炉边谈话：Sam Altman×Vinod Khosla 以及 IIT 德里的 AMA | 20 二月 2026
本页标题、字幕、摘要和阅读笔记为已保存的 AI 译文，可能有误。时间戳与来源一致，可展开查看原始字幕，并对照原视频核对。
Sam Altman 和 Vinod Khosla 在二月 2026 于 IIT 德里的对话中讨论了 AI、未来工作、学生学习以及创业机会。
转录来源：Vidleaf上的自动语音识别（可能包含错误）。AI标点由Vidleaf添加；保留原始措辞和时间戳锚点。原始转录可在下方获取。
AI生成的摘要
Sam Altman和Vinod Khosla于二月2026在印度理工学院德里分校与学生讨论人工智能、教育和未来工作。他们的对话强调好奇心、能动性和快速学习，同时考虑人工智能如何改变软件和其他活动的经济模式。学生的问题将讨论扩展到安全、职业选择、创造力和能源使用。演讲者提供预测和个人判断，包括对工作将如何变化以及过渡可能有多困难的不确定性。链接的段落保留了这些限定条件，并将给学生的建议与对未来市场和技术的预测区分开来。
要点
- [18:42] 讨论强调学会如何学习、好奇心和能动性是学生应具备的技能。
- [29:04] Altman 将 AI 安全围绕对齐、安全和广泛分布的韧性来构建。
- [39:03] 关于工作的讨论承认了不确定性以及过渡期间可能出现的痛苦变化。
根据文字记录生成。请对照原始视频核查所链接的段落。简要说明和要点由 Vidleaf 于 2026-10-21 审核。
完整概述
本批次涵盖在印度理工学院德里分校的一场会议，Vinod Khosla和Sam Altman讨论人工智能、工作的未来以及学生应该学习什么，同时还有Altman在拥挤的印度之行中的言论。印度理工学院德里分校主任描述了该学院的AI教育努力，包括专门的AI学院、课程改革、部落语言工具Adibani以及即将与OpenAI建立的合作伙伴关系。Altman将超级智能描述为一个不可思议的机会，并表示为稀缺而建立的制度不适用于富足，而Khosla认为GDP将是一个糟糕的指标，因为AI具有如此大的通缩效应；两人都强调学习如何快速学习、好奇心和能动性。Altman引用Emergent在八个月内达到100百万美元ARR的速度，称这是他见过的最快增长，并称这是有史以来开始职业生涯的最佳时机，并敦促冒险、好奇心、能动性和坚持，指出许多人，尤其是印度人，特别厌恶风险。他指出印度已经是OpenAI的第二大市场，拥有100百万ChatGPT用户，其中三分之一是学生，并表示随着时间的推移，它应该成为最大的市场；关于成本，他表示通过API解决一个困难的推理问题的成本比一年半前低约1,000倍，他不认为在18个月内还能再降低1,000倍，但预计会大幅下降。他将AI安全描述为技术对齐、新的安全基础设施，而不是阻止创新，将权力的广泛分配（他们称之为韧性）描述为最佳、最公平和最安全的路径。在由Raghav Gupta主持的学生问答&中，Altman将AI态度描述为一个连续体而非两个阵营，表示没有人真正知道工作影响，会有许多惊喜，并指出历史一再低估新工作出现速度和方式的差异，而过渡对许多人来说将非常痛苦；他还讨论了能源丰富、创造力和水资源问题，并在快速问答环节中将杀人机器人列为最被高估的恐惧，将零人公司列为最被低估的机会，将教育列为五年内变化超过医疗保健的领域，并将韧性列为关键技能。
带时间戳的章节摘要
介绍维诺德·科斯拉和萨姆·奥尔特曼
- [0:18] 主持人表示，预测未来的最佳方式是创造未来，并介绍了两位来自不同世代的嘉宾，他们亲手创造了未来。
- [0:31] 维诺德·科斯拉被描述为全球偶像和校友，他在80年代初联合创立了Sun Microsystems。
- [0:47] 在计算封闭且专有的时代，他倡导开放系统，引领了驱动早期互联网的工作站和Java的创建。
- [0:57] 他后来创立了科斯拉创投，支持旨在解决从清洁能源到数字技术等社会最大问题的想法，并且是OpenAI的第一位机构投资者。
- [1:21] 萨姆·奥尔特曼被介绍为当前科技界最具影响力的人物，指导历史上最强大的人工智能模型的开发。
- [1:30] 作为OpenAI的首席执行官，他被认为开启了生成式人工智能时代，并改变了世界与信息、创造力和智能本身的互动方式。
印度理工学院德里分校主任谈人工智能教育与合作伙伴关系
- [2:08] 萨姆·奥尔特曼和维诺德·科斯拉活动的需求超出了多格拉厅的容量，仅印度理工学院德里分校的注册人数就填满了两个大厅。
- [2:32] 小考正在进行中，一些学生在一小时内完成了两小时的考试以便参加。
- [2:50] 校外参加者受邀参加下周的学生科技活动Trist。
- [3:12] 印度理工学院德里分校将人工智能视为机遇和挑战，必须弄清楚如何培训未来的劳动力。
- [3:30] 成立了一个由教师、学生和工作人员组成的委员会，研究整合负责任的人工智能以及教师和学生使用人工智能的道德规范，作为持续的对话。
- [3:55] 印度理工学院德里分校设有专门的人工智能学院、机器智能与数据科学方向的M Tech项目以及Ph.D.项目，今年将在其阿布扎比校区启动M Tech项目。
- [4:14] 去年的课程审查确保每个学生都能接触到人工智能课程，包括人工智能与可持续发展。
- [4:45] 印度理工学院德里分校与部落事务部合作，开发了一款名为Adibani的工具，支持四种部落语言，包括桑塔利语、比里语和蒙达里语，面向无法获得此类内容的社区。
- [5:12] 印度理工学院德里分校正在与OpenAI建立合作伙伴关系，并正在寻找学生和教师与OpenAI及主权人工智能模型互动的方式，以共同开发负责任的人工智能。
- [5:56] 维诺德·科斯拉于1971年至1976年在印度理工学院德里分校学习，并在印度理工学院系统内建立了最早的计算机实验室之一。
- [6:26] 2004年，维诺德·科斯拉捐款建立了信息技术学院，以他父母的名字命名。
- [6:41] 印度理工学院德里分校是一所拥有65年历史的机构，正通过IIT Delhi 2035重新构想其未来，在基础设施上纵向发展，并寻求科斯拉以想法形式的支持。
人工智能与印度年轻建设者
- [7:50] 萨姆表示，这次与印度年轻人的会议是他此行最期待的。
- [8:00] 他看到印度的年轻人以惊人的能量用人工智能进行建设，初创公司和产品已经出现。
- [8:27] 他称这是年轻人一生中能取得的成就的最大转变。
- [8:37] 他将2047年印度独立100周年作为想象更大转变的时间视野。
- [9:11] 他说观众将进入一个拥有超级智能的世界，这是一个不可思议的机遇。
- [9:21] 他一岁的孩子永远不会知道一个他在原始智力上比计算机更聪明的世界。
- [9:48] 人们现在能用 GPU 做的事情，将成为创造新价值的方式，尽管他和该节点对此只有高层次的猜测。
- [10:09] 教育、医疗保健、工作的含义、通过机器人进行的体力劳动，以及科学发现，都显然将会发生变化。
- [10:43] 为稀缺世界建立的制度大多不太适用于富足世界，因此社会必须被重新构想。
AI 通缩与工作的未来
- [11:24] GDP 最终会被证明是一个糟糕的指标，因为 AI 的通缩效应太强了。
- [11:28] GDP 想要近似衡量的是生活质量，但它是以美元计量的。
- [11:48] 一个能恰当衡量 GDP 试图近似之物的指标，肯定会显示增长率翻倍还不止。
- [12:05] 到 2035，经济将出现巨大的通缩效应，因此应该持有 Google 发行的 100 年期债券。
- [13:08] 即便是从事 AI 工作的人，对这场变革的规模也感到有些茫然。
- [13:17] Codex 在印度的增长速度比世界上任何地方都快，使用者主要是年轻人。
- [13:32] 在 OpenAI 内部，工程师和研究员的含义在如此短的时间内发生了如此大的变化。
- [14:07] 科学发现以及围绕它的经济引擎将同时实现自动化。
- [15:21] 建议是学会如何非常、非常快地学习，因为你的专业领域并不重要。
- [15:46] 一年级学生不会学到任何 AI 不更懂的技术知识。
- [16:52] 学会使用这些工具现在是显而易见的技能，就像上学时学编程一样。
学生最应该学什么
- [17:00] 尽可能多地在边缘上推进似乎非常有价值。
- [17:12] 技术知识没那么重要了，因为电气工程师可以相当轻松地做结构工程师的工作。
- [17:35] 始终站在工具的最前沿极其重要，而且预计会成为一种复利式的加速因素。
- [17:49] 一直做项目有助于弄清楚什么将成为可能，而能力每几个月就会迅速提升。
- [18:06] Y Combinator 的座右铭“做出人们想要的东西”是一项可学习的技能，值得开一门大学课程，而想出好点子会随着刻意练习而提高。
- [18:42] 好奇心和能动性是最重要的学习内容。
- [18:53] 能动性是一项可学习的技能，而世界会把它从你身上打掉，所以你必须抵抗，并继续相信自己能让任何事情发生。
- [19:25] 能动性的回报从未如此之高，一个拥有动力、想法和意志力的人就能让不可思议的事情发生。
信念、能动性与创造未来
- [19:42] 演讲者职业生涯中最重要的学习是：在一个想法上是对的，比它是否受欢迎更重要；不受欢迎反而可能更好，因为没有竞争。
- [19:55] 长期坚持毫不动摇的信念，一个接一个地扫除障碍，在挫折之间不失去热情，就能成就了不起的事情。
- [20:04] 这就是 OpenAI 的故事，它刚起步时看起来完全疯狂，甚至在其头四五年里也像是一个非常糟糕的主意。
- [20:30] 他们在职业生涯早期培养起来的能动性——演讲者希望自己能更早培养——对实现这一切至关重要。
- [20:45] 演讲者引用了 Alan Kay 的话：预测未来的最好方式就是创造未来，并补充道：创造你想要的未来，因为现在这比以往任何时候都更有可能。
- [21:06] 演讲者是四名学生之一，他们在 IIT Delhi 创办了第一个计算机编程俱乐部，此前任何 IIT 都从未有过，因为他们想要它，并让它成真。
- [21:22] 在 1970 年代，演讲者作为一名电气工程师对人体中的电现象感到好奇，说服 Guha 教授与全印度医学科学研究所一起创办了印度首个生物医学工程项目。
- [21:48] 在 16 或 17 岁时，演讲者学到：只要你想，你就能让任何事情发生，这就是能动性的意义所在，而技术和工具让几乎一切皆有可能。
- [22:00] 能动性才是关键，此外还有对学习的好奇心，以及始终站在最前沿。
创业机会与冒险建议
- [22:17] Emergent在七月或八月推出了一款产品，上周报告称超过了100百万美元的ARR。
- [22:26] Emergent 在八个月内就达到了这一年化收入，这是演讲者见过的最快增长，甚至与 Loveable 或 Replit 相比也是如此。
- [22:41] 演讲者称这是一个特殊的时代，并认为这是有史以来开启职业生涯的最佳时机。
- [22:58] 演讲者说，大多数人，尤其是印度人，特别厌恶风险。
- [23:06] 一个哈佛商学院的案例以演讲者的话开头：“我愿意失败，这让我得以成功。”
- [23:12] 演讲者鼓励听众勇于冒险，并利用工具、好奇心和能动性。
- [23:25] 演讲者还补充了坚持，因为道路从来不是简单平坦的，你必须坚持度过艰难时期。
学生问答&与OpenAI印度负责人
Sam Altman 谈印度与 AI
- [24:33] 主持人说，这似乎是Sam此次行程紧凑的印度之行中最激动人心的一场，Sam也表示同意。
- [24:55] 观众由学习工程、计算机科学、健康科学、管理和设计的聪明年轻人组成。
- [25:08] 共提交了超过800个问题；主持人筛选了其中一些，并将邀请学生现场提问。
- [25:27] Sam在《印度时报》文章中的标题写道，AI将定义印度的未来，而印度将定义AI的未来。
- [25:43] 印度已经是OpenAI的第二大市场，拥有100百万ChatGPT用户，其中三分之一是学生，并且是全世界增长最快的市场。
- [25:54] Sam表示，随着时间的推移，印度应该成为OpenAI最大的市场，而印度的建设活力似乎比任何其他地方都增长得更快。
AI成本下降与可及性
- [26:35] Mukund问Sam，如何将智能成为廉价公用事业与高推理模型被高级订阅门槛限制这两点协调起来。
- [26:54] Sam表示，正在发生的惊人事情之一是，在给定智能水平下，AI变便宜的速度。
- [27:07] 通过OpenAI API解决一个困难推理问题的成本，比一年半前低了约1,000倍。
- [27:23] Sam不认为未来18个月还能再保持1,000倍的降幅，但他预计成本会显著下降。
- [27:35] 目标是坚持不懈地将智能成本尽可能降至接近零，并使其作为一种极低成本资产提供给全世界，可及性是这一使命的关键部分。
- [27:54] Sam指出，将会出现社会性冲击，因为社会一直习惯于智能是相对稀缺的资源，而且如果物理世界的能力也发生变化，社会对于事情将如何运作尚无共识。
- [28:28] Sam相信智能成本将呈下降趋势，这对于赋能全世界非常重要。
AI安全：对齐、安全与韧性
- [29:04] Sam Altman表示，AI安全有三个部分，第一部分是技术对齐工作。
- [29:13] 对齐方面的进展比他预期的要好，但监督比我们聪明得多的系统需要新的想法。
- [29:45] 总体而言，对齐看起来比他担心的要容易，但显然尚未完成。
- [29:53] 第二部分是一种新型安全基础设施，提示注入就是一个很好的例子。
- [30:03] 他原本打算不让OpenClaw和Codex这类工具以root权限运行，但这只持续了大约两个小时，因为批准所有操作很烦人，便利性占了上风。
- [30:30] 他认为需要一种独立于模型技术对齐的新型安全架构，以便人们可以使用智能体而不会发生疯狂的事情。
- [31:01] 第三部分也是最重要的部分，是不阻碍创新，他对此非常赞同。
- [31:07] 他勾勒了两条漫画化的路径：一家AI公司、一个国家或一个模型拥有有效的极权控制且没有安全问题，对比超级民主化访问并将强大AI交到每个人手中。
- [31:42] 在民主化的世界里，有些事情会出错，人们会以小的、并且可能是大的方式滥用AI。
- [31:50] 他主张，如果大多数人是善良的，并且只要不存在巨大的权力失衡，好人就会阻止坏人，那么AI就可以民主化。
- [32:19] 他称以广泛权力分配赋能每个人是最佳、最公平、最安全的路径，他们称之为韧性。
- [32:33] 他说，OpenAI仅仅阻止其模型告诉人们如何制造病原体已经不再奏效，因为某个开源模型会帮助某个坏人，所以社会必须准备好通过快速响应疫苗和阻断呼吸道传播的新方法来防御。
OpenAI的印度战略与创新
没有病历情况下的印度健康AI
OpenAI职业机会与印度扩张
- [35:10] 一名IIT德里计算机科学三年级学生问，IIT本科生如何直接加入OpenAI，以及OpenAI目前在印度的录用目标和未来目标是什么。
- [35:19] Sam表示，OpenAI很希望这名学生加入，无论是在印度办公室还是在美国。
- [35:30] Sam表示，未来几年将是AI发展中最激动人心的几年。
- [35:44] Sam表示，在OpenAI工作最酷的部分是看到未来，并比其他人早六个月使用Codex等工具。
- [35:55] Sam 邀请提出更尖锐、更具争议性的问题，并表示他不会因此感到冒犯。
- [36:04] Sam 表示 OpenAI 在德里设有办公室，并且刚刚宣布在班加罗尔和孟买设立办公室，团队正在快速壮大。
人类在 AI 时代的优势
- [36:30] 阿米蒂大学一名人工智能专业三年级学生提问：当 18 岁的人与 AI 竞争工作岗位时，人类拥有什么不公平的优势。
- [36:46] 观察顶尖人才如何使用 AI 令人惊叹，与一年前截然不同，有些人同时操作六台显示器，每个显示器会话中运行三个智能体。
- [36:59] 现在一个人就能完成过去一个工程师团队所做的工作。
- [37:04] 在一个人的头脑中把握整个问题，同时管理一群智能体，能够实现以往不可能达到的速度和质量。
- [37:15] 在思考系统或产品的高层战略方面，人类仍然比 AI 强得多，尽管未来这种情况可能会改变。
- [37:28] 熟练使用 AI 的人，相比 AI 本身以及不使用 AI 的人，拥有巨大优势。
- [37:37] 长期优势尚不确定，但人类似乎天生就关心他人，而对机器则漠不关心。
AI 就业恐惧与技能
AI 对就业和未来工作的影响
- [39:03] 你必须处于这个光谱上的某个位置，不可能处在任何一端。
- [39:10] 其中很大一部分是对就业的影响。
- [39:16] 没有人真正知道，包括我们在内，而且会有很多意外。
- [39:25] 有些工作会完全消失，但许多工作的变化并不会像看起来那样大。
- [39:37] 人们仍然希望由人类医生来告诉他们情况，即使 AI 给出的建议更好。
- [39:48] 创业者仍然会创办公司，只是会发展得更快，对现有工作的颠覆性更强。
- [39:59] 历史上每一次重大技术变革，我们都低估了新工作出现的速度以及它们有多么不同。
- [40:23] 如今有一些非常好的工作，在 20 年前根本不存在这个类别。
- [41:25] 我们永远不会无事可做，也永远不会达到理论上的无所事事状态。
- [41:48] 许多人最终会从事今天甚至作为一种概念都不存在的工作。
- [41:53] 这个过渡并非没有痛苦；对许多人来说，它将非常痛苦。
- [42:02] 随着社会迅速变得更加富裕，我希望我们能找到新的社会经济模式，让这个过程不那么痛苦。
Sam 谈绿色 AI 与能源充裕
- [42:32] 印度理工学院德里分校 AI 学院二年级博士生 Kaustubh 提问：OpenAI 正在采取哪些措施来实现绿色 AI 解决方案。
- [42:49] Sam 表示，距离核聚变以及真正可靠的太阳能、风能加储能，世界只有几年时间了。
- [42:57] 他进一步认为，这些将成为世界上最便宜的能源来源。
- [43:04] 当这一切发生时，经济会按其规律运转，不仅是 AI，其他所有人都会使用新型能源。
- [43:11] 他说，一个较少被讨论的点是，推动能源的巨大充裕，就像智能的充裕一样，将极大地提升世界各地的人们。
- [43:31] 他说，他们显然在环境方面关心此事并致力于此，并且将此视为对世界的责任，让能源变得充裕。
AI 与人类创造力
- [44:08] 一名学生提问：AI 正在成为一种能够生成图像、空间和整个概念的工具。
- [44:35] Sam 说，他过去对此更为担忧，当模型能够创造任何东西时，他不确定会怎样发展。
- [44:50] 在实践中，创作者普遍表示 AI 是他们创作过程中最好的工具，尽管并非所有人。
- [45:08] AI 可以缩短从想法到看到结果再到新想法的循环。
- [45:27] 创作者在意的是人类精神、方向和印记仍然留在作品中，并认为他们会创作出比以前更好的东西。
- [45:53] 人们实际上对据称由 AI 创作的艺术赋予零价值，即使它很美。
- [46:07] 由人类创作的艺术，即使使用了 AI 工具，当创作者有故事并在作品上署名时，就很有价值。
Sam Altman 谈 AI 的水和能源问题
- [46:37] 主持人指出嘉宾是一位非常受欢迎的电影导演，并期待稍后进一步交流。
- [46:43] 主持人表示时间快到了，将再提两个问题。
- [46:54] 阿米蒂大学人工智能专业三年级学生 Paridhi 被点名提出下一个问题。
- [47:05] Paridhi 提问：有人担心用于冷却服务器的水会导致水资源短缺，OpenAI 对此采取了什么措施。
- [47:27] Altman 表示，能源问题以及转向绿色能源的必要性非常真实，在 AI 发展的规模下可能成为一个重大问题。
- [47:31] Altman 表示，水资源问题已不再真实，因为数据中心基本上不再使用蒸发冷却，尽管生产半导体确实会消耗一些水。
- [47:52] Altman 表示，AI 数据中心并不是用水大户。
- [47:58] Altman 称，每个 ChatGPT 问题消耗约七加仑水的说法是一个广为流传的梗，完全不是真的，同时表示关注能源确实有道理。
Sam 谈年轻人与 AI 建议
关于AI与学生技能的快速问答&A
闭幕感谢与感恩纪念品
棒球帽读心预测与闭幕自拍
[0:03] 今天是多么美好的一天，能成为IoT Delhi社区的一员。让我们再次为我们的嘉宾热烈鼓掌。来吧。他们说预测未来的最好方式就是创造未来。今天我们有两位来自不同世代的杰出人物，他们各自通过亲手建设为世界预见了未来。我们的第一位嘉宾是一个在这些领域深深共鸣的名字。他不仅是全球偶像，也是我们自己的校友。在80年代初，他联合创立了Sun Microsystems，一家从根本上改变了硅谷结构的公司。
[0:47] 在计算封闭专有的时代，他倡导开放系统。他的愿景催生了驱动早期互联网的工作站和Java的诞生。后来，他创立了Khosla Ventures。他始终支持旨在解决社会最大问题的想法，从清洁能源到数字技术，并作为OpenAI的第一位机构投资者。女士们先生们，让我们再次为Vinod Khosla先生热烈鼓掌。与他一起。
[1:21] 我们请到了可以说是当前技术领域最具影响力的人，他正在引导历史上最强大的AI模型的发展。作为OpenAI的首席执行官，他开启了生成式AI时代，改变了世界与信息、创造力以及智能本身的互动方式。为Sam Oiltman先生热烈鼓掌。为了正式欢迎我们尊贵的嘉宾，并为这一历史性活动设定背景，我现在想请IIT德里主任Rangan Banerjee教授上台。
[2:02] 早上好。很高兴看到这么多人在此。当得知Sam Altman先生和Vinod Khosla先生要来时，我们很快意识到Dogra Hall不够用。而你们所有人，实际上我们能有两个Dogra Hall是一种荣幸，如果我们想容纳所有从IIT德里注册的人，呃...
[2:32] 我们正在进行小考。我不知道你们中有多少人跳过了小考来到这里。有人告诉我，有些人一小时就完成了两小时的考试，然后赶到这里。你们中有多少人来自IIT以外？请举手。太好了。所以我们希望你们所有人也能来参加我们的学生技术活动Trist，下周。我想借此机会说，作为一所高等教育机构，IIT德里认为AI提供了机遇，也带来了挑战。我们需要弄清楚如何培养即将进入劳动力的未来一代。为此，我们成立了一个由教师、学生和员工组成的委员会，研究
[3:35] 我们如何整合负责任的AI，以及教师和学生使用AI的伦理是什么。这是一个持续的对话。我们再次，我们有一个...专门的AI学院。我们有一个机器智能与数据科学的M Tech项目。我们有一个博士项目。我们还在建立，今年我们将启动机器智能与数据科学的M Tech项目。
[4:14] 在我们的阿布扎比校区。我们去年进行了课程审查，在那次课程审查中，我们确保学院的每个学生都能接触到AI课程，AI与可持续性。这是课程审查的特点之一。我不打算继续讲这个，因为我们想听Sam Altman先生和Vinod Khosla先生讲话。我只想说，给你们举一个例子。
[4:45] 我们与部落事务部合作，开发了一个名为 Adibani 的工具，它支持四种部落语言：桑塔利语、比里语、蒙达里语等。我们正在探索如何为社区提供人工智能，特别是在目前无法获得这类内容的社区。我们正在寻找各种途径，并且很高兴能与 OpenAI 建立合作伙伴关系。
[5:19] 我们正在探索各种方式，让我们的学生和教职员工既能与像 OpenAI 这样的公司合作，也能与主权人工智能模型互动，以便我们能够共同开发并创造负责任的人工智能，从而改善人们的生活。谢谢。我想说，我们非常高兴能邀请到 Sam Altman 先生来到这里。在人工智能和生成式人工智能的思考方面，他一直走在时代前沿。
[5:52] 我们也很高兴邀请到我们自己的 Vinod Khosla 先生。他一直走在时代前沿。作为这里的学生，我想是在 1971 到 1976 年间，他在 IIT 系统中建立了最早的计算机实验室之一。所以他在那时就走在时代前沿。在 2004 年，我想，他也走在时代前沿，他捐赠给我们研究所建立了信息技术学院，该学院以他的父母命名。
[6:34] 而我们 IIT 德里现在是一所拥有 65 年历史的机构。我们正在重新构想我们的未来，我们希望在教育方面走在时代前沿。我们正在关注 IIT 德里 2035。我们正在关注改变我们的基础设施，我们正在向垂直方向发展。我们希望以想法形式得到 Khosla 先生的支持。
[7:08] 在我们进行这种重新构想的过程中。所以，说了这几句话，我想说，像大家一样，我期待着这次对话。非常感谢。谢谢。我们应该从哪里开始？希望有人能让我们守时。Sam，也许首先，我总是很兴奋看到这么多年轻人，所以也许你可以谈谈你在这里的观察。
[7:46] 是的。对我来说，这次印度之行非常精彩，今天我还有更多令人兴奋的事情，但这是整个旅程中我最期待的事情。我观察着世界各地年轻人正在做的事情，尤其是印度，他们以惊人的能量构建人工智能。将要出现的新初创公司已经在出现。将要创造的新产品和服务，对人工智能的全面拥抱。以及那种兴奋和渴望，将所有这些能动性、能量和速度带入未来。
[8:23] 太棒了。我……我不知道你们怎么想，但我认为这是一个年轻人所能取得的成就格局中发生的最大转变。在我的一生中。看到这一点真是令人惊叹。观察人们正在做的事情简直不可思议。印度在 2047 年迎来独立 100 周年。我无法想象到那时会有比任何人想象的更大的转变。所以想象一下 2047，到那时印度会是什么样子。我们没有准备问题，我们只是想进行一场对话。
[9:04] 我的意思是，这将取决于所有这些人的行动。你们都将进入……崇拜它。带着超级智能。这将极大地改变事物，但这是一个不可思议的机会。我有一个一岁的孩子。他永远不会知道一个他比计算机更聪明的世界。永远不会。他出生得正是时候，在原始智力上永远不会被计算机超越。
[9:35] 而且，你知道，对他来说这将是完全正常的生活。对我们来说，这是一个巨大的适应。对你们所有人来说，这有点奇怪，但你们对此感到兴奋。人们现在能用 GPU 做的事情将成为世界上创造新价值的方式。所以我不知道那会是什么样子。节点也不知道那会是什么样子。我们有一些高层次的猜测。但我们不会是那些将其构建到……所有这些人们将用其创造的不可思议的新公司中的人。
[10:09] 我们都喜欢的例子是，教育显然将彻底改变，医疗保健显然将彻底改变，工作的意义将彻底改变，公司，也许几个人和一个巨大的数据中心。并构建不可思议的新型价值。体力工作的意义将改变，因为机器人可以自动化越来越多的制造业。当然，发现科学的意义也将改变。世界经济价值的增长率，生活质量的提高，我引用最近一位 OpenAI 的人对我说的话，它真的让我印象深刻。
[10:43] 是我们建立了许多本能、制度和政策结构来应对稀缺的世界。但几乎所有这些都不适用于富足的世界。除了建立所有这些新型公司、产品和服务外，社会将不得不为这个不可思议的富足世界重新构想。你对此思考了很多。是的，我花了很多时间。让我们问一些实际问题。你认为 GDP 增长率，无论经济学家今天预测什么，每个国家都有其 15 年和 25 年预测。
[11:20] 你认为它们会翻倍，还是不止翻倍？我认为 GDP 将成为一个糟糕的指标，因为人工智能具有如此大的通缩效应。我们想要衡量的，我认为 GDP 旨在近似的是生活质量。是的。以及生活质量。从技术上讲，它是商品和服务的定义。但它以美元衡量，而货币将完全不同。是的。所以我认为，一个能恰当近似 GDP 试图衡量的指标，增长率肯定会不止翻倍。
[11:53] 嗯……以名义美元衡量的GDP。那可能有点不太靠谱。其实，我想问你一个奇怪的问题。五年后的利率。2%还是20%？——我想，到2035年肯定，我们会有一个极度通缩的经济。这是世界还没有在思考的。有意思的是，谷歌刚刚发行了100年期债券。我说，到2035年，你应该持有所有这些债券，因为世界会变得如此通缩，很难想象有任何人会假装不是这样。这就是那种概念，你知道，世界还不知道变化有多大，而他们还在照常行事。我想你的一位工程师写了一份很棒的宣言。你能谈谈那个吗？
[12:54] 哪份宣言？人们会写好几份这种东西。我忘了名字。我刚才在对话中早些时候提到过。本质上就是说，世界对正在发生什么以及变化的规模一无所知。我马上就会想起来。我认为即使是从事AI工作的人，对变化的规模也感到有点茫然。Codex就是一个例子。顺便说一句，Codex在印度的增长速度比世界上任何地方都快，而且主要是年轻人在用。但Codex就是一个例子，说明每个人的直觉，尽管我们在理智上知道这一点，在情感上我们都被它击中了。
[13:32] 我们在OpenAI内部有大量Codex的使用权限，而作为一名工程师意味着什么、作为一名研究者意味着什么，变化得如此之快、如此之大。而这在整个经济中只会一次又一次地发生。呃，谢谢。还有谢谢。如果你相信——我也相信——经济和生活一年比一年更好的顶层驱动力是科学发现。那么第二重要的事情就是围绕它的经济引擎，把这些新能力带到世界上。
[14:07] 而这两者将同时被自动化。我们将自动化科学进步。我们将自动化整个经济。你想想如果这些事情真的能发生，那到底意味着什么。比我们自己能做到的快10倍、快100倍。我们并没有……我们并不是进化出对那种增长率、那种复利效应的直觉的。所以我不知道。但如果你回到这里当学生，你明天会做什么？不，我想过很多。我框定想象15年后最低限度变化的方式。
[14:44] 所以想象一下2040。我得说，我想象的生活是，现在和2040之间的差距，与今天和1970之间的差距一样大。所以回顾一下1970年的技术。这就是我对Delta的假设，而我认为那还是太保守的想法。——我认为那太保守了。——是的，所以那就是变化的规模。
[15:17] 人们问我，那我们该做什么？我说，学会如何非常非常快地学习。比如，你的专业领域并不重要。你是电气工程师还是机械工程师，基本上将变得无关紧要。事实上，我有点吓到了主任，我说：“你学的东西甚至在2030年都不会有用。”当一年级学生毕业时，你不会学到任何AI不比你更懂的东西。我知道，Sam，你是否同意我。
[15:54] 你不会更懂任何技术领域。你可能会更好地理解人们想要什么。你可能会更好地理解如何与人互动、如何激励人、如何想出创意、如何快速适应、创造价值。但是，是的，你学到的任何技术技能，它们能帮你变得更好。这仍然值得学。你知道，就像……最近有人问我，我后悔学编程吗？
[16:25] 阿门。因为显然AI是比你更好的程序员。而答案是，一点也不后悔。我也不后悔学数学。你知道，我成年后从没用过数学。学习如何思考。学习如何快速学习。这些都是宝贵的技能，但真正重要的是那些元技能。嗯，我是说，我想我们能自信地说的一项具体技能是，我上学时学编程是显而易见的事。现在学习使用这些工具是显而易见的事。
[16:56] 以及人们每一代人所获得的流畅度。如果你尽可能地在边缘上推进，那似乎非常有价值。如果你是一名大学生，你认为什么最重要？什么会有影响？那么他们拥有的技术知识就不重要了。电气工程师可以相当轻松地做结构工程师的工作。顺便说一句，我们今天既在培养结构工程师，也在培养芯片设计师和电气工程师。
[17:28] 那将会……被普遍使用，无处不在。那你认为什么重要？我，嗯，我知道我刚说过这个，但我怎么强调都不为过，站在工具的最前沿有多么重要。我预计这会成为一个复利式的加速器。所以不断做项目——只是为了弄清楚什么将成为可能。而且从现在起，每过几个月就会迅速增加。那真的很重要。
[17:59] 嗯，我希望我大学开设的那门课。嗯，Y Combinator的座右铭是做出人们想要的东西。而且我认为，对于20……一年就是现在。一直是最好的投资者模型。我希望我的学校开设一门关于这个的课。因为我认为那是一项可学习的技能。弄清楚这一点，在一个你可以让任何事情发生的世界里，弄清楚如何获得好想法，这将会变得越来越重要，而像其他任何事情一样，刻意练习似乎能改善它，那似乎很重要。
[18:40] 谢谢。是的，那么。从我个人角度看，最需要学习的是好奇心和能动性。就是告诉我你是否同意、不同意，或者详细阐述一下。我完全同意。我认为能动性也是一项可以学习的技能，世界会把它从你身上打掉，所以你必须去抵抗，继续相信自己能够弄清楚任何事情、让任何事情发生。你做得越多，就越能培养出自信。嗯，很多人在职业生涯后期才学到这一点，早点学到是非常有价值的，就像其他事情一样？
[19:19] 你只需要开始。而且你必须坚持下去，你就会在这方面建立起信心。但是……能动性的回报。显然从未如此之高。你知道，我们长期以来一直在谈论一个人借助AI能够做到什么。我们现在终于看到了。一个有干劲、有想法、有意志力的单个人。可以让不可思议的事情发生。而且。在某种意义上，我整个职业生涯中最重要的学习就是：如果你在想法上是对的，无论它多么不受欢迎，事实上，往往不受欢迎更好，因为你在做这件事时没有其他人和你竞争。
[19:55] 而且你以坚定不移的信念长期坚持下去，你只是一个接一个地清除障碍，在每次挫折之间都不丧失热情。你就能取得惊人的成就。这当然就是OpenAI的故事。现在它显然看起来像是一件伟大的事情，但当我们开始的时候，它看起来完全疯狂，就像我们一群人待在一个房间里，凭借意志力，以及一种无论如何都要做成的态度。然后甚至在OpenAI的头四五年里，它看起来也像是一个非常糟糕的主意。
[20:26] 顺便说一句，谢谢你相信它。而且，呃，我们所有人在职业生涯早期培养起来的能动性，我真希望我们更早培养出来。对让它成为现实至关重要。是的，我想说你们听到了几句引语。我想有一句可惜的是，是的，预测未来的最好方式就是发明它。我接着补充一句。那是Alan Kay说的。而我说，发明你想要的未来。
[20:57] 因为现在比以往任何时候都更有可能。关键在于能动性，而它一直都可获得，你可以让你想要的世界发生。主任说：“我是四名学生之一，我们在IIT Delhi创办了第一个计算机编程俱乐部，而此前任何IIT都从未有过任何编程。我们只是想要它，所以我们就让它发生了。我还和这里的Guha教授一起在1970中创办了生物医学工程项目，那是另一方面。
[21:30] 那就是，作为一名电气工程师，我对人体中的电现象产生了好奇，并说服Guha教授与全印度医学科学研究所一起，在印度任何地方创办了第一个生物医学项目。我举这个例子是因为我当时是16，我当时是17。如果你想，你可以让任何事情发生，这就是能动性的意义。然后技术和工具让几乎任何事情都成为可能。能动性才是关键。对学习的好奇心，以及始终保持在最前沿。
[22:05] 一个让世界发生的机构。从来没有过创办初创公司的机会。我对我们投资的这家小公司感到非常兴奋。他们几个月前联系我，Emergent，在七月或八月推出了一款产品。上周他们打电话给我，说他们已经超过了100百万美元的ARR。在八个月内。这是我见过的最快增长，甚至Loveable或Replit或任何那些公司，所以这个机会对你们所有人都存在。我知道你是否想做一个总结评论。
[22:41] 我真的说不出比这更好的了。这就像一个特殊的时刻。就像我，我认为这是有史以来开始职业生涯的最佳时机。而且，机会是不可思议的。- 是的，我用我有的最后一句引语来结束。大多数人，尤其是在印度，特别厌恶风险。有哈佛商学院。我们有一个案例，开头引用了我的话：“我愿意失败，这让我能够成功，所以我真的鼓励你们所有人去承担风险，有那么多机会可以利用，利用工具、好奇心。嗯，能动性、承担风险，然后是坚持，因为道路从来都不简单或平坦。所以你必须坚持度过艰难时期。
[23:36] 谢谢，我想有。我们到了谈话的尾声。好的。多么精彩的对话，各位。一个月前我们请到了Sunita Williams女士，学习对如此重要的人物、如此杰出的巨擘来说都是如此重要的事情，这个理念。我认为这是我们在这里学到的最令人谦卑的事情之一。现在，我们想开放现场，让学生们提出自己的问题。为此，我想邀请OpenAI印度教育负责人Raghav Gupta先生上台。
[24:10] 他目前正在领导与学术机构的合作，确保AI在所有维度上得到负责任的整合。Raghav先生。早上好。看到满座的礼堂真是不可思议。嗯，Sam，我知道你在印度有忙碌的几天和许多高级别会议，但对我来说，这似乎将是最令人兴奋的一场。你同意吗？我同意。我也很想问问学生们。你们都对这场对话感到兴奋吗？
[24:43] 我是说，拜托。现在我们可以做得更好。你们都对与Sam的这场对话感到兴奋吗？所以我们这里有一些印度聪明的年轻头脑，他们在学习工程、学习计算机科学，我想我们还有健康科学、管理、设计等等。嗯。我想你们所有人都给我们发了800多个问题。我们无法全部采纳800个，但我已经筛选出了重点团队，我们会邀请你们现场提问。
[25:18] 我先来简单开个场。Sam，你本周早些时候在《印度时报》上发表了这篇文章，标题很好地抓住了你的核心信息，文章说AI将定义印度的未来。而且，嗯，印度也将定义AI的未来。我知道在AI峰会上，你看到了很多AI发展方面的动态。你想谈谈你在那里的印象吗？印度已经是我们的第二大市场。有一亿ChatGPT用户，其中三分之一是学生。全球增长最快的编解码器市场。随着时间的推移，印度应该成为我们最大的市场。
[25:54] 不仅仅是使用像Chachi EBT这样的产品，印度的建设活力似乎比任何其他地方都增长得更快，真正创造新型产品和服务、全新公司的愿望都围绕着AI。是的，这真是太棒了。所以能来这里真是太好了。太棒了。那么接下来我们将转向学生，我们会现场开放提问。我想我们有Mukund来问第一个问题。Mukund来自IIT德里。那我们就从Mukund开始吧。
[26:28] 嗨，Sam，我是Mukund，来自德里理工学院数学系。我正在读最后一年，攻读综合硕士学位。所以我的问题是，你谈到智能正在成为一种廉价的公用事业，但我们看到专门的高推理模型被高级订阅所限制。那么你如何将广泛访问的概念与这些前沿模型的经济可行性相协调？我认为最……了不起的事情，嗯，有很多了不起的事情，我不能说最了不起的，但正在发生的事情之一是，在给定智能水平下，AI变得越来越便宜的速度。嗯，我们……如果你看一个困难的推理问题，用OpenAI API解决它一年半前和现在相比，成本下降了大约一千倍。
[27:18] 1,000倍在18个月内，这并不常见。我不认为我们能在接下来的18个月里再保持一千倍，但也许我们会。但成本会有显著下降。所以……这里的目标是坚持不懈地将智能的成本降至尽可能接近零，然后将其作为一种非常低成本资产提供给世界，嗯，出于一系列原因，可及性是其中之一。这是我们使命的关键部分。
[27:54] 围绕这一点将会出现一系列社会颠覆，就像社会的一部分可能已经改变了。谢谢。我们之前谈到了这个关于丰裕与稀缺的问题，但社会已经习惯了智能是一种相对稀缺的资源。还有，在物理世界中实现事情的能力，能源、机器人等等，如果所有这些都改变了，我不认为社会已经就事情将如何运作以及极限、自然极限将在哪里达成共识。
[28:28] 我确实认为智能的成本将会持续下降，就赋能整个世界而言，这非常重要。接下来我们将请Tejas，他也来自IIT德里。我知道主任也谈到了安全，我想Tejas，你的问题与此相关。那么请讲。嗨，Sam。我是Tejas Rajmangla。我目前是IIT德里的最后一年学生。我正在攻读计算力学学士学位。我给你的问题是，我们如何在不妨碍创新的情况下保持AI安全，比如最近发生的提示注入之类的事情？
[29:04] 我认为这里有三个方面。第一，我们必须弄清楚如何在技术上对齐AI。我们在这方面取得的进展比我预想的要好，但当我们达到那些真正比我们聪明得多的系统时，仍然有事情要做，我们如何监督那些系统，需要新的想法，当我们进入一个AI编写如此多源代码和进行如此多研究的世界时，我们无法审查所有内容。有很多曾经感觉像科幻小说的事情现在变成了明确的、我们必须解决的对齐挑战。我们在这里做工作已经有一段时间了。其他实验室也在这里做工作有一段时间了。
[29:45] 总的来说，对齐看起来比我担心的要容易，但我们显然还没有完成。所以第一类是围绕对齐的技术安全工作。第二是需要建立一种全新的安全基础设施。你提到了提示注入。我认为那是一个很好的例子。OpenClaw是一个了不起的东西。Codex是一个了不起的东西。我以为我会是一个非常有纪律的人，不让它们在我的电脑上以root权限运行。因为我认为那是一件疯狂的事情。
[30:17] 这持续了大约两个小时。然后我就想，哦，批准所有事情真的很烦人。太方便了。我只想……我就想，让它去吧。如果我愿意这样做，很多其他人也会愿意这样做。我应该是那里比较保守的人之一。所以，我认为我们需要开发一种新型的安全架构，不同于模型技术对齐，让人们能够使用这些代理而不会发生疯狂的事情。比如，你可以在你的电脑上运行一些东西，它可以了解你生活的一切，记录你的屏幕，为你使用网络等等，这真是太棒了。
[30:52] 如果它开始随意向不该知道的人透露你的秘密，那可就远没那么厉害了。然后第三点，也是最重要的一点，你提到不要阻碍创新。我完全同意。那种……这些是夸张的说法，因为每个人都处在中间地带，但世界目前正在审视的两条路径。要么我们接受，谢谢。你知道，一家AI公司或一个国家或一个模型对世界拥有某种有效的极权控制，而这是以牺牲安全为代价的。
[31:24] 根本没问题。你有一个AI统治一切。这一直是人们长期以来的愿景。另一种是我们极大地民主化访问权限，把强大的AI交到每个人手中。我们相信，如果你知道，在100个人中，有99个会是好人。一个人并不坏。好人会阻止坏人。在那个世界里，有些事情会出错。人们肯定会滥用AI。以小规模，可能也以大规模的方式。长期以来，我认为世界已经表明，如果人们基本上大多数人是好的，而且只要不存在巨大的权力失衡，好人就会阻止坏人，我们就能走向一个民主化AI的世界。
[32:03] 每个人都被提升，而这在某种程度上制约了新的强大坏力量。这是……社会长期以来的故事，它会是混乱的。事情会出错。再见。我认为赋予每个人权力并实现权力的广泛分配是最好的、最公平的和最安全的前进道路。我们称之为韧性。所以不仅仅是它，有对齐，有安全，然后还有这个更新的韧性理念，我认为OpenAI仅仅说我们将阻止我们的模型告诉人们如何制造病原体已经不再奏效，因为某个开源模型会帮助某个坏人制造病原体。
[32:42] 我们必须让社会准备好用快速响应疫苗、某种阻止呼吸道传播的新方法等等来防御它。嗯，是的，是的，这就是我想说的。所以，下一个问题来自Yash，他不在场，所以我会代表他读出来。我读到去年在印度进行了很多关于ChatGPT学习模式的研究。那么，如果OpenAI要专门为印度构建一个产品，不是适配而是首先为印度设计，你希望它解决什么问题？
[33:15] 我宁愿你们中的一个人找出那个答案。我们……OpenAI之外将会发生比OpenAI内部多得多的创新。我们的策略是构建核心团队、模型以及使用它的基本方式。ChatGPT、API、像Codex这样的代理和通用代理，并真正赋能其他所有人去构建它。我从OpenAI学到的一点是，世界上最多的创新来自初创公司，我们想要赋能他们。
[33:46] 谢谢。太好了。我想我们要稍微换个话题，转向一个与健康相关的问题，也就是，我将来自我们在Ames的细分领域。你拿着麦克风。是的，请讲，Adnish。你好。我叫Adnish。我是Ames新德里的一名三年级医学生。我的问题是关于ChatGPT健康的。我们知道ChatGPT健康连接Apple Health和医疗记录。但在印度，人们，大多数人两者都没有。
[34:18] 他们大多只有一部手机。所以我的问题是，你会如何重新思考像印度这样的国家的健康AI。我很喜欢用ChatGPT来处理健康问题。我没有把它连接到任何东西。我输入很多笔记。每次我拿到血液检查之类的，我就上传它。每次我有症状，我就输入进去。它在没有医疗记录的情况下运行得非常好。我们很乐意连接到任何在印度出现的重要平台，但ChatGPT的巨大流量中有一大部分。
[34:50] 是与健康相关的查询，而且完全没有连接器，我们已经努力让它变得非常好。我想你会喜欢下一个问题，它来自Vijay，他来自IIT德里。Vijay，你拿着麦克风。嗨，Sam。我是Vijay。我是IIT德里计算机科学系的三年级学生。所以我想知道IIT本科生如何能直接加入OpenAI，以及OpenAI在印度提供的当前和未来目标是什么？我们很希望你能加入OpenAI。
[35:27] 我们很希望你能加入OpenAI，无论是在我们印度的办公室还是在美国，我认为未来几年将是AI发展中最激动人心的几年。嗯，而且最酷的，好吧，希望作为OpenAI有很多很酷的部分，但我认为最酷的部分是你能看到未来，并比其他人早六个月使用这些工具。而这在开发过程中所促成的工作是极好的。所以我们很希望听到你的消息。顺便说一句，你可以问我更难、更有争议的问题。
[35:58] 我不会被冒犯。就像，放马过来。我喜欢那些。我想你可能已经看到，你知道，我们在德里有一个办公室。我们刚刚宣布在班加罗尔和孟买也设立办公室。所以团队正在迅速壮大。所以我们很乐意参与这场对话。好的，我们接下来去Sahaj。Sahaj，抱歉，嗯，他来自Amity大学，你拿着麦克风。嗨，Sam，我是Sahaj，我来自Amity大学，我是一名三年级学生，在Amity大学攻读人工智能专业。所以我的问题是，在当今市场上，18岁的孩子在与AI竞争工作，我们人类有什么不公平的优势？
[36:40] 很多。现在看着最顶尖的人用AI工作，真的非常不可思议，和一年前完全不同。但你看这些人，桌上摆着六台显示器，每台显示器上跑着三个智能体，一个会话接一个会话地推进。一个人就能完成过去一整个团队——嗯，很多工程师——才能完成的工作。当你能把整个问题装在一个人的脑子里，同时管理一群智能体时，你就能以以前根本达不到的速度和质量把东西做出来。
[37:15] 而且仍然有——这一点未来可能会改变——但仍然有一些事情是人类非常擅长、而AI非常不擅长的，那就是思考一个系统或一个产品之类的高层战略。熟练使用AI的人拥有巨大的优势，不仅相对于AI，也相对于其他不使用AI的人。呃，在长远的未来，我不知道这种优势具体会是什么样子，但是，嗯，我们似乎天生就关心他人，而对机器却几乎毫不在意。
[37:49] 呃，不知怎的，这感觉会成为最终答案的一部分。好。那么下一个问题与总理昨天说的话也有点关联。你知道，他说世界上有两种人，一种认为AI令人恐惧，另一种则对AI能带来的未来真正感到兴奋。我想下一个问题跟工作有点关系。当然，这个话题经常被提起，这个问题来自Guru Dhanush。那么请讲。
[38:21] 先生您好，我叫Gurdanush。我是数学与计算专业BTech二年级学生。我的问题是：对于那些担心自己的工作在未来5年内会被淘汰的劳动者，您有什么想说的？他们现在应该专注于哪些技能？任何说自己对AI只有兴奋或只有恐惧的人，我认为都不够深思熟虑。我不认为存在两个阵营。我认为人是一个连续谱，取决于你是90%兴奋、10%恐惧，还是90%恐惧、10%兴奋。
[38:52] 但是……就像任何如此迅速发生的、如此大规模的变化一样。面对这么多未知、这么多希望、这么多显而易见的危险，就是这样。你必须处在这个光谱上的某个位置。你不能处在……任何一端。其中很大一部分是这对工作会产生什么影响。呃，我要用很多其他事情来给这一点加上限定条件：没有人真正知道，包括我们自己。会有很多意外。
[39:25] 有些工作会完全消失。很多工作不会像看起来那样发生那么大的变化，因为有了AI，你知道，我们还是希望由别人来做那件事。我们仍然希望由人类医生来告诉我们情况，即使我们知道AI能给出更好的建议，或者还有很多其他例子，或者，你知道，创业者仍然会创办公司，只是公司会成长得更快、做得更多，对现有工作的冲击也会更大。然后……
[39:57] 最重要的是，历史上每一次发生重大技术变革时，我们都低估了新工作出现的速度，以及它们会有多么不同。比如说，在计算机发明之前，很难想象我的工作；在工业革命之前，更是不可能想象我的工作。嗯，现在有一些非常好的工作，20年前甚至还不作为一个类别存在，即使让500年前的人……
[40:30] 来看看今天世界上的各种工作。呃……一般来说，他们会说，关于很多工作，嗯，当然包括我的工作，你知道，这绝对是份正经工作，你干得不错。你可以，你不必工作。你就像，你有充足的食物。你能获得很好的住所。这，这是他们无法想象的工作，而且看起来不像真正的工作。我希望，100年后，甚至不用500年，我们所有人看着那些工作都会觉得，那些人富得不可思议，他们拥有自己想要的一切。
[41:03] 他们在想方设法打发时间、娱乐自己。你知道，这是一种玩家式的生活方式。他们其实并不需要这些，而这就是社会发展的历史：为彼此积累更多的集体福祉，呃……我认为这很棒。但我们永远不会无事可做。我们从未达到人们几个世纪以来一直理论化的那种状态，即技术会进步到我们只是坐着无所事事。
[41:35] 我们会找到新的事情做。我们会找到表达创造力的新方式。我们会找到对彼此有用的新方式。我们会找到新的方式，你知道，获得地位或者别的什么。好。你们中的许多人最终会从事今天甚至作为一个概念都不存在的工作。这并不意味着转型不痛苦。我预计对很多人来说会非常痛苦。嗯，随着社会迅速变得更富裕，我希望我们能找到新的社会经济模式，让痛苦减轻一些。
[42:08] 但我对长远的未来没那么担心。我认为我们会在不太长的时间内找到新的事情做。那么，我们昨天宣布了与Tata集团的合作，涉及算力、数据中心等等。我想下一个问题与此有点关系，来自IIT德里的Kaustubh。请讲，Kaustubh。嗨，Sam。我叫Kaustubh。我是IIT德里AI学院二年级博士生。我想问一个问题：OpenAI正在采取哪些措施为世界构建绿色AI解决方案？
[42:39] 要构建什么解决方案？绿色AI解决方案。真的吗？绿色，绿色。绿色，绿色。我想……谢谢。世界距离实现核聚变以及真正可靠的太阳能、风能加储能，只有几年时间了。我还认为，这些将成为世界上最便宜的能源。当这一切发生时，经济会发挥它的作用，而且不只是AI，其他所有人也都会使用新型能源。
[43:11] 嗯，这一点之所以重要，原因有很多，但较少被谈到的一点是，作为其中一部分，如果我们能像推动智能的极大丰富那样，推动能源的极大丰富，我认为这将极大地提升世界各地的人们。所以，显然，我们在环境方面关心这件事，并承诺做到这一点；我们确实从一种对世界的责任出发关心这件事，要让能源变得丰富，并从这一视角出发。
[43:44] 我们要稍微换个话题。我之前说过，我们也有来自创意产业的学生。所以下一个问题与此有关，我想是来自 。如果你有麦克风，请讲。嗨，Sam，我是来自Pearl Academy的Parisha。我正在学习室内与建筑设计。所以我的问题是，AI成了能够生成图像、空间并创造完整概念的工具。
[44:30] 而不是变得过于由机器驱动。好问题。我以前更担心这一点，你知道，我并不完全确定当模型能够创造任何东西时事情会如何发展。不过，观察实践中的情况并与创作者交谈后，不是所有人，但一般来说，人们说这是他们创作过程中作为工具有史以来最好的事情，如果他们能够探索想法，让机器帮助他们搜索空间，然后在每一层选择他们想走的方向。
[45:06] 阿门。如果我们能让他们的创意构思越来越快，从想法到看到结果、再由此产生新想法的时间越来越短。而且与创作者交谈时，无论他们是视觉艺术家、做视频的人还是做音乐的人，他们都非常在意，我也一样，人类精神、人类方向和人类印记要留在其中。但他们也都认为，他们将能够创造出比以往任何时候都更好的东西。另一个乐观的理由回到之前关于人的那一点：人确实天生非常在意他人。
[45:44] 我们看到了一个我认为可以预料、但仍让一些人惊讶的结果，那就是：如果你给某人看一张图像，如果你说，我要把这件艺术品卖给你，它是AI做的，很美，但是，你知道，是AI做的。人们赋予它的价值实际上几乎为零。嗯，如果是人做的，即使使用了AI工具，而且这是你能产生联系、有故事、投入了创作精力、有想要表达的东西的人，但AI在他们的创作过程中帮助了他们，那就很有价值。
[46:20] 但如果没有那个人有效地署名其上并说，这是我的作品，我们似乎就不在意。我认为这对人类创作者未来的处境来说，是另一个非常积极的信号。我知道你今天晚些时候要和Karan Johar讨论。所以其中一些内容也会更多地出现。如你所知，他是一位非常受欢迎的电影导演。所以我知道那也会是一场进一步的对话。我有点意识到我们可能时间快到了。我最后再回答两个问题。Sam，如果你愿意的话，最后也可以。
[46:54] 我想接下来是Paridhi。Paridhi，如果你有麦克风。好，请讲。嗨，Sam。我是来自Amity University的Paridhi。我目前在我的人工智能学位第三年。我想问您的问题是，最近有很多关于使用水来冷却服务器并导致水资源短缺的担忧。那么OpenAI正在做什么，你知道，嗯，帮助人们、帮助人们消除这些担忧，所以能源方面的担忧以及转向绿色能源的必要性，我认为是非常真实的。嗯，这是大规模AI发展可能带来的一个重大问题；我认为水方面的担忧不再真实了。曾经有一段时间，数据中心使用蒸发冷却。现在人们已经不太这么做了。
[47:47] 生产半导体显然会消耗一些水，但如果你看AI数据中心，它们并不是显著的用水大户。这一直是一个非常顽固的迷因，我不完全理解为什么。有人说：“嗯，我不用ChashuVT”，因为每次你问它一个问题，它就会用掉七加仑水之类的。这完全不是真的。我们试图解释这不是真的。不知为何它一直非常顽固。但我认为对能源的关注确实有道理。再见。好。我知道我们还有很多问题，但我意识到时间快到了，所以Ocean会来找你，请上前提问。好的。嗨，Sam。
[48:27] 我叫Oceana。我是这里交通研究与伤害预防中心的博士生。所以我的问题是，您认为年轻人现在在为AI做准备时犯的最大错误是什么？天哪，我来回答这个。我希望这次演讲没有被录制。我认为听老年人的话，是年轻人犯的最大错误。我认为你的父母显然比任何人都更爱你。他们非常善意。
[48:59]，寻求关于如何过上幸福生活或如何做个好人的建议。没有哪一个人是你应该听从的。比你的父母还多吗？嗯，但是……作为对世界未来走向的预测者，我不认为你应该信任我对变化速度有良好的直觉。嗯，这……年轻人总是最擅长弄清楚这一点。而且，你们所有人将要继承的世界将会非常不同。
[49:30] 而且你必须迅速培养自己的直觉并信任它们。但我认为传统的职业建议。可能不会那么奏效了。太棒了。感谢你回答所有这些问题。让我们来一个快速的快问快答。你可以选择用一个词或一句话来回答，但这就是你需要大致遵守的。我有大约四五个问题。第一个，那么下一家万亿美元的人工智能公司会来自印度吗？我希望如此。
[50:01] 很好。关于人工智能最被高估的恐惧。再见。就像街头上的杀人机器人。最被低估的机会？谢谢。零人公司。谢谢。五年内更大的变化，教育还是医疗保健？教育。最后，每个IIT学生今年应该培养的一项技能。
[50:38] 韧性。博士。太棒了。谢谢大家。谢谢。非常感谢您与我们的学生互动，这确实对我们所有人来说都是一个非常难忘的夜晚。而且为了确保您离开这个学院后也能记住我们。我们想赠送一份小小的感谢礼物。为此，我想请维诺德·科斯拉先生回到台上，也请主任来赠送我们的礼物。
[51:53] 我想以一句话结束，既然你给了我一顶棒球帽，谢谢。我认为在未来四年内，这顶棒球帽将能读取我的思想并直接输入计算机。我们可能会成功也可能会失败，但我们在尝试。四年。谢谢。我想——我想请萨姆现在与观众一起自拍，记录这一刻。我想请学术拓展与新倡议副院长希尔皮·夏尔马教授和迪帕克·贾恩教授上台来参与这个。
[52:29] 所以，我们正在和大家一起自拍。谢谢。
查看英文原始字幕
[0:03] What an amazing day this is to be a part of the IoT Delhi community. Let's have another huge round of applause for our guest guys. Come on. They say the best way to predict the future is by inventing it. and today we have two stalwarts from different generations who have each foreseen the future for the world by building it themselves. Our first guest is a name that resonates deeply within these worlds. He's not just a global icon, but he's also our very own alumni. In the early 80s, he co-founded Sun Microsystems, a company that fundamentally revamped the structure of Silicon Valley.
[0:47] At a time when computing was closed in proprietary, he championed open systems. His vision led to workstations that powered the early internet and the creation of Java. Later, he founded Khosla Ventures. He has consistently backed ideas that aim to solve the society's largest problems. from clean energy to digital tech and as the first institutional investor in OpenAI. Ladies and gentlemen, let's have another huge round of applause for Mr. Vinod Khosla. With him.
[1:21] We have arguably the most influential person in technology right now. steering the development of the most powerful AI models in history. As the CEO of OpenAI, he has kicked off the generative AI era, changing how the world interacts with the information, creativity, and intelligence itself. A huge round of applause for Mr. Sam Oiltman. To formally welcome our distinguished guests, and set the context for this historic event, I would now like to call upon the stage the Director of IIT Delhi, Professor Rangan Banerjee.
[2:02] Good morning. So nice to see so many of you here. And when it was known that Mr. Sam Altman and Vinod Khosla are going to come here, we quickly realized that The Dogra Hall is not going to be enough. And all of you, it's actually a privilege we could have Two dogra halls here. if we wanted to fit in all those who had registered from IIT Delhi itself, Uh...
[2:32] We have the minor exams going on. I don't know how many of you have skipped the minor exams to come here. And I was told some people have finished the two hour exam. in an hour and come down here. How many of you are from outside IIT? Please raise your hands. Excellent. So we hope that all of you would also come for our student tech event, Trist. on next week. I'd like to just Take this opportunity to say that as A higher educational institute, IIT Delhi, AI provides an opportunity, but also a challenge. And we need to figure out how we train the future generation of people who are going into the workforce. Towards this, we created a committee of faculty members, students and staff to look at
[3:35] How do we integrate responsible AI and what is ethical use of AI by faculty by students And this is an ongoing conversation. We have Again, we have a... dedicated school of AI. We have a M Tech program in machine intelligence and data sciences. We have a Ph.D. program. We've also setting up This year we'll be starting the M Tech program in machine intelligence and data sciences.
[4:14] in our Abu Dhabi campus. We did a curriculum review last year and in that curriculum review, We ensure that every student of the institute has exposure to AI in terms of courses, AI and sustainability. This was one of the features of the curriculum review. I'm not going to go on about this because we want to hear Mr. Sam Altman and Mr. Vinod Khosla. I just like to say that, just give you one example.
[4:45] We, with the Ministry of Tribal Affairs, we developed a tool called Adibani, which are in four tribal languages, Santhali, Bheeli, Mundari, and others, where we are looking at AI for the community. in communities which currently do not have access to this kind of content. We are Looking at ways and we are delighted that we are entering into a partnership with OpenAI.
[5:19] We are looking at ways where our students and our faculty will engage with both companies like open AI as well as with sovereign AI models so that we can co-develop and create responsible AI which makes a difference in people's lives. Thank you. I'd like to say that we are delighted to have Mr. Sam Altman here. He's a person who's been ahead of the curve in terms of thinking on AI and generative AI.
[5:52] We are also delighted to have with us one of our own, Mr. Vinod Khosla. who has always been ahead of the curve and As a student here, I think 1971 to 1976, During that time, He set up One of the first computer laboratories in the IIT system. And so he was ahead of the curve at that time. He was also ahead of the curve when he in 2004, I think, He donated to our institute for setting up the school of IT, which is named after his parents.
[6:34] And we at IIT Delhi are now a 65-year-old institution And we are In the process of reimagining our future, we want to be ahead of the curve in education, We are looking at IIT Delhi 2035. We are looking at Changing our infrastructure, we are going vertical. And we are hoping to have Mr. Khosla's support in form of ideas.
[7:08] as we do this reimagination. So with these few words, I'd like to just say that, like all of you, I'm looking forward to this conversation. Thank you so much. Thank you. Where should we start? Hopefully somebody will keep us honest on time. Sam, maybe, first I'm always excited to see so many young people, so maybe you can talk about your observations here.
[7:46] Yeah. This has been a wonderful trip. to India for me and I have a lot more exciting stuff today, but this is the thing I've been most looking forward to. of the entire trip. I watching what's happening with young people around the world, uh especially in india though building with incredible energy with ai The new startups that are going to happen are already happening. the new products and services that are going to get created, the just total embrace of AI. And the. the excitement and the desire to bring all of this agency and energy and, you know, speed to the future.
[8:23] is awesome. I... I don't know about you, but I think this is the biggest shift that has happened in the landscape of what a young person can accomplish. In my entire life. It's totally amazing to see. and watching what people are doing is just incredible. India has its 100th anniversary of independence in 2047. And I can't imagine a larger shift by that date than anybody would imagine. So imagine 2047 and what India might be like by then. We haven't prepped our question, we just wanted to have a conversation.
[9:04] I mean, it's going to depend what all of these people do. You all will enter... Adore it. with superintelligence. And That's going to change things a lot, but that is an incredible opportunity. I have a one-year-old kid. And he is never going to know a world where he was smarter than a computer. ever. He was born just at the right time to never outrun a computer in raw intelligence.
[9:35] And, uh, you know, for him it'll be totally normal life. For us, it's a big adaptation. For you all, it's like a little strange, but you're excited about it. What people can do with GPUs now will be the way that new value gets created in the world. So I don't know what that's going to look like. The node doesn't know what that's going to look like. We got a bunch of high level guesses. But we're not going to be the ones that build that into... uh, all of these incredible new companies that people will do with that.
[10:09] Examples that we both like are education is clearly going to totally change, healthcare is clearly totally going to change, what it means to do work is going to totally change, companies, maybe a few people and a giant data center. and build incredible new kinds of value. What it means to do physical work will change as robots can automate more and more manufacturing. Certainly what it means to discover science is going to change. The growth rate of economic value in the world the increase in quality of life, I quoted someone from Open AI said to me recently that it's really stuck with me.
[10:43] is we built up a lot of instincts and institutions policies. structure to deal with a world of scarcity. But almost none of that applies well to a world of abundance. And in addition to building all of these new kinds of companies. and products and services, society is going to have to get reimagined for this world of incredible abundance. You've thought about this a lot. Yeah, I've spent a lot of time. Let's ask some practical questions. Do you think GDP rates, whatever is projected by economists today, And every country has its 15 and 25 year forecast.
[11:20] Do you think they'd double, more than double what happens? I think GDP is going to turn out to be a terrible, metric because AI is so deflationary. What we want to measure, and I think what GDP is meant to approximate is sort of quality of life. Yeah. And quality of life. Technically, it's a definition of goods and services. But it's measured in dollars, which is money will be a whole different thing. Yeah. So I think. a proper metric of what GDP is trying to approximate, growth rate will more than double for sure.
[11:53] Um... GDP measured in nominal dollars. That probably gets kind of squirrely. Actually, to ask you a weird question. interest rates in five more years. -2% or 20%? -I think by 2035 for sure, we will have a hugely deflationary economy. that the world isn't thinking about. It's interesting because Google just issued 100-year bonds. And I said, by 2035, you should own all those bonds because The world will be so deflationary that it's hard to imagine anybody pretending. And this is sort of this notion of You know, the world doesn't yet know how large the change is and they're going on as business as usual. I think one of your engineers wrote a great manifesto Can you talk about that a little bit?
[12:54] Which manifesto? People do a few of these. I'm forgetting the name. I just mentioned it earlier in the conversation. The essentially that the world is clueless about what's going on and the magnitude of the change. It'll come to me in just a second. I think even people who work in AI, feel a little bit clueless about the magnitude of the change. Codex is an example. By the way, Codex is growing in India faster than anywhere in the world, and mostly with young people. But Codex is an example of where Everyone's intuitions, even though we knew this intellectually, emotionally we were all hit by it.
[13:32] We have lots of codecs access inside of OpenAI and what it means to be an engineer, what it means to be a researcher changed so much so quickly. That is just going to happen again and again throughout the economy. Uh, Thank you. And Thank you. if you believe, which I do, that the top level driver of the economy and of life getting better and better every year is scientific discovery. And then the second most important thing is the economic engine around that that brings these new capabilities out into the world.
[14:07] And both of those are going to get automated at the same time. We're going to automate scientific progress. We're going to automate the whole economy. And you think about what that really would mean if these things could happen. 10 times faster, 100 times faster than we've been able to do it on our own. We were not... We did not evolve to have intuitions about that kind of a growth rate. and that kind of compounding. So I don't know. But If you were back here as a student, what would you do tomorrow? No, I've thought about it a lot. The way I frame the minimum level of change to imagine 15 years from now.
[14:44] So imagine 2040. I have to say, I imagine life The delta being the same between now and 2040 as it is between today in 1970. So look back at technology in 1970. That's my assumption about Delta, and I think that's too conservative a notion. - I think that's too conservative. - Yeah, so that's the magnitude of change.
[15:17] And people ask me, so what should we do? I say, learn how to learn really, really fast. Like, your area of expertise doesn't matter. Whether you're an electrical engineer or mechanical engineer is largely going to be irrelevant. In fact, I scared the director a little bit saying, "Nothing you learn will be relevant even in 2030." When the first year students graduate, You won't have learned anything that the AI doesn't know better. I know, Sam, if you agree with me or not.
[15:54] You won't know any technical field. Better. you may understand better. what people want. you may understand much better how to interact with people, motivate people, how to come up with creative ideas, how to sort of adapt quickly. create value. But yeah, any technical skill that you learn, they help you get better at. It's still worth learning. You know, like... Someone asked me recently, do I regret learning the program?
[16:25] Amen. Because clearly the AI is better programmer than you know. The and the answer is Not at all. I also don't regret learning math. You know, I never used math as an adult. Learning how to think. Learning how to learn quickly. This was all a valuable skill, but it was all of the meta skills that, matter. Um, I mean, the one the one thing that think we could say with confidence of a specific skill to learn is when i was in school learning to program was the obvious thing. Now learning to use these tools is the obvious thing.
[16:56] and the fluency that people pick up with each new generation. if you sort of are pushing as much as possible on the edge that that seems very valuable If you're a college student, what do you think matters the most? What will matter? then the technical knowledge they have doesn't matter. An electrical engineer can do a structural engineer's job quite easily. And by the way, we're building both structural engineers, chip designers, and electrical engineers today.
[17:28] That'll be... used Pervasive everywhere. So what do you think matters? I Well, I know I just said this, but I can't emphasize enough how important I think it is to stay on the forefront of the tools. I expect that to be a compounding accelerant. So building projects-- all the time just to figure out what's going to be possible. And it's going to rapidly increase every couple of months from now on. That really matters.
[17:59] Um, The class that I wish my college had offered. um, The motto of Y Combinator is make something people want. And it is, I think, for 20... One year is now. been like the best model of any investor. And I wish My school had offered a class on that. because I think that is a learnable Skill. figuring that out and that's going to be ever more important in a world where in a world where you can sort of make anything happen figuring out how to get good ideas, which like anything else deliberate practice seems to improve, That seems to matter a lot.
[18:40] Thank you. Yeah, so. From my personal view, what's most important to learn is curiosity and agency. is tell me if you agree, disagree, elaborate on that. I totally agree. I think agency is also a learnable skill, the world kind of beats it out of you, so you have to you have to like resist and continue to believe that you can sort of figure anything and make anything happen. And the more times you do it, the more, self-belief you develop, Um, many people learn this late in their careers it is super valuable to learn early like Other things?
[19:19] You just have to start. and you have to keep doing it, and you'll develop confidence in it. But the... the returns on agency. It clearly have never been higher. You know, we've talked for a long time about what one person is going to be able to do with AI. We're finally seeing it now. a single person with drive and an idea and willpower. can make incredible things happen. And. In some sense, the most important learning of my entire career has been that if you are right on the idea, no matter how unpopular it is, in fact, it's often better to be unpopular because you have No one else you're competing with doing it.
[19:55] And you stick with unwavering conviction over a long time and you just knock down obstacle after obstacle with no loss of enthusiasm between each setback. you can accomplish amazing things. This was certainly the story of OpenAI. Now it obviously looks like a great thing, but When we started, it seemed totally crazy and it was just like a bunch of us were in a room with force of will and sort of a attitude that we were going to do it no matter what. And then even the first four or five years of opening eye, it seemed like a very bad idea.
[20:26] Thank you for believing in it, by the way. And, uh, the agency that we had all developed earlier in our careers, which I wish we had developed even earlier. was critical to making it happen. Yeah, I would say you heard a couple of quotes. I think one sad Yeah, the best way to predict the future is to invent it. I go on to add to it. That was Alan Kay who said that. And I say, invent the future you want.
[20:57] Because now more than ever, that's possible. The key is agency, and it's always been available where you can make the world you want happen. The director said, "I was part of four students who started the first computer programming club at IIT Delhi, which was any programming in any IIT had never existed. We just wanted it, so we made it happen. I also, with Professor Guha here in 1970s, started the biomedical engineering program, which is the other side.
[21:30] which is I got curious as an electrical engineer about electrical phenomena in the body and convinced Professor Guha to start the first biomedical program anywhere in India with the All India Institute of Medical Sciences. I give that as example of I was 16, I was 17. You can make anything happen if you want to, and that's what agency is about. And then technology and tools make almost anything possible. Agency is what matters. Curiosity for learning and always staying at the leading edge.
[22:05] an agency to make the world happen. There's never been an opportunity to do startups. I'm very excited by this little company we invested in. They approached me a few months ago, Emergent, launched a product in July or August. Last week they called me and said they'd exceeded a $100 million ARR rate. in eight months. The fastest I've seen anybody grow, even Loveable or Replit or any of those companies, So that opportunity exists for all of you. I know if you want to do a finishing comment.
[22:41] I can't really say anything better than that. This is like a special time. Like I this is, I think, the best time ever to be starting off. careers And yeah, the opportunity is incredible. - Yeah, I'll finish by a last quote I have. Most people, especially in India, are especially averse to risk. There's the Harvard Business School. We had a case that starts with a quote from me saying, "My willingness to fail allows me to succeed so i really encourage all of you to go be risk taking there's so much opportunity to take advantage of it take advantage of tools curiosity Um, agency, risk taking, and then persistence, because the path is never simple or smooth. So you have to persist through the hard times.
[23:36] Thank you, I think there's. We are at the end of our talk. Okay. What a wonderful conversation, guys. We had Ms. Sunita Williams a month back and The idea that Learning is such an important thing for such huge personas, for such huge stalwarts. I think that is one of the most humbling things we learn here. Now, we would like to open up the floor for students to ask their own questions. For this, I would like to invite on stage Mr. Raghav Gupta, the Head of Education for OpenAI in India.
[24:10] He is currently leading partnerships with academic institutions, ensuring that AI is integrated responsibly across all dimensions. Mr. Raghav. Good morning. It's incredible seeing a packed hall. Um, Sam, I know you have a packed few days in India and many senior meetings, but to me, it seems like this is going to be the most exciting session. Would you would you agree? I do. And I'd love to ask the students as well. Are you all excited for this conversation?
[24:43] I mean, come on. Now we can do better. Are you all excited for this conversation with Sam? So we have some of India's bright young minds here and they're studying engineering, they're studying computer science, I think we have health sciences, management, design and so on. Um. I think you all sent us like 800 plus questions. We can't take all 800, but I've shortlisted the blood teams and we'll invite you to ask them live.
[25:18] I'll kick us off a little bit. Sam, you wrote this article in Times of India earlier this week, and the headline kind of captured your message very nicely, where it said that AI will define India's future. and Um, and, India will define AI's future as well. And I know at the AI summit, you've seen a lot of what's happening around development in AI. Would you like to talk a little bit about your impressions there? India is already our second largest market. There's a hundred million. ChatGPT users, a third of them are students. Fastest growing codecs market in the world. India should be our largest market over time.
[25:54] more than just the usage of products like Chachi EBT, that the building energy in India seems like it is growing faster. than anywhere else, the desire to really create new kinds of products and services, brand new companies that are all about AI. This is Yeah, it's just amazing. So it's been great to be here. Wonderful. So we're going to go to the students next and we'll open it up live. I think we have Mukund who's going to ask the first question. And Mukund is from IIT Delhi. Let's get started with Mukund then.
[26:28] Hi Sam, I'm Mukund from the Department of Mathematics, IT Delhi. I'm in my final year pursuing an integrated master's degree. So my question is, you have talked about intelligence becoming a cheap utility, but we are seeing specialized high reasoning models being gated by premium subscriptions. So how do you reconcile the concept of broad access with the economic viability of these frontier models? I think the most... amazing thing well there's many amazing things i can't say the most one of the amazing things happening is the rate at which ai is getting cheaper at a given level of intelligence um we are If you look at a hard reasoning problem, what it would have cost you to solve it using an OpenAI API, Um, a year and a half ago versus now, it's down about a thousand VEX.
[27:18] 1,000x in 18 months, that doesn't happen very often. I don't think we'll be able to sustain another thousand X over the next 18 months, but maybe we will. But there will be a significant decrease in in cost. So. Are... goal here is to relentlessly drive the cost of intelligence down as close to zero as we can um and then make that available to the world as a very low cost asset uh and for a bunch of reasons, accessibility being one of them. That's a critical part of our mission.
[27:54] there are going to be a bunch of societal disruptions around that like part of Society has maybe changed. Thank you. We talked earlier about this issue of um, Abundance versus scarcity, but society has been used to intelligence being a relatively scarce resource. Also, the ability to make things happen in the physical world, energy, robotics, whatever, if all of that changes, I don't think society yet has a consensus on how things are going to work and where the limits, the natural limits are going to be.
[28:28] I do think the cost of intelligence is going to trend down and down and in terms of empowering the entire world, that's very important. We'll go to Tejas next who's also from IIT Delhi. I know the director also spoke a little bit about safety, and I think Tejas, your question is related to that. So go ahead, please. Hi, Sam. I'm Tejas Rajmangla. I'm currently a final year student here at IIT Delhi. I'm pursuing my bachelor's in computational mechanics. My question for you is, how do we keep AI safe without blocking innovation from things like prompt injection that are happening recently?
[29:04] I think there's three things here. One, we have to figure out how to technically align AI. We have made better progress there than I thought we might have, but there's still things to do as we get to these systems that are truly much smarter than we are, how we supervise those systems, requires new ideas as we get to a world where AI is writing so much. source code and doing so much research, we can't review all of it. There's a bunch of Things that used to feel like sci-fi that are now clear alignment challenges that we have to that we have to address. We've been doing work here for a while. Other labs have been doing work here for a while.
[29:45] On the whole, alignment has looked easier than I feared, but we're clearly not done. So bucket one is technical. safety work around alignment. Two is there's a whole new kind of security infrastructure. that needs to get built. You mentioned prompt injection. I think that's a great example. OpenClaw is an amazing thing. Codex is an amazing thing. I thought that I was going to be a very disciplined person and not let them run on my computer with root access. Because I thought that was an insane thing to do.
[30:17] And that lasted about two hours. And then I was like, oh, it's really annoying to approve everything. It's so convenient. I want to just I'm just like, let it rip. And if I'm willing to do that, a lot of other people will be willing to do that, too. I should be one of the more conservative ones there. So. I think we need to develop a new kind of security architecture different from the model technical alignment to let people be able to use these agents without insane things happening. Like, it is awesome that you can run something on your computer and it can get to know everything about your life and record your screen and use the web for you and everything else.
[30:52] It will be significantly less awesome if it starts, like, randomly telling your secrets to people it shouldn't. And then third and most important, You mentioned not blocking innovation. I super agree. The kind of like... These are caricatures because everybody is a little bit in the middle, but the kind of the two paths the world is looking at right now. is either we accept Thank you. you know, an AI company or country or model having sort of effective totalitarian control over the world for no safety.
[31:24] problems at all. You have one AI to rule them all. This has been people's vision over time. The other is that we super democratize access and we put powerful AI in everyone's hands. And we trust that if you, you know, out of 100 people, 99 are going to be good. One person is not bad. The good people will stop the bad. Some things are going to go wrong in that world. People are for sure going to misuse AI. in small and probably in big ways. for a long time the world i think has shown that if people are basically most people are good and the good people stop the bad people as long as there's not a huge power imbalance, we can get to a world where we democratize AI.
[32:03] everybody lifts up And that kind of keeps the new powerful bad forces in check. This is the story of... society for a long time, it will be messy. Things will go wrong. Bye. I think empowering everyone and having a wide distribution of power is the best, fairest and safest path forward. We call this resilience. So instead of just it, There's alignment, there's security, and then there's this newer idea of resilience where I think it no longer works for OpenAI to just say we will block our models from telling people how to make a pathogen because some open source model is going to help some bad guy make a pathogen.
[32:42] And we have to have society ready to defend against that with rapid response vaccines, sort of new ways to block respiratory transmission, whatever else. Um, yeah, Yeah, that's what I would say. So, the next question is from Yash and he is not here, so I will read it out on his behalf. I read that a lot of research behind study mode in chat GPT happened in India last year. So if OpenAI were to build a product specifically for India, not adapted but designed for India first, what problem would you want it to solve?
[33:15] I would rather one of you figure out the answer to that. We... Way more innovation is going to happen outside of OpenAI than inside of OpenAI. And our strategy is to build the core team. model and the basic way to use it. ChatGPT, the API, agents like Codex and the general agent, and really empower everybody else to build it. one of my learnings from OpenAI is the most innovation in the world. comes from startups and we want to empower them.
[33:46] Thank you. Wonderful. I think we're going to shift gears a little bit to a health related question, which is, I'm going to come from our niche from Ames. And you have the microphone. Yeah, go ahead, Adnish. Hello. My name is Adnish. I'm a third year medical student at Ames, New Delhi. And my question is about chat GPT health. So we know that Chargivity Health connects to Apple Health. and medical records. but In India, people, most people have neither.
[34:18] they have mostly just a phone So my question is, how would you rethink health AI in a country like India. I love using ChachiPetit for health. I don't have it connected to anything. I type in lots of notes. Every time I get like a blood test or something, I upload it. Every time I have symptoms, I type it in. It works really well without medical records. We'd love to connect to any platforms that... emerged in India as important platforms, but one of a huge volume of chat GPT.
[34:50] is health related queries and with no connectors at all we've worked to make it very good I think you'll like the next question, which is going to come from Vijay, who's from IIT Delhi. Vijay, you have the microphone. Hi, Sam. I'm Vijay. I'm a third year student at IIT Delhi in the computer science department. So I would like to know how would IIT undergrads be able to join OpenAI directly and what are the current and future objectives of OpenAI's offers in India? We would love for you to join OpenAI.
[35:27] we would love for you to join OpenAI, either in our office in India or in the US, I think these next couple of years are going to be the most exciting years yet in the development of AI. Um, And the coolest, well, hopefully there's a lot of cool parts about being an open eye, but I think the coolest part is you get to see the future and use the tools like six months before everybody else. And the work that that enables in the development process is awesome. So we would love to hear from you. By the way, you can ask me like harder, more controversial questions.
[35:58] I won't be offended. Like, go for it. I like those. And I think you might have seen, you know, we have an office in Delhi. We've just announced offices in Bangalore and Mumbai as well. So the team is growing rapidly. So we'd love to engage in this conversation. OK, we're going to go to Sahaj next. sahai sorry um who's from amity university and you have the microphone hi sam i'm sahaj i'm from amity university i'm a third year student pursuing artificial intelligence in ai in amateur university. So my question is in today's market where 18 year olds are competing for jobs with AI, what unfair advantage do we humans have?
[36:40] A lot. watching the very best people now work with AI. is really incredible and so different than a year ago. But you watch these people that have like six monitors on their desk, three agents per monitor, like going session by session. One person can do what a team of Um, well, a lot of engineers used to take to do. And when you can keep a whole problem in one human head and manage a bunch of agents, you are able to bring things to life at a speed and quality level you just couldn't do before.
[37:15] And there is still, this may change something in the future, but there's still something that humans are so good at. that AI is so bad at about thinking through the sort of like high level strategy of a system or a product or whatever that. people who are proficient at using AI. have a huge advantage. not just over AI, but over other people who are not doing that. Uh, In the long term future, I don't know exactly what the advantage will look like will be, but Um, We seem so wired to care about other people and care so little about machines.
[37:49] uh that Somehow that feels like it's going to be part of whatever the answer is. Okay. So the next question is a little linked to what the prime minister said yesterday as well. And, you know, he said that there are two kinds of people in the world, one who think that AI is fearful, the others who are really, you know, excited about the future of what AI can bring. And I think the next question is a little bit about jobs. And, you know, that comes up often, of course, which is from Guru Dhanush. So go ahead, please.
[38:21] Hi Sir, My name is Gurdanush. I am a second year BTech student in Mathematics and Computing. My question is, What do you say to the workers? who fear that their jobs will be outdated in next 5 years and what skills should they be focusing on now? Anyone who says they're only excited or only fearful about AI is, I think, not being very thoughtful. I don't think it's two camps. I think there's a continuum of people. based on whether you're 90% excited and 10% fearful or not. 90% fearful and 10% excited.
[38:52] But... As with any... change of this magnitude happening this quickly. And with this much unknown, this much promise, this much obvious peril, it's it. You have to be somewhere on that spectrum. You can't be sort of at... at either end. A big part of that is what the impact on jobs is going to be. And, uh, I will caveat this with lots of other things by saying that no one really knows, including us. There will be many surprises.
[39:25] Some jobs will totally go away. many jobs will not change as much as it seems like they should. with AI because we have, you know, we want someone else to be doing that. We still want a human doctor to tell us things, even if we know the AI gets better advice or any many other set of examples or, you know, entrepreneurs. will still start. companies, they'll just grow faster and they'll do more and it'll be more disruptive to existing jobs. And then...
[39:57] Most of all, Every time in history that there's been a major technological shift, we have underestimated how quickly the new jobs are going to come and how different they were going to be. it'd be very hard to think about my job, for example, before computers were invented, and impossible to think about my job before, say, the Industrial Revolution. Um, There are jobs now that are incredibly good jobs that didn't exist as a category 20 years ago even if someone from 500 years ago.
[40:30] could look at the jobs of the world today. uh... Generally speaking, they would say, about many of the jobs. um certainly my job you know definitely a real job you're doing fine. You could, you don't have to be working. You're like, you've got plenty of food. You've got access to great shelter. Like this is, This is not a job they could have imagined and it wouldn't look like real work. I hope. 100 years from now not even 500 we would all look at the jobs and be like those people are impossibly rich They have everything they could ever want.
[41:03] They are making up ways to spend their time to entertain themselves It's a you know, it's a player of game style existence. They don't really need this and That is sort of the history of society. building up more collective welfare for each other, uh... And I think that's great. But we never run out of things to do. We have never gotten to the sort of like thing that people have theorized for centuries now, which is technology is going to progress so much that we just sit around and do nothing.
[41:35] We find new things to do. We find new ways to express our creativity. We find new ways to be useful to each other. We find new ways to. you know, gain status or whatever it is. Okay. Many of you will eventually do jobs that don't exist today, even really as a concept. It doesn't mean that the transition is not painful. I expect for many people it will be very painful. Um, as society gets. richer quickly i hope we find new socioeconomic models to make it less painful.
[42:08] But I'm less worried about the long term future. And I think we'll find new things to do in not very long. So we've announced our partnership with the Tata Group yesterday and It's about compute, it's about data centers and so on. I think the next question relates to that a little bit, and which is from Kaustubh from IIT Delhi. Go ahead, Kaustubh. Hi, Sam. My name is Kaustubh. I'm a second year PhD student here at the School of AI, IIT Delhi. I wanted to ask a question about what are the steps that OpenAI is taking towards building green AI solutions for the world?
[42:39] To build in what solutions? Green AI solutions. Real? Green, green. Green, green. I think... Thank you. the world is only a few years away from fusion and really reliable solar wind plus storage. And I further think that those will become the cheapest source of energy in the world. And when that happens, the economy will do its thing and not just AI. but everybody else will use new forms of energy.
[43:11] Um, the this is important for a ton of reasons, but one that gets talked about less is as part of this, if we can drive massive abundance of energy. in the same way that we can drive abundance of intelligence. I think that will do a huge amount to elevate people around the world. So, Obviously, we care about it environmentally. and are committed to do that, we really care about it from a sort of duty to the world to make energy abundant and perspective.
[43:44] We're going to go change gears a little bit. And I was saying earlier, we have students from creative industries as well. So the next question is. is around that, and I think it's from . If you have the microphone, go ahead, please. Hi Sam, I'm Parisha from Pearl Academy. I'm pursuing interior and architectural design. So my question is, like AI became the tool that can generate images, space and create entire concept.
[44:30] rather than becoming too machine driven. Great question. I used to be more worried about this and you know i i wasn't sure totally how it was going to go when models could sort of create anything Watching what's happened in practice though and talking to creators, Not everyone, but generally speaking, people say this is the best thing ever to happen as a tool in their creative process if they can explore ideas and have a machine help them search the space and then pick at each layer which direction they want to go.
[45:06] Amen. and if we can make their creative ideation look faster and faster and faster where the time from idea to then seeing in results and then coming up with a new idea from that gets shorter and shorter. and talking to creators, whether they're visual artists or people making videos or music, They all care a lot, as do I, that the human spirit and the human direction and the human imprint is in this. But they all think they'll be able to create much better stuff than ever before. Another thing for optimism back on this earlier point about people, really are very hardwired to care about other people.
[45:44] we are seeing a, I think, predictable but still surprising to some people result, which is If you show someone an image, if you say, I'll sell you this piece of art and an AI made it and it's beautiful, but, you know, an AI made it. the value that people put on that effectively rounds to zero. um if a human made it even with ai tools and it's you know someone who you can connect to and has a story and put their creative energy into it and had something they were trying to express uh but AR helped them in their creative process, it's worth a lot.
[46:20] But without the person like effectively signing their name to it and saying like, this is my thing, we seem to not care. And I think that's another very positive sign for what it's going to be like for human creators. And I know you're on a discussion with Karan Johar later today. So some of this will come up a little bit more as well. And he's a very popular movie director, as you know. So I know this is going to be a conversation further up as well. I'm a little conscious we're maybe coming up a little bit on time. I'll take two more questions. at the end as well, Sam, if you're open to that.
[46:54] I think we have Paridhi next. Paridhi, if you have the microphone. Yeah, go ahead. Hi, Sam. I'm Paridhi from Amity University. I'm currently in third year of my artificial intelligence degree. My question to you is that recently there have been a lot of concerns regarding the use of water in cooling down the servers and that it's leading to water scarcity. So what is OpenAI doing to, you know, Nuh. help people help people uh dismiss these concerns so the the energy concern and the need to move to green energy, I think it's very real. Uh, that's, something that at scale ai is developing can be a significant issue the water concern i think is no longer real uh there was a time when data centers were using evaporative cooling That is not something that people are doing much anymore.
[47:47] There is obviously some water consumption in producing semiconductors, but if you look at AI data centers, they are not significant consumers of water. This has been an incredibly sticky meme, I don't totally understand why. There's people who say, "Well, I don't use ChashuVT," because each time you ask it a question, it uses up seven gallons of water or whatever. It's totally not true. We tried to explain it's not true. It's been very sticky for whatever reason. But I think the focus on energy does make sense. Bye. All right. I know we have a lot more questions, but I'm conscious we're coming up on time, so Ocean will come to you and please go ahead and ask your question. Yeah. Hi, Sam.
[48:27] My name is Oceana. I'm a PhD scholar at the Transportation Research and Injury Prevention Center here. So my question is, what is the biggest mistake you see young people make right now when they apparently prepare for AI? Man, I'm going to answer this. I hope this talk is not recorded. I think listening to old people. is the biggest mistake young people make I think your parents obviously love you more than anyone. They're extremely well-meaning.
[48:59] for advice about how to live a happy life or how to be a good person. There is no one you should listen to. More than your parents? Um, But... for a predictor of what the world is going to be like going forward, I don't think you should trust me for having a good intuition of the rate of change. Um, It... Young people always figure this out the best. And. the world you are all inheriting is going to be very different.
[49:30] And you'll have to quickly develop your own intuitions and trust them. But I think the traditional Career advice. It's probably not going to work as well. Wonderful. Thank you for taking all those questions. Let's do a quick rapid fire. And you can choose to answer with one word or one sentence, but that's what you need to kind of keep to. And I have like four or five questions. The first one, so will the next trillion dollar AI company come from India? I hope so.
[50:01] Nice. Most overrated fear about AI. Bye. Like killer robots in the streets. Most underrated opportunity? Thank you. zero person companies. Thank you. Bigger change in five years, education or health care? Education. And last, one skill every IIT student should build this year.
[50:38] Resilience. Dr. Wonderful. Thank you all. Thank you. Thank you so much for interacting with our students and This has indeed been a very memorable evening for all of us. And in order to ensure that you also remember us once you go out of this institute. We'd like to present a small token of gratitude. For that I'd like to call Mr. Vinod Khosla sir back on stage and also the director to present a our guests.
[51:53] I want to finish, given you gave me a baseball cap, thank you, with one comment. I think in the next four years, this baseball cap will read my mind and input directly into computers. We may succeed or fail at that, but we are trying. Four years. Thank you. I like-- I'd like to request Sam now to capture this moment with the audience with a selfie. I'd like to call up on the stage the Associate Dean of Academic Outreach and New Initiatives, Professor Shilpi Sharma and Professor Deepak Jain for this.
[52:29] So, we are doing a selfie right with everybody. Thank you.
在Vidleaf工作台中打开
搜索转录文本，选择行，复制带时间戳的引文，翻译。
署名
《炉边谈话：Sam Altman×Vinod Khosla及IIT德里AMA | 20年2026月》由IIT德里（https://www.youtube.com/@iitdelhiofficial）制作，依据CC BY 3.0（https://creativecommons.org/licenses/by/3.0/）许可。源视频：https://www.youtube.com/watch?v=IiV9q73YUxI。本页为视频的文字转录稿，添加了段落分隔与时间戳；创作者与Vidleaf无关联，也不为其背书。Vidleaf添加了AI标点和AI生成的摘要；原始措辞仍可在本页查看。
您是创作者或权利持有人吗？请求更正或移除：copyright@vidleaf.app（参见关于这些页面）。
最后更新于