Work

A comparison of the suitability of items generated by 4 large language models for pathology continuing certification