[blakers95]

And what happens when the models themselves are trained primarily on the data they produced?

[danielmarkbruce]

They will spit out crappy, copycat material which might be fake. Like 90% of the material created by humans. And then the people with something valuable to say will have something original and valuable. And we'll be back to having a search/curation/review problem on our hands.

[vkou]

I don't learn from 90% of material created by humans.

I learn from the 1% of material that is created by humans who are, to some degree, experts in their field.

Sometimes it's also wrong, but it's not because they were just lazily regurgitating rando 'net posts.