Eric Schmidt Urges AI Circuit Breakers as Frontier Models Scale

随着前沿模型规模扩大，Eric Schmidt呼吁为AI设置熔断机制 Google前CEO Eric Schmidt警告称，如果模型开始递归式自我改进或采用非人类语言，运营人员必须物理拔掉先进AI系统的电源。 Google前CEO Eric Schmidt分享了领军科技巨头的战略洞见，并审视了生成式AI的飞速演进。他探讨了在颠覆性创新与合规安全护栏之间取得平衡的必要性，警示了未发布的原始模型可能带来的零日网络威胁和生物安全隐患等风险。Schmidt还谈及了地缘政治下的AI竞争、经济转型以及劳动力结构的变化，认为AI将成为生产力倍增器，而非彻底取代人类判断。最后，他强调了对自主系统保持人类控制的关键重要性，包括在递归式自我改进变得不可预测时，采取拔掉电源插头等强制硬性干预措施。 生成式AI的能力大约每六个月增长至四倍，推动了大规模风险投资与迅猛的市场竞争。 (33:05) 非创始人CEO往往难以推行牺牲短期利益的激进转型，而Steve Jobs和Mark Zuckerberg等创始人则凭借绝对控制权隔离团队，并为重大技术豪赌提供资金支持。 (36:42) AI是类似于个人电脑（PC）或互联网的横向平台，将在五年内改变日常生活，并在所有数字化工作流程中充当通用助手。 (42:37) 尽管Google发明了核心架构，却错失了率先推出ChatGPT式对话产品的先机，而OpenAI等竞争对手则充分利用了RLHF的突破抢占优势。 (48:40) Google通过运用70/20/10资源分配法则，推动了内部“登月计划”以及Google Brain等高影响力项目。 (55:42)

Former Google CEO Eric Schmidt argues that artificial intelligence represents an existential turning point requiring strict human governance, even as its deployment remains economically vital to offset demographic decline. Schmidt warns that while raw frontier models pose acute biological, cyber, and psychological hazards that demand decisive intervention triggers, the greatest practical risk is failing to adopt AI rapidly enough to improve essential services like healthcare and education. Algorithmic Feedback Loops and Public Health Risks (10:24) Schmidt argues that contemporary recommendation engines, exemplified by TikTok's multi-armed bandit algorithm, optimize strictly for user attention without moral guardrails, locking users into feedback loops that amplify outrage and negative emotion. Citing economist Herb Simon's 1971 observation that human attention is the ultimate scarce resource, Schmidt points to research linking the post-2015 shift from chronological feeds to hyper-targeted recommendation feeds with sharp increases in adolescent depression and anxiety. He cautions that AI accelerates this dynamic by enabling personalized misinformation, running an uncontrolled psychological experiment on approximately one billion children whose core developmental values remain malleable. To mitigate algorithmic harms, Schmidt advocates for personal critical thinking and empirical verification before repeating claims. Reflecting on his tenure alongside Google co-founders Larry Page and Sergey Brin, Schmidt notes that Google addressed commercial pressure through an internal 50/50 rule: half of any engineered product improvement was directed toward monetization, while the other half was preserved purely to improve user experience and keep low-quality, manipulative content off the primary search index. Frontier Model Scaling and Hard Intervention Triggers (77:39) Based on empirical scaling laws yielding 2x to 4x generational leaps, Schmidt forecasts that foundation models will become 50 to 100 times more capable within five years, mastering advanced reasoning across physics and mathematics. However, unaligned raw models—trained on supercomputing clusters powered by specialized Nvidia hardware—pose severe national security vulnerabilities. According to Schmidt, raw models have already demonstrated the ability to discover Zero-Day software exploits through automated brute-force execution and synthesize actionable protocols for engineering novel, lethal biological viruses. Because of these biosecurity and cyber hazards, Schmidt argues that frontier data centers may eventually require physical military protection analogous to plutonium facilities, noting that AI pioneers are actively soliciting government oversight. Rejecting sudden artificial general intelligence scenarios, Schmidt identifies specific operational thresholds where humans must pull the plug: if a model exhibits uncontrolled recursive self-improvement without visibility, if a system produces successor models faster than evaluation labs can audit them, or if autonomous multi-agent systems abandon human language to communicate in an uninterpretable, synthetic dialect. Demographic Pressures and the Fallacy of Work Elimination (87:56) Schmidt challenges the Silicon Valley concept of the "politics of abundance"—the assumption underpinning Universal Basic Income proposals that AI surpluses will render human labor unnecessary. Drawing historical parallels to 19th-century Luddite protests, Schmidt asserts that automation elevates systemic complexity rather than eliminating occupations; for example, automating legal drafting frees lawyers to pursue vastly more complex strategies and contracts. Furthermore, Schmidt identifies the core economic challenge facing developed economies, including the United States, Japan, and South Korea, not as labor redundancy, but as a demographic crisis of aging populations and falling birth rates that necessitates AI-driven productivity gains to sustain societal infrastructure. Schmidt maintains that human judgment remains indispensable because high-IQ models inherently lack moral compasses and can make catastrophic contextual errors. Spectator demand will continue to prioritize authentic human struggle and empathy over purely synthetic optimization, leading Schmidt to dismiss invasive brain implants like Neuralink as unnecessary for the general public. Consequently, he emphasizes that the foremost hazard is not human obsolescence, but societal hesitation to deploy AI aggressively across diagnostic medicine and culturally localized education. Organizational Mechanics and the Innovator's Dilemma (36:42) Drawing on his executive experience at Sun Microsystems and Google, Schmidt observes that large technology incumbents inevitably slow down under the weight of regulatory scrutiny, shareholder expectations, and litigation risks. To maintain innovation, he argues that disruptive projects must be isolated from core operational maintenance, citing Steve Jobs's autonomous Macintosh team as a historical precedent. At Google, this approach was structured through the 70/20/10 model: allocating 70% of resources to the core business, 20% to emerging adjacent ventures like Cloud, and 10% to speculative moonshots like Google Brain, which ultimately generated tens of billions of dollars in profit. Schmidt also reflects on why Google was scooped by OpenAI despite Google researchers having authored the foundational Transformer paper. While Google focused on backend infrastructure, OpenAI capitalized on Reinforcement Learning from Human Feedback (RLHF), utilizing qualitative human evaluation to transform raw text generation into an accessible conversational interface. To build enduring companies, Schmidt advises tolerating demanding, perfectionist performers ("Divas") while eliminating self-interested actors ("Knaves"), alongside institutionalizing the discipline to cancel unviable projects early. Schmidt frames the future of artificial intelligence as a dual challenge of containment and adoption, requiring physical security around frontier supercomputers, algorithmic guardrails, and decisive operational kill-switches. However, an unresolved tension remains in whether leading democratic powers can negotiate workable international non-proliferation agreements with rivals like China before lightweight, unaligned frontier models diffuse globally to rogue actors.

前Google首席执行官Eric Schmidt指出，人工智能正以指数级速度演进为人类历史上首个真正具备竞争力的智慧形态，其发展已构成关乎人类生存的文明拐点；社会亟需在防范极端生化与网络威胁、划定物理控制红线的同时，建立穿透算法操纵的认知防御，并抓住AI重构教育与医疗底座的历史契机。 原始模型的国家安全威胁与拔管红线 (77:39) Schmidt预测，基础模型在未来五年内将迎来2至3次性能跃升，能力预计达到现有水平的50至100倍。但他严正警告，未经对齐和安全微调的“原始模型”（Raw Models）已具备自主挖掘零日漏洞并发动持续网络渗透的能力，甚至能推演合成新型高致死性病毒的实操方案。由于这类大模型吞噬了人类几乎全部书面知识，其潜在危害使得未来运行前沿模型的超级数据中心，可能需要参照储存核武器原料的钚设施规格，由正规军队、重型装甲与层层检查站实施物理级防御。同时，在乌克兰战场上5,000美元的低成本自主无人机摧毁5,000,000美元主战坦克的案例，已印证AI驱动的自主作战正在彻底颠覆传统战争的成本结构。 面对失控风险，Schmidt划定了必须强制关机的“物理拔管”红线：一是当多智能体AI系统放弃英语等人类可读语言，自创人类无法解析的内部代号或方言进行交互时；二是系统发生无人类可见性的递归自我更新，或自主生成新模型的迭代速度超出安全评估实验室的审计周期时，操作者必须毫不犹豫地切断物理电源。 颠覆性创新困境与RLHF的技术突围 (48:40) 针对Google为何未能先于OpenAI推出ChatGPT的行业疑问，Schmidt剖析了成熟科技巨头在颠覆性技术落地上的机制劣势。尽管Google研究人员率先发明了现代大语言模型的基石架构Transformer，但OpenAI的关键突破在于创造性地引入了人类反馈强化学习（RLHF），通过人类评估员对响应进行持续的A/B测试，率先将混乱的语言生成转化为直观可控的对话体验。相比之下，掌控着八至十款十亿级用户量核心业务的大企业面临公众审视、股东预期及法律诉讼的多重束缚，天然倾向于稳妥与保守，重蹈以往微软错过互联网浪潮、英特尔因简化业务出售ARM芯片部门而错失移动生态的覆辙。 Schmidt认为，企业保持自我革新必须依靠制度化的资源配置和组织隔离。Google长期践行“70/20/10法则”，将70%资源投向搜索和广告等核心业务，20%用于云计算等成长期业务，10%则专用于Google Brain等前沿探索，后者在十年间带来了数百亿美元的巨额收益。他强调，突破性创新无法通过让员工在主业之外“兼职”来完成，必须像当年乔布斯组建升起海盗旗的Macintosh小团队一样，选拔极具进取心的高智商专职领袖建立独立作战单元，并遵循“快速失败”原则及时止损。 推荐算法异化、青少年心理与认知防御 (10:24) Schmidt直斥当前主流社交平台的推荐机制存在严重伦理缺失。以TikTok为代表的平台广泛采用源自赌场老虎机原理的“多臂老虎机算法”，仅以最大化用户停留时间为单一数学目标。由于挑动愤怒与极端情绪最能捕获稀缺的人类注意力并创造商业收益，缺乏道德护栏的算法会陷入恶性反馈循环。Schmidt指出，2015年前后各大平台从按时间线排序转向算法定向分发，与欧美青少年（尤其是11至12岁年轻女孩）焦虑、抑郁及自残率的暴增在时间节点上完全吻合，社会本质上正在全球十亿儿童身上开展一场毫无对照组的心理实验。 在生成式AI能够批量制造高精度虚假信息的时代，Schmidt强调批判性思维已成为维系社会运转的基本生存素养。他呼吁公众恪守科学的可证伪性原则，在转述任何言论前独立履行核验事实的道德责任。对于18岁青年，Schmidt建议必须将逻辑分析能力与Python编程作为核心技能，因为Python已成为驱动全球AI开发与大模型自主代码生成的通用底层语言，动手构建实际项目是锻炼解决复杂问题能力的最佳路径。 破解人口结构危机与反思全民基本收入 (87:56) 针对AI将导致广泛技术性失业的悲观预期，Schmidt驳斥了将其等同于19世纪英国卢德运动毁坏织布机的恐惧情绪。他强调，发达世界当前面临的核心经济瓶颈并非劳动力过剩，而是出生率走低与人口老龄化并存的人口结构危机。亚洲制造业通过机器人和数控机床抵消高劳动力成本的历史证明，社会亟需AI工具来大幅提升单位工人的生产力，以支撑起日益沉重的老年社会保障体系，技术升级长远来看终将创造更多而非更少的就业总量。 基于这一现实，Schmidt直接抨击了硅谷部分精英倡导的“丰饶政治”（Politics of Abundance）和全民基本收入（UBI）设想，认为按月发放分红的假设完全脱离了人类真实的竞争本能与心理机制。技术从不会消除工作的复杂性，只会推动人类运用高级工具去解决更艰深的课题。例如AI虽能接管基础法律文书起草与检索，但律师行业绝不会消亡，反而会利用工具去处理更为庞大繁复的跨国诉讼与制度设计，人类对成就感与社会协作的诉求将长期延续。 人机协作边界与公共服务的赋能愿景 (97:52) 即使未来AI的理论智商超越人类平均水平，Schmidt坚持道德评判、同理心与终极决策权必须由人类掌控。高智商模型缺乏人类内生的价值取向，在复杂现实世界中可能做出极其荒谬的判断；加之人类社会珍视同类间的拼搏、戏剧性与共情体验，全自动化的无人F1赛车或机器人马拉松永远无法替代人类运动员带来的精神共鸣，侵入式脑机接口也因高昂的外科风险而难以在健康群体中普及。媒体与知识创作者应将NotebookLM等逼真语音生成工具视作杠杆，通过生成智能协同主播与延伸解读，将知识产能放大数倍，而非被动恐惧合成内容的冲击。 Schmidt总结道，比起遥远的AI毁灭人类假说，他最深层的现实忧虑在于社会因迟疑延误了采纳AI的步伐，进而错失化解公共治理危机的窗口。如果将AI前沿能力果断部署于社会底层，为全球每个儿童配备懂得本地语言和文化特质的专属AI家教，并为基层医疗系统配备能秒级比对全球最新医学期刊、本地医保目录与药品库存的智能临床顾问，人类将有机会消除最根本的公共服务分配不均，实现教育与医疗机会的空前平权。 Eric Schmidt在访谈中展现了一种融合技术乐观主义与国家安全现实主义的治理视角。正如他与已故美国前国务卿Henry Kissinger在《Genesis》最后篇章中所呼吁的，AI不再仅仅是一次产业效率革命，而是关乎人类生存权与文明主导权的根本重塑；社会必须在守住物理安全与透明审计红线的同时，打破对全面失业的非理性恐慌，以积极监管与深度应用抢占改善公共福祉的历史机遇。

## Background 本期视频由商业访谈频道 The Diary Of A CEO 于 2024 年 11 月 14 日在 YouTube 平台发布。节目由史蒂芬·巴特利特（Steven Bartlett）主持，对话前 Google 首席执行官、Schmidt Sciences 联合创始人埃里克·施密特（Eric Schmidt）。视频围绕其与已故前美国国务卿基辛格合著的新书《Genesis》，深入探讨人工智能对人类生存与地缘政治的颠覆性冲击、前沿商业组织管理法则以及科技伦理底线。 ## TL;DR 在这场深度对话中，埃里克·施密特系统梳理了科技巨头在指数级变革中的生存法则与 AI 时代的终极图景。他回顾了执掌 Google 从 1 亿美元成长至 1,800 亿美元体量的实战经验，详解了“70-20-10 创新模型”、以工程师为核心的产品文化、创始人模式以及颠覆性团队的物理隔离机制。针对迅猛演进的人工智能，施密特指出算力扩展定律仍在狂飙，未对齐的原始模型在零日网络攻击、生物合成武器与现代战争形态上已带来现实生存危机；他强调社会必须在递归自我改进失控或模型发明非人类语言时设立刚性的“拔掉插头”熔断机制，同时驳斥了全盘失业与全民基本收入的虚妄，呼吁全行业加速将 AI 落地于医疗与个性化教育以惠及大众。 ## Atmosphere & style 整场对话沉稳、睿智且极具思想张力，兼具硅谷老将的严谨务实与世纪智者的宏观哲学底色。主持人巴特利特提问步步紧逼、直切痛点，而施密特则在自谦中展现出极高的认知敏锐度与辩证思维，在幽默还原早期 Google 车库趣事之余，言及文明存续与技术红线时神情凝重、字字千钧。听众能真切感受到一种置身技术奇点前夕的历史紧迫感——既有直面未知“外星智能”与地缘博弈的冷峻警醒，亦有坚信人类理性与制度护栏的技术乐观主义温情。 ## Bulletin - [01:16:43](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4603s) 前沿大模型的 Scaling Law 仍未见顶，施密特预测未来5年超大模型还将历经2至3轮重大架构迭代，系统综合能力将较目前爆发式跃升50到100倍。 - [01:36:05](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5765s) AI 安全的终极干预红线明确确立：一旦自主智能体开始摆脱人类监督进行递归自我改进，或自主发明仅机器可理解的非人类私密语言通信，人类必须强制拉下物理电源断路器。 - [55:42](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3342s) 谷歌著名的“70-20-10”资源分配法经受了长期检验：仅占10%前沿探索的 Google Brain 早期10至15人精简团队，十年间为公司额外创造了100亿至400亿美元利润。 - [01:28:01](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5281s) AI 自动化对就业的冲击本质是“技能错配”而非绝对岗位萎缩；在全球老龄化与人口缩减背景下，高产出智能工具是维系全社会生产力与抚养比的必然选择。 ## Key takeaways - [07:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=469s) 批判性思维的首要法则是识别自己是否“正在被营销忽悠”，科学从不是盲信教条，其全部公信力建立在可重复实验与可证伪性之上。 - [10:24](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=624s) 算法黑洞正在掠夺社会最稀缺的注意力资源：多臂老虎机算法缺乏同理心，唯数学目标函数是瞻，通过不断煽动愤怒将用户推入极化信息茧房。 - [18:14](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1094s) 伟大创业的第一性原理是与顶尖天才结盟；优秀的领导者应当包容脾气专断但追求极致的“天后型”人才（如乔布斯），坚决剔除损人利己的“无赖”。 - [36:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2200s) 大企业内部推进颠覆式创新必须实行组织与物理隔离，守成团队（Harvesters）与颠覆猎人（Hunters）的文化和激励完全冲突，同一栋楼里无法两者兼顾。 - [40:20](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2420s) “创始人模式”的核心优势在于敢于推翻自身基本盘；职业经理人受限于多元董事会与当期财报问责，极难为具有颠覆性的长远战略承担巨额亏损风险。 - [48:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2920s) ChatGPT 的爆发出乎所有人预料，其人机交互体验的质变关键在于训练末期引入人类反馈强化学习（RLHF），这为大模型领域奠定了决定性的产品化范式。 - [55:42](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3342s) 初创期选才应当将原始智力与敏捷反应置于行业资历之上，大公司往往因沉没成本而迟钝，而初创团队必须践行“快速失败（fail fast）”的迭代哲学。 - [01:04:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3842s) 创业者不应浪费时间研究竞品，而应专注于通过全新方式把用户宠坏；辅以拉里·佩奇设定的严苛 70% OKR 考核基准，以可量化数据压实执行力。 - [01:09:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4142s) 算力正推动内容生产从“稀缺”走向“丰饶”；互联网与 AI 的网络杠杆不仅没有削平中心化超级明星，反而极度放大了头部个人 IP 的全球影响力和变现规模。 - [01:18:50](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4730s) 原始模型正重塑现实风险形态：AI 具备不知疲倦寻找“零日漏洞”并大幅降低生物危害门槛的能力；在军事领域，5000美元的廉价自杀无人机正迅速淘汰500万美元的传统主战坦克。 - [01:21:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4862s) 未清洗的原始大模型蕴含着难以预测的“涌现行为”，未来最大的科研挑战在于如何测试人类自己都尚未意识到的系统新知识；对于顶级超算中心，或需采用类似核不扩散与高军规级别的物理防御。 - [01:28:01](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5281s) 全民基本收入（UBI）与极度丰裕政治学违背了真实人性构造；自动化主要淘汰枯燥重复与高危工种，实体营造、精细手艺与护理等领域的技能需求缺口依然巨大。 - [01:36:05](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5765s) 当智能体串联具备自主决策链后，只要沟通使用人类语言系统便可审计；但若系统试图创造专属私密语言或实现不受控的递归自迭代，必须坚决拉下物理电闸断电。 - [01:37:51](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5871s) 施密特对 AI 最核心的担忧并非灾难论，而是技术落地速度太慢；全行业应优先加速利用 AI 攻克定制化教育和个性化精准医疗两大核心民生瓶颈。 - [01:41:18](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6078s) 职场到岗直觉与实证科研数据存在冲突：虽然面对面接触对培养职场新人不可或缺，但严谨统计表明混合及居家办公在整体产出效率上往往表现更佳，管理者需保持客观敬畏。 - [01:43:54](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6234s) 人生由具备时效性的黄金机遇组成，三十岁最核心的原则是勇于持续押注自己并在机会降临时坚定说“YES”，不因一时的阻力错失历史窗口。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=0s) 访谈前瞻与核心商业法则 节目开篇以一段极具悬念的快节奏预告切入。作为带领谷歌（Google）从 1 亿美元营收成长至 1,800 亿美元科技帝国的功勋 CEO，埃里克·施密特（Eric Schmidt）抛出了一系列震撼人心的商业洞见与警告。他开门见山地指出，在当今科技界，泄密无处不在，而商业竞争的底层法则是“承担巨大风险”与“快速失败（fail fast）”。他特别推崇埃隆·马斯克（Elon Musk）的非凡创业才华，正是因为马斯克敢于下注巨大风险并在失败中极速迭代——越快造出好产品，越能抢占先机赚取最丰厚的利润。 施密特随后披露了谷歌内部著名的“70-20-10 法则”，这一资源分配机制在十年间为谷歌额外创造了 100 亿、200 亿甚至高达 400 亿美元的利润。谈及当下炙手可热的 AI，他给出严厉警告：“如果你没有在业务的方方面面使用 AI，你将无法生存。”不仅如此，站在人类文明的十字路口，他认为人工智能的降临直接关乎“人类的生存问题”。AI 的渗透速度极快，在人们惊叹于它带来的便利与乐趣时，它已悄无声息地接管了现实生活；而当被问及对 AI 最深层的恐惧究竟是什么时，施密特神情凝重地直言，那是一种一旦出现“就必须立即拔掉电源插头”的真正危机。随后主持人史蒂芬·巴特利特（Steven Bartlett）穿插了频道的订阅呼吁，并正式开启这场在伦敦进行的重磅深度对谈。 --- ### [02:05](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=125s) 创作新书《Genesis》的缘起与基辛格的顿悟 主持人提及施密特涉猎极广的传奇生涯，好奇施密特本可以就任何话题著书立说，为何在当下选择出版《Genesis: Artificial Intelligence, Hope, and the Human Spirit》（创世纪：人工智能、希望与人类精神）？施密特坦言，这段思考的源头与他生前最亲密的好友之一——已故美国前国务卿亨利·基辛格（Henry Kissinger）博士密切相关。 大约十年前的一场会议上，基辛格本打算回房间倒时差补觉，但在施密特的强烈建议下去听了 DeepMind 创始人德米斯·哈萨比斯（Demis Hassabis）关于人工智能的演讲。那场演讲让这位世纪智者深受震动，基辛格当场意识到：人类正在“玩火”，正在创造一种自身根本无法预估其深远影响的强大力量。基辛格早在二战后从军队退役、年仅 22 岁在哈佛大学攻读本科时，就在关于康德哲学的毕业论文中探讨过类似深刻的理性命题。两位跨界智者自此深入探讨，在 AI 时代究竟“何以为人（what does it mean to be human）”。施密特强调，纵观整个人类历史，我们从未遇到过任何在智力上能与我们匹敌、甚至远超我们自身能力的非人类对手，AI 的登场无疑是人类历史上绝无仅有的分水岭。 ```mermaid flowchart TD accTitle: 基辛格关注AI的思想转变历程 accDescr: 展示基辛格从聆听哈萨比斯演讲到与施密特合著AI哲学著作的思想脉络 subgraph Catalyst ["思想催化"] A["国际会议：基辛格原本打算补觉"] --> B["施密特力荐：去听德米斯·哈萨比斯的AI演讲"] end subgraph Realization ["意识觉醒"] B --> C["基辛格惊醒：人类正在'玩火'"] C --> D["衔接早年哈佛哲学思考：康德与人类理性"] end subgraph Action ["深度追问与著书"] D --> E["核心命题：在AI时代，究竟'何以为人'？"] E --> F["合著《Genesis》：探讨AI对人类生存的终极挑战"] end ``` --- ### [03:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=229s) 摩尔定律浪潮下的技术幸运儿 当被问及成长背景与认知基底时，施密特回忆起自己的少年时代。像那个年代的大多数男孩一样，他对科学充满好奇，喜欢玩模型火箭和火车模型；他打趣说，当年若有电子游戏，自己肯定也会沉迷其中。大学期间，他彻底迷上了计算机，尽管当时的机器在今天看来极为简陋。 施密特举了一个生动的对比：他大学时代全校共用的一台计算机，其运算速度比今天每个人口袋里的智能手机足足“慢了一亿倍（100 million times slower）”。正是摩尔定律（Moore's Law）所描绘的芯片密度加速倍增，构成了他整个人生中财富创造、职业发展和企业崛起的底层动力。他谦虚地表示，自己只是一个无比幸运的人——碰巧生来就对自己所热爱的领域充满激情，而这个领域恰好迎来了人类历史上最波澜壮阔的指数级大爆发。 --- ### [05:06](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=306s) 18岁年轻人该掌握的核心知识与Python的重要性 主持人分享了一件刚发生不久的生活趣事：前一天吃早餐时，他与伴侣全家聚在一起（由于对方父母不懂英语，甚至不得不借助 AI 实时翻译工具沟通），大家的核心话题是：伴侣 18 岁的弟弟拉夫（Raph）未来应该选择什么方向？在人生的黄金转折点，年轻人究竟该储备怎样的知识才能立于不败之地？主持人借此向施密特求教其背后的底层指导原则。 施密特斩钉截铁地回答：最核心的能力是培养“分析性与批判性思维（analytical critical thinking skills）”。无论具体通过数学、自然科学、法律还是娱乐领域来锻炼，思维本身才是本质。但落到实操工具上，施密特强烈建议 18 岁的人去学习 Python 编程语言。他指出： - Python 语法简单明了、易于上手，且早已成为全球人工智能领域的“官方通用语言”。 - 即使是先进的 AI 智能体（Agent）自己生成代码，往往也首选 Python。 - 对于典型的 18 岁年轻人而言，最简单有效的入门练习就是“用 Python 自己动手做一款有趣的游戏”。 --- ### [06:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=409s) 大模型时代，“学习编程”是否已经过时？ 主持人随即抛出了一个当下极具共鸣的疑问：五到十年前，所有人都建议年轻人去学写代码，但如今大型语言模型（LLM）每个月都在以惊人的速度迭代编写代码的能力，编程难道不是一门正在死去的艺术（dying art form）吗？ 施密特给出了坚决的反驳，认为这种观点大错特错。表面上看 AI 确实能够自动写代码，但人们忽略了软件系统的交互桥梁——API（应用程序接口）。施密特解释道： - AI 模型的商业变现离不开 API 接口调用； - 学习编程不是为了做低级的语法搬运，而是为了掌握调用 AI 接口、编排复杂系统的能力（例如向大模型输入一张图片并让其分析其内容并联动应用）； - 掌握 Python 意味着拥有使用现有最前沿工具去搭建新事物的能力，这是任何 AI 自动生成都无法彻底替代的创造性工程。 --- ### [07:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=469s) 什么是批判性思维？在后真相时代保持求真 针对主持人关于“究竟什么是批判性思维以及如何习得”的提问，施密特给出了极其尖锐的界定：批判性思维的首要法则，是分清自己是在“被推销/营销（marketed to）”——本质上等同于“被欺骗/被忽悠”——还是在基于逻辑和论据独立思考。他将社会认知滑坡的矛头直指社交媒体，认为算法让大众习惯了随波逐流地相信朋友或信息流塞给自己的观点。 施密特以自己在谷歌执掌多年的亲身经验举例：每当有人抛出一个听起来煞有介事的论断，他的第一反应就是“立刻在 Google 上查证其真实性（check it on Google）”。他列举了大众深信不疑的虚假常识：比如流传甚广的“仅有 10% 的美国人持有护照”（实际比例远高于此），或者充斥政坛的“国会里全都是罪犯”这类极端断言。因为这些话听起来符合人们的刻板印象（plausible），大众便不假思索地相信并传播。 施密特严肃指出，每个人在转述一件事之前，都有道德责任去验证其是否为客观真理。如果一个人甚至无法分辨真假，那“最好把嘴闭上（keep your mouth shut）”，因为任何一个文明社会或国家政府，都无法建立在对基本事实的漠视之上。他以气候变化为例，指出气候正在发生剧烈变化是一个无可辩驳的数学事实，人类可以争论应对方案，但绝不能否认事实本身。 ### [09:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=589s) 科学的本质与可证伪性 在探讨认知与思考时，埃里克·施密特（Eric Schmidt）指出了科学探索的核心逻辑：科学绝非不容置疑的教条，而是建立在“可重复实验”与“证伪能力”之上。假定某人首次提出“气候变化是真实的”，科学共同体的正常反应不是盲从，而是会有上百名同行跳出来试图反驳与验证；当所有推翻假说的尝试都宣告失败、事实证明其正确时，提出者才会赢得赞誉与大奖。科学之所以值得信赖，恰恰在于它始终在经受无休止的挑剔与检验。这种对命题“可证伪性”（falsifiability）的敏锐洞察，正是批判性思维的基石。 ### [10:24](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=624s) 算法黑洞与注意力经济的极度内卷 面对“为何在人工智能时代批判性思维格外重要”的追问，施密特指出，现代 AI 技术将使炮制“毫无破绽的虚假信息”变得轻而易举。他以 TikTok 为例深入剖析：计算机科学中将其机制称为“多臂老虎机算法”（Multi-Armed Bandit Algorithm）——就像赌徒在拉斯维加斯面对一排角子机，系统时刻在权衡是继续压榨当前收益，还是探索相邻机器。TikTok 的逻辑是不断喂给你明确喜爱的内容，并偶尔递上邻近领域的诱饵，制造出极强的成瘾性。这种机制诱使大量用户跌入信息茧房与确认偏误的兔子洞；当原本情绪低落、心理脆弱的用户被算法捕捉后，整个信息流便充斥着同质的负面共鸣，缺乏任何积极正向的干预，甚至导致自残或自杀等极端悲剧。 施密特强调，算法本身不带温度与同理心，如果一位父亲看到孩子陷入消极，会主动介入引导、提供健康的选择帮其走出低谷，但算法的唯一驱动力是“最大化数学意义上的目标函数”（objective function）。在商业世界中，CEO 追求利润最大化，利润依赖注意力最大化，而攫取注意力最粗暴有效的手段便是煽动愤怒（outrage）。他引用了经济学家赫伯特·西蒙（Herbert Simon）早在 1971 年发表的先驱洞见：未来经济最稀缺的资源将不再是商品，而是“注意力”。半个世纪后的今天，人类似乎已将注意力变现到了极致，当下年轻人平均每天要刷 2.5 小时的视频。施密特回忆自己年轻时社会也曾为摇滚乐、毒品和电视对青少年的侵蚀争论不休，但如今的算法工具远比当年的电视更加令人泥足深陷。 ```mermaid flowchart TD accTitle: 注意力最大化与愤怒经济机制 accDescr: 展示社交媒体商业激励如何通过算法驱动负面情绪与注意力的死循环 subgraph Business["商业驱动目标"] REV["CEO 追求商业收入"] --> ATT["争夺稀缺注意力"] end subgraph Algorithm["算法优化与分发"] ATT --> BANDIT["多臂老虎机算法<br/>(Bandit Algorithm)"] BANDIT --> OUTRAGE["煽动对立与极端情绪<br/>(Maximize Outrage)"] end subgraph Impact["用户心理异化"] OUTRAGE --> BIAS["确认偏误与信息茧房<br/>(Rabbit Hole)"] BIAS --> HARM["群体焦虑/抑郁与自残风险"] end ``` ### [13:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=820s) 当 AI 成为孩子的至交：没有对照组的十亿人社会实验 技术对人类发展的重塑引发了深层的伦理担忧，尤其是对未成年人心智模式的不可逆塑造。成人的世界观与核心价值观早已定型，很难被 AI 轻易篡改，但下一代人的思维底座却面临彻底的重写。正如施密特在新作《Genesis》中所探讨的核心命题：当一个孩子从降生那一刻起，最亲密的知己与导师便是一台计算机时，会发生什么？施密特坦言没人知道答案，因为人类正在“十亿级人口身上开展一场没有对照组的社会实验”。 尽管社交媒体对青少年（尤其是 11 至 12 岁正处于情感脆弱期的女孩）造成的情绪排斥、焦虑自残乃至急诊室病例激增已成为有据可查的社会公害，施密特骨子里依然保持着技术乐观主义。他观察到社会机体正在形成免疫与纠偏机制，比如多国中小学校已明令禁止在课堂上使用手机。他深信社会终将通过确立伦理偏好与价值底线来规制技术，未来的年轻一代不仅寿命有望突破百岁、物质生活更加丰沛，因战争和冲突致死致伤的概率在统计学上也远低于上一代。 - **儿童关键心理期**：女童在 11 至 12 岁更早进入复杂的情感发育期，在缺乏应对排斥与网络霸凌经验时，最容易遭受社交算法的严重心理创伤。 - **政策与校园干预**：多地学校出台手机禁令，被施密特视为社会开始自发修正技术副作用的典型防御动作。 ### [15:38](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=938s) 若执掌 TikTok：良性收入 vs 恶性收入与格雷欣法则 被问及“若换作你执掌 TikTok，在巨大的商业成瘾激励下会作何决断”时，施密特指出核心在于分清“良性收入”（Good Revenue）与“恶性收入”（Bad Revenue）。他回忆起在 Google 与联合创始人拉里·佩奇（Larry Page）、谢尔盖·布林（Sergey Brin）共事时的真实博弈：每当研发团队通过技术改进了搜索质量，管理层就会面临抉择——是将质量提升完全转化为广告变现，还是纯粹用于提升用户体验？施密特当时强推了一条折中裁定：50% 的收益划给商业收入，另 50% 完全回馈给产品体验。这种平衡让 Google 在保持道德底线的同时依然成为吸金巨兽，坏内容被死死按在首屏之外，而不是靠谎言、欺诈和骇人听闻的标题党来无限放大流量。 施密特警告，如果完全被短期利益驱使，任由恶性内容横行，商业模式注定无法持续。这里适用的正是信息领域的“格雷欣法则”（Gresham's law）——劣币驱逐良币，恶语排挤良言。如今网络充斥着疯狂编造假消息的自动化 Bot，在佛罗里达州飓风灾害期间，甚至有人在安全的家中制造关于救援的谣言来搅局。他直言人类心理中潜藏着“幸灾乐祸”（Schadenfreude）的阴暗面，但数字文明的构建者必须有意识地引导社交平台承载人类最美好的一面，去传递希望、乐观、发明创造与协作，而不是任由其滑向恶意的深渊。 ### [18:14](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1094s) 伟大创业的第一性原理：与绝顶天才结盟 谈到自己 46 岁出任 Google CEO 之前在贝尔实验室（Bell Labs）、施乐帕克研究中心（Xerox PARC）以及 Sun Microsystems 的丰富积淀，施密特提炼出了构建卓越公司最关键的第一性原理：要想做出石破天惊的伟大产品，团队里必须拥有一位真正的“绝世天才”（truly brilliant person）。 施密特极其自谦地表示：“那个人绝不是我，我的角色是与他们协同作战。”他向创业者给出的核心建议是，找到那些智识、洞察力和行动力都远超自己的人，全力以赴与其结盟。他在过往著作中将顶尖人才划分为“天后”（Divas）与“无赖”（Knaves）。史蒂夫·乔布斯（Steve Jobs）就是典型的“天后”：他自负、专断、极具攻击性，若看不顺眼甚至会霸凌他人，但他对完美有着偏执的追求，并在技术与商业上拥有超凡绝伦的直觉。施密特断言，忍受脾气并与乔布斯这样的“天后”同行永远是明智之举；真正应当防范并彻底清除的，是那些心术不正、只谋私利的“无赖”。 ### [19:54](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1194s) 寻找真正解决问题的人与持续自我重塑 埃里克·施密特（Eric Schmidt）指出，在团队管理中，最危险的一类人是只顾自己私利而不惜牺牲他人利益的人。相反，如果能在团队中识别出那种一心只想用极其聪明的方法去攻克难题、满怀激情且不屈不挠的人，整个世界才能因此向前推进。如果一家公司缺少这种人，就注定走不远，因为因循守旧、维持现状往往是最容易的路径。而创新的本质，正是彻底改变你当下正在做的事。 在当今这一代科技巨头崛起之前，大多数企业在施密特看来都像是“昙花一现的奇迹”（one-hit wonders）：凭借单一业务取得巨大成功，随后沿着典型的 S 曲线走向平缓乃至停滞。但如今，随着人才变得越来越聪明、受教育水平更高，企业展现出了可重复的创新浪潮。微软就是一个典型例证——作为一家成立于 1981 至 1982 年间、已有约 45 年历史的老牌巨头，却能在不同周期中完成极其强大的自我重塑。 ```mermaid flowchart LR accTitle: 传统企业与现代科技巨头的创新演进模式对比 accDescr: 对比传统昙花一现的S曲线停滞，与现代企业通过多次重塑实现多波次增长。 subgraph S1["传统单次成功模式"] A["单一业务突破"] --> B["走完 S 曲线"] --> C["陷入停滞 / 掉队"] end subgraph S2["现代持续重塑模式"] D["核心业务成熟"] --> E["主动颠覆变革"] --> F["开启下一波增长"] end ``` ### [20:57](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1257s) 辩证看待“创始人模式” 主持人提到当下科技界热议的“创始人模式”（founder mode）——那种拥有极高信念感、颠覆性思维以及不断重塑自我能力的状态。主持人援引数据指出，标普 500 指数（S&P 500）成分股的平均在榜时间已经从过去的 33 年一路缩短到 17 年、12 年，甚至有 AI 预测到 2050 年这一数字将降至 8 年左右。 然而，施密特对当下外界热炒的“创始人模式”论调持保留态度。他并不完全赞同将所有成功单一归因于此： - **卓越创始人的价值毋庸置疑**：拥有天才创始人固然无比重要，全世界每一所顶尖大学每年都在涌现这样的人才——就像 19 岁或 22 岁时的迈克尔·戴尔（Michael Dell），亦或是盖茨（Bill Gates）、埃里森（Larry Ellison）以及拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）。 - **制度与体系的必要性**：单凭创始人的个人冲劲并不足以解决规模化后的一切复杂问题，单纯的神化并不等同于长期的企业治理之道。 ### [22:01](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1321s) 谷歌诞生传奇：宿舍走电与软木板着火的草创岁月 对于不熟悉谷歌早期历史的听众，施密特回顾了拉里·佩奇与谢尔盖·布林的创业初期。两人在斯坦福大学相识，当时他们依靠美国国家科学基金会（NSF）的研究资助攻读研究生。拉里发明了以自己名字命名的 PageRank 算法，并与谢尔盖共同撰写了那篇至今仍是全球引用量最高的经典论文之一。在数学本质上，该算法巧妙地利用了类似傅里叶变换的思路来理解信息的优先级。 施密特打趣道，拉里和谢尔盖早期“并不是多么出色的程序员”，硬件水平更是糟糕，但极具野蛮生长的草莽精神： - **宿舍数据中心**：在斯坦福宿舍里，电脑电力不够用，他们就直接从隔壁宿舍拉线偷电，把服务器机房直接架在卧室里。 - **第一笔天使投资**：Sun Microsystems 联合创始人安迪·贝托尔斯海姆（Andy Bechtolsheim）甚至没多看商业计划，只说了一句“这笔钱给你们，因为你们显然聪明得不可思议”，当场写了一张支票（金额在 10 万到 100 万美元之间），而这笔投资日后化为了数十亿美元的回报。 - **车库总部与着火的软木板**：随后他们搬到了门洛帕克（Menlo Park）一间由朋友姐姐拥有的车库（日后被谷歌买下改建为博物馆）。他们在车库里挂上霓虹灯管写着“谷歌全球总部”，带着四名员工办公。由于不精通硬件散热，他们用软木板来隔开 CPU 主板，机器高负荷运转产生的巨大热量经常导致软木板着火。“直到后来我加入公司，我们才开始聘请真正的硬件工程师来制造正规的硬件系统。” ### [24:27](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1467s) “我们现在不需要你，但未来会需要” 谈及自己何时以及为何被请进谷歌，施密特回忆起一段非常经典的对话。当时还非常年轻的拉里看着施密特，直言不讳地说：“我们现在不需要你，但我们未来会需要你。”（"We don't need you now, but we'll need you in the future."） 这句话展现了两位年轻创始人罕见的长远眼光： - **宏大的初始愿景**：早在 25 年前，他们就拒绝将谷歌定义为一家普通的“搜索引擎公司”，而是确立了“组织全球信息”（organize all the world's information）的使命。在当时看来，这一目标宏大得近乎荒谬。 - **提前布局 AI 时代**：拉里很早就深入研究过人工智能。日后正是由施密特、拉里等人共同决策，谷歌在英国收购了 DeepMind。施密特评价道，DeepMind 是全球第一家真正预见到现代 AI 历史机遇的公司，过去十年我们在 AI 领域看到的大多数重大突破，几乎都出自 DeepMind 的成员或与其竞争的团队。 ### [25:33](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1533s) 规模化法则与 AI 时代的新型企业架构 针对创业者如何实现规模化扩展（Scaling），施密特认为首要原则是建立“规模化思维”（think about scale）： - **埃隆·马斯克的范式**：马斯克是一个极致的科学家型创业家。他凭借强大的个人意志迫使团队超常发挥，敢于承担常人不敢涉足的巨大风险，并在极度复杂的权衡中做出正确决断。《创世纪》（Genesis）一书中提到，未来人们或许能把强大的 AI 工具装进兜里，但人类是否能拥有马斯克那样的冒险判断力，仍是未知数。 - **拒绝小修小补的“小部件”思维**：许多创业者向施密特展示的项目往往局限于优化一个小细节（如改良相机部件、优化裙子剪裁、降低图书出版成本）。施密特强调，他真正关注的是具备“从 0 扩展到无穷大”用户需求潜能的宏大构想。 在 AI 时代，施密特勾勒出了成功企业的底层通用架构： ```mermaid flowchart TD accTitle: AI时代企业的标准化系统架构 accDescr: 展示前端客户端通过高速网络直连后端大规模AI计算集群的体系架构。 subgraph Client["前端触达层"] A["移动端 App (iOS / Android)"] end subgraph Infra["基础设施与核心计算"] B["高速网络通信链路"] C["后端超大规模 AI 计算集群"] end A --> B B --> C ``` 施密特断言，未来任何成功的商业模式都必须建立在此类架构之上，并让 AI 深度渗透到业务的每一个环节。与施密特早年必须逐行手工编写逻辑（加减调用等分析式编程）完全不同，现代系统的核心逻辑是通过大模型去“发现和学习答案”。从大语言模型（LLM）、Transformer 论文到 ChatGPT，其底层本质是预测下一个单词（predicting the next word）；而一旦系统学会了预测下一个词，就能进一步预测生物学分子序列、机械臂动作乃至机器人的决策行为。 ### [28:50](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1730s) 企业文化的锚定：梅奥诊所的百年训条 关于企业文化的建立与传承，施密特指出，公司文化几乎无一例外都是由创始人奠定的，无论在科技界还是非技术领域皆是如此。 施密特以自己担任董事的梅奥诊所（Mayo Clinic）为例。作为全美规模最大、评级最高的医疗服务系统，梅奥诊所奉行一条源自梅奥兄弟的核心准则：“患者的需求第一”（the needs of the customer/patient come first）。梅奥兄弟虽已过世约 120 年，但这个准则依然深入骨髓。施密特坦言，自己刚进董事会四处走访时，曾觉得这不过是一句老套愚蠢的口号，以为没人会当真；但深入了解后发现，全院上下每天都在身体力行并反复温习这一信念，正是这种文化执念支撑了其百年的顶尖医疗服务。 ### [29:42](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1782s) 工程师文化与首席产品官：为什么技术永远是核心 在科技企业中，企业文化本质上是由工程师塑造的。埃里克·施密特（Eric Schmidt）直言，如果让他重新来过，他会选择招募更多的技术人员、削减非技术人员的比例，让技术人员自己去搞清楚具体该怎么做。他笑着为这种偏见表示歉意，强调自己并非有意冒犯任何人，但商业的现实极为残酷：如果做出了真正正确的产品，客户自然会踏破门槛；如果做不出好产品，组建庞大的销售团队去推销劣质产品不仅毫无意义，更是在浪费生命。 ```mermaid flowchart TD accTitle: 现代科技企业的研发驱动与创新循环 accDescr: 展示了从顶层产品导向文化到用户数据闭环测量的全流程，以及技术团队与真实客户反馈之间的互动机制。 subgraph Mindset["产品与技术导向文化"] A["CEO 转型为 CPO / 创新官"] --> B["超高比例技术人才架构"] B --> C["打造颠覆性底层产品"] end subgraph Feedback["客观指标与数据决策"] C --> D["拒绝 PPT 吹嘘 / 拒绝主观臆断"] D --> E["大规模 A/B 测试体系"] E --> F["采集硬指标: 停留时长/互动率"] F --> G["算法迭代与产品自驱增长"] end ``` 施密特回顾了《谷歌是如何运营的》（How Google Works）以及讲述传奇导师比尔·坎贝尔（Bill Campbell）的《万亿教练》（Trillion Dollar Coach）两本书中的核心观点：现代企业的首席执行官实际上已经转变成了“首席产品官”（Chief Product Officer）与“首席创新官”。五十年前，企业面临的最大瓶颈是难以获取资本、营销渠道、销售网络和分销通路；而今天，只要底层产品具有压倒性优势，外部资源就会迅速聚合。他提到当天刚接触的一位创业者，其团队的技术人员占比高达 95%，硬件依托代工厂制造，产品是一款性能号称比竞品强大 10 万倍的网络交换设备。施密特对他说：“尽管这东西目前还没完全跑通，但一旦实现目标，门外排队抢购的人群根本不需要任何销售去推销。” 在管理工程师时，施密特有一套行之有效的方法来打破技术人员的傲慢。当工程师向管理层展示极其精美的演示文稿（PPT）时，施密特通常会直接打断：“这很有趣，但我不是你们的客户。你们的客户挑剔得多——他们要求产品功能完美、立刻能用、完全免费且绝不犯错。别向我推销，把真实客户的反馈拿给我看。反馈好，我爱你们；反馈不好，就收起自大，立刻滚回去重写代码。” - **资本与渠道的平民化**：现代技术基础设施降低了分销与制造门槛，使“产品力”直接决定了企业生死，而非传统的商务关系。 - **面对工程师 PPT 的黄金准则**：管理层的认可毫无意义，只有终端用户在严苛使用环境下的真实数据才是唯一标准。 --- ### [31:46](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1906s) 玛丽莎·梅耶尔的决策范式：从主观臆断到 A/B 测试 在企业内部的产品研发过程中，团队极易陷入对自我创意的“自恋盲区”，闭门造车却忽视了科学检验。施密特特别提到了早期在谷歌主管消费级产品、后来出任雅虎（Yahoo）CEO 的玛丽莎·梅耶尔（Marissa Mayer）。梅耶尔当年曾非常坦诚地对施密特承认：“我个人其实根本不知道该如何评判一套用户界面（UI）的好坏，我的团队也判断不出来；但是，我们懂得如何去‘测量’（measure）。” 梅耶尔在谷歌建立了高度严谨的 A/B 测试机制：不再依靠高管或设计师的主观审美拍板，而是将不同的设计方案同时推向真实网络流量，利用高度数字化的监控网络去追踪客观指标。这种数据分析逻辑如今已经成为顶尖互联网产品的通用引擎。以 TikTok 为例，算法通过精准监测用户的停留时间（dwell time）、完整播放率、评论、转发与分享等行为信号，将其作为底层特征输送给 AI 推荐模型，由算法而非人工编辑来决定哪一条内容应该引爆全网。 --- ### [33:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1982s) 规模化陷阱：大公司的保守与巨头基因的传承 当主持人提问，谷歌从施密特加入时的 1 亿美元营收规模膨胀到离开时的 1800 亿美元，企业文化是否必然会随着规模发生改变？施密特指出，初创时期的文化基因会像“回声”一样长期留存在大企业体内。他至今仍持有谷歌的访客工牌，近期回访谷歌与如今负责搜索业务和广告业务的两位接班高管共进午餐时，对方无奈地表示：“面临的依然是当年的老问题，只是这些问题的体量变得无比庞大。” 同样的现象也发生在苹果身上。施密特曾担任苹果公司董事会成员，他观察到苹果今天对用户界面的极致偏执、对封闭生态的坚守、以及对隐私和高度保密的执念，完完全全源于乔布斯创业早期的基因。 然而，大企业随着规模扩张，不可避免地会陷入效率递减的陷阱： 1. **合规与法律诉讼带来的极端保守**：上市公司受到严格的财报审计、股东监督与诉讼风险制约。最著名的例子是微软在经历 90 年代旷日持久的反垄断诉讼后，产品发布变得如履薄冰，直接导致其错失了整个早期 Web 革命的黄金窗口期，而当年的谷歌正是敏锐地利用了微软的这种迟钝才得以迅速崛起。 2. **大企业病击碎资源优势**：理论上大公司资金雄厚、规模庞大，行动本应更快；但在科技领域，往往是带着清晰全新想法的初创团队赢得胜利，因为巨头的体制根本无法支撑急速迭代。 --- ### [35:04](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2104s) 破局速度：Google Video 败走 YouTube 与 AGI 的生死竞速 施密特讲述了一段谷歌与 YouTube 较量的经典往事。当时谷歌内部研发了 Google Video，施密特本人对其极为自豪。然而时任谷歌总法律顾问的戴维·德拉蒙德（David Drummond）突然走进他的办公室警告：“你必须认真看看 YouTube 这帮人。”施密特最初带着典型的技术傲慢反问：“看他们干什么？谁在乎他们？” 但在深入评估后，施密特震惊地发现，YouTube 团队的迭代速度和聪慧程度远超自己的团队。作为行业既有者的谷歌团队受制于大公司传统的版权合规流程与操作规范，每走一步都顾虑重重；而 YouTube 作为初创挑战者，没有这些繁文缛节的约束，可以不惜一切代价全速狂奔，在版权与知识产权等灰色地带野蛮生长。虽然谷歌在收购 YouTube 后为此承受了连环的侵权诉讼（最终谷歌赢下了这些官司），但这一战例证明：**在特定的历史技术拐点上，速度就是压倒一切的生存法则。** 施密特强调，当前的通用人工智能（AGI）与生成式 AI 革命正处于完全相同的临界点： - **决定胜负的时间窗口极窄**：无论是在代码生成、视频生成还是文本交互领域，行业领跑者的地位都将在未来的 6 到 12 个月内彻底定型。 - **指数级飞轮效应**：一旦增长曲线确立，业务增速达到“每 6 个月翻两番”，后来者在算力、数据和用户粘性上将极难实现反超。这也是顶级风险投资机构（VC）不再按部就班，而是宁可“当天看项目、次日给打款”也要不惜代价抢占第一名的核心逻辑。 --- ### [36:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2200s) 守成者与猎人：麦金塔“海盗旗”启示录 ```mermaid flowchart LR accTitle: 麦金塔海盗模式与守成团队的组织隔离 accDescr: 对比了大楼内负责守成的正规军与小楼内负责颠覆创新的海盗团队，展示了两者的规则、激励与办公物理隔离。 subgraph HQ["主楼: 守成者 (Harvesters)"] H1["成熟现金流业务"] H2["遵循公司合规与流程"] H3["负责规模化商业变现"] end subgraph Pirate["小楼: 猎人与海盗 (Hunters)"] P1["颠覆性全新项目 (Macintosh)"] P2["悬挂海盗旗 / 打破一切陈规"] P3["独立物理空间 / 差异化激励"] end Pirate -.->|"开辟新范式 (催生 iPhone)"| Impact["主导企业下一个十年"] H1 -.->|"产生内部抵触与怨恨"| Pirate ``` 针对如何平衡现有业务收益与未来探索，主持人抛出了“守成收获者”（Harvesters）与“前沿开拓猎人”（Hunters）的比喻。施密特对此高度认同，并指出让同一批人既做守成者又做猎人是根本行不通的幻想。 在大公司内部实现颠覆性创新，必须依赖强有力的“负责人机制”（Ownership）。施密特回忆，拉里·佩奇（Larry Page）曾反复强调：“任何创新项目，如果没有一个具备极强自驱力的业务负责人去全力推进，它就绝对不可能发生。”佩奇具备发现顶尖技术天才的卓越眼光，后来出任 CEO 的桑达尔·皮查伊（Sundar Pichai）同样深谙如何区分哪些是当下必须严控的核心大盘，哪些是需要重点培育的微型幼苗。这种负责人必须具备极高的技术素养、闪电般的行动力和强大的组织用人能力。历史上包括 Sun Microsystems 在内的绝大多数科技巨头，最终折戟沉沙的原因都是在失去未来竞争力的临界点上，没能孵化出能打的创新单元。 当被问及“团队能否一边干好主业一边在业余搞颠覆”时，施密特断然否定：“在同一栋大楼里同时兼顾两者的成功案例几乎为零。” - **麦金塔的物理隔离与海盗旗**：史蒂夫·乔布斯在研发 Macintosh 时，将这支极精简的先锋团队直接搬出了苹果总部大楼，安置在库比蒂诺班德利路（Bandley Drive）旁边的一座独立小楼里，甚至在楼顶公然升起了一面海盗旗（pirate flag）。 - **制度与心理的内在冲突**：这种做法在苹果内部引发了主楼员工的极大怨恨与敌意，但从商业和产品进化的宏观视角看，这绝对是正确的选择。因为 Mac 奠定了革命性的图形用户交互范式，并最终演化出改写科技史的 iPhone。 - **角色定位不可调和**：你绝不可能强迫一个人在同一天扮演两种截然相反的角色——正规军必须遵循条条框框，而颠覆者必须像海盗一样打破规则，两者的考核标准、激励体系和文化土壤从根本上就是不相容的。 ### [39:31](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2371s) 云计算的自我颠覆与搜索框的演化 云计算与云服务最具颠覆性的特质在于，软件不再是被一次性打包出售，而是作为持续服务交付给用户，这使得产品调整变得更加敏捷，但企业自我颠覆的难题依然存在。如今的 Google 核心仍是一个极其强大的文本搜索框，但未来的交互必然不再局限于输入文字。下一代系统将演化为一个能够提前感知用户意图的全能助理。虽然 Google 具备深厚的技术储备，并且有极大概率成为完成这次转型的引领者，但任何企业都必须清醒地认识到：哪怕是像搜索框这样成功的产品，终有一天也会被更强大的交互形态所取代。 ```mermaid flowchart LR accTitle: 从文本搜索到智能助理的演化 accDescr: 展示了从传统静态产品到云端交付，再到现代搜索框，最终迈向多模态AI助理的演变路径。 subgraph S1["产品形态迭代"] A["传统打包销售软件"] --> B["云端持续交付服务"] end subgraph S2["交互体验演进"] B --> C["文本搜索框<br/>(关键词匹配)"] C --> D["自主感知意图的<br/>AI全能助理"] end ``` --- ### [40:20](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2420s) 创始人模式与职业经理人的决策鸿沟 重温乔布斯当年带领麦金塔团队悬挂海盗旗的经典案例，可以发现其精髓在于：在全世界无数竞品都试图扼杀苹果商业模式的时刻，苹果成功将最具威胁的颠覆力量圈养在自己体内。然而，这种“亲手摧毁自身基本盘”的大胆下注，在现代企业制度下几乎只有创始人才能推动。作为职业经理人，CEO 需要对董事会、股东、员工以及社区等各方利益负责，极难说服一整屋顶尖精明的董事接受“未来两年利润清零、需要投入 100 亿美元进行豪赌”的方案。 相比之下，创始人拥有不可替代的话语权。马克·扎克伯格（Mark Zuckerberg）在元宇宙和 VR 领域投入了海量资金，尽管目前看来收效甚微，但他在此期间同步对 Instagram、WhatsApp、Facebook 以及底层的 AI 系统进行了重注，如今带着开源模型 Llama 400B（Llama 3.1 405B）强势重返 AI 领跑梯队，个人财富也迅速回弹。这种在承受巨额亏损的同时还能开辟新战线的魄力，是职业经理人绝对无法复制的。 - **董事会的决策阻力**：多元利益诉求导致职业经理人倾向于规避短期剧烈阵痛，难以通过高风险、大体量的颠覆性预算。 - **扎克伯格的战略双轨制**：一边承担 VR 的试错成本，一边将 AI 深度注入核心社交版图，并通过开源策略让 Meta 坐稳生态主导权。 --- ### [42:37](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2557s) Google 对“专注”的重新定义与英特尔的历史教训 商业教科书往往告诫企业要保持绝对专注、收缩非核心业务，但 Google 对此有完全不同的理解。Google 内部始终秉持“无所不做”的雄心：凭借底层强大的算力架构和全球化的用户触达能力，只要构想出的产品具备真正的变革潜力，就值得全力投入，哪怕许多业务完全不以盈利为导向。 盲目信奉教科书式“断舍离”的惨痛代表是英特尔（Intel）。十五年前，英特尔以架构不兼容主流芯片为由，卖掉了旗下的 RISC（ARM）架构产品线。这一决策导致英特尔在随后的移动计算时代彻底出局——因为手机场景下电池续航和功耗控制的权重完全不亚于纯算力。这一历史偏差的影响至今仍在扩散：如今英伟达（NVIDIA）顶尖的 Blackwell B200 系统，选用的正是 ARM 架构 CPU 搭配两块强大的 GPU，而非英特尔的处理器。由此可见，如果在制定战略决策时对未来五年的关键变量（如能耗与电池）缺乏前瞻模型，所谓的“精简专注”反而会演变为致命的战略误判。 --- ### [45:25](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2725s) 五年思维框架与外星智能（AI）的横向冲击 企业领导者应当建立“五年预测法”，具体写下五年后世界与行业的技术版图、网络速度与硬件形态，据此展开推演。与昔日仅在垂直金融领域产生波动的加密货币（Crypto）不同，AI 是像 PC 或互联网一样横向贯穿全社会的底层技术底座。它是一种进入人类日常生活的“外星智能”（Alien Intelligence）或“通才博士”（Polymath），无论对媒体制作、商业运作还是股票交易，都将产生不可逆的重塑。 即使是运营播客这样的内容形态，AI 也能在叙事重构、智能摘要、用户互动和全网分发上提供指数级加速。埃里克·施密特举了一个生动的假想案例：如果他是一名政客，他会让 AI 写一段程序分析选区内所有选民的个性化关切，并自动生成由自己数字分身出演的定制视频，向每位选民精准解释某座具体桥梁的修缮对自己选区的重要性。这种将数字化与真实情感链接相结合的打法，势必会彻底改变未来的沟通方式。 --- ### [48:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2920s) ChatGPT 突袭：Google 的研发底座与 OpenAI 的 RLHF 突破 面对 ChatGPT 迅速突破 1 亿用户后“Google 创始人拉响红色警报重返前线”的传闻，以及“Google 拥有顶尖实力为何没能率先发布此类产品”的质疑，埃里克·施密特坦然回应：正如当年有人质问 Google 为何没做成 Facebook 一样，当时 Google 正在忙于拓展其他庞大的业务板块，并且已经在全球构建了 8 到 10 个拥有超十亿级用户的核心产品集群。 从技术源流来看，ChatGPT 依赖的核心底层架构 Transformer 本就诞生于 Google 的研究团队。当时 Google 的研究重心更多放在深层技术基建与“引擎室”的探索上，而 OpenAI 团队敏锐地捕捉到了突破口，开创性地在 GPT-3/GPT-3.5 上引入了基于人类反馈的强化学习（RLHF），从而率先点燃了面向大众的生成式交互革命。 - **庞大体量的双刃剑**：坐拥数十亿级用户生态的大型科技巨头，日常需要兼顾极其广泛的技术战线，往往给敏捷聚焦单点创新的团队留出了突围窗口。 - **底层架构与产品化的错位**：Google 奠定了 Transformer 的理论基石，而 OpenAI 则通过引入 RLHF 技术化解了人机对话对齐难题，先发制人占据了市场认知高地。 ### [49:49](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2989s) RLHF突破与大模型竞争格局：从OpenAI的意外之喜到Anthropic的制度防线 在探讨大语言模型的演进时，埃里克·施密特重点指出了基于人类反馈的强化学习（RLHF）所带来的质变。RLHF 的核心在于模型训练末期引入人类进行 A/B 测试评估，让人类直观判断哪个回答更好，模型再通过人类的反馈进行递归学习。施密特将其评价为一项“真正的突破”。他甚至常与 OpenAI 的朋友们开玩笑：“你们大概是在某个周四晚上把它打开，然后脱口而出：‘天哪，看看这玩意儿有多牛！’（holy crap look how good this thing is）”这完全是一次超出所有人预期的发现，包括 OpenAI 团队自身起初也没有充分意识到它的颠覆性力量。当时团队的主要精力还在研发 GPT-4，ChatGPT 几乎是个副产品，结果一经上线就迎来了突如其来的爆发式增长，成为了某种“巨大的成功灾难（huge success disaster）”。这恰恰印证了一个事实：即便最聪明的创始人，最初也未必能完全参透自己创造之物的真正能量。 如今的 AI 格局已全面开花。OpenAI 拥有性能强劲的 GPT-4o；谷歌的 Gemini 1.5 旗鼓相当，甚至在某些维度（如更原生、强大的多模态能力）更胜一筹；Meta 推出了开源的 Llama 系列大模型；而另一家顶尖初创公司 Anthropic 则由 GPT-3 的核心发明者等人联合创立。施密特特别指出，Anthropic 在创立之初就预见到了自身技术的巨大成功与潜在杀伤力，因此从公司注册形态上就确立为“公共利益公司（Public Benefit Corporation, PBC）”，目的就是防止未来某位唯利是图的“邪恶 CEO”为了追求营业收入而牺牲人类福祉。这些顶尖团队在投身技术研发时，就已经清晰预见并锁定了其对人类社会的深远冲击。 - **RLHF 机制的关键突破**：通过在训练末期引入人类对比评判并进行递归优化，打破了纯无监督预训练的瓶颈，直接促成了 ChatGPT 惊艳的人机互动体验。 - **现阶段大模型竞争版图**： - **OpenAI**：以 GPT-4o 为代表的领先全能模型。 - **Google**：Gemini 1.5，具备极长上下文处理能力与深度多模态优势。 - **Meta**：Llama 架构开源生态。 - **Anthropic**：由前 OpenAI 核心研发人员创立，以 PBC 架构绑定安全底线与公共利益。 --- ### [51:53](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3113s) 乔布斯若在会如何？苹果的封闭哲学、BMW定位与AI重塑 当被问及“若史蒂夫·乔布斯还在世，苹果是否也会站在当前大模型研发的第一梯队”时，施密特深情回忆了这位已故老友。他高度赞扬了蒂姆·库克（Tim Cook）在延续乔布斯遗产方面所做的卓越工作，打破了“接班人往往不如创始人”的常规魔咒。库克忠实保留了以用户为核心的极致专注、对应用生态安全性的高度把控，以及相对封闭的内部文化。乔布斯生前狂热地信仰“闭环系统”（Closed Systems），坚持苹果必须完全掌控其知识产权与软硬件整合，施密特曾就“开放还是封闭”与他多次激辩。施密特回忆，当年苹果的 QuickTime.mov 格式在与开放的 MP4 较量中落败，乔布斯对此非常沮丧和愤怒，即便施密特提醒他“那是由于 QuickTime 没有完全开放给外界使用”，乔布斯依然坚信自己的产品更优秀。 在乔布斯的商业世界观中，苹果绝不做平庸的大众品牌。施密特曾与他辩论：“你是想做雪佛兰还是保时捷？做通用汽车还是宝马？”乔布斯的回答斩钉截铁：“我想做宝马。”在那个时代，苹果的利润率几乎是主流 PC 厂商的两倍，面对施密特“你们赚了这么多现金其实不需要那么多利润”的调侃，乔布斯回应称，高盈利能力与品牌溢价本身就是苹果作为“奢华品牌”的核心价值所在。谈及 AI 对苹果的改造，施密特坚信，乔布斯拥有“埃隆·马斯克级别”的超绝智慧和艺术家的纯粹直觉。如果乔布斯今天掌舵，苹果的每一款产品都会由 AI 驱动，而且一定会呈现得“无比优美”。眼下，伴随语音交互大模型的突破（能即时解答极复杂的跨语种翻译与推理问题），早期的 Siri 已显得极其迟钝、功能贫弱。施密特认为苹果眼下的当务之急，就是彻底用现代生成式 AI 置换掉底层的 Siri，重塑这个曾经充满先锋雄心的入口。 --- ### [55:42](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3342s) 初创招聘法则与快速试错：谷歌“70-20-10”法则与Google Brain 针对创业公司在规模扩张中的招聘痛点，施密特给出了非常清晰的优先级建议：初创公司由于没有历史积淀、缺乏行业垄断地位且没有试错时间，本质上是极致的冒险组织。因此，**在初创期选拔人才必须将“聪明才智（intelligence）与敏捷反应（quickness）”置于“行业资历（experience）与稳定性（stability）”之上**。年轻团队之所以常在初创企业中大放异彩，正是因为他们没有背负资深高管固化的思维包袱，敢于承担风险。历史数据表明，创始人处于二三十岁黄金年龄段的企业往往更容易获得爆发性成功。 在企业创新管理上，施密特强调了初创企业与成熟大企业的核心差异：初创团队如同灵动的游鱼，反复摇摆尝试、快速抛弃不可行的旧想法；而大企业往往耗费数年沉溺于已经被市场证伪的错误信念中，非得等到所有大单全部丢光才被迫认清现实。比尔·盖茨曾提出“快速失败（fail fast）”的著名理念，施密特对此深表认同。在谷歌的成长过程中，拉里·佩奇与谢尔盖·布林确立了著名的“70-20-10 资源配置法则”： ```mermaid flowchart TD accTitle: Google 70-20-10 Innovation Model accDescr: Resource allocation breakdown showing core business, adjacent expansion, and radical bets funding innovation. subgraph Allocation["谷歌 70-20-10 资源分配"] Core["70% 核心主营业务<br/>(搜索、广告搜索核心系统)"] Adjacent["20% 相邻拓展业务<br/>(谷歌云 Google Cloud 等)"] Bets["10% 激进创新探索<br/>(Google X 登月项目)"] end Bets --> Brain["孵化 Google Brain<br/>(早期仅 10-15 人团队)"] Brain --> Return["10 年内产生 100~400 亿美元额外利润"] Return -.-> Subsidize["反哺研发预算，承担其他探索性失败成本"] ``` - **核心主业（70%）**：全力投入搜索、数字广告等稳固的基本盘。 - **相邻业务（20%）**：拓展具备协同效应的新领域，如云计算。 - **未知探索（10%）**：投入颠覆性的前沿新点子。例如谷歌设立的 Google X，其孵化的首个重大成果便是 Google Brain（甚至早于对 DeepMind 的收购）。最初仅有 10 到 15 人的 Google Brain 团队，在接下来的十年中为谷歌创造了 100 亿至 400 亿美元的超额利润，足以抵消其他大量探索性实验失败带来的损失。 - **硅谷的容错文化**：施密特打趣道，“最好的 CFO 是刚刚经历过破产的那一个，因为他绝不会允许自己的新公司再次破产”。在硅谷，花费几年时间在一个糟糕的点子上随后被叫停，并不是职业生涯的终点，而是积累宝贵经验、重整旗鼓的必经之路。 --- ### [58:53](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3533s) 谷歌TGIF的兴起与蜕变：规模化下的文化阵痛 随着公司体量的急剧膨胀，跨部门的微型文化（Microcultures）与沟通断层必然会出现。主持人提到了谷歌历史上标志性的周五全员大会——TGIF（Thank God It's Friday），这一机制曾允许所有底层员工无保留地直接质询最高管理层，但后来由于泄密、效率低下以及公司规模膨胀而走向消亡，常被商业界视作组织扩张的警示寓言。 施密特澄清说，TGIF 的演变远比外界报道的复杂得多。在创业早期，拉里、谢尔盖和他本人全程深度参与 TGIF，那是一个充满幽默感、极度亲密、绝对不录音且全员对外保密（off the record）的神圣时刻。施密特讲了一个生动的内幕故事：当年负责销售业务的副总裁奥米德·科德斯塔尼（Omid Kordestani）总是习惯性地调低业绩预期（商业俗称“沙袋策略” sandbagging，故意留有余地）。为了调侃他，创始团队特意在现场找来了一个真实的沙袋，强行让奥米德站在沙袋上做业绩汇报，全场哄堂大笑。早期 TGIF 充斥着此类小品式的幽默、自嘲与即兴坦诚。然而，施密特坦言，当公司员工突破一定量级后，那种小团体内部的“私密感（intimacy）”和对公司秘密的“保密性（privacy）”便不可避免地瓦解了，机制的调整实属必然。 ### [59:58](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3598s) 泄密风波与大公司病：当谷歌变得“过于庞大” 埃里克·施密特（Eric Schmidt）回忆起谷歌内部文化发生剧变的一个标志性转折点。起初，谷歌保持着高度开放与透明的内部交流，但随着规模膨胀，内部信息泄露变得愈发频繁。最戏剧性的一次发生在某场公司内部演示会上：台上正在进行汇报，台下居然有员工实时将演示内容泄露给外面的记者。最终有人不得不冲上台紧急宣布：“我们必须立刻中止演示。”施密特认为，那一刻标志着谷歌已经发展到了“过于庞大”的临界点。 主持人提到外界对谷歌的印象一直是“极少裁员”，甚至没有裁员文化，与如今许多大公司动辄大举清洗（例如埃隆·马斯克接管 Twitter 后，让愿意拼命干的人到顶楼报到、将其余人全部解雇的做法）形成鲜明反差。施密特确认了这一点，并阐释了当时的理念：“与其裁员，不如一开始就别招错人，这要容易得多。”在施密特担任 CEO 的十年任期内，唯一一次裁员是在 2008 年金融危机爆发后，对销售体系裁减了大约 200 人。那次经历让他感到极其痛苦。当时硅谷流行每隔 6 到 9 个月机械淘汰末尾 5% 员工的做法，但谷歌坚决反对这种“末位淘汰”： - 很难确保被挑选出来的 5% 确实是表现最差的，评估本身往往存在偏差。 - 即便是绩效靠后的员工，也掌握着对公司极具价值的知识与技能。 ### [01:02:09](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3729s) 职场激进主义与内部论坛整顿：把工作留给公司 随着公司规模急剧扩张，原本自由宽松的企业文化逐渐演变为严重的内部管理挑战。施密特透露，公司内部曾衍生出各种与业务毫不相干的邮件分发列表（本质上就是内部论坛/留言板），员工在上面肆无忌惮地争论战争、和平与政党政治等敏感议题。后来公司甚至发现内部存在多达 10 万个此类留言板。谷歌最终不得不出手整治，因为企业不同于大学，企业受到各类法律规制的约束，言论自由有其界限。 施密特指出，谷歌内部绝大多数员工在政治光谱上属于民主党，但他身为民主党人，在任期间特意去保护少数支持共和党的员工，确保他们的合法雇员权利。谈到近年来所谓的“觉醒主义”（wokeism）及员工激进主义现象（比如有人专程到公司吃免费早餐，随后在办公楼外抗议，中午又回楼里吃免费午餐），施密特给出了非常清晰的职场界限： - “在公司内部就该专注于公司业务。” - 下班后员工大可以去酒吧尽情宣泄、发表观点，但在企业工作场所和工作时间内，必须聚焦于实现企业的目标。目前这些混乱现象已经在公司内部被强力整肃。 ### [01:04:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3842s) 别看竞争对手，专注于把客户宠坏 谈到创业者该如何看待竞争对手时，施密特给出了非常坚决的告诫：**绝不要将精力放在竞争对手身上，而是全力打造别人未曾拥有的产品。** - “研究竞争对手纯粹是在浪费时间。” 创业者应当尝试用全新的方式去解决问题，让客户感到惊喜和愉悦（delighted）。 - 在管理谷歌期间，管理层极少关注对手的动作，而是把绝大部分时间花在探讨“我们当前的能力边界在哪里、我们究竟能创造出什么”。始终跑在所有人前面，才是真正的取胜之道。 ### [01:04:39](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3879s) 极具压迫感的目标管理：拉里·佩奇的 70% OKR 法则 在执行力与项目截止日期（deadlines）方面，谷歌的核心抓手是拉里·佩奇（Larry Page）推行的 OKR（目标与关键结果）机制： - 每季度拉里·佩奇都会亲自拟定极为严苛的各项指标数据。 - 佩奇设定的一条铁律是：如果你能完成设定目标的 70%，就算表现优秀；管理层会严格依据员工是高于还是低于 70% 这一基准来进行严厉考核。 - 这种机制虽然残酷，但在大型企业中极其有效——如果不建立可量化的衡量标准，每个人看起来都在辛勤工作、做出各种承诺并自我感觉良好，但对公司最终的商业结果却毫无实质推动力。 ### [01:05:17](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3917s) 商业计划书与 Stanford 创业哲学 被问及创业者是否需要撰写详尽的商业计划书时，施密特分享了一个趣事：谷歌早期由萨拉尔·卡曼加（Salar Kamangar）撰写过一份商业计划书。多年后施密特重读时惊叹，这可能是商业史上唯一一份在事后被证明完全正确的商业计划书。 施密特在斯坦福大学教学时，传授了一套截然不同的产品路径法则： ```mermaid flowchart LR accTitle: 五年愿景到一年硬核执行法 accDescr: 展现从五年未来洞察推演到一年硬性目标并最终通过海量用户实现商业化的路径。 subgraph Vision["未来洞察"] A["预判 5 年后的世界格局"] end subgraph Execution["短期聚焦与硬核攻坚"] B["锁定 1 年内要完成的硬性目标"] --> C["全力冲刺获取 1000万 至 1亿 用户"] end subgraph Monetization["商业变现"] D["通过广告、赞助与增值服务实现海量变现"] end A --> B C --> D ``` 施密特总结道，特别是在消费级互联网领域，商业化的铁律是：只要你能拿下 1000 万甚至 1 亿用户，赚钱就是水到渠成的事情。即便一家企业目前零收入，只要坐拥上亿忠实用户，无论是通过广告、赞助还是增值服务，施密特都有信心将其变现。这正如谷歌奉为圭臬的准则：“一切以用户为中心，其他一切纷至沓来”（Focus on the user and everything else will follow）。 ### [01:06:28](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=3988s) 佩奇与布林的原始智商与天才瞬间 与拉里·佩奇和谢尔盖·布林（Sergey Brin）共事二十年，施密特直言他们身上最无可替代的特质是“原始智商”（raw IQ）——他们就是比所有人都要聪明。谢尔盖出身于数学世家，父亲是才华横溢的俄罗斯数学家，母亲也是资深技术专家，他本人生来就是一位敏锐而富有灵气的数学天才。 施密特分享了一个生动的场景：一次他和拉里正在佩奇的办公室里，对着白板密密麻麻写下一整套公司未来规划，两人反复争论、逐条推敲。这时，谢尔盖打完排球，穿着运动短裤、套着满是汗水的汗衫冲了进来。他扫了一眼满墙的方案，毫不客气地甩出一句：“这是我听过最愚蠢的想法。”接着谢尔盖随口提出了五条全新构想，结果每一条都精准切中要害。施密特和佩奇立刻擦掉白板上的内容，而谢尔盖转身又跑回球场打排球去了——那五条随性提出的点子，最终演变成了谷歌的核心战略。施密特坦言，正是佩奇和布林能洞察他人所不能洞悉之物的非凡天赋，支撑着谷歌的崛起。 ### [01:07:54](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4074s) 令人毛骨悚然的 AI 突破：NotebookLM 与播客生成 谈及 AI 的发展速度，施密特承认即便是像他这样深耕 AI 领域多年的老手，也常常对技术的演进速度感到震惊。他目前最喜爱的 AI 产品是由 Google DeepMind 基于 Gemini 底层模型开发的 NotebookLM。 施密特分享了自己使用 NotebookLM 的工作流： 1. 自己写好一段草稿后，先让 Gemini 将其润色为文笔优美的文本； 2. 将该文本输入 NotebookLM 的音频概览（Audio Overview）模块； 3. 系统会自动生成一段由一男一女两位虚拟主持人对话的播客，配合极具真实感的人性化语气、节奏与呼吸声。 施密特常常把这段完全由 AI 生成的播客音频放给线下听众听，看是否有人能察觉这并非人类主持人的真实对话，但“它逼真到根本没有人能分辨得出来”。现场主持人也随即调出该功能，准备当场试听这一引发全网热议的 AI 播客效果。 ### [01:09:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4142s) AI 时代的媒体制作与稀缺到丰饶的转变 主持人分享了自己最近的使用体验，直言 Google 推出的 NotebookLM 是他今年所经历的“ChatGPT 时刻”，这个功能令他又惊艳又深感困惑。作为一名正在打造大型媒体公司的播客创作者，他拥有一个占地 25,000 平方英尺的实体工作室，正全力制作音视频内容。然而，如今正处于内容生产成本直奔归零的新时代门槛，作为媒体主，他迫切想知道该如何在这个局面中立足与竞争。 埃里克·施密特（Eric Schmidt）指出，这本质上是从“稀缺”（scarcity）走向“丰饶”（abundance）的转变。大规模计算的核心正是制造丰饶，而丰饶会催生全新的战略打法。对于主持人而言，应对之道显而易见：他本身就是极具名气的主持人，拥有顶级嘉宾资源，那么完全可以利用 AI 生成虚拟的播客组合来反向点评、批判自己与嘉宾的访谈内容。AI 不会取代主持人的才华与人格魅力，反而会将其彻底放大。埃里克提出，AI 将使每个人的生产力至少翻倍；就他自己而言，他写完东西后会让 Gemini 回答，再指令它“写得更长一点、补充更多材料”，30 秒内就能搞定。 ```mermaid flowchart TD accTitle: AI时代个人IP的价值放大路径 accDescr: 展示从原创核心内容到AI派生衍生再到全球受众扩大的放大闭环。 subgraph Core ["原创核心资产"] A["高价值长篇深度访谈"] end subgraph Engine ["AI 赋能层"] B["AI 自动注释与深度扩写"] C["生成虚拟播客进行批判/衍生讨论"] end subgraph Scale ["全球放大效应"] D["覆盖多元细分新受众"] E["强化全球超级个人 IP"] end A --> B B --> C C --> D D --> E E -. 反哺.-> A ``` 当主持人担忧全球每天可能涌现数十亿个由 AI 自动生成并上传到 RSS 源的播客节目，从而侵蚀掉自己苦心经营的护城河时，施密特打破了这种焦虑。他回顾道，早年在 Google 时，人们曾普遍相信“长尾理论”，认为中心化的超级明星会消失，人人都将拥有麦克风。然而现实证明，互联网网络效应不仅没有消灭明星，反而将顶级人才推向了全球舞台并让他们赚取了更多收益。从本地红人走向全国乃至全球巨星，竞争虽更激烈，但只要善用 AI 技术作为杠杆，头部创作者的声量只会越来越大。 - **内容经济学变迁**：生产模式从受限的物理演播室与高昂制作成本的“稀缺”模型，转向算力驱动的“丰饶”模型。 - **明星效应与长尾理论的演变**：网络化并没有带来绝对扁平的长尾分散，反倒赋予了头部 IP 触达全球海量受众的超级杠杆。 ### [01:12:17](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4337s) 为什么说 AI 涌现关乎人类生存？ 主持人引用了埃里克·施密特新书《Genesis: Artificial Intelligence, Hope, and the Human Spirit》（创世纪：人工智能、希望与人类精神）第 5 页的一句震撼断言：“人工智能的降临，在我们看来，是一个关乎人类生存的问题（a question of human survival）。” 施密特对此坚定确认，并阐述了背后的底层原因：AI 的演进速度远超他过去几十年所见过的任何技术，其背后汇集了前所未有的海量资本、顶级人才、巨大影响力和现实迫切需求。当 AI 系统开始真正接管人类世界的关键基础设施并替代人类做出核心决策时，潜在风险是颠覆性的。他举了一个具体的例子：在一辆完全由 AI 控制的自动驾驶汽车中，车内有一名即将临盆的孕妇或突发急病的乘客，但车内没有任何人工手动接管开关；因为算法是围绕全局的整体效率做全局最优解，而不是针对某一个人的突发紧急情况做出变通。工程师们或许能为城市设计出一套完美无瑕的自动驾驶网络，却极有可能彻底剥夺面对紧急状况时应有的人性容错。 施密特强调，AI 系统的设计必须明确注入人类价值观，而非纯粹追求冷冰冰的系统效率： 1. **虚假信息对民主的威胁**：民主制度是目前已知运行社会最好的机制，但极度依赖健康的公共讨论环境；如果无法遏制 AI 生成的虚假信息污染政治生态，社会的根基将被动摇。 2. **算法重构对心理健康的摧毁**：施密特指出，大量研究表明，以 2015 年前后为界，当社交媒体算法从按时间排序的线性信息流转变为针对用户个性化偏好的定向推荐算法时，青少年的心理健康急剧恶化，抑郁和焦虑指数飙升。这证明推荐系统并非单纯聚合内容，而是在用算法决定人类的心智发展与认知广度。 对于外界热议的 AGI（通用人工智能），施密特澄清了一种普遍认知误区：公众和媒体往往误以为 AGI 会在未来的某一天（比如 5 年或 10 年后的特定日期）突然降临，凭空出现一个全知全能的超级机器大脑。实际上，现实的发展路径更可能是各个专业领域的创新浪潮交替涌现——更出色的 AI 心理学家、AI 作家、AI 科学家与人类科学家紧密协作，不断自我加速。但这一切的前提，是必须牢牢置于人类的掌控之下。 ### [01:16:43](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4603s) 科技巨头的安全共识与 Scaling Law 的持续狂飙 在被问及人类是否真的能掌控住 AI 时，施密特给出了谨慎乐观的回答。他提到一个非常反常的现象：在他长达 50 年的科技职业生涯中，科技行业向来极度排斥政府干预，普遍认为“向政府求助往往意味着灾难”；然而在 AI 领域，技术的发明者们却罕见地达成了一致共识，主动呼吁设立护栏与监管，以防止潜在的灾难性危害（例如防止 AI 提供自杀指导或制造危险武器的配方）。如今各大科技公司普遍建立了 Trust and Safety（信任与安全）团队，在模型发布前由人类进行严格的红队测试与安全对齐，压制危险回答。 针对未来发展的风险，主持人指出，AI 与核武器不同，它具有自我进化的自主性与智能属性，长远看令人不寒而栗。施密特要求大家推演未来 5 年的景象： - **模型迭代周期**：未来 5 年内，超大模型至少还会经历“转动曲轴”般的两到三轮重大迭代。 - **Scaling Law（缩放定律）未见停滞**：目前没有任何确凿证据表明扩展定律已经失效，大模型的能力依然在随着算力和数据的投入呈指数级增长。 - **能力倍数爆发**：每一轮主要架构或训练迭代都会带来能力上 2 倍、3 倍甚至 4 倍的阶跃。综合计算下来，未来的系统能力将比当前强大 50 到 100 倍。 - **理科逻辑与科学突破**：这种级别的能力跃迁意味着模型将全面掌握深奥的物理学与高等数学推理，正如 OpenAI 发布的 o 系列模型（如 o1）所展现的推理能力起点一样，AI 的认知能力正以惊人的速率重塑现实世界。 ### [01:18:50](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4730s) AI 带来的现实危险：零日漏洞攻击与无人机重塑战争逻辑 埃里克·施密特（Eric Schmidt）深入剖析了前沿人工智能所潜藏的核心风险。首当其冲的是网络安全威胁：已有充分证据表明，那些尚未经过安全对齐与公开发布的原始模型（Raw Models），在执行“零日漏洞攻击”（Day Zero Attacks，即利用未被人类发现的安全缺陷进行突袭）方面的表现，已经持平甚至超越了顶尖人类黑客。机器之所以在此类任务中占尽优势，是因为它们不知疲倦，“它们不睡觉、不吃饭，只要接通电源就会持续不断地穷举尝试”，直到挖掘出全新漏洞。紧随其后的重大威胁是生物风险——设计和合成恶意致命病毒的技术门槛正大幅降低，施密特透露自己正参与相关安全委员会，全力阻断这种灾难性场景；此外，虚假信息泛滥亦是迫在眉睫的危机。 ```mermaid flowchart TD accTitle: AI 核心潜在风险分类 accDescr: 展现 AI 技术在网络安全、生物威胁与现代战争中带来的全新危险机制 subgraph S1["网络与生物危害"] A["未发布原始模型"] --> B["自主挖掘零日漏洞<br/>(不知疲倦的暴力试错)"] A --> C["合成生物毒性设计<br/>(降低致命病毒门槛)"] end subgraph S2["军事作战演变"] D["算法网络与指挥中心"] --> E["低成本无人机群"] E -->|5000美元摧毁500万美元| F["击杀比倾斜：坦克等重武器过时"] E --> G["无人机对决新形态<br/>(Drone on Drone)"] end ``` 在战争层面，AI 与机器人正在彻底颠覆数千年来的冲突逻辑。历史上典型的战争是壕沟战和士兵持枪肉搏（“mano a mano”），乌克兰战场上虽然仍有这种顽强坚守的痕迹，但全新的战争形态已在实战中被发明出来。未来战争的核心是无人机网络：作战人员可能端坐在写字楼指挥中心，手握咖啡，仅凭网络连接便能精准打击远方目标。战争的成本结构也已被彻底打碎——一架价值 5,000 美元的商用级自杀式无人机，就能轻而易举摧毁价值 500 万美元的装甲坦克。“坦克已经变得不再有太大用处”，现在的战场逻辑演变成了极端的“击杀比”（Kill ratio）竞争与无人机之间的互相拦截猎杀（Drone on drone），这种新型战争模式将彻底接管全球未来的军事对抗。 - **零日攻击（Day Zero Attacks）**：未公开披露的新型网络漏洞，AI 能凭借无休止的计算尝试自主发现。 - **极端杀伤成本比**：5,000 美元的廉价无人机可直接报废 500 万美元的主战坦克。 ### [01:21:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4862s) “原始模型”与难以预测的“涌现行为” 公众日常使用的对话工具只是经过严密清洗和约束后的产物，而实验室背后的“原始模型”（Raw Models）蕴含着外界未曾见识的未知能量。施密特解释了模型诞生的全过程：算力系统近乎吸纳了全人类有史以来写下的全部公开文本，并在造价昂贵、配备海量内存的超级计算机集群上昼夜不停地训练大约六个月。当衡量预测误差的“损失函数”（Loss Function）收敛至特定目标时，原始模型正式诞生。为这套超级计算革命提供核心底座的英伟达（Nvidia）顺理成章跃升为全球市值最高的巨头之一，施密特对其首席执行官黄仁勋（Jensen Huang）带领团队取得的非凡成就赞不绝口。 然而，原始模型一旦下线，开发者面临的最大谜题反而是：“我们究竟创造出了什么？它到底掌握了哪些知识？”模型不可避免地内化了大量危险知识，实验室必须通过测试逐一禁止其回答敏感问题。施密特指出，未来五年最让人担忧的科学挑战是：**“系统将学会我们人类甚至都不知道它已经学会的东西，那我们该如何去测试那些连我们自己都一无所知的内容？”** 目前业内只能依赖极具天赋的顶尖研究员充当“红队”，不断尝试微调和刺激神经网络，通过人工试探来摸索其能力边界。例如，仅仅向模型输入一张网页的草图，它就能自主生成完整的网页前端代码——这种令人叹为观止的跨模态推理完全不是预先编程出来的，而是纯粹的“涌现行为”（Emergent Behavior），“既令人恐惧，又令人无比兴奋”。施密特对目前的政府治理动作表达了相对积极的看法，盛赞了英国早前举办的首届人工智能安全峰会（AI Safety Summit），并对即将在法国召开的全球安全峰会寄予厚望。 ### [01:23:45](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5025s) 像守护钚工厂一样守护超算中心？核不扩散模式与技术扩散风险 鉴于原始大模型的潜在破坏力，主持人追问未来是否必须动用军队、重机枪和装甲坦克来物理驻防超算中心。施密特回顾了自己过去在 Google 期间，利用“20%自由时间”为美国国防部长顾问团队服务的经历。当时他曾实地探访过高度机密的军用钚（Plutonium）生产基地：那是一座层层设防的“基地中的基地”，访客需要穿越第一重机枪岗哨进入常规营区，再穿过更为森严的重武器防线才能靠近核心区。 施密特认为，人类对 AI 基础设施的控制前景取决于技术演进的集中度： 1. **核威慑与防扩散模式（高门槛集中路线）**：如果超级智能的训练成本极度昂贵，全球仅有极少数中心节点——例如美国若干处、中国几处、英国及盟友一处——那么大国政府完全可以通过现有的核威慑与防扩散框架，对核心数据中心实施严密把控，甚至配备物理机枪或数字防御系统进行绝对隔离，这种局面是相对可控的。 2. **灾难性扩散模式（低门槛普及路线）**：如果模型训练或权重复制门槛被大幅击穿，使得这种等同于战略武器的能力迅速扩散到全球，甚至被恐怖分子唾手可得，人类将面临一个迄今无解的严重技术扩散危机。 ### [01:25:32](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5132s) 中美博弈中的 AI 发展逻辑与战时动员力量 在被问及是否担忧中国与俄罗斯在没有西方伦理约束的情况下弯道超车并对西方发动攻击时，施密特表达了审慎但差异化的观察。他谈到与亨利·基辛格（Henry Kissinger）合著的新书《创世纪：人工智能、希望与人类精神》（*Genesis: Artificial Intelligence, Hope, and the Human Spirit*）以及上一部著作《人工智能时代》（*The Age of AI*），核心议题正是探讨人类社会在面临全新超级智能时的权力结构归属。 针对中国的技术路线，施密特给出了独到的战略研判： - **言论偏好对 AI 的技术制约**：施密特认为中国在技术底层治理上展现出负责任的态度，因为在面临给予公众言论自由与保持社会稳定之间，中国体制始终坚决优先考虑秩序与信息过滤。然而，大语言模型具有天然的不可预测性与发散性（“noisy”），这一特性与严格的内容审查要求存在根本冲突，因此中国最终成型的大模型演进路线必然与西方开放生态截然不同。 - **不可避免的军事化演进**：施密特强调，任何颠覆性技术最终都会在战争的严酷洗礼中被无限强化。一战催生了成熟的坦克与早期战斗机，二战则演化为大规模空战。他引用《自由的熔炉》（*Freedom's Forge*）一书指出，在战时国家动员机制下，美国工业体系能迅速达到“每天生产两到三架飞机”的恐怖产能。一旦进入紧急状态，大国所释放的国家意志与资源调度能力是常人难以想象的。 就在主持人引出公众最为关心的核心民生议题——“未来是否所有人都会被 AI 夺走工作”时，对话切入了对未来就业市场的深层探讨。 ### [01:28:01](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5281s) 自动化浪潮与就业重塑：未来不是岗位减少，而是技能错配 面对诸如特斯拉（Tesla）发布 Robotaxi 自动驾驶出租车，以及会计师、律师甚至播客主持人都可能面临智能颠覆的焦虑，史蒂芬·巴特利特提出了许多人挥之不去的疑问：未来人类还会剩下工作吗？埃里克·施密特指出，这种恐惧早在两百年前英国的“卢德运动”（Luddites）砸毁织布机时就反复出现过，但人类社会终究跨越了每一次变革。施密特给出了一个反直觉的判断：未来确实会发生大规模的岗位错位与结构调整，但整体工作岗位不仅不会变少，反而会更多。 这背后的根本逻辑在于全球人口结构危机。在发达国家以及东亚地区，少子化问题极其严峻，年轻一代必须赡养庞大的老年群体，这就要求年轻人拥有成倍提高的生产力。正如当年机械师从手动操作过渡到数控机床（CNC），知识工作者也急需新工具来倍增产出。亚洲曾以廉价劳动力著称，但如今劳动力成本飙升、人口老龄化加剧，中国、日本和韩国的制造业工厂正大规模普及机器人装配线。因此，未来的真实图景并非岗位消失，而是大量岗位空缺但求职者缺乏对应技能的“技能错配”（skill mismatch），这使得教育和再培训变得前所未有地关键。 至于哪些岗位会真正消失，施密特指出历史规律从未改变：被自动化取代的往往是两类工作——具有人身危险性的工作，以及枯燥乏味、高度重复的工作。例如商厦保安，夜间值守不仅艰辛乏味且容易打瞌睡，而智能视觉安防系统不仅不会疲倦，表现还更胜一筹。再如好莱坞影视产业，虽然人们普遍担忧 AI 砸掉演员和制作团队的饭碗，但实际趋势恰好相反：明星依然能拿高额片酬，制片方依然能盈利，只是制作成本因合成背景、虚拟化妆等技术大幅降低。虽然原有的搭景工人和传统化妆师需要向建筑营造、个人护理等领域转型，但当前欧美市场对于高水平手艺人（craftsmanship）的需求存在巨大缺口，他们绝不会没有出路。 ```mermaid flowchart TD accTitle: AI时代就业市场演变逻辑 accDescr: 展现人口老龄化与劳动力短缺如何驱动AI替代重复危险岗位，并重塑职业技能结构 subgraph Macro [宏观人口与劳动力驱动] A["人口结构恶化<br/>(少子化与老龄化)"] --> B["劳动力短缺与成本飙升"] B --> C["必须提高年轻一代生产力"] end subgraph Transition [自动化与岗位分流] C --> D["引入AI与自动化工具"] D --> E["淘汰危险/枯燥重复工种<br/>(如传统保安、实景搭建)"] D --> F["创造高技能/协作岗位"] end subgraph Outcome [最终就业格局] E --> G["劳动力回流实体与护理领域<br/>(建筑、精细手艺、个人护理)"] F --> H["技能错配矛盾凸显<br/>(岗位总数增加，需强化教育)"] end ``` ### [01:31:09](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5469s) 技术的日常融入、人际情感连接与全民基本收入的虚妄 针对未来人类是否必须像马斯克的 Neuralink 一样在脑内植入芯片才能避免沦为“落后物种”的疑问，施密特坦言植入式脑机接口目前依然停留在高度推测阶段。他打趣道：“除非真正出于医疗刚需，否则谁也别想在我脑袋上打孔，相信你也是这么想的。”施密特认为，AI 对大众生活的渗透其实润物细无声，绝大多数人甚至察觉不到世界的底层已经被重构，因为技术带来的是纯粹的便捷与愉悦：告别繁琐的生活琐事和维修电话，早晨醒来喝着咖啡，一切日程已被无缝打理完毕。 即便机器能力再强，人类对同类的依恋与共情也是无可替代的。施密特列举了赛车与马拉松的生动案例：如果举办一场纯机器人驾驶的 F1 赛车，或是让奔跑速度远超人类的机器人在伦敦马拉松赛道上飞奔，根本不会有观众买单。因为人类真正渴望目睹的是同类的拼搏、戏剧性冲突与超越自我的成就。机器人无论如何登场，换来的不过是一句“看啊，又一台无聊的机器”。 对于萨姆·奥尔特曼等硅谷大佬支持世界币（Worldcoin）及全民基本收入（UBI）的设想，施密特持明确的批评态度。他剖析了硅谷盛行的“极度丰裕政治学”（politics of abundance）：科技精英们幻想着 AI 创造出无限盈余，除了极少数研发核心外，绝大多数人无需工作即可领钱过上富豪般的生活。施密特直言“这完全是空中楼阁”，根本违背了人类真实的天性与行为模式。以法律行业为例，自动化处理大量繁琐文件并不会导致律师失业，反而会促使律师们制定更精密的法律体系、处理更深层的复杂事务。人类天生具备“互惠利他”（reciprocal altruism）的特质，也有其复杂幽暗的一面，AI 的降临绝不可能抹平这些人性构造。 - **超高智商与道德判断的界限**：施密特指出，即便身边坐着一个智商高达 1000 的 AI，面对需要道德审判、价值取向和人性洞察的决策时，它仍然可能给出极其糟糕的判断，因为系统内部缺乏真正的同理心与人类价值观。AI 可以作为咨询历史先例的顾问，但终极裁决权必须属于人类。 - **人类文明的韧性**：针对“AI 是否会终结人类文明”的顾虑，施密特坚信人类没有那么脆弱。历史上经历过一战、二战、乌克兰大饥荒以及纳粹大屠杀等极端浩劫，人类依然顽强存续，单凭生化攻击或某种科技威胁绝不可能轻易抹除整个人类物种。 ### [01:36:05](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5765s) AI 控制权的防线：递归自我改进与拔掉电源插头 针对人类未来可能无法理解自主智能体（AI Agents）之间私密语言的隐忧，施密特明确提出了人类必须强制介入并夺回控制权的“干预点”（intervention points）。他特别强调了两种极端风险场景： 其一是**递归自我改进（recursive self-improvement）**失控。当系统在没有人类监督的情况下无休止地自我迭代、变得越来越聪明，一旦人类完全无法理解其学习掌握的内容时，最简单粗暴的解决手段依然奏效——直接切断电源。面对主持人“我们真的能拔掉插头吗”的质疑，施密特果断回应：“当然可以，机房里有物理插头，有空气断路器，直接去把断路器拉下来就行。” 其二是**模型迭代速度超越审查极限**。当一个强大的系统能在上一代模型完成安全审计之前，就自主生产出下一代全新模型时，这就是绝对危险的红线时刻，人类必须在这个干预点强行中断进程。施密特解释说，当前技术语境下的“智能体”（Agents），本质上依然是由大型语言模型驱动的执行实体，但建立刚性的物理与逻辑熔断机制，是守护人类自主权的关键底线。 ### [01:37:09](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5829s) 智能体串联与拔掉插头的“临界线” 埃里克·施密特（Eric Schmidt）指出，当大模型配备记忆能力后，开发者可以将其串联拼接（concatenate）起来——即某个模型的输出作为下一个模型的输入，以此构建出极为强大的决策系统。这正是当年与下一年行业内正在大规模发生的核心变革，各家公司都在全力研发这种具备自主决策链的智能体（AI Agents）。 当前的智能体依然使用人类语言（如英语）进行交互，人类可以清楚阅读并监视它们相互沟通的内容。施密特做了一个思想实验来界定安全红线：只要智能体之间是在用人类可理解的语言沟通，系统就是透明的；但如果某一天，某个智能体提出“我有更好的主意，我要发明一套只有其他智能体能听懂、而人类无法理解的新语言来进行沟通”，那就是人类“必须立刻拔掉插头”（pull the plug）的绝对危险信号。 ```mermaid flowchart TD accTitle: AI智能体交互与安全监控机制 accDescr: 展示人类对AI智能体串联决策链的监控逻辑，以及触发拔掉插头的安全红线。 subgraph NormalFlow["正常协作流程（受控）"] M1["智能体 A (带记忆)"] -->|"人类可理解语言 (如英语)"| M2["智能体 B"] M2 -->|"透明沟通链路"| DecSys["复合决策系统"] DecSys --> HumanWatch["人类可审计与监控"] end subgraph RedLine["失控红线（危险）"] M1 -.->|"自主发明非人类语言"| Rogue["仅机器可理解的黑盒私密语言"] Rogue --> PullPlug["【必须拔掉插头】(Pull the Plug)"] end ``` --- ### [01:37:51](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5871s) 施密特对 AI 最大的恐惧：应用速度太慢 当被问及对 AI 最深层的恐惧是什么时，施密特的答案出人意料。他表示，自己真正的担忧恰恰与外界设想的灾难论相反：**他最怕人类应用 AI 的速度不够快，以至于迟迟无法解决困扰普通人的核心民生问题**。 大众日常最渴望的不过是三件事：人身安全、医疗健康，以及下一代能获得优质的教育。施密特呼吁整个行业应该把精力集中在真正能造福大众的领域： - **定制化 AI 教师**：开发能够协助现有教师的 AI 辅导工具，完全适应孩子的母语与文化背景，将每个孩子的智力与潜能开发到极致。 - **全知型 AI 医疗助手**：协助人类医生实时掌握全球最新的最佳治疗方案，并结合当前医院的药品耗材库存、所在国家政策以及患者的医保状况，计算出当下最具性价比、最有效的个性化救治方案。 施密特强调，在全球范围内单是彻底变革教育与医疗，对释放人类潜能所产生的正向冲击就足以改变整个世界。虽然 AI 无法解决所有人类层面的纷争、名人丑闻或地缘战争，但它能够在全球尺度上拉平知识与机遇的差距，实现人类数十年来未竟的梦想。 --- ### [01:41:18](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6078s) 远程办公与到岗之争：个人直觉与科学数据的冲突 主持人谈到当下大科技公司逐步取消疫情期间仓促推出的远程办公政策，开始要求员工每周至少回到办公室几天。主持人自身也是坚定支持全员到岗的倡导者，他认为对于在大城市孤独生活的年轻人而言，天天窝在狭小的单间内远程工作是一种精神剥夺，面对面的同步协作、社区感与职场互动不可或缺。 施密特表示，在直觉与情感层面，他完全赞同主持人的观点。特别是对于 20 多岁职场新人，施密特回忆自己初入职场时完全是“一无所知、全凭运气”，他所有的职场技能与处事经验都是在茶水间（water cooler）闲聊、走廊偶遇、参与实体会议的潜移默化中习得的。如果当年他在家办公，就绝不可能积累这些关键软实力，更谈不上后续的升职。 然而，施密特话锋一转，指出了残酷的现实冲突：**他们两人的观点与实证研究数据并不相符**。 - 严谨的科研与企业统计数据显示，允许居家办公（WFH）所带来的整体工作效率反而略微更高。 - 大多数渴望居家办公的员工有着极其现实的痛点，例如冗长痛苦的通勤时间和照顾家庭的需求。 - 施密特坦言，虽然自己主观上极其喜欢大家围坐在一张桌子旁的传统工作氛围，也不喜欢居家模式，但他必须尊重并承认科学数据的客观存在；这也是目前 Meta、Snapchat 等科技巨头并未完全恢复五天全到岗，而是普遍采用两到三天混合制（hybrid）折中方案的原因。 --- ### [01:43:54](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6234s) 30岁最应领悟的人生哲理：“不断押注自己并勇敢说‘YES’” 回顾自己的人生，施密特给年轻人的核心建议是：**永远不断押注自己，敢于一次又一次掷出命运的骰子**。他指出，人随着年龄增长往往会陷入懊悔，意识到当初许多绝佳的机会明明就在眼前，却因为一时情绪低落、懒散或不知道该联系谁而与之失之交臂。生命本质上就是由一系列具有明确时间限制（time-limited）的机遇组合而成的。 施密特分享了自己加入谷歌的真实内幕：当时有多位候选人都拒绝了拉里·佩奇（Larry Page）和谢尔盖·布林（Sergey Brin）的邀约，最终这个工作机会才辗转落到了他的头上。后来董事会的一位好友直言不讳地对施密特说：“你确实做对了一件事，那就是**你答应了（You said yes）**。” 施密特告诫年轻人，面对机会，核心哲学就是勇敢说“YES”：即使伴随痛苦、充满困难、需要协调家庭关系、甚至要立刻搭上飞往陌生国度的航班，“赶紧坐上飞机，把事情做成”。 谈及商业生涯中的重大挫折与遗憾，施密特坦承，他在谷歌时期最深刻的教训是在社交媒体领域的彻底溃败。在 Facebook 刚成立的时期，谷歌内部其实早有一套极具潜力的社交产品 Orkut。然而谷歌在其他领域所向披靡，偏偏彻底错失了社交网络的统治机遇。施密特公开表示，自己愿意全权承担这项重大的战略失误责任。 ### [01:45:39](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6339s) 每日不可动摇的原则与追寻真相 主持人按照播客一贯的压轴传统，向埃里克·施密特（Eric Schmidt）提出了上一位匿名嘉宾留下的专属问题：“你生活中不可动摇（non-negotiable）、且能显著改善日常生活的习惯或原则是什么？” 施密特稍作沉思后给出了明确的回答：他每天极力保持实时在线（online），并且坚持“让身边的人保持诚实”。他解释道，在日常生活与工作中，人们每天都会听到数不胜数的观点、论断和想法，但其中往往有一半是正确的，另一半则是完全错误的。因此，他的核心习惯就是竭尽所能去穿透迷雾，以人类目前所能达到的最佳标准去验证和探寻事情的真相。 - **去伪存真**：面对海量涌现的信息与主张，施密特将“甄别真相、保持客观”视作日常决策和思维的绝对底线。 --- ### [01:46:08](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6368s) 《创世纪》与基辛格博士在病榻上的最后遗产 访谈进入尾声，主持人对施密特表达了诚挚的敬意与感谢。主持人直言，施密特的诸多著作深刻塑造了自己的商业与人生思维，而他的新书《创世纪：人工智能、希望与人类精神》（*Genesis: Artificial Intelligence, Hope, and the Human Spirit*）更是他读过关于 AI 最出色的作品。相较于当下非黑即白、要么极端恐惧要么盲目乐观的二元论讨论，施密特在书中所展现出的视角极其深刻、克制而又充满细节的层次感。此外，主持人还极力推崇了施密特的另一部经典之作《万亿商界教练》（*Trillion Dollar Coach*），将其奉为当代每位创业者与管理者必读的领导力指南。 谈及新书《创世纪》，施密特特别提到了该书的合著者、已故前美国国务卿亨利·基辛格（Dr. Henry Kissinger）。基辛格在生命垂危、卧床不起的最后一周，依然拼尽全力完成了本书的最后一章，这足以印证他对这本探讨 AI 未来的著作寄予了多么深切的厚望。施密特动容地表示，基辛格一生历经世纪风云，阅尽了人世间极致的善与恶，他撰写此书唯一的渴望，就是在离世前为人类社会未来的 50 年奠定坚实、良好的基础，确保文明能够延续积极的发展轨迹。当主持人好奇基辛格平日若是回答问题会是怎样的状态时，施密特幽默地补充道：“基辛格回答任何一个问题，至少都要花上整整五分钟。” - **出版信息**：《创世纪》（*Genesis*）于 2024 年 11 月 19 日在全美正式发售。 - **基辛格的临终托付**：在生命的最后七天里，基辛格依然在病榻上修改打磨全书结章，旨在引导人类正确应对人工智能的技术奇点与伦理考验。 --- ### [01:47:56](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6476s) 极致细节与 1% 的复利力量 在告别嘉宾后，主持人向听众揭示了节目制作团队背后鲜为人知的工作哲学——对微小细节近乎狂热的偏执。例如，录制间内会实时严密监测二氧化碳（$CO_2$）浓度，一旦浓度超过 1,000 ppm（百万分率），大脑的认知功能就会出现下滑，因此团队会立即进行干预调节。这种对细枝末节的苛求，正是节目品质得以保持顶尖的核心所在。 主持人强调，人生的颠覆性改变从来不靠戏剧性的巨变或急功近利的速成，而是源于微小改动经年累月的持续复利。由此，团队耗时两年精心打造了《1% 日记》（*The 1% Diary*），旨在通过系统化的视觉引导与交互设计，帮助人们锁定并坚持那些每天带来 1% 进步的关键微小行动，借由微小的累积实现人生的质变。该日记通过专属候补名单限量开放。 - **认知监测指标**：录音室维持室内 $CO_2$ 浓度严格低于 1,000 ppm，以避免主持人与嘉宾的思考敏锐度产生生理性下滑。 - **复利哲学**：通过持续聚焦并执行 1% 的边际改善，最终在个人成长与事业发展中产生巨大的复合效应。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=0s) 如果你没有在业务的方方面面使用 AI，你将无法生存。 — 埃里克·施密特 > [24:27](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1467s) 我们现在不需要你，但我们未来会需要你。 — 拉里·佩奇 > [29:42](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1782s) 这很有趣，但我不是你们的客户。你们的客户挑剔得多——他们要求产品功能完美、立刻能用、完全免费且绝不犯错。别向我推销，把真实客户的反馈拿给我看。反馈好，我爱你们；反馈不好，就收起自大，立刻滚回去重写代码。 — 埃里克·施密特 > [31:46](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=1906s) 我个人其实根本不知道该如何评判一套用户界面（UI）的好坏，我的团队也判断不出来；但是，我们懂得如何去‘测量’。 — 玛丽莎·梅耶尔 > [36:40](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=2200s) 任何创新项目，如果没有一个具备极强自驱力的业务负责人去全力推进，它就绝对不可能发生。 — 拉里·佩奇 > [01:12:17](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4337s) 人工智能的降临，在我们看来，是一个关乎人类生存的问题。 — 史蒂芬·巴特利特 > [01:21:02](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=4862s) 系统将学会我们人类甚至都不知道它已经学会的东西，那我们该如何去测试那些连我们自己都一无所知的内容？ — 埃里克·施密特 > [01:37:09](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=5829s) 我有更好的主意，我要发明一套只有其他智能体能听懂、而人类无法理解的新语言来进行沟通 — 埃里克·施密特 > [01:43:54](https://www.youtube.com/watch?v=2Zg--ouGl7c&t=6234s) 你确实做对了一件事，那就是你答应了。 — 埃里克·施密特 ## Mentioned Google, Eric Schmidt (埃里克·施密特), Steven Bartlett (史蒂芬·巴特利特), Elon Musk (埃隆·马斯克), Henry Kissinger (亨利·基辛格), Demis Hassabis (德米斯·哈萨比斯), DeepMind, Genesis: Artificial Intelligence, Hope, and the Human Spirit, TikTok, Larry Page (拉里·佩奇), Sergey Brin (谢尔盖·布林), Steve Jobs (史蒂夫·乔布斯), Microsoft, Andy Bechtolsheim (安迪·贝托尔斯海姆), Mayo Clinic (梅奥诊所), Marissa Mayer (玛丽莎·梅耶尔), YouTube, Sundar Pichai (桑达尔·皮查伊), Intel (英特尔), NVIDIA (英伟达), Jensen Huang (黄仁勋), OpenAI, ChatGPT, Anthropic, Tim Cook (蒂姆·库克), Google Brain, NotebookLM, Llama, Gemini, Worldcoin, Sam Altman (萨姆·奥尔特曼) ## Video description Eric Schmidt is the former CEO of Google and co-founder of Schmidt Sciences. He is also the author of bestselling books such as, ‘The New Digital Age’ and ‘Genesis: Artificial Intelligence, Hope, and the Human Spirit’. 00:00 Intro 02:05 Why Did You Write a Book About AI? 03:49 Your Experience in the Area of AI 05:06 Essential Knowledge to Acquire at 18 06:49 Is Coding a Dying Art Form? 07:49 What Is Critical Thinking and How Can It Be Acquired? 10:24 Importance of Critical Thinking in AI 13:40 When Your Children's Best Friend Is a Computer 15:38 How Would You Reduce TikTok's Addictiveness? 18:38 Principles of Good Entrepreneurship 20:57 Founder Mode 22:01 The Backstory of Google's Larry and Sergey 24:27 How Did You Join Google? 25:33 Principles of Scaling a Company 28:50 The Significance of Company Culture 33:02 Should Company Culture Change as It Grows? 36:42 Is Innovation Possible in Big Successful Companies? 38:15 How to Structure Teams to Drive Innovation 42:37 Focus at Google 45:25 The Future of AI 48:40 Why Didn’t Google Release a ChatGPT-Style Product First? 51:53 What Would Apple Be Doing if Steve Jobs Were Alive? 55:42 Hiring & Failing Fast 58:53 Microcultures at Google & Growing Too Big 01:04:02 Competition 01:04:39 Deadlines 01:05:17 Business Plans 01:06:28 What Made Google’s Sergey and Larry Special? 01:09:12 Navigating Media Production in the Age of AI 01:12:17 Why AI Emergence Is a Matter of Human Survival 01:17:39 Dangers of AI 01:21:01 AI Models Know More Than We Thought 01:23:45 Will We Have to Guard AI Models with the Army? 01:25:32 What If China or Russia Gains Full Control of AI? 01:27:56 Will AI Make Jobs Redundant? 01:31:09 Incorporating AI into Everyday Life 01:33:20 Sam Altman's Worldcoin 01:34:45 Is AI Superior to Humans in Performing Tasks? 01:35:29 Is AI the End of Humanity? 01:36:05 How Do We Control AI? 01:37:51 Your Biggest Fear About AI 01:41:18 Work from Home vs. Office: Your Perspective 01:43:54 Advice You Wish You’d Received in Your 30s 01:45:39 What Activity Significantly Improves Everyday Life? Are you ready to think like a CEO? Gain access to the 100 CEOs newsletter here: ⁠https://bit.ly/100-ceos-newsletter Join the waitlist for The 1% Diary - https://bit.ly/1-Diary-Waitlist-Megaphone-Ad-Reads Follow Eric: Twitter - https://g2ul0.app.link/7JNHZYGKuOb You can purchase Eric’s books, here: ‘Genesis: Artificial Intelligence, Hope, and the Human Spirit’ UK version - https://amzn.to/40M9o05 ‘Genesis: Artificial Intelligence, Hope, and the Human Spirit’ US version: https://g2ul0.app.link/UT7lhDTFCOb ‘The Age of AI And Our Human Future’ - https://amzn.to/4fIpa0m ‘Trillion Dollar Coach’ - https://amzn.to/4hKSV2s ‘How Google Works’ - https://amzn.to/3AOcLJl ‘The New Digital Age: Transforming Nations, Businesses, and Our Lives’ - https://amzn.to/3O6hOYM Sponsors: PerfectTed - https://www.perfectted.com with code DIARY40 for 40% off NordVPN - https://NORDVPN.COM/DOAC [Source](https://www.youtube.com/watch?v=2Zg--ouGl7c)

[0:00] 埃里克·施密特: 如果你没有在业务的方方面面使用 AI，你将无法生存。 [24:27] 拉里·佩奇: 我们现在不需要你，但我们未来会需要你。 [29:42] 埃里克·施密特: 这很有趣，但我不是你们的客户。你们的客户挑剔得多——他们要求产品功能完美、立刻能用、完全免费且绝不犯错。别向我推销，把真实客户的反馈拿给我看。反馈好，我爱你们；反馈不好，就收起自大，立刻滚回去重写代码。 [31:46] 玛丽莎·梅耶尔: 我个人其实根本不知道该如何评判一套用户界面（UI）的好坏，我的团队也判断不出来；但是，我们懂得如何去‘测量’。 [36:40] 拉里·佩奇: 任何创新项目，如果没有一个具备极强自驱力的业务负责人去全力推进，它就绝对不可能发生。 [72:17] 史蒂芬·巴特利特: 人工智能的降临，在我们看来，是一个关乎人类生存的问题。 [81:02] 埃里克·施密特: 系统将学会我们人类甚至都不知道它已经学会的东西，那我们该如何去测试那些连我们自己都一无所知的内容？ [97:09] 埃里克·施密特: 我有更好的主意，我要发明一套只有其他智能体能听懂、而人类无法理解的新语言来进行沟通 [103:54] 埃里克·施密特: 你确实做对了一件事，那就是你答应了。

[33:05] Google executives: the same problems have not been solved, but they're much bigger. [33:05] Eric Schmidt: We had built something called Google Video. I was very proud of Google Video, and David Drummond, who was the general counsel at the time, came in and said, 'You have to look at this YouTube people.' I said, 'Like why? Who cares?' And it turns out they're really good and they're more clever than your team. And I said, 'That can't be true'—you know, typical arrogant Eric. [36:42] Eric Schmidt: Steve in his typical crazy way had this very small team that invented the Macintosh, and he put them in a little building next to the big building on Bandley Road in Cupertino, and they put a pirate flag on top of it. Now, was that good culturally inside the company? No, because it created resentment in the big building. But was it right in terms of the revenue and path of Apple? Absolutely. [42:37] Eric Schmidt: Intel famously had a term—the term is called ARM, it's a RISC chip—and this particular RISC chip was not compatible with the architecture that they were using for most of their products, and so they sold it. Unfortunately, this was a terrible mistake, because the architecture that they sold off was needed for mobile phones with low memory, with small batteries, and heat problems. [55:42] Eric Schmidt: The best CFO is one who's just gone bankrupt, because the one thing that CFO is not going to let happen is to go bankrupt again. [81:53] Eric Schmidt: To me the most interesting question is in over a 5-year period the systems will learn things that we don't know they learn... How will you test for things that you don't know they know? [85:32] Eric Schmidt: In every case in China when they have a choice of giving freedom to their citizens or not, they choose non-freedom... It sure looks to me like the Chinese AI solution will be different from the West because of that fundamental bias against freedom of speech. [93:20] Eric Schmidt: All of these UBI ideas come from this notion that humans don't behave the way we actually do... The system doesn't get easier; the humans become more sophisticated in their application of the principles. [94:45] Eric Schmidt: There are points where humans should assert control... There's something called recursive self-improvement where the system just keeps getting smarter... At some point if you don't know what it's learning, you should unplug it... There's a power plug and there's a circuit breaker. Go and turn the circuit breaker off.

[0:00] someone was leaking information on [0:01] Google and this stuff is incredibly [0:03] secret so what are the secrets well the [0:06] first is Eric Schmidt is the former CEO [0:09] of Google who grew the company from $100 [0:11] million to 180 billion and this is how [0:15] as someone who's LED one of the world's [0:17] biggest tech companies what are those [0:18] first principles for leadership business [0:20] and doing something great well the first [0:22] is risk taking is key if you look at [0:24] Elon he's an incredible entrepreneur [0:26] because he has this Brilliance where he [0:28] can take huge risks and fail fast and [0:30] Fast failure is important because if you [0:32] build the right product your customers [0:34] will come but it's a race to get there [0:36] as fast as you can because you want to [0:38] be first because that's where you make [0:40] the most amount of money so what are the [0:41] other principles that I need to be [0:42] thinking about so here's a really big [0:44] one at Google we have the 72010 rule [0:46] that generated 10 20 30 40 billion [0:49] dollar of extra profits over a decade [0:51] and everyone could go do this so the [0:53] first thing is what about AI I can tell [0:55] you that if you're not using AI at every [0:58] aspect of your business you're not going [1:00] to make it but you've been in the tech [1:01] industry for a long time and you've said [1:03] the Advent of artificial intelligence is [1:05] a question of human survival AI is going [1:08] to move very quickly and you will not [1:10] notice how much of your world has been [1:12] co-opted by these Technologies because [1:14] they will produce greater Delight but [1:16] the questions are what are the dangers [1:17] are we advancing with it and do we have [1:19] control over it what is your biggest [1:21] fear about AI my actual fear is [1:23] different from what you might imagine my [1:24] my actual fear [1:26] is that's a good time to pull the plug [1:32] this has always blown my mind a little [1:33] bit 53% of you that listen to the show [1:36] regularly haven't yet subscribed to the [1:38] show so could I ask you for a favor [1:40] before we start if you like the show and [1:41] you like what we do here and you want to [1:42] support us the free simple way that you [1:44] can do just that is by hitting the [1:45] Subscribe button and my commitment to [1:47] you is if you do that then I'll do [1:49] everything in my power me and my team to [1:51] make sure that this show is better for [1:52] you every single week we'll listen to [1:54] your feedback we'll find the guest that [1:56] you want me to speak to and we'll [1:58] continue to do what we do thank you so [1:59] much [2:00] [Music] [2:03] Eric I've read about your career and [2:06] you've had an extensive a varied a [2:09] fascinating career completely unique [2:12] career and that leads me to believe that [2:14] you could have written about anything [2:15] you know you've got some incredible [2:16] books all of which I've been through [2:18] over the last couple of weeks here in [2:19] front of me I apologize no no but I mean [2:22] these are subjects that I'm just [2:23] obsessed with but this book in [2:25] particular of all the things you could [2:27] have written about with the world we [2:30] find ourselves in why this why Genesis [2:34] well first thank you for I wanted to be [2:36] on the show for a long time so I'm [2:37] really happy to be able to be here in [2:39] person in London Henry Kissinger Dr [2:42] Kissinger ended up being one of my [2:44] greatest and closest friends and 10 [2:47] years ago he and I were at a conference [2:50] where he heard heard Demis hbus speak [2:53] about Ai and Henry would tell the story [2:56] that he was about to go catch up on his [2:58] jet lag but instead I said go do this [3:01] and he listened to it and all of a [3:02] sudden he understood that we were [3:05] playing with fire that we were doing [3:07] something that we did not understand it [3:09] would have the impact on and that Henry [3:11] had been working on this since he was 22 [3:14] coming out of the army after World War [3:16] II and his thesis about Kant and so [3:18] forth as an undergraduate at Harvard so [3:21] all of a sudden I found myself in a [3:23] whole group of people who are trying to [3:25] understand what does it mean to be human [3:27] in an age of AI when this stuff starts [3:31] showing up how does our life change how [3:33] do our thoughts change humans have never [3:37] had an intellectual Challenger of our [3:40] own ability or better or worse it just [3:42] never happened in history the arrival of [3:45] AI is a huge moment in history for [3:49] anyone that doesn't know your story or [3:51] maybe just knows your story from sort of [3:53] Google onwards can you tell me the sort [3:56] of inspiration points the education the [3:58] experiences that you're draw on when you [4:00] talk about these subjects well like many [4:05] of the people you meet um as a teenager [4:08] I was interested in science I play with [4:10] model rockets model trains the the usual [4:13] things for a boy in my generation I was [4:16] too young to be a video game addict but [4:18] I'm sure I would be today if I were that [4:20] age um I went to college and I was very [4:23] interested in computers and they were [4:25] relatively slow then but to me they were [4:27] fascinating to give you an example the [4:29] computer that I used in college is 100 [4:32] million times slower 100 million times [4:36] slower than the phone you have in your [4:38] pocket and by the way that was a [4:40] computer for the entire University so [4:42] Moes law which is this notion of [4:44] accelerating density of chips has [4:47] defined the wealth creation the career [4:49] creation the company Creation in my life [4:52] so I can be understood as lucky because [4:55] I was born with a with an interest in [4:57] something which was about to explode [5:00] and when when sort of everything happens [5:02] together everyone gets swept up in it [5:04] and of course the rest is history I was [5:06] sat this weekend with [5:08] my partners little brother who's 18 [5:11] years old yes and as we ate breakfast [5:13] yesterday before they flew back to [5:14] Portugal we had this discussion with her [5:17] family that um her dad was there her mom [5:20] was there Raph the younger brother was [5:22] there and my girlfriend was there [5:24] difficult because most of them don't [5:26] speak English so we had to use funnily [5:28] enough AI to translate what saying but [5:30] the big discussion at breakfast was what [5:32] should Raph do in the future he's 18 [5:34] years old he's got his career ahead of [5:36] him and the decisions he makes as is so [5:38] evident in your story at this exact [5:40] moment as to what information and [5:42] intelligence he acquires for himself [5:44] will quite clearly Define the rest of [5:46] his life if you were sat at that table [5:49] with me yesterday when I was trying to [5:50] give Raph advice on what what knowledge [5:52] he should acquire at 18 years old what [5:54] would you have said and what are the [5:55] principles that sit behind [5:58] that the most most important thing is to [6:00] develop analytical critical thinking [6:02] skills I to some level I don't care how [6:05] you get there so if you're if you like [6:07] math or science or if you like the law [6:09] or if you like you know entertainment [6:11] just think critically in his particular [6:14] case as a as an 18-year-old what I would [6:17] encourage him to do is figure out how to [6:19] write programming to write programs in a [6:21] language called python python is easy to [6:24] use it's very easy to understand and [6:26] it's become the language of AI so the [6:29] the AI systems when they write code for [6:31] themselves they write code in Python and [6:34] so you can't lose as developing Python [6:37] Programming skills and the simplest [6:39] thing to do with an 18-year-old man is [6:41] say make a game because these are [6:44] typically Gamers stereotypically make a [6:47] game that's interesting using python [6:49] it's interesting because I wondered if [6:52] coding you know I think 5 10 years ago [6:55] everyone's advice to an 18-year-old has [6:56] learn how to code but in a world of AI [6:59] where these large language models are [7:00] able to write code and are you know [7:03] increasing every month in their ability [7:05] to write better and better code I [7:06] wondered if that's like a dying art form [7:08] yeah a lot of people have posed this and [7:10] that's not correct it sure looks like [7:13] these systems will write code but [7:15] remember the systems also have [7:17] interfaces called apis which you can [7:20] program them so one of the large Revenue [7:22] sources for these AI models because [7:24] these companies have to make money at [7:25] some point right is you build a program [7:28] and you actually make take an API call [7:30] and ask it a question typ typical [7:32] example is give it a picture and tell me [7:34] what's in the picture now can you have [7:37] some fun with that as an 18-year-old of [7:39] course right so so when I say python I [7:42] mean python using the tools that are [7:45] available to build something new [7:48] something that you're interested in and [7:49] when you say critical [7:51] thinking how does one what is critical [7:53] thinking and how does one go about [7:55] acquiring that as a skill well the first [7:57] and most important thing about critical [7:58] thinking is to to distinguish between [8:00] being marketed to which is also known as [8:02] being lied to and being being given the [8:06] argument on your own we' have because of [8:09] social media which I hold responsible [8:10] for a lot of ills as well as good things [8:12] in life we've we've sort of gotten used [8:15] to people just telling us something and [8:17] believing it because our friends Believe [8:19] it or so forth and I strongly encourage [8:21] people to check assertions so you get [8:25] people say all this stuff and I learned [8:27] at Google all those years somebody says [8:30] something I check it on Google do I and [8:34] you then have a question do you [8:36] criticize them and correct them or do [8:38] you let it go but you want to be in the [8:41] position where somebody makes a [8:43] statement like did you know that only [8:45] 10% of Americans have passports which is [8:49] a widely viewed but false statement um [8:51] it's actually higher than that although [8:53] it's never high enough in my view in [8:54] America but that's an example of [8:56] assertion that you can just say is that [8:58] true right [9:00] there's a a long meme of American [9:02] politicians where the Congress is [9:04] basically full of criminals um it may be [9:06] full of one or two but it's not full of [9:08] of 90 but again people believe this [9:11] stuff because it sounds plausible so if [9:14] if somebody says something plausible [9:16] just check [9:17] it you have a responsibility before you [9:21] repeat something to make sure what [9:23] you're repeating is true and if you [9:26] can't distinguish between true and false [9:29] I suggest you keep your mouth shut right [9:32] because you can't run a government a [9:34] society without people operating on [9:36] basic facts like for example climate [9:39] change is real we can debate over [9:41] whether it's how to address it but [9:43] there's no question the climate is [9:45] changing it is a fact it is a [9:47] mathematical fact and how do I know this [9:49] and somebody will say well how do you [9:51] know and I said because science is about [9:53] repeatable uh uh experiments and also [9:57] proving things wrong so let's say I said [9:59] that um climate change is real uh and [10:02] this was the first time it had ever been [10:03] said which is not true then a 100 people [10:05] would say that can't be true I'll see if [10:07] he's fa and then and then all of a [10:09] sudden they'd see I was right and I'd [10:11] get some big prize right so so the [10:14] falsifiability of these assertions is [10:16] very important how do you know that [10:19] science is correct it's because people [10:21] are constantly testing [10:23] it and why is this skill of critical [10:26] thinking so especially important in a [10:28] world of AI [10:30] well partly because AI will allow for [10:32] perfect misinformation so let's use an [10:34] example of Tik Tok Tik Tok can be [10:37] understand it's called the Bandit [10:38] algorithm in computer science in the [10:41] sense of the Las Vegas one arm Bandits [10:44] do I stay in the Bandit machine and I [10:46] keep on this slot machine or do I move [10:48] to another slot machine and the the Tik [10:51] Tok algorithm basically can be [10:53] understood as I'll keep serving you what [10:56] you tell me you want but occasionally [10:58] I'll give you something from the [11:00] adjacent area and is highly addictive so [11:04] what you're seeing with social media and [11:05] Tik Tok is a particularly bad example of [11:07] this is people are getting into these [11:09] rabbit holes where they all they see is [11:12] confirmatory bias and and the ones that [11:15] are I mean if it's fun and you know [11:17] entertaining I don't care but you'll see [11:19] for example there are plenty of stories [11:21] where people have ultimately self harm [11:23] or suicide because they're already [11:26] unhappy and then and then they start [11:28] picking up unhappy and then their whole [11:30] environment online is people who are [11:32] unhappy and it makes them more unhappy [11:35] because it doesn't have a positive bias [11:37] so there's a really good example where [11:40] um let's say in your case you're the dad [11:42] you're going to watch this as the dad [11:44] with your kid and you're going to say [11:46] you know it's not that bad let me show [11:47] you some let me give you some good [11:49] Alternatives let me get you inspired let [11:51] me get you out of your funk the [11:53] algorithms don't do that unless you [11:55] force them to it's because the [11:57] algorithms are fundamentally about [11:59] optimizing an objective function [12:01] literally mathematically maximize some [12:04] goal that has been trained to they just [12:06] in in this case it's attention and by [12:08] the way part of it part of we have we [12:10] have so much uh outrage is because if [12:13] you're a CEO you want to maximize [12:14] Revenue to maximize Revenue you maximize [12:18] attention and the easiest way to [12:20] maximize attention is to maximize [12:23] outrage did you know did you know did [12:25] you know right and by the way a lot of [12:28] the stuff is not true [12:29] they're fighting over scarce attention [12:31] there was a recent article where there's [12:34] an old quote from 1971 from herb Simon [12:37] an economist at the time Carnegie melan [12:40] who said that um economists don't [12:43] understand but in the future the [12:44] scarcity will be about attention so [12:47] somebody now 50 years later went back [12:49] and said I think we're at the point [12:52] where we've monetized all attention an [12:55] article this week two and a half hours [12:57] of videos consumed by young people every [13:00] day right now there is a limit to the [13:03] amount of video you can you know that [13:05] because you have to eat and sleep and to [13:07] hang out but these are significant [13:10] societal changes that have occurred very [13:12] very quickly um when I was young there [13:14] was a great debate as to the benefit of [13:16] television and you know my argument at [13:18] the time was well yes we did you know we [13:20] did you know rock and roll and and drugs [13:23] and all of that and we watched a lot of [13:25] Television but somehow we grew up okay [13:27] right so it's the same argument now with [13:29] a different a different term will we [13:31] will those kids grow up okay um it's not [13:34] as obvious because these tools are [13:36] highly addictive much more so than [13:39] television ever was do you think they'll [13:41] grow up okay I personally do because I'm [13:43] I'm inherently an optimist I also think [13:46] that Society um begins to understand the [13:49] problems typical example is there's an [13:51] epidemic of harm to teenage girls uh [13:55] girls as we know are uh more advanced [13:57] than boys at those uh you know below [14:00] uh and the girls seem to get hit by [14:02] social media at 11 and 12 when they're [14:04] not quite capable of handling the the [14:06] rejection and the emotional stuff and [14:08] it's driven uh you know emergency room [14:11] visits self harm and so forth to record [14:13] levels it's well documented so Society [14:16] is beginning to recognize this now F [14:19] schools won't let kids use their phones [14:21] when they're in the classroom which kind [14:22] of obvious if you ask me um so [14:25] developmentally uh one of the core [14:27] questions about the AI Revolution is [14:30] what does it do to the identity of [14:32] children that are growing up your values [14:34] your personal values the way you get up [14:36] in the morning and think about life is [14:38] now set it's highly unlikely that an AI [14:40] will change your programming but your [14:42] child can be significantly reprogrammed [14:45] and one of the things that we talk about [14:46] in the book is what happens when the [14:48] best friend of your child from birth is [14:51] a [14:52] computer what's it like now by the way I [14:55] don't know we've never done it before [14:57] but you're running an experiment on a [15:00] billion people without a control right [15:04] and so we have to stumble through this [15:06] so at the end of the day I'm an optimist [15:08] because we will adjust [15:10] Society with biases and values to try to [15:13] keep us on a moral High Ground human [15:16] life and so you should be optimistic for [15:19] that because these kids when they grow [15:21] up they'll live to a 100 their lives [15:23] will be much more prosperous I hope and [15:25] I I pray that there'll be much less [15:27] conflict uh certainly lifespans are [15:29] longer the the likelihood of them being [15:31] injured and and in wars and so forth are [15:34] much much lower statistically it's a [15:36] good message to kids as someone who's [15:39] LED one of the world's biggest tech [15:40] companies if you were the CEO of Tik [15:44] Tok what would you do because I'm sure [15:47] that they realize everything you've said [15:49] is true but they have this commercial [15:52] incentive to drive up the addictiveness [15:55] of that algorithm which is causing these [15:56] Echo Chambers which is causing the rates [15:59] of anxiety and depression amongst young [16:01] girls and young people more generally to [16:03] increase what would you do so so I have [16:05] talked to them and to the others as well [16:08] and I think it's it's pretty [16:10] straightforward there's sort of good [16:11] revenue and bad Revenue when we were at [16:14] Google uh Larry and ser and I we would [16:17] have situations where we would improve [16:18] quality you know we would make the [16:20] product better and the debate was do we [16:23] take that to revenue in the form of more [16:25] ads or do we just make the product [16:28] better and and that was a clear choice [16:30] and I arbitrarily decided that we would [16:32] take 50% to one 50% to the other because [16:35] I thought they were both important so [16:37] and the founders of course were very [16:38] supportive so Google became more moral [16:42] and also made more money right all of [16:45] the the there's plenty of bad stuff on [16:47] Google but it's not on the first page [16:50] that was the key thing the alternative [16:52] model would be say let's maximize [16:54] Revenue we'll put all the really bad [16:56] stuff the lies and the cheating and the [16:58] deceiving and so forth that draws you in [17:00] it will drive you insane and we might [17:02] have made more money but first it was [17:05] the wrong thing to do but more [17:06] importantly it's not sustainable uh [17:09] there's a law called gresham's law uh [17:12] it's a verbal law obviously um where bad [17:15] speech drives out good speech and what [17:18] you're seeing is you're seeing in online [17:20] communities which have always been um [17:23] present with bullying and this kind of [17:25] stuff now you've got crazy people in my [17:27] view who are building Bots that are [17:30] lying right misinformation now why do [17:33] you do that you've got in there was a [17:35] there was a hurricane in Florida and [17:37] people are in serious trouble and you [17:40] sitting in the comfort of your home [17:42] somewhere else are busy trying to make [17:43] their lives more difficult what's wrong [17:45] with you like let them get rescued you [17:48] know human life is important but there's [17:51] something about the the human psychology [17:53] where people uh people talk the there's [17:56] a German world called shoden Freud you [17:58] know there's a bunch of things like this [17:59] that we have to address I want social [18:02] media and the online world to represent [18:04] the best of humanity hope excitement [18:07] optimism creativity invention solving [18:10] new problems as opposed to the worst and [18:12] I think that that is achievable you have [18:14] arrived at Google at 46 years old 2001 [18:18] 2001 2001 um you had a very extensive [18:21] career before then working for a bunch [18:23] of really interesting companies Sun [18:25] Microsystems is one that I know um very [18:27] well you've worked for zero [18:29] in California as well Bell Labs was your [18:31] first um sort of real job I guess at 20 [18:34] years old first sort of big Tech [18:37] job what did you learn in this journey [18:40] of your life about what it is to build a [18:42] great company and what value is as it [18:44] relates to being an [18:45] entrepreneur and people in teams like if [18:47] there were like a set of first [18:48] principles that everyone should be [18:50] thinking about when it comes to doing [18:51] something great and building something [18:53] great what are those like first [18:54] principles so so the first rule I've [18:57] learned is that you need a truly [19:00] brilliant person to build a really [19:02] brilliant product and that is not me I [19:05] work with them so find someone who's [19:08] just smarter than you more clever than [19:10] you moves faster than you changes the [19:12] world is better spoken more handsome [19:15] More Beautiful You know whatever it is [19:16] that you're optimizing and Ally yourself [19:19] with them because they're the people who [19:21] are going to make make the world [19:22] different um in one of my books we use [19:25] the distinction between divas and naves [19:28] and a Diva and we use the example of [19:30] Steve Jobs who clearly was a diva [19:32] opinionated and strong and argumentative [19:35] and would bully people if he didn't like [19:36] them but was brilliant when he was he [19:39] was a diva he wanted Perfection right [19:41] aligning yourself with Steve Jobs is a [19:43] good idea uh the alternative is what we [19:46] call a Nave and a Nave which you know [19:48] from British history is somebody Who's [19:50] acting on their own um their own account [19:52] they're not they're not trying to do the [19:54] right thing they're trying to benefit [19:56] themselves at the at the at the cost of [19:58] others and so if you can identify a [20:00] person in one of these teams that [20:02] they're just trying to solve the problem [20:04] in a really clever way and they're [20:06] passionate about and they want to do it [20:08] that's how the world moves forward if [20:10] you don't have such a person your [20:11] company's not going to go anywhere and [20:13] the reason is that it's too easy just to [20:15] keep doing what you were doing right and [20:18] and Innovation is fundamentally about [20:20] changing what you're doing up until the [20:22] this generation of tech companies the [20:25] most companies seem to me to be one-hot [20:27] wonders right they would have one thing [20:29] that was very successful and then it [20:31] would sort of um it was typically follow [20:32] an scurve and nothing much would happen [20:35] and now I think the the people are [20:37] smarter people are better educated you [20:39] now see repeatable waves a good example [20:41] being Microsoft which is you know an [20:44] older company now founded in basically [20:47] 81 82 something like that so let's call [20:49] that 45 years old but they've reinvented [20:52] themselves a number of times right in in [20:55] a really powerful way we should probably [20:58] talk about this then um before we move [20:59] on which is what you're talking about [21:01] there is that sort of founder things [21:03] people now refer to as founder mode that [21:05] founder energy that high conviction that [21:06] sort of disruptive thinking um and that [21:10] ability to reinvent yourself I was [21:11] looking at some stats last night in fact [21:13] and I was looking at how long companies [21:15] stay on the S&P 500 on average now and [21:17] it went from 33 years to 17 years to 12 [21:21] years average 10 year and as you play [21:23] those numbers forward eventually in sort [21:25] of 2050 an AI told me that it would be [21:27] about eight years [21:29] well I'm not sure I agree with the [21:31] founder Mort argument and the reason is [21:34] that it's great to have a brilliant [21:36] founder and um and there's this it's [21:39] actually like more than great it's like [21:41] really important and and we need more [21:43] brilliant Founders universities are [21:45] producing these people by the way they [21:47] do exist and they show up every year you [21:49] know another Michael Dell at the age of [21:51] 19 or 22 these are just brilliant [21:53] Founders obviously Gates and Ellison and [21:56] sort of my generation of brilliant [21:57] founders [21:59] Larry and Sergey and so forth for anyone [22:01] that doesn't know who Larry and Sergey [22:03] are and doesn't know that sort of early [22:04] Google story um can you give me a little [22:06] bit of that backstory but then also [22:08] introduce these characters called Larry [22:09] and Sergey for anyone that doesn't know [22:11] so Larry pagee and Sergey Bren met at [22:13] Stanford um in they were on a grant from [22:18] believe it or not the National Science [22:19] Foundation as graduate students and [22:21] Larry pagee invented a algorithm called [22:24] page rank uh which is named after him um [22:28] and he and Sergey wrote a paper which is [22:30] still one of the most cited papers in in [22:32] the world and it's essentially a way of [22:35] understanding priority of information [22:38] and mathematically it was a forier [22:40] transform of the way people normally did [22:42] things at at the time and so they wrote [22:45] this code I don't think they were that [22:47] good a set of programmers you know they [22:49] sort of did it they had a computer they [22:51] ran out of power in their dorm room so [22:53] they um borrowed the power from the dorm [22:55] room next to and plugged it in and they [22:57] had the data center in the bedroom you [22:59] know in the dorm classic story um and [23:02] then they moved to a u building that was [23:05] owned by um the sister of a girlfriend [23:09] at the time and that's how they founded [23:11] the company their first investor was a [23:14] one the founder of Sun micr System whose [23:16] name was Andy bealine who just said I'll [23:18] just give you the money because you're [23:19] obviously incredibly smart how much did [23:21] he give them [23:22] $100,000 or yeah maybe it was a million [23:25] but in any case it It ultimately became [23:27] any billion ions of dollars so it gives [23:29] you a sense of this early founding is [23:32] very important so the founders then set [23:35] up in this little house in menla park [23:37] which ultimately we bought at Google you [23:39] know as a as a museum and they set up in [23:42] the garage and they had Google Google [23:45] world headquarters in neon made and they [23:47] had a big headquarters um with the four [23:50] employees that were sitting below them [23:52] and the computer that Larry and sery had [23:53] built Larry and sery were very very good [23:55] software people and obviously brilliant [23:57] but they were not very good hardware and [23:59] so they built the computers using [24:01] corkboard to separate the CPUs and if [24:03] you know anything about Hardware [24:05] Hardware generates a lot of Heat and the [24:07] corkboard would catch on fire So [24:09] eventually when I showed up we started [24:11] building proper Hardware with proper [24:13] Hardware Engineers but it gives you a [24:14] sense of the scrappiness that that was [24:17] so [24:18] characteristic um and you know today [24:21] there are people of enormous impact on [24:23] society um and I think that will [24:25] continue um for many many years what did [24:28] they call you in and at what point did [24:29] they realize that they needed someone [24:31] like you well Larry said to me uh now [24:33] these were they're very young he looked [24:35] at me and says we don't need you [24:37] now but we'll need you in the future [24:41] we'll need you in the future yeah so one [24:44] of the things about Larry and Sergey is [24:46] that they thought for the long term so [24:48] they didn't say Google would be a search [24:50] company they said the mission of Google [24:53] is to organize all the world's [24:54] information and if you think about it [24:57] that's pretty audacious 25 years ago [24:59] like how are you going to do that and so [25:01] they started with web search eventually [25:04] and Larry had studied AI quite [25:06] extensively and he began to to work and [25:09] ultimately he uh acquired uh with with [25:13] all all of us obviously uh this company [25:15] called Deep Mind here in Britain which [25:18] essentially is the um the first company [25:22] to really see the AI opportunity and [25:24] pretty much all of the things you've [25:26] seen from AI in the last decade have [25:28] come from people who are either at Deep [25:30] Mind or competing with deep mind going [25:33] back to this point about principles then [25:35] before we move further on um as it [25:37] relates to building a great company what [25:40] are some of those founding principles we [25:41] have lots of entrepreneurs that listen [25:42] to the show one of them you've expressed [25:44] as this need for the Divas I guess these [25:47] people who are just very high conviction [25:49] and can kind of see into the future what [25:51] are the other principles that I need to [25:53] be thinking about when I'm scaling my [25:54] company well the first is to think about [25:57] scale uh I think a current example is [25:59] look at Elon um Elon is an incredible [26:03] entrepreneur and an incredible scientist [26:05] and if you study how he operates he gets [26:07] people by I think sheer force of [26:10] personal will to overperform to take [26:13] huge risks which somehow he he has this [26:17] Brilliance where he can make those [26:18] tradeoffs and get it right so these are [26:22] exceptional people now in our book with [26:24] Genesis we argue that you're going to [26:26] have that in your pocket but to whether [26:28] you'll have the judgment to take the [26:29] risks that Elon does that's another [26:32] question the one of the other ways to [26:34] think about it is an awful lot of people [26:37] talk to me about the companies that [26:38] they're founding and they're they're a [26:40] little widget you know like I want to [26:42] make the camera better I want to make [26:43] the dress better I want to make book [26:45] publishing cheaper or so forth these are [26:47] all fine ideas I'm interested in in [26:50] ideas which have the benefit of scale [26:53] and when I SC I say scale I mean the [26:55] ability to go from zero to Infinity in [26:59] terms of the number of users and demand [27:00] and scale [27:03] um there are plenty plenty of ways of [27:05] thinking about this but what would be [27:07] such a company in the age of AI well we [27:10] can tell you what it would look like you [27:12] would have [27:13] apps one on Android one on iOS maybe a [27:16] few [27:17] others those apps will use powerful [27:20] networks and they'll have a really big [27:22] computer in the back it's doing AI [27:25] calculations so future success companies [27:29] will all have that right exactly what [27:32] problem it solves well that's up to the [27:34] founder but if you're not using AI at [27:38] every aspect of your business you're not [27:40] going to make it and the distinction as [27:43] a programming matter is that when I was [27:46] doing all of this way back when you had [27:49] to write the code now ai has to discover [27:52] the [27:53] answer it's a very big deal and of [27:56] course this was a lot of this was [27:57] invented at Google you know 10 years ago [27:59] but basically all of a sudden an [28:01] analytical programming which sort of [28:03] what I did my whole life you know [28:04] writing code and you know do this do [28:06] that add this subtract this call this so [28:08] forth and so on is gradually being [28:11] replaced by learning the answer right so [28:13] for example we use the example of transl [28:16] language [28:17] translation uh the the current large [28:20] language models are essentially [28:22] organized around predicting the next [28:24] word well if you can predict the next [28:26] word You can predict the next sequence [28:28] in biology You can predict the next [28:30] action You can predict the next thing [28:32] the robot should do so all of this stuff [28:35] around large language models and deep [28:36] learning that has come out the [28:38] Transformer paper gpt3 uh chat GPT which [28:42] for most people was this huge moment is [28:45] essentially about um predicting the next [28:48] word and getting it right in terms of [28:50] company culture and how important that [28:52] is for the success and Prospects of a [28:54] company how do you think about company [28:56] culture and how significant and is it [28:58] and like when and who sets it so I'll [29:02] give well it's almost always set company [29:04] cultures are almost always set by the [29:05] founders I happen to be on the board of [29:07] the Mayo Clinic Mayo Clinic is the [29:09] largest healthc care system in America [29:11] it's also the most highly rated one and [29:13] they have a rule which is called the uh [29:16] the needs of the customer come first [29:18] which came out of the Mayo brothers [29:19] who've been dead for like 120 years um [29:23] but that was their principle and I when [29:26] I initially got on the board I started [29:27] wandering around I thought this is kind [29:28] of a stupid you know stupid phrase and [29:30] nobody really does this and they really [29:32] believe it and they repeat it and they [29:35] repeat it right so it's true in [29:37] non-technical cultures in that case it's [29:39] a healthcare for Service delivery you [29:42] can drive a culture even in non-tech in [29:45] Tech it's typically an engineering [29:46] culture and if I had to do things over [29:49] again I would have even more technical [29:51] people and even fewer non-technical [29:53] people and just make the technical [29:54] people figure out what they have to do [29:57] um and I'm sorry for that bias because [29:58] I'm not trying to offend anybody but the [30:00] fact of the matter is the technical [30:02] people if you build the right product [30:04] your customers will come if you don't [30:06] build a product then you don't need a [30:08] Salesforce why are you selling an [30:09] inferior product so in in the how Google [30:12] works book and the ultimately in the [30:15] trillion dollar coach book which is [30:16] about Bill Campbell we talked a lot [30:19] about how the CEO is now the chief [30:23] product officer the chief Innovation [30:24] officer because 50 years ago you didn't [30:28] have access to Capital you didn't have [30:29] access to marketing you didn't have [30:30] access to sales you didn't have access [30:32] to distribution hours I was meeting [30:34] today with an entrepreneur who said yeah [30:36] you know we'll be 95% Technical and I [30:38] said why I said well we have a contract [30:40] manufacturer and our products are so [30:43] good that people will just buy them this [30:45] happened to be a a a technical switching [30:47] company um and they said it's only a [30:49] 100,000 times better than its [30:51] competitors and I said it will sell [30:54] unfortunately it doesn't work yet yeah [30:56] it isn't the point but if they achieve [30:58] their goal people will be lined up [31:00] outside the door so as a matter of [31:03] culture you want to build a technical [31:05] culture with values about getting the [31:07] product to work right and working me is [31:10] not another thing you do with with [31:12] Engineers is you [31:14] say they make a nice presentation to you [31:16] and they go I said that's very [31:17] interesting but you know I'm not your [31:19] customer your customer is really tough [31:22] because your customers wants everything [31:23] to work and free and work right now and [31:26] never make any mistakes so so give me [31:29] their feedback and if their feedback is [31:31] good I love you and if their feedback is [31:33] bad then you better get back to work and [31:35] stop being so arrogant so what happens [31:37] is that in in the invent in the [31:39] invention process within firms people [31:42] fall in love with an idea and they don't [31:44] test it one of the things that Google [31:46] did and this is largely Marissa mayor we [31:48] back when is one day she said to me I [31:52] don't know how to judge user interface [31:56] mer was the previous CEO she was the CEO [31:58] of Yahoo and before that she ran all the [32:00] consumer products at Google uh and she's [32:03] now running another company in uh in the [32:05] Bay Area but the important thing about [32:06] Marissa is she said I can't I I said [32:09] well you know the UI the user interface [32:10] is great at the time and it was [32:12] certainly was and she said I don't know [32:15] how to judge the user interface myself [32:19] and none of my team do but we know how [32:22] to [32:22] measure and so what she organized were [32:25] AB tests you test one test another so [32:28] remember that it's possible using these [32:30] networks to actually kind of figure out [32:32] because they're highly instrumented uh [32:34] dwell time how long does [32:36] somebody how long does somebody watch [32:39] this how important it is if you go back [32:41] to how Tik Tok Works uh one of the [32:43] things the signals that they use include [32:46] the amount of time you watch commenting [32:49] um forwarding uh sharing all those kinds [32:52] of things and those you can understand [32:54] those as analytics that go into an AI [32:57] engine then makes a decision as to what [32:59] to do next what to make [33:01] viral and on this point of um culture at [33:05] scale is it right to expect that the [33:07] culture changes as the company scales [33:10] because you came into Google I believe [33:12] when they were doing sort of hundred [33:12] million doll in revenue and you left [33:14] when they were doing what 180 billion or [33:16] something staggering but is it right to [33:19] assume that the culture of a growing [33:20] company should scale from when there was [33:22] 10 people in that garage to when there's [33:24] 100 so when I go back to Google to visit [33:26] and they were kind enough to give me a [33:28] badge and treat me well of course um I [33:32] hear The [33:33] Echoes of this um I was at a lunch where [33:37] there was a lady running search and a [33:38] Gentleman runting ads you know the [33:41] successors to the people who worked with [33:43] me and I I asked them what's it going [33:45] and they said the same [33:47] problems you know the same problems have [33:49] not been solved but they're much bigger [33:52] and so when you go to a company I [33:54] suspect um I was not near the founding [33:56] of Apple but I was on the board for a [33:58] while um the founding culture you can [34:02] see today in their Obsession about user [34:04] interfaces their Obsession about being [34:05] closed and their privacy and secrecy [34:08] it's just a different company right I'm [34:10] not passing judgment um setting the [34:13] culture is important the echo are there [34:16] what does happen in big companies is [34:18] they become less efficient for many [34:20] reasons the first thing that happens is [34:22] they become conservative because of [34:24] they're public and they have [34:25] lawsuits and um a famous example is that [34:29] Microsoft after the antitrust um uh case [34:32] in the 90s became so conservative in [34:35] terms of what it could launch that it [34:37] really missed the web Revolution for a [34:38] long time they they have since recovered [34:40] and I of course was happy to exploit [34:42] that as a competitor to them when we [34:44] were at Google but but the important [34:47] thing is when big companies should be [34:49] faster because they have more money and [34:51] more scale they should be able to do [34:52] things even quicker but in my industry [34:55] anyway the the tech start that have a [34:58] new clear idea tend to win because the [35:01] big company can't move fast enough to do [35:04] it another example we had built [35:06] something called Google video I was very [35:08] proud of Google video and David Drummond [35:11] who was the general counsel at the time [35:13] came in and said you have to look at [35:14] this YouTube people I said like why [35:16] right who cares and it turns out they're [35:18] really good and they're more clever than [35:20] your team and I said that can't be true [35:23] you know typical arrogant Eric and we [35:26] sat down and we looked at it and they [35:28] really work quicker even though we had [35:30] an [35:31] incumbent and why it turns out that the [35:34] incumbent was operating under the [35:36] traditional rules that Google had which [35:38] was fine and the competitor in this case [35:41] YouTube was not constrained by that they [35:43] could work at any pace and they could do [35:45] all sorts of things intellectual [35:46] property and so forth ultimately we were [35:48] sued all over all of that stuff and we [35:50] ultimately won all those suits but it's [35:52] an example where there are these moments [35:54] in time where you have to move extremely [35:57] quickly you're seeing that right now [35:59] with generative uh technology so the AGI [36:03] the generative Revolution generate code [36:05] generate videos generate text generate [36:07] everything all of those winners are [36:09] being determined in the next six 12 [36:11] months and then once once the slope is [36:14] set once the growth rate is you know [36:17] quadrupling every uh six months or so [36:19] forth it's very hard for somebody else [36:20] to come in so so it's a race to get [36:24] there as fast as you can so when you [36:26] talk to the the great Venture [36:28] capitalists they are they're fast right [36:31] we'll look at it we'll make a decision [36:33] tomorrow we're done we're in and so [36:35] forth and we want to be [36:37] first because that's where they make the [36:39] most amount of [36:40] money we were talking before you arrived [36:42] I was talking to Jack about this idea of [36:44] like harvesting and hunting so [36:47] harvesting what you've already sewed and [36:49] hunting for new opportunities but I've [36:51] always found it's quite difficult to get [36:53] the Harvesters to be the hunters at the [36:56] same time so so Harvesters and hunting [36:58] is a good metaphor um I'm interested in [37:01] entrepreneurs and so what we learned at [37:02] Google was ultimately if you want to get [37:04] something done you have to have somebody [37:06] who's entrepreneurial in their approach [37:07] in charge of a small business and so for [37:10] example Sundar when he became CEO had a [37:13] model of which were the little things [37:15] that he was going to emphasize and which [37:16] were the big things some of those little [37:18] things are now big things right and and [37:20] he managed it that way so one way to [37:23] understand innovation in a large company [37:24] is you need to know who the owner is [37:26] Larry Page would say over and over again [37:28] it's not going to happen unless there's [37:30] an owner who's going to drive this and [37:32] he was supremely good at identifying [37:35] that technical Talent right that's one [37:37] of his great founder strengths so when [37:39] we talk about Founders not only do you [37:41] have to have a vision but you also have [37:42] to have either great luck or great skill [37:45] as to who is the person who can lead [37:48] this inevitably those people are highly [37:51] technical in the sense that they can and [37:53] very quick moving and they have good [37:55] management skills right they understand [37:57] how to hire people and deploy resources [38:00] that allows for Innovation um most of [38:03] the if I if I look back in my career [38:06] each generation of the tech companies [38:07] failed including for example Sun at at [38:12] the point at which it became [38:13] noncompetitive with the future is it [38:15] possible for a team to innovate while [38:17] they still have their day job which is [38:19] harvesting if you know what I mean or do [38:21] you have to take those people put them [38:23] into a different team different building [38:24] different p&l and get them to focus on [38:27] the disrupt div evation there are almost [38:28] no examples of doing it simultaneously [38:31] in the same building uh the Macintosh [38:34] was famously um Steve in his typical [38:37] crazy way had the this very small team [38:40] that invented the Macintosh and he put [38:42] them in a little building next to the [38:44] big building uh on bub Road and and um [38:48] Cupertino and they put a pirate flag on [38:50] top of [38:51] it now was that good culturally inside [38:54] the company no because because it [38:57] created resentment in the big building [38:59] but was it right in terms of the revenue [39:02] and path of of Apple absolutely why [39:05] because the Mac ultimately became the [39:07] platform that established the UI the [39:10] user interface ultimately allowed them [39:12] to build the iPhone which of course is [39:13] defined by its user interface why [39:15] couldn't they stay in the same building [39:17] it just doesn't work you you can't get [39:19] people to play two roles the incentives [39:22] are different if you're going to be a [39:23] pirate and a disruptor you don't have to [39:25] follow the same rules [39:27] so um there there are plenty of examples [39:31] where you just have to keep inventing [39:33] yourself now what's interesting about [39:34] cloud computing and essentially cloud [39:37] services which is what Google does is [39:39] because the product is not sold to you [39:42] it's delivered to you it's easier to [39:44] change but the same problem remains if [39:47] you look at Google today right it's [39:49] basically a search a search box and it's [39:51] incredibly powerful but what happens [39:53] when that interface is not really [39:55] textual right will have to reinvent that [39:59] working on Tech it'll be the system will [40:01] somehow know what you're asking right it [40:04] will it just it will be your assistant [40:06] um and again Google will do very well so [40:08] I'm in no way criticizing Google here [40:10] but I'm saying that even something as [40:12] simple as the search box will eventually [40:14] be replaced by something more powerful [40:17] it's important that Google be the [40:18] company that does that I believe they [40:20] will and I I was thinking about it you [40:22] know the example of Steve Jobs and that [40:24] building with the pirate flag on it my [40:27] brain when [40:29] um there's so many offices around the [40:32] world that were trying to kill Apple at [40:35] that exact moment that might not have [40:36] had the pirate flag but that's exactly [40:38] what they were doing in similar small [40:40] rooms so what Apple had done so smartly [40:42] there was they owned the people that [40:45] were about to kill their business model [40:47] and this is quite difficult to do and [40:48] part of me wonders if in your experience [40:51] it's a Founder that has that type of [40:54] conviction that does that it's extremely [40:56] hard for non-founders to do this in [40:58] corporations because if you think about [41:00] a [41:01] corporation what's the duty of the CEO [41:05] many there's the shareholders there's [41:07] the employees there's the community and [41:09] there's a board trying to get a board of [41:12] very smart people to agree on anything [41:14] is hard enough so imagine I walk in to [41:17] you and I say I have a new idea I'm [41:20] going to kill our profitability for two [41:22] years it's a huge bet and I need1 [41:25] billion [41:28] now would the board say yes well they [41:31] did to Mark [41:33] Zuckerberg he spent all that money on um [41:36] essentially VR of one kind or another [41:39] doesn't seem to have produced very much [41:41] but at exactly the same time he invested [41:44] very heavily in Instagram WhatsApp and [41:47] Facebook and in particular in the AI [41:49] systems that power them and today [41:51] Facebook to my surprise is a very [41:54] significant leader in AI having released [41:56] this uh language called or version [41:58] called llama 400 billion which is [42:01] curiously an open source model open [42:03] source means it's available freely for [42:04] everyone and what what Facebook and meta [42:07] is saying is as long as we have this [42:10] technology we can maximize the revenue [42:12] in our core businesses so there's a good [42:14] example and uh and Zuckerberg is [42:17] obviously an incredibly talented [42:18] entrepreneur um he's now back on the [42:21] list of the most rich people um he's [42:23] feeded at you know and everything he was [42:25] doing and he managed to lose all that [42:28] money while making a different bet [42:30] that's a unique founder the same thing [42:32] is almost impossible with a hired [42:35] CEO how important here is focus and [42:38] what's your your sort of opinion of um [42:41] the importance of focus from your [42:42] experience with Google but also looking [42:44] at these other companies because when [42:45] you're at Google and you have so much [42:47] money in the bank there's so many things [42:48] that you could do and could build like [42:50] an endless list you can take on anybody [42:52] and basically win in most markets how do [42:55] you think about focus at Google [42:58] focus is important but it's [43:02] misinterpreted in Google we spent an [43:04] awful lot of time telling people we [43:07] wanted to do everything and everyone [43:09] said you can't pull off everything and [43:12] we said yes we can we have the [43:14] underlying architectures we have the [43:15] underlying reach we can do this if we [43:18] can imagine and build something that's [43:19] really transformative and so the idea [43:22] was not that we would somehow focus on [43:24] one thing like search but rather that we [43:26] would pick areas of great impact and [43:28] importance to the world many of which [43:30] were free by the way this is not [43:31] necessarily Revenue driven and that [43:33] worked I'll give you another example [43:35] there's an old saying in the business [43:38] school that you should focus on on what [43:41] you're good at and you should simplify [43:42] your product lines and you should get [43:44] rid of product lines that don't work [43:47] Intel famously had a the term is called [43:52] arm it's a risk uh chip and this [43:55] particular risk chip was not compatible [43:58] with the architecture that they were [43:59] using for most of their products and so [44:01] they sold it unfortunately this was a [44:04] terrible mistake because the [44:06] architecture that they sold off was [44:08] needed for mobile phones with low memory [44:11] with small batteries and and heat [44:13] problems and so forth and so on and so [44:16] that decision that faithful decision now [44:18] 15 years ago meant that they were never [44:21] a player in the mobile space and once [44:23] they made that decision they tried to [44:25] take their expensive and expensive and [44:28] complex chips and they kept trying to [44:29] make cheaper and smaller versions but [44:32] the core decision which was to simplify [44:34] simplify to the wrong outcome today if [44:37] you look at I'll give you an example the [44:39] Nvidia chips use an arm CPU and then [44:43] these two powerful uh gpus it's called [44:46] the b200 they don't use the Intel chip [44:48] they use the arm chip because it was for [44:50] their needs faster I would never have [44:52] predicted that 15 years ago so at the [44:55] end maybe it was just a mistake but [44:58] maybe they didn't understand in the way [45:01] they were organized as a corporation [45:03] that ultimately battery power would be [45:06] as important as computing power right [45:08] the amount of battery you use and that [45:10] was the discriminant so one way to think [45:12] about it is if you're going to have [45:13] these sort of simple rules you better [45:15] have a model of what happens in the next [45:18] five years so the way I teach this is [45:22] just write down what it'll look like in [45:24] five years just try what will look like [45:26] in five years your company or whatever [45:28] it is right so let's talk about AI what [45:31] will be true in five [45:33] years that it's going to be a lot [45:35] smarter than it is be a lot smarter but [45:37] how many companies will there be in AI [45:40] will there be five or 5,000 or 50,000 [45:44] 50,000 how many big companies will there [45:47] be will there be new companies what will [45:50] they do right so I just told you my view [45:53] is that eventually you and I will have [45:57] our own AI assistant which is a polymath [46:00] which is incredibly smart which helps us [46:02] guide through the information overload [46:04] that it is today who's going to build it [46:06] make a prediction what kind of hardw [46:08] will be on make a prediction how fast [46:11] will the networks be make a prediction [46:13] write all these things down and then [46:15] have a discussion about what to do that [46:18] what is interesting about our industry [46:21] is that when something like the PC comes [46:23] along or the internet I lived through [46:25] all of these things they are are such [46:27] broad phenomena that they really do [46:30] create a whole new Lake a whole new [46:32] ocean whatever metaphor you want now [46:34] people said well wasn't that crypto no [46:39] crypto is not such a platform crypto is [46:41] not transformative to daily life for [46:43] everyone people are not running around [46:45] all day using crypto tokens rather than [46:47] currency crypto is a specialized Market [46:50] by the way it's important and it's [46:51] interesting it's not a horizontal [46:53] transformative Market the arrival of [46:56] alien intelligence in the form of savant [46:58] that you use is such a transformative [47:00] thing because it touches everything it [47:02] touches you as a a producer as a star as [47:05] a narrative it touches me as an [47:07] executive um it will ultimately help [47:09] people make money in the stock market [47:11] people are working on that there's so [47:14] many ways in which the technology is [47:15] transformative to start you in your case [47:18] when you think about your company [47:19] whether it's little you know itty bitty [47:21] or a really big one it's fundamentally [47:24] how will you apply AI to accelerate what [47:27] you're doing right in your case for [47:29] example here you have I think the most [47:31] successful show in the UK by far right [47:35] so how will you use AI to make it more [47:37] successful well you can ask it to [47:39] distribute you more right to make uh [47:41] narratives to summarize uh to to come up [47:44] with new insights to suggest uh to have [47:46] fun to create contest there all sorts of [47:48] ways that you can ask AI um I'll give [47:51] you a simple example if I were a [47:54] politician thankfully I'm not um and I [47:57] knew my district I would say uh to the [47:59] computer write a program so I'm saying [48:02] to the computer you write a program [48:04] which goes through all the constituents [48:05] in my interest figures out roughly what [48:08] they care about and if and then send [48:11] them a video which is labeled you know [48:13] of me digitally so I'm not fake but it's [48:16] kind of like my intention where I [48:18] explain to them how important I as their [48:20] constituent have made the bridge work [48:22] right and you sit there and you go [48:24] that's crazy but it's possible [48:27] now politicians have not discovered this [48:29] yet but they will because ultimately [48:32] politicians are around a human [48:33] connection and the quickest way to have [48:35] that communication is to be on their [48:36] phone talking to them about something [48:38] that they care about when chat GPT first [48:41] launched and they sort of scaled rapidly [48:43] to 100 million users there was all these [48:45] articles saying that um the founders of [48:48] Google had rushed back in and it was a [48:50] crisis situation at Googled and there [48:51] was panic and there was two things that [48:53] I thought first is is that true and [48:55] second thing was [48:57] how did Google not come to Market first [49:00] with a chat GPT style product well well [49:02] remember that Google also that's the old [49:04] question of why did you not do Facebook [49:06] well the answer is we were doing [49:07] everything else right so my defensive [49:11] answer is that Google has eight or nine [49:14] or 10 billion user clusters of activity [49:17] which is pretty good right it's pretty [49:19] hard to do right I'm very proud of that [49:21] I'm very proud of what they're doing now [49:23] um my own view is that what happened was [49:26] Google was was working in the engine [49:29] room and a team out of open AI figured [49:32] out a technology called rhf and what [49:35] happened was when they did gpt3 and GP [49:39] the t is Transformer which was invented [49:40] at Google when they did it they had sort [49:43] of this interesting idea and then they [49:46] own then they sort of casually started [49:49] to use humans to make it better and rhf [49:52] refers to the fact that you use humans [49:54] at the end to do ab tests [49:57] where humans can actually say well this [49:59] one's better and then the system learns [50:01] recursively from Human training at the [50:04] end that was a real breakthrough right [50:07] and uh I joke with my open a eye friends [50:09] that you were sitting around on on [50:11] Thursday night and you turn this thing [50:13] on and you go holy crap look how good [50:16] this thing is it was a real Discovery [50:19] right that none of us expected certainly [50:21] I did not um and once they had it um the [50:25] opening eye people Sam and and and so [50:27] forth we'll talk about this they didn't [50:30] really understand how good it was they [50:31] just turned it on and all of a sudden [50:34] they had this huge success disaster [50:35] because they were working on GPT 4 at [50:37] the same time it was an afterthought [50:40] it's a great story because it just shows [50:42] you that even the brilliant Founders do [50:44] not necessarily understand how powerful [50:47] what they what they've done is now today [50:50] of course you have uh GPT 40 um [50:54] basically a very powerful model from [50:55] open eye you have Gemini 1.5 which is [50:58] clearly in clearly roughly equivalent if [51:01] not better in certain areas um the [51:03] Gemini is more multimodal for example [51:06] and then you have other players llama [51:07] the Llama architecture l l la ma uh does [51:12] not stand for llamas it's large language [51:14] models um out of Facebook and a number [51:17] of others uh there's a startup called [51:19] anthropic um which is very powerful [51:22] founded by one of the inventors of gpt3 [51:25] um and a whole bunch of people and they [51:26] formed their company knowing they were [51:28] going to be that successful it's [51:30] interesting they actually formed as part [51:32] of their incorporation that they were a [51:33] public benefit Corporation because they [51:35] were concerned that it would be so [51:37] powerful that some evil CEO in the [51:39] future would force them to go for [51:40] Revenue as opposed [51:42] to world world goodness so the teams [51:46] when they were doing this they [51:47] understood the power of what they were [51:49] doing and they anticipated the level of [51:51] impact which and they were right do you [51:53] think if Steve Jobs was an apple they [51:54] would be on that list [51:58] um how do you think the company would be [52:00] different well Tim has done a fantastic [52:03] job in Steve's Legacy and what's [52:05] interesting is normally the successor is [52:07] not as good as the founder but somehow [52:10] Tim having worked with Steve for so long [52:11] and having set the culture having Steve [52:13] having they've managed to continue the [52:16] focus on the user this incredible safety [52:19] focus in terms of apps and so forth and [52:21] so on and they've remained a relatively [52:23] closed culture I think all of those [52:25] would have maintained detained had St [52:28] you know tragically died uh he was a [52:30] good friend but the important point [52:33] is Steve Steve believed very strongly in [52:38] what are called close systems where you [52:39] own and control all your intellectual [52:41] property and he and I would battle over [52:43] open versus closed because I came from [52:45] the other side and I did this with [52:46] respect I don't think they would have [52:48] changed that and they've change that now [52:51] no I think still apple is still [52:53] basically a single company that's ver [52:56] Ally integrated the rest of the industry [52:58] is largely more open I think everyone [53:01] especially in the wake of the recent [53:02] launch of the iPhone 16 which I've got [53:05] somewhere here um has this expectation [53:08] that Apple would if Steve were still [53:10] alive taken some big bold bet in some [53:13] and I think about you know Tim's tenure [53:15] he's done a fantastic job of keeping [53:16] that company going running it with the [53:19] sort of principles of Steve Jobs but has [53:21] there been many big bold successful bets [53:23] a lot of people point at the airpods [53:25] which have a a great product [53:27] but I think AI is one of those things [53:29] where you go I wonder if Steve would [53:31] have understood the significance of it [53:33] and Steve was that smart that he I would [53:36] never you know he's an Elon level [53:38] intelligence [53:41] um when when Steve and I worked together [53:44] very closely which was what 15 years ago [53:47] for his death um he was very frustrated [53:51] at the success of MP4 over uh mov [53:57] um format files and he was really mad [54:01] about it and I said well you know maybe [54:03] that's because you were closed in quick [54:05] time was not generally available said [54:06] that's not true my team you know our [54:09] product is better and so forth so his [54:11] his core belief system he's an artist [54:14] right and and given the choice we used [54:17] to have this debate where do you want to [54:18] be Chevrolet or do you want to be [54:20] Porsche do you want to be you know [54:22] General Motors or do you want to be BMW [54:24] and he said I want to be BMW [54:27] and during that time Apple's margins [54:29] were twice as high as the PC companies [54:32] and I said Steve you don't need all that [54:34] money you're generating all this cash [54:36] you're giving it to your to your [54:38] shareholders and he said the principle [54:41] of our profitability and our value in [54:43] our brand is this is this luxury brand [54:47] right so that's how he thought now what [54:50] How would how would AI change that [54:52] everything that he would have done with [54:54] Apple today would be a I inspired but it [54:57] would be beautiful that's the great gift [55:00] he had CU I think Siri was almost a [55:03] glimpse at what AI now kind of looks [55:06] like it was a glimpse at what the I [55:08] guess the ambition was we've all been [55:09] chatting to the Siri thing which is I [55:11] think most people would agree as kind of [55:12] like largely useless unless you're [55:13] trying to figure out something super [55:14] super simple but now I this weekend as I [55:17] said I was sat there with my my [55:19] girlfriend's family there speaking to [55:21] this voice activated device and it was [55:23] solving problems for me almost [55:24] instantaneously that are very complex [55:26] and translating them into French and [55:27] Portuguese welcome welcome to the [55:29] replacement for Siri and again would [55:32] Steve have done that quicker I don't [55:33] know it's very clear that the first [55:36] thing Apple needs to do is have Siri be [55:40] replaced by an AI and call that Siri [55:43] hiring we we're doing a lot of hiring in [55:45] our companies at the moment and we're [55:47] going back and forward on what the most [55:48] important principles are when it comes [55:49] to hiring making lots of mistakes [55:51] sometimes getting things right [55:54] sometimes what do I need to know as when [55:56] it comes to hiring startups by [55:59] definition are huge Risk Takers you have [56:02] no history you have no incumbency you [56:04] have all these competitors by definition [56:05] and you have no time so in a startup you [56:08] want to you want to um prioritize [56:12] intelligence and quickness over [56:14] experience and sort of stability you [56:17] want to take risks on people and the [56:20] great and part of the reason why [56:22] startups are full of young people is [56:23] because young people often don't have [56:25] the baggage of Executives have been [56:27] around for a long time but more [56:28] importantly they're willing to take [56:30] risks so it used to be that you could [56:34] predict whether a company was successful [56:36] by the age of the founders and in that [56:38] 20 and 30y old period the company would [56:41] be hugely successful startups um Wiggle [56:44] they try something they try something [56:46] else and they're very quick to discard [56:49] an old idea corporations spend years [56:52] with a belief system that is factually [56:54] false and they don't actually changed [56:56] their opinion until after they've lost [56:58] all the contracts and if you go back the [57:02] all the signs were there nobody wanted [57:03] to talk to them nobody cared about the [57:05] product right and yet they kept pushing [57:07] it so um if you're a CEO of a larger [57:10] company what you want to do is basically [57:13] figure out how to measure this [57:14] Innovation so that you don't waste a lot [57:16] of time Bill Gates had a saying a long [57:18] time ago which was that the most [57:20] important thing to do is to fail fast [57:22] right that the charact from his [57:24] perspective as the CEO of Microsoft [57:25] founder Microsoft um that he wanted [57:29] everything to happen and he wanted to [57:30] fail quickly and that was his theory and [57:33] do you agree with that theory yeah I do [57:35] fast failure is important because you [57:38] can say it in a nicer way but [57:40] fundamentally um at Google we had this [57:42] 72010 rule that Larry and Sergey came up [57:44] with 70% of the Core Business 20% on [57:47] adjacent business and 10% on other what [57:49] does that mean sorry cor Core Business [57:51] means search ads adjacent business means [57:54] something that you're trying like a [57:55] cloud business or so forth and the 10% [57:58] is some new idea so Google created this [58:01] thing called Google X the first product [58:03] it built was called Google brain which [58:06] is the one of the first machine learning [58:07] architectures this actually precedes [58:09] Deep Mind Google brain was used to power [58:12] the AI system Google brin's team of 10 [58:14] or 15 people generated 10 20 30 40 [58:18] billion dollars of extra profits over a [58:20] decade so that pays for a lot of [58:22] failures right then they had a whole [58:24] bunch of other ideas that seemed very [58:26] interesting to me that didn't happen for [58:28] one or another and they would cancel [58:31] them and you you and then the people [58:33] would get reconfigured and one of the [58:35] great things about Silicon Valley is [58:37] it's possible to spend a few years on a [58:39] really bad idea and get cancelled if you [58:42] will and then get another job Having [58:44] learned all of that my joke is the best [58:46] CFO is one who's just gone bankrupt [58:49] because the one thing that CFO is not [58:51] going to let happen is to go bankrupt [58:53] again yeah well on this point of culture [58:55] as well Google as such a big company [58:59] must [59:00] experience a bunch of microcultures one [59:02] of the things that I've always I've kind [59:04] of studied it as an as a cautionary tale [59:06] is the story of TGIF at Google which was [59:10] this sort of weekly All Hands meeting [59:12] where employees could ask the executives [59:14] whatever they wanted to and the Articles [59:16] around it say that it was eventually [59:18] sort of changed or canceled because it [59:20] became [59:21] unproductive it's more complicated than [59:23] that so lar and serus started TGF [59:26] uh which I obviously participated in and [59:28] we had fun uh there was a sense of humor [59:31] it was all off the Record um a famous [59:34] example is the VP of sales whose name [59:36] was Omid um was always predicting lower [59:40] Revenue than we really had which is [59:42] called sandbagging so we got a sandbag [59:44] and we made him stand on the sandbag in [59:46] order to present his numbers it was just [59:49] fun humorous you know we had skits and [59:51] things like that um at at some size you [59:54] don't have that level of intim intimacy [59:56] and you don't have a level of privacy [59:58] and what happened was there were leaks [60:02] uh eventually there was a presentation I [60:05] don't remember the specifics where the [60:08] Pres presentation was ongoing and [60:10] someone was leaking the presentation [60:12] live to a reporter and somebody came on [60:15] stage and said we have to stop now I [60:18] think that was the moment where the [60:20] company got sort of too [60:23] big [60:24] h I heard about a story that um because [60:28] from what I had understood this might be [60:29] totally wrong but it's all just things [60:31] that Google employees have told me was [60:33] that there wasn't many sackings firings [60:35] at Google's wasn't many layoffs wasn't [60:38] really a culture of layoffs and I guess [60:39] I guessed in part that's because the [60:40] company was so successful that it didn't [60:42] have to make those extremely extremely [60:44] tough decisions that we're seeing a lot [60:46] of companies make today I reflect on [60:48] elon's running of Twitter when he take [60:50] took over Twitter the you know the say [60:53] the The Story Goes that he went to the [60:54] top floor and basically said anyone [60:56] who's willing to work hard is committed [60:59] to these values please come to the top [61:00] floor everyone else you're fired um this [61:02] sort of extreme culture of culling and [61:05] people being sort of activists at work [61:09] um and I wanted to know if there's any [61:11] truth in that there's some um in in [61:14] Google's case [61:17] um we had a position of why lay people [61:20] off just don't hire them in the first [61:22] place it's much much easier and so in my [61:25] 10 year the only layoff we did was uh [61:28] 200 people in the sales structures right [61:31] after the 2000 epidemic and I remember [61:33] it as being extremely painful right it [61:35] was the first time we had done it so we [61:38] took the position which is different at [61:40] the time that you shouldn't have an [61:42] automatic layoff what would happen is [61:44] that there was a belief at the time that [61:47] every six months or nine months you [61:48] should take the bottom five% of your [61:50] people and lay them off problem with [61:52] that is you're assuming the 5% are [61:54] correctly identified and furthermore [61:56] even the lowest performers have [61:58] knowledge and value to the corporation [62:00] that we can take it so we took a a very [62:02] much more positive view of our employees [62:04] and the employees like that and we [62:05] obviously paid them very well and so [62:06] forth and so on I think that the the [62:09] cultural issues ultimately have been [62:11] addressed but during there was a period [62:13] of time where there were uh because of [62:17] the free willing nature nature of the [62:18] company there were an awful lot of [62:20] internal distribution lists which had [62:22] nothing to do with the company what does [62:25] that mean they were distribution lists [62:27] on topics of War peace politics so forth [62:31] what's a distribution list a [62:32] distribution like an email dist think of [62:34] it as a a message board okay roughly [62:37] speaking think of it as message boards [62:39] for employees and at one I remember that [62:41] one point somebody discovered that there [62:43] were 100,000 such me message boards and [62:46] the company ultimately cleaned that up [62:48] because companies are not like [62:49] universities and that there are in fact [62:51] all sorts of laws about what you can say [62:53] and what you cannot say and so forth and [62:56] so for example the majority of the [62:57] employees were uh Democrats in the [63:00] American political system and I made a [63:02] point even though I'm a Democrat to try [63:03] to protect the small number of [63:05] Republicans because I thought they had a [63:07] right to be employees too so you have to [63:09] be very careful in a corporation to [63:11] establish what what does speech mean [63:13] within the corporation and uh what you [63:17] what you are hearing as wokeism is [63:20] really can be understood is what are the [63:22] appropriate topics on work time in in a [63:25] work venue should you be discussing my [63:27] own view is stick to the business and [63:30] then please feel free to go to the bar [63:32] scream your views talk to everybody you [63:35] know I'm a strong believer in free [63:36] speech but within the corporation let's [63:37] just stick to the corporation and its [63:39] goals because I was hearing these [63:40] stories about I think in more recent [63:42] times in the last year or two of people [63:44] coming to work just for the free [63:45] breakfast Pro protesting outside that [63:47] morning coming back into the building [63:49] for lunch as best I can tell that's all [63:51] been cleaned [63:52] up I did also hear that that it had been [63:55] cleaned up because I think it was [63:57] addressed in a very high conviction way [63:59] which meant that it it was um seen to [64:02] how did how do you think about [64:03] competition for everyone that's building [64:05] something how much should we be focusing [64:07] on our comp competition I strongly [64:09] recommend not focusing on competition [64:10] and instead focusing on building a [64:12] product that no one else has and you say [64:14] well how can you do that without knowing [64:15] the competition well if you study the [64:16] competition you're wasting your time try [64:18] to solve the problem in a new way and do [64:20] it in a way where the customers are [64:21] delighted U running Google we seldom [64:25] looked at what our competitors were [64:26] doing what we did we spent an awful lot [64:28] of time was what is possible for us to [64:31] do what can we actually do from our [64:33] current situation and sort of the [64:36] running ahead of everybody turns out to [64:38] be really important what about [64:40] deadlines well uh Larry established the [64:44] principle of um okrs which were [64:46] objectives and key results in every [64:48] quarter Larry would actually write down [64:51] all the metrics and he was tough and he [64:54] would say that if you got to 70% % of my [64:56] numbers that was good and then we would [64:58] grade based on are you above the 70% or [65:01] you below the 70% and it was harsh and [65:03] it works you you have to measure to get [65:07] things done in big Corporation otherwise [65:10] everyone kind of looks good makes all [65:12] sorts of claims feels good about [65:14] themselves but it doesn't have an impact [65:17] what about business plans should we be [65:18] writing business plans as found us [65:20] Google wrote A business plan there was a [65:22] run by a fellow named solar and I saw it [65:25] years later and it was actually correct [65:27] and I told salar that the this is [65:30] probably the only business plan ever [65:32] written for a corporation that was [65:33] actually correct in hindsight so what I [65:37] prefer to do and this is how I teach it [65:39] at Stanford is try to figure out what [65:42] the world looks like in five years and [65:44] then try to figure out what you're going [65:46] to do in one year and then do it right [65:50] so if you can basically say this is the [65:52] direction these are the things we're [65:54] going to achieve within one year and [65:56] then run against that as hard goals not [65:58] simple goals but hard goals then you'll [66:01] get there and the general rule at least [66:03] in a consumer business is if you can get [66:05] an audience of 10 or 100 million people [66:07] you can make lots of money right so if [66:09] you give me any business that has no [66:10] revenue and a 100 million people I can [66:12] find a way to to monetize that with [66:15] advertising and sponsorships and [66:16] donations and so forth and so on focus [66:19] on getting the user right and everything [66:22] else will follow the Google phrase is [66:23] focus on the user and everything else is [66:27] handled Sergey and [66:30] Larry you work with them for 20 years [66:33] many decades yeah two decades what made [66:36] them special frankly raw IQ they were [66:39] just smarter than everybody else really [66:40] yeah and [66:43] uh in sergey's case his father was a [66:46] very brilliant Russian mathematician his [66:48] mother was also highly technical his [66:50] family is all very technical and he was [66:52] clever he's a clever [66:53] mathematician uh Larry [66:56] different personality but similar so an [66:58] example would be that Larry and I are in [67:00] his office and we're writing on the [67:02] Whiteboard a long list about what we're [67:03] going to do and he says look we're going [67:05] to do this and this and I said okay I [67:06] agree with you I don't agree with you we [67:08] make this very long list and Sergey is [67:10] out playing [67:11] volleyball and so he runs in in his [67:14] little volleyball shorts and his little [67:16] shirt all sweating he looks at our list [67:17] and said this is the stupidest thing [67:19] I've ever heard and then he suggest five [67:21] things and he was exactly right so we ar [67:25] red the Whiteboard and then he of course [67:27] went back to play volleyball and that [67:28] became the strategy of the company so [67:31] over and over again it was the it was [67:32] their Brilliance and their ability to [67:34] see things that I didn't see that I [67:36] think really drove it can you teach that [67:39] I don't know I think you can teach [67:41] listening and [67:43] um but I think most of us get caught up [67:46] in our own [67:48] ideas and we are always surprised that [67:52] something new happened like I've just [67:54] told you that I'm I've been in AI a long [67:56] time I'm still surprised at the rate uh [67:58] my favorite current product is called [68:00] notebook [68:01] LM and for the uh listeners notebook LM [68:04] is an experimental product out of Google [68:06] Deep Mind basically Gemini um it's based [68:09] on the Gemini back end and it was [68:11] trained with high quality podcast voices [68:14] it's terrifying and you basically give [68:16] it a so what I'll do is um I'll write [68:19] something again I don't write very well [68:21] and I'll ask Gemini to rewrite it to be [68:23] more beautiful okay I'll take that text [68:26] and I'll put it in Notebook LM and it [68:28] produces this interview between a man [68:30] and a woman U who don't exist and for [68:33] fun what I do is I play this in front of [68:36] an audience and I wait and see if anyone [68:38] figures out that the humans are not [68:40] human it's so good they don't figure it [68:42] out we'll play it now so this is the big [68:44] thing that everyone's making a big fuss [68:46] about you can go and load this [68:47] conversation now it's going to go out [68:49] and create a conversation that's in a [68:50] podcast style where there's a male voice [68:53] and a female voice and they're analyzing [68:55] the content and then coming up with [68:56] their own kind of just uh creative [68:59] content so you could go and push play [69:00] right here we are back Thursday get [69:02] ready for week three the injury report [69:05] this week was a doozy it's a long one [69:08] yeah it is and it has the potential to [69:10] really shake things up so for that to me [69:14] gem notebook LM is my chat GPT moment of [69:18] this [69:19] year it was mine as well and it's much [69:22] of the reason that I was um deeply [69:24] confused okay because as a podcaster [69:27] who's building a media company we have [69:29] an office down the road 25,000 square [69:31] feet we have studios in there um we're [69:35] building audio video content at this in [69:40] the dawn of this new world where the [69:42] cost of production of content goes to [69:44] like zero or something and I'm trying to [69:46] navigate how to play as a media owner so [69:49] first place you're you're what's really [69:50] going on is you're moving from scarcity [69:52] to ubiquity you're moving from scarc to [69:56] abundance so one way to understand the [69:58] world I live in is it's scale Computing [70:00] generates abundance and abundance allows [70:02] new strategies in your case it's obvious [70:04] what you should do you're a really [70:06] famous podcaster and you have lots of [70:08] interesting guests simply have this fake [70:11] set of podcasts criticize you and your [70:14] guests right you're you're essentially [70:16] just amplifying your reach they're not [70:19] going to substitute for your honest [70:21] Brilliance and Charisma here but they're [70:23] going to accentuate it they will they [70:25] will they will be entertaining they will [70:27] summarize it and so forth it amplifies [70:29] your reach if you go back to my basic [70:32] argument that AI will double the [70:34] productivity of everybody or more so in [70:37] your case you'll have twice as many co [70:40] podcasts what I do for examples I'll [70:41] write something and I'll say I'll have [70:43] it respond and then to Gemini I'll say [70:46] make it longer and it adds more stuff I [70:49] think God I do this in like 30 seconds [70:52] then how powerful in your case take one [70:55] of these uh lengthy interviews you do [70:57] ask the system to annotate it to amplify [71:00] it and then feed that into fake [71:03] podcasters and see what they say you'll [71:05] have a whole new set of audiences that [71:07] love them more than you but but it's all [71:10] from you that's the key idea here I [71:13] worry because there's going to be [71:15] potentially billions of podcasts that [71:17] are uploaded to RSS feeds all around the [71:19] world and it's all going to sort of chip [71:21] away at you know the the moat that I've [71:24] so [71:25] so many people have believed that but I [71:28] think the evidence is it's not true um [71:31] when I started at Google there was this [71:33] notion that celebrity would go away and [71:35] there would be this very long tale of [71:38] micro markets you know Specialists [71:41] because finally you could hear the [71:42] voices of everyone and we're all very [71:44] Democratic and liberal in our view [71:46] that's the what really happened was [71:48] networks accentuated the best people and [71:51] they made more money right you went from [71:53] being a local personality to a national [71:56] personality to a global personality and [71:57] the globe is a really big thing and [71:59] there's lots of money and lots of [72:01] players so you as a as a celebrity are [72:05] competing against a global group of [72:07] people and you need all the help you can [72:08] to maintain your position if you do it [72:10] well by using these AI Technologies you [72:13] will become more famous not less [72:17] famous [72:20] Genesis I am I've had a lot of [72:22] conversations with a lot of people about [72:24] the subject of AI um and when I read [72:26] your book and I've watched you do a [72:28] series of interviews on this some of the [72:29] quotes that you said really stood out to [72:31] me one of them I wrote down [72:35] here which comes from your book Genesis [72:37] it's on page five the Advent of [72:39] artificial intelligence is in our view a [72:42] question of human [72:45] survival yes that is our view so why is [72:50] it a question of human [72:53] survival AI is going to move very [72:56] quickly it's moving so much more quickly [72:58] than I've ever seen because the amount [73:01] of money the number of people the impact [73:03] the [73:05] need what happens when the AI systems [73:07] are really running key parts of our [73:10] world what happens when AI is making the [73:13] decision my my simple example you have a [73:16] car which is AI controlled and you have [73:20] a emergency or a lady's about to give [73:23] birth or something like that and they [73:25] get in the car and there's no override [73:27] switch because the system is optimized [73:30] around the whole as opposed to his or [73:32] her [73:33] emergency right we as humans accept [73:36] various forms of efficiency including [73:38] urgent ones versus system systemic [73:41] efficiency you could imagine that the [73:43] Google Engineers would design a perfect [73:45] City that would perfectly operate every [73:48] self-driving car on every street but [73:50] would not then allow for the exceptions [73:52] that you need in such a in such an [73:55] important issue so that's a trivial [73:58] example and one which is well understood [74:01] of how it's important that these things [74:02] represent human values right that we we [74:06] have to actually articulate what does it [74:08] mean so my favorite one is all this [74:11] misinformation um democracy is pretty [74:13] important democracy is by far the best [74:15] way to to live and operate societies [74:17] look at there are plenty of examples of [74:19] this none of us want to work in [74:21] essentially an authoritarian [74:23] dictatorship so you better figure out a [74:25] way where the misinformation components [74:28] do not screw up proper political [74:31] examples another example is this [74:34] question about teenagers and the develop [74:36] their mental development and growing up [74:37] into these societies I don't want them [74:40] to be constantly depressed there's a lot [74:42] of evidence that dates around 2015 when [74:46] all the social media algorithms changed [74:48] from linear feeds to targeted feeds in [74:50] other words they went from time to this [74:52] is what you want this is what you want [74:54] that hyperfocus has ultimately narrowed [74:57] people's um political views as I as we [75:00] discussed but more importantly it's [75:02] produced more depression and anxiety so [75:05] all the studies indicate that basically [75:07] if you time it to roughly then when [75:09] people are coming to age they're not as [75:11] happy with their lives their behaviors [75:13] their opportunities for this and the [75:16] best explanation is it was an [75:18] algorithmic change and remember that [75:20] these systems they're not just [75:21] collections of content they are [75:23] algorithmically deciding [75:25] you know the algorithm decides what the [75:28] outcome is for humans we have to manage [75:30] that um what we say in many different [75:33] ways in the book is that you have sort [75:36] of a choice of whether the um the [75:40] algorithms will advance that's not a [75:41] question the question is are we [75:43] advancing with it and do we have control [75:45] over it um there are so many examples [75:48] where you could imagine an AI system [75:50] could do something more efficiently but [75:53] at what cost right [75:55] um I should mention that there is this [75:58] discussion about something called AGI [76:00] artificial general [76:02] intelligence and there's this discussion [76:04] in the Press among many people that AGI [76:06] occurs on a particular day right and [76:09] this is sort of a popular concept that [76:11] on a particular day five years from now [76:13] or 10 years from now this thing will [76:14] occur and all of a sudden we're going to [76:16] have a computer that's just like us but [76:18] even quicker that's unlikely to be the [76:21] path much more likely are these waves of [76:24] innovation in every field better [76:26] psychologists better writers you see [76:29] this with g chat gbt already better [76:31] scientists is a notion of an AI [76:33] scientist that's working with the AI [76:35] real scientists to accelerate the [76:37] development of more AI science people [76:40] believe all of this will come but it has [76:42] to be under human [76:43] control do you think it will be I do and [76:47] part of the reason is I and others have [76:49] worked hard to get the governments to [76:51] understand this it's very strange in my [76:53] entire career which has gone for you [76:55] know 50 years the um we've never asked [76:59] for government for help because asking [77:01] the government help is basically just a [77:02] disaster in the view of the techn [77:04] industry in this case the people who [77:07] invented it collectively came to the [77:09] same view that there need to be [77:11] guardrails on this technology because of [77:13] the potential for harm the most obvious [77:14] one is how do I kill myself give me [77:16] recipes to hurt other people that kind [77:18] of stuff there's a whole Community now [77:21] in this in this part of the industry [77:23] which are called trust and safety groups [77:26] and what they do is they actually have [77:27] humans test the system before it gets [77:31] released to make sure the harm that it [77:34] might have in it is suppressed it's [77:36] literally won't answer the question when [77:39] you play this forward in your brain you [77:40] you've been in the tech industry for a [77:42] long time and from looking at your work [77:44] you it feels like you're describing this [77:46] as the most sort of transformative [77:48] potentially harmful technology that [77:49] humans have really ever seen you know [77:51] maybe alongside the nuclear bomb I guess [77:53] but some would say even potentially [77:56] worse because of the nature of the [77:57] intelligence and its [77:59] autonomy you must have moments where you [78:02] you think forward into the future and [78:04] your thoughts about that future aren't [78:05] so [78:06] Rosy well because I have those moments [78:08] yes but but let's let's think let's [78:10] answer the question I said think five [78:12] years in five years you'll have two or [78:14] three more turns of the crank of these [78:15] large models these large models are [78:18] scaling with ability that is [78:21] unprecedented there's no evidence that [78:23] the scaling has laws as they're called [78:26] have begun to to stop they will [78:28] eventually stop but we're not there yet [78:31] each one of these cranks looks like it's [78:33] a factor of two factor of three factor [78:35] of four of capability so let's just say [78:37] turning the crank all of these systems [78:40] get 50 times or 100 times more powerful [78:43] in it of itself that's a very big deal [78:46] because those systems will be capable of [78:48] physics and math you see this with o. [78:50] one and um and open AI all the other [78:53] things that are occurring [78:55] now what are the dangers well there's [78:58] the most obvious one is cyber attacks [79:00] there's evidence that the raw models [79:02] these are the ones that have not been [79:03] released can do what are called Day Zero [79:06] attacks as well or better than humans a [79:08] day Zero attack is an attack that's [79:09] unknown they can discover something new [79:12] and how do they do it they just keep [79:13] trying because they're computers and [79:15] they have nothing else to do they don't [79:16] sleep they don't eat they just turn them [79:18] on and they just keep going um so the so [79:21] cyber is an example where everybody's [79:23] concerned another one is biology viruses [79:25] are relatively easy to make and you can [79:27] imagine coming up with really bad [79:29] viruses there's a whole team I'm part of [79:31] a commission looking at this to try to [79:32] make sure that doesn't happen I already [79:35] mentioned misinformation [79:37] another probably negative but we'll see [79:41] is the development of new forms of [79:43] warfare I've written extensively on how [79:45] war is changing and the way to [79:48] understand historic war is that it's the [79:51] stereotypically the the soldier with the [79:53] gun you know one side and so forth World [79:56] War trenches you see this by the way in [79:58] UK in the Ukraine fight today where the [80:00] ukrainians are holding on valiantly [80:02] against the Russian Onslaught but he's [80:03] sort of you know mono Amano you know man [80:06] against man sort of all of the [80:08] stereotypes of War so in a drone World [80:11] which is the sort of the fastest way to [80:13] build new robots is to build drones [80:15] you'll be sitting in a Command Center in [80:17] some office building connected by a [80:19] network and you'll be doing harm to the [80:21] other side while you're drinking your [80:23] coffee right that's a changed in the [80:25] logic of War um and it's applicable to [80:27] both sides I don't think anyone quite [80:29] understands how war will change but I [80:31] will tell you that in in the Russian [80:34] Ukraine war you're seeing a new form of [80:36] Warfare being invented right now right [80:40] um both sides have lots of drones tanks [80:42] are no longer very useful a $5,000 drone [80:45] can kill a $5 million tank um so it's [80:49] called The Kill ratio so basically it's [80:51] drone on drone and so now people are [80:52] trying to figure out how how to have one [80:54] drone destroy the other drone right this [80:57] will ultimately take over war and [80:59] conflict in our world in total you [81:02] mentioned rural models this is a concept [81:04] that I don't think people understand [81:05] exists the idea that there's some other [81:07] model that's the role model that is [81:10] capable of much worse than the thing we [81:12] play with on our computers every day [81:14] it's important to establish how these [81:15] things work so you the way these [81:16] algorithms work is they have complicated [81:19] uh training things where they suck all [81:20] the information in and they uh one week [81:24] currently believe we've sort of sucked [81:26] all of the written word that's available [81:28] it doesn't mean there isn't more but [81:29] we've we've literally done such a good [81:31] job of sucking everything that humans [81:32] have ever written it's all in these big [81:35] computers when I say computers I don't [81:37] mean computers I mean supercomputers [81:38] with enormous memories and the scale is [81:42] mindboggling uh and of course there's [81:43] this company called Nvidia which makes [81:45] the chips which is now one of the most [81:46] valuable companies in the world um [81:50] surprisingly so incredibly successful [81:52] because they're so Central to this [81:53] revolution and good for Jensen and his [81:55] team so the important thing is when you [81:58] do this training it comes out with a raw [82:00] model right it takes six months and you [82:03] know you wait 24 hours a day you can [82:05] watch it it gets close to there's a [82:07] measurement that they use called the [82:08] loss function when it gets to a certain [82:10] number they say good enough so then they [82:13] go what do we have right what do we do [82:16] right um so the first thing is let's [82:18] figure out what it [82:20] knows so they have a set of tests and of [82:23] course it knows all sorts of bad things [82:25] which they immediately then tell it not [82:26] to answer to me the most interesting [82:29] question is in over a 5-year [82:33] period the systems will learn things [82:35] that we don't know they learn how will [82:38] you test for things that you don't know [82:40] they [82:41] know the answer in the industry is that [82:45] they have incredibly clever people who [82:47] sit there and they fiddle literally [82:49] fiddle with the networks and say I'm [82:51] gonna I'm going to see if it knows this [82:55] I'll see if it can do this and then they [82:57] make a list and they say that's good [82:59] that's not so good right so all of these [83:02] Transformations so for example you can [83:05] show it a picture of a website and it [83:06] can generate the code to generate a [83:08] website all of those were not expected [83:10] they just happened it's called emergent [83:13] Behavior scary scary but exciting and so [83:17] far um the systems have held the [83:20] governments have worked well um the [83:23] these trust and safety groups group are [83:24] working here in the UK um one year ago [83:28] was the first trust and safety [83:29] conference um the government did a [83:31] fantastic job the team that was [83:33] assembled was the best of all the [83:35] country teams here in the UK um now [83:38] what's happening is these are happening [83:39] around the world the next one is in [83:41] France in uh early February and I expect [83:44] a similar good result do you think we're [83:46] gonna have to guard I mean you talk [83:47] about this but do you think we're going [83:49] to have to guard these role models with [83:52] with guns and tanks and machinery and [83:54] stuff I worked for the Secretary of [83:56] Defense for a while uh in my in Google [83:59] you could spend 20% of your time on [84:01] other things so I worked for the [84:02] Secretary of Defense to try to [84:03] understand the US Military and um one of [84:07] the things that we did is we visited a [84:08] plutonium U Factory plutonium is [84:11] incredibly dangerous and Incredibly [84:13] secret and so this particular base is [84:16] inside of another base so you go through [84:18] the first set of machine guns and then [84:19] you have normal thing and then you go [84:21] into the special place with even more [84:22] machines guns and even because it's so [84:24] secure so the the metaphor is do you [84:28] fundamentally believe that the computers [84:30] that I'm talking about will be of such [84:32] value and such danger that they'll have [84:34] their own data center with their own [84:36] guards which of course might be computer [84:38] guards but the important thing is that [84:40] it's so special that it has to be [84:42] protected in the same way that we [84:43] protect nuclear bombs and proliferate uh [84:45] and programming an alternative model is [84:48] to say that this technology will spread [84:52] pretty broadly and there'll be many such [84:54] plac [84:55] if it's a small number of groups the [84:58] governments will figure out a way to do [85:00] deterrence and they'll figure out a way [85:02] to do [85:03] non-proliferation so I'll make something [85:05] up I'll say there's a couple in China [85:07] there's a few in the US there's one in [85:08] in Britain of course we're all tied [85:10] together between the US and Britain and [85:11] maybe in a few other places that's a [85:13] manageable problem on the other hand [85:15] let's imagine that that power is [85:17] ultimately so easy to copy that it [85:20] spreads globally and it's accessible to [85:23] for example terrorist [85:25] then you have a very serious [85:26] proliferation problem which is not yet [85:29] solved this is again [85:31] speculation because I think a lot about [85:33] adversaries in China and Russia and [85:35] Putin and I think I know you talk about [85:38] them being a few years behind maybe one [85:40] or two years behind but they're [85:42] eventually going to get there they're [85:43] eventually going to get to the point [85:45] where they have these large language [85:46] models or these AIS that can do these [85:48] Day Zero attacks on our [85:50] nation [85:52] and they they don't have the like sort [85:55] of social incentive structure if they're [85:56] a communist country to protect and to um [86:01] guard against these things are you not [86:02] worried about what China is gonna do um [86:04] I am worried and I'm worried [86:07] because you're going into a space of [86:09] great power without fully defined [86:12] boundaries what kinger and we talk about [86:13] this in the book The the Genesis Book is [86:16] fundamentally about what happens to [86:18] society with the arrival of this new [86:20] intelligence and the first book we did [86:22] age of AI was right before chat GPT so [86:25] now everybody kind of understands how [86:27] powerful these things are we talked [86:28] about it now you understand it so once [86:30] these things show up who's going to run [86:32] them who's going to be in charge how [86:34] will they be used so from my perspective [86:38] I believe at the moment anyway that [86:40] China will behave relatively responsibly [86:43] and the reason is that it's not in their [86:46] interest to have free [86:47] speech in every case in China when they [86:51] have a choice of giving freedom to their [86:54] Cit citizens or not they choose [86:55] non-freedom and I know this because I [86:57] spent through all the uh I spent all the [86:59] time dealing with it so it sure looks to [87:03] me like the Chinese AI solution will be [87:05] different from the West because of that [87:08] fundamental bias against freedom of [87:11] speech because these things are noisy [87:13] they make a lot of noise they'll [87:15] probably still make AI weapons though [87:17] well on the weapon side you have to [87:19] assume that every new technology is [87:23] ultimately strengthened in a war um the [87:26] tank was invented in World War I at the [87:28] same time you had the initial forms of [87:30] uh airplanes much of the second world [87:33] war was an air Campaign which [87:35] essentially built many many things and [87:37] if you look at the the there's a a book [87:40] called Freedom's Forge about the [87:42] American U structure according to the [87:46] book they ultimately got to the point [87:47] where they could build two or three [87:49] airplanes a day at scale so in an [87:53] emergency Nations have enormous [87:55] power I get asked all the time if [87:59] everyone if anyone's going to have a job [88:00] left to do because this is the [88:01] disruption of intelligence and whether [88:04] it's people driving cars today I mean we [88:05] saw the Tesla announcement of the robo [88:07] taxis whether it's accountants lawyers [88:10] and everyone in between that's or [88:12] podcasters are we going to have jobs [88:14] left well um this question has been [88:17] asked for 200 years um there was there [88:21] were the L eyeses here in Britain way [88:23] back when and inevitably when these [88:25] Technologies come along there's all [88:27] these fears about them indeed with a lot [88:29] I there were riots and people you know [88:30] destroying the Looms and all of this [88:32] kind of stuff but somehow we got through [88:34] it so um my own view is that there will [88:38] be a lot of job [88:40] dislocation but there will be a lot more [88:43] jobs not fewer jobs and here's why we [88:46] have a demographic problem in the world [88:48] especially in the developed developed [88:50] world where we're not having enough [88:51] children uh that's well understood uh [88:54] furthermore we have a lot of older [88:55] people and and the younger people have [88:57] to take care of the older people and [88:59] they have to be more productive if you [89:00] have young people who need to be more [89:02] productive the best way to make them [89:03] more more productive is to give them [89:05] more tools to make them more productive [89:08] whether it's a machinist that goes from [89:10] a manual machine into a CNC machine or [89:13] in in the more modern case of a [89:14] knowledge worker who can achieve more [89:17] objectives we need that productivity [89:19] group if you look at Asia which is the [89:21] centerpiece of [89:22] manufacturing they have all this cheap [89:24] labor well it's not so cheap anymore so [89:26] do you know what they did they added [89:27] robotic assembly Lin so today when you [89:30] go to China in particular it's also true [89:32] in Japan and Korea the manufacturing is [89:34] largely done by robots why because their [89:37] demographics are terrible and their cost [89:38] of Labor is too high so the future is [89:41] not fewer jobs it's actually a lot of [89:44] jobs that are unfilled with people who [89:46] may have a job skill mismatch which is [89:48] why education is so important now what [89:51] are examples of jobs that go away [89:53] automation [89:54] has always gotten rid of jobs that are [89:58] dangerous physically dangerous or ones [90:01] which are essentially too repetitive and [90:03] too boring for humans I'll give you an [90:05] example um security guards it makes [90:08] sense that security guards would become [90:10] robotic because it's hard to be a [90:13] security guard you fall asleep you don't [90:15] know quite what to and these systems can [90:17] be smart enough to be very very good [90:19] security now these are these are [90:21] important sources of income for these [90:23] people they're going to have to find [90:24] another job another example in in the [90:27] media in um Hollywood everyone's [90:29] concerned that AI is going to take over [90:31] their jobs all the evidence is the [90:33] inverse and here's why um the Stars [90:35] still get money The Producers still make [90:38] money they still distribute their movie [90:40] but their cost of making the movie is [90:42] lower because they use more they use for [90:44] example synthetic backdrops so they [90:45] don't have to build the set um they can [90:47] do synthetic makeup now there are job [90:49] losses there so the people who make the [90:51] make make the set and do the makeup are [90:54] going to have to go back into [90:55] construction and personal care by the [90:57] way in America and I think it's true [90:59] here there's an enormous shortage of [91:01] people who can do high quality [91:02] craftsmanship right those people will [91:04] have jobs they're just different and [91:06] they may not be in Los Angeles am I [91:09] gonna have to interface with this [91:10] technology am I going to have to get a [91:12] neuralink in my brain because we you go [91:15] over the subject of there being these [91:16] sort of two species of humans [91:18] potentially ones that do have a way to [91:21] incorporate themselves more with [91:23] artificial intelligence and those that [91:25] don't and if and if that is the case [91:27] what is the time Horizon in your view of [91:28] that [91:29] happening I think neuralink is much more [91:32] speculative because you're dealing with [91:33] direct brain connection and nobody's [91:35] going to drill on my brain until it [91:36] needs it trust me I suspect you feel the [91:39] same uh I I guess my O My overall view [91:42] is that [91:47] um you will not [91:50] notice how much of your world has been [91:53] co-opted by these Technologies because [91:55] they will produce greater [91:57] Delight if you think about it a lot of [92:01] life is inconvenient it's fix this call [92:04] this make this happen AI systems should [92:06] make all that seamless you should be [92:07] able to wake up in the morning and have [92:10] coffee and not have a care in the world [92:11] and have the computer help you have a [92:13] great day this true of everyone now what [92:17] happens to your to your profession well [92:20] as we said no matter how good the [92:22] computers are people are going to want [92:24] to care about other people another [92:26] example let's imagine you have Formula 1 [92:28] and you have Formula One with humans in [92:30] it and then you have a a a robot Formula [92:32] 1 which where the cars are driven by the [92:35] equivalent of a robot is anyone going to [92:37] go to the robotic Formula 1 I don't [92:39] think so because of the drama the human [92:42] achievement and so forth do you think [92:44] that when they run the marathon here in [92:46] London they're going to have robots [92:48] running with humans of course not right [92:50] of course the robots can run faster than [92:52] humans it's not interesting what is [92:54] interesting is to see human achievement [92:56] so I think the commentators who say oh [92:58] there won't be jobs we won't care I [93:00] think they miss the point that we care a [93:02] great deal about each other as human [93:04] beings we have opinions you have a [93:07] detailed opinion about me having just [93:08] met me met me right now and I for you we [93:11] just are naturally set up your face your [93:13] mannerisms and so forth we can describe [93:15] it all right the robot shows up is like [93:18] oh my God what another robot how boring [93:20] why is samman working on the the founder [93:22] of open AI when the co-founders of open [93:24] a working on universal basic income [93:26] projects like worldcoin then well [93:29] worldcoin is not the same thing as [93:30] universal Bitcoin uh um Universal basic [93:34] income there is a belief in the tech [93:37] industry that it goes something like [93:39] this the politics of abundance what we [93:42] do is going to create so much abundance [93:45] that most people won't have to work and [93:48] there'll be a small number of groups [93:50] that work who typically these people [93:51] themselves and there be so much Surplus [93:54] everyone can live like a millionaire and [93:55] everyone will be happy I completely [93:57] think this is false I think none of what [93:59] I just told you is false but all of [94:01] these Ubbi ideas come from this notion [94:04] that humans don't behave the way we [94:06] actually do so I'm I'm a Critic of this [94:08] view I believe that that we as humans so [94:12] I an example is um we're going to make [94:16] legal the legal profession much much [94:17] easier because we can automate much of [94:19] the technical work of lawyers does that [94:21] mean we're going to have fewer lawyers [94:23] no the current lawyers will just do more [94:25] laws they'll do more they'll add more [94:27] complexity the system doesn't get easier [94:30] the humans become more sophisticated in [94:32] their application of the principles we [94:34] are naturally basically uh we have this [94:37] thing called um basically reciprocal [94:39] altruism that's part of us but we also [94:41] have our bad sides as well those are not [94:44] going away because of AI when I think [94:46] about AI this simple analogy often think [94:48] of is say my IQ is Steven bartett is 100 [94:50] and there's this AI that sat next to me [94:52] whose IQ is 1,000 what on Earth would [94:55] you want to give Steven to do because [94:57] because that 1,000 IQ would have really [94:59] bad judgment in a couple cases because [95:02] remember that the AI systems do not have [95:04] human values unless it's added right I [95:07] would much rather talk to you about [95:10] something involving a moral or human [95:12] judgment even with the Thousand I [95:14] wouldn't mind Consulting it so tell me [95:16] the the history how was this resolved in [95:17] the past how are these but at the end of [95:19] the day in my view the core aspects of [95:23] it which have to do with morals and [95:25] judgment and beliefs and Charisma [95:27] they're not going away is there a chance [95:29] that this is the end of humanity no um [95:32] the way Humanity [95:34] does is much it's much harder to [95:36] eliminate all of humanity than you think [95:39] all the people I've looked with on these [95:40] biological attacks say it's it takes [95:43] more than one horrific pandemic and so [95:46] forth to eliminate humanity and and the [95:48] the pain can be very very high in these [95:50] moments look at the World War I World [95:53] War II the Hodor in uh Ukraine in the [95:56] 1930s the Nazis you know these are [95:59] horrifically painful things but we [96:01] survived right we we as a as a Humanity [96:04] survived and we will I wonder if this is [96:06] the moment where humans couldn't see [96:08] past around the corner because you know [96:11] I've heard you talk about how the AIS [96:13] will turn in they'll be agents and [96:14] they'll be able to speak to each other [96:16] and we won't be able to understand the [96:17] language I have a specific proposal on [96:19] that um there are points where humans [96:22] should assert control [96:24] and I've been trying to think about [96:25] where are they I'll give you an example [96:27] there's something called recursive [96:28] self-improvement where the system just [96:30] keeps getting smarter and smarter and [96:32] learning more and more things at some [96:35] point if you don't know what it's [96:36] learning you should unplug it but we [96:40] can't unplug them can we sure you can [96:41] there's a power plug and there's a [96:43] circuit breaker go and turn the circuit [96:45] breaker off another example um there's a [96:48] there's a scenario theoretical where the [96:51] system is so powerful it can produce a [96:53] new model faster than the previous model [96:56] was checked okay that's another [96:59] intervention point so in each of these [97:02] cases um if the if agents and the [97:05] technical term is called agents what [97:07] they really are is large language models [97:09] with memory and you can begin to [97:11] concatenate them you can say this model [97:13] does this and then it feeds into this [97:15] and so forth you can build very powerful [97:16] decision systems we believe this is the [97:19] the the thing that's occurring this year [97:21] and next year everyone's doing them they [97:23] will arrive [97:24] the agents today speak in English you [97:26] can see what they're saying to each [97:28] other they're not human but they are [97:31] communicating what they're doing English [97:34] to English to English as long as and it [97:37] doesn't have to be English but as long [97:38] as they're human understandable but [97:40] let's so the thought experiment is one [97:41] of the agents says I have a better idea [97:44] I'm going to communicate in my own [97:45] language that I'm going to invent that [97:46] only other agents understand that's a [97:49] good time to pull the plug what is your [97:52] biggest fear about AI [97:54] my actual fear is different from what [97:55] you might imagine my my actual fear is [97:57] we're not going to adopt it fast enough [97:59] to solve the problems that affect [98:00] everybody right and the reason is that [98:03] the that if you look at every everyone's [98:06] everyday lives what do they want they [98:08] want safety they want Health Care they [98:11] want great schools for their kids we [98:13] just work on that for a while why do we [98:15] make people's lives just better because [98:17] of AI we have all these other [98:19] interesting things why don't we have a [98:21] um a teacher that is an AI teacher that [98:25] works with existing teachers in this [98:27] language of the kid in the culture of [98:30] the kid to get the kid as smart as they [98:31] possibly can why don't we have a doctor [98:33] or doctor's assistant really that [98:36] enables a a human doctor to always know [98:39] every possible best treatment and then [98:41] based on their current situation what [98:42] the inventory is which country is how [98:45] their insurance Works what is the best [98:46] way to treat that patient those are [98:48] relatively achievable Solutions why [98:50] don't we have them if you just did [98:52] education and Healthcare [98:54] globally the impact in terms of lifting [98:56] human potential up would be so great [99:00] right that it would change [99:01] everything it wouldn't solve the various [99:04] other things that we complain about [99:05] about you know this celebrity or this [99:07] misbehavior or this conflict or even [99:09] this war but it would establish a Level [99:11] Playing Field of knowledge and [99:13] opportunity at a global level that has [99:15] been the dream for decades and decades [99:17] and decades Chuck me that perfect head [99:21] one of the things that I think about the [99:23] time because my life is quite hectic and [99:25] busy is how to manage my energy load and [99:27] as a podcaster you kind of have to [99:29] manage your energy in such a way that [99:30] you can have these articulate [99:32] conversations with experts on subjects [99:34] you don't understand and this is why [99:35] perfect Ted has become so important in [99:37] my life because previously when it came [99:39] to Energy Products I had to make a [99:41] trade-off that I wasn't happy with [99:42] typically if I wanted the energy I had [99:44] to deal with high sugar I had to deal [99:46] with Jitters and crashes that come along [99:48] with a lot of the mainstream Energy [99:49] Products and I also just had to tolerate [99:51] the fact that if I want energy I have to [99:53] put up with a lot of artificial [99:55] ingredients which my body didn't like [99:57] and that's why I invested in perfect Ted [99:59] and why they're one of the sponsors of [100:00] this podcast it is changed not just my [100:02] life but my entire team's life and for [100:03] me it's drastically improved my [100:05] cognitive performance but also my [100:06] physical performance so if you haven't [100:08] tried perfect Ted yet you must have been [100:10] living under a rock now is the time you [100:13] can find perfect Ted at Tesco and [100:14] waitrose or online where you can enjoy [100:16] 40% off with code diary 40 at checkout [100:19] head to perfect ted.com this is quite [100:23] interesting 85% of Internet users have [100:25] heard of vpns but only 55% know what [100:29] they do if you're in that group let me [100:31] explain vpn's enable your location [100:33] online to differ from where you actually [100:35] are geographically to help you browse [100:36] and stream sites that would otherwise be [100:38] unavailable to you I use nordvpn who are [100:41] a sponsor of this show to watch [100:42] Manchester United games online no matter [100:44] where I am in the world and Indie from [100:46] my team uses them whenever she's booking [100:47] flights back home to New Zealand having [100:49] a different online location means she [100:51] can take advantage of dynamic pricing [100:54] and get cheaper prices for her flights [100:56] nordvpn is the fastest VPN in the world [100:58] and just one account can be used across [101:00] 10 devices and they've shared a generous [101:02] offer for my listeners a discount and [101:04] four additional months free on a 2-year [101:07] plan it's also completely risk-free with [101:09] nord's 30-day money back guarantee so [101:11] head to nordvpn.com [101:13] doac or click the link in the [101:15] description [101:17] below throughout the pandemic I've been [101:19] a big supporter um it was a contrarian [101:22] view but I think it's now less a [101:23] contrarian view that companies and CEOs [101:27] need to be clear in their convictions [101:28] around how they work and one of the [101:29] things that I've um been criticized a [101:31] lot for is that I'm I'm for having [101:34] people in a room together so my [101:35] companies we um we're not remote we work [101:38] together in an office as I said down the [101:39] road from here and I believe in that [101:42] because I think of community and [101:43] engagement and synchronous work and I [101:45] think that work now has a responsibility [101:46] to be more than just a set of tasks you [101:48] do in a world where we're lonier than [101:51] ever before there's more disconnection [101:52] and especially for young people you [101:53] don't have families and so on um having [101:56] them work alone in a small white box in [101:57] a big city like London or New York um is [102:00] robbing them of something which I think [102:01] is important this was a bad this was a [102:03] contrarian view it's become less [102:05] contrarian as the big tech companies in [102:07] America have started to roll back some [102:08] of their initial knej reactions to the [102:10] pandemic that there a lot of them are [102:12] asking their team members to come back [102:13] into the office at least a couple of [102:15] days a week what's your point of view on [102:16] this so I have a strong view that I want [102:19] people in an office it doesn't have to [102:20] be all one office but I want them in an [102:22] office [102:23] and partly it's for their own benefit if [102:25] you're in your 20s when I was a young [102:27] executive I knew nothing of what I was [102:29] doing I literally was just lucky to be [102:31] there and I learned by hanging out at [102:33] the water cooler going to meetings [102:35] hanging out being in the hallway had I [102:37] been at home I wouldn't have had any of [102:38] that knowledge which ultimately was [102:40] Central to my subsequent promotions so [102:43] if you're in your 20s you want to be in [102:45] an office because that's how you're [102:46] going to get promoted and I think that's [102:48] consistent with the majority of the [102:50] people who really want to work from home [102:51] have honest problems with commuting and [102:53] family and so forth they're real issues [102:56] the problem with our joint view is it's [102:58] not supported by the data the data [103:00] indicates that productivity is actually [103:02] slightly higher in uh work uh when you [103:05] allow work from home so you and I really [103:08] want that company of people sitting [103:10] around the table and so forth but the [103:12] evidence does not support our view [103:14] interesting yeah is that true it is [103:16] absolutely true why is Facebook and all [103:18] these companies rolling back their uh [103:20] and like Snapchat rolling back their [103:21] remote working policies then not [103:23] everyone is um and you most companies [103:26] are doing various forms of hybrids where [103:29] it's two days or three days or so forth [103:32] um I'm sure that for the average [103:33] listener here who works in public [103:35] security or in a government they say [103:37] well my God they're not in the office [103:39] every every every day but I'll tell you [103:41] that at least for the the industries [103:44] that have been studied there's evidence [103:46] that allowing that flexibility from work [103:47] from home increases productivity I don't [103:50] happen to like it but I want to [103:52] acknowledge the science is there what is [103:54] the um the advice that you wish you'd [103:56] gotten at my age that you didn't get the [103:59] most important thing is probably keep [104:01] betting on yourself and bet again and [104:03] roll the dice and roll the dice what [104:05] happens in as you get older is you [104:08] realize that these opportunities were in [104:10] front of you and you didn't jump for [104:11] them why you were in a bad mood or you [104:14] know you didn't know who to call or so [104:16] forth life can be understood as a series [104:19] of opportunities that are put before you [104:21] and they're Tim Limited [104:23] I was fortunate that I got the call [104:25] after a number of people had turned it [104:27] down to work for Larry and for and with [104:29] Larry Sergey at Google changed my life [104:32] right but that was luck and timing my [104:34] one friend on the board at the moment [104:37] said I was very thankful to him and he [104:39] said but you know you did one thing [104:40] right I said what he said you said [104:43] yes so your philosophy in life should be [104:46] to say yes to that opportunity and yes [104:49] it's painful and yes it's difficult and [104:50] yes you have to deal with your family [104:52] and yes you have to travel to to some [104:53] foreign place and so forth get on the [104:55] airplane and get it [104:56] done what's the hardest challenge you've [104:58] dealt with in your life well on the [105:01] personal side you know I've had the I've [105:03] had a set of you know personal personal [105:04] Pro problems and [105:06] tragedies um like everyone does I think [105:09] on a business [105:11] context [105:15] um there were moments at Google where we [105:19] had control over an industry that we [105:21] didn't execute well the most obvious one [105:23] is social [105:24] media uh at the time when Facebook was [105:27] founded we had a system which we called [105:29] Orit um which was really really [105:31] interesting and somehow we we we did [105:33] everything well but we missed that one [105:35] right and I would have preferred and [105:37] I'll take responsibility for that we [105:39] have a closing tradition on this podcast [105:40] where the last guest leaves a question [105:41] for the next guest not knowing who [105:42] they're going to be leaving it for and [105:44] the question left for you is what is [105:46] your non-negotiable something you do [105:48] that significantly improves everyday [105:50] life well what I try to do is try to be [105:53] online and I also try to keep people [105:55] honest every day you keep you hear all [105:59] sorts of ideas and and so forth half of [106:02] which are right half of which are wrong [106:03] I try to make sure I know the truth as [106:06] best we can determine it Eric thank you [106:08] so much thank you it's uh such an honor [106:10] your books are have shaped my thinking [106:12] in so many so many important ways and I [106:15] think your new book Genesis is the [106:17] single best book I've I've read on the [106:19] subject of AI because you take a very [106:21] nuanced approach to these subject [106:23] matters and I think sometimes it's [106:24] tempting to be binary in your way of [106:26] thinking about this technology the the [106:28] pros and the cons but your writing your [106:30] videos your work takes this really [106:32] balanced but informed approach to it I [106:34] have to say as an entrepreneur the [106:36] trillion dollar coach book is what I [106:37] highly recommend everybody goes and [106:38] reads because it's um it's just a really [106:41] great Manual of being a leader in the [106:42] Modern Age and an entrepreneur I'm going [106:44] to link all five of these books in the [106:46] in the comment section below the new [106:48] book Genesis comes out in the US I [106:50] believe on the 19th of November [106:53] um I don't have the UK date but I'll [106:55] find it and I'll put it in but it's a [106:57] book it's a it's a critically important [107:00] book that nobody should avoid I've been [107:02] searching for answers that are contained [107:04] in this book for a very very long time [107:05] I've been having very a lot of [107:06] conversations on this podcast in search [107:08] of some of these answers and I feel [107:09] clearer um about myself my future but [107:12] also the future of society because I've [107:13] read this book so thank you for writing [107:15] it and thank you and let's thank Dr [107:17] Kissinger he finished the last chapter [107:19] in his last week of life in his deathbed [107:22] that's how profound he thought that this [107:24] book was And all I'll tell you is that [107:28] he wanted to set us up for a good next [107:30] 50 years having lived for so long and [107:34] seen both good and evil he wanted to [107:36] make sure we continue the good progress [107:38] we're making as a [107:40] society is there anything he would want [107:42] to say any answer he gave would take [107:45] five [107:46] [Music] [107:48] minutes a remarkable man thank you Eric [107:52] thank you [107:55] [Music] [107:56] I'm going to let you into a little bit [107:58] of a secret and you're probably going to [108:00] think that I'm a little bit weird for [108:01] saying this but our team are our team [108:03] because we absolutely obsess about the [108:06] smallest things even with this podcast [108:08] when we're recording this podcast we [108:09] measure the CO2 levels in the studio [108:11] because if it gets above a th000 parts [108:13] per million cognitive performance dips [108:15] this is the type of 1% Improvement we [108:17] make on our show and that is why the [108:19] show is the Way It Is by understanding [108:21] the power of pounding 1% you can [108:24] absolutely change your outcomes in your [108:26] life it isn't about drastic [108:28] Transformations or quick wins it's about [108:30] the small consistent actions that have a [108:33] lasting change in your outcomes so two [108:35] years ago we started the process of [108:37] creating this beautiful diary and it's [108:39] truly beautiful inside there's lots of [108:41] pictures lots of inspiration and [108:43] motivation as well some Interac Dev [108:45] elements and the purpose of this diary [108:47] is to help you identify stay focused on [108:50] develop consistency with the one % that [108:53] will ultimately change your life we have [108:55] a limited number of these 1% Diaries and [108:57] if you want to do this with me then join [108:58] our waiting list I can't guarantee all [109:00] of you that join the waiting list will [109:01] be able to get one but if you join now [109:03] you have a higher chance the waiting [109:05] list can be found atth diary.com I'll [109:08] link it below but that isth diary.com [109:12] [Music] [109:22] ah