On Possibilities and Challenges of GPT‐Assisted Learning Environments

Contemporary education demands a student‐centered and competence‐oriented approach. Learners must tackle and solve problems independently and in line with constructivist learning theories. Accordingly, instructional roles must evolve from mere knowledge transmission to facilitating students' autonomous achievement of learning objectives—a paradigm often described as the shift from “teaching to learning.” This, however, leads to a seemingly insurmountable problem: As class sizes grow, lecturers quickly reach the practical limits for personalized support. Moreover, consistent with the principles of “Universal Design for Learning,” students benefit most when assistance is available whenever they find the time to work independently on specific topics or challenges. One promising solution is the deployment of generative pre‐trained transformers (GPTs). ChatGPT, a prominent large language model (LLM), has demonstrated remarkable capabilities for knowledge delivery. Furthermore, it enables the creation of custom GPTs tailored to particular tasks, suggesting their potential as on‐demand support for students. In this contribution, we first formalize the behavior an LLM must exhibit to effectively function as personalized support, grounded in contemporary constructivist learning theories. We then conduct an exploratory case study to explore to what extent current GPT models can meet these criteria. In this case study, we evaluate both the strengths and limitations of using an unmodified ChatGPT model for university‐level mechanics problems. Without directed prompting, the model struggles to align with constructivist principles and can be misled. Additionally, its inherent and intentional randomness (controlled via the temperature parameter) may yield inconsistent responses to identical queries. We demonstrate that by developing a customized GPT—with targeted instructions and suitable data–these issues can be addressed. In our case study, the model shows promising results suggesting that reliable support for students in accordance with constructivist theory at an expert level may be achievable.