I can now go and incorporate this detector into the retraining pipeline for my evil language model or put it at the end of my architecture to emit only human-li

I can now go and incorporate this detector into the retraining pipeline for my evil language model or put it at the end of my architecture to emit only human-like results (as labeled by the detector). I don't see how detectors can win this cat and mouse game.