The Earth is Flat? Unveiling Factual Errors in Large Language Models