AI language models hold different political biases, according to a study from the University of Washington, Carnegie Mellon University, and Xi'an Jiaotong University. The researchers tested 14 models and found that OpenAI's ChatGPT and GPT-4 were the most left-libertarian, while Meta's LLaMA was the most right-wing authoritarian. The political compass test helped identify the models' political leanings, though the researchers noted that it's not a perfect method. The researchers urged companies to be aware of the biases of AI models in order to make their products more fair.
