Fine-tuning Large Language Models for Educational Applications: Use Cases, Architectures, and Techniques

Large Language Models (LLMs) have significantly expanded the breadth of artificial intelligence applications in education by automating or facilitating complex tasks such as question generation and feedback provision. However, general-purpose LLMs may not meet the pedagogical and domain-specific demands of educational tasks. Fine-tuning has emerged as a possible solution to bridge this gap. This article provides an overview of how different LLM architectures and fine-tuning techniques have been applied to the full spectrum of educational tasks. We provide examples from recent literature to illustrate these methods, their effectiveness, the data available for training, and the evaluation methods. Our synthesis reveals that fine-tuning can enhance LLM performance in educational tasks. However, such improvement is constrained by technical, infrastructural, and theoretical limitations. We also conclude that current implementations have so far fallen short of providing personalized education at the individual level, and that there are difficulties keeping pace with the advancements of the industry.