Generative retrieval pushes the corpus into the paramet...

Generative retrieval and RAG are complementary. Generative retrieval fails on private or post-cutoff data and lacks citation capabilities, while RAG-based indexes fail when query wording does not match any document posting lists. By running both and unioning the results, you can significantly lift coverage. Relying solely on generative retrieval due to model scaling is a mistake, as scale improves the model’s internal representation but does not provide the random access required for reliable retrieval.