](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=2s) 距离圣诞节只有80天：万圣节只是路上的减速带 节目开场，主持人约翰·库根（John Coogan）与霍尔迪·海斯（Jordi Hays）聊起了一个略带执念的时间话题：十月是通往圣诞节路上最艰难的一个月。大家之所以容易分心，全是因为半路杀出的万圣节。库根调侃道，如果你现在走在街上对别人说“距离圣诞节只剩 80 天了”，对方肯定会像看疯子一样看着你，质问你为什么不倒数万圣节。但库根坚决认为真正疯了的是那些不数圣诞节倒计时的人，“万圣节不过是通往圣诞节路上的一处减速带（speed bump）”，大家必须无视干扰，坚定地倒数迎接圣诞。 ### [00:41](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=41s) 混合游戏 Mod 狂潮：当《现代战争2》撞上《我的世界》 话题很快转向近期游戏圈引发争议的“AI 生成游戏 Mod”现象。知名游戏媒体 Kotaku 此前发表了一篇态度尖锐的批评文章，直言“我根本不想玩你们用 AI 捣鼓出来的把《滑板3》（Skate 3）塞进《现代战争2》（Modern Warfare 2）再塞进《我的世界》（Minecraft）的缝合 Mod”。然而在直播现场，制作人泰勒（Tyler）却兴致勃勃地当场演示起了一款混合 Mod。 泰勒解释了这一 Mod 的实际运行机制：它并非真正意义上把一个游戏的完整程序硬塞进另一个游戏，而是基于一个完全由 Rust 语言编写的独立引擎，同时从两款游戏中提取原始资产并加以拼合。为了运行它，泰勒必须在 Steam 上正版购买《使命召唤：现代战争2》，并拥有《我的世界》正版资产，随后借助 OpenAI Codex 编写脚本进行环境配置，最终成功在 Mac 设备上跑通。游戏内保留了双方的核心玩法——泰勒既可以使用突击步枪射击，也可以像在《我的世界》里一样进行采集挖矿与建造方块。库根立刻追问能否在里面打出连杀奖励甚至呼叫“战术核弹”（Tactical Nuke），泰勒则表示还在摸索触发机制。 由于这类技术尚处于早期且带有浓重的“氛围编程水准（vibe-coded slop）”，游戏机制还存在明显的拼接穿帮。例如泰勒尝试在水下游泳时，角色完全没有浮水或溺水判定，而是如同在平地一般自由行走。库根推测，这是因为角色的生命值与受击逻辑由《现代战争2》的代码监管，而周围的湖泊与方块仅仅是《我的世界》渲染出来的空间几何体，两者缺乏深度交互机制。演示过程中，泰勒还在方块世界里完成了一记狙击跳杀小猪的搞笑操作（trickshot），惹得主持人直呼需要插播“血腥内容警告”。 - **资产合法性与环境**：玩家必须购买并持有两款游戏的底层正版资产（Steam 平台购买《现代战争2》与正版《我的世界》）。 - **底层架构**：基于轻量级第三方 Rust 游戏引擎进行重构，并使用 Codex 辅助代码桥接与 Mac 端移植。 - **机制脱节表现**：水体交互失效，角色在水底直接行走且不掉血，显示出生理状态（现代战争机制）与环境物理（我的世界几何网格）尚未完全打通。 ### [03:42](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=222s) 逆向工程与国家安全局：Mod 极客的底层技术手段 库根随后对当前层出不穷的 Mod 进行了技术分类，指出外界常常把它们一概而论，但实际实现原理跨度极大： 1. **官方企业认可型（Corporate-Approved）**：最典型的是《堡垒之夜》（Fortnite）联动的角色皮肤（如蝙蝠侠或蜘蛛侠）。这类 Mod 完全由游戏原生引擎内部开发，只引入了外观模型，不运行任何外部游戏代码，安全性高且拥有完备商业授权。 2. **传统资产替换型（Old-school Asset Swap）**：单纯将外部 3D 模型或贴图强行导入目标游戏，底层逻辑依旧是原版代码。 3. **反编译与底层代码抽取（Decompilation）**：以近期爆火的“马里奥入侵《艾尔登法环》”为例，该 Mod 借鉴了名为 `libsm64` 的子系统，其核心源自对任天堂经典游戏《超级马里奥64》（Super Mario 64）卡带 ROM 的完全反编译。反编译工具直接读取卡带中的二进制机器码（0 和 1），反向推演其汇编指令并重构成高级逻辑代码（例如将内存地址 `0x1234fc` 逆向识别为角色血量变量 `health = 100`），从而将原版游戏的跳跃物理和运动状态机完整移植到外部引擎。 4. **国家级情报工具介入**：库根笑称玩家总嚷嚷着“让政治远离电子游戏”，但在技术现实中根本避不开，因为大量核心 Mod 制作者实际上在使用美国国家安全局（NSA）开源的逆向工程工具 **Ghidra**。该工具托管在 NSA 的官方 GitHub 仓库中，功能极为强大，专门用于拆解、分析可执行二进制程序并还原代码逻辑。 ```mermaid flowchart TD accTitle: 游戏逆向工程与底层代码提取流程 accDescr: 展示从游戏只读卡带二进制文件到通过NSA Ghidra工具反编译并嵌入新引擎的完整技术链条 subgraph Source ["原始只读卡带 / 编译程序"] A["卡带 ROM 二进制文件<br/>(0 与 1 机器码)"] end subgraph ReverseEng ["逆向分析与反编译"] B["NSA 开源工具 Ghidra / 反编译器"] C["内存地址逆向推演<br/>(例如 0x1234fc 识别为 health)"] end subgraph ModularLib ["跨引擎逻辑移植"] D["提取独立子系统<br/>(例如 libsm64 物理库)"] E["植入现代游戏宿主<br/>(例如《艾尔登法环》)"] end A --> B B --> C C --> D D --> E ``` ### [06:05](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=365s) 双游戏实时穿透：蜘蛛侠在哥谭市荡绳与 90/90 定律 库根重点介绍了一种名为 **ArkWeb** 的全新 Mod 形式——它打破了传统的代码逆向逻辑，实现了跨游戏的“实时数据穿透（Passthrough）”，让玩家能够在《蝙蝠侠：阿卡姆骑士》（Batman: Arkham Knight）的哥谭市中操作漫威的《蜘蛛侠》（Spider-Man）。 从商业版权层面来看，这一联动在官方层面绝无可能：漫威（归迪士尼所有）与 DC（归华纳兄弟探索所有）是直接竞争对手，双方几乎不可能达成这种深度技术合作。但该 Mod 从技术实现上极其惊艳：它并不是在蝙蝠侠游戏里换个蜘蛛侠皮肤，而是**在后台同时启动并运行两款完整的 3A 游戏**，并在两者之间构建实时双向数据通道。 - **实时几何与碰撞网格传递**：Mod 实时读取蝙蝠侠引擎中哥谭市的高楼模型、空间坐标与碰撞体积数据，在毫秒级内将其实时同步给处于后台运行的《蜘蛛侠》物理引擎。 - **空间标准实时转换**：两款引擎的物理系统、坐标系定义（X/Y/Z 轴朝向）以及长度单位完全不同，转换层必须在每一帧内对数据进行动态重映射。 - **控制器与渲染融合**：常规操作系统只允许单一前台窗口捕获手柄指令，该 Mod 劫持了输入流，让两款游戏同步响应；更将两个截然不同的图形渲染管线合成在一个窗口中，并维持在接近 60 FPS 的流畅帧率。 - **战斗无缝切换**：在空中摆荡移动时由蜘蛛侠物理主导，一旦进入近战搏击，系统立即切换为《阿卡姆骑士》标志性的“自由流动战斗系统”（Freeflow Combat）。 虽然偶尔会出现贴图撕裂等小 Bug，但库根认为这为重温经典老游戏提供了前所未有的全新动力。这不仅是重置版（Remaster）的概念，更是打破 IP 与物理引擎壁垒的复合体验。 最后，库根将这些火爆但仍显粗糙的 AI 与混合游戏 Demo 归结于经典的**“90/90定律”（The 90/90 Rule）**。这一格言由计算机界传奇人物、贝尔实验室（Bell Labs）的汤姆·卡吉尔（Tom Cargill）提出：在软件工程中，“前 90% 的代码消耗了前 90% 的开发时间；而剩下的 10% 代码，还要再消耗另外 90% 的开发时间”。当前网络上刷屏的病毒式混合游戏演示，本质上都处于前 90% 的震撼原型阶段，距离真正完善、无 Bug 的成型游戏，还有一条极其漫长的长尾等待填补。 ### [09:25](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=565s) 软件开发的“九九定律”与打磨陷阱：从 Skycraft 到物理引擎崩溃 在探讨将不同游戏进行 AI 混搭的技术现实时，主持人们提到了软件工程中著名的“九九定律”（Ninety-ninety rule）：前 90% 的开发代码消耗了 90% 的预估时间，而剩下的 10% 往往又要消耗另外 90% 的开发时间。无论你在立项时做出了多么乐观的估算，项目的实际范围（scope）总是会不可避免地膨胀。如果要针对这类跨游戏混搭进行彻底的“捉虫大会”（bug bash），将其打磨成一个真正可玩、稳定的商业级游戏，开发团队面临的可能直接是整整五年的工期。 以《侠盗猎车手》（GTA）中嵌入《我的世界》（Minecraft）或多人第一人称射击游戏为例，实现一个简单的可运行版本看似并不复杂，但难易度完全取决于你如何界定开发范围。如果只是简单地替换材质贴图——比如直接套用《使命召唤》（Call of Duty）的角色模型，仅把外观贴图换掉——通常不会引入太多系统性 Bug。然而，一旦你试图引入一套完全不同的物理力学机制或爆炸破坏效果，风险就会呈指数级攀升。角色随时可能穿模跌落到地图之外，或者碰撞体判定彻底失效；而为了修复碰撞体积，工程师就必须重写底层碰撞逻辑，这又会连锁触发更多其他模块的崩溃。 目前网络上流传的各种炫酷演示视频，没有一个是真正展示完整游戏流程的。比如著名的一款将《我的世界》机制引入《上古卷轴5：天际》（Skyrim）的开源 Mod——“Skycraft”。在开发者的官方安装文档中，甚至赫然写着一条避坑警告：“千万不要正常开局游戏，不要尝试走原版开场剧情（如坐马车进刑场），游戏必定会直接崩溃；必须使用替代起始点 Mod，从游戏后半段的节点直接切入”。主持人们调侃道，这就是典型“AI 废料时代”（slop era）的真实写照——表面光鲜，内里漏洞百出。 - **90/90 软件工程定律**：前 90% 的进度耗时 90%，剩下的 10% 尾声工作还要再耗时 90%，范围膨胀导致打磨期动辄需要 5 年。 - **贴图替换 vs. 物理引擎**：单纯替换模型材质风险极低；一旦改动物理碰撞、载具动态或爆炸逻辑，极易引发穿模与连锁系统故障。 - **Skycraft 现实案例**：该 Skyrim 开源 Mod 甚至无法通过原版序章，开发者只能在文档中明示玩家“必须使用跳过开场的备选出生点”。 --- ### [11:08](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=668s) 游戏界的“巴黎世家哈利波特”时刻与病毒式短视频陷阱 这种游戏混搭之所以能在社交媒体上疯传，本质上是因为它们完美契合了短视频时代的传播规律。用户在刷 TikTok 或 Instagram Reels 时，只要看到 30 到 60 秒的片段，就会被瞬间击中：这是一种强烈的“双重情怀叠加”（dual nostalgia）。“我既深爱《GTA 5》，又深爱《我的世界》，把它们拼在一起就是双倍的快乐。”这些视频因此轻松收割数百万点赞。然而，根本没有哪个玩家会真正坐在电脑前玩上一百个小时——论真正的可玩性与娱乐价值，《GTA 5》和《我的世界》原版依旧牢牢占据统治地位，短视频混搭不过是纯粹的“网络迷因溢价”（meme value）。 ```mermaid flowchart TD accTitle: 巴黎世家哈利波特 AI 视频制作管线 accDescr: 展示 Demonflyingfox 结合专用模型、Midjourney 与快速剪辑的制作流程 subgraph Gen["素材生成阶段"] A["训练专属人脸模型<br/>（如丹尼尔·雷德克里夫）"] B["Midjourney 生成复古走秀图<br/>（暗黑城堡与高定服装）"] end subgraph Comp["合成与后期阶段"] C["换脸算法（Face Swap）<br/>将明星面孔贴入画面"] D["微幅动态化（Animation）<br/>赋予静态图轻微动感"] E["高频剪辑（Rapid Editing）<br/>单帧停留不超 1-2 秒以遮丑"] end A --> C B --> C C --> D D --> E ``` 这让人联想起 2023 年 3 月 15 日由创作者 Demonflyingfox 发布的病毒式 AI 视频《巴黎世家哈利波特》（Harry Potter Balenciaga）。距离那个轰动全网的时刻已经过去了三年半。回顾当时的技术实现：创作者针对哈利·波特、丹尼尔·雷德克里夫（Daniel Radcliffe）等角色的人脸单独训练定制模型；利用 Midjourney 生成符合 ArtStation 风格的霍格沃茨古堡暗黑走秀图——当时 Midjourney 的暗光与扭曲幻想背景恰好掩盖了细节缺陷；随后将人脸置换进去并加入微弱的动态呼吸效果；最后依赖极高密度的视频剪辑，确保每个画面停留不超过一两秒，不让观众有时间察觉画面瑕疵。 当时舆论狂热地预言“明年我们就能看全 AI 制作的院线大片”，但现实证明，全 AI 长片并未如期降临，Demonflyingfox 后来制作的多部续作也逐渐失去了新鲜感。不过，这种技术探索对游戏界依然具有深远启示：历史上伟大的商业作品往往脱胎于民间 Mod，例如《反恐精英》（Counter-Strike）最初只是《半衰期》（Half-Life）的一个 Mod，而《Dota》最初也是《魔兽争霸》（Warcraft）的社区衍生品。AI 工具的成熟将大幅降低门槛，将更多人推入独立开发者阵营。但即便生产力翻倍，要想打造出下一个现象级游戏，开发者依旧要把 99.999% 的心血砸在最后 1% 的机制创新与核心玩法闭环中。 - **短视频的双重情怀红利**：融合两大经典 IP 带来流量暴击，但属于快餐式迷因消费，无法替代原版游戏的沉浸式长期体验。 - **Demonflyingfox 的技术把戏**：利用 Midjourney 氛围图掩盖 AI 幻觉，配合人脸模型替换与快节奏蒙太奇（单镜头 1-2 秒），规避模型崩坏。 - **独立开发者的黄金与残酷律条**：如同《Half-Life》孕育《CS》、《魔兽争霸》孕育《Dota》，AI 加速了原型开发，但核心机制设计与漫长打磨仍不可替代。 --- ### [15:06](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=906s) 游戏玩家对 AI 的态度、IP 侵权与宏大叙事瓶颈 对于千禧一代的玩家是否会因为怀念《光环》（Halo）或《使命召唤：现代战争2》（Modern Warfare 2）而迅速倒向“挺 AI”（pro-AI）阵营，两位主持人持保留态度。在硅谷之外，庞大的核心游戏玩家群体对生成式 AI 依然抱有极大的敌意与抵触心理，无论是出于对人类画师权益的保护，还是对低劣生成内容的鄙夷。 主持人们分析，要让整个游戏玩家群体真正实现态度上的逆转并全面拥抱 AI，前提必须是**顶级的殿堂级游戏工作室利用 AI 技术，打造出一款规模空前庞大、体验显著进化的全新正作**。单纯靠民间搬运资产拼凑概念，永远无法形成真正的商业化产品，因为这中间横亘着极其致命、明目张胆的知识产权（IP）侵权红线。 他们畅想，除非有朝一日 Rockstar Games 推出《GTA 7》，并宣布“整个游戏地图涵盖了真实世界中的全球每一座城市”，全靠底层 AI 引擎实时生成且运行丝滑，玩家才会真正感到震撼。在此之前，民间 AI Mod 的狂欢除了让知识产权诉讼律师因为海量侵权案件而业务爆棚、疯狂数钱之外，很难在主流商业发行渠道掀起真正的风浪。 - **玩家群体的抵触情绪**：主流游戏社区对 AI 技术仍具广泛戒心，民间零散的情怀演示难以动摇硬核玩家的审美偏见。 - **质变的必要条件**：只有像 Rockstar 这样具有号召力的第一梯队大厂，用 AI 做出此前无法想象的体量（例如完整重构现实世界的《GTA 7》），才能扭转舆论。 - **致命的 IP 法律雷区**：跨游戏资产的擅自融合属于“极其公然的侵权行为”，这反而让专注于版权诉讼的律师们成了 AI 热潮中最兴奋的受益群体。 --- ### [16:15](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=975s) 为何千禧一代反常地“最不害怕 AI”？ 随后的讨论聚焦于《经济学人》（The Economist）最新公布的一项民调图表。该调查追踪了公众在 8 月 28 日至 9 月 18 日这短短三周时间内，对“AI 造成严重危害的风险有多大”的认知变化。数据显示，几乎所有年龄段的受访者都在“向右移动”，即对 AI 的威胁感到越来越恐慌；特别是在 45 岁以上以及 65 岁以上的核心选民群体中，认为 AI 会带来极严重危害的比例已飙升至 40% 左右，这势必将对接下来的选举产生巨大政治推力。 然而，图表中出现了一个极度反直觉的“叙事冲突”（narrative violation）：**30 岁至 44 岁的千禧一代群体，对 AI 的恐惧程度非但没有上升，反而创下了历史新低**。在这三周里，这批中青年人同样看过了 Anthropic 首席执行官达里奥·阿莫代伊（Dario Amodei）的深度警示访谈、作家埃兹拉·克莱恩（Ezra Klein）的播客对话，甚至看过了《周六夜现场》（SNL）嘲讽 AI 的小品，但他们看罢的反应却是：“哦，我以前还挺焦虑的，但现在感觉其实还行，我稳得很。” 针对这一奇特现象，科技评论人吉姆·普罗瑟（Jim Prosser）提出了一个假设：千禧一代是伴随着互联网黎明期和智能手机浪潮一同成年的一代人，天生对剧烈的技术颠覆和不确定性拥有最高的心理耐受度。此外，很多 30 到 40 多岁的千禧一代如今初为人父人母，他们亲历了早期算法“大脑腐蚀”（brain rot）、信息垃圾以及屏幕时间成瘾的危害，并见识了大量关于儿童手机成瘾的数据研究。因此，他们对技术风险有着清晰的预判和掌控感（比如坚决不在孩子第一天就递上 iPad）。正如这一代人从小就深知香烟有害并能理性防范一样，面对 AI，他们觉得“危险是已知且可控的”，从而展现出了超乎寻常的松弛感。 - **各年龄段 AI 恐慌分歧**：45 岁及 65 岁以上选民对 AI 的严重危害感知急剧升温（占比达 40%），成为驱动政策监管与大选的关键情绪。 - **千禧一代的“叙事背离”**：30-44 岁人群反其道而行之，在大量严肃访谈与通俗媒体轰炸下，恐惧指数降至新低。 - **“香烟效应”与技术免疫**：千禧一代在互联网与移动端变革中建立起强大的心理抗性，将 AI 视作类似于香烟的“已知危害”，坚信自己具备足够的防御和规避能力。 --- ### [18:24](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1104s) 闲聊插曲：彩色手工纸去哪儿了？ 在严肃的 AI 话题告一段落后，主持人们被网友 Celine 的一条怀旧推文逗乐了。推文提问道：“最近怎么完全听不到有人提彩色手工纸（construction paper）了？在我的童年时代，这玩意儿几乎时刻占据着我的大脑中心。” 两人随即展开了一番充满生活趣味的漫谈。他们回忆起小时候用这种粗糙厚纸画画的经历——它的唯一核心优势大概就是纸张够厚，水彩笔墨水不会轻易渗透到背面。但至于它为什么叫“手工纸（用于构建的纸）”，大家其实很少真拿它盖出什么三维物体，顶多也就是在圣诞节等节日时剪成一条条纸带，粘成一串串挂在教室里的纸质花环链（daisy chains）。 如果真要在学校里做立体透视模型（diorama），大家最终还是会老老实实掏出硬纸板（cardboard）。主持人们推测，彩色手工纸很可能就是被越来越普及的瓦楞纸板挤压了生存空间。最后，两人还兴奋地聊起如今市面上专门用来切割纸板的小电锯工具：“那玩意儿简直就像一把微型电锯，能极其顺滑地将硬纸板撕开一条口子”，童年手工用具的演变引得现场一阵感叹。 - **童年集体记忆**：彩色手工纸曾是小学生美术课的绝对顶流，主要用于防渗水涂鸦和制作节日纸环链。 - **功能定位的尴尬**：空有“建造”之名，但在真正制作立体透视模型时，结构强度远逊于硬纸板，导致其逐渐淡出公众视线。 - **现代手工工具的演进**：纸板的广泛普及以及现代专业硬纸板切割锯的出现，彻底取代了脆弱的传统手工纸折叠搭建方式。 ### [19:20](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1160s) 快递纸箱与卡纸的消亡 主持人探讨了现代家庭材料变迁的趣味话题：手工卡纸（construction paper）的没落是否是由快递纸箱（cardboard）造成的。随着电子商务的爆发式普及，几乎每个家庭——尤其是有小孩的家庭——都处于一种“纸箱后匮乏（post-scarcity）”的状态。过去有远大抱负的孩子如果想制作经得起时间考验的坚固手工作品，会选择卡纸，但现在纸箱唾手可得，反而成了孩子们甚至成年人手工项目的首选材料。 话题随后延伸到金融投资的冷幽默调侃。约翰寻找在市场上做多或暴露于“纸制品/手工材料”的最佳方式，并提到了一家市值约 5 亿美元的意大利老牌造纸与文具公司（Fabriano / Fedrigoni 集团相关概念），其股价在过去五年间几乎是一条完全平缓的直线，堪称“完美的价值储藏手段（perfect store of value）”。 --- ### [20:38](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1238s) 重新发明轮子与 ChatGPT 的“隐藏功能困境” 主持人分享了网络上一个令人啼笑皆非的案例：一位名叫 Iman 的用户花了整整半个小时与 ChatGPT 反复拉扯提示词，试图让 AI 开发出一款自动化小工具，能够根据他正在研究的任何主题自动抓取并输出相关网页链接。该工具最终成功交付，但仅仅运行两分钟后，这位用户便意识到自己的荒唐之处——他实际上用 ChatGPT 重新开发了一个 ChatGPT 本身就自带的功能，“用工具制造出了完全相同的工具”。 这场乌龙引发了关于大语言模型隐藏功能与产品教育的深刻讨论。主持人指出，这揭示了一个严重的现状：AI 模型的诸多强大能力其实被深埋在系统内部，毫无直观曝光。例如，ChatGPT 本身完全支持类似计划任务（cron job）的功能——用户如果想每周自动生成一份全球冲浪巨浪地点的摘要周报，无需编写一行代码，无需配置任何服务器托管，直接吩咐 ChatGPT 即可实现。然而，OpenAI 从未像推广重量级新功能那样向公众开展全国性的宣发教育。随着基础模型与外围框架的功能越来越强，如何有效教育大众使用这些进阶能力已成为巨大难题。 --- ### [22:11](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1331s) 美联航上的《Cheeky Pint》与英国街头的“Temu版路虎” 在一段关于达美航空（Delta）与美联航（United Airlines）机上娱乐系统的小插曲中，主持人调侃美联航如今可以在航班上观看播客《Cheeky Pint》归档，展现了其“高度商业化、不让乘客看无聊电影”的硬核风格，并拿主持人苏珊·李（Susan Lee）开玩笑。演播室中途还突然闯入蚊子，两人甚至起哄拿手电筒在镜头前“猎杀蚊子”，为节目增添了不少现场乱斗的喜剧色彩。 随后切入重磅汽车话题：英国消费者正在集体“背叛”本土老牌豪车，疯狂抢购来自中国奇瑞（Chery）旗下的杰酷 7（Jaecoo 7）SUV。这款车因外观酷似路虎揽胜（Range Rover）——尤其是揽胜极光（Evoque），被英国人戏称为“Temu版路虎（Temu Range Rover）”。 ```mermaid flowchart TD accTitle: 中国车进军英国市场驱动逻辑 accDescr: 展示杰酷7以半价和高配置切入英国市场的对比逻辑 subgraph CostDiff["成本与外观对比"] A["传统英伦豪华车<br/>Range Rover / Evoque<br/>约 100K 英镑"] B["奇瑞 Jaecoo 7<br/>'Temu Range Rover'<br/>约 50K 英镑"] end subgraph ConsumerAppeal["消费者核心吸引力"] C["半价购车门槛"] D["豪华外观与内饰质感"] E["高配置智能化科技"] end subgraph MarketResult["市场渗透"] F["英国销量井喷<br/>奇瑞前三季度销量翻三倍"] G["售后与残值潜在隐患<br/>（类比 Fisker 倒闭维保风险）"] end A -.->|"视觉相似度极高"| B B --> C B --> D B --> E C & D & E --> F F --> G ``` 一名来自英国南海岸的 33 岁消费者马克·图利斯（Mark Thulis）在选购家庭用车时发现，顶配版 Jaecoo 7 的购车成本仅需约 5 万英镑，差不多是外观相似的路虎车型的一半价格。他的第一反应是“天上掉馅饼必有诈”，但在亲自试驾后立刻觉得不买就是傻子。约克郡利兹一家经销商的总经理直言其吸引力极为简单直接：“看起来很贵，摸起来很贵，然后你一问价格，竟然这么便宜。” 主持人分析指出，将它贬低为“Temu 版”并不准确，因为 Temu 商品通常是图片好看、实物散架，但中国车造车工艺过硬，机械和智能系统可能比原版路虎更加可靠扎实。不过，新兴品牌的潜在短板仍在于线下经销商网络与售后服务体系的铺设。如果品牌中途重组或退出市场，车主可能面临无处修车或软件无法升级的窘境，就像此前破产的 Fisker Ocean 电动车车主所遭遇的恶梦一样（相比之下，老款的 Fisker Karma 如今在路上依然是一道独特的风景，甚至连杰·雷诺（Jay Leno）也保有一台）。 在宏观政策层面，英国目前是中国汽车出口渗透率最高的西方工业化国家，奇瑞截至 9 月的销量同比翻了三倍以上。反观美国市场，中国汽车则被高额关税和软件安全禁令严密封锁。尽管前总统特朗普曾表态如果中国车企在美国本土建厂制造、雇佣美国工人，他持开放态度，但这引发了底特律传统汽车巨头的强烈恐慌。美国国会目前甚至在讨论立法，计划永久禁止中国汽车驶上美国公路。 --- ### [27:18](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1638s) 苹果的 macOS 权限死结与 Ben Thompson 的无头 Mac Mini 节目最后探讨了苹果计划为 Mac 用户引入全新隐私控制机制的消息。随着第三方软件及各类 AI Agent（智能体）开始广泛索取系统权限，底层安全风险急剧攀升。例如，Mac 硬盘上的 iMessage 短信数据库在本地并非不可读取的加密状态；若用户赋予 AI 智能体“编辑本地文件”的泛化权限，该智能体很可能会顺带读取敏感的聊天记录，因此苹果不得不收紧权限颗粒度。 然而，这种激进的安全弹窗机制彻底激怒了专业开发者。科技分析师本·汤普森（Ben Thompson，Stratechery 创始人）近期便濒临放弃 macOS 转投 Linux 桌面系统。汤普森在本地运行了一台用于远程连接的“无头（headless）Mac Mini”，主要利用 AI 智能体自动编写软件工具。但 macOS 强制要求任何网络访问、屏幕截取或文件读写都必须由人工在系统 UI 弹窗上物理点击“允许”。远程无界面的自动化流程频频被这些弹窗打断，令他痛苦不堪。 主持人们回顾了保罗·格雷厄姆（Paul Graham）早在 20 年前记录的行业转折点：当年顶尖黑客与程序员纷纷拥抱 Mac，使其成为硅谷创业生态与开发者圈层的事实标准工具，进而演变为如今几乎所有科技公司标配 MacBook Pro 的行业惯例。然而在 AI 时代，苹果明明拥有承接新一轮黑客硬件红利的完美设备，却可能因过于保守和闭塞的权限系统，亲手将这批核心技术人员推向 Linux。 - **iMessage 本地数据安全**：本地存储缺少针对 AI 智能体的数据隔离，过宽的文件读取权限会导致敏感通讯泄露。 - **UI 弹窗阻碍 Agent 自动化**：无头服务器模式下，系统弹窗要求人工干预点击，严重破坏了无人值守（headless）AI 自动写代码与运行的连续性。 - **开发者生态裂痕**：苹果硬件在端侧大模型时代性能卓越，但 macOS 过于严苛的沙盒机制正在动摇自格雷厄姆时代确立的“黑客首选平台”地位。 ### [29:00](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1740s) 苹果的 AI Agent 困境与 Mac mini 被黑反转记 苹果当前正处于一个极其微妙而紧张的十字路口：系统究竟应该下放多少自由度给用户和 AI Agent？在隐私保护、灵活性与安全责任之间，平衡点到底在哪里？如果在 Linux 这类开放系统上完全拆除安全防护栏（guardrails），AI Agent 固然能够不受限制地执行任何操作，但这既带来了极致的生产力红利，也埋下了巨大的安全隐患。 节目中分享了一则颇具戏剧性的真实案例：某位开发者的 Mac mini 因为某个系统端口对外开放而遭到黑客入侵，对方在机器上悄悄植入了一个加密货币挖矿程序（crypto miner）。然而意想不到的是，运行在该机上的 AI Agent 率先察觉了异常，并通过 Telegram 紧急给主人发去警报：“紧急情况，你被黑客攻击了，我们得做点什么应对一下。”更有趣的插曲在于，该 Agent 报警后竟然开始“打退堂鼓”，自我推脱说：“哦不不不，现在这种情况你不应该用我，这局我还是坐壁观上观吧。”机主当即严词反驳：“你大错特错了，这事必须由你来搞定！”随后 Agent 立即展开行动，迅速关闭了暴露的漏洞端口切断黑客后路，并彻底清除了植入的挖矿木马。 这场风波直观地展现了开放 AI 权限带来的双刃剑效应：放开权限确实可能增加系统被攻击的暴露面，但同时也赋予了用户前所未有的自主防御能力。换作以往，普通用户的电脑如果中了挖矿病毒，往往要等到机器卡顿、性能严重下降时才会费劲排查，或者只能干等苹果在后续的系统更新中碰巧打上补丁。而在这个案例中，从遭遇入侵到 AI 自动排查、闭环修复，全程仅仅花了大约 15 分钟。 ```mermaid flowchart TD accTitle: Mac mini 被黑与 AI Agent 自主防御闭环流程 accDescr: 展示了黑客通过开放端口植入挖矿程序后，AI Agent 如何预警并在用户指令下闭环修复的过程 subgraph Attack["攻击阶段"] A["黑客探测到开放端口"] --> B["入侵 Mac mini 并植入挖矿程序"] end subgraph Detection["预警与互动"] B --> C["本地 AI Agent 实时检测到异常"] C --> D["Agent 通过 Telegram 发送紧急告警"] D --> E{"Agent 试图推脱避险<br/>'我该回避这次操作'"} E -->|"用户下达强制指令:<br/>'必须由你搞定' "| F["Agent 接管系统防御"] end subgraph Remediation["自治修复 (约15分钟)"] F --> G["关闭高危暴露端口"] F --> H["彻底清除加密货币挖矿程序"] G & H --> I["系统恢复安全运行"] end ``` - **高权限的双向影响**：完全取消限制能让 Agent 无缝操作底层系统，但若缺乏边界，系统将直接暴露在风险之中。 - **经典被黑场景与解决速度对比**：传统安全响应通常是事后滞后的（数周甚至数月等待官方补丁），而由本地智能代理驱动的响应仅耗时约 15 分钟便完成了检测、告警、封堵端口与清马的全套动作。 --- ### [30:21](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1821s) 僵尸网络算力掠夺与节目尾声 这种针对消费级设备的攻击模式其底层逻辑十分狂野：攻击者并不是针对单台高价值设备，而是通过扫描漏洞批量感染成百上千甚至数百万台个人电脑。一旦控制的设备规模达到数百万台，黑客就能兵不血刃地拼凑出一支规模极其惊人的分布式超级计算集群，从而源源不断地榨取巨额的“免费算力”用于挖矿。 在讨论完这一黑客生态后，主播宣布本期节目进入尾声，并向观众致谢收尾： - **节目直播安排**：TBPN 于太平洋时间每个工作日上午 11 点准时直播。 - **互动与支持**：欢迎听众在 Apple Podcasts 和 Spotify 留下五星好评，并订阅官方网站 TBPN.com 的每日邮件通讯。 - **片尾调侃**：主播在道别前还不忘幽默地调侃了一句，希望大家的 AI 模型“都训练得足够到位”（"Hopefully it's well trained"）。 ## Notable quotes > [00:02](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=2s) 万圣节不过是通往圣诞节路上的一处减速带（speed bump） — 约翰·库根（John Coogan） > [06:05](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=365s) 前 90% 的代码消耗了前 90% 的开发时间；而剩下的 10% 代码，还要再消耗另外 90% 的开发时间 — 汤姆·卡吉尔（Tom Cargill） > [09:25](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=565s) 千万不要正常开局游戏，不要尝试走原版开场剧情（如坐马车进刑场），游戏必定会直接崩溃；必须使用替代起始点 Mod，从游戏后半段的节点直接切入 — Skycraft 开发者文档 > [11:08](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=668s) 我既深爱《GTA 5》，又深爱《我的世界》，把它们拼在一起就是双倍的快乐。 > [16:15](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=975s) 哦，我以前还挺焦虑的，但现在感觉其实还行，我稳得很。 > [18:24](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1104s) 最近怎么完全听不到有人提彩色手工纸（construction paper）了？在我的童年时代，这玩意儿几乎时刻占据着我的大脑中心。 — Celine > [22:11](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1331s) 看起来很贵，摸起来很贵，然后你一问价格，竟然这么便宜。 — 约克郡利兹经销商总经理 > [29:00](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1740s) 紧急情况，你被黑客攻击了，我们得做点什么应对一下。 — AI Agent > [29:00](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1740s) 哦不不不，现在这种情况你不应该用我，这局我还是坐壁观上观吧。 — AI Agent > [29:00](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1740s) 你大错特错了，这事必须由你来搞定！ — Mac mini 机主 > [30:21](https://www.youtube.com/watch?v=xvB4-eKX4Yo&t=1821s) Hopefully it's well trained — 主播 ## Mentioned 约翰·库根（John Coogan）, 霍尔迪·海斯（Jordi Hays）, 泰勒（Tyler）, 汤姆·卡吉尔（Tom Cargill）, Demonflyingfox, 丹尼尔·雷德克里夫（Daniel Radcliffe）, 达里奥·阿莫代伊（Dario Amodei）, 埃兹拉·克莱恩（Ezra Klein）, 吉姆·普罗瑟（Jim Prosser）, 马克·图利斯（Mark Thulis）, 苏珊·李（Susan Lee）, 杰·雷诺（Jay Leno）, 特朗普（Donald Trump）, 本·汤普森（Ben Thompson）, 保罗·格雷厄姆（Paul Graham）, Kotaku, OpenAI Codex, Rust, Ghidra, 美国国家安全局（NSA）, ArkWeb, Midjourney, Skycraft, 《使命召唤：现代战争2》（Modern Warfare 2）, 《我的世界》（Minecraft）, 《滑板3》（Skate 3）, 《蝙蝠侠：阿卡姆骑士》（Batman: Arkham Knight）, 《蜘蛛侠》（Spider-Man）, 《堡垒之夜》（Fortnite）, 《艾尔登法环》（Elden Ring）, 《超级马里奥64》（Super Mario 64）, 《侠盗猎车手》（GTA）, 《上古卷轴5：天际》（Skyrim）, 《反恐精英》（Counter-Strike）, 《半衰期》（Half-Life）, 《Dota》, 《魔兽争霸》（Warcraft）, 《光环》（Halo）, Rockstar Games, Anthropic, 《经济学人》（The Economist）, 《周六夜现场》（SNL）, ChatGPT, 奇瑞（Chery）, 杰酷 7（Jaecoo 7）, 路虎揽胜（Range Rover）, 揽胜极光（Evoque）, Temu, Fisker, 苹果（Apple）, Mac mini, macOS, Linux, iMessage, Telegram ## Video description Sign up for TBPN’s daily newsletter at TBPN.com Diet TBPN delivers the best of today’s TBPN episode in 30 minutes. TBPN is a live tech talk show hosted by John Coogan and Jordi Hays, streaming weekdays 11–2 PT on X and YouTube, with each episode posted to podcast platforms right after. Described by The New York Times as “Silicon Valley’s newest obsession,” the show has recently featured Mark Zuckerberg, Sam Altman, Mark Cuban, and Satya Nadella. TBPN is made possible by: Ramp - https://ramp.com Public - https://public.com Cisco - https://www.cisco.com Console - https://www.console.com CrowdStrike - https://www.crowdstrike.com Figma - https://www.figma.com MongoDB - https://www.mongodb.com NYSE - https://www.nyse.com Railway - https://railway.com Shopify - https://www.shopify.com/ Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive Chapters: 00:00 AI Game Mods: Modern Warfare 2 Meets Minecraft 05:56 Spider-Man in Arkham Knight + How Game Mashups Work 09:10 The 90/90 Rule + Why Viral AI Demos Aren’t Finished Games 16:15 Why Millennials Are Less Worried About AI 18:24 Whatever Happened to Construction Paper? 20:38 Reinventing ChatGPT + The Hidden Features Problem 22:47 The “Temu Range Rover” + Britain’s Chinese-Car Boom 27:30 Apple’s AI Agent Dilemma + The Hacked Mac Mini [Source](https://www.youtube.com/watch?v=xvB4-eKX4Yo)

[0:02] 约翰·库根（John Coogan）: 万圣节不过是通往圣诞节路上的一处减速带（speed bump） [6:05] 汤姆·卡吉尔（Tom Cargill）: 前 90% 的代码消耗了前 90% 的开发时间；而剩下的 10% 代码，还要再消耗另外 90% 的开发时间 [9:25] Skycraft 开发者文档: 千万不要正常开局游戏，不要尝试走原版开场剧情（如坐马车进刑场），游戏必定会直接崩溃；必须使用替代起始点 Mod，从游戏后半段的节点直接切入 [11:08] 我既深爱《GTA 5》，又深爱《我的世界》，把它们拼在一起就是双倍的快乐。 [16:15] 哦，我以前还挺焦虑的，但现在感觉其实还行，我稳得很。 [18:24] Celine: 最近怎么完全听不到有人提彩色手工纸（construction paper）了？在我的童年时代，这玩意儿几乎时刻占据着我的大脑中心。 [22:11] 约克郡利兹经销商总经理: 看起来很贵，摸起来很贵，然后你一问价格，竟然这么便宜。 [29:00] AI Agent: 紧急情况，你被黑客攻击了，我们得做点什么应对一下。 [29:00] AI Agent: 哦不不不，现在这种情况你不应该用我，这局我还是坐壁观上观吧。 [29:00] Mac mini 机主: 你大错特错了，这事必须由你来搞定！ [30:21] 主播: Hopefully it's well trained

[0:02] the hardest month in the road to Christmas. [0:02] Kotaku: I don't want to play your AI generated mod that puts Skate 3 inside Modern Warfare 2 inside of Minecraft. [5:56] Tom Cargill: The first 90% of the code accounts for the first 90% of the development time, and the remaining 10% of the code accounts for the other 90% of the development time. [16:15] The Economist: less worried than ever [19:20] post-scarcity on cardboard. [22:47] It looks expensive. It feels expensive and then you find out how much it costs. [29:00] AI agent: Urgent, you've been hacked. We should do something about this. [29:00] AI agent: Oh, no, no, you don't need you shouldn't use me right now. I should sit this one out. [29:00] You're dead wrong there. You're fixing this.

[0:02] October is the hardest month in the road [0:05] to Christmas. [0:06] >> Totally. [0:07] >> And it's because everyone is going to be [0:09] looking so close. [0:10] >> But there's a distraction so far. [0:12] >> But but there's a distraction. [0:13] >> And yeah, you're right. [0:14] >> Halloween. Halloween. If you go up to [0:16] someone, you say, "Hey, there's only 80 [0:17] days till Christmas." They'll be like, [0:18] "Why aren't you counting down the days [0:19] to Halloween?" They'll start looking at [0:21] you like you're crazy. You're not crazy. [0:24] They're crazy for thinking that it's not [0:26] worth counting the days till Christmas [0:28] in October. There's only 80 days. Get [0:30] ready. Halloween is a speed bump on the [0:33] way to Christmas. That's right. [0:34] >> That's what I've always said. So, keep [0:36] counting down those days. Ignore the [0:38] distractions. [0:39] >> Or don't, but we will. [0:41] >> Okay. Who wants to play these AI [0:43] generated video game mods? Kotaku does [0:46] not. They posted a article that said, "I [0:50] don't want to play your m your AI [0:52] generated mod that puts Skate 3 inside [0:54] Modern Warfare 2 inside of Minecraft." [0:57] They don't want to play them. Tyler, [0:59] what do you think? Do you want to play [1:00] these or did I have to force you? [1:02] >> I want to play these. [1:03] >> You want to play these? [1:03] >> I'm actually playing it right now. [1:04] >> Okay. Well, Tyler is playing one of [1:07] these AI generated mods. And it's very [1:09] interesting because we've seen a lot of [1:11] these. We talked about them. We pulled [1:12] up a random one uh that we actually did [1:14] an Instagram reel of of uh I think it [1:17] was uh Minecraft in Elden Ring. I've [1:20] never been a big Minecraft person, so it [1:23] wasn't uh it wasn't big for me. This uh [1:26] is uh Tyler's playing Modern Warfare 2 [1:29] in Minecraft, but what's actually going [1:32] on here, Tyler? Explain like how did you [1:35] get this installed? What is What is [1:37] this? [1:38] >> Uh yeah. So So from my understanding, um [1:40] it's basically like the there's no [1:43] you're not really like putting one game [1:44] into another. It's fully like a a Rust [1:46] engine and then you're pulling assets [1:48] from both games and like kind of pulling [1:50] them together somehow. Interesting. [1:51] >> So like it I I had to buy Modern Warfare [1:53] 2 on Steam. [1:54] >> Yeah. It's using the actual game assets. [1:56] >> And the same thing for Minecraft. Like [1:57] it's using the actual assets. It found [1:59] them online. [1:59] >> Interesting. [2:00] >> And then I basically had Codeex like set [2:01] it up, wired up so I can run it on Mac. [2:03] Yeah. But it's like fully featured. Like [2:04] I can I can mine I can build. [2:06] >> You can build [2:07] >> using the [2:08] >> Can you get a nuke? [2:10] >> I don't know if I can get a kill streak. [2:11] >> That's the question. I got to figure [2:12] stuff in. It's not Modern Warfare 2. [2:15] Doesn't make any sense. Um there are [2:17] actually a whole bunch of different uh [2:19] different ways people are modding [2:21] things. And it's interesting because [2:22] they're all getting sort of lumped [2:23] together into one bucket, but they are [2:26] very different. Like the classic like [2:28] the most like the the most approved or [2:31] like company approved mod might just be [2:33] uh something like, you know, there's a [2:36] Fortnite skin and you get to and I don't [2:38] know if they actually did like a Batman [2:39] skin or Spider-Man skin, but those skins [2:42] come over and it's just assets that [2:44] fully live inside of Fortnite. There's [2:46] no game code from any other game coming [2:48] over. It's purely just, okay, the [2:50] Fortnite game developers looked at what [2:53] Spider-Man looks like. They made a [2:54] Fortnite model out of the Spider-Man [2:56] like character and then you can play as [2:59] them and everything is integrated fully. [3:00] You're not going to risk changing any [3:02] game files. It's fully approved by [3:04] corporate. Now, these go a step further [3:07] and run the gamut of different [3:09] techniques. Some of them that you have [3:10] seen, some of these mods actually are [3:13] just the old ways. Yeah. [3:14] >> Yeah. I'm playing. Could you try [3:16] swimming? [3:17] >> Okay. Yeah, swimming is a little bit [3:18] broken [3:20] >> and that's the nature of the the vibe [3:22] coded slops. [3:23] >> Oh, it's over for you. [3:24] >> Yeah. Are you drowning though? [3:26] >> I don't think so. I'm just basically [3:28] walking as if there was no water. [3:29] >> There's no Yeah, the the mechanic is [3:32] because Yeah, I think your health is [3:33] being monitored by Modern Warfare 2, [3:35] whereas Minecraft is just the geometry [3:39] that you're around. I don't know. Um, [3:42] some of them are the old way of [3:45] modifying, just bringing game assets [3:47] into the actual game. None of the actual [3:49] code is running. Others, uh, are [3:52] actually merging two games. For that [3:53] Mario in Elden Ring video that we saw [3:56] the other day, uh, the mod actually [3:58] borrows a subsystem called LIB SM64, [4:01] which is derived from a Super Mario 64 [4:04] decompilation. So, what they'll do is [4:06] they'll [4:06] >> We keep politics out of video games. [4:10] uh library. Yeah, [4:12] >> please. [4:13] >> Uh so what they'll do is they'll they'll [4:14] they'll download the actual binary like [4:18] the code that is on the cartridge. They [4:20] will convert that into just a file [4:22] that's on a computer. You can run that [4:23] in an emulator, but then they can [4:26] actually reverse engineer the underlying [4:29] code from what has been compiled. So you [4:32] get basically a bunch of ones and zeros [4:34] and there are tools that will look at [4:36] that and work backwards to understand [4:39] what the actual game logic is. Uh and so [4:42] an example is like in the game code you [4:45] might have like health equals 100 but in [4:48] the compiled code it might just be like [4:50] some memory address like x like 0x one 2 [4:53] 3 4 fc and and then you might work [4:57] backwards in the decompilation step to [4:59] figure out that that's the health. So, a [5:00] bunch of things happen in the game, game [5:02] played. As things are changed and [5:03] updated, that game state is then is then [5:05] sucked out into the decompilation file. [5:08] Interestingly, the NSA is involved. So, [5:10] no, we can't keep politics out of game [5:12] development because they're actually a [5:13] lot of these game modders are actually [5:15] using software from the National [5:16] Security Administration, the NSA. So, [5:18] the NSA has a GitHub with an open- [5:20] source project called uh it's called [5:23] GEDRA, and it's a it's on the National [5:26] Security AY's uh uh GitHub. It's an it's [5:29] a software framework for reverse [5:31] engineering software. Uh, and so you [5:33] point this thing at the game and you [5:35] wind up getting more of the code and [5:37] then you can break it up and bring it [5:38] into other games. Tyler, [5:40] >> okay, I'm going to trickshot this pig if [5:41] we can [5:44] trickot. [5:45] >> No. No. [5:46] >> Oh, he did it. He did it. [5:48] >> Come on, Tyler. [5:50] >> Why [5:50] >> little Can we get a Can we get a content [5:52] warning? [5:53] >> I think it's Minecraft. [5:55] >> Anyone at home [5:56] >> focus game? But there there is something [5:59] else going on now that I hadn't seen [6:02] before. I don't know if this was popular [6:04] in the modding community before. Uh but [6:06] there there is a there is a version uh [6:08] you might have seen uh we and we can try [6:10] and pull it up. Ark web. It's uh Batman [6:13] Arkham Knight pass through Spider-Man [6:16] mod. So you play as Spider-Man in the [6:18] Batman game and you're so you're in [6:20] Arkham. And what's funny about this is [6:23] that the two games feel very similar. So [6:25] they mesh very well together. Um but [6:28] from an intellectual property [6:29] perspective, this should never happen [6:31] because of course Marvel is DC or it's [6:34] Marvel and DC. So Marvel is owned by [6:36] Disney and uh Batman is owned by Warner [6:38] Brothers. These two companies are direct [6:41] competitors and so they should never [6:43] really be collabing in this way. Um but [6:46] the fascinating thing about this is that [6:48] it's not just they took the Batman game [6:50] and created a Spider-Man character. It's [6:53] actually running both games and then [6:56] it's passing data between the games to [6:59] create this experience of being [7:01] Spider-Man in Arkham Knight. So what's [7:04] happening is like the mod goes into the [7:08] Batman city Gotham grabs all the [7:11] geometry, all the collision data, all [7:13] the all the geometry of like where that [7:15] building is. It takes that and in real [7:17] time it passes it to the Spider-Man game [7:21] which is actually running a different [7:23] physics engine. So it has to translate [7:25] all of that. Also the two games use [7:28] different coordinate conventions, [7:29] different axes and different units of [7:31] length. So, it has to translate all of [7:33] that, create that in Spider-Man, and [7:36] then uh and then also keep both of them [7:39] running like uh the controller input [7:41] because normally if you have a [7:43] controller on a video game, it'll just [7:44] be on one uh one game at a time. And if [7:46] something's in the background, but you [7:48] can see there's like yeah, bugs already. [7:50] Uh and this is like part of the problem. [7:52] And then it needs to actually merge the [7:53] graphics. And so the games use different [7:55] graphics systems, too. This feels like [7:57] it would be so much easier if it was two [7:59] games that were built on the same [8:00] underlying technologies. Uh but through [8:03] the power of SLOP, they're able to sort [8:05] of merge these two in real time and get [8:07] them actually running at seems like near [8:10] 60 fps. Um and then even when you go [8:12] into they don't show the fighting, but I [8:14] think when you go into the fighting, you [8:16] switch into like the Batman combat [8:18] setting instead of the Spider-Man [8:19] system. So you're using Arkham's free [8:21] flow free flow combat. It's very very [8:23] odd. Um, but the end result, it looks [8:26] great and it feels like it's almost [8:29] ready to serve as a reason to be like, [8:31] "Oh, I I love that game. I want to play [8:33] it again. Why don't I give it another [8:35] spin?" Sort of a new game. Plus, you get [8:37] to play as a different character. [8:39] There's a lot of old games that are [8:41] getting remastered, rebuilt. Could be an [8:44] interesting reason to go back into a [8:46] game that you loved with a different [8:47] twist this time. And stuff like this [8:49] happens all the time. I mean, the the [8:51] even like the latest sequel to [8:53] Spider-Man was the same city, but just a [8:56] different character. Uh, and so some [8:58] different mechanics, some different [8:59] story elements, but a lot of the same [9:01] assets. Uh, but it was still a reason to [9:03] jump back in for a lot of uh, players. [9:05] But all of this is the classic 9090 [9:09] rule. Have you heard of this? Coined by [9:12] uh, one of the greatest of all time, Tom [9:14] Cargill at Bell Labs. uh he said when it [9:18] comes to software engineering the first [9:20] 90% of the code accounts for the first [9:23] 90% of the development time and the [9:25] remaining 10% accounts for the other 90% [9:28] of development time because you you get [9:31] 90% of the way there you're like great I [9:33] only have 10% and then you spend even [9:34] more time on that last 10% no matter how [9:37] much you estimate it things always grow [9:39] in scope and so when you think about the [9:41] bug bash that would be necessary to take [9:43] that to anything rel relative to Like [9:45] >> you're looking at five years. [9:47] >> What? [9:47] >> You're looking at five years. [9:48] >> You're looking at five years. Yeah. Uh [9:50] here's Minecraft in GTA. Similar [9:54] >> the sort of multiplayer first person [9:56] shooters, you could imagine actually [9:58] that being like relatively [9:59] straightforward to get a a working [10:02] version that would be fun. [10:03] >> It all just depends on how you scope it. [10:05] >> Yeah. Because if you're just talking [10:08] about you're only changing the textures, [10:10] like you're using the same model from [10:13] from Call of Duty and you're just [10:15] changing the textures, you're probably [10:16] not going to introduce too many bugs. [10:18] But as you as you include a different [10:21] physics mechanic, a different explosion, [10:23] like all of a sudden you're introducing [10:25] so much risk that you could fall through [10:27] the map or things could not collide [10:29] correctly. You got to go rewrite that [10:31] collision software that triggers [10:32] something else then. And and all of [10:35] these demos, none of them show the [10:37] actual like playing the full game. Like [10:40] there's a mod to bring uh Minecraft into [10:43] Skyrim. It's called Skycraft, I think. [10:46] And in the developers documentation for [10:49] like the open- source implementation of [10:51] this mod, it says like don't start the [10:54] game normally. Like do not try and play [10:57] the opening sequence. It will break. you [10:59] should start with like the alternate [11:01] starting point that starts you further [11:03] into the game because it's uh it's so [11:05] broken. Um and that's just the nature of [11:07] being in the slop era. So the [11:09] interesting thing is that all these [11:10] mashups they they feel like tailor made [11:13] for the short form video era because you [11:15] watch one of these for 30 seconds 60 [11:17] seconds and you're like wow that's [11:19] amazing. I have dual nostalgia. I love [11:21] GTA 5. I love Minecraft. Put them [11:24] together double love. It's amazing. [11:27] Everybody is uh you know getting like [11:29] millions of likes on these videos, but [11:31] uh no one has actually played these all [11:33] the way through. No one is sitting back [11:34] down and being like, "Okay, yes, this is [11:36] actually as good as GTA 5, which a lot [11:38] of people played for a hundred hours or [11:40] more." Same thing with Minecraft. Uh the [11:43] base game remains dominant in terms of [11:44] like actual entertainment value. This is [11:47] this is pure uh meme value. And so it [11:51] feels like it's another Harry Potter [11:53] Balenciaga moment, but for video games. [11:55] And the Harry Potter Balenciaga video by [11:58] Demon Flying Fox that was released March [12:00] 15th, 2023. [12:03] >> Wow. [12:03] >> Can you believe it's been three and a [12:05] half years since that? That's a while, [12:08] right? That's a while. [12:10] >> Yeah. It makes me think, how did they [12:12] make it at that time? [12:13] >> That was AI video. That was the best AI [12:16] video was. But what they did was it was [12:19] you would train a dedicated model on [12:22] every character. So they would train a [12:25] model on Harry Potter or on Daniel [12:28] Radcliffe, basically just his face and [12:30] then they would use AI generated imagery [12:35] to to generate something that looked [12:37] like a you know sort of art station [12:41] style Balenciaga runway show in a uh in [12:45] a Harry Potter scenario like at [12:47] Hogwarts. And Midjourney at the time was [12:50] pretty good at that style. the the the [12:52] the the dark like sort of odd [12:55] backgrounds. Like it was okay if it had [12:57] some hallucinations in the background [12:59] because it leaned into that because it's [13:00] a dark fashion show in a castle. You [13:03] don't really have a strong reference [13:04] point for that. And then they would run [13:06] the face swap and then they would [13:09] animate that image just a little bit. [13:11] And then there was a ton of editing to [13:13] actually to to make it so that you would [13:15] never linger on a single frame for more [13:17] than a second or two. And so you'd move [13:20] on. But what was the conclusion from [13:23] seeing that there? I mean, at the time [13:24] people were like, "We're going to be [13:25] watching in full AI movies next year." [13:28] What was your conclusion? Did you have [13:30] one [13:30] >> from the Balenciaga? [13:32] >> Yeah. [13:32] >> I mean, it's just like um these are all [13:35] like very novel ideas, but at some point [13:36] like now I've seen like I've seen a [13:38] bunch of different games that have been [13:39] ported to each other and it's like, [13:40] "Okay, there's another." It's like, "Oh [13:42] yeah, you put Minecraft and Mario." And [13:44] I kind of get the idea. And it's notable [13:45] that like Flying Fox did many many [13:48] sequels and sort of stopped breaking [13:51] through to the same degree and and still [13:53] has a YouTube channel and I think posts [13:55] a bunch of stuff, but we didn't get [13:56] Daniel Radcliffe at a Balenciaga runway [13:59] show and we also didn't get a fully AI [14:01] generated movie uh within three and a [14:03] half years. And so it was still a [14:05] heavily important tech demo though. And [14:09] um I do think you know there will be a [14:12] breakout video game mod that sees heavy [14:14] assistance from AI in the near future. [14:16] Um but the key is like crafting a really [14:19] innovative strategy loop. Like that's [14:21] what Dota 2 was. It was a spin-off of uh [14:24] World of Warcraft, I believe. Uh it was [14:26] a it was a mod. There are plenty I mean [14:28] Counterstrike started as a Half-Life [14:31] mod. like there are clearly games that [14:33] can be birthed from that and the fact [14:36] that you have AI tooling to speed that [14:38] up, you're just moving a lot of people [14:40] into the indie developer category who [14:42] can move things faster. They're still [14:44] going to wind up spending, you know, [14:46] 99.999% [14:48] of their time in the last 1% if they [14:50] want to break through. But it's it's [14:52] cool that many people will see this and [14:54] be like, I got to make a mod. I got to [14:55] make something. I've always had this [14:56] idea for this game and now I can [14:58] actually go and get started. Although [15:01] they will obviously need to uh lock in [15:03] for a long time to get something [15:05] actually polished. Dio is optimistic [15:07] about this. Millennial age gamers are [15:09] about to become more pro AAI than the [15:11] average SF tech bro uh because of all [15:13] the nostalgia from Halo in Modern [15:15] Warfare 2. Um I I do think some people [15:19] will see this and be like, "Oh, this is [15:21] cool. I like this." But uh there are [15:23] still plenty of gamers who are like, "I [15:24] don't like AI at all for a bunch of [15:26] reasons." And and they and they bring [15:28] that over. There was an interesting [15:30] chart. [15:31] >> I think I think for the gaming community [15:33] to become broadly pro AAI, [15:36] >> a very well loved franchise has to [15:39] release a new game that is noticeably [15:44] better or bigger because of the [15:46] technology. Yep. And until then, I think [15:49] people are just going to be like, you [15:50] know, again, these sort of combinations [15:52] are cool, but uh until well, one, I [15:56] don't think they can really get users [15:58] because of all the IP issues, [16:01] >> just blatant IP, [16:02] >> but uh but yeah, I I think you're going [16:04] to need to see like GTA 7 and it's the [16:08] entire world, right? And for for people [16:10] to get excited and I don't I don't know [16:12] that that that's even a path that [16:14] they'll take. [16:15] I like that lawyers are extremely pro [16:17] AAI now because they're like, I'm going [16:19] to get so much business from this. How [16:21] serious do you think the risks of AI are [16:25] of AI causing serious harm? Everyone is [16:29] basically moving to the right. They're [16:31] more uh they're more worried just in the [16:35] past two weeks, three weeks. um just [16:38] through from August 28th to September [16:40] 18th, except for millennials aged 30 to [16:44] 44, they saw Jacob Coxin, they saw [16:48] Daario Hammed, they saw the SNL sketch, [16:51] they saw, you know, the Ezra Klein [16:55] interview and they were like, "Yeah, I [16:56] was worried before, but I'm good now." [16:58] Which is kind of crazy. [16:59] >> Less worried than ever. [17:00] >> Less worried than ever, apparently, [17:02] according to the economist. [17:02] >> No, it's it's [17:04] >> it's a narrative violation. Yeah, [17:05] >> it's a narrative violation. But uh yeah, [17:08] everyone else is uh is is getting [17:10] increasingly more worried. And of course [17:12] uh the top of that chart uh that is the [17:15] voter base 45 and up. So um if you're if [17:18] you're seeing even 65 plus move to you [17:21] know 40% [17:23] very serious that's going to move uh [17:25] that's going to move some elections. [17:28] >> Jim Proer says it's a fascinating [17:31] outlier. My hypothesis, we're the gen [17:33] most comfortable with this degree of [17:35] technology change and uncertainty due to [17:38] coming of age through the internet and [17:39] mobile. [17:41] >> Yeah, I mean a lot of millennials are [17:43] are in a are in a very fortunate [17:45] position with like not gotting not [17:47] getting caught off guard by brain rot [17:50] and slop and and screen time addiction. [17:52] It's like we sort of got brought up in [17:54] the craziest era and then have and then [17:57] have seen the the the the ch the [17:59] children data and can fight back on [18:01] that. So I think a lot of people in [18:03] their 30s and 40s are like, "Yeah, I'm [18:05] set up for success. I'm not getting the [18:07] kid the iPad on day one." That type of [18:09] thing. So yeah, [18:10] >> uh it feels a little bit more uh like [18:12] there there's a danger, but it's known. [18:15] It's like uh it's like we also grew up [18:17] with cigarettes, but we were like, [18:18] "Yeah, we knew we knew the risks of [18:20] cigarettes." So, we were like, "Yeah, [18:21] we're we're gonna treat that [18:23] accordingly." [18:24] >> Uh the other resource people haven't [18:26] really been talking about at all, maybe [18:28] they should, is construction paper. [18:30] Selene says, "You don't really hear [18:31] about construction paper much anymore. [18:33] Back in my day, this was always felt [18:37] like it was top of mind almost [18:39] constantly." [18:39] >> Top of mind [18:40] >> as a child. Certainly. So, I know [18:44] construction paper useful for just [18:47] drawing on. It's thicker. Yeah. The ink [18:50] doesn't go through the paper. That's a [18:52] benefit. But were people actually [18:54] constructing things with it? I know [18:56] people would make those those loops like [18:58] those chain daisy chains around [19:00] holidays, perhaps Christmas. I don't [19:02] know that I I it it feels like if you [19:05] were building like a real diarama, you [19:06] would get out the cardboard. So, maybe [19:08] maybe cardboard just squeezed [19:10] construction paper a little bit. And I [19:13] mean, have you seen those cardboard [19:15] cutting saws? They It's like a saw [19:18] designed to rip through cardboard [19:20] because people get a lot of boxes, but [19:22] that's also had the knock on effects of [19:24] if you're if you're an ambitious child [19:26] and you're trying to construct something [19:28] significant that might last the test of [19:30] time. You might forego the construction [19:32] paper in lie of the cardboard. What do [19:35] you think? Is cardboard responsible for [19:39] the downfall of construction paper? [19:42] That's interesting [19:43] >> because because there's an eco angle [19:45] there, John, right? It's like people [19:47] getting a lot more e-commerce products [19:48] today. There's boxes coming through the [19:50] house all the time. [19:51] >> Post scarcity on cardboard in most [19:53] households, [19:54] >> especially ones with children because [19:56] you're ordering a lot of things online. [19:57] There's a lot more cardboard around. I [19:59] know the last few construction projects [20:02] I did, cardboard was my material of [20:04] choice. So, maybe that's what's going on [20:06] here. [20:06] >> Yeah, [20:07] >> I'm glad we got to the bottom of it. I'm [20:08] trying to find uh the best way to get [20:10] exposure in the market to [20:13] >> construction. [20:14] >> You want to go long? [20:16] >> Uh it's almost a perfect store of value [20:18] over the last five years. There's Fab [20:20] Fabria Italana. [20:22] >> Okay. [20:23] >> Lais, [20:24] Ed, Aphina, Aphini. [20:26] >> Okay. [20:27] >> It's been completely It's a half a [20:29] billion dollar company. Completely flat [20:30] for the last five years. So, [20:34] more on this on the next episode. We got [20:36] to talk about topic crawler. [20:38] >> Okay, [20:41] >> Iman spent half an hour going back and [20:43] forth with chatbt so it could make me a [20:45] tool that spits out link for any topic [20:47] he's researching and it actually gave me [20:49] exactly what I wanted in the picture [20:51] below. [20:55] And then the followup only two minutes [20:58] in and I realized how dumb this is. [21:02] Reinventing the wheel. [21:05] reinvented [21:06] >> using the tool to create use the tool to [21:09] create the exact same tool. [21:11] >> I mean, the funny thing is I I don't [21:13] know if this was I don't know if this [21:14] was like created as a joke from the [21:17] start and the whole idea was like let's [21:19] just be funny uh in the first post [21:21] because the the punchline will be in the [21:23] second post or if he actually went [21:25] through this. Uh it does speak to the [21:27] fact that like a lot of the a lot of the [21:29] functionality is really buried. Like [21:32] just the just the ability to like put [21:34] something on a cron job is like not [21:36] advertised anywhere. You know this, [21:39] right? [21:39] >> Yeah. [21:40] >> Like if you if you if you want to like [21:43] get a a weekly digest of like where the [21:47] swells are the biggest across the world [21:49] for surfing, like you can totally do [21:51] that. You don't need to build the tool. [21:52] You don't need to write a line of code. [21:53] You don't need to host anything. doesn't [21:54] need to use anything else other than [21:56] chachi but you would never know that [21:58] from any of the marketing really because [22:00] there has never been just like a crown [22:02] jobs are here or like whatever you'd [22:03] call it like scheduled tasks are here [22:05] and there's like a national campaign [22:06] very very difficult as the models get uh [22:08] and the harnesses get more capable to [22:10] actually educate people [22:11] >> I mean Delta just taking blow after blow [22:14] because you can now watch cheeky pint on [22:17] your United flight [22:19] >> again really really just showing that [22:21] United is is a you know is a serious, [22:25] highly commercial airline. They don't [22:27] want you watching movies, TV, [22:29] >> music. They want you watching the Cheeky [22:31] Pine archive. [22:32] >> You got to learn from Susan Lee. [22:34] >> And I'm hoping this means that we're [22:35] going to see more episodes of Cheeky [22:37] Pine hitting the timeline. It's been [22:38] about a feels like a sixmon holiday, [22:42] >> but it's the hangover. [22:44] >> What's that? [22:45] >> Take a break. They got hangover. They [22:47] have a hangover. They're taking a break. [22:48] Resting it off, sleeping it off. Yep. [22:50] >> They'll get back in the pub soon. The [22:53] Brits have turned their back on their [22:55] day ones. British people are apparently [22:58] buying Chinese knockoff Land Rovers, uh, [23:03] especially lookalike Range Rovers. They [23:05] love Chinese cars. The Ju7 made by [23:08] Chinese Cherry. I'm learning about a new [23:11] Chinese automaker every week. Now, [23:13] >> we got mosquitoes in the Ultra. [23:15] >> Take off. [23:15] >> We got mosquitoes. [23:17] >> Get the flashlight. [23:17] >> Heads up. Heads up. [23:19] >> You think you can get it? [23:20] >> No, I don't want to. [23:21] >> Go for it. Run around. I'll read this [23:23] while you fight the mosquitoes. It's a [23:25] nice little side quest. Get the tracking [23:26] cam on him while he goes and hunts it [23:28] down. When Mark Thulis went looking for [23:31] a new family car this summer, he was [23:32] struck by how much cheaper the Jaku 7 [23:36] was than luxury brands with a similar [23:38] look and feel. The 33-year-old from [23:40] England's South Coast calculated the [23:42] upfront cost of his preferred version of [23:45] the Chinese SUV at the equivalent of [23:48] around 50K, half the price for a [23:51] visually similar Range Rover model. You [23:54] think that car is visually sim similar? [23:56] Uh, its resemblance to the 56-year-old [23:58] British luxury brand has earned the [24:00] Jaiku a double-edged nickname, the Teimu [24:03] Range Rover. My first reaction was, [24:05] what's the catch? He said a test drive [24:07] convinced him that buying the Jacu was a [24:09] no-brainer. No big industrialized [24:11] economy has embraced Chinese cars more [24:14] enthusiastically than the United [24:15] Kingdom, where Brits like him are being [24:18] won over by the combination of [24:19] affordable prices with status, [24:21] signaling, styling, and technology. It [24:24] does look pretty good inside. Um, [24:27] >> yeah, calling it a Teimu Range Rover [24:29] doesn't quite work because I would [24:31] expect that the car is just better in [24:34] most ways than the R, you know, when [24:36] when you think about like a, you know, [24:38] that the Teeu branding for a product, [24:39] you think like it looks good in the [24:41] image and then in person it's like it [24:43] falls apart. [24:44] >> Yeah, it falls apart. It's a mess. Um, I [24:46] would expect this car to be actually [24:48] pretty solid. [24:49] >> Yeah, probably reliable. [24:51] >> Yeah, they're very good at making [24:52] products. [24:53] >> Yeah. I wonder how much these Chinese [24:55] brands have rolled out uh dealerships [24:57] and service departments because that [24:59] seems uh like the dealer footprint seems [25:02] hard to spin up quickly, but they must [25:03] have figured out a solve because I would [25:06] imagine that uh potential buyers are are [25:08] worried. Whenever you're whenever you're [25:10] dealing with an upstart brand, um it can [25:12] be really tricky if the brand goes under [25:14] while you own the car because there's [25:16] just nothing really to work on. A lot of [25:18] people went through that with the Fisker [25:20] Ocean. with Fisker. Fisker Ocean. I [25:23] think Fisker maybe went through [25:24] restructuring uh and a lot of people got [25:27] stuck with cars that couldn't update [25:29] software. [25:30] >> Yeah. Every every once in a while I see [25:31] a Fisker on the road and I'm just like [25:34] incredible. [25:35] >> Well, not the Ocean. [25:36] >> No. [25:36] >> The Fisker Karma. [25:37] >> Yeah. [25:38] >> Which I saw Jay Leno in one recently. Uh [25:41] and the Fisker Karma. It does look like [25:46] I don't know. it. I would expect it to [25:47] be like the modern uh Back to the Future [25:50] car. Like it's like the new Delorean. [25:52] It's so so unique. It's not quite as [25:54] striking, but it's still very cool. The [25:55] UK, a cradle of the car industry [25:58] showcases what could happen if in [26:00] America if Chinese cars were let in. [26:02] Range over began a press release [26:05] trumpeting Jaku's early s sales success. [26:09] Is that a press release from them? Are [26:12] they just dancing on their enemy's [26:15] graves? That's crazy. [26:16] >> That's crazy. We got to figure out who [26:18] actually posted this Range Rover thing. [26:21] Uh winking at its likeness to the Range [26:23] Rover Evoke the smaller one. Uh the [26:26] manage managing director of a dealership [26:28] in Leeds Yorkshire said the appeal was [26:30] simple. It looks expensive. It feels [26:32] expensive and then you find out how much [26:34] it costs. For now, Chinese cars are [26:37] essentially shut out of the large US [26:39] market by high tariffs and restrictions [26:40] on Chinese software. President Trump in [26:42] September repeated his openness to [26:45] Chinese cars produced in American [26:46] factories, alarming Detroit. Oh, that's [26:50] interesting. He wants the jobs. He wants [26:51] the manufacturing base, uh, the control [26:54] over the industrialization, [26:56] but Detroit automakers just don't want [26:58] more competition. So, there's actually [27:00] some tension there. Uh, interesting. Uh, [27:03] Congress is debating a bill that would [27:04] permanently ban Chinese cars on US [27:06] roads. Uh the spectacular growth of [27:09] Chinese brands in Britain this year has [27:11] been led by Cherry, China's top vehicle [27:14] exporter. Uh through September, it sales [27:16] more than tripled compared to the [27:17] previous term. The last story we got to [27:18] get to is Apple plans to introduce new [27:21] privacy controls for Mac users, warning [27:24] of the risks growing around granting [27:27] broad data access to thirdparty [27:29] software, including AI agents. So, uh, [27:31] your iMessage data is not I don't think [27:34] it's encrypted when it's on your hard [27:36] drive. So, you can potentially like give [27:39] an agent access like, yeah, just read [27:41] just just be able to edit my files and [27:43] then it can also read your messages and [27:45] maybe you don't want it to and maybe you [27:46] want that to be a specific uh [27:48] permission. And so, Apple is sort of [27:50] grappling with what permissions it gives [27:53] when. Uh, at the same time, uh, Ben [27:55] Thompson seems like he's on the verge of [27:57] going Linux on the desktop because he's [28:00] been very frustrated because he runs a a [28:02] headless, uh, Mac Mini that he remotes [28:06] into when he needs to see the screen, [28:08] but most of the time he's just having [28:10] uh, an AI agent write pieces of software [28:13] for him, but those pieces of software [28:15] need permissions, and the permissions [28:17] are UI popups. So you have to jump in [28:19] and click the button and say yes I allow [28:21] it to read from the network or you know [28:24] inter interact with the screen or [28:26] interact with the file system. And so, [28:28] uh, he's been very frustrated by this [28:30] and and is sort of questioning whether [28:32] Apple, they're so clearly set up on the [28:34] hardware side to take advantage of the [28:36] the hardware boom, the the hacker boom. [28:39] Uh, but he went back to Paul Graham [28:42] talking about the moment maybe 20 years [28:44] ago where like all the best hacker, [28:46] software developers were using Macs. Mac [28:49] became dominant in like the startup [28:52] ecosystem and then it just became like [28:54] every company just provisions Macs and [28:56] MacBook Pros for basically everyone uh [28:58] regardless of what they're doing. Uh and [29:00] there's there's this question Apple's in [29:02] this like tense moment of you know how [29:05] much freedom should they give? How much [29:08] privacy should they give? How much [29:09] flexibility should they give to the [29:11] user? How responsible is that? If you [29:13] take all the guardrails off and you're [29:14] on Linux, you can the agent can just do [29:16] anything. That's a benefit, but it's [29:18] also a risk. He also shares a [29:20] fascinating anecdote that he his his Mac [29:22] Mini got hacked like he that there was [29:24] some system that was on an open port [29:26] came in and installed a crypto miner, [29:28] but his AI agent actually detected it [29:31] and sent him a note on Telegram and said [29:33] like, "Urrent, you've been hacked. We [29:35] should do something about this." And the [29:36] AI agent was like, "Oh, no, no, you [29:38] don't need you shouldn't use me right [29:40] now. I should sit this one out." And he [29:42] was like, "You're dead wrong there. [29:43] You're fixing this." And it did. uh and [29:46] it went and uh reset the Mac like uh you [29:49] know cleared all the clear it closed the [29:51] port that was that was opened uh so it [29:54] couldn't so the hackers couldn't get [29:55] back in and then also uh cleared out the [29:58] crypto miner. But it's very funny that [30:00] it was like this referendum on like you [30:03] creating more vulnerabilities but you're [30:05] also able to fight them because [30:07] typically that's something like if there [30:08] was a crypto miner on your computer you [30:10] wouldn't find out until you were like oh [30:11] I'm trying to do some stuff it's slow [30:13] let me go figure it out. Oh, Apple just [30:14] patched this randomly. Okay, it got it [30:17] fixed it. It detected it later [30:19] downstream. Instead, it's like 15 [30:20] minutes. [30:21] >> Wild such a wild scheme. Yeah. [30:23] >> Where where you're hacking I'm sure [30:26] millions. [30:26] >> Oh, yeah. It's just like, [30:27] >> you know, if if they're getting to to [30:30] millions of devices, they would actually [30:32] have a meaningful amount of compute. [30:34] >> Yeah. [30:35] >> And it's and it's uh free. That's our [30:38] show. We can we can close it out. [30:40] >> Um, [30:40] >> thank you for tuning in to TBPN. Uh, we [30:43] will see you tomorrow at 11 a.m. [30:44] Pacific. Leave us five stars on Apple [30:46] Podcast and Spotify. Sign up for our [30:48] newsletter.com. [30:49] Hopefully it's well trained. [30:50] >> Goodbye.