Can Large Language Models Play Text Games Well? Current State-of-the-Art and Open Questions