Evan Hubinger on Inner Alignment, Outer Alignment, and Proposals for Building Safe Advanced AI

It’s well-established in the AI alignment literature what happens when an AI system learns or is given an objective that doesn’t fully …

source