As ChatGPT said itself, GPT models can give responses that are fluent, relevant and coherent but factually wrong, as they don't perform fact-checking or verification when generating a response.