OpenAI CFO Deploys Custom GPTs to Run 200-Person Finance Division

OpenAI CFO部署定制GPT管理200人财务团队 AI智能体自动化处理日常审计与全球税务申报，以扩展精益财务运营，而在数十亿美元的融资中，人际信任仍然至关重要。 OpenAI CFO Sarah Friar分享了她从工程与采矿研究走向Salesforce、Square、Nextdoor以及OpenAI等公司财务领导岗位的非线性职业路径。她概述了步入AI时代的核心原则——好奇心、适应力与善意，并探讨了如何利用恐惧和肾上腺素来推动行动与提升代表性。谈及财务工作时，Friar重点介绍了OpenAI如何运作一个“AI原生”财务部门，利用定制GPT处理投资者关系、实现100%的审计覆盖率以及全球税务申报自动化。最后，她强调尽管技术飞速进步，在融资与领导力方面，人际信任、人际交往和共同价值观依然不可或缺。 OpenAI 首席财务官 Sarah Friar 分享了自己从工程转向金融的非线性职业经历，并强调在 AI 时代，好奇心、适应力与善意是职场人驾驭变革的核心原则。 (6:41) OpenAI 财务团队通过每季度举办内部黑客松推动技术落地，曾在融资期间开发了定制投资者关系 GPT，通过吸收路演材料和审计数据快速拟定尽调回复。 (14:27) 通过利用 AI 智能体自动化处理全球跨国报税流程并实现 100% 全量发票审计，OpenAI 仅以约 200 人的精干财务团队便支撑起全球业务，并将资源拓展至经济研究与定价等战略领域。 (19:27) Friar 强调“人最终投资的是人”，人际信任在融资中无法被技术替代；同时她指出自 ChatGPT 推出以来软件工程师队伍规模增长约 6%，反驳了 AI 会彻底消除专业岗位的观点。 (23:17)

OpenAI Chief Financial Officer Sarah Friar argues that artificial intelligence will not eliminate human employment, but will instead automate routine administrative burdens, transform corporate finance into an advisory discipline, and elevate the importance of human trust in high-stakes decisions. Nonlinear Careers and Navigating Technological Shifts (1:43) Drawing from her background growing up in Northern Ireland and studying materials science, Friar emphasizes that professional careers rarely follow linear trajectories. Her path traversed roles at Arthur Andersen, McKinsey & Company, Goldman Sachs, Salesforce, Square, and Nextdoor before she became CFO of OpenAI. When preparing students and young professionals for an AI-driven workforce, Friar focuses on curiosity, adaptability, and kindness rather than rigid career plans. Addressing workforce anxiety, Friar rejects extreme claims that AI will either have zero economic impact or entirely eliminate jobs. Citing OpenAI's internal economic research team, she notes that while AI agents will usurp repetitive, mundane tasks, baseline occupations will adapt and generate new industries. Friar also notes that software engineering populations expanded by roughly 6% following ChatGPT's late 2022 release, suggesting that AI tools lower barriers to technical execution rather than eradicating technical roles. Building an AI-Native Corporate Finance Organization (14:27) Within OpenAI, Friar's finance organization of approximately 200 people uses quarterly internal hackathons to foster tool experimentation and overcome operational inertia. In practice, the team has implemented AI agents across core finance workflows, including tax compliance, treasury, quote-to-cash, and investor relations. During high-stakes capital fundraising, the finance team deployed a custom GPT loaded with pitch decks, financial models, and audit files to quickly draft responses to investor due diligence inquiries while leaving final verification to human experts. To prevent model hallucination and maintain integrity, Friar stresses instituting strict system guardrails. For example, OpenAI's investor relations GPT is instructed to maintain a factual, positive tone without lapsing into aggressive sales rhetoric, and it is explicitly programmed to state when it lacks necessary information. Friar also highlights how AI shifts financial audits from checking tiny sample sets (such as 10 out of 1,000 invoices) to evaluating 100% of transactions, improving financial controls and accounting precision. Automating Administrative Tax Work and Strategic Reallocation (19:27) Global tax compliance serves as another primary example of AI automation inside OpenAI. With 90% of ChatGPT's weekly active users located outside the United States, managing shifting global filings presents a significant administrative workload. OpenAI utilizes AI agents to fetch new tax form templates directly from overseas portals, parse the PDF structures, and populate required figures automatically. This shift removes manual data-entry burdens, allowing the tax team to operate primarily in a strategic and supervisory capacity. Friar points out that reducing administrative overhead enables lean teams to scale impact and direct resources into specialized, non-traditional finance roles. At OpenAI, this includes establishing dedicated internal pricing and economic research teams, with the latter publishing open research regarding enterprise automation and workforce trends. The Irreplaceability of Human Relationships in Capital Markets (23:17) Despite aggressive internal adoption of automation, Friar maintains that core human interpersonal dynamics remain irreplaceable in finance. In high-stakes investing and capital allocation, she contends that investors ultimately back people, relying on shared values, integrity, and direct personal chemistry that machines cannot replicate. Applying this perspective to education, Friar advocates for universities to adapt assessment models to encourage tool-assisted learning and curiosity rather than trying to wall off technology. She cites collaborative academic deployments, such as OpenAI's work with the University of Oxford's Bodleian Library to transcribe 15th-century historical manuscripts, as evidence of how institutions can integrate AI productively. Friar concludes that finance teams must deliberately build structured space to test new AI capabilities, though questions remain on how educational systems and traditional corporate hierarchies will adapt assessment and governance as automated agents become ubiquitous across core business operations.

OpenAI 首席财务官 Sarah Friar 指出，人工智能正推动企业财务从繁琐的手工流程全面转向“AI 原生”运作模式，常规重复性任务将被智能体接管，从而释放人力投向经济研究与定价等战略领域。然而 Friar 与加州大学首席投资官 Jagdeep Singh Bachher 同时强调，尽管技术将审计、税务与尽调分析推向全量与实时化，人际信任与价值观仍是投融资合作中不可替代的核心要素。 构建 AI 原生财务与全量审计转型 (16:00) Friar 阐述了将 OpenAI 内部打造成“未来公司”样板的战略设想，旨在全面建立覆盖报价到现金（quote-to-cash）、订单到付款（order-to-payment）、股权管理、税务及资金调拨的 AI 原生财务体系。目前 OpenAI 仅维持约 200 人的全球财务团队，为化解员工因日常琐事缺乏学习新工具时间的痛点，团队推行了季度内部黑客松机制，为员工提供专门的时间窗口来试验新技术。在具体业务方面，全球税务合规展现出极高的自动化成效：由于 ChatGPT 90% 的周活跃用户位于美国境外（其中非洲为增速最快的大洲），面对全球各司法管辖区每年频繁微调的税务表单，OpenAI 部署了专门的 AI Agent。这些智能体能够自动抓取包括美国国税局（IRS）和英国税务局在内的官方门户表单模板，解析 PDF 版面并自动填报数据，使得内部税务团队彻底摆脱繁琐的数据录入，转向审核与战略咨询。 在审计流程方面，传统财务审计受限于人力和时间成本，长期依赖对小样本数据的抽样检查（例如在 1,000 张发票中仅抽查 10 张）。Friar 指出，在 AI 算力充足的环境下，智能体已能够对所有发票执行 100% 的全量审查。这种全面覆盖消除了传统抽样盲区，显著提升了财务控制标准与会计核算的精准度。Bachher 认为这种模式契合加州大学投资办公室“少即是多”的理念，即精干的小型团队能够借助 AI 技术实现规模化的业务影响力。 投资者关系定制智能体应用与边界 (14:27) 在实际资本运作层面，Friar 分享了 OpenAI 在高额融资周期中应用定制化 GPT（Custom GPT）支持投资者关系的案例。财务团队构建了专用于投资者关系（IR）的智能体，将商业计划书、财务预测模型以及审计材料全部接入系统。该工具能够快速草拟出详尽且高准确度的尽职调查回复，极大地缩短了周转时间，同时由人类专业人员在最后环节进行审核把关。Friar 提及赴韩国与投资人会面的经历：在现场收到两页全韩文的复杂提问清单时，她使用手机拍摄文件并传入定制 IR GPT，系统即刻生成了精准翻译和对应的回复方案。 与此同时，Friar 强调为 AI 智能体设定明确行为边界和系统记忆的重要性，以防止产生幻觉。OpenAI 的投资者关系智能体在底层指令中被严格限制为秉持客观、事实与积极基调，明令禁止进入激进推销模式。智能体还被专门配置了诚实面对知识盲区的机制，当面对未掌握的信息时需直接表明“不知道，但后续会跟进”，这也印证了 Friar 早期在财报电话会议中获得的核心职业建议，以此确保与外部投资者沟通的高诚信度。 技能重构、就业分化与新兴战略职能 (19:27) 伴随 AI 将财务人员从行政录入负担中解脱出来，OpenAI 的财务组织得以向非传统的高价值战略领域延伸。团队内部增设了专门的经济研究团队（Economic Research Team）与独立的定价团队（Pricing Team），负责深入分析宏观环境及制定商业化策略，相关研究成果亦通过公开博文对外发布。Bachher 与 Friar 讨论了技术对劳动力市场的冲击，Friar 坚决否定了“AI 毫无影响”或“AI 将摧毁所有工作”这两种二元极端论调。她援引 OpenAI 经济研究团队的成果说明，常规且重复的机械性劳动确实应当且正在被智能体接管，但现有职业也会相应转型，同时将催生全新的产业形态。 针对专业人才被取代的担忧，Friar 以工程领域为例指出，自 2022 年底 ChatGPT 发布以来，软件工程从业人员数量实际上增长了约 6%。AI 降低了编程的技术门槛，使早年掌握 COBOL 等陈旧语言的非现役技术人员也能重新高效编写代码；医学领域的放射科医生同样未被技术淘汰。因此，技术的扩散更多是赋能个体扩大产出，而非消灭整个专业工种。 人机协作中的信任基石与教育评价重塑 (23:17) 在探讨技术局限时，Bachher 引用了加州大学投资办公室“人机结合”的文化准则，询问融资中何种要素无法被机器替代。Friar 明确主张“归根结底是人在投资人”（people invest in people），技术无法动摇基于共同价值观、人际化学反应与深层信任的合作基础。Bachher 回顾了造访 Friar 简朴站立式工位并赠送价值观专著的经历，并提及 OpenAI 获得的 1,220 亿美元融资轮次。面对科技变革引发的焦虑，Friar 建议将恐惧转化为如同急诊注射肾上腺素般的行动力量，尤其是女性领导者在科技舞台上应勇于展现影响力，践行榜样作用。 面对高校和年轻一代，Friar 提炼出“好奇心、适应力与善良”三大核心素养，主张以探索未知取代技术恐惧，如在使用 Codex 等智能体报错时直接指令其自我修复，并坚信不求回报的善意具有职业复利。针对高等教育领域的 AI 应用，Friar 介绍了 OpenAI 在 Lia Balsky 推动下开展的 GenAI Next Program，并提及与牛津大学博德利图书馆合作利用 AI 转录 15 世纪历史手稿的项目。在学术诚信议题上，她援引佐治亚理工学院负责人的观点建议，大学不应将精力耗费在防范学生利用 AI 撰写申请文书上，而应主动改革考核框架，转向鼓励学生利用工具进行自主探索式学习。 对话表明，AI 驱动的财务转型重点在于重塑人机协作流程，将财务人员提升为高阶战略与风控决策者。然而，讨论未深入探讨在全量自动化审计与财税申报中若发生系统性漏洞时的权责归属标准，也未涉及中小型机构在缺乏充足自研资源时复刻此类 AI 原生架构的技术门槛与部署成本。

## Background 本期内容由 OpenAI 官方 YouTube 频道于 2026 年 7 月 3 日发布。视频记录了 OpenAI 首席财务官 Sarah Friar 与加州大学首席投资官 Jagdeep Singh Bachher 在 OpenAI Forum 上的对谈，深入探讨了 AI 对财务体系的深刻重塑，以及面对快速变革的职场环境，年轻人所需的关键素养与人际信任的持久价值。 ## TL;DR 对话围绕 AI 时代的职业发展与财务革新展开。Sarah Friar 结合自己从材料工程走向顶级科技高管的非线性经历，提出应对技术洪流的三大支柱——好奇心、适应力与善良。在财务实践方面，OpenAI 约 200 人的精简团队借助自定义 GPT 与智能体，实现了审计从样本抽检向全量核查的飞跃，并将繁琐的全球税务填表全面自动化，从而释放出精力统领宏观经济与定价研究。针对技术替代的普遍恐慌，两人强调 AI 实际上降低了技术门槛并重构了分工，而在高风险融资与机构决策中，算法永远无法替代面对面的人际连接与诚信。 ## Atmosphere & style 整场对谈兼具思想深度与人文温度，氛围坦诚、轻松且富有亲和力。两位资深掌舵人没有摆出居高临下的布道姿态，而是像在家庭餐桌旁娓娓道来，巧妙穿插着童年拆吸尘器多出螺丝钉、顶尖独角兽 CFO 只能用升降工位等幽默自嘲。从探讨技术焦虑、肾上腺素转化，再到对青年学子的真挚期许，对话洋溢着对人际信任与善意的坚守，带给听众一种既务实清醒又充满力量的从容感。 ## Bulletin - [21:34](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1294s) OpenAI 仅以约 200 人的精简财务团队支撑全球十亿级用户体量，并将 AI 释放的行政人力重组为由 CFO 直辖的经济研究与定价策略团队。 - [17:41](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1061s) OpenAI 财务正全面转向 AI 原生运营：审计从传统的抽样检查跨越至 100% 全量核验，全球税表申报亦实现由 AI 解析版式并全自动预填数据。 - [29:00](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1740s) 针对 AI 替代职业的恐慌，数据显示自 2022 年底 ChatGPT 问世以来全球软件工程师总数逆势增长约 6%，技术演进降低了编程门槛而非消灭职业。 - [19:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1167s) ChatGPT 仅有 10% 的周活跃用户位于本土美国，90% 分布在海外，其中非洲是目前增速最快的大洲。 ## Key takeaways - [01:43](https://www.youtube.com/watch?v=76ABuBGX4IY&t=103s) Sarah Friar 的职业生涯呈非线性演进，从早年在安达信的财务轮岗、麦肯锡的采矿业分析，到 Salesforce 与 Square 的实战积累，核心在于沿途沉淀底层的通用基本原语（primitives）。 - [05:30](https://www.youtube.com/watch?v=76ABuBGX4IY&t=330s) 职业规划应摆脱假想中的外部评价与虚荣头衔，认识到“每个人都在看自己的电影，没人看你的戏”，扎实深入业务底层的实操才能带来本质成长。 - [06:42](https://www.youtube.com/watch?v=76ABuBGX4IY&t=402s) 新一代步入 AI 时代的年轻人需具备三大核心特质：以好奇心打破对未知的恐惧、以适应力应对高频迭代，以及在人际协作中坚守不求即时回报的善良。 - [06:42](https://www.youtube.com/watch?v=76ABuBGX4IY&t=402s) 人机交互范式已从静态单向操作转向具备双向反馈与自愈能力的智能体交互，用户甚至可以通过自然语言命令系统自我修复报错与配置。 - [09:31](https://www.youtube.com/watch?v=76ABuBGX4IY&t=571s) 面对技术冲击不应陷入非黑即白的二元论：枯燥重复的底层岗位理应被智能体自动化，多数专业岗位将发生深层变革，同时也会催生前所未见的新职业。 - [12:12](https://www.youtube.com/watch?v=76ABuBGX4IY&t=732s) 将面对高压与未知的生理恐惧和肾上腺素激增，转化为高昂的专注力与代表多元化使命感的大局动力。 - [14:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=867s) OpenAI 财务团队通过季度黑客松推动内部工具创新，打造了遵循“客观不夸大”与“坦诚不知即不幻觉”原则的投资者关系专属 GPT，并成功应用于跨语言投资沟通实战。 - [17:41](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1061s) AI 让财务审计从人力匮乏时代不得已采用的“样本抽检”逻辑，根本性跨越为算力充沛时代对发票数据的“100% 全量逐笔核查”，彻底改变内部风控底座。 - [19:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1167s) 税务团队借助大模型自动识别各国官网动态变化的 PDF 税表并自动预填，使财务专业人才彻底脱离低效填表，转向业务战略与数据真实性把控。 - [23:18](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1398s) “归根结底，人投资的是人”：无论算法如何先进，融资与合作决策的根本基石依然建立在面对面的信任、正直与价值观共鸣之上。 - [25:46](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1546s) 高等教育机构应鼓励全面实验并推行课堂 AI 原生化，考核体系应从被动的防范作弊转向考察深层思考力与先进工具协同能力。 - [25:46](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1546s) 学生应善用 AI 保持学习与解题的心流，但日常工具辅助绝不能替代对底层原理的真正认知与独立思考训练。 - [29:00](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1740s) 正如 AI 未能取代放射科医生一样，生成式 AI 大幅降低了代码编写门槛，让更广泛的人群重获编程创造力，全球开发者生态反而进一步扩张。 ## Detailed walkthrough ### [00:09](https://www.youtube.com/watch?v=76ABuBGX4IY&t=9s) 论坛开场与对话背景：AI 时代的财务变革 OpenAI 编辑总监 Mark Murray 为本期论坛拉开帷幕，隆重介绍了两位重量级对话嘉宾：OpenAI 首席财务官（CFO）Sarah Friar，以及加州大学（University of California）首席投资官（CIO）Jagdeep Singh Bachher。当前正值人工智能深刻重塑金融行业的关键历史节点，本次对话旨在探讨 AI 如何改变财务日常工作、技术能辅助完成哪些复杂任务，以及对于新一代从业者而言哪些核心素养正变得愈发不可或缺。 现场还特别向正在收看直播的加州大学学生们致意，欢迎这些对金融与前沿科技充满热情的未来人才加入讨论。随后，Jagdeep 接过主持接力棒，他感慨加州大学的学生们正是思考如何与 AI 共存未来的核心群体，而 Sarah 无疑是探讨 AI 与金融交汇点最理想的对话对象。在切入具体技术红利与机遇之前，Jagdeep 将第一个问题抛向了 Sarah 独特的个人成长史：她究竟是如何对财务产生兴趣，并一步步走向这一全球顶尖科技企业财务掌舵人位置的。 --- ### [01:43](https://www.youtube.com/watch?v=76ABuBGX4IY&t=103s) 从北爱尔兰拆吸尘器到硅谷高管：曲折多元的职业进阶之路 Sarah 回顾了自己充满戏剧性与“之字形”（zigzaggy）前行的职业轨迹。她成长于动荡时期的北爱尔兰（The Troubles），虽然外部环境复杂，但童年的她天生充满好奇心。她曾把母亲的胡佛牌（Hoover）吸尘器大卸八块，虽然最后重装时总会多出一颗螺丝钉、怎么也找不到原位，但她却自信地得出结论——这机器设计得“太冗余了”，少颗螺丝照样能转。这段经历一度让她坚信自己的宿命是当一名工程师。然而，为了筹措大学学费，她在 18 至 19 岁那年选择去安达信（Arthur Andersen）做见习会计，自此一脚踏入财务世界。安达信随后资助了她的大学学业，作为回报，她每个假期都会回到公司，轮流在税务、咨询甚至是破产清算等各个部门轮岗实战，打下了扎实的财务底层功底。 即便手握材料科学硕士的光鲜文凭，Sarah 的职业道路依然没有画成直线。毕业后她加入麦肯锡（McKinsey），并因一篇关于“如何从硫化矿中提取黄金”的硕士毕业论文，成为了当时唯一被派往南非采矿业前线的年轻商业分析师（BA）。这段经历让她得以俯瞰真实商业世界的底层运转逻辑。随后，麦肯锡资助她攻读商学院，将她带到了西海岸。在面对加州大学的听众时，Sarah 幽默地打趣自己不敢说出母校名字，怕引来伯克利学生的起哄——原来她去的是斯坦福（Stanford）。商学院毕业后，她入职高盛（Goldman Sachs）深耕股票研究，并在那里确立了自己对企业实际运营的热爱。 当她自认为羽翼丰满、信心满满地准备直接出任 CFO 时，一位资深猎头在午餐会上面无表情地给她泼了一盆冷水：“你根本没资格当 CFO，你甚至连 CFO 是干什么的都不知道。”虽然当头一棒，但 Sarah 将其视为人生中最珍贵的棒喝时刻。猎头建议她先去向顶尖的行业领袖学艺。于是，她加入 Salesforce，师从当时的明星 CFO Graham Smith。正是这段扎实的实操历练，最终为她后来出任 Square CFO、Nextdoor 首席执行官（CEO）以及执掌 OpenAI 财务大局奠定了不可动摇的基石。 ```mermaid flowchart TD accTitle: Sarah Friar 职业演进路径 accDescr: 展示 Sarah Friar 从早期教育到出任 OpenAI CFO 的关键阶段与转折点 subgraph Phase1["基础探索与工科背景"] A["北爱尔兰成长<br/>机械拆解好奇心"] --> B["安达信实习与资助<br/>轮岗税务/清算"] B --> C["材料科学硕士<br/>硫化矿提金论文"] end subgraph Phase2["战略与资本市场磨砺"] C --> D["麦肯锡商业分析师<br/>派驻南非采矿业"] D --> E["斯坦福商学院 MBA"] E --> F["高盛股票研究<br/>明确运营志向"] end subgraph Phase3["运营实战与高管进阶"] F --> G["Salesforce 财务实战<br/>师从 Graham Smith"] G --> H["Square CFO"] H --> I["Nextdoor CEO"] I --> J["OpenAI CFO"] end ``` - **求学自筹机制**：18-19 岁通过在安达信全职工作一年筹集学费，并获得后续大学全额赞助与寒暑假跨部门轮岗机会。 - **学术专业背景**：拥有材料科学与工程硕士学位，硕士论文专门研究从硫化矿中提炼黄金的冶金工艺。 - **核心职业忠告**：职业生涯往往是非线性的，关键在于沿途沉淀了哪些“基本原语”（primitives）与基础技能，将自己擅长的、热爱的以及能维持生计的能力三者合一。 --- ### [05:30](https://www.youtube.com/watch?v=76ABuBGX4IY&t=330s) 撕掉头衔的虚荣：别为别人的观众席表演，深入企业实操 针对 Jagdeep 关于“成为 CFO 是既定目标还是猎头激将法”的提问，Sarah 坦然承认，从高盛离开时，周围所有人都对她说“你做过股票研究员，顺理成章应该去当 CFO”。当时年轻的她一度被这个头衔所俘获，觉得挂上 CFO 的名号听上去体面又风光，能向外界证明自己正在做一番了不起的大事。 然而她随后意识到，这种被他人眼光绑架的虚荣是极其幼稚的。Sarah 分享了一段发人深省的心得：“我们每个人都是自己电影里的主角，而其他所有人都在忙着看他们自己的电影，根本没人在看你的戏。”因此，永远不要为了假想中的外界评价去做职业抉择。此外，猎头当年的当头棒喝十分中肯——从外部审视一个职位（如发布财报、对外发声等光鲜表象），与深入企业内部参与脏活累活的底层运作完全是两码事。真正看清优秀管理者如何在业务肌理中穿针引线，才是成长不可替代的捷径。 --- ### [06:42](https://www.youtube.com/watch?v=76ABuBGX4IY&t=402s) 餐桌上的代际对话：拥抱 AI 的三大支柱与“让它自我修复”的交互变革 回到家庭生活，Sarah 透露自己的女儿目前上大三、儿子刚上大一，正值同现场学生完全相同的年龄阶段，家庭餐桌上的日常话题时刻围绕着 AI、学业与未来职业展开。在谈到年轻人如何面对 AI 时代的浪潮时，她提炼了三个核心关键词：**好奇心（Curiosity）**、**适应力（Adaptability）** 和看似出人意料的 **善良（Kindness）**。 在深入阐释“好奇心”时，Sarah 指出恐惧往往源于未知，唯有主动跳入水中尝试才能破除壁垒。即便作为 OpenAI 的财务掌门人，她也常常面临技术飞速迭代带来的本能恐慌——公司几乎每天都在发布新功能，当同事冷不丁询问她“有没有试过某个新模型”或“是否在 Codex 里加载了某种新技能（skill）”时，她也会瞬间心跳加速，害怕自己落伍。 她分享了一个极具代表性的生动趣事：在她刚开始使用 OpenAI 旗下具备智能体特性的开发平台 Codex 时，尝试下载配置某项功能却报错卡壳。她沮丧地走进办公室准备找技术团队求援：“我试了半天根本跑不通。”结果她的执行副总裁兼助理反问她：“你有没有直接要求它把自己修好？”Sarah 当时十分诧异，半信半疑地输入了一句“请修复你自己”（Please fix yourself），不可思议的是，系统就像施了魔法一样自动修复了报错并正常运转起来。 Sarah 强调，这种人机交互范式的转变是革命性的：过去人们遇到故障，就像她小时候拆解吸尘器一样，死物绝不会主动开口告诉你多余的螺丝该放哪；而现在的 AI 技术却能协同人类、反哺人类，甚至指导使用者去驾驭它自己并实现自我诊断修复。随后她对比了自己的童年：当年在信息匮乏的北爱尔兰，推销员上门卖给父母一套《大英百科全书》（Encyclopaedia Britannica），就成了她童年里最快乐的大事；而如今的青年学子指尖所连接的，是远超百科全书千万倍的知识海洋与智能工具，懂得利用好奇心去驾驭这种馈赠，将是这个时代青年最强大的竞争壁垒。 - **育儿生活写照**：育有一女（大三，21 岁）一子（大一），日常关于职场与 AI 冲击的讨论正是家庭对话的核心。 - **应对时代的三大核心特质**：好奇心（勇于尝试并击碎对未知的恐惧）、适应力（拥抱高频迭代）、善良（维系人际信任与长远协作）。 - **人机交互新范式**：技术工具从传统的“静态单向操作”跨越至“双向反馈与自我修正”，用户可以通过自然语言直接命令智能体完成报错诊断与配置自愈。 ### [09:31](https://www.youtube.com/watch?v=76ABuBGX4IY&t=571s) 职场适应力与善良的力量：从家庭变迁看 AI 时代的职业演进 在探讨年轻人步入职场应具备的能力时，Sarah Friar 强调了第二大特质——“适应力”（Adaptability）。她分享道，如今试图给自己的孩子提供职业建议，就像当年父母试图指导她一样无力。她的母亲在农场长大，后来成为一名护士；外祖父则是电工，曾前往北爱尔兰拉设电线，两人都在 16 岁时便离开了学校。Jagdeep Singh Bachher 随之打趣道：“你们现在正好非常需要电线和电力。”Friar 笑着赞同，并借此打了个生动的比方：想象一下当一户人家的电灯第一次被点亮时的震撼，这与如今的 AI 变革有着极强的类比性。父母虽然是令人无比感激的社区楷模，但在职业上除了能理解“医生”这个传统职业之外，对其余现代行业一无所知。如今身为家长的我们面对急速变化的 AI 时代也是如此。 关于 AI 对就业的影响，Friar 直言自己绝非“非黑即白”的二元论者：她既不认同“AI 毫无影响、一切照旧”的盲目乐观，也不同意“AI 会抢走人类所有工作”的极度悲观。OpenAI 的经济研究团队所做出的研究表明，未来就业结构将呈现多层分化： 1. 部分高度日常化、重复枯燥（Routine and mundane）的底层岗位确实会被智能体（Agents）取代，且这类工作“本就应该被自动化”； 2. 绝大部分岗位将发生翻天覆地的转变，包括她自己的 CFO 工作，在引入智能体工作流程短短一年后就已出现“天翻地覆”（Night and day）的变化； 3. 将涌现出大量人类目前甚至无法想象的全新职业，正如当年的祖辈和母辈无法想象今天的科技金融高管角色一样。 最后，Friar 着重强调了第三个关键词——“善良”（Kindness）。她真诚地告诫在大学里即将步入职场的年轻学子们，切莫低估单纯行善的力量。人类社会是一个反复互动的系统，在不求即时回报的前提下去主动帮助他人、展现同理心，这些善意未来都会成倍、成桶地（in bucket loads）回馈到自己身上。 --- ### [12:12](https://www.youtube.com/watch?v=76ABuBGX4IY&t=732s) 驾驭恐惧与肾上腺素：将紧张转化为舞台力量 面对 Jagdeep Singh Bachher 提出的“如何克服对拥抱 AI 的恐惧”这一疑问，Friar 分享了自己的独特法则：拥抱并利用“肾上腺素”（Adrenaline rush）。她提到自己的哥哥是一名急诊室医生（Bachher 随即附和“我的妻子也是急诊室医生”，两人笑称急诊医生简直是“特殊材料制成的”）。哥哥曾告诉她，当病患生命垂危时，医生会直接给病人注射一剂肾上腺素，通过这种剧烈的刺激把系统重新“电击唤醒”。因此，每当 Friar 在生活或工作中感到害怕时，她会立刻进行心态转换：第一，告诉自己“感受这股肾上腺素吧，我正鲜活地活着”；第二，思考如何驾驭这种情绪并将其转化为正向动力。 ```mermaid flowchart TD accTitle: 恐惧与肾上腺素转化机制 accDescr: 展示 Sarah Friar 如何将心理恐惧转化为舞台呈现动力的思维链路 subgraph FearTrigger ["恐惧触发阶段"] A["高风险任务 / 关键舞台演讲"] --> B["生理恐惧与肾上腺素飙升"] end subgraph MentalReframing ["认知重塑阶段"] B --> C["自我觉察：感受心跳与生命力"] B --> D["抽离自我：着眼于更大格局 (Greater Good)"] end subgraph ActionOutput ["行动输出阶段"] C --> E["转化为高昂专注力"] D --> F["展现科技女性代表性 (Diversity)"] E --> G["出色完成演讲与推介"] F --> G end ``` Friar 举例说明，就在前一天上午的一场高规格公开演讲前，她感到压力巨大并伴随着强烈的肾上腺素激增。在那一刻，她没有陷在自我焦虑中，而是把思维抽离出来，着眼于更广阔的意义：“目前科技界舞台上的女性代表实在太少了。俗话说‘看不见就无法成为’（You cannot be what you cannot see）。我要带着这份恐惧和肾上腺素走上台，向大家展示科技领域的多元性。”这种为了大局而将恐惧转化为使命感的策略，正是她克服心理阻碍的核心法宝。 --- ### [14:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=867s) OpenAI 财务团队的 AI 原生演进：自定义 GPT 在投资者关系中的实战 回到金融与财务的核心议题，Friar 指出，OpenAI 财务团队的愿景是“从自身做起，成为未来企业的典范”。两年前她刚加入时，团队从 ChatGPT 和自定义 GPT（Custom GPTs）起步。针对员工因业务繁重而抱怨“没时间学习新工具”的痛点，Friar 在团队内部推行了季度黑客松（Hackathons），通过“先慢下来，才能走得更快”（Slow down to go fast）的方式，为员工提供专门探索工具的时间与空间。 其中一个具有里程碑意义的成果，是在一轮重大融资期间团队打造的“投资者关系专属 GPT”（IR GPT）。团队将公司路演 PPT、尽职调查档案等全套资料输入给模型，让其快速掌握所有背景信息。在设定提示词与核心记忆边界时，Friar 明确立下了两条铁律： - **基调平衡**：要客观求真、适度具备吸引力，但绝不能过度推销（"a little salesy but not too salesy"），确保面对投资人时的诚信与精准度； - **杜绝幻觉与坦诚不知**：如果缺少对应数据，模型必须明确回答“不知道”，这契合了 Friar 在财报电话会上学到的最重要经验——“不知道就明确说不知道，随后跟进补齐”。 在韩国的一次投资人会议上，该工具展现了惊人的实战价值。对方一见面便递给她两页密密麻麻的韩语问题清单。Friar 当场拿出手机，利用 IR GPT 拍照上传，模型立即完成了精准翻译并给出了详尽答案。尽管当时的正式沟通仍依赖同声传译，但这种技术已让人清晰看到：利用双向语音交互直接与全球投资者无缝对话的未来，早已近在咫尺。 --- ### [17:41](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1061s) 从“样本抽检”到“全量核查”：AI 赋予财务审计的全面颠覆 Friar 进一步阐述，OpenAI 目前正在全面转向“AI 原生”（AI-native）的财务运营模式，深入重塑包括从报价到收款（Quote-to-cash）、订单到付款（Order-to-payment）、股权管理（Equity）、税务（Tax）以及资金管理（Treasury）在内的所有底层流程。每一项细分职能都在挖掘自身业务痛点，思考如何利用大模型技术提升洞察力、准确率和执行效率。 在审计领域，Friar 提出了一个根本性的范式转移： - **匮乏世界（World of scarcity）下的传统审计**：受限于人力与时间成本，审计师在核查成百上千份发票时，通常只能依赖“抽样检查”（Sampling）。如果 1000 份发票中抽查的 10 份均无异常，审计师便基于统计推论假定其余 990 份发票同样合规。这种控制机制本质上是人力短缺之下的权衡妥协。 - **充沛世界（World of abundance）下的 AI 审计**：由具备自动化能力的智能体介入后，系统可以毫无压力地完成 1000 份发票的 100% 全量逐笔核验。这不仅彻底改变了企业内部风控机制的底层逻辑，而且让财务控制的确定性与数据精度达到了前所未有的高度。 ### [19:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1167s) OpenAI 税务团队的自动化变革：从“手动填表”到“审核与战略思考” OpenAI 内部最深度拥抱 AI（AI-pilled）的团队，出人意料地竟是通常被认为最保守的税务团队。由于 OpenAI 业务极其全球化——目前 ChatGPT 每周活跃用户（WAU）中只有 10% 位于美国，高达 90% 分布在海外，且增速最快的大洲是非洲——团队每年年底都要面对向全球各个国家和地区税务机关提交海量税表的艰巨任务。 在过去，这项工作异常繁重：员工必须前往美国国税局（IRS）或英国皇家税务与海关总署（HMRC）等各国官网下载税表。更棘手的是，这些表格每年都会悄悄发生格式微调，比如在某个角落新增一个勾选框。如今，OpenAI 通过自家 AI 技术彻底重塑了这一流程：AI 能够自动解析下载的 PDF 表格布局与结构，结合税务人员事先核算好的内部底层财务数据，全自动完成所有表格的填写。 这一变革让税务团队彻底从机械低效的“在 PDF 格子里填数据”中解脱出来，进入纯粹的“核对与复核”模式，未来甚至有望实现全自动申报。团队得以将宝贵时间投入到真正需要专业智慧的核心问题上：数字是否精准？该方案是否最契合公司业务战略？是否有更新颖的呈报方式？正如 Sarah Friar 所强调的，高校培养财务人才绝不是为了让他们毕业后去手动填写 PDF 表格。 ```mermaid flowchart TD accTitle: OpenAI 税务自动化流程图 accDescr: 展示从官方下载 PDF 表格、AI 识别解析，再到结合财务数据自动填表与人工审核的流程。 subgraph Download["1. 表格获取与解析"] A["各国税务官网 (IRS / HMRC 等)"] --> B["获取最新 PDF 税表"] B --> C["AI 理解与解析表格版式"] end subgraph Automation["2. 智能填充与计算"] D["内部核心财务与税务核算数据"] --> E["AI 自动映射并填入对应单元格"] C --> E end subgraph Review["3. 人工复核与战略决策"] E --> F["完全自动预填的全球税表"] F --> G["税务团队转向复核与战略把控"] G --> H["未来愿景: 全自动一键申报"] end ``` - **全球用户分布格局**：ChatGPT 仅有 10% 的周活跃用户在本土美国，90% 用户位于海外，非洲是当前增长速度最快的地区。 - **流程质变**：以前需人工紧盯每年表格细微变动逐格填写；现在全球所有税表全部实现自动预填，税务专家专注于业务战略与数据真实性。 --- ### [21:34](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1294s) 精简团队释放深度洞察：OpenAI 财务部的独特组织架构 加州大学首席投资官（CIO）Jagdeep Singh Bachher 提到了 UC Investments 投资哲学的第一条准则——“少即是多”（Less is more），并询问面对全球超 10 亿用户的庞大体量，OpenAI 的财务团队究竟有多少人。Sarah Friar 透露，OpenAI 的整个财务团队大约仅有 200 人，规模极其精简。 然而，在 AI 工具大幅压缩了基础合规与繁杂行政事务后，Friar 得以打破常规 CFO 的组织架构边界，在能够输出高价值深度洞察的领域设立了编制更大、更特殊的团队。例如，OpenAI 财务部内部直接下设了“经济研究团队”（Economic Research Team）和“定价策略团队”（Pricing Team）。在传统企业中，CFO 通常被日常核算与杂务压得喘不过气，根本无力统管这类职能。 Friar 结合了早年在高盛（Goldman Sachs）从事证券研究的深厚积累，依托 OpenAI 独一无二的底层宏观数据，将全球社会对 AI 发展的极度好奇转化为极具洞见的研究报告。这些研究不仅能赋能内部投资人关系与商业客户沟通，也向公众公开分享。Friar 甚至幽默地向观众推荐：“大家若有空，可以打发自己的 AI Agent 去检索拜读一下 OpenAI 经济研究团队的官方博客文章。” - **团队规模**：支撑全球级业务的 OpenAI 财务团队仅约 200 人。 - **差异化组织架构**：CFO 旗下直接统领经济研究团队与定价团队，将 AI 释放的生产力转化为深度市场洞察与研究能力。 --- ### [23:18](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1398s) 融资中永远不变的核心法则：“归根结底，人投资的是人” 围绕 UC Investments 投资框架中的第九条原则“人与机器结合”（Human meets machine），Bachher 询问 Friar：在 AI 工具几乎重构了所有融资与路演分析流程的当下，究竟有什么东西是自始至终未曾改变的？ Friar 坚定地表示，她在融资领域的座右铭永远是：“归根结底，人投资的是人”（In the end people invest in people）。无论算法与工具多么先进，人与人之间的信任（Trust）与正直诚信（Integrity）始终是不可动摇的基石。她回忆起 Bachher 当初前往 OpenAI 拜访的场景，直言双方关系的突破契机就是面对面坐在同一个房间里交流。 当时 Bachher 带来了一本凝聚加州大学投资理念的“价值观之书”，Friar 至今仍将其摆放在自己的办公桌上。她笑着自嘲道，自己其实根本没有独立的豪华办公室，甚至作为顶级独角兽的 CFO 也只是分配到了一张升降工位（Stand-up desk），打趣学生们这就是大家奋斗成为 CFO 后的真实写照。 Friar 深刻反思道，融资的本质绝非仅仅为了拿钱（尽管资金是维持运转的血液），更关键在于选择与什么样的人为伍。优秀的合作伙伴带来的远超资本： - 他们能带来卓越的见识与启发，就像那本价值观手册带来的思想激荡； - 他们主动提供切实的生态资源，例如探讨如何与加州大学等大规模学术科研机构对接； - 他们提议通过轻松创新的播客对话形式增进连接（正如现场正在录制的内容）。 这也是为什么 Friar 始终将“善意”（Kindness）列为三大核心素养之一。在商业与人际交往中善待他人，即便当下并未达成投资合作，但在长达 5 年、10 年乃至一生的漫长职业长河中，彼此很可能会以意想不到的方式再度交汇相遇。 --- ### [25:46](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1546s) 高校与年轻人的 AI 生存之道：全面实验、重构教学与培养思考力 针对大学应如何培养下一代人才管道（Pipeline）与未来 OpenAI 员工的问题，Friar 结合 OpenAI 与全球顶尖高校的合作实践给出了三项核心建议： 1. **全面部署与大胆实验（Deploy, be curious, and learn）**： OpenAI 设有由 Lia Balsky 带领的杰出教育团队（EDU team），并推展了面向高校的“GenAI Next Program”。例如在 Friar 的本科母校牛津大学（Oxford），OpenAI 与拥有悠久历史的博德利图书馆（Bodleian Library）展开深度合作，探索利用大模型解析 15 世纪古籍手稿的全新可能。Friar 建议高校管理层效仿前沿企业，“先让一千朵花盛开”（Let a thousand flowers bloom），在过早制定严密管制前，先让师生自由探索、测试边界。 2. **积极将课堂“AI 原生化”（AI-ify your classes）并重构考核体系**： 教育界最初普遍担忧学生会使用 AI 进行作弊。Friar 分享了她与佐治亚理工学院（Georgia Tech）校长的经典对话。当时招生主管惊慌失措地冲进校长室喊道：“天哪，学生们要在入学申请文书上用 AI 作弊了！”校长却从容回应：“你在说什么呢？他们以前本来就在作弊。”因为家境优渥的学生本就可以花钱请顶尖升学顾问润色，或者由邻里高人代笔。因此，教育者的职责不再是四处抓“作弊”，而是设计出真正能考察学生真实特质与深层思考的试题；如果学生懂得善用 ChatGPT 辅助表达，这恰恰证明了他们的探索好奇心与先进工具运用能力。 3. **借助工具保持心流，但不可放弃核心思考力**： Friar 以自己正在读大三、主修化学的女儿为例：女儿在写作业卡壳时，不再被动干等助教周一早上的答疑时间，而是实时用 ChatGPT 理清思路，让自己的学习“心流”（Flow）不被打断。然而，女儿清醒地意识到，期末闭卷考无法使用 AI，日常的工具辅助必须沉淀为真正的个人认知掌握度。AI 是思考的倍增器，但学生绝不能在学习过程中放弃“学会如何思考”的本质。 ### [29:00](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1740s) 破除被替代的恐慌：从放射科医生到软件工程师的 AI 新生 谈及技术对职业的冲击，人们总会想起那个著名的预测——当年各界曾言之凿凿地预言 AI 将让全世界的放射科医生彻底失业，然而现实中如今的放射科医生数量反而比以往任何时候都要多。软件工程领域正在重演同样的规律。尽管外界不断充斥着“程序员即将消亡”的悲观论调，但数据给出了相反的答案：自 2022 年底 ChatGPT 问世以来，全球软件工程师的整体从业人数实际上增长了约 6%。技术的突破并没有消灭职业，而是显著降低了技术门槛，使得更庞大的人群获得了开发与编写代码的能力。 对于非纯技术背景或早已远离一线开发的人来说，这种赋能尤为真切。Sarah Friar 笑着分享了自己的亲身经历：她在十几二十岁接受工程训练时曾经写过代码，但当时使用的是早已过时的 COBOL 语言。过去来到办公室时，她常常只能感叹“要是我现在还能写代码该多好”，而如今借助生成式 AI，她兴奋地发现：“你知道吗？我又可以重新写代码了，这感觉太酷了！”她强调，当前最紧要的是教会人们放下恐惧心理，主动去拥抱、实验并学会成为“AI 原生代”（AI native），同时保持对未知的好奇心。大学也应打破传统壁垒，大胆设立以往不曾有过的跨学科（cross-disciplinary）培养项目，为下一代人才开拓更广阔的生长空间。 - **放射科医生的反例**：早前关于 AI 将取代全人类放射科医生的断言并未成真，目前的从业人数比以往更多。 - **软件工程师就业增长**：自 2022 年 ChatGPT 推出以来，全球软件工程师总人数逆势上升约 6%。 - **重获编程能力**：Sarah Friar 早年曾用 COBOL 编程，如今借助 AI 助手重新获得了编写代码的能力。 - **教育与人才建议**：不畏惧技术变革、积极动手实验、培养“AI 原生思维”，并大力倡导高校推进跨学科探索。 --- ### [29:57](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1797s) 面对面的人际连接与投资信任：Jagdeep 的投资经历与致谢 加利福尼亚大学首席投资官 Jagdeep Singh Bachher 随即将话题拉回人际交往的核心价值。他回忆道，几个月前自己初次拜访 Sarah 时，OpenAI 正在进行极其密集的巨额融资谈判（Jagdeep 提到“募集一千亿美元”，Sarah 随即打趣纠正为“122”）。尽管 Jagdeep 在做决策前也专门用自己的 ChatGPT 模型跑过“我们是否应该投资”的分析模拟，但真正促使他下定决心的，却是双方线下见面的时刻。 ```mermaid flowchart TD accTitle: 机构投资决策驱动要素 accDescr: 对比纯 AI 算法分析与线下人际交互在关键投资决策中的不同作用 subgraph Algo["AI 辅助量化验证"] A["ChatGPT 模型推演与数据测算"] end subgraph Human["真实人际互动与信任构建"] B["面对面深入交流"] C["对加州大学机构文化的深度好奇与尊重"] D["探索面向未来学生的校企合作潜力"] E["真诚的领导力、热情与善意"] end Algo --> F["投资决策敲定与深度战略结盟"] Human --> F ``` Jagdeep 直言，直到走出会议室的那一刻，他才真正坚定了入股 OpenAI 的决心。打动他的不是冷冰冰的数据，而是 Sarah 身上展现出的感染力、充沛的热情，以及她对加州大学的历史文化、机构使命和双方合作所能带给学生的深远意义所表现出的由衷好奇与透彻理解。他感慨道，尽管作为投资人见过了无数顶尖的 CFO 和行业领袖，但虽然只与 Sarah 见过三次面，她身上流露出的善意与由衷的感恩之心令他深受触动。作为一家伟大机构的领航者与财务掌门人，Sarah 的表现堪称典范。Jagdeep 认为两人的交流才刚刚拉开序幕，希望这只是系列对话的第一期，并期待未来能为广大 UC 学子带来更多思想碰撞。 --- ### [31:25](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1885s) 论坛圆满收官与后续活动预告 OpenAI Forum 主持人 Mark 重新接入并为主持环节作结。他代表加利福尼亚大学和 OpenAI 社区，对 Sarah Friar 与 Jagdeep Singh Bachher 带来的坦诚、深刻且富有人情味的精彩对谈表示由衷感谢，并感谢所有线上参与的听众。 Mark 随后预告了 OpenAI Forum 近期即将举办的重磅活动： - **企业 AI 落地前沿对话**：本周四将邀请 OpenAI 首席经济学家与贝恩咨询公司（Bain & Company）的合伙人共同登台，深入剖析企业端（Enterprise）采用与部署 AI 技术的实际路径。 - **体育娱乐产业落地实践**：下周二将迎来布鲁克林体育娱乐公司（BSE Global，旗下拥有布鲁克林篮网队 Brooklyn Nets、纽约自由人队 New York Liberty 以及巴克莱中心 Barclays Center）的高管团队，现场分享体育与场馆巨头如何在内部全面部署与应用 AI。 主持人最后在互动区留下了后续活动的报名链接，并期待大家在 OpenAI 论坛后续的思想盛宴中再次相聚。 ## Notable quotes > [01:43](https://www.youtube.com/watch?v=76ABuBGX4IY&t=103s) 你根本没资格当 CFO，你甚至连 CFO 是干什么的都不知道。 — 资深猎头 > [05:30](https://www.youtube.com/watch?v=76ABuBGX4IY&t=330s) 我们每个人都是自己电影里的主角，而其他所有人都在忙着看他们自己的电影，根本没人在看你的戏。 — Sarah Friar > [06:42](https://www.youtube.com/watch?v=76ABuBGX4IY&t=402s) 请修复你自己 — Sarah Friar > [12:12](https://www.youtube.com/watch?v=76ABuBGX4IY&t=732s) 感受这股肾上腺素吧，我正鲜活地活着 — Sarah Friar > [12:12](https://www.youtube.com/watch?v=76ABuBGX4IY&t=732s) 目前科技界舞台上的女性代表实在太少了。俗话说‘看不见就无法成为’。我要带着这份恐惧和肾上腺素走上台，向大家展示科技领域的多元性。 — Sarah Friar > [14:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=867s) 从自身做起，成为未来企业的典范 — Sarah Friar > [14:27](https://www.youtube.com/watch?v=76ABuBGX4IY&t=867s) 不知道就明确说不知道，随后跟进补齐 — Sarah Friar > [21:34](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1294s) 大家若有空，可以打发自己的 AI Agent 去检索拜读一下 OpenAI 经济研究团队的官方博客文章。 — Sarah Friar > [23:18](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1398s) 归根结底，人投资的是人 — Sarah Friar > [25:46](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1546s) 先让一千朵花盛开 — Sarah Friar > [25:46](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1546s) 你在说什么呢？他们以前本来就在作弊。 — 佐治亚理工学院校长 > [29:00](https://www.youtube.com/watch?v=76ABuBGX4IY&t=1740s) 你知道吗？我又可以重新写代码了，这感觉太酷了！ — Sarah Friar ## Mentioned Sarah Friar, Jagdeep Singh Bachher, Mark Murray, Graham Smith, Lia Balsky, OpenAI, 加州大学 (University of California / UC Investments), 安达信 (Arthur Andersen), 麦肯锡 (McKinsey), 高盛 (Goldman Sachs), Salesforce, Square, Nextdoor, 牛津大学 (Oxford University), 斯坦福大学 (Stanford University), 佐治亚理工学院 (Georgia Tech), 博德利图书馆 (Bodleian Library), 贝恩咨询公司 (Bain & Company), BSE Global, 美国国税局 (IRS), 英国皇家税务与海关总署 (HMRC), ChatGPT, Codex, Custom GPTs, IR GPT, GenAI Next Program, COBOL, 《大英百科全书》 ## Video description OpenAI CFO Sarah Friar joined UC CIO Jagdeep Singh Bachher for an OpenAI Forum conversation on AI’s impact on finance and careers. Friar reflected on her nonlinear path from engineering to finance leadership, emphasizing curiosity, adaptability, and kindness as essential skills for students entering a changing workforce. She described how OpenAI’s finance team is using tools like custom GPTs for investor relations, audit support, and more. While AI is transforming finance tasks and expanding what lean teams can do, Friar stressed that human trust and relationships remain central, especially in fundraising and leadership. You can find out more about the OpenAI Forum here: https://forum.openai.com [Source](https://www.youtube.com/watch?v=76ABuBGX4IY)

[1:43] 资深猎头: 你根本没资格当 CFO，你甚至连 CFO 是干什么的都不知道。 [5:30] Sarah Friar: 我们每个人都是自己电影里的主角，而其他所有人都在忙着看他们自己的电影，根本没人在看你的戏。 [6:42] Sarah Friar: 请修复你自己 [12:12] Sarah Friar: 感受这股肾上腺素吧，我正鲜活地活着 [12:12] Sarah Friar: 目前科技界舞台上的女性代表实在太少了。俗话说‘看不见就无法成为’。我要带着这份恐惧和肾上腺素走上台，向大家展示科技领域的多元性。 [14:27] Sarah Friar: 从自身做起，成为未来企业的典范 [14:27] Sarah Friar: 不知道就明确说不知道，随后跟进补齐 [21:34] Sarah Friar: 大家若有空，可以打发自己的 AI Agent 去检索拜读一下 OpenAI 经济研究团队的官方博客文章。 [23:18] Sarah Friar: 归根结底，人投资的是人 [25:46] Sarah Friar: 先让一千朵花盛开 [25:46] 佐治亚理工学院校长: 你在说什么呢？他们以前本来就在作弊。 [29:00] Sarah Friar: 你知道吗？我又可以重新写代码了，这感觉太酷了！

[1:43] Headhunter (recounted by Sarah Friar): You have no business being a CFO. You don't even know what a CFO does. [1:43] Sarah Friar: We're all the protagonist in our own movie and everybody else is watching their movies, no one's watching your movie. [6:41] Sarah Friar's Executive Assistant: Did you just ask it to fix itself? [6:41] Sarah Friar: I neither believe that AI will have no impact... And I also don't believe AI is going to take away all of our jobs... There are some jobs that will probably get usurped by an agent. And many of those jobs probably should because they're jobs that were very routine and mundane. [12:12] Sarah Friar: you cannot be what you cannot see [14:27] Sarah Friar: time to slow down to go fast [16:00] Sarah Friar: I don't know, but I will follow up. [16:00] Sarah Friar: inches away [19:27] Jagdeep Singh Bachher: less is more [23:17] Jagdeep Singh Bachher: human meets machine [23:17] Sarah Friar: in the end, people invest in people

[0:09] Hi everybody. Welcome to the Open AI [0:11] forum. I'm Mark Murray, editorial [0:13] director here at Open AI. [0:15] We're joined today by Sarah Friar, Open [0:18] AI's chief financial officer, and [0:20] Jugdeep Singh Bocker, chief investment [0:22] officer at the University of California. [0:25] Today's conversation comes at a [0:27] fascinating moment for finance. Sarah [0:29] and Jugdeep will talk about how AI is [0:31] changing finance work, what kinds of [0:34] tasks AI can help with, and what skills [0:36] are becoming increasingly important. [0:39] I also want to give a quick shout out to [0:41] the UC students tuning in today. We're [0:43] excited to have UC students interested [0:45] in finance careers joining us along with [0:48] our forum community. And with that, I'll [0:51] turn it over to Jugdeep, who will [0:52] interview Sarah for this conversation. [0:58] >> And Sarah, thank you for agreeing to do [1:00] this. [1:01] >> I'm very excited. [1:02] >> I'm very excited, too, cuz we have with [1:05] us [1:06] a lot of students from the University of [1:08] California, a very important group of [1:11] individuals that are thinking about the [1:12] future with AI. [1:15] And I thought, you know, who better to [1:18] have a conversation around AI and [1:20] finance [1:21] than yourself. So, to start with, [1:25] before we get into more detailed [1:27] questions around the the benefits and [1:29] all the opportunities that come around [1:31] with AI, let's start with your career. [1:34] How did you get to [1:36] get interested in finance, [1:38] and ultimately get to this very [1:40] important role [1:42] in finance? [1:43] >> Well, first of all, hello students. I'm [1:45] super excited to be talking to you today [1:47] as well. As the mom of teenagers, we'll [1:49] come to that in a moment, or actually [1:51] not even a teen. I have a 21-year-old [1:53] now. [1:54] Um, it is an incredible time in your [1:56] life. I'll I'll [1:58] take the clock back a little, but I will [2:00] not I'll make it short. I grew up in [2:01] Northern Ireland. I grew up during the [2:03] troubles, not the best time, but you [2:05] know, very unique time. I um was always [2:09] super curious as a kid. I was the kid [2:11] that took my mom's, you know, called [2:12] them vacuum cleaner Hoover apart. I [2:15] could never figure out where the last [2:16] screw would go, but clearly recognized [2:18] that it'd been over built. It didn't [2:20] need that last screw. It still worked. [2:22] But I thought that would turn me into an [2:23] engineer. And that was the path I took. [2:26] I was an engineer in college. But right [2:28] ahead of college, I had to um had to [2:30] figure out a way frankly to pay my way [2:32] through college. And so I took a year [2:35] out working for Arthur Andersen. And [2:37] that took me into the world of finance. [2:38] So I was a trainee accountant [2:40] effectively for that year when I was 18 [2:43] to 19. And then Arthur Andersen [2:45] sponsored me through college. And so I [2:47] got to go in every holiday and work in [2:50] different parts of their organization, [2:52] which took me through everything from [2:54] tax to consulting. They even had a [2:56] bankruptcy um [2:58] uh division that I spent a summer in. So [3:00] it was really a fascinating way to learn [3:02] about finance. [3:03] That said, I kind of exited college with [3:05] my shiny bright master's degree in [3:07] material science and thought I would [3:09] still do this um world of more [3:11] engineering. [3:12] Um [3:13] it didn't quite, you know, careers are [3:16] long and also quite zigzaggy. They're [3:18] they're never the straight line that you [3:20] expect. So I ended up going to McKinsey. [3:22] McKinsey sent me off to South Africa. I [3:24] was the only young BA at the time that [3:26] had spent any time in the mining [3:28] industry of all things cuz I wrote my [3:30] thesis on how to extract gold out of [3:32] sulfide ores. [3:33] And hence with McKinsey, I got this [3:36] wonderful purview into like how [3:37] businesses really work. And it was like [3:39] an extension. It was just really like [3:40] continuing at school frankly. I feel so [3:42] grateful for that time. [3:44] Short you know, quick kind of after [3:46] that, McKinsey sent me to business [3:48] school. That brought me out here to the [3:49] West Coast. I will not say where I went [3:52] cuz you'll all boo me, but somewhere [3:53] here on the West Coast that might be a [3:56] little bit of a of an adversary to Cal [3:58] Berkeley. Um but I came out [4:01] >> you [laughter] have to tell us. [4:01] >> I came out to Stanford for 2 years. I'm [4:03] sorry. I should have come to Berkeley. [4:05] >> I shouldn't have asked you that. [4:06] >> [laughter] [4:06] >> Uh [4:07] Uh spent 2 years in the West Coast and [4:09] then stayed. Um I joined Goldman, really [4:11] got to learn about banking, about equity [4:13] research. Um and then I actually really [4:16] realized I'm I loved operating. I wanted [4:19] to be inside a company. A very natural [4:21] path given my background was into the [4:22] world of finance. One of the best pieces [4:25] of career advice I got at the time was I [4:27] thought I was all ready to be a shiny [4:29] bright new CFO. And I went and had lunch [4:31] with a headhunter who at the end of [4:33] lunch said, "You have no business being [4:36] a CFO. You don't even know what a CFO [4:37] does." And I was very taken aback. And [4:40] by the way, those are good moments in [4:41] your life where someone kind of calls [4:44] you on it. And they said, "The best [4:45] thing you could do is go work for a [4:47] great CFO and learn the tools of your [4:49] trade." So, I joined Salesforce and I [4:51] worked for an amazing CFO, um [4:54] uh Graham Smith, incredible individual, [4:57] great mentor, great coach, great friend. [4:59] And I really learned a enough at that [5:01] point um to eventually project myself [5:04] into Square where I was the CFO of [5:05] Square, became CEO of Nextdoor, and then [5:08] finally the CFO of OpenAI. So, I tell [5:10] the story a little bit more long cuz I [5:12] think it's good to see that you start [5:14] out thinking you're going one direction, [5:15] you may end up pivoting, but always [5:17] think about what are the primitives or [5:19] what are the basic skills you're [5:21] learning along the way. What are you [5:22] great at? What gives you pleasure? What [5:25] what actually can pay the bills? And try [5:27] to bring all those things together. [5:29] >> Incredible. Now, did you want to be a [5:31] CFO? [5:32] >> Uh [5:33] >> Or was that the headhunter [5:35] challenging you to become a CFO? [5:37] >> No, at the time I think when I came out [5:38] of Goldman, I everyone kept telling me, [5:41] "Oh, you've been an equity research [5:42] analyst. You should go be a CFO." And um [5:46] and it sounded like a really good title. [5:48] And this is like another good learning [5:49] in your career life. It's like don't get [5:51] enamored of titles or what other First [5:54] of all, what other people think. When I [5:56] left Goldman, I felt like I had to go do [5:57] something and CFO sounded like I was [5:59] doing something. [6:01] And you know [6:01] >> Sounds cool. [6:02] >> I'm like, why? And it was cuz other [6:04] people will you know, like might look at [6:06] it if I don't go do that. [6:08] Nobody cares. Like the good news is [6:10] we're all the protagonist in our own [6:12] movie and everybody else is watching [6:13] their movies, no one's watching your [6:14] movie. So never do things because of [6:16] what you think other people will think. [6:18] Um, I think the other reason the [6:20] headhunter said that though at the time [6:22] was truly like I didn't know what a CFO [6:25] did. I thought I knew from the outside. [6:28] And seeing what someone's job is like, [6:30] you know, when they they tell you the [6:32] the external facing things is very [6:34] different from the guts of the [6:35] operation. And so getting inside a [6:38] company and seeing what operators do, [6:39] that is the best way to learn. [6:41] >> Incredible. Now, you're having [6:43] conversations at dinner with your kids, [6:46] those that are at home. [6:47] >> Mhm. [6:47] >> And some on the phone call. [6:50] What are the discussions around careers [6:52] and ambitions and jobs and the impact of [6:56] AI [6:57] to the future of [6:59] young Americans? [7:00] >> Yeah. So I mean, I'm right in this right [7:02] now with probably the folks listening on [7:04] on the call here on the on the web [7:07] webinar. My daughter is a junior in [7:09] college, my son's a freshman. So [7:12] we talk about this all the time. I've [7:13] got one home at the moment which I'm [7:15] excited about, one about to come home. [7:18] I use three words when I talk about I [7:20] came out AI career future. Number one is [7:23] curiosity. Number two is adaptability. [7:26] And the third I will always say to [7:28] everyone, which is kindness, which is [7:29] probably an unexpected third in some [7:32] ways. [7:33] The first is um curiosity. You know, [7:36] dive in, try stuff. Don't be afraid of [7:39] things. Like fear usually comes when we [7:41] don't understand or we're not sure what [7:43] something will be. I face this every day [7:46] at work, right? We live in an [7:47] environment at OpenAI where we probably [7:49] launch something new every single day. [7:51] And I can have that feeling after a few [7:53] weeks where someone [7:56] um in the org will ask me, "Am I using [7:58] XYZ?" or "Have I tried this other [8:00] thing?" or "Have I downloaded a skill in [8:03] Codex?" And [8:04] you know, I can almost feel the fear in [8:07] me coming forward cuz I'm like, "Oh my [8:08] goodness, I don't know what they're [8:09] talking about." And then I go off and I [8:11] try to download it and then the thing [8:13] doesn't work. And um [8:15] and I remember coming to work right [8:16] after I started using Codex, which is [8:18] our agentic platform, and kind of [8:20] saying, "We need to get someone in my [8:21] team to come. I tried this and it didn't [8:23] work." And my assistant, my EVP, [8:26] actually turned me and said, "Did you [8:27] just ask it to fix itself?" And I was [8:29] like, "What are you talking about?" And [8:30] he's like, "Just say, 'Please fix [8:32] yourself.'" And it's like magic, and it [8:35] did. [8:35] >> Wow. [8:36] >> And so, I think this is a big shift, by [8:37] the way, in using technology, where [8:39] being able to ask the technology to help [8:42] you use the technology, whether it's to [8:44] fix itself or to suggest ways to use it, [8:46] that is a such a night and day change [8:48] from when I was taking my mom's vacuum [8:51] cleaner apart and it was not telling me [8:53] where to put the final screw back in [8:54] again. So, huge change just in how we [8:57] use and feel and intuitively work with [8:59] technology. So, curiosity, number one. I [9:02] think I've always been very curious by [9:04] nature, and so I think now of, you know, [9:07] back in Northern Ireland when the man [9:10] showed up at our front door and sold my [9:11] parents the encyclopedic Encyclopedia [9:14] Britannica, was like the best day of my [9:16] life because I lived in a place where [9:19] there wasn't a ton of ways to educate [9:21] yourself. All of you on this call right [9:23] now, you have access to not just the [9:25] Encyclopedia Britannica, but like so [9:27] much knowledge. [9:28] >> Right. [9:28] >> And making use of that, I think, is such [9:30] a gift. [9:31] Second is adaptability. You know, trying [9:34] to talk with my my right now about the [9:36] career for them, [9:38] I'd be like my parents trying to talk to [9:39] me about my career. [9:40] My My dad My mom grew up on a farm. She [9:44] was a nurse. My dad um you know his dad [9:47] was an electrician. Actually came to [9:48] Northern Ireland to put up the [9:49] electricity wires, which I think is a [9:51] really cool arc thinking about AI. And [9:54] um [9:54] >> And you need a lot of that right now. [9:55] >> You do. And um but think about like what [9:57] it was like when the lights turned on [9:59] for the first time in someone's house. [10:00] Like I you know I think there's an a [10:02] really strong analogy here. Um Both of [10:05] them left school at 16. Um they're [10:07] amazing parents. They are the most [10:09] amazing community people. I'm so [10:11] grateful to them. But them trying to [10:13] give me career advice was just not [10:15] happening cuz they didn't know. They [10:17] wanted me to be a doctor, which they [10:18] understood. Um but every other career [10:21] they weren't really sure about. And so I [10:23] think that's so much true today. Like as [10:25] we try to give our kids advice on the [10:27] career, AI is changing a lot. I don't [10:31] I'm not in binaryisms. I neither believe [10:33] that AI will have no impact, don't [10:35] worry, we'll all be fine, everything [10:37] will be steady as she goes. And I also [10:39] don't believe AI is going to take away [10:41] all of our jobs. Our economic research [10:43] team has actually done a very good piece [10:45] of research that suggests it's much more [10:47] There are some jobs that will probably [10:50] get usurped by an agent. And many of [10:53] those jobs probably should because [10:55] they're jobs that were very routine and [10:57] mundane. [10:58] There are jobs that are absolutely going [11:00] to change. My job is night and day [11:02] different even a year on from starting [11:05] to use agentic um outcomes in my work. [11:08] And then there are jobs that we cannot [11:10] even imagine that haven't been created [11:12] yet. Like my mom couldn't imagine my [11:15] job. My grandmother worked on a farm. [11:18] She definitely couldn't have imagined [11:19] what I do today. And so I think that [11:22] adaptability is key. And then I'll just [11:24] end with kindness. I do think that [11:27] no matter what you're doing in life, [11:29] like [11:30] learning that to be kind, to be [11:32] empathetic, like in the end, I think so [11:35] much of humanity, like it's a repeated [11:37] thing, it comes around over and over [11:38] again. Like helping people before they [11:42] you know, they're not giving you [11:43] anything, but you're doing something for [11:45] them. I think in your life, you will be [11:48] shocked how much it pays back to you [11:50] over and over again. If you're a young [11:52] student right now sitting in college, [11:54] you're coming into the workforce for the [11:56] first time, do not underestimate the [11:58] power of just doing something kind for [12:01] someone. Along the way, it will return [12:04] in bucket loads to you. So, those are my [12:06] three words that I drive home to my [12:08] kids, and I hope folks listening maybe [12:10] take something away on that front. [12:12] >> Incredible. Can I just dive in on fear [12:14] for a moment? [12:15] >> Mhm. [12:15] >> Is there one thing you would suggest as [12:18] a tip on how to overcome one's fear [12:20] >> Mhm. [12:21] >> on embracing AI? [12:23] >> Um [12:25] So, I have one of my kind of rules of [12:28] the road for how I feel about my life is [12:30] about adrenaline. And when I when you're [12:33] afraid, you get that adrenaline rush. I [12:35] actually like getting an adrenaline [12:37] rush. [12:38] >> Okay. [12:38] >> I My brother is an ER doctor. He [12:40] actually did what my parents wanted to [12:41] do, became a doctor. They understand [12:43] what he does. [12:44] >> My wife is an ER doctor. [12:45] >> It's and they're they're built from [12:47] different material than the rest of us. [12:50] Um [12:50] He always says like in the end, if it's [12:53] all going poorly, what do they do to [12:55] bring you back? They stick a needle of [12:58] adrenaline into you, and that's like [12:59] part of trying to shock the system back [13:01] into moving. So, if you want to feel [13:03] alive, feeling that adrenaline rush. So, [13:06] usually when I find myself in a moment [13:08] of fear, I actually try to flip it back [13:11] to number one, oh, feel that adrenaline, [13:13] I'm alive. And number two, how can I [13:16] harness this moment of what I'm feeling [13:19] and try to turn it into something [13:20] positive. Um and I feel this all the [13:23] time, right? I I had to give a talk [13:26] yesterday morning, I think. Um I don't [13:29] often get a ton of adrenaline when I'm [13:30] presenting, but this was like it felt [13:32] like a higher stakes moment. And so in [13:34] that moment, right before I take the [13:36] stage, it's a really good moment to feel [13:38] the adrenaline rush. And frankly, in my [13:40] head, what I'm thinking in those moments [13:41] is usually um it's a little off topic, [13:44] but bear with me for a second. There [13:47] just not that many women on stages right [13:49] now in tech. And so I'm going to take my [13:51] fear and my adrenaline rush, and I'm [13:53] going to do this because I can't You [13:55] cannot be what you cannot see. And so [13:59] going to go out and do that pitch, do [14:01] that stage show, because I just think [14:04] it's good to show diversity in [14:05] technology. So it's a very minor example [14:08] of how I try to harness fear and how I [14:10] try to kind of work through it by just [14:12] taking myself out of the equation a [14:14] little bit and thinking about a greater [14:16] good, a greater moment that I can create [14:18] with that fear. [14:19] >> Thank you for sharing that, especially [14:21] on recognizing those that might not [14:24] feel like they're represented in the [14:26] audience. [14:27] Um now, where do you see AI enabling [14:30] people to do more [14:32] particularly in the finance industry, [14:34] since that's the focus today, but [14:37] there's so much more we can do. [14:38] >> Yeah. So on the finance side, I mean [14:40] what we have really seen happen in our [14:43] my finance world here at OpenAI is we're [14:45] pushing ourselves every day to become [14:48] the company of the future beginning with [14:50] finance. That is one of our core [14:51] objectives. So about 2 years ago when I [14:55] joined, we started small. So we were [14:57] using at the time ChatGPT and custom [15:00] GPTs. Um we were using them in small and [15:03] big ways, and one of the things I did [15:05] for my team was to create hackathons. So [15:08] every quarter, we would create some [15:10] space for the team to slow down to go [15:12] fast. Because when I would poll people [15:14] on like, "Why aren't you Why aren't you [15:16] using this new tool?" they were like, "I [15:18] don't have any time to learn it. I'm too [15:20] busy getting my work done." And that to [15:22] me is always symptomatic of people who [15:24] actually need a breath to slow down um [15:27] so that then they can come back and use [15:28] tools that will accelerate them. So, [15:30] that was the beginning was those [15:32] hackathons. Um one of my real magical [15:36] moments is we were in the middle of a [15:37] big fundraise. We created a custom GPT [15:40] for investor relations. What we did is [15:42] effectively fed ChatGPT all of our [15:45] materials. So, our a company [15:46] presentation, our all of the diligence [15:49] materials we'd created um we would we [15:52] started to update it and then very [15:54] quickly we have at our fingertips a [15:56] custom GPT that helps us answer any [15:58] investor questions. So, you know all [16:00] those answers you got that you thought [16:01] were so good, they were coming from our [16:03] custom GPT. But, that aside [16:06] I [16:07] >> Without any hallucinations. [16:08] >> With no well, it's really important to [16:10] give the you know, whatever if it's our [16:13] agents today or to give your memory a [16:15] lot of direction [16:17] >> Right. [16:17] >> about things that you don't want it to [16:19] do or say. Like in the case of our IR [16:21] GPT, we want it to be very factual a [16:25] little salesy but not too salesy, right? [16:27] We want it to have a positive um [16:29] personality but not to go into mass [16:31] sales mode cuz with investors we want to [16:33] be very high integrity, you know, very [16:36] on the mark. We also want to make sure [16:38] that if it doesn't know that it just [16:41] says it does not know, which is the best [16:43] piece of advice I ever got as a CFO on [16:45] an earnings call was to say "I don't [16:48] know, but I will follow up." So, I try [16:50] to give my custom GPT that personality. [16:53] Um I will run the clock forward to about [16:56] about a month ago I was in South Korea [16:58] in an investor meeting and when I walked [17:00] in the door they presented me with two [17:02] pages of questions in Korean. And so, I [17:05] pulled my phone out which had my custom [17:07] GPT on it. I snapped a photograph um and [17:10] I asked it to just give me the answers. [17:12] So, of course it knew to translate, gave [17:15] it me all the answers. I actually wanted [17:17] to then have ChatGPT talk back in the [17:20] meeting in Korean. Um my translator [17:22] wanted to make sure that we still went [17:24] through the translator, but we are [17:25] inches away from being able to just [17:28] communicate in a bi-directional way, [17:30] English to Korean and back, so that you [17:32] can have a very fruitful conversation [17:35] with an investor utilizing my um IRGBT. [17:38] So, that that was a very early example. [17:41] But, today we are really going AI native [17:44] in the finance world. [17:46] >> Okay. [17:46] >> So, making sure that every process we [17:49] do, right? When you're a CFO, you'll [17:50] start to hear these terms like quote to [17:52] cash, [17:54] um uh order to payment, um [17:57] equity, [17:58] uh tax, treasury, right? These are kind [18:00] of the building blocks of any strong [18:03] finance organization. And so, within [18:05] each of those, we look at what are the [18:08] core tasks each is trying to do, and [18:11] then we're constantly looking for ways [18:13] where we can embrace our own technology [18:15] to make us more productive, more [18:17] insightful, do things faster, um even [18:21] with our accountants, I've been spending [18:23] a lot of time talking to them about [18:26] something that I learned back when I was [18:29] 18, 19 years old, which is most [18:32] accountants, when you're doing an audit, [18:34] you do something called sampling. [18:36] And why do you do sampling? You do [18:37] sampling because there's not enough [18:39] people and time to actually go read [18:41] every single invoice, for example. [18:43] >> Right. [18:44] >> So, you instead say there's a thousand [18:46] invoices in this period, I'm going to [18:48] sample 10. And if the 10 I sample are [18:52] what I expect to find, then I'm going to [18:55] check this audit. I expect now that the [18:58] other 990, did I say a thousand? are [19:02] correct. But, you do that because you [19:04] live in a world of scarcity. Scarce [19:06] resources, scarce people, scarce time. [19:09] In a world of abundance, which is what [19:10] AI brings us, [19:12] The agent can actually check all 1,000 [19:14] invoices. And so it really kind of [19:16] starts to shift the control. I actually [19:19] think it will make controls and and [19:21] precision and so on much stronger. [19:24] I'll give you one other example. So, in [19:27] our tax team, our tax team is probably [19:29] our most AI pill team, which is funny [19:31] cuz tax people are often your most [19:32] conservative. But we've done something [19:34] magical here at OpenAI. Our tax team [19:36] loves our tools. So, one of the things [19:38] they have to do at the at the end of [19:40] every tax year is they have to think [19:43] about all of the tax forms that we need [19:45] to fill in and um and put in to every [19:49] tax department in the world. Like we're [19:52] a very global company. Um today only 10% [19:56] of ChatGPT weekly active users are in [19:58] the United States. So, 90% are outside [20:01] the US. Our fastest growing continent [20:03] right now is Africa, for example. [20:05] >> Wow. [20:05] >> So, imagine what it would be like [20:07] without the use of of our own [20:10] technology. You would have people whose [20:12] job would be to go to the IRS website or [20:15] the equivalent, you know, the Inland [20:16] Revenue if you're in the United Kingdom, [20:19] to go scour through to find the tax [20:22] form. And these things have a sneaky [20:24] habit of changing slightly every year. [20:27] So, you might think you've nailed it and [20:28] then they'll add a new box somewhere. [20:31] So, now it's going to pull that down, [20:33] usually in a PDF. [20:35] It will understand the shape of the [20:36] form. You can give it the the data on [20:39] the other side of like here here's what [20:41] we all calculated as tax people, but [20:44] like the mundane task is the writing it [20:46] in in all the right boxes. So, today we [20:49] have now totally automated that. So, [20:52] every tax form in the world is [20:54] automatically pre-filled, which means [20:56] that my tax team moves into checking [20:59] mode. That saved them a lot of time. And [21:01] ultimately, as we get more and more [21:03] confident, our hope is that we we even [21:05] just file in an automated way, which [21:08] gives the tax team more time to spend on [21:10] are the numbers correct? Are we doing [21:12] the right thing for the business? Are [21:13] there some interesting new ways that we [21:16] could present our taxes? But to do what [21:19] they went to school for, the people on [21:20] this call. [21:22] But they you all are not in school right [21:23] now to learn how to fill in PDFs. And so [21:26] I think that is a really interesting [21:28] shift in my world and I could give you [21:32] hundreds of examples like that. [21:34] >> So at [21:35] At UC Investments, we have our culture, [21:37] which is the UC Investments way. And the [21:39] first pillar of it is less is more. And [21:41] as I'm listening to you, it sounds like [21:43] you can do so much more [21:45] >> Mhm. [21:46] >> with a lean and a small team. [21:48] >> Yeah. [21:48] >> How many people do you have on the [21:49] finance team for such a global [21:52] enterprise with over a billion [21:54] >> It's a [21:54] >> active users? [21:55] >> Today we are finance team is probably [21:57] about 200 people [21:59] >> Incredible. [21:59] >> in that zone. Yeah, we are we are quite [22:01] lean. Now, we [22:05] also though are in certain areas [22:08] probably bigger than maybe a normal [22:10] finance team. And that comes back to [22:13] with the technology, we can drive a lot [22:15] more insights. So, for example, in my [22:17] finance team today, I have an economic [22:20] research team. I have a pricing team. [22:22] Like these often will not sit in the [22:24] CFO's world because the CFO's overloaded [22:27] with all the other things. [22:29] But actually I'm able to grow in certain [22:31] areas where actually I deem there being [22:33] a lot of room for real insight. So, [22:36] economic research is a great example [22:38] where wouldn't normally put an I've [22:40] never had an economic research team as a [22:42] CFO. But it kind of now helps me use my [22:45] old research learning when I was at [22:47] Goldman Sachs and pair that to the fact [22:50] there's so much curiosity in the world [22:52] right now about what's happening with [22:54] AI, paired with the fact that we at [22:56] OpenAI have access to a lot of data. I'm [22:59] being able to kind of create insights [23:01] that help us with investors, they help [23:03] us with customers, they help us with [23:05] just the curious. And so, if you all [23:08] have a minute in your day, you can send [23:10] your agent off to find the OpenAI [23:12] economic research team's blog post, and [23:14] I highly recommend them. [23:15] >> Incredible. Now, [23:17] the [23:18] ninth pillar is human meets machine. [23:21] >> Mhm. [23:21] >> And you've described fundraising [23:24] and all these incredible tools that [23:26] helped. [23:27] What's one thing that [23:29] hasn't changed in the fundraising [23:31] process [23:32] >> Mhm. [23:33] >> with [23:34] AI? [23:36] >> So, [23:37] certainly for now, I still believe one [23:41] of my mantras on fundraising is in the [23:43] end people invest in people. [23:45] >> Okay. [23:46] >> And I think it's so much about trust and [23:50] integrity, right? When you came to [23:52] visit, [23:53] like our breakthrough moment was just [23:54] the fact we stood in the same room [23:56] together. [23:57] >> Right. [23:57] >> And uh if I recall at the time, um you [24:01] brought in your amazing book of values, [24:03] by the way, which I have sitting in my [24:05] in my office right now. Let's say [24:07] office, my room. I don't really have an [24:08] office. [24:09] >> You just have a desk. [24:10] >> And I love it. I have a desk. [24:12] Way to go, people. That's what you've [24:13] got to look forward to. [24:14] You you will become the CFO, and you [24:16] will get be granted a stand-up desk. [24:19] Um [24:20] but what I loved about that moment is I [24:23] remember in fundraising, [24:26] it's not about the money that you get. [24:29] Money is important to run your business, [24:31] but it's the people you surround [24:32] yourself with. Because those people [24:35] don't just bring you money, they bring [24:37] you many, many other things. They teach [24:40] you, like your book teaches me, the way [24:42] you have shown up in the world teaches [24:44] me. The fact that you said, "I want to [24:47] come in and help you think about how do [24:49] you talk to um you know, a large-scale [24:51] educational establishment, like the [24:53] UCs?" Um you said, "You know what? We [24:55] could do like a podcast. It'd be super [24:57] fun." We're doing it right now. [24:58] >> Here we are. [24:59] >> And that's what has not changed to me [25:02] for fundraising, and it's why I always [25:04] put kindness in one of my top three, [25:06] because I think when you treat people [25:08] well, when you take the time to get to [25:09] know them, in the moment you might not I [25:12] don't know if we were going to raise [25:13] money from you all. I didn't know if [25:15] you'd want to invest in us. I didn't [25:16] know if you'd be the right investor for [25:18] us on the other side. But it's a long [25:21] game, and you may find not at that [25:23] moment, but maybe it's in 6 months or in [25:25] a year, or maybe it's in 5 years, but [25:27] it's incredible how many people come [25:29] around in your career again, and I think [25:32] the human-to-human has not changed for [25:34] me. [25:34] >> I should have just sent it 2 years [25:35] prior. [25:36] >> [laughter] [25:37] >> That that [25:38] that that when I actually had a chance [25:40] to to to meet Sam Altman for the first [25:43] time, and he had said, "This is a great [25:44] opportunity." Now, [25:46] uh what advice would you have for [25:49] students and universities as they're [25:52] growing their talent and their pipeline? [25:55] >> Mhm. [25:55] >> Of [25:56] future OpenAI employees, future [25:59] contributors to the economy, what skills [26:02] do they need to learn? [26:03] >> Yeah, so I mean, first of all, we've we [26:06] work with a lot of universities around [26:08] the globe. I do think the UCs have been [26:10] really ahead of their time, and we're [26:11] really appreciative for the fact that [26:13] you do work closely with us. Um our EDU [26:16] team under Lia Balsky is phenomenal. [26:19] Um we have a program called um uh the [26:22] GenAI Next Program, where we on it to [26:25] work with many in um you uh [26:27] establishments around the world. One One [26:29] that's near and dear to my heart is [26:30] Oxford, which is my alma mater for my [26:32] undergrad. Um we've done a bunch of work [26:34] with them with the Bodleian Library, [26:35] which is I think the oldest library in [26:37] the world, certainly one of the biggest. [26:39] And are there ways to bring back, you [26:40] know, manuscripts from like the 1400s or [26:43] whatever, and actually understand them? [26:45] So, I think for universities, one piece [26:47] of advice is just, you know, deploy and [26:50] be curious and learn. And And as I say [26:53] with enterprises as well, sometimes you [26:55] have to kind of let a thousand flowers [26:57] bloom first and let people experiment [27:00] before you try to lock it in. So, that's [27:01] kind of one piece of advice. I think the [27:04] second thing is [27:06] um frankly, you know, how are you going [27:08] to AI-ify your classes? I I push um [27:12] uh teachers on this all the time. I [27:14] think there was original fear of like [27:16] students are going to go around teach [27:18] cheating. Um I remember having a [27:20] wonderful conversation uh with the head [27:23] of um Georgia Tech. And he said, you [27:26] know, he had this moment where his head [27:28] of admissions had run in the door and [27:29] said, "Oh my goodness, they're all going [27:31] to cheat on their admissions essays." [27:32] And he was like, "What are you talking [27:33] about? They already cheat." And he I was [27:35] like stopped in his tracks like, "What [27:36] do you mean they're cheating?" He's [27:38] like, "Well, you know, if you're wealthy [27:40] enough to go get the counselor or your [27:41] good neighbor down the street is helping [27:43] you, is that cheating or is that just [27:46] using the resources at your fingertips?" [27:47] So, in this case, our job is not to try [27:50] to call people out for cheating, but [27:52] instead to set admissions essays that [27:55] really tell us about the student. And if [27:56] they're using ChatGPT to help with that, [27:59] we should encourage that because that [28:01] shows they're curious and they're using [28:03] the tools at their fingertips. So, I [28:04] think the same is true for in-classroom [28:06] participation, too. It's how do we [28:08] utilize the tools to keep extending the [28:10] thinking without having people cheat [28:12] themselves, which is you still have to [28:14] learn how to think. Like my daughter, as [28:16] I said, is a junior, she's doing [28:18] chemistry. I see her working all the [28:20] time with chat because she'll say, [28:22] "Instead of waiting for my TA on a [28:24] Monday morning to help me solve this [28:25] problem I'm stuck at, if I can keep [28:27] going when I'm in flow, that's a good [28:30] thing. And in the end I have to take [28:31] finals and I can't use chat then. So, [28:34] it's helping me learn, but in the end I [28:36] have to show that I have actually [28:38] reached a degree of mastery." So, I [28:40] think that's the second thing. You know, [28:41] one is experiment, two is really kind of [28:44] try to shift it into how to make it [28:45] AI-native. Um Um, then I think the final [28:48] thing I would say [clears throat] is [28:50] it's not in some ways like there was a [28:52] lot of concern back in the day about [28:54] radiologists, right? And people have [28:55] talked about that over and over again. [28:56] That somehow, you know, particularly at [28:58] the time um machine learning would take [29:00] all radiologists out of the world. They [29:02] don't. There's more of them today than [29:04] that. It's the famous study. But even I [29:06] was looking at software engineers [29:07] recently, there are more software [29:09] engineers now too since 2022 when [29:11] ChatGPT arrived. I think the software [29:14] engineer population is actually up about [29:15] 6% since then despite the you know, [29:18] there's a lot of chatter about its [29:19] demise. In reality, what's happening is [29:22] many more people can do things like [29:24] develop code than could before, right? I [29:27] did code development back in my teens [29:29] and 20s as an engineer, but on Cobalt. [29:32] So like I come to work now and I'm like [29:34] I wish I could code. You know what? I [29:36] can code again. It's super cool. So I [29:39] think it's like teaching people not to [29:41] be fearful, to embrace it, to [29:43] experiment, to learn how to be AI [29:45] native. Um, and then also just to be [29:47] curious about what's next. Like what are [29:49] the new programs that universities might [29:51] not have put in place. I think there's a [29:53] lot of space right now for [29:54] cross-disciplinary [29:56] um outcomes as well. [29:57] >> Incredible. Now look uh [29:59] when I came to meet you [30:01] a few months ago, you were in the middle [30:03] of raising a hundred billion dollars. [30:05] >> 122. [30:06] >> 122. [30:08] I had run my ChatGPT model on should I [30:11] invest? [30:13] But candidly, not until when I showed up [30:15] and I met you in person. [30:18] The energy, the excitement you had, the [30:21] fact that you were curious about who we [30:23] are, [30:24] and actually understanding our culture [30:25] and our institution, and what it meant [30:27] to OpenAI. [30:29] All of those things, as soon as I left [30:30] here, gave me the excitement to want to [30:32] be an investor in OpenAI, but also [30:35] recognizing there's so much OpenAI in [30:37] partnership with the UC can do for the [30:39] future of students. [30:41] We both get to meet a lot of great [30:44] investors, CFOs, leaders. [30:48] You've always been kind and I've only [30:49] met you now twice before. This is the [30:51] third time. [30:52] But your kindness comes through. [30:54] Your gratitude is heartfelt and genuine [30:57] and I really appreciate it. And I would [30:59] say in addition to this being a great [31:01] company, [31:03] it's great to have you as the CFO and [31:05] the leader of the institution. So, thank [31:06] you for doing this today. And hopefully [31:08] to all the UC students, [31:10] this was just part one [31:12] of another podcast or conversation we [31:15] can have. I I feel like we're just [31:16] getting started, warmed up. [31:18] >> We are. [31:18] >> So, thank you very [clears throat] much. [31:19] >> Yeah, appreciate it. [31:21] >> Wonderful. Back to you, Mark. [31:25] >> That was awesome. Thank you for that [31:26] great conversation, Sarah and Jagdeep. [31:29] We appreciate everyone taking the time [31:31] to tune in, including the UC and Open AI [31:33] Forum community. And we hope you join us [31:36] for our future upcoming Forum events. [31:38] This Thursday, we're hosting our chief [31:39] economist and partners from Bain and [31:41] Company for a discussion on enterprise [31:43] AI adoption. Then next Tuesday, we're [31:46] hosting the Brooklyn Sports and [31:48] Entertainment, the owners of the [31:49] Brooklyn Nets, the New York Liberty, and [31:51] Barclays Center for a conversation on [31:53] how they deploy AI internally. And we're [31:55] going to be dropping in links to [31:56] register for upcoming events. We hope to [31:59] see you back at the Open AI Forum very [32:01] soon.