Actually, given that models are trained on data generated by humans, the ideal way to get decent responses are by treating them as such. Many times when I'm aware that a model gave an incorrect response, or I'm not sure and skeptical, a simple "Are you sure" or "Explain your answer" does wonders.