As AI Meets Gaokao: Fine-Tuning GPT-4o for Enhanced Performance in Objective Questions

This study explores the application of artificial intelligence (AI) in education by fine-tuning a large language model, GPT-4o. We fine-tuned GPT-4o and proposed an integrated prompt engineering framework to accommodate the diverse formats and cognitive demands of Gaokao questions. Experimental results show that the fine-tuned GPT-4o achieves an accuracy of 84.04% on objective questions, demonstrating substantial gains in language-related tasks. However, performance remains limited in science subjects requiring multi-step logical reasoning and precise numerical computation. These findings highlight both the promise and limitations of LLMs in educational contexts. This work offers practical insights into optimizing AI for academic use and lays a foundation for future development of intelligent, personalized learning tools aimed at reducing educational inequality and improving global access to quality education.