But nowadays we should have an advantage by using high fidelity, purpose built training simulators to do this training. But we don't.
I agree 100%. I have had the rare "privilege" to live through a (self induced) upset with subsequent successful recovery in an airliner some time ago. In spare time in a sim, i later duplicated the event, only to realise how far away from reality the machine simulated the event. At that time we blamed it on technology with the hope that newer generations would improve.
Lately i was planned on a brand new sim with state of the art programming. You guessed right, i tried again and realised that it was closer, but still too far away from what reality shoved down my spine as to be able to state, that a sim would be valuable enough to train upsets.
It still is not realistic enough, unfortunately.
The cost cutting by putting almost all the flying training into simulators still has its limits.
Upsets and close to the edge flying is still not covered satisfactorily in synthetic training.