Đánh giá hiệu suất giải toán xác suất của ChatGPT (phiên bản GPT-5) dựa trên các bộ sách Toán 12 theo Chương trình 2018

Tóm tắt

Conditional probability is a central topic in the Grade 12 Mathematics curriculum, demanding a high level of logical reasoning. Amid the rapid advancement of large language models in education, this study evaluates the accuracy and reasoning quality of GPT-5 on Grade 12 probability problems through a mixed-methods approach - combining quantitative and qualitative analysis - applied to a dataset of 513 problems drawn from current Grade 12 Mathematics textbooks. Accuracy rates were statistically quantified, while qualitative analysis identified characteristic reasoning patterns and error types. The findings indicate that GPT-5 correctly solved 84.01% of the textbook problems. The model demonstrated proficiency in formula application but exhibited notable difficultiesin reasoning, particularly in tasks requiring the construction and interpretation of tree diagrams. These findings provide teachers with empirical evidence regarding the reliability of GPT-5 as a instructional tool. The study recommends extending the evaluation to other mathematical topics, investigating student–ChatGPT interactions in authentic learning contexts, and conducting controlled experimental studies to establish causal evidence for the pedagogical integration of AI in mathematics education.