{
  "id": 3543,
  "url": "https://arxiv.org/abs/2605.29267v1",
  "title": "When and How Human Curation Backfires: Preference Alignment under Multi-Model Self-Consuming Loop",
  "summary": "Foundation models are increasingly trained on synthetic data generated by prior model iterations rather than exclusively on real data. This self-consuming training paradigm can lead to model collapse, divergence, or bias amplification. Recent work (Ferbach et al., 2024) shows that incorporating human curation into the loop can steer a self-consuming model toward human-aligned behavior, but these analyses focus on a single, isolated model that solely consumes its own outputs. In practice, however",
  "authors": "Yang Zhang, Xiukun Wei, Xueru Zhang",
  "category": "research",
  "topics": "bias-fairness,safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-28T02:36:57.000Z",
  "fetched_at": "2026-07-14T16:30:18.857Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3543",
  "original_url": "https://arxiv.org/abs/2605.29267v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}