Preliminary evaluation of the potential of commercially available large language models in diagnosing skin tumours

Abstract

To date, the most commercially available large language models have refused to evaluate clinical pictures because of the medical contexts. In our study, only BingAI Creative mode provided decisions about whether images were malignant or benign and provided a diagnosis for skin tumours, both with relatively low accuracy rates of 58% for classification and 3% for diagnosis.

Topics & Concepts