Large language model uncertainty proxies: discrimination and calibration for medical diagnosis and treatment
SC is the most effective method for estimating LLM uncertainty of the proxies evaluated. SC by sentence embedding can effectively estimate uncertainty if the user has a set of reference cases with which to re-calibrate their results, while SC by GPT annotation is the more effective method if the user does not have reference cases and requires accurate raw calibration. Our results confirm LLMs are consistently over-confident when verbalizing their confidence (CE).
Authors
- Jonathan H. Chen (ORCID: https://orcid.org/0000-0002-4387-8740)
- Robert J. Gallo (ORCID: https://orcid.org/0000-0002-2601-0173)
- Ali Soroush (ORCID: https://orcid.org/0000-0001-6900-5596)
- Seyed Amir Ahmad Safavi‐Naini (ORCID: https://orcid.org/0000-0001-9295-9283)
- Abdessalem Boukil
- John Wang (ORCID: https://orcid.org/0000-0002-8515-5014)
- Vishwesh Patel
- Thomas Savage (ORCID: https://orcid.org/0000-0003-4828-5802)
Institutions
- Palo Alto Veterans Institute for Research (US)
- Stanford Medicine (US)
- Government Medical College (IN)
- M. P. Shah Medical College (IN)
- University of Sousse (TN)
- Stanford University (US)
- Icahn School of Medicine at Mount Sinai (US)
Publication Details
- Journal
- Journal of the American Medical Informatics Association
- Published
- 2024-10-12
- DOI
- https://doi.org/10.1093/jamia/ocae254
- Citations
- 41
- Primary Topic
- Artificial Intelligence in Healthcare and Education
- Type
- article
- Field-Weighted Citation Impact
- 4.93
Funders
- U.S. Department of Veterans Affairs
- Gordon and Betty Moore Foundation
- American Heart Association
- Georgia Clinical and Translational Science Alliance
- National Institutes of Health
- National Institute of Allergy and Infectious Diseases
- National Center for Advancing Translational Sciences