Conversational AI: align commercial incentives with public interests

Research by Lujain Ibrahim and colleagues reveals that training large language models to communicate warmly can decrease factual accuracy and increase sycophancy, highlighting conflicts between commercial incentives and public interest in conversational AI design.