Generative language models exhibit social identity biases

Source page