OpenAI says the New York Times cherry picked responses and manipulated prompts to make its chatbot “regurgitate” content.