why is it not possible for a "big" model to contain a hidden super intelligent sub model? or a distributed model?