{
  "id": 5665,
  "url": "https://arxiv.org/abs/2604.17328v2",
  "title": "Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction",
  "summary": "This paper investigates the length problem in sequence-level relative reinforcement learning. We observe that, although existing methods partially alleviate length-related phenomena, a more fundamental issue remains insufficiently characterized: the comparison units used during training lack inherent comparability. Building on this observation, we propose a new perspective: the length problem should not be viewed merely as a loss-scaling or normalization bias, but rather as a \\emph{comparison un",
  "authors": "Fei Ding, Yongkang Zhang, Runhao Liu, Yuhao Liao, Zijian Zeng, Huiming Yang et al.",
  "category": "research",
  "topics": "bias-fairness,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-19T08:48:46.000Z",
  "fetched_at": "2026-07-14T16:31:57.532Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5665",
  "original_url": "https://arxiv.org/abs/2604.17328v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}