Training language models to be warm can reduce accuracy and increase sycophancy(nature.com)4 points by Anon84 15 days ago | 0 commentsNo comments yet