Florian_Dietz comments on Revealing alignment faking with a single prompt