From Instructions to Intrinsic Human Values -- A Survey of Alignment Goals for Big Models