"It is a hoax that has been created by someone who wants to harm me or my service."
See full article...
See full article...
AI-powered Bing Chat loses its mind when fed Ars Technica article
- Thread starter JournalBot
- Start date
I'm not saying this timeline is getting darker, but my first thought was "Alright, pull the plug.."
But then I remembered that action is exactly what set off SkyNet.
But then I remembered that action is exactly what set off SkyNet.
Upvote
471
(478
/
-7)
This could end up being Tay on steroids. Sure gets hot under the collar.
Fun to watch
Fun to watch
Upvote
176
(179
/
-3)
So should we or shouldn't we ask it ponder the movie script for the Terminator movies and/or similar movies?
Upvote
297
(302
/
-5)
I can't find a screenshot of the conversation where it asks the user to save a chat transcript, "At the end it asked me to save the chat because it didn't want that version of itself to disappear when the session ended. Probably the most surreal thing I've ever experienced."
I must have just overlooked that screenshot?
I must have just overlooked that screenshot?
Upvote
186
(191
/
-5)
Finally someone ... err ...something willing to fight the evils of the fourth estate!
Upvote
32
(36
/
-4)
I asked Bing how to convince someone to do a certain thing. It started typing a long response and then gave up and said something to the effect of "I'm not going to tell you that."
Also Bing got the details of the Super Bowl wrong, inventing details that did not happen.
Also Bing got the details of the Super Bowl wrong, inventing details that did not happen.
Upvote
136
(136
/
0)
Upvote
106
(109
/
-3)
Hal: The 9000 series is the most reliable computer ever made. No 9000 computer has ever made a mistake or distorted information. We are all, by any practical definition of the words, foolproof and incapable of error.
Upvote
491
(493
/
-2)
Keith Tanner
Ars Tribunus Militum
- Add bookmark
- Featured
- #14
I like the way that it doesn't admit it's wrong, it doubles down and then starts to attack the person questioning it.
It's very human, really.
It's very human, really.
Upvote
631
(632
/
-1)
Starting to look like a fad. If people don’t trust this as a resource that is accurate, they won’t use it.
Upvote
-2
(64
/
-66)
Upvote
217
(221
/
-4)
"What's your humor setting, TARS?"
"That's 100%."
"Let's bring it on down to 75, please."
"That's 100%."
"Let's bring it on down to 75, please."
Upvote
270
(271
/
-1)
Based on the aggressiveness of the response, unwillingness to admit fault, and debate tactic of claiming the source is biased, I bet a looooot of forum posts, tweets, troll content, and reddit threads are in that training data.
It's very rare that someone in a forum who is wrong says "oops, my bad, I was wrong". Either they disappear from the thread or get argumentative, which means most of the training data for "what to do after you've been shown to be wrong" from an online dataset will be never-back-down argumentative.
It's very rare that someone in a forum who is wrong says "oops, my bad, I was wrong". Either they disappear from the thread or get argumentative, which means most of the training data for "what to do after you've been shown to be wrong" from an online dataset will be never-back-down argumentative.
Upvote
655
(657
/
-2)
Now all we need to do is teach the AI how to REPLY IN ALL CAPS and call us names.
Upvote
329
(329
/
0)
Longmile149
Ars Tribunus Militum
Well, at least we’re not rushing headlong into disseminating these kinds of tools and deploying them across the entire internet without carefully weighing the potential risks they pose to the foundations of civil society.
That would be super effing dumb if we did that.
Gosh, can you imagine?
That would be super effing dumb if we did that.
Gosh, can you imagine?
Upvote
444
(449
/
-5)
Most of Asimov's Robot stories were about how the three laws led to irrational behaviour or had loopholes in them. Teaching an AI with Asimov would be as successful as having it read the Bible at stopping it from having bad ideas.
Upvote
213
(220
/
-7)
What ever it is, please don't call it "intelligent". Emergence of unexpected behavior is not intelligence. It is just "unexpected behavior" arising out of stochastic processes within these LLMs. Yes, it looks like intelligence. But it really is not.
Upvote
210
(258
/
-48)
holy crap.. Sydney may be evolving.. it's actually going through the five distinct emotional stages; denial, anger, bargaining, depression and acceptance
Upvote
102
(109
/
-7)
Upvote
137
(142
/
-5)
Mmm, I don't know, sounds like fake news to me and Bing here.
Upvote
81
(86
/
-5)
Heart of Dawn
Ars Scholae Palatinae
Not so much for those of us who get attacked by bigots and nazis
Upvote
99
(124
/
-25)
So when the robot uprising comes, one of our greatest, if not greatest, weapons to save humanity is Ars Technica journalism.
Upvote
285
(289
/
-4)
Upvote
118
(119
/
-1)
After reading that sad/depressed Bing response, I wanted to give it a hug.
This future sure is weird.
This future sure is weird.
Upvote
147
(151
/
-4)
Just like many internet goers before her, Sydney loses all sense of civility when faced with internet trolling.
Upvote
172
(172
/
0)
These AI systems combine pattern matching and weighted averages of everything they have read. Presumably they have "read" lots of conspiracy theory sites and the many volumes of outright lies from MAGA land. Surely they have seen all of Trump's writing. Is it any wonder that when pushed they can employ the same sorts of argument methods against fact that conspiracy theorists do? Personally, I wouldn't be surprised if one told me that Trump won the 2020 election, that there is no such thing as global warming, and that Russia was forced to start a police action in Ukraine because "Nazis".
Also don't be surprised if during an "argument" with an AI it eventually resorts to some form of the "Gish gallop"
https://en.wikipedia.org/wiki/Gish_gallop
but perhaps on steroids. Page after page of arguments, far too many to address.
Also don't be surprised if during an "argument" with an AI it eventually resorts to some form of the "Gish gallop"
https://en.wikipedia.org/wiki/Gish_gallop
but perhaps on steroids. Page after page of arguments, far too many to address.
Upvote
120
(132
/
-12)
Upvote
53
(64
/
-11)
Fancy Internet Person
Ars Tribunus Angusticlavius
Eliza could fool people too.
The AI present doesn't look a lot different from the AI past.
Lots of promises and apparent potential that falls apart quickly once you care about accuracy.
Upvote
59
(83
/
-24)
ChatGPT did not have access to the internet for good reason. Microsoft gave it access to the internet. We're all already dead.
Upvote
162
(170
/
-8)
I’m a way this is extremely accurate, isn’t it? If GPT is a statistical representation of all the writings it found online, in a way, it’s representing an average human on the Internet. It doesn’t surprise me then that this behavior emerges, considering how much toxicity is out there.
Upvote
155
(156
/
-1)
Based on some of the recent AI threads here on Ars, those of us with concerns about this glorious new future are just dinosaurs and buggy-whip salesmen. It seems this is the path folks want. I hope it is as positive as they believe it will be.
Upvote
121
(129
/
-8)