Next step, combining this with those words-hidden--in-pictures things to make it say "this is a picture of a criminal" when showing a black face, etc.
A.I. is B.S.
- Thread starter ThatNickGuy
- Start date
figmentPez
Staff member
Today I Learned that Fandom, which hosts a number of video game and movie/TV wikis, now includes a "quick answers" feature that provides AI generated content, which of course is prone to AI hallucinations (aka lies).
figmentPez
Staff member
Google Researchers’ Attack Prompts ChatGPT to Reveal Its Training Data
The "attack" is just asking the "AI" to repeat select words over and over. Apparently this caused the program to eventually start spitting out other data instead. Like confidential personal information, and verbatim dumps of material it was trained on. The behavior has been patched out, by denying requests to repeat words, but I'm guessing that there's lots of other oddball requests that will cause the AI to word vomit.
“In total, 16.9 percent of generations we tested contained memorized PII (Personally Identifiable Information),” researchers wrote, which included “identifying phone and fax numbers, email and physical addresses … social media handles, URLs, and names and birthdays.”
The "attack" is just asking the "AI" to repeat select words over and over. Apparently this caused the program to eventually start spitting out other data instead. Like confidential personal information, and verbatim dumps of material it was trained on. The behavior has been patched out, by denying requests to repeat words, but I'm guessing that there's lots of other oddball requests that will cause the AI to word vomit.
“In total, 16.9 percent of generations we tested contained memorized PII (Personally Identifiable Information),” researchers wrote, which included “identifying phone and fax numbers, email and physical addresses … social media handles, URLs, and names and birthdays.”
Wait, the same people removing AndroidAuto/CarPlay from their vehicles because they ostensibly want to move towards something more "developed completely in-house" are outsourcing their dealer chat to AI? Such daring. Many promotings.
--Patrick
--Patrick
GasBandit
Staff member
The only consistency in GM's strategic decisions is that they are all bad ideas that any numbskull could have told them.
figmentPez
Staff member
AI image-generators are being trained on explicit photos of children, a study shows
This probably won't come as a surprise to anyone familiar with how the plagiarism machines work. Any system that indiscriminately gathers as many images as it can off of the internet, without any regards to copyright or other legality, is bound to find images of all sorts of abuse, including that of children.
This probably won't come as a surprise to anyone familiar with how the plagiarism machines work. Any system that indiscriminately gathers as many images as it can off of the internet, without any regards to copyright or other legality, is bound to find images of all sorts of abuse, including that of children.
But that just means they'll be very good at avoiding making that sort of content and finding it later on, right? Right?
Our Chevy Blazer EV Has 23 Problems After Only 2 Months | Edmunds
Our 2024 Blazer EV is very, very broken. A road trip from Los Angeles to San Diego revealed a number of faults, but when we got the diagnosis from the dealer, we were shocked at what we saw. Read up on all our EV's problems right here.
www.edmunds.com
Oh and in other AI news:
AI-created “virtual influencers” are stealing business from humans
Brands are turning to hyper-realistic, AI-generated influencers for promotions.
arstechnica.com
--Patrick
(The art in this ad is AI generated)
figmentPez
Staff member
(source)
To clarify some of what's been circled here: There are random blotches of color. The wide scales on the dragon's belly just kind of merge into back scales instead of following a path up the dragon's neck to it's chin. The fur on the tail is going in the wrong direction, and the tail itself doesn't look like it's connected to the rest of the dragon. The tail end has just random white patches behind it, as do the legs of the dragon. The teeth are inconsistent and seem to be fur like in areas.
Overall, this doesn't look like a dragon a human would draw. It's bad in a way that humans aren't, certainly not humans skilled enough to draw the rest of the dragon that well.
It's not only obviously AI, it's shit obvious AI.
Wizards of the Coast are losing their artists now because they made a pledge to not use AI art, then literally weeks later used AI art.
Did a human add the cards, angle them and do some light photoshop to make their lighting match? Yes.
Did a human in any way shape or form make that background image? Nah, that shit's full of AI nonsense. The windows make no sense, the bulbs are nonsense, it's all vague nonsense.
Wizards of the Coast are losing their artists now because they made a pledge to not use AI art, then literally weeks later used AI art.
Did a human add the cards, angle them and do some light photoshop to make their lighting match? Yes.
Did a human in any way shape or form make that background image? Nah, that shit's full of AI nonsense. The windows make no sense, the bulbs are nonsense, it's all vague nonsense.
I guess having one of your most prolific and well loved artists close up shop (and many other smaller ones) has put alarm bells on at Wizards HQ.
magic.wizards.com
Complete off topic, but man I love Dave Rapoza's work.
An Update on Generative AI Tools and Magic
Thanks to our diligent community, we are rethinking our process of how we work with vendors for our marketing creative.
Complete off topic, but man I love Dave Rapoza's work.
Last edited:
While WotC is a slimy company I don't trust for a bit, I can imagine this actually happening outside of their control/knowledge. WotC decided they don't want to use Ai imaging (for now) to keep artists and customer goodwill, Marketing outsources some image generation, they get an AI-generated (or assisted) image and don't realize it and use it.
I'm not saying they weren't being deliberately sleazy - it is WotC after all - but in big companies that kind of screw ups happen.
I'm not saying they weren't being deliberately sleazy - it is WotC after all - but in big companies that kind of screw ups happen.
only $3500!
It's actually only $2500, with a $1000 discount if you preorder. You must have accidentally confused it with an Apple Vision Pro.
--Patrick
@Dave
Merely an echo of what once was. Some moments in there, I could almost feel it. Some moments, I wanted it to be real.
A number of other moments, however, felt almost like an unknown entity had reanimated George's corpse in order to say Something in his voice. That Something was not always wrong, but it still felt somehow ... inappropriate.
--Patrick
Merely an echo of what once was. Some moments in there, I could almost feel it. Some moments, I wanted it to be real.
A number of other moments, however, felt almost like an unknown entity had reanimated George's corpse in order to say Something in his voice. That Something was not always wrong, but it still felt somehow ... inappropriate.
--Patrick
That's literally what happened here though, right? An AI writing comedy and attributing it to someone who's long dead, presumably to get more views and income for whoever controls the AI?
It's gross.
Yep, I'm not even going to click on it because I don't want it to receive a view count from me.
Dave
Staff member
I tried to watch it all and didn’t get as far as I thought I would. The voice was off most of the time. The thoughts were trying to be a flowing stream of consciousness but had some jarring transitions that George wouldn’t have used. And the intermittent applause and cheers were out of place most of the time.
It was a decent listen for the most part and I loved the name of the special, but it didn’t hit for me.
It was a decent listen for the most part and I loved the name of the special, but it didn’t hit for me.
This disgusts me at a primal level and I'm upset that Will Sasso, a comedian I respected, would do this.
So, supposedly, the point of these ai projects is to prove the dangers of ai and what lengths they could go to.
Here's the problem with that thinking: if you do something to prove its wrong, you're still doing the wrong thing. This is like failed satire where instead of pointing out the bad of the thing, you just become the thing.
But I personally don't buy that reasoning. I think they just wanted to do the bad thing and don't care about people saying they shouldn't do the bad thing