A company wants an FM to answer from policies updated every day and cite the current source. Which approach is the best initial baseline?

1. Fine-tune once and assume weights track every future policy change.
2. Increase temperature to generate newer policy versions.
3. Use only a static prompt containing obsolete policies.
4. RAG with a maintained policy corpus and evaluated retrieval.