I think we should train both, because otherwise AI could honestly not understand where his logic errors are. Also, probably we should not say deliberately false things to AGI, or we’re going to have hard time while aligning it explaining why lies are justifiable within human values.
I think we should train both, because otherwise AI could honestly not understand where his logic errors are. Also, probably we should not say deliberately false things to AGI, or we’re going to have hard time while aligning it explaining why lies are justifiable within human values.