Framing Bias in a Large Language Model: A Diagnostic Accuracy Study of Prompt Effects on ChatGPT’s Melanoma Classification

- Mendeley Supplemental Figure 1. Representative examples of the test. Each image was presented six times under different instructions: a neutral baseline prompt, and five framed prompts.
- Mendeley Supplemental File 1. Detailed Materials and methods (Study design, Model Access and Interaction, Dataset Details, Prompting Procedure, Outcome Measures and Statistical Analysis) of the study