Evaluating Large Multimodal Models for Inclusive Mathematics Education

This study investigates the potential of Large Multimodal Models (LMMs) in generating educational materials tailored to students with learning disorders, specifically dyslexia and dyscalculia, in higher education mathematics courses. By analyzing outputs from ChatGPT o1, Llama 3.1, Phi4, DeepSeek r1, and Claude Sonnet 3.5, this research evaluates the effectiveness of AI-generated learning resources in maintaining both mathematical correctness and pedagogi-cal usability. Using a teacher-centered prompting approach, we request model-generated adaptations for problem-solving lessons on limits and derivatives. Expert evaluators assess the outputs through a Likert scale to determine their alignment with inclusive teaching strategies. The findings provide insights into the strengths and limitations of LMMs in fostering accessible and pedagogically sound mathematics education while emphasizing the need for AI-human collaboration in instructional design.