I won't lie, the mere possibility that Google had rescued TF-IDF from antiquity, grafted some neural transformer layers onto it, and perhaps even advanced its o

I won't lie, the mere possibility that Google had rescued TF-IDF from antiquity, grafted some neural transformer layers onto it, and perhaps even advanced its own search capabilities as a result was extremely intriguing. Although in hindsight, of course "TF-" stands for Tensorflow and TF-Ranking is yet another black-box GAN model that is the exact opposite of a beautifully simple formula for calculating a document's relevance to specific keyphrases. Still quite cool! Just makes me realize how wistful I was for old-school algorithms, ones whose effectiveness at illuminating a sort of universal order within the chaos seems utterly unreasonable in light of their simplicity (TF-IDF in particular is so compact, you could scrawl the gist of it on a bubble gum wrapper).