{
  "id": 18440,
  "url": "https://arxiv.org/abs/2608.10209v1",
  "title": "Evaluation-Conditioned Training: Teaching Models to Generalize to Stronger Oversight Regimes",
  "summary": "Feedback signals used to train Large Language Models (LLMs) are the primary driver of their behavior and our main lever for instilling alignment with human values and objectives. However, a key limitation of current post-training methods is the inability of human annotators and automated reward functions to faithfully capture the feedback we would like to give. We introduce Evaluation-Conditioned Training (ECT), a post-training framework that uses natural language to condition each training samp",
  "authors": "Alec Harris, Kasey Corra, Archie Chaudhury, Yixiong Hao",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-08-10T20:22:13.000Z",
  "fetched_at": "2026-08-12T05:10:43.828Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/18440",
  "original_url": "https://arxiv.org/abs/2608.10209v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}