Development of a Student Tutoring Chatbot by Instruction Tuning a Publicly Available Large Language Model

The higher education landscape is constantly evolving thanks to technological innovation, and the use of advanced language models represents one of the most promising frontiers in machine learning. This research project focuses on the adaptation and use of modern publicly available large language models in order to create highly efficient automatic tutoring tools for university students. The main objective is to focus on supporting foundational teachings, with the aim of providing a solid learning basis to students and facilitating their study process. The central core of this research project consists in training and specializing a large language model, previously trained on very large textual corpuses, so that it is able to provide relevant and in-depth answers to student questions relating to the topics covered during the course lectures. This adaptation process involves the integration of specific data from university courses, in order to refine the model's ability to understand academic contexts and technical terminologies. The automatic tutoring approach based on large language models offers several opportunities and advantages. First of all, it allows students to access personalized support available at any time, thus helping to fill any conceptual gaps or clarify doubts. Furthermore, the automatic tutoring system can be designed to offer detailed explanations, practical examples and additional resources, enriching the students' learning experience. However, this project also faces significant challenges. The correct interpretation of student questions, the production of coherent answers and the management of ambiguity are complex aspects that require careful training and particular care. Furthermore, evaluating the effectiveness of the automatic tutoring system is a crucial step, as it requires collaboration between experts in pedagogy, computational linguistics and university teachers. To achieve the goals of this project, key steps will be followed, including collecting and annotating training data, designing appropriate evaluation metrics and implementing intuitive interfaces for students. It is also planned to carry out controlled experiments to analyze the effectiveness of the system and collect feedback from users. Ultimately, this project aims to exploit the potential of modern large language models to revolutionize university learning.