Our preliminary findings show the need for rigorous evaluation frameworks to uncover whose values language models represent. We encourage using this methodology to assess interventions to align models with global, diverse perspectives. Paper:

→ View original post on X — @anthropicai