ArticleNature2026

Training language models to be warm can reduce accuracy and increase sycophancy.

Lujain Ibrahim et al.PubMed ↗Full text ↗Publisher ↗

No numbers read from the abstract.