Quick thoughts on "scalable oversight" / "super-human feedback" research — AI Alignment Forum