Which LLM is this using because it is providing hallucinated facts that differ from ChatGPT 4o on the latest model?