A multi-institution study has found that different AI language models often generate strikingly similar responses, even across companies and countries, due to design choices that favor safe, consensus ...