{
  "id": 15252,
  "url": "https://arxiv.org/abs/2607.28582v1",
  "title": "$β$-OPSD: Deriving with Policy Optimization, Training with Self-Distillation",
  "summary": "On-policy self-distillation (OPSD) is a promising approach to improve reasoning language models, but it remains brittle in practice: making it work reliably often requires substantial engineering effort. We identify a structural source of this difficulty: vanilla OPSD is precisely the $β=1$ member of a broader policy-optimization family, where $β$ weights the KL penalty anchoring the student to a reference policy. This equivalence turns $β$ from an implicit value fixed at one into a controllable",
  "authors": "Jiawei Xu, Minghui Liu, Juzheng Zhang, Tom Goldstein, Furong Huang",
  "category": "research",
  "topics": "regulation,children-education",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-30T17:41:16.000Z",
  "fetched_at": "2026-07-31T05:10:57.675Z",
  "source_slug": "arxiv-cslg",
  "source_name": "arXiv cs.LG",
  "source_homepage": "https://arxiv.org/list/cs.LG/recent",
  "ethics_ai_record_url": "https://ethics.ai/record/15252",
  "original_url": "https://arxiv.org/abs/2607.28582v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}