William_S comments on The case for aligning narrowly superhuman models