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

(nature.com) by Anon84 | May 3, 2026 | 0 comments on HN