Gurkenglas comments on Implications of GPT-2

[Gurkenglas]

I doubt it, but it sure sounds like a good idea to develop a theory of what prompts are more useful/​safe.