Large language model’s utility in helping pathology professionals
Hinpetch Daungsupawong, Viroj Wiwanitkit · American Journal of Clinical Pathology · 2023
We follow the topic on “Assessment of a Large Language Model’s Utility in Helping Pathology Professionals Answer General Knowledge Pathology Questions.”1 The purpose of the study was to assess the utility and performance of ChatGPT 4.0, a big language model, in the field of pathology. The responses generated by ChatGPT were scored by subject matter experts using a database of general pathology questions previously provided to pathology residents. The grading criteria emphasized accuracy, thoroughness, and the potential time savings for pathologists when composing responses. The study discovered that ChatGPT performed similarly in Anatomic Pathology (AP) and Clinical Pathology (CP). However, it was discovered that a higher percentage of relevant material was omitted in AP answers than in CP ones. While the study emphasizes the potential benefits of employing ChatGPT in pathology, it is vital to recognize the research’s limitations and potential biases. The study relied on a deidentified inquiry database, which may not fully represent the breadth and complexity of pathology inquiries seen in real-world circumstances. Subjectivity and potential variances in the judgment of accuracy and completeness are introduced by the grading process by subject matter experts. Furthermore, the study does not provide concrete examples of questions or responses, making evaluation difficult. Furthermore, the study focuses exclusively on the correctness and completeness of the responses, without taking into account any potential restrictions or errors that may develop as a result of relying on an artificial intelligence model. It is vital to assess the reliability and potential biases of ChatGPT, as well as the necessity for human pathology expertise and judgment.