{
  "id": 505,
  "url": "https://arxiv.org/abs/2606.28707v1",
  "title": "BV-Blend: Uncertainty-Weighted Historical Baselines for Stable Critic-Free RL with Verifiable Rewards",
  "summary": "Critic-free reinforcement learning with verifiable rewards (RLVR), exemplified by Group Relative Policy Optimization (GRPO), avoids training a value function (critic) and reduces memory and compute overhead relative to critic-based PPO pipelines for aligning large language models. However, GRPO-style advantage estimation depends on prompt-local (within-prompt-group) reward statistics and can be unstable. In particular, when all rollouts in a prompt group receive identical rewards, the within-gro",
  "authors": "Yupeng Chang, Yuan Wu, Yi Chang",
  "category": "research",
  "topics": "regulation",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-27T03:25:53.000Z",
  "fetched_at": "2026-07-14T14:14:37.244Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/505",
  "original_url": "https://arxiv.org/abs/2606.28707v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}