Is it possible to catch those hallucinations using another LLM with a strong fact checker prompt with sources provided in output for human validation?