Perceptions of large language models among predominantly STEM undergraduate students in higher education

Large language models (LLMs) are increasingly present in higher education, yet large-scale empirical evidence examining students’ perceptions, reliance patterns, and self-assessed learning-related experiences with these systems remains limited. Using a two-stage measurement and structural modeling approach, this study first identifies and validates latent dimensions of undergraduate students’ perceptions of LLM use and then tests how these perceptions are associated with student characteristics, including personality traits, gender, education level, course level, usage frequency, subscription status, and computer science major status. Survey data from a STEM-dominant institutional convenience sample of 653 undergraduate students at a large public university were analyzed using factor analyses and structural equation modeling. Results indicate that students view LLMs as valuable for understanding complex material and completing assignments, though relatively few report strong reliance on them for academic success. Most students express confidence in evaluating the accuracy of LLM outputs, with significant variation linked to personality and usage patterns. These findings highlight students’ perceived benefits and risks of LLM use and point to future research on strategies for supporting independent learning and critical evaluation, but they should be interpreted as cross-sectional self-reported perceptions rather than evidence of objective learning outcomes or causal effects.