UTILIZATION OF GENERATIVE ARTIFICIAL INTELLIGENCE AS AN ARITHMETIC LEARNING ASSISTANT: INTERACTIVE STEP-BY-STEP INSTRUCTION STRATEGY

• Main Author: Winanda Adliya, Difa
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83321
• Institution: Institut Teknologi Bandung

This undergraduate thesis research discusses the utilization of generative artificial intelligence for creating an AI chatbot as a mathematics learning assistant. The testing of this chatbot in this thesis is limited to arithmetic lessons. The presence of generative artificial intelligence poses a unique challenge in the field of education today. The background section explains that the AI chatbot as a mathematics learning assistant should be able to provide step-by-step guidance interactively. This thesis research employs prompt engineering, fine-tuning, and Retrieval Augmented Generation (RAG) methods on the GPT-3.5 language model to create an AI chatbot as a mathematics learning assistant. The prompt engineering method used in this research demonstrates that providing the main question context and step-by-step solutions to the GPT-3.5 model can significantly enhance the chatbot’s response quality. In this thesis, it is shown that the accuracy of the GPT-3.5 model on the Multi- Arith dataset is 96.03%. With fine-tuning methods, the accuracy of the fine-tuned GPT-3.5 model on the MultiArith dataset is 99.48%. Additionally, the step-by-step solutions generated by the fine-tuned GPT-3.5 model are more structured than those of the base model. This thesis also explains the use of the RAG method for retrieving step-by-step solutions from pre-prepared documents. With the writing of this thesis, the author hopes that this research can assist teachers in creating AI chatbots as mathematics learning assistants.