Study at the Swiss Federal Institute of Technology Lausanne (EPFL)

How well does generative artificial intelligence like ChatGPT perform in exams? This question was investigated by a team at EPFL led by Antoine Bosselut. They presented the GenKI variants GPT 3.5 and GPT 4 with exam questions from 50 courses covering a wide range of STEM disciplines, including computer science, mathematics, biology, chemistry, physics and materials science. Result: GPT 4 answered an average of 65.8 percent of the questions correctly when formulated in the style of an AI layperson. With a better prompting strategy, the machine even achieved 85.1%. According to the researchers, the results speak for a revision of the assessment design at degree course level in higher education.

more