It is possible to make meaningful progress on deceptive alignment using experiments on current models
It is possible to make meaningful progress on deceptive alignment using experiments on current models