Even when the models are trained on data that should coax the outcome they want to see they're still getting undesirable results. Reality is simply refusing to bend to their will. I wonder at what point they either just fudge it or give up.