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

(nature.com)

4 points | by Anon84 12 hours ago ago

No comments yet.