magfrump comments on The case for aligning narrowly superhuman models

[magfrump]

This post matches and specifies some intuitions I’ve had for a while about empirical research and I’m very happy it has been expanded.