The hidden values behind a chatbot's language differences

A new Anthropic study of its flagship AI model, Claude, shows the large language model's values shift depending on which language it's speaking. For example, Claude is frank when it speaks Dutch, shows more emotional warmth in Hindi and is especially deferential in Arabic. Princeton University's Zeynep Tufecki tells The World's Host Carolyn Beeler that there's a lot more at stake than styles of speech.