I’m bullish on RAG since you’ll always need a way to work with new information without retraining or fine tuning an LLM. Even as humans we essentially do RAG with google