This is (in my opinion) one of the most important pieces of empirical research to come out about generative AI’s actual/likely effects on politics. @shelbygrossman.bsky.social and team are right to warn that AI built to parrot back our beliefs to us can be an effective vehicle for propaganda.

CNcaseynewton.bsky.socialFeb 21, 2024 1:08am

A new study out of Georgetown and Stanford found that propaganda generated by GPT-3 (not 4!) was persuasive to almost half of the people who read it. I spoke to one of the authors and wrote about the implications www.platformer.news/ai-generated...

Thank you for being a cheerleader for this paper!

Yep. We’re boned. The AI technology is largely open source. It’s trained on data scraped from anywhere (social media) without regard to ethics or copyright. So if you can train the AI to perform a particular task (radicalize a target group) on scrape-able data who is going to stop you? (No one.)

There are so many AI powered accounts on Twitter now. It’s alarming

Makes me wonder if Musk made Grok so that Russia wouldn’t have to work to bypass GPT’s ethical constraints.

So much of what the Russian IRA was doing from 2015-2017 was clumsy, stupid experimentation. They hit on a few successful formulae (mostly by pretending to be American women of color) largely by accident. I saw hundreds of failed efforts for each success. AI accelerates that experimental cycle.

24% working without GPT-3 is also pretty worrying

The "But I've got a link (that handily confirms even my wet dream biases)!!!!!11one" demo is going to drink this Kool Aid deeply and repeatedly. But they did this for Z-grade blog posts as well...

I imagine "persuasive" here means "confirms people's prior opinions." The volume of material like this will effectively crowd out anything that might possibly *change* anyone's opinions. This algorithmic bullshit will further segregate the public into groups that can't communicate.