AI Companies Are Getting the Culture War They Deserve
nymag.com
nymag.com
Really? That's the best we can do here? Showing a couple of tweets from a so-called "very vocal" group (gaslighting much?) doesn't do justice to the blatant racism that is present in the image generator.
You can see a video of what's actually happening here. It will refuse to generate white people, responding instead with lectures, while the same prompts asking for black people instead will be just fine. When asking for medieval knights, etc, it will show black kings with white people bowing to it. It has shown George Washington as a black person, etc.
Says the highest voted comment on the post...
There is no win with AI image generation, anyone can ask it to do anything nefarious at any time... Artists can technically draw any picture they want, so I really don't know why they're always trying to do PR control over tools.
Instead, they need to apply the logic to improve account control in order to censor and properly tag explicit and offensive material. I still see snuff videos on TikTok and Reddit, I'm worried about kids seeing some of the horrible stuff I've seen online, at this point Ai should be flawless in IDing and tagging negative content. Ai should flawlessly ID the people sharing nefarious content without proper context and even handle tech support and appeals for wrongful cases quickly... That's the real type of use case for AI, not limiting and constantly extinguishing what can be generated by creative minds as much as they do.
When you ask for a photo of a viking and 75% of the results are Asian/African/native-american, and the only picture of a "viking" is Netflix adaptation version with dreadlocks, leather bracelets and dirt over his face then whatever anti-bias thing they were doing clearly is doing the opposite of what it's supposed to and it's objectively making their product worse.
Stable Diffusion will happily make me Germans, Scandinavians and black people, no matter whether their job description includes raiding coastal cities or not. And it will do so whenever asked for. And without annoying moralizing lectures. Also it runs on my computer and they can't take it away from me tomorrow because they've deemed cats eating fish to be "harmful" overnight. (context: Gemini sometimes refuses to make pictures of animals eating animals, but it's not consistently refusing to)
This mishap is Silicon tech bros engaging in cultural colonialism imposing their own narrow view on the world because they think the hallucinogens they took on holiday somehow made them more enlightened and better than those plebs slaving away outside of their citadel.
Funnily enough they took it offline because it was generating black soldiers when you asked it to generate a picture of a Nazi. Which ironically is more realistic than the black Scandinavians that Google doesn't have a problem with. The Nazi's did indeed have African, Asian and Middle-Eastern soldiers, and even Jewish ones.
Google doesn't care about truth or history. They are doing exactly what Orwell described in Winston's job: alter [the perception of] the past to control the future. I happen to think the ends don't justify the means, and Google has become Evil.
Because why? Because "that's different"?
Anyway, I was just pointing out that you aren't the only one capable of blanket dismissing the other side.
Assange is a hacker who's indiscriminately leaked tons of classified info, much of which had little to do with corruption like the unredacted cables and the Vault 7 documents. He's (allegedly) conspired with Chelsea Manning and Lulzsec to basically get whatever he can get his hands on [1][2].
btw, Navalny died in a penal colony serving a 19 year sentence. Chelsea got out after 7.
[1] https://www.justice.gov/opa/pr/wikileaks-founder-charged-com... [2] https://www.justice.gov/opa/pr/wikileaks-founder-charged-sup...
I'm actually appalled at the way the Russian government acted here, but let's not pretend the US government is all nice and fuzzy and always plays by the rules. Even when the rules are their own.
The complaint is more that AI companies are misrepresenting what their AI is and can do, leading to unrealistic expectations.
For starters, most local models I've seen don't stuff prompts with DEI requirements, and actually do a reasonable job of following instructions
Meta's local models contain alignment and censoring, yet it is nowhere near the scale of Gemini's bias. I hopefully don't need to remind you that a large number of local models are based on llama.
Mistral has raised over $300 million. I'd say that warrants being worried about getting sued, yet you don't see people memeing about their model's alignment.