On the Essence and Prospect: An Investigation of Alignment Approaches for Big Models