> They can't be better then the average human though because of training data

Is this based on the belief that an LLM can only represent an "average" human being?

If posting good messages is automated then the AI will post a good question and another AI will answer it and the humans will look and see nothing extra to contribute.

It is not a meme, it's an xkcd: https://xkcd.com/810/