Rethinking Deep Alignment Through The Lens Of Incomplete Learning