Stanford faculty Matt Lungren and Justin Norden host Epic's Senior Vice President of R&D Seth Hain for a discussion on how large language models are reshaping clinical medicine. The conversation covers how new systems like GPT-5 compare against earlier models in medical benchmarking, and how Epic has built specialized predictive tools, including a model called Cosmos, trained on large clinical datasets to forecast patient outcomes. The three speakers trace how these tools are moving into real clinical workflows inside hospital systems, and debate a genuine risk: that clinicians leaning on AI outputs could lose diagnostic skill over time. They also discuss what a usable AI interface looks like for a doctor mid-shift, as opposed to a developer testing a model in isolation. The tone is practitioner-to-practitioner, grounded in specific products and deployment problems rather than abstract AI hype, aimed at clinicians, developers, and healthcare administrators making near-term decisions.