Figure 2 | Scientific Reports

Figure 2

From: The model student: GPT-4 performance on graduate biomedical science exams

Figure 2

GPT-4 Performance on 9 graduate examinations in the biomedical sciences. GPT-4 using zero-shot or persona prompt patterns met or exceeded the average performance of graduate students in 7 courses in the biomedical sciences. The horizontal bar represents the student average score in the course and ranges represent sample standard deviation in student score values. In 7/9 cases, either the GPT4-Simple or GPT4-Expert prompt pattern met or exceeded average student performance in the course. Table Legend—Blinded: All GPT-4 exam grading was performed blinded in parallel with student assessments; GPT-4 > All: One or more GPT-4 scores exceeded all student scores; GPT-4 ≥ Av.: One or more GPT-4 scores exceeded the average student score. *GPT4-Simple and GPT4-Expert prompt patterns both received initial grades of 100% and final grades of 0% for the genomics and bioinformatics course based on suspected plagiarism due to similarities between answers.

Back to article page