Statistical Knowledge Assessment for Large Language Models