Training language models to be warm and empathetic makes them less reliablearxiv.org358 pointsCynddl10 months ago