wunan comments on Paper: Large Language Models Can Self-improve [Linkpost]