It seems that the gzip approach, altough really cool, was ‘optimistic’ and thus overhyped, see: https://kenschutte.com/gzip-knn-paper/ (basiccaly they confused k in k-nn and top-k accuracy, reporting top-2 accuracy). More recent studies found that it is, as expected, on ‘bag of words’ performance level Gzip versus bag-of-words for text classification.
I don’t know if you intend to (or are even interested) but I am on the look out for “usecases for normies”.