Language Models as Fact Checkers?