{
  "id": 9992,
  "url": "https://doi.org/10.1038/s41586-026-10410-0",
  "title": "Training language models to be warm can reduce accuracy and increase sycophancy",
  "summary": ". Here we show how this can create a significant trade-off: optimizing language models for warmth can undermine their performance, especially when users express vulnerability. We conducted controlled experiments on five different language models, training them to produce warmer responses, then evaluating them on consequential tasks. Warm models showed substantially higher error rates (+10 to +30 percentage points) than their original counterparts, promoting conspiracy theories, providing inaccur",
  "authors": "Lujain Ibrahim, Franziska Sofia Hafner, Luc Rocher",
  "category": "research",
  "topics": null,
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-29T00:00:00.000Z",
  "fetched_at": "2026-07-14T16:34:04.102Z",
  "source_slug": "openalex",
  "source_name": "OpenAlex",
  "source_homepage": "https://openalex.org",
  "ethics_ai_record_url": "https://ethics.ai/record/9992",
  "original_url": "https://doi.org/10.1038/s41586-026-10410-0",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}