Generative Feedback for Code Learning: A Study on LLM-Driven Formative Assessment

Abstract

This study investigates the integration of large language models (LLMs) into programming education, emphasizing their role in enhancing learning motivation, teaching efficiency, and formative feedback. On the student level, frequent LLM usage significantly improved learners' programming comprehension, self-regulation, and problem-solving willingness. On the instructor level, LLMs alleviated repetitive workloads, enabling teachers to shift their focus toward creative guidance and interdisciplinary design. Correlation analysis confirmed a strong positive relationship between LLM interaction frequency and student performance, affirming the pedagogical efficacy of AI assistance. A closed-loop mechanism was established, integrating students, teachers, and AI into a feedback ecosystem driven by student behavior, real-time model response, and adaptive instructional strategies. Through case studies, metric-based evaluation, and interactive logs, this research highlights the transformative potential of LLMs in building a dynamic, data-informed programming education framework.