Should You Trust ChatGPT With Your Health?

影片總結

OpenAI’s Karan Singhal says ChatGPT reaches 300 million weekly health users, while AI could help cut care delays and costs.

OpenAI health AI lead Karan Singhal tells a16z’s Jay Rughani that more than 300 million people use ChatGPT for health and wellness each week, often to prepare for visits and interpret results between appointments. He recounts Nathan Levent’s account of AI helping his family navigate symptoms that led to his young son’s aggressive leukemia diagnosis. Singhal says OpenAI trains healthcare expertise into general-purpose models and evaluates them with hundreds of physicians, prioritizing safety when users want reassurance but need urgent care. He describes proactive health assistants as companions to doctors, not replacements, and points to Epic integration and an AdventHealth report of roughly 80% lower administrative workflow costs. He also notes that medical guidance can take 5–17 years to reach everyday practice, while AI may help surface evidence faster, reduce paperwork, and flag risks such as antibiotic overuse. Looking ahead, he expects preventive care and AI-assisted drug discovery and clinical-trial prioritization to improve access and treatment.

章節

1. 0:00Intro: $5.3 trillion in healthcare spending and OpenAI’s health-intelligence mission
2. 1:00Nathan Labenz’s son: GPT-5 Pro helped navigate an aggressive leukemia diagnosis
3. 7:57A doctor in your pocket: Penda Health’s medication-interaction co-pilot
4. 14:20Teaching AI who it’s talking to: Healthbench and feedback from hundreds of physicians
5. 19:44Medical consensus and its 5–17-year lag: bringing current research to care
6. 24:09Medical consensus and its 5–17 year lag: Epic integration and healthcare adoption barriers
7. 28:03The mental health problem: OpenAI’s Mental Health Bench measures positive outcomes
8. 30:59Information vs actual care: UCSF, AdventHealth’s 80% savings, and chronic-care support
9. 38:21A “Move 37” for health: Panda Health’s antibiotic alert and insights from longitudinal data
10. 43:39Driving down the cost of care: GPT-6 Luna, 34× cheaper performance, and workflow savings
11. 50:27Procedural care and what's next: AI prevention, robotics timelines, and personalized medicines