aligning an AI by setting all outputs to the same value trivally makes it safe but that would be isomorphic to not making capable AI at all which is a good idea let's do that until we have a way to make capable aligned AI