seems like what you're suggesting to token efficiency is to simply use less of it?