{
  "id": 4440,
  "url": "https://arxiv.org/abs/2605.12667v2",
  "title": "ODRPO: Ordinal Decompositions of Discrete Rewards for Robust Policy Optimization",
  "summary": "The alignment of Large Language Models (LLMs) utilizes Reinforcement Learning from AI Feedback (RLAIF) for non-verifiable domains such as long-form question answering and open-ended instruction following. These domains often rely on LLM based auto-raters to provide granular, multi-tier discrete rewards (e.g., 1-10 rubrics) that are inherently stochastic due to prompt sensitivity and sampling randomness. We empirically verify the stochasticity of auto-raters that can propagate and corrupt standar",
  "authors": "Nirmal Patel, Fei Wang, Inderjit S. Dhillon",
  "category": "research",
  "topics": "regulation,safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-12T19:17:14.000Z",
  "fetched_at": "2026-07-14T16:30:59.238Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4440",
  "original_url": "https://arxiv.org/abs/2605.12667v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}