MIRI updates Scott Garrabrant and Rohin Shah debate one of the central questions in AI alignment strategy: whether we should try to avoid human-modeling capabilities in the first AGI systems. Scott gives a proof of the fundamental theorem of finite...
