@nirgal I'm sure this was totally an oversight as well. 🤣
Uhh...
Writers for The Daily Stormer, a neo-Nazi website directed at disaffected white teenage boys, joined in to promote the prank soon after it was launched, as did former KKK Grand Wizard David Duke and other politically organized racists. Duke, for example, ran a headline on his website declaring, "It's OK to Be White Campaign Sweeping the Nation!" That appeared on November 2, a full day before Carlson aired his segment.
You're mad about the AI actually being really accurate and picking up that a specific phrase was being mi-used negatively for one race? That's actually impressive on the AI's part. Really dying on a shitty hill here man.
You're talking to people who were freaking out over Taylor Swift. It's useless man.
Yeah we get it, everything to the left of Stalin is tEh fAr rIGhT to people like you.
Calling AI "woke" over an error they quickly saw and shut down to fix is pretty far right. And if not far right, it's pretty dumb at best. Of all the things to cry about. Like I said, it's as foolish as crying about Taylor Swift. I call it
"touch grass politics".
I'm near the center personally speaking.
Trump said it best when speaking of the anti-trans movement. Paraphrasing:
"You people, you never heard of trans before me. Then I come up here, I make it a big deal, and you go nuts for it!"
*crowd goes nuts, Trump laughs*
"Yeah, the trans are ruining this country"
Touch grass, indeed.
They don't have any good policies and are unpopular as hell, so they need to make stuff up and use tactics. We've been here before with the Red Scare, Prohibition, War on Drugs, Satanic Panic, and other things that make God-fearing moms and dads and old people lie awake at night...it was all fake and wasteful then, and this shit is fake and wasteful now.
We have bigger things to worry about than AI being woke and trans people and whatever.
And I'm sure all of you made just as big of a deal when instagram filters and ai generation were white-washing everyone for a while, riiiight?
Like obviously it's just an overcorrection with poor implementation and coding, not some secret exposed woke agenda. Let's not draw the stupid and wrong conclusion to make this non-issue seem like an issue.
I tend to not assign malice to things that so often are explained by stupidity. These systems are monstrously complicated to the point that the people who build and train them do not fully understand how they work or can fully explain why a certain result occurred.
If anyone here has tried to build something complicated, you know. You **** up constantly. It is literally "finding 10,000 ways to make something not work"
Shit, Boeing can't even remember to bolt door plugs on fully. The software approach even worse. Just look at games. It's basically "release it broken and we'll patch it". Google is under tremendous pressure to get SOMETHING out to compete w/ MS, openAI, etc...
Just as all rational people expected, it wasn't intentional:
'We messed up': Google admits blunder with black Nazi AI (msn.com)
He added: “We haven’t fully understood why it leans left in many cases and that’s not our intention."
No reason to get triggered before waiting for more info.
also LOL
"Examples post last week showed the bot saying it was “difficult to say” whether Adolf Hitler or Elon Musk had a more negative impact on society in conversations."
looks like they built a damn good troll bot for elon sycophants.
@robertos: oh well then. Google said it wasn't intentional, and I see no reason to question that, as a totally rational person, because large corporations never lie or spin once caught out on something! 🤣
@robertos: Rational just means you think in a logical, reasonable way, rather than allowing emotion to cloud your judgment.
AI generates images based on algorithms using a large data set of images it bas been fed for reference.
So if AI is now generating black popes, black nazis, black 17th century european monarchs, but not white ones, which of the following explanations is the most logical to you?
A) This AI tool was fed a bunch of actual pictures of black nazis, black popes, black 17th century monarchs, but none, or few of white people in the same roles?
B) This tool was fed pictures of real nazis, popes, monarchs, but a parameter was set in the algorithm to prioritize showing "diverse" people? Aka: "not white people".
C) random glitch that none of Google's many developers, engineers, or beta testers managed to uncover before releasing to the public?
So if AI is now generating black popes, black nazis, black 17th century european monarchs, but not white ones, which of the following explanations is the most logical to you?
Gemini was definitely capable of doing depictions of white people before the shutdown. Get your facts straight dude.
Even images from right wing NYP specifically to show the issues, still show white people being generated.
The AI was trained on all peoples of all walks of life and simply wasn't capable of making a logical conclusion like "Don't draw black Nazi's" because it isn't yet capable of that exception. I really have no idea why you want so badly for this to be some elaborate conspiracy when it so clearly isn't.
And on a related note, Meta now also has the same issue with it's generator. Are we going to assume that Meta is now a woke leftist conspiracy as well?
The AI was trained on all peoples of all walks of life and simply wasn't capable of making a logical conclusion like "Don't draw black Nazi's" because it isn't yet capable of that exception. I really have no idea why you want so badly for this to be some elaborate conspiracy when it so clearly isn't
So, I guess you're going with option C then?
C) random glitch that none of Google's many developers, engineers, or beta testers managed to uncover before releasing to the public?
Seems like a pretty big oversight when your algorithm somehow manages to generate every color of nazi except the actual color they were, especially when there shouldn't be any actual images out there of anything other than white nazis to feed it for reference.
But sure, go on believing there was no parameter at all set enforcing "diversity". While we're at it, I have some property on the moon to sell you. 🤣
"Why it matters:AI makers are trying to counter biases and stereotyping in the data they used to train their models by turning up the "diversity" dial — but they're over-correcting and producing problematic results."
I'm sure that "diversity dial" just appeared out of nowhere. A random glitch, and not intentional at all!
So if AI is now generating black popes, black nazis, black 17th century european monarchs, but not white ones, which of the following explanations is the most logical to you?
That's just more evidence this was not intentional, as the company has stated. Someone "woke" wouldn't want an AI making Black Nazis 😂🤣.
@appariti0n said:
@Vaasman:
I'm sure that "diversity dial" just appeared out of nowhere. A random glitch, and not intentional at all!
Diversity on it's own isn't woke, that's crazy talk. The extreme extent of how much this dial it was turned by the AIs learning, which is what caused this controversy, wasn't intentional. That is until you prove otherwise.
What a weird thing to still be mad about now that they said it wasn't on purpose and are actively trying to fix it. They even shut it down. What more do you want?
@Vaasman: "It wasn't programmed intentionally to not show white people OK??? It just has a diversity dial programmed in, that was turned up so high that it was barely showing any white people. Geez, stop with the conspiracies!"
That's just more evidence this was not intentional, as the company has stated. Someone "woke" wouldn't want an AI making Black Nazis 😂🤣.
Logical fallacy. Just because some of the end results were unexpected, doesn't mean the action that caused it was/wasn't intentional.
The extend of how much it was turned by the AIs learning, which is what caused this controversy, wasn't intentional. That is until you prove otherwise.
What more proof do you need? The article Vaasman linked acknowledges the existence of what is effectively a "diversity dial", and them admitting it was basically turned up too high. How is that not intentional?
That's a specific parameter that can be set to show more or less white people.
"Why it matters:AI makers are trying to counter biases and stereotyping in the data they used to train their models by turning up the "diversity" dial — but they're over-correcting and producing problematic results."
I'm sure that "diversity dial" just appeared out of nowhere. A random glitch, and not intentional at all!
You're right, it didn't appear out of nowhere, it appeared because AI models have historically engaged in whitewashing or stereotyping everyone from lack of training data and excess of stock imagery. Except no one called it a right-wing white nationalist conspiracy from Dall-e or Chat GPT or Stable Diffusion, because the people affected realized it was just a bug or algorithm issue to be worked on.
Now the Gemini and Meta models are pulling from a more diverse crowd that makes up their global userbase, and less from stolen or purchased stock images ,so they inevitably will generate more diverse depictions. The only issue here is they do it without more careful consideration of context. It's baffling to me that you can't understand that a model designed to generate literally everything and everyone will do just that and might not do exactly what you want when it's essentially random number generation.
That's just more evidence this was not intentional, as the company has stated. Someone "woke" wouldn't want an AI making Black Nazis 😂🤣.
Logical fallacy. Just because some of the end results were unexpected, doesn't mean the action that caused it was/wasn't intentional.
Your entire argument is a logical fallacy. You're saying it was intentional without proof. They took it down to fix it since it wasn't intentional, what's the issue.
@Vaasman said:
@appariti0n said:
@Vaasman:
I mean, right in your own article:
"Why it matters:AI makers are trying to counter biases and stereotyping in the data they used to train their models by turning up the "diversity" dial — but they're over-correcting and producing problematic results."
I'm sure that "diversity dial" just appeared out of nowhere. A random glitch, and not intentional at all!
You're right, it didn't appear out of nowhere, it appeared because AI models have historically engaged in whitewashing or stereotyping everyone from lack of training data and excess of stock imagery. Except no one called it a right-wing white nationalist conspiracy from Dall-e or Chat GPT or Stable Diffusion, because the people affected realized it was just a bug or algorithm issue to be worked on.
Now the Gemini and Meta models are pulling from a more diverse crowd that makes up their global userbase, and less from stolen or purchased stock images ,so they inevitably will generate more diverse depictions. The only issue here is they do it without more careful consideration of context. It's baffling to me that you can't understand that a model designed to generate literally everything and everyone will do just that and might not do exactly what you want when it's essentially random number generation.
Yeah it's all quite simple.
But again, Google admitted it was an unintentional mistake and more importantly have taken it down to fix it. So what's with people still crying?
Your entire argument is a logical fallacy. You're saying it was intentional without proof. They took it down to fix it since it wasn't intentional, what's the issue.
Maybe familiarize yourself with what a logical fallacy actually is. What you implied, which is that intention can always be determined by results, most certainly is one.
You may disagree with me asserting that this was intentional, but it's not a logical fallacy. There will never be ironclad "proof" unless google releases their source code the public, which we both know isn't going to happen. The best we can rely on is evidence. And the fact that there is confirmed existence of a "diversity dial", and admission that it was "overtuned", and numerous people who all experienced the same thing (including myself, as I actually tried gemini for myself), makes it pretty logical and reasonable to assume this was a feature, and not a bug.
Said lead developer's history of blatant anti-white rhetoric via tweets doesn't exactly strengthen the case that this was all just a "glitch" either.
Log in to comment