Assessing the Reliability of Large Language Model Knowledge