Trends of Using AI Services Between Students of Educational Study Programs at the University of Maribor

Since the widespread availability of ChatGPT, the utilization of large language models (LLMs) among students has notably changed. This study investigates LLM usage trends during 2023-2024 and 2024-2025 at the University of Maribor, covering Faculties of Arts, Education, and Natural Science and Mathematics. Initially, 20% of students subscribed to LLM services in 2023-2024; however, this figure significantly dropped to 4% in 2024-2025. Verification practices also decreased-from 70% to 45%-with 26% of students accepting LLM-generated outputs without verification. The primary uses of LLMs remained consistent: concept explanation, supplementary information access, and basic information retrieval. Translation declined notably from third to sixth most common application. Interestingly, LLM use did not significantly rise during examination periods. Subscription rates varied by year of study, with fourth-year students highest at 9%, followed by fifth-year at 7%. These shifts indicate changing perceptions of LLM reliability, emphasizing the urgency of training educators to integrate AI effectively. Furthermore, this evolving reliance underscores the need to strengthen students' critical thinking in AI-assisted learning environments.