Decomposing treatment effect estimation reveals asymmetric clinical knowledge in large language models: analysis code, frozen prompts and aggregate results
Code, frozen prompts and aggregate results from a study of language-model number-needed-to-treat estimates against 539 reference values.
Shuhan He · Amal Mohamed · Cory Rohlfsen · Brian W. Locke · Brian Sheridan · Natasha Akomeah · Alex Sorrick · Andrew S. Parsons · Ali Raja · Jarone Lee · David Liebovitz