That statement makes no sense. When you loglinearize your model, you still need to know the steady state values of the original loglinear equations for the parameter values you now need to set. But you are introducing an additional complication in that Dynare will not check the correctness of your steady state computations when you enter log-linearized equations.