abramdemski comments on The case for aligning narrowly superhuman models