Are Large Language Models Good Statisticians?