We’re not even close to the limits of AI optimization so finding the theoretical bounds is going to have to wait