Obviously, error rate of ChatGPT should be measured by comparing to human diagnoses and when it plateaus, we should investigate a possibility that the number is driven mostly by human errors.