Way more thoughtful than the GPT-2 model’s take on the question: https://openai.com/blog/better-language-models/#sample8