Safety vs Capabilities

Distinguishing making AI more capable from making it safer.

Which statement best distinguishes AI safety work from AI capabilities work?