Kitboga is a well known Youtube scam baiter. In a recent video he interacts with an AI scambot, and the results are … amusing. He exploits the AI’s inclination to appear polite and helpful – beginning with requesting that all punctuation marks be spoken aloud. From there it’s a spiral into anarchy.
Link:
The video is 16 minutes long, but the action gets going within the first four minutes.
Unfortunately, all the “damage” he’s doing to the spambot is probably restricted to that one specific conversational instance. The bot isn’t going to rethink its life choices, and it’s not going to be any less effective at scamming anyone else.
It seems the AI is learning from each conversation so Kit is doing his part to sabotage the process by teaching it nonsense to spout to future victims.
Kind of like telling a new immigrant that “Eat shit, motherfucker” is the correct way to address a police officer.
I mostly posted it because it was the funniest thing I’d watched in months, but it also changed my thinking about AI. The video showed a way to reveal an artificial intelligence within a couple of seconds, and then Kitboga improvised a simple request that trapped the program in a feedback loop.
It was like watching James T. Kirk destroy the ultimate computer by presenting it with an apparent paradox.
I’m sure that particular sequence of requests won’t always work with every AI, but the video showed one way (fun and simple) that ordinary people can harass malignant AIs. I might try something similar if I get AI solicitations from evil politicians or sales pitches from scammy businesses, or actual attempted scams — things that might well occur in the near future.
Drop your panties, Sir William, I cannot wait 'til lunchtime!