Nathan C. Hurley

Hello! I am an anethesia resident at The University of Wisconsin, Madison.

I earned my MD and PhD at Texas A&M University in 2022. I studied in the lab of Bobak Mortazavi and defended my dissertation, “Deep Semi-Supervised and Multi-Stage Learning for Medical Applications” in May 2021.

My research focuses in graduate school were the development and application of machine learning algorithms to estimate probability of specific patient outcomes. I also explored the discovery of phenotypes within large heterogeneous populations and in elucidating the relationships that these phenotypes have on outcomes.

My current work is in exploring the amazing potential of Large Language Models (LLMs) for applications in medicine. The invention of LLMs has the potential to be as big of an invention as the internet. How do we safely use these in medicine? It would be irresponsible not to use such a powerful tool to improve patient care. However, irresponsibly using LLMs for clinical decisions could directly cause patient harm. How do we train and validate models so that we can trust them with decisions that have real-world applications?

To see some of my work in attempting to probe how far we can trust these models, see this repo where GPT-3.5 performs poorly at recommending for or against neuraxial anesthesia or this repo where several LLMs perform surprisingly well at several transfusion-related tasks].