读书摘要✂：
「techreview： Meena was trained on a whopping 341GB of public social-media chatter—8.5 times as much data as OpenAI’s GPT-2. http://bit.ly/2uHZDTP 」