PHYSICS LEARNING BASED ON CONCEPTUAL ERROR PREDICTION USING LARGE LANGUAGE MODEL AS A COGNITIVE ASSISTANT FOR SCHOOL STUDENTS

Abstract

Physics learning based on the prediction of conceptual errors is a critical aspect of physics education, as persistent student misconceptions often hinder the development of deep conceptual understanding. This challenge highlights the need for adaptive learning approaches capable of identifying and anticipating students’ conceptual errors at an early stage. This study aims to examine the effectiveness of implementing a Large Language Model as a cognitive assistant in predicting conceptual errors and enhancing students’ conceptual understanding and learning engagement. The research employed an experimental method with a quasi-experimental design. Data were collected through conceptual understanding tests, learning engagement observation sheets, and instructional documentation, and were analyzed using descriptive statistics and inferential analysis through an independent samples t-test. The results indicate that the experimental group achieved a higher mean post-test score in conceptual understanding than the control group (M = 81.9; SD = 6.7 vs. M = 68.1; SD = 7.4), with a statistically significant difference (t(58) = 6.05, p < 0.001) and a strong effect size (Cohen’s d = 0.93). In addition, the level of conceptual errors among students in the experimental group decreased significantly, while learning engagement was classified as high (M = 4.18) compared to the control group (M = 3.34). These findings demonstrate that a Large Language Model can function effectively as a cognitive assistant by predicting patterns of conceptual errors and providing adaptive scaffolding in real time. This study makes a significant contribution to the development of artificial intelligence–based physics learning models, extends the application of constructivist theory and cognitive load theory within intelligent digital learning environments, and enriches the international literature on the use of Large Language Models in science education. Furthermore, the findings are expected to serve as a reference for the development of adaptive physics learning and for future research on integrating Large Language Models as pedagogical agents at the school level.