Running some local models and wondering what insanity was consuming 100k+ tokens to respond to "hello world" was rather eye-opening, yeah. Full of pointless fluff.