[apt-apt-apt-apt]

I was going to suggest emotional leetcode, but LLMs do well on this.

When given a conversation about Alice and Suzy having a one-upmanship conversation (my husband rich, my kid is a genius) and what emotions they are feeling, and what Suzy could have said instead to improve the conversation, it gave accurate responses (e.g. they're feeling insecure, competitive, envy).

[II2II]

That type of question could also turn people off. We already have too many discussions where people are quick to jump to conclusions and attribute intent, rather than asking basic questions.