Roxana Daneshjou MD/PhD
@RoxanaDaneshjou
Assistant professor of Biomedical Data Science @StanfordDBDS and Dermatology @StanfordMed | AI/ML & precision health | @Rice_BioE alum | @pdsoros Fellow 2014
ID:20898564
https://profiles.stanford.edu/roxana-daneshjou 15-02-2009 07:28:31
24,8K Tweets
21,2K Followers
4,8K Following
A good point from Roxana Daneshjou MD/PhD. When writing a radiology (or other healthcare) LLM paper that requires prompts, how should the Methods be constructed? Should the prompt engineering method be explained? Tested? Comes up often in radiology.
Maria (Maria Antoniak) is a rock-star researcher, and it was a dream to work with her, Aakanksha Naik ✈️ EMNLP 2023, Carla S. Alvarado, and Lucy Lu Wang on this project rethinking LLMs for maternal health -- with lessons for other medical contexts
Chatting with Cyril Zakka, MD about the lack of rigor in LLM/VLM research for healthcare. If you run a prompt multiple times and use the best output for calculating your metric (accuracy, sensitivity, etc), that feels equivalent to p-hacking.
Large language models (LLMs) like ChatGPT can propagate false race-based medical information. Work by Roxana Daneshjou MD/PhD nature.com/articles/s4174…