right, but massive scale production use in the google crawler to index the entire internet when that was at the bleeding edge was state of the art before the art was even really recognized as an art.

i don't even think they called it ANN. it was high performance, scalable deduplication. (which is, in fact, just fast/scalable lossy clustering)

collaborative filtering was kind of a cute joke at the time. meanwhile they had lsh, in production, actually deduplicating the internet.