Is generalized least squares that general?