Training language models to be warm can reduce accuracy and increase sycophancynature.com2 points·0in··0 commentsOpen articleSaveView on HN