A new study suggests AI models like ChatGPT and Claude consistently overestimate how rational humans really are, leading them ...