A Meta-Model Analysis of Students’ Autonomous Interaction With General-Purpose LLMs

With the rise of general-purpose Large Language Models (LLMs) tools like ChatGPT, Gemini, and Copilot, students and teachers are increasingly integrating these technologies into learning, raising concerns about their impact on education. This study examines, over the course of an entire course module, how students learn through the use of noneducational LLMs, and introduces a meta-model that maps theses uses onto structured learning stages. A mixed-methods study was conducted with fifty-one engineering students, using surveys, semi-structured interviews and qualitative coding analysis to identify key LLMs usage patterns. Results indicate that 92% of students use LLMs autonomously in academic contexts, primarily for writing assistance, information retrieval, and ideation. The application of the meta-model shows that students use LLMs more or less depending on the pedagogical formula used by the teacher. They are more often used for homework and teamwork than in simulations or lectures. It also shows that students use LLMs for high cognitive tasks that require critical thinking, computational thinking and problem-solving skills. Future research should measure non-educational LLM’s actual impact on these skills and explore cross-disciplinary differences in LLMs adoption.