The whole point of a vector database seems like something you could just build into a regular database or a long context model. I mean, if the model can read everything at once, why bother with separate retrieval? My opponent probably has a fair point, but that's just my gut feeling.