Large Language Models are Superpositions of All Characters: Attaining Arbitrary Role-play via Self-Alignment Paper • 2401.12474 • Published Jan 23 • 33