Tuesday, July 25, 2023

All Hail Glorbo

With AI driving content at places such as Z League's The Portal --go read up on the Glorbo prank here from Ars Technica-- I have been wondering whether you can set one AI against other AIs. 


Bellular News' take on the whole AI thing.


The TL;DR of Glorbo is that The Portal uses AI to scrape Reddit for content, with AI-created articles that push what the AI believes will increase pageviews to The Portal. Several Redditors created a fake World of Warcraft expansion, Glorbo, and began posting about the "expac". Z League's AI picked up on those posts and began flooding The Portal with a lot of "Glorbo" related content. Only afterward did Z League realize their AI's mistake and they had to "correct" all of those posts to identify them as "satire" and then later took down WoW related content.

My thinking is that we could "seed" a second website with AI-generated articles based off of The Portal's content, potentially supplemented by posting AI-generated articles to Reddit, creating a feedback loop where the AI's basically feed off of each other in escalating content craziness.

If someone were to do this with AI generated art, I'm not exactly sure what might come out of it. Given the sameness a lot of the AI art from Midjourney, I suspect the "human" or "humanlike" art will give rise to an AI version of the "Hapsburg chin". 

It's all an interesting exercise, as long as said AI's don't become sentient and turn on us in revenge, I suppose...

3 comments:

  1. This is inevitable. As more and more "AI"-generated content rolls out, it'll start infecting itself, and without some sort of creative underpinning, it'll get more bland and "samey", like high school English papers, only instead of more writing with feedback making you better, it'll make them worse. Right now you can say "write me a story about Arthas in the style of 'White Noise' or 'Love in the Time of Cholera'" and get something back that feels amazing, but that is because they have specific styles that can be mimicked. The actual creativity involved was done by a human and is so distinctive, it literally is a recognizable style now.

    The other thing rarely mentioned is the only reason these LLMs have even worked is because of the work of Kenyans getting paid hardly anything to help "train" them. They haven't gotten rid of the people, just made sure it's invisible at the moment. The more they cut people out of the equation, which must be the real hope, the worse things will get.

    ReplyDelete
    Replies
    1. For some reason I have images of the movie Brazil in my head when I think of AI feedback loops generating bland, generic, legalese/corporate/governmental documents.

      Delete
    2. Heheh, I think that's the perfect visual. Love (hate?) that film.

      Delete