Large language models validate misinformation