Towards Modeling Learner Performance with Large Language Models