Comments on prediction by nonlinear least squares methods