Is Mathematical Problem-Solving Expertise in Large Language Models Associated with Assessment Performance?