Purpose: This study aims to evaluate the problem-solving performance of GPT-4o, a widely used large language model (LLM), on the Korean Level 2 Sports Instructor Licensing Examination (SILE), a national certification exam assessing theoretical knowledge in sports science.Methods: A total of 673 text-based multiple-choice questions from the 2020-2024 SILE were analyzed, covering seven subject areas: sports sociology, sports education, sports psychology, Korean sports history, exercise physiology, sports biomechanics, and sports ethics.Questions from each subject were presented to GPT-4o as a single prompt, following a zero-shot multi-problem evaluation format.The model's answers were compared with official answer keys to calculate accuracy by year, subject, and cognitive level.Descriptive and inferential statistics were used to analyze performance, including analysis of variance and Tukey's honestly significant difference test.Results: GPT-4o achieved an average annual accuracy of 91.2%, surpassing the 60% passing threshold in all years, with no significant differences across years (p=0.633).Subject-specific accuracy ranged from 83.0% (Korean sports history) to 98.8% (sports biomechanics).Accuracy varied by cognitive level, with the highest performance in knowledge recall and comprehension (92.4%) and lowest in calculation (72.7%).Conclusion: GPT-4o demonstrated consistently high performance on the SILE, indicating strong domain knowledge and reasoning capabilities in sports science.These results highlight the potential of LLMs for applications in sports education, certification preparation, and evaluation design.However, limitations remain in handling culturally specific knowledge and quantitative problem-solving, suggesting the need for cautious and context-aware integration into real-world assessment systems.
Park et al. (Mon,) studied this question.