Utilizing Adaptive Learning with GPT-4 for Introduction to Python: Effects on Accuracy and Time-to-Solution

Santivanez D.; Oliva K.; Wong L. · Engineering, Technology and Applied Science Research (2026) · DOI: 10.48084/etasr.13246

Resumen

Programming instruction in universities often struggles to adapt to individual needs and sustain motivation. This study presents Code Showdown, a web application that combines adaptive learning with GPT-4 to strengthen Python programming skills through automatically generated challenges and immediate personalized feedback. The development followed four phases: (i) selecting the pedagogical approach, (ii) choosing the language model, (iii) integrating GPT-4 for challenge generation and formative feedback, and (iv) building a scalable web architecture. The proposed approach was evaluated with 40 undergraduates (two groups of 20 individuals) over two weeks. Group E1 studied with traditional resources, whereas Group E2 used Code Showdown in individual and multiplayer modes. Using explicit metrics—Improvement Score (IS, %) and Improvement in Response Time (IRT, %)—Group E2 achieved higher means across difficulty levels: IS of 18.9±4.2% (Easy), 40.7±4.7 % (Intermediate), and 58.6±5.1 % (Hard), and IRT reductions of 32.4±6.3%, 28.0±5.8%, and 28.3±6.1 %, respectively. A t-test showed that these differences were statistically significant (p<0.05). Satisfaction was high (ISO/IEC 25010-based survey), with more than 80% positive ratings and a recommendation score of 4.9±0.3/5. These findings suggest that adaptive, GPT-4-assisted practice may enhance accuracy and efficiency in introductory Python while maintaining engagement through gamification and real-time competition. Licensed under a CC-BY 4.0 license | Copyright (c) by the authors

Instituciones: Universidad Peruana de Ciencias Aplicadas