New benchmark SocialReasoning-Bench measures AI agent alignment with user interests. Agents exhibit competence but struggle to consistently optimize user outcomes, even with explicit instructions. This highlights a critical gap in developing truly user-centric AI systems.
Opening Kapyn…