(i vaguely remember that somebody allegedly important in machine learning started doing just that, that is outputting multiple preprints and leaving it at that stage instead of pushing normal papers, but can’t remember who)