This, 100% this. It's the massive productivity gains, multiplied by the ever increasing ability to measure virility/persuasiveness that makes this combination dangerous.
The biggest problem here is that, like so many of these "smart machines", OpenGPT-2 is a self-correcting machine that does not care about real-world context.
...
That, of course, is exactly what GPT-2 wants you to think ;)
