SimuLyte: A Generative Simulated Learner in Bloom's Taxonomy

Artificial intelligence (AI) plays an important role in education, particularly in the profiling of learners and the evaluation of instructional strategies. The scarcity of learners' data and privacy issues present ongoing challenges in AI-based modeling, as it heavily relies on existing datasets. Large Language Models (LLMs) have emerged as a promising approach in simulating human behavior to mitigate these constraints. LLM-based platforms could act as efficient solutions for assessing educational methods without real classroom interventions. In this work, we introduce SimuLyte, a simulated learner built as a customized GPT based on GPT-4o, which imitates diverse cognitive personae aligned with the revised Bloom's taxonomy. SimuLyte performed 24% better when its bloom level matched or exceeded the question's. Additionally, increased confusion level caused SimuLyte to attempt more for answering each question, while its accuracy dropped significantly after 0.6. According to our observations, SimuLyte closely reflects Bloom's taxonomy accompanied by confusion level, and could be used as a groundwork for scalable learner modeling.