Seeing Things Nearly Each Other in Latent Space

Commercial deployments of Lateen diffusion models probably have already deduplicated data sets. What you do is you basically query that vector database and you ask it for okay You know, here's a new vector which may or may not already be in here. There's two kinds of nearest neighbor. Proximate nearest neighbor algorithms are like pretty state of the art right now And there's increasing interest in this whole thing Allow you to scale sublinearly But it means that you won't get as like the results you get back weren't necessarily be accurate. The problem with that is even the best can and algorithms exact can and algorithms scale linearly in the amount of data So you can imagine like