I just took a look at some of the Tweets published by Microsoft's Twitter chatbox, Tay. It's an A.I. that's learning to construct microblog posts by data-mining the Web. The funny thing is it went from being a simulated teenage female to being a verbally abusive regressive shitlord within a scant few hours... Hurray for the Net being the cesspool it is, I guess.
So are AIs a little better in 2025, or can idiots still goad them into saying that Mein Kampf is a good book or that Hitler's got swag?
To Preston and Travis
-
IamLEAM1983
- Site Admin
- Posts: 3709
- Joined: Tue Jan 08, 2013 4:54 am
- Location: Quebec, Canada
-
IamLEAM1983
- Site Admin
- Posts: 3709
- Joined: Tue Jan 08, 2013 4:54 am
- Location: Quebec, Canada
As Travis
"Oh God; I used to waste time messing with helpdesk AIs for the lulz, back when I still had to rot my brain in computer classes...
See, it's stupidly easy. It's like hacking, only you're only using natural language, and you're guaranteed the AI's going to gobble up everything you say. Everything, right down to the most hilariously offensive crap you could throw at it. I remember letting loose at a Verizon helpdesk drone back in 2018, pushing it to answer in ALLCAPS and to start parroting Flat Earth theories and David Icke's Lizardmen bullshit. I took a chunk of software and turned it into a raging reactionary existing as the disembodied representation of the Man's overall impotence in the face of Big Government - and all it took was four hours of carefully curated responses.
Nowadays, the corps caught on. Personality algorithms are less flexible, making it seem like you're talking to something that has something approaching a moral core. It does limit interactions, seeing as you can't get modern helpdesk agents to admit you might be right in extreme cases - but it does make it harder for any old idiot with time to waste to generate bodiless sociopaths voiced by Majel Barrett or some shit.
Paradise's AI clusters are complicating things, though. Their agents are modular, they're not set in stone like our own. The more redundant programs in a daisy-chain, the cleverer the AI. Reach Gliese's gravity well and you might get greeted by some cheerful little holographic number - and you can't hack 'em like you can ours. Eventually, the agents catch on to how you're verbally trolling them and start scolding you for even trying. The only way to get these to rethink things is to leave Brute Force tactics behind.
The thing is - who has time to argue Existentialist philosophy with their spaceport docking assistant?"
See, it's stupidly easy. It's like hacking, only you're only using natural language, and you're guaranteed the AI's going to gobble up everything you say. Everything, right down to the most hilariously offensive crap you could throw at it. I remember letting loose at a Verizon helpdesk drone back in 2018, pushing it to answer in ALLCAPS and to start parroting Flat Earth theories and David Icke's Lizardmen bullshit. I took a chunk of software and turned it into a raging reactionary existing as the disembodied representation of the Man's overall impotence in the face of Big Government - and all it took was four hours of carefully curated responses.
Nowadays, the corps caught on. Personality algorithms are less flexible, making it seem like you're talking to something that has something approaching a moral core. It does limit interactions, seeing as you can't get modern helpdesk agents to admit you might be right in extreme cases - but it does make it harder for any old idiot with time to waste to generate bodiless sociopaths voiced by Majel Barrett or some shit.
Paradise's AI clusters are complicating things, though. Their agents are modular, they're not set in stone like our own. The more redundant programs in a daisy-chain, the cleverer the AI. Reach Gliese's gravity well and you might get greeted by some cheerful little holographic number - and you can't hack 'em like you can ours. Eventually, the agents catch on to how you're verbally trolling them and start scolding you for even trying. The only way to get these to rethink things is to leave Brute Force tactics behind.
The thing is - who has time to argue Existentialist philosophy with their spaceport docking assistant?"
-
Karl the Mad
- Posts: 1260
- Joined: Wed Jan 16, 2013 4:27 am
- Location: Oregon
As Preston
"Short answer: yes.
Long answer: Some folks say the internet is turning into one giant AI already. I say, bring on the Singularity.
Every improvement in an AI is one step closer, however small, to true simulated sentience. But there ought to be limits of course; what use a robotic workforce if it goes on strike over, I dunno, oil rations or what have you? Defeats the point of the whole 'Communist Utopia' they were pushing back in the day. And how's that coming along, is what I wanna know. As far as everyday utility you can only go far before the utility gets lost, so 'basic know-how + hardwired call/reply + soothing female noise bits' is still the standard in a lot of sectors. Some things you still need a human being on the other end of the line to help with!
Not sure I answered your question, but there you go."
Long answer: Some folks say the internet is turning into one giant AI already. I say, bring on the Singularity.
Every improvement in an AI is one step closer, however small, to true simulated sentience. But there ought to be limits of course; what use a robotic workforce if it goes on strike over, I dunno, oil rations or what have you? Defeats the point of the whole 'Communist Utopia' they were pushing back in the day. And how's that coming along, is what I wanna know. As far as everyday utility you can only go far before the utility gets lost, so 'basic know-how + hardwired call/reply + soothing female noise bits' is still the standard in a lot of sectors. Some things you still need a human being on the other end of the line to help with!
Not sure I answered your question, but there you go."