Rob Bensinger comments on A central AI alignment problem: capabilities generalization, and the sharp left turn

[Rob Bensinger]

Nate’s follow-up post is now up: On how various plans miss the hard bits of the alignment challenge.