Interview, Fireside Chat
Should You Trust ChatGPT With Your Health?
- OpenAI anticipates resolving security and compliance bottlenecks, including achieving HIPAA compliance, within approximately two years to enable enterprise offerings, while noting that robotics and direct physical care integration are longer-term developments.
- In the short term, AI is expected to augment rather than replace clinical workflows by facilitating navigation, logistics, and proactive flagging of issues like rising A1C levels to prevent expensive deterioration, with a specific focus on the "last mile" of care between doctor visits.
- The timeline for clinical guidelines to penetrate practice is predicted to compress from the current 5 to 17 years down to weeks, days, or hours by integrating biomedical literature directly into the moment of care.
- Within the next five to ten years, AI is expected to drive significant impacts through the generation of new targets or molecules and the prioritization and validation of clinical trials, accelerating the validation bottleneck in drug discovery.
- Cost trends indicate a rapid reduction in health intelligence expenses, with models like GPT-6 Luna predicted to outperform previous iterations like GPT-5.6 on benchmarks while being 34 times cheaper, eventually reaching a point where the cost is "too cheap to meter."
- The industry is expected to shift away from adversarial billing processes toward a "handshake" of ground truth adjudicated by AI, alongside the emergence of ambient intelligence in clinical rooms to transcribe notes and eliminate friction.
- While health intelligence currently lacks a macro effect on the $5.3 trillion total spend, the speaker expects overall compression of spend over the coming years through widespread integration and the commoditization of health information.
- Safety and trust are projected to increase as the "Overton window" shifts, with specific efforts to establish shared benchmarks with licensed mental health experts to measure performance and ensure models can recognize their own uncertainty.
- The vision includes the deployment of proactive personal health assistants acting as a "quarterback" for scheduling and logistics, supported by agents capable of analyzing existing longitudinal and EHR data to generate new insights.
- The speaker foresees "Chastity" (ChatGPT) for clinicians undergoing dramatic improvements in performance and reliability, with a target of 300 million weekly active health users as a starting point for abundant, free or nearly free access.
- Entrepreneurs are advised to build on improving models rather than waiting for perfection, as the activation energy for using these tools will decrease through product design addressing current security concerns.
- The speaker predicts that AI will help identify "Move 37" moments—predictions that seem brilliant in hindsight but were previously impossible for humans to integrate—effectively raising the ceiling of human health capabilities.