Addressing mistakes of ChatGPT from an educational perspective
View Abstract View PDF Download PDF

Keywords

Artificial intelligence, ChatGPT, Educational application, Large language model, Mistakes.

How to Cite

Leung, C. H., & Chan, W. T. Y. . (2025). Addressing mistakes of ChatGPT from an educational perspective. Asian Journal of Contemporary Education, 9(1), 11–26. https://doi.org/10.55493/5052.v9i1.5301

Abstract

This paper explores the efficacy of ChatGPT, a generative artificial intelligence in educational contexts, particularly concerning its potential to assist students in overcoming academic challenges while highlighting its limitations. ChatGPT is suitable for solving general problems. When a student comes across academic challenges, ChatGPT may conveniently provide a solution. However, ChatGPT may create mistakes in its responses that become detrimental to the learning process. The study discusses the technical principles underlying ChatGPT and identifies factors that contribute to incorrect responses. It categorizes various errors such as outdated information, mathematical inaccuracies and cognitive biases illustrating these with concrete examples. The analysis reveals that although ChatGPT is trained on extensive datasets, it can produce statistically plausible but factually incorrect answers. This can hinder the learning process and lead students astray if they rely solely on its outputs. The paper offers practical recommendations for students and teachers to mitigate the risks associated with using ChatGPT. These include verifying sources of information, formulating questions in diverse formats, supplementing queries with additional context, and reporting inaccuracies to OpenAI. Implementing these strategies can enhance the educational utility of ChatGPT while minimizing potential pitfalls. Understanding both the strengths and weaknesses of ChatGPT is a proper way of using a new technology efficiently.

https://doi.org/10.55493/5052.v9i1.5301
View Abstract View PDF Download PDF

Downloads

Download data is not yet available.