← back
arXivZiyun Qiao, Yue Min, Ruining Chen, Yujun LiThu, Jul 2, 2026, 7:51 AM PDT
score 17.0

Hierarchical data labels improve AI training data mixing

Original: HERMES: A Multi-Granularity Labeling Substrate for Pre-training Data Mixtures

Source: arxiv.org

Writing ELI5 summary…