Flame: Factuality-Aware Alignment for Large Language Models

3 points| milliondreams | 1 year ago |arxiv.org