I often wondered if this strategy could work with email/phone scammers as well?
Have an LLM write positive email responses, stringing them along as long as possible.
Imagine the Nigerian prince of scams... instead of getting 0.1% response rates that he can work on, he gets 99% responses of encouraging replies stringing him along. He would drown in false positives.
On the other hand, the Nigerian prince of scams would probably just use his LLM:s to take care of the answers, and we'll have email storms of LLM:s stringing along LLM:s.
> On the other hand, the Nigerian prince of scams would probably just use his LLM:s to take care of the answers, and we'll have email storms of LLM:s stringing along LLM:s.
I already see that happening in other ways.
On a forum I frequent there's been a user who instantly replies to new articles with a long irrelevant blahblah with a bunch of bulletpoints and emoji. And a while later another user started insta-replying to everyone with the same kind of crap.
Both were banned after a while and the forum now has a stricter "No AI" policy but even that caused some uproar, some users say they want it so they can be more elaborate without having the linguistic skills. But personally I'd love to see it go because most of these replies added nothing to the conversation.
You can do nice things with AI, yes. But it needs a human to tell it what to do. If you just put in a generic prompt "Write something insightful about this article" it's going to provide nothing but slop.
Have an LLM write positive email responses, stringing them along as long as possible.
Imagine the Nigerian prince of scams... instead of getting 0.1% response rates that he can work on, he gets 99% responses of encouraging replies stringing him along. He would drown in false positives.
On the other hand, the Nigerian prince of scams would probably just use his LLM:s to take care of the answers, and we'll have email storms of LLM:s stringing along LLM:s.