The other labs already censor their models. Everyone is trying to find the sweet spot where performance and ‘alignment’ are both maximized. This seems no different