How did they measure and train for warmth and empathy? Since they are using two adjectives are they treating these as separate metrics? Ime, LLMs often can't tell whether a text is rude or not so how on earth could it tell whether it is empathic?

Disclaimer: I didn't read the article.