ParentFull threadmnicky·May be related to this from METR evaluation:> GPT-5.6 Sol’s detected cheating rate was higher than any public model we have evaluatedView on HN