Influence of a Large Language Model on Diagnostic Reasoning: A Randomized Clinical Vignette Study