Tweeted By @togelius

This is actually an argument _for_ releasing language models like the GPT-2. The sooner everyone understands that we can now generate surface-level correct and coherent text, but not process the text semantically, the sooner we'll stop stupid ideas like autograding essays. https://t.co/gPlGRQnJuW

— Julian Togelius (@togelius) March 10, 2019