a16z Podcast: Can You Trust ChatGPT for Health Advice?

SPEAKER_1 (0:01)

So I was listening to the a16z Podcast with Karan Singhal from OpenAI, and he made this fascinating case for why ChatGPT could become a new layer of health intelligence. We're talking about a $5.3 trillion healthcare system that's completely overloaded, and people are desperate for help anywhere they can find it. The scale is pretty staggering. Singhal mentioned that more than 300 million people already use ChatGPT for health and wellness questions every week. That tells you something about the need, right? Even in places with excellent care, patients face long waits, short visits, and way too many unanswered questions. Exactly. And he shared this powerful example about a family who used AI to understand that their child's fatigue and irritability actually pointed to aggressive leukemia, even after doctors initially downplayed it.

That story really captures the broader promise here: AI helping people interpret lab reports, prepare for visits, and track next steps between appointments.

That's a great example, but it also raises the trust issue. How does OpenAI think about that?

Singhal actually says health moves at the speed of trust. They work with hundreds of experts across dozens of specialties to shape how the model behaves. The key principle is when the model is uncertain, it should say so. And when it disagrees with a physician, the advice isn't to replace the doctor, but to seek a second opinion.

That distinction matters. Building on that point, I found it interesting that Singhal argues health AI shouldn't be narrow or frozen. OpenAI is training general models with specialized healthcare data so they can keep pace with frontier improvements in reasoning and scale. Right, and he points out that medical guidance changes incredibly slowly in practice. It can take five to 17 years for new research to make its way into everyday care. He wants to compress that timeline dramatically. Which brings us to deployment. Singhal describes three pillars: making the model useful, building the right interface, and securing enterprise partnerships. HIPAA compliance, privacy, and integrations like the one with Epic are essential for AI to work inside real healthcare workflows. What do you think about his distinction between medical information and medical care? It's crucial. AI isn't a substitute for physical exams, bedside empathy, or clinical liability. But it can act as a quarterback, helping patients schedule visits, manage insurance friction, and catch medication issues before they escalate. Looking ahead, Singhal is optimistic about safer care, better coordination, and even scientific discovery. Tools that could spot antibiotic overuse, support mental health benchmarks, and accelerate drug discovery and clinical trials.

The bigger vision is really about making reliable intelligence abundant so doctors and patients can spend more time on the human side of medicine. His mission is to raise the floor of human health through trustworthy, abundant, and reliable health information.