Crawling the Internal Knowledge-Base of Language Models