Beyond these specific recommendations, our work provides an overarching framework for thinking about model evaluations in a statistical way. Read the full paper here:

→ View original post on X — @anthropicai