Siri Co-Founder, Ex-Olympian Launch GOLD AI to Fight Metabolic Sickness

Siri联合创始人与前奥运选手联手推出GOLD AI对抗代谢疾病 通过合成AI教练应用奥运级量化自我方案可规模化逆转代谢疾病，因为日常行为干预的影响超过了基因因素。 前奥运自行车手Sky Christopherson与Siri联合创始人Tom Gruber共同推出了健康平台GOLD AI，旨在通过规模化推广精英运动员指导方法，应对广泛的代谢健康危机。Christopherson详细阐述了以数据驱动的N-of-1单病例实验如何创造场地自行车世界纪录，并助力原本不被看好的美国女子场地自行车队斩获奥运奖牌。Gruber解释了将确定性科学算法与对话式AI数字分身相结合，如何将原始生物识别追踪转化为易于获取的实时日常动力。此次展示重点介绍了关键硬件合作伙伴关系、主动式智能体教练模型，以及将奥运专业经验推广给大众的策略。 超过60%的美国成年人患有慢性疾病，主要诱因是普遍存在的代谢功能障碍，其中仅有12%的人代谢健康。 (0:03) Tom Gruber提出了“人文主义AI”，强调人工智能应当应对慢性健康等关键人类挑战，而非处理琐碎任务。 (0:03) Sky Christopherson强调了传统体育和创业文化往往美化过度工作和睡眠剥夺，而忽视了生理健康。 (2:55) Christopherson采用N-of-1量化自我方法，发现将睡眠环境温度降至60华氏度可以恢复深度慢波睡眠，并助其创造了场地自行车世界纪录。 (2:55) 美国女子场地自行车队应用纵向数据科学追踪睡眠、进餐时间和光照暴露，克服了1比50的劣势赔率并赢得奥运奖牌。 (9:50)

Former elite cyclist Sky Christopherson and Siri co-founder Tom Gruber argue that the widespread chronic health crisis can be mitigated by translating Olympic-level 'N-of-1' biometric tracking into scalable, personalized AI guidance that emphasizes daily behavioral motivation over passive data tracking. The Chronic Disease Crisis and Metabolic Syndrome (0:03) Siri co-founder Tom Gruber frames personal health as an urgent application for 'humanistic AI,' arguing that artificial intelligence should address society's most difficult problems rather than minor conveniences. Gruber highlights that over 60% of adult Americans live with at least one chronic illness, such as cardiovascular disease, hypertension, or type 2 diabetes. Underlying this trend is metabolic syndrome, which clinicians evaluate using five objective markers: blood sugar, blood lipids, visceral fat, blood pressure, and related metrics. According to Gruber, failing three or more of these markers indicates severe illness, a threshold currently met by 42% of the United States adult population. Overall, 87.5% of adult Americans are metabolically unhealthy, leaving only 12% (1 in 8) within normal clinical ranges. Among older populations not taking prescription medications such as statins or blood pressure drugs, Gruber notes that as few as 2% remain metabolically healthy. N-of-1 Biometrics and Personal Recovery (2:55) Sky Christopherson details his transition from the US National Track Cycling Team—where systemic doping led clean athletes to retire in protest—to the tech startup sector. In the technology industry, Christopherson encountered a culture celebrating severe sleep deprivation, tracking enterprise metrics while neglecting physical health. This lifestyle culminated in severe physical ailments and an emergency room admission for a suspected heart attack. Prompted by cardiologist Dr. Eric Topol's research on the 'quantified self' and findings that daily behavior outweighs genetics in longevity, Christopherson launched a personalized 'N-of-1' tracking experiment using continuous glucose monitors and sleep headbands. While training at age 35 to break a track cycling world record, data showed regular drops in his deep slow-wave sleep. By cooling his mattress to 60°F, Christopherson restored his deep sleep cycles and broke the world record, which had previously been set by an athlete later banned for testosterone use. Olympic Breakthrough and the Scalability Bottleneck (9:50) Five months before the 2012 London Olympics, the underfunded US Women's Track Cycling team entered the competition as 50-to-1 underdogs with no official support staff. Christopherson applied the N-of-1 quantified self methodology to the squad, relying on the athletes' spouses for daily logistics while data scientists tracked sleep, glucose, heart rate, and training loads. Multivariate analysis revealed that precise meal timing stabilized glucose levels and removing hotel light bulbs eliminated late-night light pollution, safeguarding recovery. The intervention culminated in the team cutting three seconds off their lifetime best and edging out Australia by 0.08 seconds in the semifinals to secure the first US women's track cycling medal in 20 years. While chronicled in the documentary *Personal Gold*, Christopherson found that delivering this manual, intensive coaching model could not scale to meet broad consumer demand without software automation. Architecting Scalable Conversational AI for Health (26:06) To transition from manual coaching to an automated platform, Christopherson and Gruber established GOLD AI. Gruber explains that standard health recommendations are already well established; the true barrier is real-time personalization and sustained motivation. Drawing from Siri's synthetic assistant framework, the platform shifts user focus away from retrospective graphs toward immediate, forward-looking guidance answering what an individual should do over the next few hours. The system utilizes a hybrid technical architecture: core health logic and schedule adjustments are handled by deterministic, patented algorithms, while commercial large language models (LLMs) provide natural language interactions and avatar interfaces. Rather than committing to a single frontier LLM, the company prioritizes conversational capability and compute cost efficiency. Operating via an agentic background loop, the virtual coach monitors real-time biometric streams—incorporating data from wearables like Oura and emerging continuous sensors from partners like BioLink—to proactively prompt users when disruptions occur. GOLD AI seeks to adapt elite athletic optimization for general metabolic health by pairing deterministic physiological models with proactive conversational agents. The venture's long-term impact depends on successfully converting high-touch athletic protocols into everyday consumer habits, managing compute costs per user, and validating emerging continuous biosensors like aptamer-based patches.

前奥运自行车选手Sky Christopherson与Siri联合创始人Tom Gruber提出，应将以人为本的人工智能与“N=1”量化自我干预相结合，把原本服务于顶级运动员的高阶生理优化经验转化为面向大众的个性化代谢健康管理与动机支持系统。 代谢健康危机与以人为本的人工智能 (0:03) Siri联合创始人Tom Gruber提出“以人为本的人工智能”理念，主张将计算资源投入到解决人类健康这一最严峻的全球性危机中。他引用的数据显示，超过60%的美国成年人患有至少一种慢性疾病（如2型糖尿病、高血压或心血管疾病）。在临床上，代谢综合征依据血糖、血脂、内脏脂肪、血压等5项客观指标评估，若有3项及以上超标即被认定为严重患病，目前已有42%的美国成年人达到这一患病标准。 更广泛的健康基线数据显示，高达87.5%的美国成年人处于代谢不健康状态，仅有12%（约八分之一）处于正常生理区间；在不依赖降脂或降压药物的老年群体中，代谢健康的比例更是低至约2%。Gruber指出，根据Eric Topol博士关于长寿与超级老人的前沿研究元分析，日常行为习惯对健康结局的决定性影响显著超越先天遗传因素，这凸显了对日常代谢生活方式进行动态干预的紧迫性。 突破世界纪录与奥运奖牌的N=1实验 (2:55) Christopherson回顾了其从国家队场地自行车手到科技创业者的历程。在经历1996年奥运周期违规药物泛滥、佩戴“clean”白臂章抗议并退役后，他进入西雅图科技初创圈，却因推崇剥夺睡眠的加班文化导致身体崩溃，甚至因疑似心脏病发作送医急救。受心脏病学家Eric Topol博士关于数字健康革命的启发，他以自身为受试者展开“N=1”量化实验，利用连续血糖监测仪（CGM）和Zeo脑电头带追踪生理数据。在35岁备战打破场地自行车世界纪录期间，他针对深慢波睡眠衰减问题，采用调至60华氏度（约15.6摄氏度）的水冷床垫进行干预，成功恢复深睡眠并打破了由用药选手保持的世界纪录。 在伦敦奥运周期，面临50比1劣势且毫无赞助支持的美国女子场地自行车队引入了该套方案。由于缺乏预算，团队动员运动员配偶承担后勤杂务，并与数据科学家合作利用圆形关系图分析多维纵向数据，针对性实施了进餐时间平抑血糖波动、拧下酒店高光灯泡杜绝光照干扰等干预。该队最终在半决赛中以0.08秒优势逆转澳大利亚队，将最好成绩缩短3秒并斩获该项目20年来的首枚奥运奖牌。然而，纪录片《Personal Gold》公映后引来的海量大众咨询，暴露出人工高频指导难以规模化扩展的瓶颈。 确定性算法与大语言模型的双层架构 (26:06) 为解决一对一专家指导无法服务数百万人的难题，Gruber结合Siri依托虚拟助手扩展至每日约10亿次调用的经验，指出通用健康常识早已明确，现代AI的核心价值在于两项能力的突破：贴合日常生活的实时个性化适配，以及提供持续的情感动机。系统不应停留在展示历史图表的被动数据堆砌，而应像导航系统一样聚焦未来数小时，回答“接下来应在何时采取何种行动”。 GOLD AI为此构建了双层架构：底层依托包含8项核心专利的自研确定性算法承载运动与医疗逻辑，上层调用大语言模型（LLM）生成交互式虚拟化身。该架构可针对突发事件（如高碳水饮食后提示轻度运动以抑制血糖峰值，或因摄入咖啡因及酒精打乱节奏）全天候动态重算行动计划。此外，团队将女子车队实践中验证的情感支持与被关怀感转化为交互功能，以解决大众从“知晓健康建议”到“持续坚持执行”之间的心理阻碍。 硬件生态集成与主动代理型交互应用 (32:01) 在底层技术与商业推进方面，GOLD AI并未绑定特定的前沿基础大模型。Gruber与Christopherson在问答中说明，由于医疗与运动逻辑由确定性模型独立运算，大语言模型的选择重心在于自然语言交互的流畅度、语境维护能力与算力成本，从而确保消费级市场的经济可行性。在交互机制上，AI具备主动式代理（Agentic）运行闭环，依托后台24小时运行机制，在用户未完成计划活动或生理指标异常时主动发起关怀提醒。 硬件生态方面，GOLD AI正与BioLink合作开发基于DNA适配体（DNA aptamer）的贴片传感器，以期连续追踪葡萄糖、皮质醇、乳酸、睾酮和雌激素5类生理标志物，并推进与智能穿戴设备Oura Ring的数据集成。商业落地上，公司正在推进与美国国家队（Team USA）的合作，吸纳奥运选手成为AI教练性格原型，目前已完成种子轮前融资并开启旧金山湾区新一轮融资以吸纳早期用户。 GOLD AI展示了利用算法和虚拟化身将昂贵的精英运动监测技术普及至大众代谢健康管理的构想，但其发展仍面临客观制约：该方案的高度个性化依赖BioLink新型贴片等连续多标靶硬件的量产与精确度，同时，将顶级运动员的严苛自律通过大语言模型的主动式提醒迁移至普通大众，其长期依从性与真实临床改善效果仍有待持续验证。

## Background 该视频由 YouTube 频道 Ai4 于 2026 年 8 月 21 日发布。视频内容为 Siri 联合创始人 Tom Gruber 与 GOLD AI 首席执行官、前奥运自行车选手 Sky Christopherson 的现场对谈，正式发布基于个性化 AI 与奥运冠军经验的健康应用 GOLD，旨在将顶尖运动健康理念带给大众。 ## TL;DR 面对现代社会高达 42% 成年人患有代谢综合征的严峻健康危机，前奥运自行车选手 Sky Christopherson 结合自身从过度透支到通过“量化自我”打破世界纪录并带领美国女子车队逆袭夺牌的真实经历，联合 Siri 联合创始人 Tom Gruber 共同打造了 GOLD AI。该应用摒弃传统健康软件滞后的历史图表，基于 8 项专利确定性算法与四大健康支柱，以如同实时导航般的动态调度指导用户“何时做何事”。产品通过真实奥运选手为原型的 AI 对话式化身与 24/7 主动式 Agent，将原本昂贵的精英级运动健康指导规模化普及给普通大众。 ## Atmosphere & style 整场对谈兼具深刻的危机警示与振奋人心的竞技叙事，开场即用严峻的代谢健康数据击中现代人的生活痛点。随着 Sky 讲述队内对抗兴奋剂的暗淡岁月、硅谷创业透支崩塌，再到绝境中依靠微观数据实现 0.08 秒奥运奇迹的跌宕起伏，现场气氛由沉重逐渐转为热血与震撼。Tom Gruber 的宏大人文视角与工程沉淀，与 Sky 的运动员坚韧特质形成了完美互补，让观众既能感受到硬核科技与运动科学的严谨，又能体会到充满温度的情绪共鸣与陪伴感。 ## Key takeaways - [00:03](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=3s) 全球代谢健康陷入严峻危机，美国仅12.5%的成年人代谢指标完全正常，技术界亟需从追求短期商业利益转向利用AI攻克慢性病等人类根本健康难题。 - [03:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=180s) 盲目套用针对服药者的传统高强度训练模板，加之科技创业圈将睡眠剥夺浪漫化的文化，印证了忽视个体生理底盘与健康指标所带来的严重代价。 - [06:35](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=395s) “量化自我”个体实验表明，通过连续数据监测锁定环境温度对慢波深睡眠的抑制并引入水冷床垫干预，能以纯自然恢复的方式打破世界纪录。 - [09:48](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=588s) 在兴奋剂丑闻摧毁车队、女子队毫无官方科研支持且夺牌赔率仅1:50的绝境下，全套“N of 1”数据驱动模型成为备战伦敦奥运的核心转折点。 - [12:07](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=727s) 机器学习关联分析确立了健康与运动表现的四大支柱（睡眠恢复、膳食营养、运动训练、心态压力），微观数据干预能带来决定性的边际增益。 - [15:48](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=948s) 数据驱动方案助力美国女子队以不可思议的0.08秒优势逆转夺牌并打破20年奖牌荒，纪录片《Personal Gold》引发了公众对专业健康指导普及化的迫切需求。 - [19:13](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1153s) 传统高接触（High-touch）人类专家指导受限于人力难以普惠大众，促成了奥运冠军团队与Siri联合创始人携手探索规模化教练方案。 - [22:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1336s) 实现从“AI即个人助理（Siri模式）”到“AI即个人教练”的范式跃迁，旨在通过合成替身将少数精英专享的顶级教练能力赋能给普通大众。 - [24:11](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1451s) 改善代谢健康的核心在于行为改变而非盲目加压，长期医学研究证实后天生活习惯对健康长寿的决定性影响远超先天基因。 - [26:06](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1566s) 现代健康AI的价值在于超越冷冰冰的极客式量化自我，将泛化的健康常识转化为针对当下的“实时动态调控”与“持续心理激励”。 - [28:18](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1698s) 摒弃传统健康软件回顾滞后图表的误区，引入顶级运动员的前瞻性调度思维，核心回答用户“接下来几小时内该做什么、何时去做”。 - [30:08](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1808s) 系统采用类似GPS导航的动态调度机制，当日程变动时自动重算饮食与活动安排，并结合反直觉的代谢干预保护线粒体细胞能量与夜间睡眠。 - [32:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1936s) 采用双层技术架构设计：由8项专利的确定性算法确保生理干预指令的严谨可靠，由大语言模型（LLM）与奥运数字化身提供有温度的共情交互。 - [35:20](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2120s) 消费级健康应用并不依赖前沿尖端大模型，核心在于确定性健康算法与极低LLM推理成本的平衡，以确保产品的大规模普及与平价可用。 - [37:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2220s) 针对健康数据碎片化痛点，AI Agent需具备7×24小时后台持续监测与主动介入调控的能力，并提供随叫随到的实时通话支持。 ## Detailed walkthrough ### [00:03](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=3s) 代谢综合征危机：当今社会的慢性健康陷阱 会议以一个看似轻松的互动开场：台下有多少人早晨醒来时感到精力充沛、神清气爽？Siri 的联合创始人 Tom Gruber 随即接过话题，将这一日常感受引向了更深层的严肃议题。Gruber 回顾了自己前一天关于“人文主义人工智能”（Humanistic AI）的演讲，再次呼吁 AI 界不应只盯住唾手可得的商业利益，而必须利用技术去攻克关乎人类命运的重大核心难题——其中首当其冲的便是全球性的健康危机。以美国为例，超过 60% 的成年人正长期遭受糖尿病、高血压、心血管疾病等慢性病的折磨，而绝大多数慢性病的根源，都指向了同一个底层病理状态：代谢综合征（Metabolic Syndrome）。 代谢综合征并非玄学，而是由临床医学长期定义的五项客观生理指标构成的评估体系：空腹血糖、血脂、内脏脂肪（腰围/体脂分布，而非单纯的体重数值）以及血压等。在长达数十年的临床医学标准中，如果一个人在这五项指标中有三项超出正常范围，就会被判定为临床意义上的“重病患者”。令人震惊的是，目前已有 42% 的美国成年人处于这种严重的患病状态，演变成了一场无声却致命的慢性危机。人类在漫长的进化中本应拥有健康的代谢系统，然而现状却是：仅有八分之一（约 12.5%）的美国成年人能够维持正常的代谢功能；而在 Gruber 所在的年龄段，不依靠降压药、他汀类降脂药等药物干预能够保持代谢健康的比例甚至仅有惨淡的 2%。面对这场严峻的健康危机，Gruber 正式引出连续创业者、前奥运自行车运动员 Sky Christopherson，由他分享如何从竞技体育的至暗时刻走向数字化健康革命。 - **慢性病蔓延现状**：美国超过 60% 的人群患有至少一种慢性疾病，42% 的人口满足代谢综合征的重症诊断标准（五项关键生理指标中有三项异常）。 - **正常代谢极为罕见**：仅有 1/8（12.5%）的成年人代谢指标完全正常；在老年群体中，未借助处方药物而实现代谢完全健康的比例仅占 2%。 ```mermaid flowchart TD accTitle: 代谢综合征与健康危机演进逻辑 subgraph S1["临床指标与诊断"] A["5大核心生理指标<br/>(血糖 / 血脂 / 血压 / 内脏脂肪等)"] --> B{"是否有 3 项及以上超标？"} B -- "是 (42% 美国人群)" --> C["确诊为临床重病 / 代谢综合征"] B -- "否 (仅 12.5% 完全健康)" --> D["正常代谢机能 (老年无药健康仅 2%)"] end subgraph S2["后果与慢性疾病"] C --> E["心血管疾病"] C --> F["2型糖尿病"] C --> G["高血压及其他退行性病变"] end ``` --- ### [03:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=180s) 从奥运体制到创业至暗：传统模板训练与恶性加班文化的代价 Sky Christopherson 登台后直言，如果扣除极少数职业运动员，普通大众的代谢健康状况恐怕比“八分之一”的数据更加糟糕。Sky 回忆了自己在 1996 年亚特兰大奥运会前夕入选美国国家自行车队“96计划”（Project 96）的往事。当时队伍虽然配备了顶级装备，但训练模式却完全照搬从前东德购买的固定训练模板（Template Training）——而这套模板本质上是专门为配合违禁药物（兴奋剂）打造的高负荷方案。Sky 与兰斯·阿姆斯特朗（Lance Armstrong）师从同一位教练，队内逐渐分裂为两派：一边是系统性使用兴奋剂的选手，另一边则是坚持干净参赛的运动员。坚持不服药的 Sky 与队友们感到无比痛苦与愤怒，甚至在比赛时用马克笔在白袖标上写下“CLEAN”（干净）以示抗议。在严苛模板的过度消耗下，未服药选手的身体频繁出现伤病和感染，心灰意冷的 Sky 最终选择退役，彻底离开自行车运动。 退役后，Sky 先后前往加州大学圣地亚哥分校（UC San Diego）和西雅图投身科技创业，却一头扎进了硅谷盛行的恶性创业文化中。在这个圈子里，通宵达旦、剥夺睡眠被当成“荣誉勋章”互相炫耀。作为一名前运动员，Sky 感到极为荒谬：“在商业世界里，所有 KPI 都聚焦在公司业务指标上，完全没有人关注核心团队和创始人的身体健康 KPI，但对运动员而言，身体机能才是唯一的底盘。”长期透支最终带来了毁灭性的身体崩塌。几年后的一天，Sky 因剧烈不适被救护车紧急送医，医护人员甚至一度贴上电极片怀疑他突发心肌梗死。虽然最终排除了心梗，但他的血压高得惊人，多项生理指标全面亮起红灯。雪上加霜的是，由于长期久坐电脑前，他患上了严重的肛裂，导致细菌侵入血液循环，整个人陷入了彻底的极度衰竭与绝望之中。 - **前东德训练模板的毒性**：盲目照搬专为服药选手设计的高强度模板，导致坚持自然训练的纯净运动员频繁发生免疫崩溃与运动损伤。 - **创业文化的错位关注**：过度追求商业 KPI，忽视创始人与核心团队的生物学健康指标，将睡眠剥夺和慢性透支浪漫化。 - **健康危机的爆发**：长期的久坐与高压导致 Sky 出现疑似心梗症状、严重高血压及局部组织感染引发的全身性代谢耗竭。 --- ### [06:35](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=395s) 量化自我与个体实验：利用数据打破世界纪录 正当 Sky 处于人生低谷时，一场 TEDx 演讲彻底改变了他的轨迹。著名数字医学先驱 Eric Topol 发表了题为《数字化生活：个人数字健康革命》（Living by Numbers: The Personal Digital Health Revolution）的演讲，系统阐述了“量化自我”（Quantified Self）的理念——即利用新兴的消费级连续监测设备（如睡眠监测、连续血糖监测等），建立个人健康的数据看板。受到启发的 Sky 主动邀请 Eric Topol 担任顾问，启动了一场针对自身的“N of 1”（个体单样本）生物追踪实验。随着身体各项指标逐步恢复，Sky 设立了一个大胆的目标：以 35 岁的年龄打破场地自行车世界纪录。 在严密监测训练的过程中，Sky 借助 Zeo 脑电睡眠头带追踪自身的慢波睡眠（深睡眠）。然而数据却显示，从 1 月、2 月到 3 月，随着训练推进，他的深度睡眠时长竟出现了逐月下降的诡异趋势。现场观众猜测这可能是训练压力或心理焦虑所致，但 Sky 通过排除法锁定了真正的变量：随着春夏季节交替，由于西雅图的住处没有安装空调，室温持续上升阻碍了体温下降。为了验证这一假设，Sky 引入了水冷降温床垫，在夜间将床铺温度强行降至 60 华氏度（约 15.5 摄氏度）。数据反馈立竿见影——慢波深度睡眠时间出现报复性反弹，白天的精神状态和体能储备也随之彻底逆转，彻底消除了午后的疲惫感。最终，35 岁的 Sky 凭借科学的数据化恢复策略成功打破了世界纪录。更具讽刺意味的是，该纪录的原保持者曾因使用睾酮类兴奋剂被终身禁赛，却因规则漏洞保留了该项纪录；Sky 用纯粹的数据干预与自然恢复，完成了对药物作弊者的终极正名。 - **量化自我理念引入**：将原本用于商业的 KPI 思维转移至人体生物学指标，利用可穿戴设备实现连续数据闭环。 - **温度与深睡眠干预**：利用水冷床垫将睡眠环境温度精准控制在 60°F（约 15.5°C），直接逆转了由于夏季室温上升导致的慢波睡眠（Slow-wave sleep）流失。 - **打破药物神话**：在 35 岁高龄、零药物辅助的情况下，打破了由终身禁赛兴奋剂选手保持的世界纪录。 ```mermaid flowchart LR accTitle: 个体量化实验与深睡眠恢复流程 subgraph DataTracking["连续数据监测"] A["Zeo 脑电头带<br/>监测慢波深睡眠"] --> B["发现异常趋势:<br/>1月至3月深睡眠逐月下降"] end subgraph Hypothesis["假设与归因"] B --> C{"排查核心诱因"} C -- "排除训练与压力" --> D["锁定环境因素:<br/>西雅图无空调导致室温上升"] end subgraph Intervention["干预与突破"] D --> E["引入水冷床垫<br/>(降温至 60°F / 15.5°C)"] E --> F["深睡眠时长报复性反弹<br/>身体恢复力与日间精力大幅提升"] F --> G["35岁打破世界纪录<br/>击败兴奋剂保持者"] end ``` --- ### [09:48](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=588s) 绝境突围：伦敦奥运前夕的女子自行车数据救赎 在 Sky 迎来个人突破的同时，美国场地自行车界却因兴奋剂丑闻陷入了彻底的毁灭。兰斯·阿姆斯特朗与其前队友弗洛伊德·兰迪斯（Floyd Landis）在幕后彻底决裂。面对阿姆斯特朗摧毁其名誉的威胁，兰迪斯毅然选择向媒体全盘托出车队系统性服药的内幕。随着阿姆斯特朗在奥普拉访谈中做出毫无悔意的公开承认，美国男子场地自行车队树倒猢狲散，整个国家队体系被完全解散。距离 2012 年伦敦奥运会仅剩五个月的时间，全美场地自行车项目仅存的火种落在了女子团体追逐赛队伍身上。然而，这支女子队伍面临着极其残酷的现实：没有任何官方资金支持、全队只有一名主教练、没有任何科研或后勤保障人员，外界给出的夺牌赔率低至 1 赔 50。 面对绝境，Sky 拨通了前国家队队友、女子车队核心成员珍妮·里德（Jennie Reed）的电话，提议将这套行之有效的“量化自我 N of 1”数据实验全套复制给女子车队，帮助她们备战伦敦奥运会。队伍在西班牙设立了简陋的训练营，由于没有预算聘请专业保障团队，随行前往西班牙探亲的队员丈夫们被当场“抓了壮丁”——这些原本以为只是来度假加油的家属，不得不承担起全部的数据采集、后勤支持与日常保障工作。Sky 现场向观众展示了关于该项目的纪录片片段，揭示了他们如何通过实时监测每一位女性运动员的微观生理数据，逐步在落后、匮乏与不被看好的绝境中挖掘出突破人类体能极限的微小边际增益。 - **阿姆斯特朗丑闻的连锁冲击**：男子车队全军覆没并解散，美国场地自行车项目在伦敦奥运会前五个月仅剩女子队伍孤军奋战。 - **极度悬殊的资源差距**：女子车队夺牌赔率仅为 50:1，无任何官方资金与科研支持，由主教练与队员家属（丈夫们）临时组建后勤团队。 - **数据驱动的奥运实验**：将个体单样本（N of 1）量化健康模型应用于女子运动员的高强度备战，通过精准的睡眠、恢复与负荷监控实现逆风翻盘。 ### [12:07](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=727s) 数据过载与多维关联分析：四大健康支柱的早期干预 在备战奥运的高压节奏下，Sky Christopherson 与运动员之间形成了极其紧密的互信与协作。日常流程被精细切割为“训练、数据分析、进食、拉伸、按摩、再训练、再分析”的无休止循环，宛如将战士全副武装送上赛场。为了让运动员在训练之余得到充分的休息与恢复，Sky 几乎包揽了所有后勤杂务，从买菜、做饭、洗碗到打扫卫生和洗衣服。然而，当他们试图用手提电脑记录和处理海量生理指标时，数据量很快超出了个人处理能力的极限。Sky 意识到他们正站在重大发现的边缘，却受困于分析瓶颈，于是他联系了数据科学团队负责人 Stefan 寻求算力与分析支持。 Stefan 的团队利用机器学习技术对长期的多模态、多变量数据展开了深度挖掘，并绘制出一张圆环关系图谱（Circle Relationship Diagram）。图谱中不同维度之间的连线宽度直观展示了各项指标的关联强度。数据显示，运动员的训练表现、饮食摄入、睡眠质量以及环境压力之间存在着极其显著的高相关系数。基于这些强信号，团队确立了健康与运动表现的四大核心支柱：**睡眠与恢复（Sleep & Recovery）**、**膳食与营养（Meals & Nutrition）**、**运动与训练（Exercise & Training）**，以及**心态与压力负荷（Mindset & Stress / 皮质醇水平）**。 ```mermaid flowchart TD accTitle: 四大支柱多维数据关联与干预模型 subgraph Pillars ["核心生理与行为支柱"] A["睡眠与恢复<br/>(Sleep & Recovery)"] B["膳食与营养<br/>(Meals & Nutrition)"] C["运动与训练<br/>(Exercise & Training)"] D["心态与压力<br/>(Mindset & Stress / Cortisol)"] end subgraph Interventions ["客观数据驱动的精准干预"] E["环境控光：卸除睡房灯泡阻隔蓝光"] F["动态调控：依据血糖曲线调整进餐时机"] end A <--> B B <--> C C <--> D D <--> A A --> E B --> F E & F --> G["赛场终极竞技表现提升"] ``` 团队随即针对这些客观数据信号展开了微观层面的科学干预： - **血糖波动调控**：监测日间血糖骤降（Blood sugar crash）现象，精准调整餐食搭配与进餐时机，平抑能量波动。 - **睡眠光环境干预**：为避免运动员睡前接触酒店房间的强光刺激，团队甚至亲自进入房间“把多余的灯泡拧下来”，确保无多余光源干扰褪黑素分泌。 - **综合表现提升**：这些细微干预的累加直接转化为赛道表现的稳步上升，让团队在动身前往伦敦时底气倍增。 --- ### [15:48](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=948s) 伦敦奥运逆转奇迹：0.08 秒的微光与纪录片《Personal Gold》 带着这套前沿的数据驱动方案，美国女子场地自行车队以不被看好的“绝对黑马（Underdogs）”身份奔赴 2012 年伦敦奥运会。在面对传统自行车强国澳大利亚、英国和法国的激烈竞争中，姑娘们一路杀入半决赛，迎来与澳大利亚队的决死一役。解说席直言，美国队若想保留争夺奖牌的希望，必须将历史最好成绩提升整整 3 秒以上。 比赛开局极为不利，澳大利亚队展现出凶猛攻势，赛程过半时美国队一度落后达 2 秒之多，现场解说和观众一度认为败局已定。然而在最后关头，队员 Sarah Hammer 带领全队发起绝地反扑，时间差距被一寸寸抹平。最终冲线时刻，美国队以 3 分 16 秒的成绩逆转取胜，领先优势仅有不可思议的 **0.08 秒（8/100 秒）**——短如一次眨眼。Sky 与团队在围栏边看到计分板亮起美国队排名第一时，甚至怀疑计时系统出现了故障。这一战打破了美国女子场地自行车长达 20 年的奥运奖牌荒。 这段由科技驱动、挑战生理极限的传奇经历随后被制作成长篇纪录片《Personal Gold》（时长 90 分钟），版权由 Netflix 购得，并在下一届奥运会黄金时段播出，团队还举办了 70 余场线下放映会。 --- ### [19:13](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1153s) 从个人精力的瓶颈，到苹果园区偶遇 Siri 联合创始人 纪录片的广泛传播引发了公众意想不到的强烈共鸣。在多场展映会后，大量观众涌向 Sky，激动地拿出自己的健康数据与体检报告，急切地询问同一个问题：“我们普通人在日常生活中该如何做到这一点？我怎样才能拥有这样的突破并感觉更好？”面对大众对专业健康指导的渴望，Sky 却陷入了深深的无力感与瓶颈期。当时他甚至连一份宣传单页都拿不出，只能被动留下个人邮箱，随即被雪崩般涌入的咨询与数据请求彻底淹没——一对一的高接触（High-touch）专家支持根本无法单靠人力向成千上万的大众普及。 ```mermaid flowchart LR accTitle: 从奥运高接触支持到大规模 AI 化的演进路径 A["奥运队高接触支持<br/>(1 对 1 专家团队)"] --> B["公众需求爆发<br/>(面临人力瓶颈与数据过载)"] B --> C["苹果总部引荐<br/>(会面 Siri 联合创始人 Tom Gruber)"] C --> D["构建 AI 个人教练<br/>(赋能数百万人实时精准指导)"] ``` 带着“如何将少数精英享有的高规格支持规模化放大至千万人”的命题，Sky 前往苹果公司与健康与健身团队的 Jake Blanik 以及 Jules Arie 会面。会议尾声，苹果团队向 Sky 推荐了常驻园区的一位传奇人物——Siri 联合创始人 Tom Gruber，称他极度热衷于具有前瞻性、突破边界的特殊项目。当即发短信确认后，Tom 恰好在园区，两人便约在苹果员工咖啡厅见面。 这次对话给 Sky 留下了极其震撼的印象。在充斥着冰冷算力与芯片的数据世界里，Tom Gruber 展现出了宏大的思考格局与深厚的人文关怀（Humanistic approach）。两人一拍即合，自此保持长期深入探讨，Tom 随后正式出任 GOLD AI 的创始顾问，共同探索利用人工智能攻克规模化教练服务这一难题。 --- ### [22:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1336s) “AI 即个人助理”的范式跃迁：打造人人可及的 AI 个人教练 作为兼具工程背景与前沿应用视野的 AI 先驱，Tom Gruber 阐释了他对“规模化（Scale）”的独到见解。在传统工程语境下，“规模化”往往意味着向硬件砸钱、堆砌成千上万的 CPU 与 GPU；然而在个性化健康领域，你无法简单地通过堆砌硬件来复制一个真实的人类专家。必须从产品隐喻和交互范式上寻找突破口——创造一个能够承载该项服务功能的“合成替身（Synthetic Placeholder）”。 Tom 回顾了 Siri 的诞生逻辑：其灵感可追溯至 1987 年苹果公司展示的“知识导航员（Knowledge Navigator）”概念片。当年开发团队立志让每个人都能拥有专属的个人助理（Personal Assistant）。这一构想得到了 Steve Jobs 的全力支持与推动，不仅打破了传统语音工具的局限，更确立了“AI 作为个人助理”的经典范式，至今该产品每天仍承载着约 10 亿次的交互，且各大主流基础模型实验室仍沿用这一架构。 如今，Tom 与 Sky 将这一成功范式迁移至大健康与运动领域，提出了 **AI 即个人教练（AI as a Personal Coach）** 的全新理念： - **应用前景广阔**：教练指导（Coaching）在健康管理、运动健身、心理健康等垂直领域具有极高的落地价值。 - **超越单点工具**：不再局限于零散的数据记录，而是以拟人化、伴随式的智能体形式，将顶尖奥运级别的综合方法论转化为普通人每天触手可及的实时指引。 ### [24:11](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1451s) 代谢健康的核心在于行为改变 讲座深入探讨了如何将奥运级别的专业指导转化为大众日常的代谢健康管理。这并不是教普通人每天去做多少个俯卧撑或跑多少圈，而是指导大家在维持正常生活与工作的同时，像奥运选手一样训练并获取顶级的代谢健康。这个挑战与“用 AI 寻找癌症疗法”截然不同——它本质上是一个人类行为学问题，AI 助手的真正使命是以规模化的方式，逐一改变每个人的日常健康行为。 创始顾问埃里克·托波尔（Eric Topol）在关于“超级老人”（SuperAgers）和长寿研究的大型元分析中指出，生活行为对健康的影响远远超出了学界以往的认知。过去人们往往侧重于先天基因与后天培养的争论，但科学数据表明，行为是决定长寿的最主要且可主动改变的因素。除了戒烟这种显而易见的常识，如何精准管理睡眠、运动与一日三餐，才是决定代谢健康的关键所在。 - 核心命题并非提升竞技极限，而是通过行为干预将普通人的代谢健康提升至奥运级水准。 - 埃里克·托波尔的研究证实：在决定健康与长寿的维度上，后天行为习惯比先天基因发挥着更具决定性的主导作用。 --- ### [26:06](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1566s) 超越极客式的“量化自我”：AI 的真正价值在于个性化与动机赋能 大众并不需要现代 AI 来重复那些人尽皆知的泛化建议，比如“多运动”或“少吃碳水”。过去的“量化自我”（Quantified Self）模式往往只是技术极客的自嗨，堆砌了大量冷冰冰的数据指标。如果想让健康干预走向主流大众，AI 的核心价值在于两点：**极度个性化的实时方案**与**持续的情绪激励**。 真正的金牌教练之所以不可替代，是因为他们能根据每个运动员当下的状态做出微调——针对昨晚没睡好或午后血糖骤降提供即时调整，并给予强大的心理支持。现实中聘请私人教练价格昂贵，且一周最多只能见上一两次，而 GOLD AI 的愿景则是打造一个全天候 24/7 陪伴在身边的个性化代谢健康助理，将量化反馈与行动激励深度结合。 - 泛化的健康知识已经普及，AI 的突破口在于将“通用建议”转化为“针对你当下的个性化行动”。 - 传统量化自我止步于数据记录，而教练的核心价值在于提供基于量化模型的反馈并激发行动力。 --- ### [28:18](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1698s) 摒弃滞后的历史图表：聚焦“接下来该做什么、何时做” 在 Google 图片中搜索任何健康或健身 App，展现在眼前的几乎全是无休止的数据曲线和历史图表，全部在回顾用户“过去做了什么”。但职业运动员的训练思维完全不同：运动员关注的是未来数小时的时间窗口，提前规划好下一餐吃什么、穿什么装备、安排怎样的日程、是该小憩还是冥想，以确保在关键节点拥有最佳的体能状态。 斯凯（Sky Christopherson）指出，年轻运动员可以把日程排得完美无瑕，但他现在已经年过五十且是两个幼儿的父亲，面临着完全不同的家庭与职场挑战。生活充满了不可预知的变化，因此健康产品的核心假设不应是回顾过去，而是极度简明地回答用户最核心的问题：“我现在该做什么，以及何时去做？”（"What should I do and when?"） ```mermaid flowchart TD accTitle: 从回顾性追踪到前瞻性调度的产品逻辑演进 subgraph Traditional["传统健康 App 模式"] A["收集过去数据"] --> B["生成历史趋势图表"] B --> C["用户自行解读历史表现"] end subgraph GoldAI["GOLD AI 核心引擎"] D["锁定未来数小时窗口"] --> E["动态日程生成与实时重算"] E --> F["明确指令：何时做何事"] F --> G["生活突发变动"] G -. 实时自适应重新调度.-> E end ``` - 现有健康软件普遍深陷“回顾性追踪”（Retroactive Tracking）的误区，充斥着过去一周或一个月的滞后图表。 - 运动员式的思维是前瞻性的超前规划，以未来 2 至 3 小时为行动窗口来锁定能量水平。 --- ### [30:08](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1808s) 如同实时导航般的动态生活调度与代谢科学 GOLD AI 的交互体验被设计得如同车载 GPS 导航系统一样简单直观：“2点向左转做这件事，3点做那件事”。当真实生活发生冲突（例如临时开会无法在下午2点锻炼）时，系统会自动后移并重新计算后续的所有安排，无缝适配现实约束。 系统底层融入了深厚的运动科学机制，甚至能处理违反直觉的代谢场景。例如，若用户血糖偏高且刚吃了一顿大餐，最科学的代谢干预不是静坐，而是立刻进行适量运动以平抑血糖尖峰；在摄入咖啡因或酒精等非理想物质时，系统也不会生硬禁止，而是指导用户在最佳时间窗口饮用，保护细胞线粒体自发的能量生成机制，避免过度压榨肾上腺，进而守住夜间的核心睡眠质量。 - **动态自适应**：如同导航重新规划路线，当日程变动时，后续运动、进食与休息安排自动顺延调整。 - **反直觉的代谢干预**：针对大餐后的血糖波动，即刻安排轻量运动能产生显著的代谢优化效果。 - **线粒体与能量保护**：科学规划咖啡因摄入时间窗，旨在从细胞线粒体层面构建天然精力，防止肾上腺疲劳并保护睡眠。 --- ### [32:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1936s) 对话式 AI 化身、核心算法专利与奥运生态合作 汤姆·格鲁伯（Tom Gruber）强调，最自然的终极交互界面就是“对话”。通过现代生成式 AI 技术，团队打造了可规模化扩展的 AI 数字化身教练。核心的代谢与健康建议由严格的量化算法模型生成，绝非让大语言模型（LLM）去胡乱猜测“中午该不该吃火腿三明治”；大语言模型的真正用武之地，在于赋予这些严谨科学指令以富有人情味、具备共情力与激励性的对话形态，引导用户成为更好的自己。 斯凯回顾道，当年自行车队夺冠后，队员们表示除了数据干预外，最重要的是感受到“有人真正在乎并时刻支持着我们”。GOLD AI 团队不仅将这套奥运经验沉淀为 8 项核心算法专利，还与 Strava 联合创始人马克·盖尼（Mark Gainey）交流，致力于将奥运多元运动的趣味性融入产品。目前团队正与美国国家队（Team USA）推进合作通道，引入现役及潜力奥运选手作为 AI 教练性格原型；在硬件传感端，团队正与创新设备伙伴 Biolinq 合作，利用其基于 DNA 适体（DNA aptamer）技术的连续传感贴片，实现高精度的生理数据无缝监测。 - **8 项核心专利算法**：支撑底层健康决策，确保干预建议基于前沿运动科学而非大模型的随机幻觉。 - **双层架构设计**：硬核算法负责输出精准的科学干预指令，LLM 与数字人化身负责提供共情、温度与激励交互。 - **战略生态合作**： - 与 Team USA 建立人才合作，将真实奥运选手转化为多样化的 AI 教练虚拟化身。 - 携手 Biolinq 开发基于 DNA 适体技术的连续体征监测贴片，提供下一代实时传感支持。 ### [34:37](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2077s) 硬件生态合作、融资计划与早期用户招募 Sky Christopherson 深入探讨了 GOLD AI 在生物传感与硬件生态方面的拓展方向。除了目前常见的数据指标外，团队正着眼于葡萄糖、皮质醇、乳酸、睾酮和雌激素等新型生物标记物的连续监测流，并非常期待与生产这类传感器的厂商展开合作。此外，智能戒指品牌 Oura 也是他们正在积极沟通对接的天然合作伙伴。 在商业推进层面，Sky 透露自己已经完成了种子轮前（Pre-Seed）融资，紧接着将在旧金山湾区启动新一轮融资。他向现场观众发出邀请，无论是希望提供产品反馈的早期体验用户，还是潜在的投资人，都可以前来交流合作，共同塑造这款产品的未来走向。 - 计划接入更多连续生物传感器，涵盖葡萄糖（Glucose）、皮质醇（Cortisol）、乳酸（Lactate）、睾酮（Testosterone）及雌激素（Estrogen）等。 - 已与 Oura 等健康监测硬件品牌建立对接与合作洽谈。 - 已完成 Pre-Seed 轮融资，并在湾区正式开启新一轮融资通道。 --- ### [35:20](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2120s) 大模型选型考量与消费级低成本架构平衡 在观众问答环节，有参会者询问 GOLD 团队目前使用的是哪家大语言模型（LLM），或是计划自研专用模型。现场虽有一定回音干扰，但嘉宾们对模型选型与系统架构给出了非常清晰的定位：系统并不依赖当月最新最前沿的“尖端模型”（Frontier Models），其核心诉求主要在于自然的对话流利度，以及处理用户当前各项健康指标与历史对话上下文的能力。 ```mermaid flowchart TD accTitle: GOLD AI 技术架构与成本控制策略 subgraph Logic["核心计算层"] A["确定性算法引擎<br/>（Deterministic Algorithms）"] B["核心健康模型与规则逻辑"] A --- B end subgraph Conversational["交互与上下文层"] C["大语言模型（LLM）"] D["上下文管理：当前健康指标 + 历史对话"] C --- D end subgraph Strategy["消费级产品化挑战"] E["模型推理成本极低化"] F["保持高度个性化交互"] E --- F end Logic --> Strategy Conversational --> Strategy ``` 团队特别指出，GOLD 最根本的健康分析算法完全是确定性的（Deterministic Algorithms），基础非常扎实。对于 LLM 部分，由于整个 AI 行业演进极快，团队尚未最终锁定某一款特定模型，而是需要在推理预算与功能表现之间寻找最佳平衡点。这类消费级健康应用必须做到极致的平价与可扩展性，如何在保持个性化体验的同时尽可能压低模型调用成本，是当前所有同类团队共同面临的核心挑战。 - 核心健康推断由确定性算法驱动，具备极高可靠性与严谨性。 - LLM 主要承担对话表达与上下文整合（实时生理指标 + 长期对话记录）。 - 尚未绑定特定 LLM 提供商，优先考量极低调用成本与规模化扩展能力的权衡。 --- ### [37:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2220s) 碎片化健康数据整合与 7×24 小时主动式 Agent 机制 一位此前与 Sky 交流过的参会者分享了自己的痛点：目前市面上的健康追踪工具过于碎片化，用户需要穿梭于各个 App 之间记录数据，因此非常认可 GOLD 整合健康全貌的使命；同时他也提出了一个核心问题：团队如何看待 AI Agent（智能体）在产品中的实际落地形态。 Tom Gruber 对此进行了解答，阐明了 Agent 封装在健康场景中的工作方式。该产品的设计理念是让 AI 助手提供 7×24 小时全天候守护，运行在持续的监控与决策循环（Loop）中。当检测到生理数据或行为异常时，系统能主动发起介入，例如“我注意到你今天还没完成某项活动，建议你现在做些调整”。此外，Agent 还必须做到即时随叫随到，用户需要支持时只需一键发起通话即可获得实时指导。 - 解决当前个人健康数据散落在不同软硬件中的碎片化困境。 - Agent 运行全天候后台循环，具备主动提醒与干预能力（Proactive Intervention）。 - 支持一键直连通话，提供随时待命（On-tap）的实时语音交互。 ## Notable quotes > [03:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=180s) 在商业世界里，所有 KPI 都聚焦在公司业务指标上，完全没有人关注核心团队和创始人的身体健康 KPI，但对运动员而言，身体机能才是唯一的底盘。 — Sky Christopherson > [06:35](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=395s) 数字化生活：个人数字健康革命 — Eric Topol > [12:07](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=727s) 把多余的灯泡拧下来 — Sky Christopherson > [19:13](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1153s) 我们普通人在日常生活中该如何做到这一点？我怎样才能拥有这样的突破并感觉更好？ — 现场观众 > [22:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1336s) AI 作为个人助理 — Tom Gruber > [28:18](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1698s) 我现在该做什么，以及何时去做？ — Sky Christopherson > [30:08](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1808s) 2点向左转做这件事，3点做那件事 — Sky Christopherson > [32:16](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=1936s) 有人真正在乎并时刻支持着我们 — 美国女子自行车队队员 > [37:00](https://www.youtube.com/watch?v=gHTbjVsNqGw&t=2220s) 我注意到你今天还没完成某项活动，建议你现在做些调整 — Tom Gruber ## Mentioned Tom Gruber, Sky Christopherson, Eric Topol, Lance Armstrong, Floyd Landis, Jennie Reed, Sarah Hammer, Jake Blanik, Jules Arie, Mark Gainey, Stefan, Siri, GOLD AI, Apple, UC San Diego, Zeo, Personal Gold, Netflix, Team USA, Biolinq, Strava, Oura, Ai4 ## Video description Tom Gruber - Co-Founder, Siri Sky Christopherson - CEO, GOLD AI Witness the world premier launch of GOLD — the app that turns medal-winning innovation into simple, real-time guidance for your health and fitness goals. Always by your side with AI and supported by real coaches. All ages. All levels. GOLD is an AI and human-powered health app that brings Olympic medal-winning innovation into everyday life. It uses AI personas modeled after real Olympians to deliver world-class expertise in a way that feels personal, motivating, and simple to follow—along with optional live coaching when you want human support. From meals and movement to sleep and recovery, GOLD turns proven methods into real-time guidance that helps anyone improve health, well-being, and quality of life, one day at a time. See the product on stage in conversation with Olympic athlete Sky Christopherson and AI expert Tom Gruber, co-creator of the first mainstream AI product that is used billions of times daily. [Source](https://www.youtube.com/watch?v=gHTbjVsNqGw)

[3:00] Sky Christopherson: 在商业世界里，所有 KPI 都聚焦在公司业务指标上，完全没有人关注核心团队和创始人的身体健康 KPI，但对运动员而言，身体机能才是唯一的底盘。 [6:35] Eric Topol: 数字化生活：个人数字健康革命 [12:07] Sky Christopherson: 把多余的灯泡拧下来 [19:13] 现场观众: 我们普通人在日常生活中该如何做到这一点？我怎样才能拥有这样的突破并感觉更好？ [22:16] Tom Gruber: AI 作为个人助理 [28:18] Sky Christopherson: 我现在该做什么，以及何时去做？ [30:08] Sky Christopherson: 2点向左转做这件事，3点做那件事 [32:16] 美国女子自行车队队员: 有人真正在乎并时刻支持着我们 [37:00] Tom Gruber: 我注意到你今天还没完成某项活动，建议你现在做些调整

[2:55] Sky Christopherson: Living by Numbers: The Personal Digital Health Revolution. [0:03] Tom Gruber: humanistic AI [32:01] Sky Christopherson: What should I do and when?

[0:03] I just have a quick question. How how many in  general, how many of you wake up in the morning [0:08] feeling refreshed and energized? I'm just curious.  Okay, it's good. Okay, hand it over to Tom here. [0:18] So, why is he asking you about whether you  feel good? I don't know. It's a feel-good talk, [0:22] right? Actually, maybe not. But it's interesting.  We're going to make it relevant. U I'm I'm Tom [0:27] Gruber. Yesterday, some of you might have seen  my talk on humanistic AI. Hi. Um, in which I made [0:33] the case that we should be trying to attack the  hard interesting important problems for humanity [0:38] now with AI um rather than just going after the  lowhanging fruit and um one of them is health. [0:44] Health is a big issue for the world. Now, how big?  Well, okay, this is talking about the US. 60% more [0:53] than half of us live with a chronic disease like  diabetes, hypertension, cardiovascular disease. [1:01] Most of us are sick. And most of these diseases,  there's an underlying condition called metabolic [1:07] syndrome. And so today we're going to talk about  metabolic syndrome and what we might be able to do [1:12] about it. Now this is the problem with metabolic  syndrome. It's it's it's not a it's not a new age [1:18] thing. There's five objective measures. Probably  most of you gotten these done in your workup [1:23] when you go to the doctor every year. There's  some blood tests for blood sugar, blood fats, [1:28] and so on. They weigh you, but actually it's  not your weight. It's your as visceral fat that [1:32] matters and your blood pressure. Now, if you you  think about it, uh well, if according to clinical [1:41] research has gone on for decades, if you're out  of range for three of these, you are very sick. [1:48] are considered clinically very sick and 42% of us  are that sick today. It's really a crisis and we [1:57] don't hear much about it because it's a slowmoving  crisis, right? But then think about it another [2:01] way. We evolved to be healthy. We evolved to have  a normal body and a normal body has all these [2:08] in range. And so think guess how many of adult  Americans are normal or have healthy metabolic [2:16] function. Any guesses? Anybody want to shout  out something? Well, I'll just guess. Actually, [2:23] I know it's one in eight. One in eight. Hardly any  of us are metabolically healthy. And if you're if [2:32] you're my age, according to some reports, only 2%  of people my age are metabolically healthy without [2:39] using drugs. That is, you know, blood pressure  medicine, statins, and things like that. Now, [2:44] it's crazy. So, it's a crisis. It's a real crisis.  So when we're thinking about what we can do with [2:49] AI to help humanity, this is one of those areas.  So I'm going to turn over the the mic to Sky [2:55] Chrisopson. He's an entrepreneur and he'll tell  you the story of how he became an entrepreneur in [3:00] health and how he got here. Here you go. Thank  you, Tom. So yeah, one in eight metabolically [3:07] healthy and I'm concerned that maybe that  one in the population could be athletes. [3:13] So really we have to kind of remove that part  and you know we look at the broader population [3:18] in general and those numbers are are very  concerning. So I had the honor of being on the [3:25] US cycling team right before the Atlanta Olympic  Games. I was part of pro what they called project [3:31] 96 which had us going into those Olympics with  the the best equipment, some of the best training [3:37] programs but was all very focused on template  training. Actually, we purchased it from the [3:43] Eastern East German uh program and and used that  as our training template, right, which was really [3:49] designed to be used with drugs. And speaking on  of that topic, this was so I I was teammates with [3:56] Lance Armstrong. We shared the same coach and on  the on the national team that year and it became [4:02] a a division. So half the team, you know, now we  learned later, was was doping and using drugs. the [4:09] other half was, you know, ve very bitter and, you  know, very uh resentful about that and it created [4:15] this tension and this split and for me it was very  frustrating because being one of the holdouts, I [4:20] mean literally we were my teammates and I on the,  you know, the clean side, we were going to races [4:26] wearing white armbands with a magic marker written  clean. You know, that's that's how dirty cycling [4:33] was at the time. So out of protest, right? So I  retired and I was very frustrated. I mean we it [4:41] was common to get sick and injured and I just felt  so discouraged. So wanting to put sport completely [4:48] behind me, I retired, went to UC San Diego, went  to Seattle and and did a startup. And I kind of [4:54] dove head first into the startup culture, right?  where you have badges of honor for not sleeping, [5:01] pulling all nighters, you know, these these kind  of very unhealthy behaviors are are celebrated. [5:06] And and one quick side note, just as an athlete  coming into the the business and entrepreneur [5:12] world, one of the things that blew my mind was  how all the KPIs are about the business and [5:19] not about the health of the core team and the  founders. And and as an athlete, that's that's [5:24] your only concern, right? is is is your is your  core. So that was remarkable remarkable to me, [5:30] right? But I participated in it and it's tough to  say that I I actually deteriorated my health over [5:39] a few year period and ended up in an ambulance  thinking I was having a heart attack. They told me [5:47] they had the leads on me and they said, you know,  we we're sorry to say, but we we think you're we [5:52] think you're having a heart attack. And I I went  in and and fortunately I was okay. They scanned [5:59] everything and the doctor said though back to  these metrics, my blood pressure was high. I I had [6:05] very concerning metrics. They wanted to put me on  medication and I had other issues from I'll just [6:12] tell you guys. I don't care. I was sitting so much  at the computer that I actually had a they call a [6:19] fissure. I get a fissure, right? and and uh what  it does is um on your butt and it lets uh bacteria [6:27] into your bloodstream and and I was absolutely  exhausted and losing all hope. So, right just [6:35] like a blessing from God, I went to a TED talk.  You guys remember TEDex? They're doing all these [6:41] great community events. Eric Toppel, he was there  giving a a talk. It was called living by numbers, [6:50] the personal digital health revolution. And the  idea was kind of this quantified self model that [6:56] you would use personal data in all these new  devices that we had to track sleep, you know, [7:01] overnight to track blood glucose, to continuously  track a lot of these things that we never had as [7:07] athletes. And that's perfect for monitoring again  those KPIs of more of you as a as a founder or as [7:14] an individual. So I I set up a kind of an  experiment with Toppel's help. He became [7:20] an adviser. The idea was going to be to measure  myself using all these new consumer devices and I [7:28] I started getting my health back and I quickly  kind of adjusted the target and said, "Okay, [7:34] I want to try to break a world record." And and  as I as I was I started training for this at the [7:41] turn of the year and one one thing here is that  I noticed in tracking the deep sleep and this is [7:46] from a Zo headband with with slowwave sleep right  month to month January, February, March I saw my [7:54] deep sleep was dropping as I was training for this  world record. Anyone have any ideas why deep sleep [8:01] would be dropping monthto monthth? Anybody have  any ideas? Just just yell it out. Anybody live in [8:08] Seattle? That's a big clue. Yeah, that's a very  that's a very great hypothesis. Stress, anxiety, [8:17] the pressure of training. I thought maybe I mean  we didn't know right until we tested it. Somebody [8:23] gave me the idea that it could be related to the  to the room temperature because actually I didn't [8:28] have air conditioning. It was getting hotter into  the towards the summer. So to test this hypothesis [8:34] on this NM1 experiment, I had a water cooling  mattress topper that that drops it down to like 60 [8:41] degrees. You get really cold when you're sleeping.  And look at the difference in the data after that [8:47] intervention. That deep sleep, that's recovery.  You know, you have guys doing the tour to France [8:53] who are EPO and they're there sleeping in hot  rooms with none of this kind of stuff and none [8:59] of this data, but they're doing drugs, right? So  remarkable, right? In terms of the the boost and I [9:06] you feel it even the next, you know, the next day  where you usually feel that that heaviness in the [9:10] afternoon, you want to take a nap, just solid  energy, right? So I finished preparations. I I [9:17] went to the um record attempt and I got it and I  broke the world record at 35 years old. Thank you. [9:29] So, man, one of probably the coolest thing I  did in my career because not only got my health [9:34] back and got my hope back, but this record was  held by a doper who because of a technicality, [9:41] they they left the record for him. He had a  lifetime ban for testosterone use and I got [9:46] it from him. So pretty cool thing. But now the bad  news was in terms of going to the Olympics again. [9:53] The dis the track team I'm on the veladrome right  the sprinter the the men's team was disbanded more [9:59] or less. Lance Armstrong and all those dominoes  fell. They have the thing they call the Merida. [10:06] They all kept the secret for the doping. And  on the inside, I told you this yesterday, [10:11] uh, Floyd Landis and Lance got into a kind of  a fight behind the scenes, and, uh, Floyd said, [10:18] "Floy, I'm going to go tell the media everything."  Lance said, "Go ahead and try it. I'll I'll ruin [10:23] your reputation. I'll tell him you're an  alcoholic. You know, I'll totally burn you." [10:27] Floyd did it anyway. He went to the the press.  Next thing you know, Lance is on Oprah and he's, [10:34] you know, confessing not an apology, but just kind  of, hey, everyone was doing it. Can you really [10:40] blame me kind of thing? So, Lance is out. We  have the Olympics coming up in like five months, [10:47] and the only hope left is the women's track  cycling team. No funding, one coach, no staff. [10:56] That's my teammate right there, Jenny Reid, second  to the left. So I called Jenny. I said, 'Hey, [11:02] why don't we try this quantified self experiment  to try to help you guys get ready for the London [11:08] Olympics? They had 50 to1 odds against them,  right? Almost no chance of of winning a medal. [11:15] So we go there and we start setting them up with  all this this this great uh N of one tracking uh [11:21] experiments. Their husbands were really the only  other people around. We had no money for staff, [11:27] right? So these guys, these great husbands,  thought they were coming on vacation to Spain [11:32] to kind of cheer their wives on. Little did they  know, we're gonna have to put them to work. So as [11:38] we were doing all the data collection and and  and training, uh these guys filled in all the [11:43] other gaps. And uh we're going to show you guys  just a there's there was a film made about this [11:48] which I'll I'll mention in a few minutes, but uh  I'll give you guys a couple excerpts from the film [11:54] that was made about this project so you can kind  of see how we pulled together and then an example [11:59] of of the insights from the data from these women.  It's step by step each of the husbands and and we [12:07] bonded and and learned with each other and got  got it taken care of. train, [music] analyze, [12:13] eat, stretch, massage, train, analyze, eat. It was  constant. It was uh it was like sending her out to [12:20] battle, making sure she was fully prepared  to to dominate. When she wasn't training, [12:25] she was resting and recovering, and it was my  job to fill in all the other gaps. The grocery [12:32] shopping, the dishes, the cooking, cleaning,  the laundry. Yeah. I, you know, stayed busy. [12:42] There's the B. We were storing and analyzing  all this data on our laptops. And the problem [12:47] was we were generating so much data  that we couldn't analyze it anymore, [12:52] which was a huge problem because we were finally  getting potentially to the good stuff here. So I [12:58] set up a call with Stfan. As you guys are experts  in your field, we really good in analyzing [13:06] large amount of data and this is a really exciting  project and if we can help we would like just to [13:11] do that we would so I was excited now because we  were going to crunch and analyze and visualize [13:17] this data at a level that we never could before  we have something for you you know it and we [13:24] also get uh he leads our data scientist team  on the board let me show you this um so this [13:30] a circle relationship diagram so this shows for  each athlete training diet sleep environmental [13:36] data and how all those things are interrelated.  You see the width of the connection creates how [13:42] much relation the different data sets have to  each other and that's the fascinating piece. [13:48] We have a bunch of really high uh corrodation  coefficients. Strong signal. Wow. Yeah, [13:57] it's really fascinating. We didn't see anything  like this before. One way you could look at this, [14:03] I guess, is that you know seeing that everything  is interconnected. Yeah. So you can see what [14:10] influences in the end basically performance.  Wow. Well, our team here is all fired up and [14:17] um Thank you. The only thing we can do is  analyze data. If that somehow helps you guys, [14:23] I mean we well we hold our fingers crossed and  we will be [music] glued in front of the TV and [14:28] we will yell for you as loud as we can. Good luck  in London. Okay. Thank you. So that's an example [14:37] of an insight that as you can see these guys were  pretty excited about. They said they'd never seen [14:41] anything like that before. So this was pre-AII  days. This was machine learning and we were [14:47] looking at this you know multi multimodal data  multivariate data longitudinally and there were [14:53] these strong interconnections and the pillars  sort of arose it became clear the pillars of [14:59] sleep and recovery meals nutrition exercise their  training and then the mindset and the stress let's [15:08] say cortisol level right those things strongly  interconnected so we did interventions right [15:14] in each into those categories. For example, um  blood sugar dropping and crashing during the day, [15:20] adjusting the the meals and the meal timing,  right? Because we're getting objective data [15:25] and and improving the performance. Even things  like light exposure, looking at bright lights [15:30] in their hotel rooms before bed and going in  and actually unscrewing light bulbs so they [15:35] they couldn't turn the lights on, right? So, so  all these interventions and we started seeing a [15:40] rise in their their performance. So, we were we  were packing up. We had no idea what to expect. [15:48] We're going to the London Olympics as total  underdogs. It's Great Britain and Australia [15:53] and France and all of these countries that  are just dominant. And we're going in and we [15:58] we made it to the semifinal ride. And I'll just  show you the clip so you can guys can see what [16:04] happened. The business in United States against  Australia. This was the race that we were either [16:11] going to walk out of that stadium knowing that  we were going to have an Olympic medal or not. [16:20] For any hope of a medal, the Americans will  have to beat their lifetime best by 3 seconds [16:25] here. And they're off. [cheering] Dealers love to  get in for that gold medal right off their bra. [16:34] But man, Australia came out hot. The radio putting  it out Australia and it just got worse with with [16:44] a K in. It was I think we were down by 2 seconds  and this gap was looking like it was not going to [16:50] be closed. No way. Something special came out. The  girls, they rallied and uh the time just started [16:59] coming down fast. pull it back and Sarah goes  for it and that time continues going down 642 [17:12] back very strongly. Can Australia react as she  did in qualifying as they were coming down the [17:17] straightaway. I'm sitting in my corner  where I was for the simulation where I [17:21] was for every training. I know they heard me  go for it at the head of the US lineup USA. [17:34] Can Australia react action?  Am I thriving this team? USA [17:38] very tight across the line. And  Australia have been beaten 316. [17:47] [applause] [17:53] Amazing. Oh, I'm just so proud. I can't I  can't tell you how much adversity. I think [17:59] that's the coolest thing I've ever  seen in sporting. The fight. I mean, [18:02] the they the rally they came back from behind.  It's just so cool. Made it all worth it. [18:12] It's the first US [music] women's track cycling  medal in 20 years. I think that's something that [18:19] we can, [music] you know, hopefully transfer over  to the general population. For anybody out there, [18:27] anything you put your heart and soul  into, anything's [music] possible. [18:41] 8100s of a second. That's like a blink of  an eye. And I was there on the rail next to [18:48] Brandon. We looked at the scoreboard and US  popped up number one. I thought there it was [18:54] broken. I thought there was a timing problem and  we were looking at each other kind of confused [18:59] like wait is, you know, it's like reality kind of  reshuffleling. So yeah, and just as Sarah said, [19:05] you know, anything you set your heart and soul to,  anything's possible. And that spirit, you know, [19:12] fortunately we captured it and created this uh  film in Hollywood afterwards called Personal [19:16] Gold. It was a feature documentary 90 minutes and  uh Netflix acquired it. We did 70 events. It was [19:23] broadcast at the next Olympics prime time. And  one of the cool other coolest things was going [19:30] to these screenings and just seeing people's  reactions, right? And they would come up to us [19:36] afterwards with a very common question and it was  how can I how can we do this in our lives? Right? [19:42] How can I do that? How can I feel better? How can  I have that that feeling or those breakthroughs? [19:47] Now the problem and I had so many great stories  and they would give me I' I felt helpless [19:53] because we inspired them and then I didn't have  anything to leave them with. I mean not even like [19:58] a pamphlet or whatever. [laughter] It's like you  know here's my email address, right? like stay in [20:04] touch. And so some of these people would they were  showing me data. They were wanting guidance. And I [20:10] got instantly bogged down. I just couldn't right.  I got bottlenecked. And that goes into the key [20:15] question here. And this is the exciting thing.  And we have the co-founder of Apple Siri here [20:21] to help us navigate this this this question here  is like how to scale how to scale that hightouch [20:28] you know you uh support that we did with the team  to the the dozens the hundreds the thousands the [20:36] millions and so I actually had a meeting with  Apple we have an we have an current Apple Siri [20:42] team member here which is another cool thing you  want to stand up and say hi here. [laughter] This [20:48] is so cool. We have the we have the se the the  original Siri and we have the current Siri AI. [20:54] And I was at Apple meeting with with the health  and fitness team there, Jake Blanik and Jules [20:59] Arie. And after the meeting, they said, "You know  what?" But they said, "You should talk to this, [21:04] we have this legend here around the Apple  campus. His name's Tom Gruber. He's really [21:09] into these special projects and kind of pushing  the envelope." And they said, "You know what? [21:13] he might he he's around, you know, sometimes.  You want us to text him? And I was like, "Yeah, [21:19] sure." So, they text, they get a text back and  they said, "He's here today." And I said, "Wow, [21:26] okay. Can can I meet him?" And they said, "Yeah,  pop over to the Apple uh cafe and he can meet you [21:31] for coffee." And I was like, "Oh my god, this is  so cool." Like, you know, the co-founder of Siri. [21:37] Tom and I sit down and what immediately stuck out  to me, I mean, talk about big thinking. I was so [21:44] inspired, right, by you having this humanistic  approach. There's so much heart like just did [21:50] you feel the guys feel the heart, right, of that?  And I feel like Tom has that which in the world of [21:56] data and silicone, you know, it's it's a it's it's  a gem and and he was really wanting to do some [22:01] some special projects there. So, we we stayed in  touch, kept talking. Now he's my founding advisor [22:07] and me Tom maybe you could you could give a few  um just to help us navigate here this question [22:15] of how do we scale that it's great because I can  put a foot in both camps you know like trying to [22:22] do something good in application space but being  an engineer um scale that when you hear that word [22:28] scale it man usually that means you know throw  money in CPUs and GPUs and that the problem and [22:32] you know just do a lot more of it but you can't  you you can't put money and GPUs into him. It's [22:38] not going to work. Right? So, we had He's like,  "How do we scale this stuff?" Because I'm I'm not [22:43] worth doing if you're going to have 10 clients.  Let's have a million, whatever. Right. Right. So, [22:48] I know I know something about scale because we  did it with Siri. Uh we decided to do AI for [22:53] everyone. And we didn't get everyone, but we got  it. It's used about a billion times a day. So, [22:57] it's a fairly decent chunk of people online have  have used it or do use it. Um, and it's been [23:03] replaced by the newer, you know, newer stuff.  But, um, I don't know if any of you recognize [23:08] the bow tie guy from the 1987 Apple Navigator  video, and this is my version of that. Um, but, [23:14] you know, so that was our model is basically,  yeah, you can scale because you can, you know, [23:17] can produce a a placeholder for the service. So,  in our case, it was personal assistant, right? [23:23] We wanted to give everyone a personal assistant.  And, you know, you can't you can't just, you know, [23:29] grow a bunch of personal assistants. You have to  do it synthetically. And so it was the metaphor [23:33] of our it wasn't just a voice app. It was really  a whole assistant metaphor. And Steve Jobs loved [23:37] that idea and wanted to really empower us to  do the whole thing. So that was the beginning [23:42] of you know AI as an assistant. And I'm happy to  see actually that metaphor is still around. You [23:46] see all the major foundation labs and using this  assistant model as the way of delivering AI. Now, [23:52] in particular, what Sky and I talked about was  AI as a personal coach. And it's one of these, [23:58] it's one of these great opportunities. Coaching  in all kinds of things from wellness, health, [24:02] mental health care, and so on. Coaching is going  to be a really great application space for AI. Uh, [24:07] it already is. But actually, in the case of  coaching, let's go back to slide one. You know, [24:11] we're not talking about coaching for performance  in Olympics. We're talking about coaching for [24:17] performance and metabolic health. And what does  that mean? Well, in this case, it's not so much [24:22] about how many push-ups to do or how many laps  to run. It's it's really about advising you in [24:28] your daily life how to train like an Olympian,  get metabolic health like an Olympian, but a but [24:35] being a normal person, having a normal life.  And so, that's the metaphor we're using here, [24:40] that essentially, well, a personal coach, if  you could hire a personal life coach or personal [24:45] metabolic coach, this is the kind of thing we want  to be able to put in an assistant. And here the [24:50] problem is this or the opportunity is this is that  it's actually not a problem like you would have AI [24:56] cure find a cure for cancer. It's not that kind  of a problem. It's a behavior problem. And so the [25:02] the role of the assistant is actually to change  human behavior at scale one person at a time. [25:09] And the same core drivers apply. I mean these are  the things that drive metabolic health. Yeah. You [25:15] guys remember that interconnected insight from  the N of let's call it N of4 athlete data. The [25:22] more we researched afterwards, the more we found  those are overlapping with all of us. And our [25:27] founding adviser, Eric Toppel, just finished  a big meta review for the book Long uh Super [25:33] Agers. And they found that behavior was much more  influential than the field ever thought before. [25:40] They thought genetics, you know, nature versus  nurture. And and that's incredible. In fact, [25:45] Toppo Tppel is like the quantified self guy or  one of them. And uh he then did another book later [25:50] showed that in longevity behavior is the dominant  preventable changeable thing in longevity as well. [25:56] And we're talking, you know, of course behavior  some behavior like don't smoke of course is no [25:59] no brainer. But behaviors like how to manage your  sleep, exercise, and meals. It's a big deal. Now [26:06] you you might think, okay, so we're going to  show you like another quantified self app with [26:10] AI driving all the metrics. And it turns out that  you don't need modern AI to figure to advise you [26:17] on what when to eat, what to eat, what to do, all  that. You don't you especially don't need it for [26:21] like uh uh what to what how many push-ups to do  or whatever. Okay? But what you need AI to do is [26:27] advise you on your life the whole life. And really  the AI is going to come in not in giving better [26:32] advice because we kind of know what the good  advice is. It's to make it personalized for you [26:39] because you can read this stuff. Okay. Everyone  knows you should exercise more. You should eat low [26:44] carb, you know, diets, whatever. Everyone knows  the general advice. But actually, the way that [26:48] Sky coached the women's biking team to a medal  was personalized to every athlete. You are not [26:55] sleeping well. We're going to change that. You  are getting a glucose drop in the middle of the [26:59] day. We're going to change that. And so on. Not  just change it once, but every day. And this is [27:04] sort of like quantified self was for G nerds. You  know, most of the people who do that were nerdy [27:09] and geeky. Um and now but if you're gonna have a  mainstream effect on a population, we need to move [27:13] beyond just getting the numbers. And so it's the  second piece of that is motivation. So delivering [27:19] feedback based on a quantitative model, but doing  it in a way that actually motivates people to act [27:26] on it. And that is actually what a real coach  does. If you shell the money for a personal [27:30] trainer, the most of all, first of all, you're  only going to see them one or twice twice a week [27:34] maybe. But even if you did that, what they want to  do is advise you on, you know, help you motivate, [27:40] right? But if you could do a personal assistant  coach for metabolic health every day, 247, well, [27:46] that's our vision. So now we're going to give it  some concrete clothes here in this application [27:52] called Gold AI. Yes. So, Gold AI was a we created  a company and the core DNA will be to take that [28:02] inspirational Olympic feeling which you guys  might have felt a little bit there [laughter] in [28:07] that in that win and capture that, right? And  sort of channel that into supporting everybody [28:13] for breakthroughs in their health and fitness,  whatever your goal is. And if you guys, you can [28:18] do this, right? You go to Google, you can type in  health or fitness app. Click images just as like [28:24] a visual landscape. You're going to see numbers  and graphs. Just add infin item, right? That's the [28:32] current landscape, numbers and graphs. And it's  all about what you did in the past. And obviously [28:38] they try to say, okay, you have these trends in  the past, you know, over the last week and month. [28:42] As athletes, it always kind of it always kind  of blew my surprised me that because what we do [28:48] is we plan we plan on the horizon of a few hours  ahead. Typically, we just make sure we're always, [28:53] you know, dressed with the right thing. We're  we're we've eaten the right meal. We're going in, [28:58] we have, you know, the right things scheduled and  we have the right energy level. If we don't take [29:02] a nap or meditate or, you know, there different  techniques we would use, but it's it's about that [29:09] time horizon right in front of you. That's where  the rubber meets the road and that's we feel like [29:14] is just it's not being captured right now with  this retroactive tracking. So in designing a [29:19] product the core question we we're embracing this  is our hypothesis. What should I do and when? And [29:26] you're going to see a radically different product  design here. All this stuff I just showed you. And [29:32] it's just that time horizon. It's that simple,  right? Well, it's not simple. It's extremely [29:37] complicated but but it appears simple. and it's  day and night and it's those key targets and [29:44] and most importantly it's like athletes make the  perfect schedules but as now that I'm in my 50s [29:51] and I'm a father of two toddlers it's a completely  different daily puzzle as a lot of you guys know [29:58] and your professional responsibilities and all  of that and it's always changing and so the key [30:03] thing is what do we do when those behaviors change  and we adapt and we update those we let me add an [30:08] example So for example, a simple thing and you may  not you know I was surprised to find out. So if [30:14] you take a go back one slide just like that that  if you look at the a daily schedule of what you're [30:18] going to be doing. Okay. Um if you let's say that  you're a Olympic athlete and you're training at [30:23] the actual worldass level and you're going to be  pounding these carbs and then going out there and [30:27] you can measure this energy. That's not the same  advice for everyone else. But every since it's [30:31] personalized it can be the right advice. But  interestingly, let's say that you're actually [30:35] having some blood sugar issues and you have a  giant meal. The unintuitive answer is go out and [30:42] do some exercise right after the meal. [laughter]  You know, it's weird, but that's actually the best [30:46] thing for your metabolic health in that particular  case. Yeah. Or like when you you know, and what [30:50] kind of a meal depends on how what you did. Was  it a high energy workout or was it more of a [30:54] endurance thing? Um you get more of a carb loaded  versus a fat or a protein thing. And these are uh [31:01] these are known to exercise science and they're  known to quantified self the world. But what [31:06] we're trying to do here is make it into a simple  thing. It's kind of like as simple as a na at a [31:10] navigation system. It just goes, you know, make  a left turn, two o'clock, do this, then 3:00 do [31:16] that, and then if you slide things around because  life happens. Sorry, I can't work out at two. It [31:22] just slides everything else out to work to work  with its constraints. Yeah, absolutely. And even [31:29] supporting all of our real world things like  alcohol, caffeine, things that are not optimal, [31:36] but supporting those and always guiding you and  supporting you and in uh you know, let's say [31:41] like caffeine, you know, when enjoy it, but for  what times for you, you know, when when to drink [31:46] it and when to not drink it and how to preserve  your sleep and and build your energy naturally [31:51] in the mitochondria of your cells, which is the  foundation of our energy, you know, not squeezing [31:56] the adrenals with caffeine. So all of these things  you know this we we can build this very complex [32:01] model to support all of these things but in the  end and this is where Tom came in is the ultimate [32:08] simplicity and interface is a conversation and and  that's becoming the core design of this product [32:16] you probably so just imagine you know having a  real coach uh who doesn't live with you how are [32:22] you going to interact with that coach you're going  to call him or her up right you're going to take [32:25] make a phone call so now with modern AI We can  do this virtually. We can have uh an AI scaled [32:32] agent avatar, right, that can talk to you uh and  give you all that advice. The interesting thing is [32:38] the advice that it's giving you is algorithmically  driven. It does it's not we're not asking Claude, [32:44] should I have a ham sandwich? That we know from  the from the like Eric Tobble research. But what [32:50] it is saying is putting it in a human form  like in a conversational way that people can [32:55] accept. And then of course layering in all the  motivational things you'd want to do with a real [33:00] coach and and the and the character AIs that are  being driven for mental health and not for mental [33:07] addiction are full of these great ideas about how  to encourage people to be the better versions of [33:12] themselves. And we're going to use those kinds  of ideas in the conversational interface. Yeah. [33:16] because the you know the the equation half of it  is you know what should you do and when and it's [33:21] all the data but the other half is that motivation  and the inspiration and someone just being there [33:26] for you. I mean the women in the end when we asked  them what were the biggest things they said yeah [33:30] those interventions mattered the other half of  it was that you cared and you were you were there [33:36] for us and and that was a big part of it. So we  want to roll that into the company building this [33:42] you know Olympic expertise as as the cultural  DNA into this. We had eight patents on the core [33:48] algorithms which came from that innovation and a  lot of top science that you're not going to find [33:52] in the in the language models. And what we use  the LLM for and these other you know AI tools [33:58] are really to scale do what they're good at which  is scaling conversations scaling you know avatars. [34:04] Um the I spoke to the co-founder of Strava Mark  Gayy recently and he said we have an opportunity [34:09] to make this really fun. The Olympics are fun. You  have all the different sports and the different [34:14] athletes and I have a partnership uh in the works  with team USA to kind of make a pipeline for all [34:20] the Olympians and Olympic hopefuls to come in  and be potential coaching personalities. And [34:26] then finally other partnerships we have a very  innovative device partner named Biolink and they [34:32] have a patch sensor that's DNA opterrimer based.  It can do continuous sensing like the stuff we tal [34:37] talked about before in new areas glucose cortisol  lactate testosterone estrogen and continuous [34:49] streams of these. We we we would love to partner  with that sensor. Aura is another natural partner [34:54] we've been speaking with. So, we are where we  are as a company. Uh, I've I've raised preede. [35:01] We're doing a financing round right after this  starting in the Bay Area. And if you'd like to [35:07] join us as an early adopter, I would love your  feedback and your input as we shape this. And uh, [35:15] obviously if you're an investor, talk to us as  well. And we would and how much time we have for [35:20] Q&A. Thanks very much. We do have about 3 minutes  for your questions. Any burning questions? Oh, [35:25] any feel free. Even non-burning questions, come  on. Come up and we'll we'll repeat the question. [35:29] If you don't have the mic, just shout it out. I'm  wondering which um LLM model are you guys using [35:35] or are you you know trying to find one on your  own? I couldn't hear because of the echo. Oh, [35:42] which which LLM model that we're using? It's a  little early to know which right now. Uh it's I [35:47] don't think it requires frontier like a you know  this month's frontier. It's mainly conversational [35:53] fluency and and being able to handle context  a reasonable amount of context you know like [35:58] what's the current numbers on the on the user and  what's the history of the conversation. Yeah the [36:02] core our core algorithms are deterministic. So  th those those are very solid uh algorithms and [36:09] the other stuff you know I think we'll work out in  terms of budget and capability. It's all evolving [36:15] so fast. Yeah, we'd love to know part of it. These  kind of these kind of applications are going to be [36:20] uh we're going to have to learn with everyone  else making these kind of applications. How do [36:23] you build on top of this amazing conversational  ability that's enabled by LLMs but but also give [36:28] a product out that's almost free? Like it has to  be consumer so it's be super cheap and scalable. [36:33] So all these companies including this one have  to work out that tradeoff and how do you keep [36:38] the model model costs down while enabling that  sort of personalized touch? Do you have some [36:44] ideas in that area? Oh, really? But it sounds  like you know exactly which LL models you are [36:50] using are still like undetermined, right?  Well, I'm sorry because really bad echo, [36:56] but we don't have we haven't committed  to a model yet to which one yet. Yeah, [37:00] that's what you're asking. Thank you. Uh I met Sky  actually yesterday. Um I love the idea of this. I [37:05] just actually want to see if how I can participate  to kind of give like user experience feedback. Um [37:10] because I myself have just fragmented use of  apps that just track everything about my data, [37:16] about my health. And I think one of the biggest  missions that you guys are doing is is just being [37:21] able to have an idea of like how my health and  my metrics are going. Um, so I guess I guess I [37:27] do have a question, but how do you see agentic  aspects of this being a part of the part of [37:33] the product that you have like in terms of like  having an AI agent do something in this space? [37:40] Yeah, I take that and it's a good question about  how does how does an agent wrapper around AI [37:45] uh work here? Uh yeah, this this model is that  the assistant is always around 724 and so it's on [37:51] a loop of some kind checking and it might even  do proactive like hey I noticed you didn't do [37:56] X you might want to do Y that kind of thing. So  it's going to be agented to do that that kind of [38:01] use case. Um and in general it needs to be on tap  like whenever they want to make a phone call just [38:05] call the agent. Yeah guys, thank you so much. Come  and talk to us after if you have more questions.