Tag
This paper introduces Internal Coherence Maximization (ICM) to generate persona-specific examples for aligning AI with diverse human values without human supervision, demonstrating that coherent examples generalize better across benchmarks.