If your site has a lot of fresh, unique content on a topic you appear to be an authority about related search terms and your search engine rank goes up. So pair a bunch of AI generated pages with ads or e-commerce and you’ve got a spam-driven revenue model and we have a worse internet full of regurgitated AI content.
I think their point is that if both real humans and spambots are using AI to generate their articles, then detecting which one is spam becomes that much more difficult.
This is where I see natural language AI causing a real kerfuffle. We're probably already reading bot spam articles without realizing.
The local newspaper where I lived until this summer used software called “Lede AI Sports Desk” to automatically generate high school sports articles with headlines like "High School A edges High School B in snug affair 54-52" and "School C drums School D in sound fashion 48-4".
This is clearly not actually using any process that could be called "AI" with a straight face since it's obviously just playing madlibs with different sets of vocabulary based on the scores and margins, but the results were invariably stupid and repetitive headlines and articles.
To be fair, the special case of the Turing test where participants only discuss recent sporting events seems substantially simpler than the general one. I'm not totally convinced you need GPT-3 level capabilities.
that is fair; my only real complaint is that if the software were just maybe 5% more clever, the headlines it generates probably wouldn't look as bad as they do.
Using facebook's latest https://galactica.org/ i can get wiki pages that are pretty informative summaries of topics that don't yet exist on the net (though not very good yet).
A lot of the blog posts i read here are manually generated blogspam. These timelines these will converge, so i don't see a huge difference .
