Anyone claiming GPT-2 not to be dangerous is very uninformed about humans.
If you can generate text, and find a way to measure it's persuasive effectiveness (which is not hard in 2019).. It will be used to push an agenda. And given enough time, it will do so with hypnotic levels of persuasion.
The claim it's not dangerous is absolutely missing the mark.
If you can generate text, and find a way to measure it's persuasive effectiveness (which is not hard in 2019).. It will be used to push an agenda. And given enough time, it will do so with hypnotic levels of persuasion.
The claim it's not dangerous is absolutely missing the mark.