Making AI chatbots helpful weakens their ability to simulate human behavior, large-scale study finds

A comprehensive study involving 208,000 participants and 26 million responses found that training language models to enhance their chatbot capabilities weakens their ability to replicate human behavior. This effect worsens with each new model generation, suggesting that current training methodologies may be ineffective. Even demographic profiling techniques provide little to no improvement in individual predictions. These findings indicate a need to rethink AI training strategies, as the gap between human and machine performance widens, impacting practical applications in the market.