CourseAssist: Pedagogically Appropriate AI Tutor for Computer Science Education
Project Overview
The document outlines the implementation of CourseAssist, an AI tutoring system designed specifically for computer science education, which leverages large language models (LLMs) to deliver scalable and effective student support. It highlights the necessity for pedagogically appropriate responses from AI systems, advocating for a constructive integration of LLMs into educational practices rather than outright prohibitions. CourseAssist enhances its tutoring capabilities through techniques such as retrieval-augmented generation and user intent classification, which contribute to improved relevance and effectiveness in tutoring sessions. The findings indicate that CourseAssist significantly outperforms baseline LLMs, including GPT-4, in terms of usefulness, accuracy, and educational appropriateness. This suggests that generative AI can play a crucial role in education by providing personalized assistance that aligns with pedagogical goals, ultimately fostering better learning outcomes for students in computer science.
Key Applications
CourseAssist, an AI tutoring system
Context: Computer science courses at a large public university, targeting students enrolled in programming classes.
Implementation: CourseAssist uses LLMs integrated with retrieval-augmented generation and user intent classification to align responses with course materials and learning objectives.
Outcomes: CourseAssist significantly outperformed GPT-4 in usefulness, accuracy, and pedagogical appropriateness, providing better support for student learning.
Challenges: Concerns over student overreliance on AI tools, miscomprehension of generated code, and risks of inaccurate answers.
Implementation Barriers
Ethical/Integrity
Students may over-rely on LLMs for generating code solutions, hindering their understanding and learning.
Proposed Solutions: Implementing user intent classification and pedagogical frameworks to ensure AI responses promote learning and comprehension.
Project Team
Ty Feng
Researcher
Sa Liu
Researcher
Dipak Ghosal
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Ty Feng, Sa Liu, Dipak Ghosal
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai