When AI Models Take Instructions a Bit Too Literally (And Why That's Actually Hilarious)