If you read the paper carefully it appears that this LLM got way better results than humans even without the socialmedia background check training on the target.
Fascinating! AI can be 6 times more effective in changing people's views than humans by... checks notes... lying the hell out of everything (like pretending to be a rape victim).
How were they so effective? Before replying, another bot stalked the post history of the target, learned about them and their beliefs, and then crafted responses that would perfectly one-shot them.
Very unethical experiment, but it shows the true power of what unscrupulous people can achieve with these tools.
https://regmedia.co.uk/2025/04/29/supplied_can_ai_change_your_view.pdf
Discussion
No replies yet.