Why does this get downvoted so heavily? It’s my experience running LLM in production. At scale hallucinations are not a huge problem when you have reference material.