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

• Posted 3 hours ago by Anon84
• 2 points