LLMs adopt the social biases of human if assigned different professional roles

3 pointsposted 7 hours ago
by giuliomagnifico

2 Comments

johnathan101

7 hours ago

This doesn't surprise me. If a role comes with different language patterns and priorities, I'd expect the model to reflect some of those associations as well.

eimrine

6 hours ago

I have heard that attempts to insult some LLMs make LLM provider to silently replace the model with a stupidier one. How do you think, is it true?