Investigating the utilization and impact of large language model-based intelligent teaching assistants in flipped classrooms

The swift advancement of artificial intelligence, especially large language models (LLMs), has generated novel prospects for improving educational methodologies. Nonetheless, the successful incorporation of these technologies into pedagogical methods, such as flipped classrooms, continues to pose a challenge. This study investigates the implementation and impact of intelligent teaching assistants utilizing LLMs in flipped classroom environments to bridge this gap. We created a teaching assistant system that employs retrieval-augmented generation (RAG) and other sophisticated methods for the injection of personal knowledge and establishes an intelligent framework for the collaborative work of assistants. In a bachelor's course in computer vision, we utilized a quasi-experimental design, randomly assigning 60 students to an experimental group (utilizing intelligent assistants) and a control group (receiving traditional instruction). During a six-week period, we analyzed the disparities between the two groups in terms of academic performance, learning engagement, learning strategies, and additional factors. The findings indicate that the experimental group markedly surpassed the control group on these metrics (p < 0.05). The intelligent assistants enabled tailored instruction and improved the efficacy of flipped classrooms, while enhancing students' learning experiences and outcomes and alleviating teachers' workload. This study illustrates the capacity of LLM-driven intelligent teaching assistants to facilitate educational reform and offers innovative concepts and empirical evidence for the future advancement of intelligent education.