Surely identifying just one file type (or two, as in your example) is a much simpler task that shouldn’t rely on horribly inefficient and imprecise “AI” tools?