Abstract
This study analyzed errors and improvements in mathematical problem-solving using ChatGPT-4o to explore the potential applications of generative AI like ChatGPT in education. To achieve this, the research examined ChatGPT-4o’s accuracy across five major units of the 7th-grade ‘Mathematics 1’ textbook, identified error types based on unit characteristics, and evaluated the effectiveness of prompt engineering. ChatGPT-4o demonstrated an accuracy rate of 51.76% on major unit assessment problems from 10 versions of the 7th-grade ‘Mathematics 1’ textbook under the 2015 Revised Curriculum. Among the units, the ‘Expressions and Equations’ unit showed the highest accuracy rate, while the ‘Statistics’ unit had the lowest. In terms of error types, ‘problem recognition errors’ were most common in the ‘Numbers and Operations’ and ‘Expressions and Equations’ units. ‘Graph recognition errors’ were predominant in the ‘Coordinate Plane and Graphs’ and ‘Statistics’ units, while ‘diagram recognition errors’ were frequent in the ‘Geometry’ unit. Prompt engineering improved accuracy for identical problems but had little effect on similar problems. A comparative analysis with earlier models also revealed that ChatGPT-4o showed improved accuracy. Despite current limitations, the rapid development of ChatGPT suggests its potential for use in classroom settings in the near future, emphasizing the need for further research.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call