Presumably referring to the recent set of alignment incidents (which my guess is create more widespread beliefs that current RLHF + RLVR + Constitutional AI do not produce sufficiently aligned AI agents for the capability levels we are about to reach in the absence of a substantial slowdown).
Presumably referring to the recent set of alignment incidents (which my guess is create more widespread beliefs that current RLHF + RLVR + Constitutional AI do not produce sufficiently aligned AI agents for the capability levels we are about to reach in the absence of a substantial slowdown).
Yup. These observations.