{
  "id": 4534,
  "url": "https://arxiv.org/abs/2605.11182v2",
  "title": "The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes",
  "summary": "On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offering dense token-level supervision on trajectories sampled from the model's own policy. However, existing results on their effectiveness remain mixed: while OP(S)D has shown promise in system prompt and knowledge internalization, recent studies also report instability and degradation. In this work, we present a comprehensive empirical study of when OP",
  "authors": "Siqi Zhu, Xuyan Ye, Hongyu Lu, Weiye Shi, Ge Liu",
  "category": "research",
  "topics": "regulation",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-11T19:44:59.000Z",
  "fetched_at": "2026-07-14T16:31:03.580Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4534",
  "original_url": "https://arxiv.org/abs/2605.11182v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}