The Use of Large Language Models in Education
Wanli Xing, Nia Nixon, Scott A. Crossley, Paul C. Denny, Andrew Lan, John C. Stamper, Zhou Yu · International Journal of Artificial Intelligence in Education · 2025
Large language models (LLMs) are based on deep neural networks and are often designed using transformer architectures.These models consist of hundreds of millions to billions of parameters and are pre-trained with vast quantities of language data.In recent years, LLMs have achieved significant advancements across a wide range of natural language processing (NLP) tasks including language generation, summarization, comprehension, and classification (Brown et al., 2020).Contemporary models, such as GPT-4 (OpenAI, 2023) and Llama (Touvron et al., 2023), have demonstrated remarkable abilities to understand and generate human-like text and provide access, respectively, through a proprietary API service and an opensource implementation, making them versatile tools for various applications, including education.Since LLMs demonstrate transferability through inheriting semantic and contextual understanding from pretraining, they fit well in the context of learning engineering and learning analytics by providing reusable and scalable technical architectures appropriate for various subjects (e.g., math, Scarlatos & Lan, 2023, Shen et al., 2021; science, Cooper, 2023; medicine, Luo et al., 2022).Early integrations of LLMs into educational settings have demonstrated promising results to augment learning through item response and student knowledge tracing models for open-ended questions (Liu et al., 2022), socio-emotional support (Li & Xing, 2021), equity and inclusion support (Nixon et al., 2024), automatically generating