OpenAI Trains Frontier Models to Act as Clinical Safety Net

OpenAI训练前沿模型以构筑临床安全网 急剧下降的医疗智能成本与主动型AI智能体将化解碎片化护理的行政负担，构建起共享的临床安全网，而非取代医生。 OpenAI健康主管Karan Singhal概述了公司的双重使命：通过提供充沛且值得信赖的医疗智能，提升全球医疗保健的底线与上限。OpenAI并未构建孤立的垂直领域模型，而是将临床推理能力直接融入通用的前沿架构中，从而打造能够跨碎片化系统充当环境式安全网的工具，并将临床指南的普及应用周期从数十年压缩至数小时。Singhal强调，AI旨在统筹后勤事务、解析纵向健康记录并为临床医生赋能，而非取代实体医疗服务的交付；同时，智能成本的大幅骤降有望催生新型疗法并彻底改变预防性医疗。 全球每周有超过3亿人使用ChatGPT咨询健康与保健问题，其中绝大多数来自美国以外地区，且通过免费层级访问服务。 (0:00) OpenAI将医疗智能直接整合到通用前沿模型中，而非开发垂直医疗模型，从而避免了特定领域模型因脱离通用推理扩展曲线而面临的过时风险。 (6:40) OpenAI推出了GPT-6 Luna，在HealthBench Professional临床医生基准测试中超越了前代旗舰5.6 Sol，同时API成本降低至后者的三十四分之一。 (42:40) OpenAI已部署了直接集成至电子病历（EHR）的功能，包括Epic侧边栏Copilot，使临床医生能够在不中断现有工作流的情况下，执行自动化病历审查并获取AI指导。 (26:45)

OpenAI health lead Karan Singhal argues that conversational and agentic artificial intelligence can bridge severe structural gaps in healthcare by serving as an ubiquitous, proactive backstop for patients and clinicians without replacing licensed medical practitioners. Unified Frontier Architecture Outperforms Vertical Models (6:40) Singhal rejects the idea of training isolated, vertical healthcare models, contending that domain-specific forks inevitably fall behind the rapid reasoning leaps driven by frontier scaling. Instead, OpenAI trains medical knowledge directly into general foundation models by incorporating global physician datasets and clinical evaluations across pre-training, post-training, and reinforcement learning. On OpenAI's clinician-oriented HealthBench Professional benchmark, its lowest-cost API model, GPT-6 Luna, surpassed the prior flagship model, 5.6 Sol, despite being 34 times cheaper. Calibrating Personas, Safety, and Medical Epistemics (14:45) Because healthcare relies heavily on trust, models are trained with feedback from an international network of hundreds of doctors to modulate outputs across three distinct user types: lay consumers, expert lay users managing chronic diseases, and licensed clinicians. Singhal notes that clinical safety overrides consumer satisfaction metrics, requiring models to issue urgent warnings even when users prefer reassurance. To address scientific uncertainty and compress the historical 5-to-17-year delay in translating clinical guidelines to bedside medicine, OpenAI uses its HealthBench and Mental Health Bench frameworks to evaluate appropriate medical hedging and non-crisis emotional support. Bridging Healthcare Fragmentation Through Computer-Use Agents (12:35) Rather than vertically integrating care delivery, OpenAI positions its systems as ambient software "quarterbacks" that navigate fragmented administrative workflows, clinical scheduling, and chart reviews. To overcome legacy software bottlenecks and cloud virtual machine deployments where modern API connectors do not yet exist, Singhal highlights computer-use agents that interact directly with graphical interfaces. Early deployments, including an Epic Systems electronic health record sidebar and enterprise implementations at AdventHealth, demonstrate how AI can synthesize complex records without requiring clinicians to juggle disparate applications. Longitudinal Insights and Early Preventative Interventions (40:00) Singhal anticipates that combining longitudinal electronic health records with conversational chat logs will reveal non-obvious diagnostic connections—analogous to AlphaGo's counterintuitive "Move 37"—such as flagging global antimicrobial resistance risks or elevated blood biomarkers like HbA1c. In the near term, Singhal expects AI's primary macroeconomic contribution will come from precision targeting and preventative timing to curb costly downstream hospitalizations, whereas autonomous medical robotics remains a distant milestone constrained by safety validation and reimbursement models. While Singhal projects that the convergence of computational drug design and AI-optimized clinical trials could deliver tailored therapeutics within 5 to 10 years, macro-level cost deflation across the $5.3 trillion healthcare sector remains contingent on resolving administrative friction, such as algorithmic billing disputes between payers and providers.

OpenAI健康业务负责人Karan Singhal指出，全球医疗支出已超5.3万亿美元却普遍存在资源稀缺，OpenAI正通过将专业医学训练直接融入通用前沿大模型，扮演健康领域的“四分卫”，在不取代医生执业的前提下实现降低医疗门槛与突破能力上限的双重目标。 托底与登顶：前沿智能如何破解医疗稀缺 (0:00) 全球每年医疗健康支出超过5.3万亿美元，若视作独立经济体其体量相当于世界第三大GDP，但体系内仍面临严重的医疗摩擦，例如美国有超1亿人缺乏固定的初级保健医生，平均专科候诊时间超过40天。Singhal解释，OpenAI的医疗使命围绕“托底”（Raise the floor）与“登顶”（Raise the ceiling）展开：一方面每周已有超过3亿活跃用户通过ChatGPT获取健康与日常保健信息，大模型能在就医前后填补15分钟匆忙面诊留下的认知空白；另一方面则追求此前无法实现的新能力与前沿科学突破。 技术路线之争：为什么通用模型优于专用模型 (6:40) 针对医疗AI应走垂类专用路线还是通用路线的疑问，曾主导研发专用医疗模型的Singhal指出这是伪二元对立。孤立的垂类小模型会因为错过前沿通用底座在推理、常识和复杂问题求解上的指数级迭代而迅速过时。OpenAI选择在通用基础模型的预训练、后训练对齐以及强化学习各个阶段，全面注入由全球数百名执业医师提供的高质量临床数据。通过与全球专家协作并开发HealthBench等评测工具，模型既能借力前沿规模效应，又能针对普通大众、长期病患和资深临床专家精准调节沟通术语与逻辑深度。 安全底线与消除17年临床指南落地时滞 (20:00) 为了在涉及重大健康的场景中防范谄媚倾向，OpenAI组建了跨越数十个国家与学科的全球医生网络，确立以临床安全性为最高准则；即便用户寻求安抚，模型在识别出高危征兆时也会强制要求线下就医。此外，医学研究成果及临床指南从发表到全面融入常规医疗实践通常存在5至17年的严重滞后。通过开发面向医疗专业人员的专用工具并在Epic等主流电子病历（EHR）系统中嵌入原生侧边栏，OpenAI试图把这一知识转化鸿沟从数十年压缩到数天甚至几分钟的面诊窗口内。 从被动问答到主动助手：医疗的“第37手” (30:00) Singhal明确表示OpenAI无意自建实体医疗机构替代执业医师，而是定位为整个医疗生态的协调者与“四分卫”。通过安全隔离健康数据并利用计算机使用（Computer Use）技术绕过陈旧孤立的虚拟机系统，AI能充当纵向主动健康管家，在幕后核对处方交互风险、监测长期指标（如糖化血红蛋白上升）并提醒复查。类比AlphaGo击败李世石时下出的划时代“第37手”，Singhal认为前沿AI整合长达数年的病历与对话记录后，有能力预判出超越传统人类认知、却在事后被证明极其深邃的早期健康风险与盲区。 边际成本递减与药物研发、临床试验长远展望 (42:40) 在宏观经济层面，Singhal列举了API成本骤降的实例：最新推出的GPT-6 Luna在极具难度的HealthBench Professional评测中表现超越了此前顶尖的5.6 Sol，而调用成本降低了34倍。针对医保控费与机构间利用AI自动报销对轰导致的行政资源浪费，Singhal认为诊室环境下的环境智能（Ambient Intelligence）能提供客观的实时记录，从而在长远上消除摩擦。至于手术和物理层面的医疗机器人自动化，他认为受制于技术与支付监管框架，短期数年内难成主流；未来5至10年最具变革性的突破将是AI深度介入全新生物靶点与药物分子的生成，以及针对人体临床试验的队列智能匹配与流程优化。 当前低成本智能尚未在宏观层面扭转高昂的医疗支出曲线，真正的结构性节约仍取决于将AI深度嵌入慢性病管理与行政审批流程；同时，面对缺少绝对共识的医学前沿，如何在系统内部维持认识论上的谦逊与严谨对齐，仍是持续演进的核心挑战。

## Background 本期播客由 a16z 合伙人 Jay Rughani 主持并于 2026 年 10 月 20 日发布在网页与 RSS 平台上，特邀嘉宾为 OpenAI 医疗 AI 研究负责人 Karan Singhal。根据页面介绍，每周有超过 3 亿人使用 ChatGPT 咨询健康与养生问题，节目深入探讨了 OpenAI 为何将医疗能力直接注入通用前沿大模型、如何携手数百位医学专家保障安全，以及 AI 在整合电子病历、辅助医生临床决策和降低医疗智能成本方面的长远愿景。 ## TL;DR 对话系统拆解了 OpenAI 在医疗健康领域的战略布局：坚持走“通用前沿大模型深度融入专业临床训练”的路线，而非构建容易脱离前沿梯队的独立小模型。通过每周触达 3 亿用户的免费分发机制，AI 致力于成为患者两次门诊间隙的主动型健康向导（Quarterback），并在幕后充当临床医生的安全网（Safety Net），有效消除抗生素滥用或用药冲突等疏漏。面对极其割裂的医疗系统，具备“计算机使用”能力的智能体正在跨越传统软件与虚拟机壁垒，而诸如与 Epic 的深度集成则消除了工作流摩擦。此外，针对大模型推理成本的断崖式下跌（如 GPT-6 Luna 相比前代降本 34 倍），双方深入探讨了医疗智能大宗商品化对现有医保报销博弈的重塑，以及 AI 加速临床试验和新药研发等长期突破潜力。 ## Atmosphere & style 整场对话洋溢着前沿科技从业者特有的理性务实与深切的人文关怀，语调沉稳、严谨且充满启发性。Jay Rughani 作为投资人展现出敏锐的行业洞察，频繁抛出诸如医患利益冲突、医疗“打包”属性及商业对抗等尖锐现实问题；而作为研究负责人的 Karan Singhal 则以极其客观的科研逻辑从容拆解，既不盲目鼓吹颠覆人类医生，也不回避算法不确定性带来的安全挑战。聆听全程仿佛置身于一场高密度的闭门思想研讨，从身患白血病儿童的真实求医困境，逐步跃升至AlphaGo“第 37 手”式的超人类直觉洞见，兼具宏大科技愿景的激荡感与直面医疗顽疾的沉静力量。 ## Bulletin - `00:00` 全球医疗年支出超 5.3 万亿美元，但仍面临严重资源匮乏；目前每周有超过 3 亿人使用 ChatGPT 咨询健康与养生问题，其中绝大多数为免费用户。 - `05:00` OpenAI 放弃独立垂直医疗小模型路线，选择在通用前沿底座的每一训练阶段深度注入全球医生团队的专业临床数据，以借力 Scaling Laws 的快速迭代。 - `43:24` OpenAI 推出的 GPT-6 Luna 在高难度临床工作流基准 HealthBench Professional 上超越了上代顶级模型，同时将 API 调用成本大幅降低 34 倍。 - `20:00` 临床指南从发表到全面落地平均需耗时 5 至 17 年，AI 的核心价值在于将这一转化周期压缩至数周甚至数小时，并无缝嵌入 Epic 等核心电子病历系统。 ## Key takeaways - `00:00` 医疗健康是 OpenAI 兑现 AGI 使命的关键抓手，整体战略聚焦于“提升基准底线”（提供近乎免费的可信医疗智能）与“拓宽能力上限”（开启全新医疗可能）。 - `05:00` 相比独立医疗小模型容易脱离前沿梯队，将医疗专业数据与评估体系直接注入通用前沿大模型，能最大化吸收底层推理与规模扩展红利。 - `08:30` 用户接入的个人健康医疗信息设立了严格的安全隔离，严禁用于商业广告定向推送或后续基础模型的通用预训练。 - `10:00` AI 定位为嵌入全流程的医疗“安全网”，在幕后提供第二诊疗意见、预警药物冲突，而非阻断患者与真实医生的连接。 - `11:58` 医疗系统高度碎片化且缺乏现代 API，具备“计算机使用”（Computer Use）能力的 Agent 可以直接操作虚拟机上的遗留 EHR，打通挂号、理赔与随访流程。 - `14:48` 模型采用动态语言适配机制，既能用浅显大白话为普通患者拆解指标，也能以高术语密度配合专科医生进行深度临床推理。 - `16:32` 在用户心理安慰诉求与临床客观风险冲突时，模型评估与训练栈设定了明确底线：即便患者抗拒，也必须坚定预警潜在风险并建议就医。 - `20:00` 面对医学事实的不确定性，OpenAI 引入 HealthBench 评估模型的自知之明与“对冲表达”（Hedging）能力，避免无共识下的武断作答。 - `25:00` 通过支持 HIPAA 合规并在 Epic 界面内嵌侧边栏 ChatGPT，OpenAI 旨在消除数据孤岛并大幅降低临床医生采纳 AI 的“激活能”。 - `26:50` Mental Health Bench 将心理健康评估从防范自残等“安全底线”扩展到衡量模型能否主动提供积极的心理干预与支持。 - `30:00` 健康信息咨询与法定医疗护理（处方权、触诊、法律责任）之间界限分明，OpenAI 定位为全流程协调的“四分卫”，将医疗交付的最后一英里赋能给合作伙伴。 - `32:38` 随着认知窗口转移，患者在就诊前核对 AI、医生在门诊借助 ChatGPT 复核病历，正促成医患双方对诊疗决策产生双向安全感。 - `38:08` 类似 AlphaGo 的“第 37 手”，AI 凭借对长达十年跨维度纵向病史的整合分析，能提出反人类直觉但事后证明极具远见的诊断预测。 - `47:04` 当前医保与医疗机构之间使用 Bot 进行相互博弈的“军备竞赛”属于短期内耗，未来将由诊室环境智能采集并直接生成标准化病历，实现自动裁决清算。 - `47:04` 对医疗创业者的核心建议是“押注底层模型会变得更强大”，切忌围绕当前模型的暂时缺陷构筑护城河，应深耕高质量护理交付闭环。 - `53:30` 未来 5 至 10 年 AI 对医疗的最大颠覆将体现在新药研发：不仅在于生成新靶点和分子，更在于打通临床试验候选药物筛选与验证的吞吐瓶颈。 ## Detailed walkthrough ### `00:00` 医疗系统的稀缺性与 OpenAI 的医疗使命 Jay Rughani 开篇指出，全球在医疗健康领域的支出高达 5.3 万亿美元以上；如果将其视为一个独立国家，它将成为全球第三大经济体（按 GDP 计算）。然而，在如此庞大的支出规模之下，医疗领域依然面临着极其严重的资源匮乏。面对这一现状，OpenAI 的使命是确保通用人工智能（AGI）造福全人类，并将健康领域视为兑现这一愿景的关键支柱。 ```mermaid flowchart TD accTitle: OpenAI AGI Mission and Health Pillars accDescr: 展现 OpenAI 推进 AGI 的三大方向以及医疗在其中扮演的双重角色 subgraph Mission["OpenAI 推进 AGI 的三大方向"] D1["构建与部署 AGI<br/>（高能力模型、产品与生态伙伴）"] D2["防范负面安全风险<br/>（Downside Safety）"] D3["创造正面价值<br/>（Making Upside Happen）"] end subgraph HealthStrategy["医疗领域的战略落脚点"] H1["提升基准底线（Raise the Floor）<br/>提供免费或极低成本的可信医疗智能"] H2["拓宽能力上限（Raise the Ceiling）<br/>开启此前无法实现的新医疗可能"] end D3 --> HealthStrategy D2 -.->|"解决医疗中的关键安全问题"| HealthStrategy ``` OpenAI 医疗 AI 研究负责人 Karan Singhal 解释说，公司从三个维度来推进 AGI 战略： - **构建与部署 AGI**：打造前沿模型、开发配套产品并推进合作伙伴生态。 - **防范负面安全风险**：在模型行为的各维度设立严密的安全防护网。 - **创造正面价值**：医疗是实现这一目标最具体、最切实的抓手，同时也能够顺带攻克医疗应用场景中特有的安全难题。 针对如何实现这一医疗目标，Karan Singhal 明确总结道： > “One is that we can raise the floor of human health and give people access to trustworthy, abundant, reliable health intelligence for free or nearly free, and really raise the ceiling as well and make things possible that weren't possible before.” 在“提升基准底线”方面，最直观的体现便是消费级产品的普及。目前每周有超过 3 亿人使用 ChatGPT 咨询健康与养生类问题，且这一群体仍在快速扩大。即便在美国或医疗资源密集的旧金山湾区，想要预约专科医生往往也需要等待数周甚至数月；好不容易排上号，患者与医生面对面的时间通常只有短短 15 分钟。在这仓促的问诊中，患者往往准备不足，离开诊室后才想起还有五六个问题没来得及问。OpenAI 希望 AI 能介入这两次就诊之间的漫长空白期，作为患者全流程照护中的向导，协助他们梳理问题、理解医嘱并在就诊间隙持续获取可靠支持。 --- ### `05:00` 医疗现实痛点、白血病真实案例与“通用底座+专业训练”架构 现实中的医疗短缺不仅体现在专科就诊上。Jay Rughani 补充了一组数据：全美有超过 1 亿人缺乏固定的初级保健医生（PCP）；即便有家庭医生，预约往往也需要等待 40 天以上。Jay 分享了自身的亲身经历：他上周看皮肤科医生前特意在 Apple Notes 里列出了所有疑问，但 15 分钟的问诊结束后依然遗漏了问题，而下一次复诊甚至已经排到了 4 个月之后。 为了具象化 AI 对现实生命的改变，Karan Singhal 分享了《The Cognitive Revolution》播客主持人 Nathan Labenz 的真实经历。当时 Nathan 带着年幼的儿子在外度假，孩子突然表现出极度疲劳与烦躁不安。在度假期间，他们借助 AI 分析症状，AI 反复提示他们情况非常严重、必须立刻前往医院排查。最终，孩子被确诊为一种侵袭性极强的白血病。在随后的整个诊疗旅程中，面对晦涩复杂的化验单，甚至在专科医生尚未抽出时间解释之前，Nathan 都在频繁使用当时的 GPT-5 Pro 模型协助解读报告和梳理流程，为整个家庭争取了宝贵的处置时机并带来了巨大的心理宽慰。 针对“为什么不用专用的医疗小模型，而是用写代码、做旅游攻略的通用模型来做医疗”的架构疑问，曾亲自研发过专用医疗模型的 Karan Singhal 从底层逻辑给出了明确解答： ```mermaid flowchart LR accTitle: Specialized Models vs Specialized Training in General Models accDescr: 对比传统独立专用医疗模型与通用前沿大模型融合医疗专业训练的演进路径差异 subgraph ForkPath["路径 A：独立专用小模型（旧范式）"] S1["分叉出专用小模型"] --> S2["前期在医疗基准上有优势"] S2 --> S3["受限于算力与数据规模<br/>快速脱离 Frontier 梯队"] end subgraph FrontierPath["路径 B：通用大模型融合专业训练（OpenAI 路线）"] F1["前沿通用基座模型<br/>（规模扩展与通用推理）"] T1["全流程注入全球医生专业数据"] F1 & T1 --> F2["兼具前沿推理能力与深厚临床专业性"] end ``` Karan Singhal 指出，行业曾尝试构建垂直专用小模型，但往往很快就会被通用前沿大模型甩在身后，因为前沿通用模型的底层推理突破和规模定律（Scaling Laws）演进速度太快。OpenAI 选择的是在通用前沿模型的每一个训练阶段都深度注入全球医生团队提供的专业医疗数据与评估体系： > “If you kind of forked and said, hey, we're going to train a kind of smaller specialized model that is just for healthcare, like I used to and like many people have, what you'd find is that model would just kind of fall behind the frontier relatively quickly, because the pace of the frontier and the general model improvements is so fast.” --- ### `08:30` 口袋里的全科健康助理与主动式照护愿景 针对未来人手一个“装在口袋里的私人全科医生”的设想——即拥有全科专科知识、完全个性化、记住病史与偏好、并在最佳时机提供健康行为干预的智能体——Karan Singhal 阐述了 OpenAI 的长远定位与边界划分。他明确强调，AI 绝不是为了替代真实的医生或阻隔人们看医生，而是扮演一个全天候、主动式的私人健康助理。 为了实现高度个性化和主动服务，必须确保用户能够放心将其个人健康数据接入模型。Karan 指出，OpenAI 针对个人健康信息设立了极度严苛的隐私安全边界：用户接入的健康数据受到严格保护，绝不会被用于模型训练，也绝不会用于广告定向推送。这构成了构建高度个性化、主动式健康 AI 的核心基石，也指引着技术未来从被动解答走向主动规划与跨机构协同的部署方向。 - **每周使用规模**：超过 3 亿人每周在 ChatGPT 上进行健康与养生相关咨询。 - **初级保健短缺**：全美有超过 1 亿人无法获得常规的初级保健医生服务。 - **专科就诊痛点**：通常问诊时间仅 15 分钟左右，常规复诊等待周期常达数周至数月（如 40 天乃至 4 个月）。 - **隐私承诺边界**：接入的个人健康医疗信息严禁用于广告定位投放或后续的模型通用预训练。 ### `10:00` 免费分发策略与医疗“安全网”愿景 Karan Singhal 阐明了 OpenAI 在医疗能力分发上的核心立场：尽可能广泛地普及给所有用户，而不仅仅面向付费订阅者。每周使用 ChatGPT 咨询健康与养生问题的 3 亿活跃用户中，绝大多数都是免费版用户，且大部分生活在美国之外。OpenAI 将这种“分发广度”视作推进主动型个人健康助手（proactive personal health assistant）时的核心前提。 为了具象化技术带来的改变，Jay Rughani 重提了 Greg Brockman 此前在节目中分享的一个真实案例：Greg 的一位朋友在住院治疗时，医生正准备为她注射某种药物；她在 ChatGPT 中核对该药物并结合自己的病史进行比对，发现该药物可能会引发严重的不良反应。当她将这一提示给医生看时，医生当场认同并坦言：“我当时没有足够的时间仔细翻阅你的完整病史来注意到这一点，非常感谢你能提出来。”Jay 借此发问：如果 AI 足够廉价且便捷，是否应当让它为每一个临床场景提供第二、第三乃至第四诊疗意见？ Karan 对此极为赞同，他将这种机制构想为一种无缝融入诊疗流程的“安全网”： > “我非常喜欢这种为患者和临床医生构筑‘安全网’的理念，它就在诊疗全程中默默守候，不总是横加干预或充当阻隔各方的中间人，而是真正思考如何增强现有的体验并在其基础上添砖加瓦。” 这种主动式 Agent（proactive agent）不仅能在幕后提示医生潜在的药物相互作用，也能在患者端发挥关键效用。例如，当糖尿病患者的糖化血红蛋白（A1C）出现异常上升时，系统能主动捕捉并标记该信号，提示进行针对性的复查与干预。 --- ### `11:58` 碎片化医疗生态中的自主 Agent 与“计算机使用” ```mermaid flowchart TD accTitle: AI Agents Navigating Fragmented Healthcare Ecosystem accDescr: 流程图展示了个人 AI Agent 如何突破碎片化的遗留系统与虚拟机壁垒，代表患者自主执行预约、医保报销及复核账单等复杂任务。 subgraph User["用户意图与授权"] A["个人目标设定<br/>(挂号 / 医保核查 / 随访)"] B["授权访问凭证<br/>(个人应用 / 信用卡 / 诊疗记录)"] end subgraph Agent["AI 个人 Agent"] C["规划与工具调用<br/>(自主规划 / 计算机操作 Computer Use)"] end subgraph LegacySystem["碎片化医疗生态 (现状)"] D["虚拟机 (VM) 托管的旧版 EHR 系统"] E["缺乏现代 API / MCP 连接器的传统软件"] F["医保理赔与医疗账单系统"] end User --> C C -->|"突破数据孤岛"| LegacySystem LegacySystem -->|"反馈执行结果"| G["患者诊后闭环与错误账单争议协商"] ``` Jay Rughani 指出，诸如 Instinct、Muse、GrokBot 以及 OpenAI 研发的各类智能体，代表了近期崛起的个人 Agent 浪潮。这类应用不仅停留在问答层面，更能够获取用户的系统权限、应用甚至信用卡信息，自主执行一连串复杂任务。若将这一能力投射到复杂而割裂的医疗系统中，Agent 可以替患者预约挂号、确认保险覆盖范围、引导出院后的随访医嘱执行，甚至能替患者与保险公司交涉并修正账单错误。 面对 Jay 询问 OpenAI 是否正朝着这一方向布局，Karan 给予了高度肯定的回应。他坦言，不论是从患者还是医生的角度来看，医疗服务中最棘手的顽疾始终是**系统的极度碎片化与各环节脱节**： > “医疗系统是一个你绝不能指望它明天就能全量接入 MCP（Model Context Protocol）连接器的行业。” 真实医疗场景中充斥着各类遗留软件，很多医疗机构的主力电子病历系统（EHR）甚至至今仍架设在云端的虚拟机（VM）中运行。具有讽刺意味的是，这种基于虚拟机的运行模式反而为具备“计算机使用”（computer use）能力的 AI Agent 提供了绝佳切入点。Karan 认为，传统上阻隔在患者、临床医生与医疗数据之间的层层壁垒，正是由于 Agent 能够直接模拟人类操作界面，未来将迎来大规模的瓦解。 --- ### `14:48` 信任的建立：针对不同用户角色的语言适配 随着讨论深入，Jay 将核心话题转移到了“信任”维度。同一款底层基座模型，既需要为毫无医学背景的普通大众解释化验单指标，又需要辅助受过高强度专业训练的专科医生分析复杂罕见的临床病例。AI 究竟如何根据用户身份（Persona）精准调整回答的呈现方式？ Karan 强调： > “正如你所言，医疗是一个必须‘以信任为速度’前行的领域。” 为了让技术真正落地，OpenAI 不仅需要衡量并提升模型能力，更需要通过以下几个维度协同推进： - **分层人群匹配**：模型必须细致区分完全缺乏医学常识的普通用户、长期患病并积累了大量护理经验的“资深患者”（expert lay user），以及临床一线医生和研究人员。 - **术语密度调节**：精准把控医学黑话与专业术语（technical jargon）的出现频率。对普通用户采用易于理解的大白话拆解，对专业医生则保持高密度的严谨学术语汇。 - **全球专家评测闭环**：通过引入全球多国医疗专家的反馈数据，系统化建立评测基准，持续校验模型在面对不同角色时的叙述方式是否得当，并将这些评估进展主动向公众与专业群体沟通。 --- ### `16:32` 分歧处理：平衡用户心理诉求与绝对临床安全 Jay 抛出了一个极为棘手的产品张力问题：患者在与 ChatGPT 沟通自身症状时，内心往往渴望听到安慰与“一切正常”的保证。从消费级产品的用户净推荐值（NPS）、用户留存与黏性来看，迎合患者并给予心理抚慰似乎更有利；但从严肃医学角度出发，患者此时最需要听到的可能恰恰是敲响警钟的警告。AI 该如何在用户体验与临床严谨性之间取得平衡？ Karan 解释了 OpenAI 在技术路线上的演进。如果仅仅在内部团队招募三五名全职医生来裁定这些充满不确定性的边缘病例，在统计学上是严重失效的：“事实证明，如果你向五位医生询问同一个疑难问题，你很可能会得到八种截然不同的回答。”为此，OpenAI 在立项之初便建立了一个涵盖全球数百名专家的网络，覆盖几十个医学专科、不同语种及执业国家，旨在汇聚群智，避免将安全评测与训练对齐的标准系于少数几个医生的个人经验之上。 当面对用户的心理迎合诉求与客观临床风险发生冲突时，OpenAI 的底线非常坚决： > “在评估栈和训练栈的全流程中，我们设计了专门追踪这类棘手情境的评估机制：即使患者并不想听，或者一心只想寻求某种替代疗法，模型也必须坚定地指出‘这存在安全风险，你应该立刻就医’。” 针对 Jay 提出的另一类现实场景——当患者拿着 ChatGPT 的建议去向主治医生求证，而医生给出了相反意见时该如何应对——Karan 直言，这与传统医疗场景中医生之间的分歧并无二致。他建议患者应当行使正当的知情权，去寻求另一位专业医生的第二诊疗意见（second opinion），无论最初的分歧是来源于另一位医生还是来源于 ChatGPT，寻求独立的临床再评估永远是最合理的路径。 ### `20:00` 医学共识的滞后与模型不确定性处理 在医疗健康领域，信任的核心不仅在于共情，更在于事实上的绝对正确性。主持人 Jay Rughani 提出，医学的大量领域并非像数学那样非黑即白，科学本身是一个持续探索的过程，临床领域也常充斥着分歧与未解之谜，这使得 AI 将回答锚定于“绝对真实”变得极为复杂。OpenAI 的健康负责人 Karan Singhal 分析了当前医学共识落地的现实机制：从科学家发表生物医学论文、提出改良方案，到全球各权威机构发布流程图式的临床指南（Clinical Guidelines），再到临床医生真正将其应用于日常诊疗，往往存在着巨大的转化瓶颈。 学术研究表明，临床指南平均需要 5 到 17 年才能真正渗透进全球基层医生的日常实践中。Karan 强调，没有人希望患者接受的是落后了 5 到 17 年的滞后治疗，而 AI 最大的价值之一就是将这种从前沿共识到临床落地的传播周期压缩到数周、数天甚至数小时。这正是 OpenAI 开发面向医护人员的产品（如 ChatGPT for Healthcare 和 ChatGPT for Clinicians）背后的关键原则，旨在将最新的生物医学文献直接注入 15 分钟的床旁诊疗过程中。至于那些确实存在医学不确定性且尚未达成共识的领域，Karan 强调模型必须学会诚实识别自身的盲区： > “训练模型使其能够识别自身的不确定性。在经过检索和获取信息之后，你希望模型能够明白：‘嘿，这里确实没有足够的信息’，并且能够坦诚地说明这一点。” —— Karan Singhal OpenAI 公开的评测基准 HealthBench 便专门纳入了对模型在面对不确定性时“对冲表达（Hedging）”能力的评测，以避免模型在没有公认结论的情况下给出武断答案。 ```mermaid flowchart TD accTitle: 医学共识传统落地流程与 AI 加速对比 accDescr: 展示生物医学文献如何经过临床指南转化为日常实践，以及传统 5-17 年滞后与 AI 缩短至数小时的对比。 subgraph 传统传导路径["传统医学共识传导路径（耗时 5 - 17 年）"] A["生物医学文献与前沿研究"] --> B["专业机构制定临床指南 (流程图)"] B --> C["日常临床实践与基层医生采纳"] end subgraph AI介入赋能["ChatGPT 介入点诊疗 (Point of Care)"] D["ChatGPT for Clinicians / Healthcare"] -.->|"即时检索与整合前沿文献"| C end ``` - **临床指南滞后周期**：多项权威研究显示，前沿临床指南在全球医生日常诊疗中的实际渗透周期长达 5 至 17 年。 - **床旁诊疗落地**：在典型的 15 分钟门诊会诊中，AI 可即时调用并总结前沿医学证据，将共识传导时间缩短至数小时或数天。 - **不确定性评估**：HealthBench 评估基准中明确包含了对模型自知之明与对冲表述（Hedging）的指标测试。 --- ### `22:30` 服务三大群体与降低“激活能” Jay 指出，OpenAI 在设计医疗产品时似乎始终围绕着三类核心诉求方：患者（普通大众）、签约使用的独立临床医生，以及医疗机构或健康产业企业。针对如何同时满足这些迥异的需求，Karan 将 OpenAI 的策略拆解为三大支柱：第一是让模型本身在专业能力上足够强；第二是构建合适的外围架构或产品形态（Harness）；第三则是推进合作伙伴关系与部署落地（Deployment Motion）。目前全球每周有超过 12 亿人使用 ChatGPT，OpenAI 的战略重心之一就是把处于技术前沿的医疗智能免费开放给庞大的免费层级用户，从底层实现健康智能的丰沛与普惠。 然而，仅有强大的模型是远远不够的。Karan 指出，如果使用该模型改善健康的“激活能（Activation Energy）”过高，用户和机构便不会采取行动。OpenAI 在产品设计上的绝大部分精力，都在于扫除障碍、降低这层激活能： > “你可以拥有一款非常出色的模型，但如果使用该模型来改善健康的激活能太高，人们就不会去用它。” —— Karan Singhal 在面向医院等医疗系统的企业端场景中，最大的阻碍往往是安全与合规。回顾两年前，医疗机构因担忧隐私而无法接入大模型，OpenAI 耗费大量精力投入安全合规建设，成功支持 HIPAA（健康保险可携性与责任法案）合规并构建起医疗级的数据隐私保护屏障。与此同时，团队也逐步学会了“讲医疗体系的行话”，改变了早期技术人员与医疗专家对话时的生疏状态，让各方在建立信任的前提下逐步采纳 AI。 --- ### `25:00` 打破数据孤岛：与 Epic 系统的深度集成 医疗行业长期以来面临的最大顽疾之一便是系统割裂（Fragmentation）。临床数据、诊疗流程彼此脱节，严重阻碍了 AI 产生实际生产力。针对这一痛点，Karan 提到了与全球领先的电子病历（EHR）巨头 Epic 的深度合作。 以往医生若想使用 AI，往往需要在不同的软件系统或屏幕间来回切换、手动复制粘贴患者信息，这种摩擦阻力打消了大多数人的使用意愿。OpenAI 正在打破这种壁垒，让 ChatGPT 直接嵌入 Epic 的核心工作流程之中。 - **工作流侧边栏集成**：在 Epic 的诊疗界面内，医生可以直接通过内嵌的侧边栏呼出并调用 ChatGPT，无需离开日常工作环境。 - **生态连接器（Connectors）**：除 Epic 外，OpenAI 正在开发并陆续发布多种医疗连接器，进一步贯通分散在各系统中的医疗记录与操作节点。 - **消除集成壁垒**：通过将智能无缝整合至现有工作流程，AI 才能从纸面上的技术原型真正转变为临床医生随时可得的助手。 --- ### `26:50` 心理健康基准：从安全底线迈向积极干预 除了生理健康的严谨诊断，涉及心理与精神健康的需求往往更具主观性和复杂性。Jay 提出了一个具体的交互场景：如果用户向系统求助“我希望在接下来的几周里感觉好一点”，由于含义极其宽泛，AI 应当如何理解、引导并采取行动？ Karan 解释称，无论产品载体如何变化，OpenAI 的首要任务是筑牢“安全底线（Safety Floor）”，确保在用户处于危机时刻或极度痛苦时提供严密的安全防护与即时危机干预。然而，仅有防御性措施还远远不够，OpenAI 近期推出了专门的“心理健康基准（Mental Health Bench）”，试图从单纯防守转向衡量 AI 能否主动带来积极的心理支持： > “借助 Mental Health Bench，我们真正采取了一种不仅限于衡量安全底线的方法，而是去思考模型如何能够主动帮助‘拉高上限’——如何提供积极的心理健康体验，从仅仅担忧安全底线，转向衡量这些积极的增益价值。” —— Karan Singhal 在心理健康领域，专家的观点分歧往往比躯体医学更加剧烈。为了构建权威的行业标准，OpenAI 联合了全球数百位持证心理健康专家，针对各类棘手复杂的场景定义模型的理想行为规范，最终凝聚出行业共识。Karan 强调，正如 HealthBench 与 HealthBench Professional 一样，Mental Health Bench 的终极目标是为全行业树立一把统一的度量衡，让所有模型开发者、医疗机构与心理健康专家都能基于同一把“尺子”，共同推动心理健康 AI 向上突破。对话随后转向了“医疗信息（Medical Information）”与真正的“医疗照护（Medical Care）”之间的核心界限。 ### `30:00` 医疗护理的“打包”属性与 AI 的边界定位 面对日常的健康疑问，大语言模型（LLM）正迅速成为更快、更好、更便宜的获取解答与方案的途径。然而，Jay Rughani 指出，一旦单纯的医学信息无法彻底解决问题，用户就会立刻被推入传统医疗体系之中：患者可能需要处方药、需要看专科门诊、要做核磁共振（MRI）或各类扫描复查，或者需要医生进行面对面的身体触诊。这一流程受制于严格的执业医师执照与既有监管体系。Jay 强调，医疗护理本质上是一个“打包产品（bundled product）”——它不仅包含诊断、分诊和护理方案，还涵盖医生的同理心、心理抚慰、信任支撑，以及医疗团队所承担的法律责任。 针对“AI 是否会提供这全部服务”的提问，Karan Singhal 明确表态：“我并不认为 AI 能够或将会提供这一切。你说的很对，健康信息与实际的医疗护理之间存在着明确界限。”（“I don't think AI can or will provide all of that. And I think you're right that there's this distinction between health information and medical care.”）OpenAI 对自身角色的定位是赋能整个医疗生态，因为未来绝不可能由单一公司来承担所有的医疗救治工作。OpenAI 的目标是理解患者医疗路径中的“最后一英里”，通过模型能力将其大幅简化，“让真正对患者产生影响的最后一英里，在我们的所有合作伙伴眼里变成仅仅剩下最后 100 米”。 ```mermaid flowchart LR accTitle: AI 在医疗生态中的角色定位与边界 accDescr: 展示 AI 从作为健康入口提供信息，到赋能生态伙伴并辅助传统医疗体系的路径划分 subgraph Entry ["OpenAI 核心入口与辅助"] A["ChatGPT 前端入口<br/>(健康问答 / 随身四分卫)"] B["个人智能体 Agent<br/>(日程规划 / 流程协调)"] end subgraph Partners ["生态赋能 (最后100米)"] C["初创企业与医疗机构伙伴<br/>(承接医疗闭环)"] end subgraph CareSystem ["传统医疗体系核心"] D["临床医生面诊 / 触诊"] E["处方权 / 影像检查 (MRI)"] F["法律责任与情感抚慰"] end A --> B B -->|"导流与协作"| C C --> D C --> E C --> F ``` - **健康信息与医疗护理的划分**：AI 胜在低成本、高效率的健康信息咨询与分诊建议，但不能等同于需要承担法定医疗责任的临床医疗实践。 - **健康“四分卫”（Quarterback）角色**：OpenAI 凭借庞大的用户规模扮演着健康前沿入口的角色，未来通过个人智能体（Personal Agents）帮助用户规划日程、整理碎片化物流，但挂号接诊、开具处方等具体事务依然依赖合作伙伴网络。 --- ### `32:38` 面向临床医生的落地合作与工作流重塑 在企业端与临床医生的实际落地中，Karan 分享了 OpenAI 与医疗机构合作的具体成果。针对临床医生的 ChatGPT 产品推出后，用户留存率高得令人惊喜。医生们正在最前线探索该技术的日常用法，其中最具代表性的场景便是病历回溯与图表审查（Chart Review）。以往医生需要频繁在不同系统之间切换标签页，体验极其碎片化，而大模型的引入让整个诊疗工作流变得平滑顺畅。 在健康系统层面，AdventHealth 成为了 OpenAI 在工作空间广泛部署智能体（Agents）的标杆级先锋。Karan 盛赞其取得了令人惊叹的落地成果，认为许多健康系统都应将其视作榜样。当 Jay 问及“如果患者走进诊室，看到医生的电脑屏幕上开着 ChatGPT，究竟该生气还是高兴”时，Karan 观察到社会大众的认知态度正在发生实时变化。 > “在实践中我们发现，尤其是随着时间推移，无论是由临床医生使用它，还是由患者用它来进行某种‘合理性核对’，双方在诊疗全过程中有 AI 的参与实际上都可能会感到更加安全。” —— Karan Singhal（“And so I think what we're finding in practice, especially over time, is that both parties might actually feel safer with AI in their care journey, whether it's their clinician using it or the patient using it as kind of like a sanity check.”） 过去那种医生抗拒患者拿网上信息看病、患者质疑医生依赖工具的对抗心态正在消退。随着认知窗口（Overton window）的转移，AI 逐渐变成了医患双方向彼此验证、防范差错的安全保护垫。 --- ### `35:10` 诊室对话的人性回归与诊疗间期的主动干预 如果患者在就诊前已通过 ChatGPT 做足了功课、整合好了自身数据，而医生也在就诊前通过自动化工具将病历全套跑了一遍，那么面对面的门诊究竟该谈些什么？Karan 认为，这恰恰能将医疗推向最核心的“人性层面”：医学中充满着不确定性，需要床旁关怀（bedside manner）、同理心、情感安慰以及必不可少的体格检查。门诊绝不仅仅是静态信息的比对，获取病史、深入探查以及给予患者心理支撑，是人类不可替代的需求。 谈到两次就诊之间的“诊疗间歇期”，Jay 询问 AI 是否会越界提供主动管理。Karan 澄清，OpenAI 依旧会严格专注于健康信息范畴，但对“主动预防性（Proactive）”的未来充满期待： - **主动指标异常提醒**：例如当患者的化验单显示糖化血红蛋白（HbA1c）偏高、处于前驱糖尿病或糖尿病状态时，AI 能主动提醒其预约心脏科或内分泌科医生进行用药核对。 - **慢性病依从性管理**：及时捕捉并标记患者的用药依从性（Medication Adherence）问题，在慢性病全周期管理中产生巨大影响力。 - **合作生态闭环**：平台负责发现异常与智能提醒，后续的具体疾病管理与处方调整依然交给外部专业团队，而不是由 OpenAI 独家包揽所有业务。 --- ### `38:08` 医疗领域的“第 37 手”（Move 37）：超越人类直觉的洞见 Jay 提到了 AlphaGo 在围棋、国际象棋中下出人类大师从未设想过的全新定式，从而反向拓宽了人类对棋局理解的现象，询问医疗领域是否也会迎来类似的认知突破。Karan 透露，OpenAI 医疗团队近期在“提升人类健康天花板（raising the ceiling of human health）”方面的核心精神动力，正是探寻属于医疗领域的“第 37 手”。 > “所谓‘第 37 手’源自 AlphaGo 在与李世石对弈中的一步棋。在当时那个瞬间，所有观看这场 AI 与世界围棋冠军对决的人类专家都觉得那是一记非常愚蠢、令人意外的臭手。但事后复盘证明，那是一步极度绝妙且充满洞见的下法，如今人类才刚刚开始理解并从中学习其规律。” —— Karan Singhal（“And this idea of a Move 37 is it comes from AlphaGo and a game that the AlphaGo system... played with Lee Sedol, where there was a specific move that just kind of in the moment seemed like a really dumb move and a surprising move for all the human experts watching this game... but it ended up being this remarkably brilliant and insightful move that humans are now just beginning to understand and learn from the patterns of and things like this.”） Karan 指出，OpenAI 正在探索如何在医疗场景中实现这样的跨越——模型给出的预测或临床线索，在当下的医生看来可能完全不符合传统直觉甚至显得突兀，但事后复盘却证明极其精准、极具前瞻性与临床价值。尽管目前该探索仍处于非常早期的阶段，但其作为临床医生的“安全网（Safety Net）”，在医疗交付现场辅助排查隐性风险的应用中，已经展现出带来巨大突破的潜能。 ### `40:00` 突破人类认知极限的纵向数据分析与科研新范式 OpenAI 与 PandaHealth 的合作研究揭示了 AI 在临床纠偏方面的具体能力。在实际临床场景中，医生有时容易过度开具抗生素，而模型在识别这类模式时，会主动提醒避免滥用抗生素，引导医生从关注孤立的局部病例转向关注全球性的抗生素耐药性问题。Karan Singhal 指出，未来更大的突破在于模型能够整合患者长达十年的完整纵向历史数据，涵盖电子病历（EHR）数据、长期聊天交互记录等海量多维信息。人类医生受限于认知带宽，根本无法在大脑中完整融合如此跨度的数据，而模型能在此基础上作出以往看似不可能的健康预测与风险评估，这类突破足以被称为类似围棋“第 37 手”（Move 37）那样的标志性时刻。 针对消费级健康领域中大量缺乏定论或处于“奥弗顿之窗”（Overton window）变动中的健康疑问，AI 正在重塑医学证据的生成方式。Jay Rughani 提出，公众对于许多新兴生活方式与健康习惯充满好奇，但传统医学研究往往缺乏确凿证据。Karan 对此表示，智能体（Agent）最大的价值在于打破现有数据分析的瓶颈。医疗系统中积累了海量的 EHR 记录、长期纵向档案和理赔报销数据，但大多被沉睡封存在各个机构的数据仓库中，未被充分挖掘提炼。借助大模型激活这些存量数据并生成新洞见，将直接推高人类整体健康的上限。 ```mermaid flowchart TD accTitle: 医疗数据与智能体知识生成闭环 accDescr: 展示从存量异构医疗数据经由智能体分析，再向患者、医生及科研人员三方赋能的知识生成流程。 subgraph Inputs["医疗存量异构数据"] D1["电子病历 (EHR)"] D2["十年期纵向病史"] D3["理赔与聊天交互数据"] end subgraph Engine["智能体分析引擎"] AG["AI 医疗智能体"] end subgraph Output["赋能三大群体"] C1["患者 / 用户：早期风险预警"] C2["临床医生：避免抗生素滥用与误诊"] C3["科学家：加速临床验证与科学发现"] end D1 --> AG D2 --> AG D3 --> AG AG --> C1 AG --> C2 AG --> C3 ``` Karan 强调，在传统的“患者/用户”和“临床医生”两类服务对象之外，还存在至关重要的第三类群体——科学家。AI 不仅能提出科学假设，更核心的价值在于加速验证流程中的重重瓶颈。正如 Karan Singhal 所说：“我非常看好这样一个未来：AI 有助于加速验证环节中的诸多瓶颈，我认为这些验证环节最终会成为瓶颈，而不是提出假设本身。” - **临床辅助案例**：在 PandaHealth 合作研究中，模型有效识别了局部抗生素过度处方，提示医生权衡全球耐药性风险。 - **跨维度综合能力**：模型打破了人类认知负荷上限，可无缝融合长达 10 年的 EHR、聊天及个人生活数据，提前预警健康隐患。 - **科研瓶颈转移**：医疗科学的推进重心正从“提出假设”转移至“临床与实验验证”，AI 智能体正成为破解验证延迟的关键工具。 --- ### `43:24` 医疗智能成本骤降：GPT-6 Luna 与 34 倍降本增效 医疗技术长期面临极其沉重的经济包袱。Jay Rughani 给出了一组宏观经济数据：美国每年的医疗保健支出超过 5.3 万亿美元，这一数额若作为一个独立国家计算，将位列全球第三大经济体（仅次于美中）。医疗行业不仅是美国经济体量最大的部门，也是最大的雇主，吸纳了超过 2300 万名从业人员，其通胀增速长期大幅超越整体通胀水平。面对这一沉重现实，业内对 AI 带来的“医疗智能成本趋近于零”寄予厚望。 Karan 解释道，OpenAI 践行其使命的核心路径，就是在推进前沿模型能力极限的同时，让这种前沿能力被全球所有人广泛且低成本地使用。这一理念直接体现在 API 模型的定价与性能突破上。OpenAI 近期推出的 GPT-6 Luna 是其 API 上最便宜的前沿模型。在针对复杂临床工作流的最新高难度评测集 HealthBench Professional 上，GPT-6 Luna 的表现击败了仅在一个月前还是顶尖水准的 5.6 sol 模型，而其调用成本却比后者便宜了整整 34 倍。这种降本曲线的速度，远远超越了历史上任何一项新技术的成本迭代曲线。 Karan 认为，一旦医疗智能达到临床医生与患者可以完全信赖的水平，其边际成本就会如同计算与网络一样被彻底“大宗商品化”（Commoditization）。正如 Karan Singhal 所总结的那样：“OpenAI 的一个核心判断是：在给定的健康智能水平下，其成本将会出现断崖式下跌。”但这并不意味着 5.3 万亿美元的医疗开支会立刻缩减，因为当前巨额开销大量分布在慢性病管理、医疗文书记录、后台行政流转等环节，大模型技术必须深入工作流集成并大规模推广落地，才能真正释放其在经济层面的降本潜力。 --- ### `47:04` 医保与机构的 AI“军备竞赛”及创业者的技术押注 Jay Rughani 提到了近期媒体的热点报道：例如《纽约时报》披露医疗服务提供方（Provider）的 AI 与医保支付方（Payer）的 AI 正在展开“算法对抗”，而蓝十字蓝盾协会（Blue Cross Blue Shield Association）的一项分析显示，双方各自使用的计费机器人与文书处理机器人反而导致行业开支增加了大约 10 亿美元。这种在“AI 降低成本”与“AI 增加内耗开销”之间的荒谬矛盾引发了广泛关注。 对此，Karan 认为这种支付方与提供方之间的 AI 军备竞赛只是短期的阵痛与干扰，长期来看完全没有存在的必要。在未来的诊疗室中，环境智能（Ambient Intelligence）技术将全程在场，自动倾听、转录并直接生成高精度的临床病历细节。所有的诊疗经过、临床决策依据都以客观、高保真的形式留存，供 AI 直接进行审核与报销裁决，只需辅以极少量的争议申诉机制即可。这种透明化的自动裁决系统将彻底消解双方依靠对抗性 Bot 相互扯皮的根基。 ```mermaid flowchart LR accTitle: 环境智能终结医患报销对抗流程 accDescr: 展示环境智能从诊疗现场采集信息，直接生成标准化病历并由 AI 自动裁决，消除传统对抗性博弈。 subgraph Clinic["诊疗现场"] P["医生与患者"] AI_Mic["环境智能感知 / 录音转录"] end subgraph Adjudication["自动化仲裁"] Notes["结构化高保真临床病历"] Judge["AI 自动化报销裁决"] end subgraph Resolution["结果交付"] Pay["直接清算结算"] Appeal["极少数人工申诉流程"] end P --> AI_Mic AI_Mic --> Notes Notes --> Judge Judge --> Pay Judge -.->|"小概率争议"| Appeal ``` 面向正在医疗健康赛道构建应用的创业者，Jay 询问在底层医疗智能成本趋近于零、能力将提升百倍以上的预期下，专业垂类应用应当如何定位。Karan 指出，OpenAI 的大部分医疗影响力并非直接源于其自研的终端产品，而是通过合作伙伴在“交付护理的最后一公里”得以实现。创业者最核心的战略是顺应模型能力的演进趋势。正如 Karan Singhal 给创业者的建议：“我认为许多医疗从业者正处于一种光谱上——要么押注模型不会变好，要么保持中立，要么持消极态度；而我认为，最重要的事情是去押注模型会变得更强大。”创业者应当积极拥抱智能体、计算机操控（Computer-Using Agent）等前沿能力，投身于高质量护理交付与全流程陪伴监控，从而在医疗系统的认知护理产能扩张中构筑真正的壁垒。 - **算法内耗与破局**：针对当前 Provider 与 Payer 各自训练 Bot 相互博弈、反增 10 亿美元成本的怪象，环境智能全流程留痕与自动裁决被视作终局解法。 - **最后一公里战略**：OpenAI 将自身定位为基础能力提供方，将庞大的 5.3 万亿美元医疗体系改造空间交由垂直领域的合作伙伴深耕。 - **核心创业哲学**：在构建医疗专业应用时，切忌基于“现有模型缺陷”构筑静态护城河，而应站在“大模型能力必将更强”的确定性趋势上设计系统。 ### `50:00` AI 对操作性医疗程序的影响与手术机器人的长远展望 医疗支出中有很大一部分流向了程序性/操作性治疗（procedural care），例如影像学检查、血液检测、后续随访检查或外科手术等实质干预。面对 Jay Rughani 关于“AI 医疗智能是否会加剧此类操作性程序的瓶颈，或者 OpenAI 是否会涉足机器人领域”的提问，Karan Singhal 阐述了短期与长期的不同演进路径。在短期内，AI 的核心价值在于更精准地定位和筛选哪些人真正需要接受此类程序，并为患者提供关键信息以判断何时该寻求深入治疗。未来一到两年内最显著的宏观变量将是更优质的前瞻性预防：通过引导患者在早期采取正确行动，直接避免病情恶化为需要昂贵外科手术的严重阶段。 至于机器人手术等更深度的物理交互，Karan 认为这属于更长期的演进方向，不仅取决于底层技术的突破，还受制于医保报销机制（reimbursement）和行业准入与整合的极高门槛。Karan 明确指出：“My expectation is that it's not like a next few years kind of thing.”（我的预期是，这绝不是未来几年内就能迅速普及的事情。） - **短期价值定位**：AI 帮助精准判断程序性治疗的必要性，并推动前瞻性、预防性医疗决策。 - **避免高额下游成本**：如果患者借助 AI 尽早完成正确的健康干预，就能省去许多原本不可避免的高额手术或检查费用。 - **机器人技术瓶颈**：医疗机器人的落地需要兼顾技术成熟度、医保支付政策以及与现有医疗机构流程的长期整合。 --- ### `51:48` 提升底线与抬高上限：ChatGPT Health 的两大演进支柱 Jay 回顾了两人讨论过的模型智能演进、建立用户与医生信任以及从“提供医疗信息”跃迁至“交付医疗服务”等诸多议题，并探讨随着智能体（Agents）和个人助理能力日益成熟，ChatGPT Health 未来究竟会呈现何种面貌。Karan 指出，OpenAI 目前专注于两个层面的战略目标：其一是“提升人类健康的底线”（raising the floor），让高度可靠、值得信赖的健康智能广泛惠及所有临床医生与患者。其终极目标是让 AI 成为诊疗流程中的安全基石，促成双向的信赖感。 ```mermaid flowchart LR accTitle: ChatGPT Health 的双重演进方向 accDescr: 展示提升健康底线与抬高能力上限的两大战略维度及具体实现路径。 subgraph Floor["提升底线 (Raise the Floor)"] F1["普及值得信赖的医疗智能"] --> F2["患者与医生双向信赖 AI 介入"] end subgraph Ceiling["抬高上限 (Raise the Ceiling)"] C1["主动型个人健康助理"] --> C3["预防昂贵的健康恶化"] C2["解锁海量沉睡医疗数据"] --> C4["挖掘全新临床洞察"] end ``` Karan 在阐述这种信任重塑时强调：“And so getting to the point where, like we talked about, all patients and clinicians just feel safer by having AI in the care journey. Like when you see the chat GPT, when the doctor is using chat GPT, you actually feel safer and vice versa.”（因此，我们要达到这样一种状态：正如我们讨论过的那样，所有患者和临床医生在诊疗旅程中有 AI 参与时都会感到更加安全。比如当你看到医生在使用 ChatGPT 时，你反而会觉得更加安心，反之亦然。）实现这种信任不仅依赖卓越的模型本身和严格的评估（evals），还取决于如何部署这些产品并降低临床使用的激活阻力（activation energy）。 第二大支柱则是“抬高能力的上限”（raising the ceiling），即让 AI 实现过去无法企及的能力： 1. **全天候的主动型健康助理（Quarterback）**：深度嵌入用户的健康管理全流程，既能处理预约、行程等繁杂流程，又能敏锐捕捉患者当下的健康风险，在疾病恶化为昂贵的严重问题前实施阻断。 2. **激活沉淀已久的存量数据**：攻克原本封存在各类孤岛或“纸箱”里的碎片化病历与检验记录，释放出足以改善患者预后的全新临床洞察。 --- ### `53:30` 5至10年展望：从 AI 靶点生成到临床试验优化的新药突破 Jay 感叹 ChatGPT 面世仅仅约三年时间，就已经取得了令人震撼而谦逊的飞速进展，在此背景下谈论未来 5 到 10 年的演变甚至带有某种超现实的学术色彩。面对“未来 5 到 10 年内可能会有哪些今天看来近乎科幻的功能变为现实”这一提问，Karan 承认对技术长远走势的精准预测极具挑战，但他认为对人类健康产生最大实质性影响的领域将是新一代疗法（treatments）的诞生，其动力来自于两项关键能力的结合： 1. **AI 驱动的全新靶点与分子生成**：利用 AI 针对疑难病症生成全新的潜在靶点或有前途的候选药物分子。 2. **AI 驱动的候选药物优先级排序与临床试验验证**：医学界想要推进至临床试验的候选药物数量，远远超出了现存临床试验名额的承载上限。AI 能够优化并筛选出最优质的候选方案，让临床试验运转得更加高效和顺畅。 Karan 预测道：“My guess is in that kind of time frame, you're talking about kind of like improved medicines, more personalized medicines for people, things like this, and that will only be possible because you invested in those two pillars.”（我的推测是，在那样的时间跨度里，我们将见证更优质的药物和更加个性化的疗法问世，而这完全归功于我们在这两大支柱上的持续投入。） - **药物研发瓶颈的解决**：不仅在于分子设计的加速，更在于疏通临床试验槽位有限的结构性瓶颈。 - **个性化医疗时代**：结合患者多组学数据与 AI 优化试验流程，将极大缩短新一代疗法从研发到临床落地的周期。 --- ### `54:55` 结语与投资免责声明 在探讨完未来医疗的宏伟愿景后，Jay 表示这场对话让他对未来医疗前景感到前所未有的乐观，并向 Karan Singhal 的分享致以感谢。节目尾声照例向听众给出了渠道关注指引（X 平台及 Substack），并宣读了标准的内容与投资免责声明：播客内容仅作信息参考，不构成法律、商业、税务或投资建议，亦不能作为评估任何证券投资的依据，且 a16z 及其关联机构可能在播客讨论的相关企业中持有投资权益。 ## Notable quotes > `43:24` OpenAI 的一个核心判断是：在给定的健康智能水平下，其成本将会出现断崖式下跌。 — Karan Singhal > `22:30` 你可以拥有一款非常出色的模型，但如果使用该模型来改善健康的激活能太高，人们就不会去用它。 — Karan Singhal > `47:04` 我认为许多医疗从业者正处于一种光谱上——要么押注模型不会变好，要么保持中立，要么持消极态度；而我认为，最重要的事情是去押注模型会变得更强大。 — Karan Singhal > `14:48` 医疗是一个必须‘以信任为速度’前行的领域。 — Karan Singhal > `11:58` 医疗系统是一个你绝不能指望它明天就能全量接入 MCP（Model Context Protocol）连接器的行业。 — Karan Singhal > `40:00` 我非常看好这样一个未来：AI 有助于加速验证环节中的诸多瓶颈，我认为这些验证环节最终会成为瓶颈，而不是提出假设本身。 — Karan Singhal > `20:00` 训练模型使其能够识别自身的不确定性。在经过检索和获取信息之后，你希望模型能够明白：‘嘿，这里确实没有足够的信息’，并且能够坦诚地说明这一点。 — Karan Singhal ## Mentioned OpenAI, ChatGPT, ChatGPT for Healthcare, ChatGPT for Clinicians, ChatGPT Health, GPT-5 Pro, GPT-6 Luna, 5.6 sol, HealthBench, HealthBench Professional, Mental Health Bench, Jay Rughani, Karan Singhal, Nathan Labenz, Greg Brockman, The Cognitive Revolution, Epic, AdventHealth, PandaHealth, Apple Notes, Instinct, Muse, GrokBot, AlphaGo, 李世石, Blue Cross Blue Shield Association, The New York Times, a16z, HIPAA ## Video description More than 300 million people use ChatGPT for health and wellness questions each week. OpenAI’s Karan Singhal thinks that’s only the beginning: the longer-term goal is AI that can help people navigate their health proactively, while making trustworthy medical intelligence dramatically more accessible. Karan joins a16z Partner Jay Rughani to explain why OpenAI trains healthcare expertise directly into its general-purpose frontier models, how it works with hundreds of medical experts to evaluate safety and disagreement, and why the company sees AI as a complement to doctors rather than a replacement for medical care. They also discuss integrations with Epic, AI copilots for clinicians, and what happens as personal agents begin navigating appointments, health records, and other fragmented parts of the healthcare system. Looking further ahead, Karan makes the case that the cost of medical intelligence could continue to fall dramatically, allowing AI to act as a safety net for patients and clinicians, surface insights from years of health data, and potentially accelerate the discovery and validation of new treatments. [Source](https://mgln.ai/e/1344/afp-848985-injected.calisto.simplecastaudio.com/3f86df7b-51c6-4101-88a2-550dba782de8/episodes/e451186f-f0cc-49a5-b14c-c146b2c680bf/audio/128/default.mp3?aid=rss_feed&awCollectionId=3f86df7b-51c6-4101-88a2-550dba782de8&awEpisodeId=e451186f-f0cc-49a5-b14c-c146b2c680bf&feed=JGE3yC0V)

[43:24] Karan Singhal: OpenAI 的一个核心判断是：在给定的健康智能水平下，其成本将会出现断崖式下跌。 [22:30] Karan Singhal: 你可以拥有一款非常出色的模型，但如果使用该模型来改善健康的激活能太高，人们就不会去用它。 [47:04] Karan Singhal: 我认为许多医疗从业者正处于一种光谱上——要么押注模型不会变好，要么保持中立，要么持消极态度；而我认为，最重要的事情是去押注模型会变得更强大。 [14:48] Karan Singhal: 医疗是一个必须‘以信任为速度’前行的领域。 [11:58] Karan Singhal: 医疗系统是一个你绝不能指望它明天就能全量接入 MCP（Model Context Protocol）连接器的行业。 [40:00] Karan Singhal: 我非常看好这样一个未来：AI 有助于加速验证环节中的诸多瓶颈，我认为这些验证环节最终会成为瓶颈，而不是提出假设本身。 [20:00] Karan Singhal: 训练模型使其能够识别自身的不确定性。在经过检索和获取信息之后，你希望模型能够明白：‘嘿，这里确实没有足够的信息’，并且能够坦诚地说明这一点。

[24:00] Karan Singhal: The way we think about this is you can have a model that's really good, but if the activation energy to use the model to improve health is too high, then people won't do it. [47:10] Karan Singhal: And so what I'd say to our partners is build on top of models that are really capable. And the most important thing is bet that those models will become more capable. So bet on the future. [50:00] Karan Singhal: If you make the right moves sooner and AI helps you navigate that kind of journey, then you can hopefully avoid a lot of the downstream costs of the most expensive procedures.

[0:00] 5.3 plus trillion we spend on healthcare. If it were its own country, it'd be the third largest GDP. And yet we have a ton of scarcity. OpenAI's mission is to ensure AGI benefits all of humanity and give people access to trustworthy, abundant, reliable health intelligence. One of the stories that I was most touched by was the story of Nathan LeBenz. He basically had this kind of long saga with his son where at some point he was suffering a lot of fatigue and irritableness and AI helped them navigate the process. And it turns out he was diagnosed with an aggressive form of leukemia. Some of the work that you've done around mental health, there's a lot of complexity. How do you help guide a user on that, given that could mean so many things? We've just been working with a bunch of licensed mental health experts across the world to think about the right ways to define model behavior and a bunch of tricky scenarios and ensure that we can get to a level of consensus where we can actually measure performance. Let's talk about trust. A person can be speaking to an LLM really hoping to hear that everything is okay, but from a medical perspective, they need to hear perhaps something very different. How do you teach an AI to recognize the difference? What happens when that AI becomes proactive, personalized, and increasingly connected to the rest of your health care? In this episode, A16Z partner Jay Rougani sits down with Karan Singhal, who leads health AI research at OpenAI, to talk about the company's vision for making trustworthy health intelligence widely accessible. Karan explains why OpenAI is training medical expertise into its general purpose models, rather than building a separate health care model, how it works with hundreds of physicians to evaluate those systems, and why the goal isn't to replace doctors. Instead, he sees AI becoming a safety net and a quarterback, helping patients understand their health, helping clinicians catch things they might otherwise miss, and eventually navigating more of the fragmented work between visits. They also get into what happens as medical intelligence becomes dramatically cheaper, how AI could shorten the years it can take for new medical knowledge to reach clinical practice, and the possibility of a healthcare Move 37, an insight that initially surprises human experts, but ultimately changes how we understand or treat disease. Hey, everyone, welcome back to the podcast. I'm Jay Rugani, a partner at A16Z, leading investments in healthcare and biotechnology companies. I'm joined by Karin Singhal of OpenAI. He leads health AI research at OpenAI and we got a lot to talk about Thanks for being here Thanks for having me I want to start with the big question There was a report earlier this year you all have talked about it publicly hundreds of millions of people each week use ChatGPT for health-related questions. What are they asking? What are the kinds of questions they're asking that they're not getting from our main healthcare system? Yeah, let me zoom out and talk a little bit about why we're investing in health in the first place. So OpenAI's mission is to ensure AGI benefits all of humanity. And so we see that at a high level happening in three ways. The first is building and deploying AGI. So building capable models, building products around them, deploying them, partnerships. The second is preventing downside safety of various kinds. The third is making upside happen. And so we view health as a really, really concrete way to make upside happen and also work on some of the safety problems that come up in healthcare along the way. And so really, we're really invested in making sure basically two things. One is that we can raise the floor of human health and give people access to trustworthy, abundant, reliable health intelligence for free or nearly free, and really raise the ceiling as well and make things possible that weren't possible before. And so the best way to see the kind of raising the floor is really to think about how we've been working in the consumer space. And so as we reported, there's more than 300 million people weekly using Chachi PT for health and wellness types of questions And we take that as an incredibly important responsibility to get that right And that been rapidly growing over time I think one of the things that you see is just if you think about the experience of getting healthcare, even in the US, even in the Bay Area, if you're looking to get a specialist appointment, even here in the Bay Area, sometimes it can take weeks or months. You might spend a lot of time figuring out how to get an appointment and scheduling and all these kinds of things. And after all that time, maybe get 15 minutes with the doctor where you don't feel adequately prepared and you maybe walk out of it thinking, oh, I had six more questions that I wanted to ask. And our hope is that AI can be part of this care journey with you. And health is not something that is just kind of restricted to these kind of 15 minute intervals. And AI can actually help you navigate your care journey. And so that actually reflects how people are using AI on the consumer side today. People are coming in with questions and concerns, information that they'd like to have. They want to figure out how to navigate their care visits or ask questions or figure out how to follow up things that the doctor told them or things like this. And we're finding that in between these visits is when AI can help people most in the consumer side. And we're making similar investments on the clinician side and on the enterprise side as well. Yeah, it's amazing. We have an extraordinary health care system. If I got really sick, there's no country I'd rather be in than the U.S. And yet we have a ton of scarcity. [5:00] You've seen some of these statistics. Over 100 million people don't have access to a regular primary care doc. Even the ones who do often have to wait 40 plus days to see that doctor. Sometimes you meet the doctor exactly as you said, and it's a quick 15-minute visit, and you have more questions. I saw a dermatologist last week about something. I wrote down all my questions beforehand in an Apple Notes, and I still forgot to ask something. And now I'm scheduling a follow-up, and it's another four months out. It happens to all of us. We've all felt that. Is there a specific patient story or recent case that you all have seen that kind of help us imagine where the world is going and where your work is taking us? One way in which the work of our team is gratifying is just seeing the numbers and seeing how many people are using the tech and how useful it seems to be for them. One other way is just hearing from people, right? So we get people coming in, sending us emails or whatever it is, reaching out in various ways multiple times a week saying, here's how this has helped me. I think one of the stories that I was most touched by was the story of Nathan Levens and his son. So he hosts the Cognitive Revolution podcast and kind of told a story in that podcast. where at some point he was kind of suffering a lot of fatigue and irritableness and things like this, his very young son. And AI helped them kind of navigate the process of realizing that this was like a pretty serious thing while they were on vacation and deciding to take them for testing and repeatedly telling them like, hey, take this seriously. And it turns out he was diagnosed with an aggressive form of leukemia. And throughout his care journey, I mean, Nathan reports kind of using GVD5 Pro at that time quite a bit for helping navigate the process, helping understand lab reports and things like this, even before a specialist had time to kind of help interpret. And I think it brought them a lot of peace of mind. So OpenAI is building health intelligence and health intelligence into your model at every step. But it's the same model that folks use to write code or plan a vacation. Can you help folks understand why you're building it that way from an architectural standpoint? Why is a general purpose AI the right foundation for something as complicated as medicine? Yeah, I mean, people often ask us like this kind of general versus specialized model debate. And I used to work on specialized models and actually in a previous role. And so I definitely have kind of experience on kind of both sides of this question. I think the kind of most important thing here is like what we're talking about here is actually specialized training in a general model, which is, I think, actually the best of both worlds. So what we do for Chats GPT and all the models that we ship in Chats GPT, across every stage of model training for Chats GPT, we actually have specialized health care data added to each phase of training. of our models in the last year. As a result is now you have a model that benefits from all the scale and general improvements in intelligence and reasoning of the big frontier models and also the benefits from the specialized training that we get from, for example, integrating data that we get from working with a bunch of physicians across the globe. And so I think you get the best of both worlds with this approach. If you kind of forked and said, hey, we're going to train a kind of smaller specialized model that is just for healthcare, like I used to and like many people have, what you'd find is that model would just kind of fall behind the frontier relatively quickly. because the pace of the frontier and the general model improvements is so fast. And if you bet on scale, what we found as OpenAI as an industry is that things will get better. And so what we're doing is kind of hitching a ride on that kind of scale frontier and leveraging the fact that we have the ability to kind of integrate evals and training data into the models to make the models both specialize and also benefit from the general capability. I want to almost start imagining what's possible with the technology that you're building. Obviously, where it is today is amazing, but you're not stopping and there's a lot of work to go. I think many people are allured by the idea of a personal doctor in your pocket, something that is perfectly knowledgeable. has knowledge across all medical specialties, so you don't necessarily need to go see a cardiologist and a dermatologist for different questions. Something that is perfectly personalized to you, has all your medical history and context, remembers everything that you've said, remembers your medical history, but also your preferences, and is also able to motivate you and nudge you so that the information that it gives you is given to you at the right time in a way that can nudge you towards better health behaviors or adhering to the protocol that your doctor gave you. Is that sort of the experience you're building at OpenAI in the fullness of time? And how close are we to that? Yeah, that's something that we're really excited about. We are very excited about this vision of kind of a proactive personal health assistant. I don't think that's any kind of replacement for actually having a doctor, having access to a doctor, but really thinking about the ways in which we can take that experience of getting health information and making it more personal to kind of your context is a really important part of kind of our investment in health and chat security, which is kind of consumer offering that we have that enables users to connect their health information for that information to be used kind of securely, not to be used for training data or for targeting ads or anything like that. And that's a big part of how we think about how we're going to move towards this world of really like increasingly personal access and proactive access. It's also been how we've been thinking about deployment. So we've actually from the beginning... [10:00] been deploying this as widely as possible to our free users as well as our pro users. And so we have been Of the kind of 300 million weekly active health users, the vast majority of them are free users. And we take that and most of them are outside the U.S. as well. And so we take that kind of question of distribution here really, really centrally. And so as we're thinking about this kind of next frontier of proactive personal health assistant, we're going to bring that to bear as well. I think use cases help us imagine, help people imagine what's actually possible, what all this work is contributing to. The idea of a personal doctor in your pocket, I think, is exciting. When Greg Brockman came on the podcast, he spoke a little bit about this situation he had with a friend of his was in the hospital and her physician was about to administer a certain medication. She double-checked that medication in ChatGPT, realized based on her medical history that she would have some drug adverse event that may come from that. She showed that to her doctor. Her doctor agreed and said, I didn't have enough time to look through your medical history to notice that. So thank you so much for raising that. Shouldn't we do that for everyone all the time? If it's cheap enough and convenient enough, can't we do a second, third, fourth opinion for every clinical scenario? Is that where we're going? Yeah, I love this idea of a kind of safety net for both patients and clinicians that is just kind of there in the care journey and not always interfering or being the middleman, but like really thinking about the ways in which you can augment the existing experience and build on it. it kind of came in and proactively suggested like hey this might be off like you know this there might be this medication interaction that you might not be might not be tracking can we you know potentially adjust that and things like this i think we're going to see the same thing happen on the patient side and i'm very very excited about this idea of like kind of like a proactive person proactive agent in your care journey really helping you and making sure that like you know if your a1c rises and you're diabetic like that you get like a review of flagged for kind of appropriate review. You mentioned sort of personalized proactive agents. Personal agents in general have kind of emerged on the scene and from a popularity perspective over the last few months. Instinct, Muse, GrokBot, OpenAI's Dots. You can give these AI apps an objective for folks who are not familiar and it can execute a series of tasks. So it's not just, It can access all of your applications that you give it permission to. It can access your credit card information if you give that to them. And it can execute an objective on your behalf. One can't help but imagine the possibilities there in navigating our wonderful but often disjointed healthcare system, scheduling a appointment and making sure your insurance is covered for that appointment, covering that appointment, you know, helping you and guiding you to follow a care protocol that your doctor gave you post-discharge, even negotiating with your insurance company on a medical billing error. healthcare system. Yeah. Is that where you're kind of going with your work? Should we look forward to that from OpenAI? I'm very excited about that vision. I think if you look at, from both the patient and clinician perspective, one of the most difficult things about the healthcare system and how it serves people is this kind of question of like fragmentation and kind of lack of connection of the ecosystem. And healthcare is also one of those industries where you shouldn't expect everything to have an MCP connector tomorrow, right? It is a place where there's legacy software. best connectivity with AI or software in the first place. And many people are, you know, running their primary EHR software in the healthcare ecosystem on like a VM, for example. So like virtual machine that they're running in the cloud. Yeah, exactly. And so there's actually like an interesting component where like a virtual machine actually makes things easier in some ways for AI agents. But this is absolutely ripe for agents and computer use. So I definitely think that this is like a place where a lot of the barriers that have traditionally existed between they need and healthcare data in general, which is often kind of a bottleneck to the whole system working well for patients, I think a lot of those will dissolve. Okay, so we're talking about the use cases. You're helping us dream the dream. Before we keep going down that road, let's talk about trust. Obviously, trust is the most important here. The same underlying model can help me understand my lab results or something that I heard from a doctor, but it can also help [15:00] a highly trained, specialized physician make sense of a very complicated case. What do you change in the answer? so that it's legible to each of those two personas or any other persona that might be engaging with the same underlying model. Yeah. I mean, as you said, health is a place where... you know, it moves to the speed of trust. And so you need to be able to not only make the models better and measure them, but also kind of communicate and bring the world along. And so that's been a very important part of our thinking and our strategy for both how we work with patients and how we work with end users and how we work with clinicians. And so, um, you know, The brief answer to your question is basically we evaluate and train the models by working with experts across the world and getting feedback in various ways to kind of tailor their behavior to whether a person is a lay user or kind of an expert lay user that has been like navigating their care for a while, a complex care journey, or a clinician or other kinds of user personas. of how the model behaved and tailored its level of kind of use of technical jargon and things like this to whether a person was a lay user or a clinician. And since then, we've both gotten better at measuring whether the models are able to cater to different kinds of users and also training the models towards that. And so... A lot of our approach has really been like, not only do, how do we make the models better for these different users, but how do we bring them along? And part of that is how we do evaluations and communicate about them as well. So there's a few components to trust. So we talked about sort of, tailoring the answer to the user. Another big component is disagreement. a person can... be speaking to an LLM or speaking to chat and, you know, be sharing all of their symptoms, but really hoping to hear that everything is okay. So from a consumer user NPS perspective, net promoter score perspective, they want to hear reassurance, but from a medical perspective, they need to hear perhaps something very different. How do you teach an AI to recognize the difference and manage different product objectives there, either from consumer engagement and retention versus sort of correctness and sort of clinical quality? Yeah, this is super central to how we do both evaluation and training. And so zooming out a little bit and the question of like agreement and disagreement around how we respond and how we evaluate. Like one thing you could do and one thing that you could kind of base your approach on here is like we could bring in a few clinicians into, let's say, the opening high health team and say like, hey, you guys figure out how to navigate these kind of uncertain cases. And, you know, we'll ask you about these uncertain cases and you'll help us build evaluations and training data and stuff like this. and leave it to those few physicians to figure it out. This is actually not the best approach. So it turns out if you ask five physicians or something like this, any question, you'll get like eight different answers. And so one of the first things that we did when we first started focusing on health at OpenAI was really build this kind of global physician network of hundreds of experts that we work with kind of across the globe, across dozens of specialties and languages and countries of practice to think about the right ways to kind of aggregate the wisdom of experts so that we weren't relying, disagreement, we weren't relying on one or two or three physicians to kind of adjudicate. So that's one part of it is kind of like thinking about evaluation and training data coming in from different experts rather than relying on one or two people. But then the second part of your question is like, how do you balance kind of competing objectives, right? Let's say the physicians and the aggregate wisdom of the physicians is pointing in one direction. And then maybe the user actually wants you to say something comforting, even when the situation is not comforting. It's basically a prioritization of safety. And so that's, It's very important to us that we take our obligation very seriously when there's hundreds of millions of health conversations coming in. And so across the evaluation stack and training stack, we have evaluations that track these kind of difficult situations where the model should actually go and say, like, hey, this is unsafe, you should seek care, even if the user doesn't want to hear it or wants to get some kind of alternative treatment or something like this. And that's a very important part of how we do evaluation. evaluation working with physicians, but also seeing pretty dramatic model improvements across the board in those areas. There's another instance of disagreement, which is me as the consumer. I ask a medical question of ChatGPT. I then show that response to my doctor during a visit. My doctor loves me, by the way. And then they disagree. They disagree with the response. How should I handle that situation? Yeah, I mean, I think you should handle it similarly as if, you know, disagreement was happening in some other way as well. Like people do disagree about different things. And I think like you have a right as a patient to get, you know, ask for another doctor to take a look and get a sense and go to another doctor to get a second opinion. And yeah, that's what I would advocate for whether or not you're also using chest. [20:00] you need to get health information as well. Yeah. Thank you. Makes a ton of sense. There's another element of trust, which is just absolute correctness. And a lot of the... Thank you. a lot of the work that you all have done It seems to be, at least I'm hearing from you, seems to be about grounding answers in the world's medical literature and evidence. Some parts of medicine, there's not a lot of agreement or disagreement about, but there is much of medicine that's still being discovered or being studied. And there is disagreement even in the medical field. And for that... For those kinds of questions, it's difficult to ground it in truth when we're still discovering truth. Science is a process, etc. How do you think about correctness in those kinds of instances? Yeah, it's a really great question. I mean, it's important with or without AI. So if you kind of zoom out and look at like, how is medical consensus distributed today? Right? Like, what is the process of that today? You basically have a bunch of people writing papers, biomedical papers of various kinds, suggesting kind of improvements or treatments or things like this that could improve the quality of care. if you're a cardiologist and you see a patient with X, Y, and Z, here's what you should probably do. And so these clinical guidelines are roughly, you can consider them kind of flow charts that are put out each year by the different kinds of bodies across the world that describe like roughly what you should do in different scenarios based on the latest research. And then the final step is, so you have the biomedical literature, and then you have the clinical guidelines, and then you have like, what do people actually do in practice? And so, The final step, it turns out, is a massive bottleneck. So there are these studies on how long does it take for clinical guidelines to penetrate the practice of everyday doctors across the world. And in the past, people have found that this takes, depending on the study, like 5 to 17 years for that penetration to actually happen, which is massive. You don't want to be practicing medicine 5 to 17 years behind. down to weeks or days or even hours, right? And so this is one very important principle behind our work in Chachapiti for healthcare and Chachapiti for clinicians, which are clinician and healthcare professional facing offerings where we're bringing the latest biomedical literature in the moment of care and allowing people to very quickly in the course of like, let's say a 15 minute visit at the point of care, get answers to their questions. So that's one way of answering a question, which is like, you know, in practice, you know, while there is medical, a lack of medical consensus in many areas, census, but it's not being practiced. And so one thing you can do is really dramatically reduce the amount that's happening. This is one way of answering your question. A second way of answering your question is how to handle the quick places where there's still genuine medical uncertainty. And there, I mean, there's a couple things. So one is you want to train the modelist to be able to recognize their own uncertainty. After After browsing and getting information and things like this, you want the models to be able to know, hey, there just isn't enough information and to be upfront about that. And so, you know, that's been an important part of how we've done evaluation. That's actually like a public aspect of HealthBench, which includes kind of evaluations on hedging, for example, when there is this kind of uncertainty. So that's one important thing. The second thing, I mean, this kind of goes to this question of like, how do you raise the ceiling of human health is like. really thinking about the ways that models can actually accelerate the process of doing science and actually pushing the medical frontier. And so maybe we'll chat about that later. Hearing you talk, it feels like you're always thinking about one of three constituents. It's either the patient, it might be the individual clinician that signed up, or it might be the enterprise, whether it's a healthcare provider enterprise or some other enterprise. How do you think about designing the product at the meet the needs of all these different users. Yeah, I mean, I think maybe there's three parts to that. One is making the models really, really good for these users. The second is building the right harness or product around it. And the third is really like the deployment or partnership motion to kind of reach out to the right, reach out to the users. And like, by default, I think a lot of the time, the models are often pretty good these days. And so we talked about our approach of evaluation, working with clinicians and things like this, we are seeing pretty dramatic improvements and kind of performance and reliability. tier and bringing free tier performance to the frontier. So, you know, of the 1.2 billion people who use ChachiPT weekly, I'm giving them a model for health that is just like incredibly useful at the frontier for free. So that's been a big part of our strategy, obviously, for making healthcare, health intelligence abundant. The second thing is our kind of product offerings. And so, The way we think about this is you can have a model that's really good, but if the activation energy to use the model to improve health is too high, then people won't do it. So a lot of our product efforts are really thinking about it. [25:00] about these constituents that you're talking about and thinking about the ways in which we can reduce the activation energy for people to use compute and intelligence to improve human health. And so... That goes along the lines of our work in health and Chachi PT, Chachi PT for healthcare on the enterprise side and Chachi PT for clinicians. A lot of the way we design these offerings is basically to think about what are the barriers and bottlenecks and activation energies for people to be able to use the models. And so, for example, Chachi PT. you know, what are the security and compliance bottlenecks in providing an enterprise offering of ChashyBT to healthcare systems? So that was like one question that we were thinking about a lot two years ago. And so we thought, okay, we need to kind of invest in security and compliance. We need to make sure we're able to offer HIPAA compliance. We need to make sure that we're able to offer the kinds of privacy protections that healthcare organizations need. So that we're able to handle the healthcare data that they need and be able to offer services that they trust. like roughly two years ago to like roughly a year ago. And we kind of just go after there, from there, like based on the bottlenecks that we see to kind of adoption. The third thing is kind of partnerships and deployment. And again, it's really like, you know, a few years ago, I think opening, I was very early to speaking the language of healthcare. And when we interacted with healthcare partners, I think we were early in those conversations. So I think one of the things that we've brought to bear since then is really, you know, thinking about each of those how to speak their language and how to kind of bring them along this journey. Because even if the models get really smart, if nobody's like really trusting them or using them, that's not really achieving our mission. And is that you're alluding to a little bit some of the work you've done with Epic to kind of integrate the product directly into Epic? Maybe share a little bit more about that, because a lot of the folks who spent a lot of time in health care were very interested to see that level of collaboration with with Epic. Yeah, I mean, this is a very important part of our approach. Like we talked about, fragmentation in healthcare has been one of the issues. And the fact that there is kind of data and workflows that are disconnected from each other, I think, is a big obstacle to the use of AI productively in healthcare. workflow itself, you can have like a little sidebar where you can actually use ChatGPT as well. And so that's one thing. We're also working on like different other kinds of connectors that we've also announced recently and we have more coming. And so there is quite a bit of work that we're doing to make sure that we are kind of breaking down these barriers, the kind of healthcare ecosystem being connected. And I think Epic is just one of many examples where we're identifying those bottlenecks and trying to fix them. The last thing I want to hit on with respect to trust And this is another area where there's a lot of complexity and there may be other objectives that you need to solve for beyond the clinical complexity that we talked about. If I ask my OpenAI dot, hey, I want to feel better over the next few weeks. How do you even interpret that question? How do you help act on that? How do you help guide a user on that, given that that could mean so many things? Yeah, I mean, regardless of the surface, we are super committed to like helping users when they're having a difficult time or when they're in crisis. And so, for example, for DOTS, we've been investing the same kinds of safeguards that we've been putting in all of our product surfaces in that product surface as well. So that's an important part of our approach is kind of like a baseline safety floor. And then, you know, with Mental Health Bench, we've really taken this approach of not just measuring the safety floor, but thinking about the ways in which the models can actually like proactively help kind of raise the ceiling. help provide positive mental health experiences and move from just worrying about the safety floor to measuring those kinds of upsides. And so again, the approach is actually very much mirroring our work on HealthBench, where we've just been working with a bunch of licensed mental health experts across the world to think about the right ways to define model behavior in a bunch of tricky scenarios and ensure that we can get to a level of consensus. People definitely disagree even more here than physical health. we can actually measure performance and create this kind of like shared benchmark for the industry. And so we released that benchmark very recently. And similar to HealthBench and HealthBench Professional, the goal has really been, can we create this benchmark that kind of raises the ceiling for everybody and kind of creates across all model developers and the healthcare industry and mental health experts like this shared ruler or measuring stick that we can use to kind of make sure that we're moving in the right direction. Okay, so we've spoken about intelligence and the use cases that are coming around intelligence approving. We've spoken about trust, which is critical. Let's talk about medical care. And I want to draw the distinction between medical information and medical care. I think for medical information, I think if you have a health concern, if I have a health concern, I'm going to talk about medical information. [30:00] about myself or something that I'm dealing with, chat and other LLMs are increasingly the better, faster, cheaper way to get an answer and a resolution to that question. But if in the scenarios where medical information alone doesn't get me to resolution, I then, at least today, am thrusted into the traditional medical care system. And so if I need to go get a prescription, if I need to go see an in-person specialist, if I need to get some follow-up or some MRI or some other form of scan done. If I need a clinician to examine me in some way, I still need to navigate the traditional healthcare system. And a lot of that work is... sort of governed and regulated by the U.S. medical license and sort of other infrastructure that we have in our healthcare system. It's the multi-turn conversation where it's not just a single question resolution. It's, I got to go talk to a few people. I need to go get a few things done. We need to collect some information. In many ways, medical care is a It's a bundled product. You know, we're getting the diagnosis and triage and the care plan. We're getting the empathy from the clinician, the assurance and the support. We're getting the legal liability associated with that and what they're taking on. Will AI provide all of that? Where will it stop short? Where do you think where do you think this is all going? I don't think AI can or will provide all of that. And I think you're right that there's this distinction between health information and medical care. And so I think the way we see our role here is really to enable the whole ecosystem because there's no future where any one company is delivering all the medical care. Right. And so... Our role here is really to think about what is the last mile of medical care really look like for people? Can we really understand that? And can we make that easier and easier for everybody so that the last mile of actually having impact on patients just looks like the last 100 meters for all of our partners? And so that's kind of how we think about it. It's really making it very easy for our partners to use AI to support everybody in the care journey. I think there is stuff that OpenAI can do relatively uniquely here because of the front door to health that it has. a lot of these kinds of questions or help you navigate this journey or, you know, as we get increasingly helpful personal agents, help you with some of the scheduling and logistics and things like this and just act as a... quarterback for your health. That said, I don't think it'll resolve all these challenges that we'll have, at least not in the short term. And I think to a large extent, we're looking to enable a lot of our partners that are delivering care that are overwhelmingly using our models in various ways to make these operations happen smoother and easier so that patients are not left in this kind of fragmented experience. So as a consumer, should the experience that I look Sort of the quarterback care that you described, will I be talking to a clinician in the interface at any point? Or would you perhaps route me to a doctor and help me find that? Because given that's a whole challenge, just scheduling and making those visits happen, what should I look forward to? I think you'll see, like, we will be... hopefully very helpful in helping you navigate the process of finding care. I don't think we'll be in the business of doing all the things that you mentioned. But that's where we're really relying on all of our startup and care provider organizations and partners and things like this. So, coming back to a question I mentioned earlier, which is you are building that future, but you're building it with partners. You're building it for consumers and patients. You're building it for clinicians. You're building it for enterprises. For clinicians and enterprises, give us some texture on some of the tools that you're launching and some of the collaborations that have really worked. Yeah, we've had a few really exciting partnerships and we've had some really amazing launch partners for Chachi PD for healthcare. We've had lots of users using Chachi PD for clinicians and really enjoying it. It's been the retention on the offering has been like a very pleasant surprise. People seem to be really enjoying it. and they've been really on the front lines of thinking about, for example, clinicians using it for chart review or things like this and thinking about the right ways in which the experience of caring for a patient doesn't need to be this super fragmented. You need to switch tabs or switch experiences and things like this. And so they've been some of our power users on that front. I think another one is AdventHealth, and they've actually been some of our power users in thinking about deploying agents across the workspace, actually across all of OpenAI. They've been a leader. they've been really, they've had amazing results. A lot of health systems, I think, should look to them. I think they're really impressive. If I go into my doctor's office and I see ChatGPT open up, on their screen? Should I be upset? Should I be happy? How do you think people will react to that? [35:00] I hear differing things and I think it's actually changing over time. And so, I think this is an interesting thing, whereas as the Overton window is shifting in real time, I think there might have been some level of concern before. But I think increasingly a lot of, On both sides, there's this question of whether clinicians would push back against patient use of AI and whether patients would push back on clinician use of AI in the way that you're describing. But I think both parties are actually finding it incredibly useful for their own personal purposes. And so I think what we're finding in practice, especially over time, is that both parties might actually feel safer with AI in their care journey, whether it's their clinician using it or the patient using it as kind of like a sanity check. go. Okay, so then taking that one step further, if a patient has done all their research before a visit, And they have all their health data integrated. They've asked a bunch of questions in ChatGPT. And then the clinician before the visit has all automated through loops, has run the entire patient case through ChatGPT. And then the visit happens. What should they talk about? Okay. My hope is that they talk about the human aspect of healthcare. I mean, there's still like a lot of uncertainty. There's a lot of empathy. There's a lot of bedside manner that goes into the human aspects of healthcare. There's also a lot of physical interaction that goes into it. I don't think, you know, I don't think all the information exists for the AI, just information there, theoretically, all the information exists for the AI to just figure out what all should be done. And part of the reason for a visit is to get more information and do the process of history taking investigation and things like this. Even from a cognitive perspective, the elements of kind of empathy and getting comfort and getting that second check, I think will still be something that people demand. Can... What do you think AI in general or let's say Dots would do in between visits? And will you think you'll cross the line into providing care in between visits and then share that back with the doctor? Yeah, I don't think we'll I think we'll we'll we'll definitely be focused on health information. But I think the whether it's Dots or any other product service, I think we're really excited about this kind of future of heading more in this proactive direction that you were outlining. again like this example of like someone's a one someone gets a lap test back whether a1c is higher and they're pre-diabetic or diabetic you know can you flag for them like hey maybe you should get a medication check and go to your cardiologist or whatever and think about the right ways in which you can um flag to users medication adherence issues or other kinds of issues. I think those kinds of issues are places where AI can have potentially a really high health impact, especially in these kind of chronic care journeys. But I don't think it'll be in the business of managing all these things by ourselves. Another thing I'm curious about is in the fields of AI applied to other domains and situations like let's say, games like chess and Go. As AI got really good at chess and go, they also revealed other strategies and ways of playing the game that even the best experts in those domains were not exploiting or not pursuing. And that helped us learn more about these games and also helped us learn more about AI. understanding a clinical scenario that can give us some optimism. Yeah, I mean, I think we've seen this at smaller scales, but let me zoom out and say, like, you know, one thing that's been motivating our team in general recently around raising the ceiling of human health is this notion of, you know, bringing to bear a kind of a Move 37 for health. And this idea of a Move 37 is it comes from AlphaGo and a game that the AlphaGo system, which plays this kind of complicated board game, played with Lisa Dole, where there was a specific move that just kind of in the moment seemed like a really dumb move and a surprising move for all the human experts watching this game between this AI system and this world expert and go. and insightful move that humans are now just beginning to understand and learn from the patterns of and things like this. Yeah. And this is a very interesting example. I think... So this idea of how do you get to a Move 37 in health, of a move or a prediction where the models predict something that a human might not have been able to predict, but in hindsight seems correct and brilliant and insightful and impactful. So I think we're very early, and I think you're going to see much more of this in the future than in the past. But I think, you know, I mean, one kind of example is, again, thinking about the ways in which these systems have been deployed as safety nets to clinicians, where clinicians are actually delivering care. And in the moment, the AI... [40:00] actually flags kind of like, hey, something might be off here, like our study with PandaHealth, and thinking about the patterns of results that we saw there. As an example, you know, in that context, the clinicians were a little bit prone to overuse of antibiotics. And a very common pattern that the models flagged is just kind of like avoiding this overuse of antibiotics and thinking less about the local case and more about the kind of global issue of resistance to antibiotics. So that's like one specific example. But again, I think you're going to see more of these examples of models taking in like a longitudinal history of a user over 10 years across all their data, across the HR data, across the chat data, all these things and making predictions that seemed impossible before because a human can't even integrate all that data in their head. And making predictions about their health and their risks and things like this that seemed like they were impossible. And I think that would rise to the occasion of being called a MOOT 37. Yeah. I mean, there are there's so many things that consumers are curious about that. At least at one point in time, we weren't sure if it was healthy for you. And then to your point, the Overton window changed and there was sort of more evidence that was sort of conclusive in one direction. Sometimes the views on some of these things have changed. All these sort of new emerging phenomenons that perhaps are understudied. Do you anticipate as people are sort of entering this information into ChatGPT or other LLMs that we'll have more evidence that we can generate to study some of these things that folks are curious about, but we don't really know, or we don't have conclusive evidence one way or the other on where they stand. Yeah, I mean, I think one of the most interesting things about agents is they kind of solve a lot of the bottlenecks in analyzing data that already exists, right? There's various places... across the healthcare ecosystem where data is being generated, EHR data, longitudinal histories, claims, whatever, where data is being generated and it is kind of just stored there. And the insights from it are not fully mined or used to generate new knowledge. And so I think one of the frontiers for how health might improve in the future and how the ceiling of health might raise will be thinking about the ways in which we can leverage the data that is just kind of out there in boxes in various places and have models generate new insights based on this data. future. I'm also optimistic about a future where, you know, you mentioned these two constituencies, the user, the patient, and the clinician. I think there is a third constituency, which is the scientist, right, which was helping kind of advance the frontier of knowledge in the space. And I'm also very optimistic about a future in which, you know, AI helps accelerate a lot of the bottlenecks and validation, which I think will end up being the bottlenecks as opposed to kind of proposing all. Switching gears slightly, you've also shared thoughts around just the cost of intelligence as an input to medical intelligence coming way down. And, you know, at some point it'll be too cheap to meter. I spent a lot of time and we as a team spent a lot of time trying to understand how these technologies will make an impact on our health care system that – is inflating well ahead of inflation. 5.3 plus trillion we spend on healthcare. If it were its own country, it'd be the third largest GDP. It's the biggest part of our economy. It's also the biggest employer in our economy. 23 million Americans work in healthcare. There's so much... There's just so much we spend on healthcare. What do you sort of make of the work that you're doing and the impact that it'll have on the cost of care? I think this is one of the most important ways that we can pursue our mission is to not only make our models better at the frontier, but make that widely accessible to everybody. So that's obviously a big part of how we've been able to deploy health intelligence to a lot of people across the world for free. I think it's also a big part of how we think about API and how good our models are there for health and pricing there and stuff like this. One recent update was... we recently released GPT-6 Luna. And GPT-6 Luna is our current cheapest model on the API, or cheapest frontier model on the API. And what we found with GPT-6 Luna was that it was actually doing better on our benchmark HealthBench Professional, our current latest benchmark of clinician workflows that are really difficult, selected for difficulty for our current models across all the different ways in which clinicians are using our models and the models still could use some improvement. it performed better than 5.6 sol, which was literally our best model like a month ago. And so it performs better despite being 34 times cheaper. And we've had a couple of those moments even in the past year. And there's been many trends showing that AI compared to other technologies that have been introduced in the past has been reducing a cost at a given level of capability or usefulness faster than many of those other technologies. And so one bet we've had at OpenAI is that the cost of health intelligence will plummet at a given level of health intelligence. the level of health intelligence where clinicians and [45:00] feel like they can trust in the care journey, they're starting to get there. So that's like a fixed level of intelligence in some sense. Hopefully the bar after that raises over time as well. My expectation is once we get close to that level of intelligence, the cost for that level of intelligence also dramatically reduces like it has been. As the hope there is that we get kind of this commoditization of health intelligence, right? We don't want to hoard that health intelligence. We want everybody to use it. It's my hope that that contributes. But, you know, to answer your question about the whole... of healthcare spending. I don't think all of that is about health intelligence. I don't think like health intelligence has had yet a macro effect on the problem that you're describing. If you look at like the whole of spend of health, there is quite a bit of areas though, where if you look at, for example, chronic care management or documentation or administration, a lot of kind of administrative workflows of various kinds, back office workflows, there's a lot of money spent in ways that probably don't need to be spent that way. over the coming years. I don't think that's happened yet. And I don't think that just happens from making health intelligence abundant. I think that happens from integration and workflows and deployment and scaling that up and really making sure that these models are valuable in the economically valuable tasks that people are using them for. But I think it's a prerequisite to making that future happen. More care, less paperwork is sort of where we're heading. At least there's a lot of the cost savings. I'm sure you saw some of the headlines over the last few weeks. There's a New York Times article that basically provider AIs and payer AIs are fighting. And there was some Blue Cross Blue Shield Association analysis that showed that the respective billing bots and paperwork bots effectively increased spend by a billion dollars or something like that. research on health and trying to bring down the cost of medical intelligence to near zero. And then you see headlines like that. What do you make of that? I think this might be one of those areas where there might be some kind of short-term pain around this kind of provider-payer-arms race, but in the long run, there's no reason this has to exist, right? And so really, I think you're entering a world where a lot of visits actually have some kind of ambient intelligence in the room, listening, transcribing, things like this. There's notes, there's clinical notes, there's AI details of clinical notes and things like this. And so you have a lot of the details you need to adjudicate visits, right? Right, right. And then an AI helps adjudicate that situation. And maybe there's like some limited appeals process or things like this. But I don't think you need to have all this process where you have like an AI arms race between pairs and providers. I think they can just be completely avoided. So I view that as a bit of a short-term distraction. There's a lot of entrepreneurs that are building technology and healthcare that are likely watching this. And they're probably trying to figure out... As you continue to advance your work and you bring down the cost of medical intelligence to near zero, and this becomes this sort of ubiquitous utility that flows through all of our healthcare interactions for consumers, for clinicians, for the administrative part of the healthcare technology or the healthcare stack, what should folks be building on top? What should specialist applications be doing now? assuming that this gets 100x plus better. Thank you. Yeah, it's a great question. Like I mentioned, we think about a lot of our health impact through OpenAI as not being directly through the products that we offer, but actually through our partners and that last mile of delivering care. And so what I'd say to our partners is build on top of models that are really capable. And the most important thing is bet that those models will become more capable. So bet on the future. I think many people in healthcare are on some kind of spectrum between betting against the models, betting neutrally for the models and betting negatively. with the models or betting for the models. And I think I would say, you know, bet on the models getting better. That's probably the most important thing. And think about the ways in which a lot of the technology that's being developed, whether that's, you know, agents or computer users or other kind of frontier tech will accelerate healthcare and bet on that and build towards that future. I also think a lot of medical care will actually like become more accessible to patients with AI. You know, we talked about this idea of kind of a quarterback and dissolving a Just being focused on delivering really good quality care and having AI be part of monitoring that care journey, I think is a very important part of that as well. So zooming out to the overall health care system and now just looking at it at large, we talked about this 5.3 trillion, a lot of that. spend is paperwork between providers and payers and other stakeholders within the five P's of healthcare. A lot of that is cognitive care that you're arguing can be augmented and enhanced and sort of we can bring on a lot more cognitive care capacity online, which is exciting from an access perspective. [50:00] And then there's a lot of spend in healthcare that is procedural, either requiring some imaging, some blood work, some follow-on examination, or it's some surgical intervention. Where do you see sort of the intelligence work that you're doing in the medical space impacting procedural care? And is that gonna be the ultimate challenge Is that going to be the new sort of or an enhanced or perhaps an exacerbated bottleneck? Or do you expect to do a lot of work there as well, robotics, etc.? Yeah, I mean, I think in the short term with procedural care, I think you want a better targeting when that's needed, right? And I think AI can help, you know, provide information that is helpful for knowing when to seek that kind of care as well. Probably the biggest macro factor in the next year or two will just be better kind of proactive, preventative stuff, right? and get that procedure right. If you make the right moves sooner and AI helps you navigate that kind of journey, then you can hopefully avoid a lot of the downstream costs of the most expensive procedures. So that might be like one macro factor to kind of look out for is my guess. I think the question of kind of robotics and care, I think is a bit of a longer term one, both from a technology perspective and kind of like a reimbursement and kind of industry integration perspective. My expectation is that it's not like a next few years kind of thing. - Very cool. Okay, we've covered a lot of ground. We've talked about intelligence improving, we've talked about trust, we've talked about moving from medical information to medical care. help us understand what we can look forward to as things continue to unfold, as agents and personal agents and the underlying models become increasingly performant. What can we look forward to from ChatGPT Health? Yeah, I mean, I think there's like really two high level things that we're really focused on making happen. One is, you know, Again, thinking about the ways in which we can raise the floor of human health and make trustworthy, reliable health intelligence abundant for all clinicians and all users and patients. And so getting to the point where, like we talked about, all patients and clinicians just feel safer by having AI in the care journey. Like when you see the chat GPT, when the doctor is using chat GPT, you actually feel safer and vice versa. Getting to that point, again, that's a question of not just the models being good and evals being good, but also how we deployed and the activation energy around that. the products and things like this. I think a second thing is really thinking about raising the ceiling. So what are the ways in which AI can actually do things that were not possible before? So one of them actually, I think, is this kind of like proactive health, proactive personal health assistant that is part of your care journey, is helping you navigate these moments, being your quarterback, helping you schedule and all these things, but also flagging the places where you can intervene in your health today and prevent a lot of expensive deterioration later in the future. And I think that's going to be an important part how we think about improving the future of health for people, as many people as possible. And I think the second thing is just what you mentioned before, which is really thinking about the ways in which we can unlock the value of existing data that is sometimes stored in boxes of various kinds and thinking about the ways in which that can be leveraged to produce new insights that benefit patients. It is striking and incredible and humbling to realize that, chat GPT, I think came out three years ago, roughly. And so it's been a lot of progress in a very, very short period of time. So to even talk on a five or a 10 year basis feels academic. But can you give us some predictions? I mean, is there a feature that we might have that feels science fiction today, but we could have one day, three, five, ten years from now? Yeah, I mean, it is super, super hard to predict the future these days. So, um, This is a difficult one, but what I would say is I think the thing that might have the most impact on human health in that kind of time frame in the next five to 10 years is probably going to be newer treatments that come out of maybe two things. One is AI-guided generation of new targets or molecules that could be interesting for new treatments. And the second is AI-guided prioritization and validation. people want to suggest to be used in clinical trials and things like this is way more than the number of slots available in clinical trials. So thinking about the ways in which models can help prioritize and run these clinical trials so that they just run smoother and faster. My guess is in that kind of time frame, you're talking about kind of like improved medicines, more personalized medicines for people, things like this, and that will only be possible because you invested in those two pillars. That is awesome. Well, we've covered a lot of ground. I'm leaving this conversation even more optimistic than I was before. So Karin, thank you so much for coming on the A16Z podcast. Thank you for having me. [55:00] and Spotify. Follow us on X at A16Z and subscribe to our Substack at a16z.substack.com. Thanks again for listening and I'll see you in the next episode. As a reminder, the content here is for informational purposes only. It should not be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments podcast. For more details, including a link to our investments, please see a16z.com forward slash disclosures. Thank you.