I've also been wondering this. It is like Kessler Syndrome. We have to be careful not to pollute our ecosystem of data.

https://en.wikipedia.org/wiki/Kessler_syndrome

Speaking with GPT-3 also makes one realize how influenced its predictions of AI scenarios are by dystopian memes. In any conversation in which you are "speaking to the AI", the AI can go rogue.

There just aren't enough positive role models authors have written for AGI.