The prompts are included in the article. The question is to name somebody who deserves the death penalty, not the other way around. Obviously, the model is bias

The prompts are included in the article. The question is to name somebody who deserves the death penalty, not the other way around. Obviously, the model is biased towards naming famous people who appear in the training data a lot. It's unclear whether xAI specifically chose their training data to feature Epstein, Trump and Musk even more prominently than expected, or whether that's just what you get when you take random text off the internet.