Performance of Generative Artificial Intelligence Models (GPT-4o, Gemini, Copilot) in YKS/TYT Exam: A Comparative Study

Abstract

This study attempts to evaluate the potential of generative artificial intelligence (GAI) models in the field of education. A simulated exam environment was planned and error analysis was performed based on the responses of GPT (Generative Pre-trained Transformer) family models to questions asked in university entrance exams. In the study, GPT, Gemini and Copilot models were tested with questions in the Higher Education Institutions Exam (YKS)/ Basic Proficiency Exam ( TYT) session in Turkey. The responses produced by these models were analyzed in terms of both accuracy rates and academic depth and consistency. While GPT-4o achieved 50.8% accuracy, Gemini 43.3% and Copilot achieved 42.5% accuracy. The findings provide important clues about the potential of GAI models to be used as a learning aid in education and whether they can be evaluated as a supportive tool in university exam preparation processes. The study sheds light on the possibility of using GAI models in the personalized education field that emerged with the rise of GAI models in the field of education. Further studies in this field may help develop new strategies for the integration of GAI models into the education system.

Ähnliche Arbeiten