Two proposed changes before AI systems become significantly more capable

A post quotes a call for two urgent changes: AI companies should draw more on safety expertise from other fields, and new science is needed before building significantly more capable systems to ensure those models and their successors make safe choices when no one is looking.