New AI Model Generates Biased Responses Based on Different Stereotypes

Researchers created a large language AI model called OpinionGPT that generates biased outputs.

OpinionGPT is a variant of Meta's Llama 2 model, fine-tuned to respond as 11 different bias groups.

The model was refined on AskX subreddit data meant to represent different biases.

The resulting outputs often reflect stereotypes more than measurable real-world biases.

OpinionGPT is available online for public testing but may generate false, inaccurate or obscene content.