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

• Posted 1 hour ago by 0in
• 1 points