Predicting Learning Performance with Large Language Models: A Study in Adult Literacy