| Sumario: | Background: In recent years, the integration of large language models has brought significant opportunities for advancing personalised learning in higher education. However, little attention has been paid to how students of different performance levels benefit from such tools, especially in creative disciplines. Objectives: This study explored how two generative AI models, ChatGPT 4 and DeepSeek, support high‐ and low‐performing college design students in personalised learning and influence their AI‐assisted design performance, creative self‐efficacy, and engagement. Methods: This study adopts a comparative quasi‐experimental design to examine the differential effects of two generative AI tools, ChatGPT and DeepSeek, on personalised learning outcomes in design education. Rather than serving as a passive control, both tools function as AI‐assisted learning conditions with distinct interaction styles and feedback structures. Results: Results show that, compared with conventional creative methods, personalised learning based on both GAIs significantly improves student outcomes. ChatGPT users demonstrated higher creative self‐efficacy, while DeepSeek users exhibited greater engagement and AI‐assisted design performance. Although self‐efficacy and engagement were influenced by students' initial performance levels, AI‐assisted design performance improved consistently across groups. Semi‐structured interviews revealed that high‐performing students used ChatGPT to expand imagination and experiment with abstract concepts, while low‐performing students benefited from DeepSeek's structured prompts for clearer direction. Conclusions: These different effects can be understood using Cognitive Load Theory, which shows that DeepSeek's organised outputs helped low‐performing students focus better, while ChatGPT's open‐ended feedback encouraged creative thinking for high‐performing learners. Overall, the findings underscore the necessity of differentiated AI integration strategies in design education, taking into account learner characteristics and instructional goals. Summary: What is currently known about this topic? ○Generative AI, such as ChatGPT, has been shown to improve personalised learning outcomes in higher education, including creativity, motivation, and academic achievement.○Existing research has largely emphasised overall effects but overlooked the differences among students with different performance levels (high vs. low performers).○In design education, creativity and self‐efficacy are crucial, but research on how GAI can simultaneously impact these two types of students is still in its infancy.What does this paper add? ○This study is the first to compare the differential effects of two different GAI tools (ChatGPT and DeepSeek) in design learning.○The results showed that ChatGPT was more effective in enhancing creative self‐efficacy, while DeepSeek was more effective in enhancing AI‐assisted design performance and engagement.○Students' prior performance level moderated self‐efficacy and engagement but did not affect the overall improvement in AI‐assisted design performance.Implications for practice or policy ○Instructors should consider adopting differentiated AI integration strategies○Instructional design should be tailored to student characteristics.○Policies should emphasise equitable AI integration and encourage universities to establish standardised and diverse AI teaching application models.
|
|---|