arXivZhiYan Hou, Xinyu Tang, Hongyan An, Jianjin Zhang, Weizhen Wang, Yunyun Han, Gengsheng Li, Xiangzhao Hao, Haiyun Guo, Wenbin Hu, Jinqiao Wang, Yafeng DengThu, Aug 6, 2026, 9:29 AM PDT
score 14.7
New training method boosts AI math reasoning without extra compute
Original: DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models
Source: arxiv.org ↗
Writing ELI5 summary…