unless they have internally built models that are of much higher intelligence than what we have today, this seems like premature optimization