Upset recovery training in a simulator will always have its limitations.
Lack of real 'g', software limitations outside of the data envelope which may not represent the aircraft are just 2 of the issues.
However, the idea is to recognise that earlier recovery may mitigate the possibility of the aircraft ending up in an unrecoverable scenario.
So, recognising and practicing nose high and nose low recovery combined with a roll component within existing and known software envelopes is better than no training at all.
I don't know of any ALPA that opposes this type of training.