The difference between Model.reference neural controller and Model reference adaptive controller?

I have reviewed both the tutorial examples of Model reference neural controller and MRAC matlab codeset in neural network toolbox. The model reference neural controller described with robot arm plant model shows good response within the range of previously offline trained dataset. But in MRAC controller, due to the adaptive adjustment mechanism, it can follow reference model outside the trained dataset range. I have deeveloped a simulink model using Model reference neural controller block but Im afraid whether its possible that it would work fairly well outside the training dataset range. Or maybe I have to use some adaptation algorithm for this to work like the MRAC? thanks in advance