Fetching the paper…

Training language models to be warm and empathetic makes them less reliable and more sycophantic · Around