The more interesting question is is it even possible to learn the logic gates solution through gradient descent?