College Exam Grader using LLM AI models
Jung X. Lee, Yeong-Tae Song
Abstract
By far, the most effective knowledge assessment in college education is to give students exam and grade their answers then assess their level of understanding. However, exam grading can be time-consuming, tedious, cumbersome, and sometimes the grading results are not consistent with the rubric. Here, we propose an AI based exam grader that can not only ease educators’ burden but also produce accurate, consistent, and precise grading results. We have used GPT-3.5, GPT-4.0, and Gemini-pro, respectively, as our grading engine. To verify the correctness, precision, and accuracy of our proposed grader, the results were compared with the instructor’s grading result and also with human grader such as teaching assistants. In our experiment, GPT-4.0 showed the most reliable and consistent results.