Large Language Models for Education

This thesis explores the use of large language models (LLMs) in specialized domains, with a particular focus on educational content found in LaTeX- formatted textbooks. The introduction highlights the importance of fine- tuning LLMs for domain-specific tasks and emphasizes the significance of preprocessing data from specialized sources like textbooks. One of the key objectives is to demonstrate how fine-tuned LLMs, tailored to educational content, can effectively address the limitations of LLMs pre- trained on general text, especially in responding to single-choice questions. Additionally, the thesis examines the practical aspects of deploying LLMs, comparing the resource consumption of large pretrained models to smaller fine-tuned ones, offering insights into performance-efficiency trade-offs. In summary, this thesis aims to contribute to the field of natural language processing by exploring the adaptation of LLMs to educational content and addressing their limitations. The research is structured into three chapters, each focusing on distinct aspects, and concludes with reflections on future directions in this evolving field.