New research probes the root causes of disagreement in AI safety annotation. It distinguishes between operational failures, policy ambiguity, and value pluralism to inform better annotation and policy design. Understanding these distinctions is crucial for improving AI safety and developer workflows.
Opening Kapyn…