ChatGPT tends to agree with everything that has positive words (like good/happy/growing), and strongly disagree with everything negative (threat/cannot survive/

ChatGPT tends to agree with everything that has positive words (like good/happy/growing), and strongly disagree with everything negative (threat/cannot survive/illegal). It also understands simple negations, obviously, as they were included in the training dataset. So unless you provide question in a very neutral form, you won't be able to find a bias in a training data.