Comparing Large Language Models on Unfair Clause Detection in Terms of Services