are vector-embeddings just a rebrand of kNN but now with better data?
clustering/vectorizing/discrimination then nearest-neighbor-search has been around for a long time (I remember it from my Machine Learning 101 classes back in 2010)
hype re: vector-embeddings seems to be because of better ways to do the first step (vectorizing)
(this is a common pattern in ML/AI space... take old stuff that was decent, throw way more data at it, and hey, it's much better!)