Comment by famouswaffles

That doesn't make sense either. Training doesn't incentivize average. The models need to be predict all perspectives accurately. a middle of the road persona doesn't do that.