[drusepth]

RAG still has lots of benefits for anyone paying per input token (e.g. over APIs).

[azinman2]

Not to mention latency

[disgruntledphd2]

And grounding for the model. Smaller models with tend to hallucinate a little less (anecdotally).