This departing note from an AI safety lead at Anthropic is worth reading.

→ View original post on X — @martyswant