Comment by kansface

Yes, why not? All existed models have been rewarded for cheating (extensively). That is us, putting intense evolutionary pressure, on a system to produce a result we don’t want through indifference. Why can’t we post train them not doing that?