Alignment Targets and The Natural Abstraction Hypothesis — AI Alignment Forum