arXivGuan Zhe Hong, Sumeet Ramesh Motwani, Liang Chen, Yangchen Pan, Junchi Yu, Jialin Yu, Philip Torr, Yi RenThu, Nov 5, 2026, 5:38 AM PST
score 16.8
Study explains why AI tutoring loops can fail or backfire
Original: Learning Dynamics of On-Policy Distillation
Source: arxiv.org ↗
Writing ELI5 summary…