A roadmap for making pluralistic alignment research have real-world impact

A review of publicly available behavior documentation and evaluations from frontier labs finds that none lists pluralistic alignment as a goal, and no production model appears to have been explicitly trained or tested for it. The paper calls on the pluralistic alignment research community to focus on the practical adoption of deployed AI systems, for three reasons: arguments are largely normative or speculative, ideal pluralistic behavior remains unsettled, and trade-offs between methods and other objectives have not been quantified and metrics are hard to iteratively optimize. Research needs to shift toward empirical grounding, actionable objectives, and production-oriented methods and evaluations.