|
mystes posted:It's really loving stupid that Google had such an overreaction to the response to Gemini sometimes making people more diverse than they actually should be. All these CHUDs are getting themselves worked up about a conspiracy to make their AI god woke and missing the glaringly obvious: It was a haphazard attempt to paper over a fundamental flaw in generative AI. These models just reflect back and summon all our own biases and prejudice. There's no intelligence there, so self awareness. Just a fun house mirror of humanity. In this case though, the rush to put a product out there and juice the stock price backfired in a really hilarious and stupid manner.
|
#
?
Mar 5, 2024 15:55
|
|
|
|
| # ? Apr 29, 2024 02:35 |
|
|
kliras posted:i'm old enough to remember when google photo tagged black people as gorillas (jesus christ that was almos ten years ago), and it's clear that google have learned nothing since
|
|
#
?
Mar 5, 2024 15:57
|
|
|
mystes posted:I don't think this is accurate at all. Google intentionally put in work to make sure it didn't have the problems that other AI had in terms of being biased too far toward generating white people. The problem is that dumb people were able to intentionally prompt it to generate images that didn't make sense and use that to start a bullshit controversy that is now being exploited by right wingers to demand that AI be made more right wing.
|
|
#
?
Mar 5, 2024 15:59
|
|
|
mrmcd posted:All these CHUDs are getting themselves worked up about a conspiracy to make their AI god woke and missing the glaringly obvious: It was a haphazard attempt to paper over a fundamental flaw in generative AI. These models just reflect back and summon all our own biases and prejudice. There's no intelligence there, so self awareness. Just a fun house mirror of humanity. In this case though, the rush to put a product out there and juice the stock price backfired in a really hilarious and stupid manner. The fact that in some cases the result won't make sense because there weren't black presidents in the 1800s or something is not actually an issue and treating it as an issue is playing into the hands of chuds. The expectation that the results of generative ai will always be 100% historically accurate doesn't make any sense and google should be responding by pointing that out, not turning the racism dial up while looking back at the crowd for approval. And again, Sergey Brin has literally interpreted the problem as Gemini being too far to the left and said they need to fix that, so if you think there's going to be any resolution here that's not simply turning up the racism dial you're probably wrong. Even if some of the people complaining were people on the left who got it to generate images of black nazis or something, chuds are the only people who have actually won in this situation because the upshot is somehow "generating images of black people is bad". mystes fucked around with this message at 16:03 on Mar 5, 2024 |
|
#
?
Mar 5, 2024 15:59
|
|
|
Cybernetic Vermin posted:this except the opposite, only way you lump that up into an acronym you throw around is if you're some bigot shithead Sorry sarcasm because the people who are doing work in the area of making orgs less racist see DEI as a smokescreen preventing real change at orgs
|
|
#
?
Mar 5, 2024 16:02
|
|
|
Cheen posted:not knowing what DEI is means you should reflect inward on if you're an ally of marginalized groups im a minority op, so i think im good
|
|
#
?
Mar 5, 2024 16:02
|
|
|
fart simpson posted:im a minority op, so i think im good
|
|
#
?
Mar 5, 2024 16:03
|
|
|
mystes posted:I don't think the result is even problematic. People just have idiotic ideas about AI. It's a tool for generating images that in the case of people will default to generating a fairly diverse group of possibilities. This is fine. Yeah, but in the specific case of the Gemini fiasco it really looks like they were worried about bad PR like "It only shows white dudes when you ask for pictures of doctors" and were either loving with the models or doing behind the scenes prompt injections to correct for that bias in the training data. But because this was a botched rush job (and they fired any buzzkill AI safety people who might slow things down) it ended up also doing things like racially diverse Nazis soldiers and Native American George Washington.
|
|
#
?
Mar 5, 2024 16:06
|
|
|
mrmcd posted:behind the scenes prompt injections to correct for that bias in the training data Word is that this is what they were doing.
|
|
#
?
Mar 5, 2024 16:13
|
|
|
ultrafilter posted:Word is that this is what they were doing. Yeah adding code that is basically "scan for words that indicate humans, then silently append 'racially diverse, multiracial, black, Asian, native American' to the prompt" is 100% plausible as something an engineer would do when a VP is screaming at them to RELEASE THIS NOW WE GOTTA LAUNCH AND NO EMBARRASSING OUTPUTS!! and also produce the kind of wacky images everyone was mocking Gemini for.
|
|
#
?
Mar 5, 2024 16:23
|
|
|
Don't turn around, uh-oh DEI Kommissar's in town, uh-oh You're in his eye and you'll know why The more you bitch, the stock price will go high Alles klar, DEI Kommissar?
|
|
#
?
Mar 5, 2024 16:25
|
|
|
mystes posted:Do they have a photo of you on the company website as the token white person? not on the website but a while back they put my picture up on the wall of Quality Stars for the quarter. i was the only white guy out of those 50 or so photos
|
|
#
?
Mar 5, 2024 16:42
|
|
|
getting blasted in the dome by the regimental dei commissar for not including pronouns in my email sig
|
|
#
?
Mar 5, 2024 17:18
|
|
|
|
|
#
?
Mar 5, 2024 17:24
|
|
|
ultrafilter posted:Word is that this is what they were doing. a friend likes loving around with ai image generators and yeah, if you put in a prompt of "person holding up a sign saying" it would frequently contain "ethnically ambiguous", "white", "asian", "african american" etc (often misspelled as always) the "boo hoo google DEI" thingi s even funnier in the context of that dude wearing that hosed up shirt that looked like a naked woman's chest. talk about a meeting that could've used a DEI commissar 
|
|
#
?
Mar 5, 2024 17:27
|
|
|
fart simpson posted:im a minority op, so i think im good Me to (iq > 200)
|
|
#
?
Mar 5, 2024 17:48
|
|
|
kliras posted:nor did elevating some insane redditor who typed up a manifesto uh they elevated him all the way to "former employee" firing that idiot into the sun was the correct move
|
|
#
?
Mar 6, 2024 03:55
|
|
|
Cheen posted:Me to (iq > 200) it’s a tough existence
|
|
#
?
Mar 6, 2024 03:56
|
|
|
dall-e also modifies prompts to add ethnic ambiguity or whatever to outputs; in their defense, they also have tighter controls on what is acceptable output. if you ask for a 1940s German soldier it just tells you to gently caress off. I agree that this is a stupid manufactured controversy and I also agree that google is handling it poorly by accepting the framing of the AI being woke or whatever.
|
|
#
?
Mar 6, 2024 05:51
|
|
|
the African Vikings are only the very obvious canary in the coalmine. would anyone notice if Google tweaked the prompt, or better yet filtered the training data, to make Palestinians look angry and violent and Israelis peaceful and sad? e: i also think that the power to control the output of a lovely picture generator is way, way less important than the power to "tune" search results, which they have monopolized for close to two decades but as a card-carrying member of the "break up the tech giants" club, I'm glad for anything that makes people realise these guys are not morally neutral service providers but an American corporation with a specifically obscenely-rich-Californian outlook on how the world should be organised NihilCredo fucked around with this message at 07:46 on Mar 6, 2024 |
|
#
?
Mar 6, 2024 07:40
|
|
|
kitten emergency posted:dall-e also modifies prompts to add ethnic ambiguity or whatever to outputs; in their defense, they also have tighter controls on what is acceptable output. if you ask for a 1940s German soldier it just tells you to gently caress off. poo poo, could probably dig through HN posts from the other times this exact same controversy happened and find eerily similar discussions, with subsequent posts shaped the same way and the model name find & replaced. Like layers in the soil.
|
|
#
?
Mar 6, 2024 07:55
|
|
|
dorkwood 18 minutes ago | parent | context | prev | next [–] | on: OpenAI and Elon Musk At first I thought "why would Elon choose to slow down progress like this?" And then I realized that he probably knows some things that I don't. He's always one step ahead, even if he doesn't seem like it. I wouldn't be surprised if he's in control of OpenAI a year from now. Having Elon as the brains and Altman as the salesman would be the ultimate dream team. reply
|
|
#
?
Mar 6, 2024 18:15
|
|
|
fritz posted:dorkwood 18 minutes ago | parent | context | prev | next [–] | on: OpenAI and Elon Musk barf
|
|
#
?
Mar 6, 2024 18:27
|
|
|
https://twitter.com/EsqTim/status/1547047570425380864
|
|
#
?
Mar 6, 2024 18:28
|
|
|
fritz posted:dorkwood 18 minutes ago | parent | context | prev | next [–] | on: OpenAI and Elon Musk Patriots are in control. Trust the plan. wwg1wga
|
|
#
?
Mar 6, 2024 18:28
|
|
|
hereme888 7 days ago | parent | prev | next [–] Even your benign comment got down-voted. The hatred and intolerance of this NPC army knows no end. reply
|
|
#
?
Mar 6, 2024 18:36
|
|
|
mrmcd posted:Patriots are in control. Trust the plan. wwg1wga lmao it's exactly this. Elon is actually the shadow CEO of every company and is sending us acrostic messages in patch notes
|
|
#
?
Mar 6, 2024 19:21
|
|
|
fritz posted:dorkwood 18 minutes ago | parent | context | prev | next [–] | on: OpenAI and Elon Musk guess we found elong’s hn profile name
|
|
#
?
Mar 6, 2024 20:07
|
|
|
mystes posted:I don't think this is accurate at all. Google intentionally put in work to make sure it didn't have the problems that other AI had in terms of being biased too far toward generating white people. The problem is that dumb people were able to intentionally prompt it to generate images that didn't make sense and use that to start a bullshit controversy that is now being exploited by right wingers to demand that AI be made more right wing. hopefully, between the right wing complaining about it generating too many nonwhite people and the left wing complaining about it generating only white people, these companies will simply throw up hands and stop investing in AI image generation
|
|
#
?
Mar 6, 2024 21:05
|
|
|
mystes posted:I don't think this is accurate at all. Google intentionally put in work to make sure it didn't have the problems that other AI had in terms of being biased too far toward generating white people. The problem is that dumb people were able to intentionally prompt it to generate images that didn't make sense and use that to start a bullshit controversy that is now being exploited by right wingers to demand that AI be made more right wing. issue kind of is that the thing they did is entirely insufficient and actually ridiculously primitive compared to the process they used to put that bias into the model to start with. putting "end war" stickers on the bombs kind of poo poo.
|
|
#
?
Mar 7, 2024 00:49
|
|
|
Juul-Whip posted:hopefully, between the right wing complaining about it generating too many nonwhite people and the left wing complaining about it generating only white people, these companies will simply throw up hands and stop investing in AI image generation I have shamelessly used the Google fracas to argue that while there are uses for LLMs in our line of work (I'm messing with them as first-stage review tools for things like API design to standards), image stuff is a minefield we should never go near tracecomplete fucked around with this message at 01:00 on Mar 7, 2024 |
|
#
?
Mar 7, 2024 00:57
|
|
|
_heimdall 9 hours ago | prev | next [–] Easy solution here is for every traveler to refuse to buy seats on a Boeing plane. If we honestly believe the company was at fault and has been negligent in their safety protocols, it seems like an easy answer to just not support their business. reply
|
|
#
?
Mar 7, 2024 13:54
|
|
|
yeah just don't get the treat. easy. arrghghggghh aaaa
|
|
#
?
Mar 7, 2024 14:36
|
|
|
whoknowsidont 9 hours ago | parent | context | flag | favorite | on: Leadership is a hell of a drug >It's also easy to jump on the "management sucks" bandwagon. Global warming is real, the sky is blue, etc. >This person sounds like an energy vampire. The only thing worse than grumpy people are the posiopaths who keep spreading through organizations. As one other commenter put it, the emperor has no clothes. And a lot of us are _really_ tired of pretending like he does
|
|
#
?
Mar 7, 2024 18:29
|
|
|
legit misread that as "posigoths"
|
|
#
?
Mar 7, 2024 18:41
|
|
|
how do grumpy yosposiopaths rank?
|
|
#
?
Mar 7, 2024 19:08
|
|
|
alephnan 4 hours ago | parent | prev | next [–] > can only see it through your cultural lens Thanks for this. I read this as you presuming me to be a Westerner Social Justice Warrior making gender progressive criticism. Rereading my comment, I can see how it may be interpreted this way interpreted. To be clear, my original comment was not a criticism of Toriyama Akira, nor a suggesting Japan needs to be more gender progressive. The fact that people project this issue and respond so defensively is an indicator that gender issues remain a thorn in Japan. > According to Shonen jumps own statistics, about 40% of their readers are women Do they report if 40% of their executives or artists are women? In general, we shouldn't take a company or institutions words at face value and look at what they actually do. According to China Evergrande and FTX's own sheets, they had health balance sheets. By China and Japan's own statistics, they have conflicting numbers on geopolitical matters. I'm not saying Shonen Jump is in the same group as shady companies, but they are a for profit organization. Of course they are increasingly financially incentivized to not limit their customer base. If they were serious about being universal, they can change their name. Let's not evade around the fact that Shonen doesn't literally means 'boy'. I'm not suggesting that Shonen Jump needs to change their name, I'm just saying if they actually cared about gender equality, other then profiting off half the population, you'd see this reflected more. Even if this cherry picked statistic is true, there could be more to it. I acknowledge Shonen Jump is increasing its female reader base, but that doesn't mean the bulk of their content is male centric. Maybe there's one series that appeals to female audience. > They are very well aware of that fact and takes that in to account when they accept new series in to the magazine. So I'm not wrong to suggest that the industry and company has been, atleast historically, male centric? > have a theme, which explicitly is "friendship, effort and victory" It seems most of the victories are through physical combat. From my western "cultural lens", that's usually a sign of male centricity. > their reader, guys or girls, old or young Would love to see a statistic on what percentage of Shonen Jump's readers are 60 year old ladies. > I think you're down voted because you have a shallow understanding of the topic... don't fully understand how the Manga culture My understanding is shallow because the subject is not deep. It's only complex if you to do mental gymnastics to defend and explain away Japan's male patriarchy. Since you want to portray that culture is more progressive then it is, and it is I who is ignorant to how progressive the culture is, can you enlighten me why all female artist collectives like this exist? https://en.wikipedia.org/wiki/Clamp_(manga_artists). Are there major publishing houses ran by such groups of women? > (and somehow think mentioning that on a post he died, is appropriate)
|
|
#
?
Mar 8, 2024 14:27
|
|
|
mystes posted:It's really loving stupid that Google had such an overreaction to the response to Gemini sometimes making people more diverse than they actually should be based on historical contexts which it doesn't know about or whatever. non whites existing is left wing bias
|
|
#
?
Mar 8, 2024 15:00
|
|
|
re: mongolia famine warning Venkatesh10 8 hours ago | prev [–] With all the money free flowing in the world and with experience to games like city builder, Builderment.. we can easily share them food, resources from tropical regions. reply
|
|
#
?
Mar 8, 2024 20:36
|
|
|
|
| # ? Apr 29, 2024 02:35 |
|
|
incredible. sometimes these comments just land perfectly and I know I will remember them forever
|
|
#
?
Mar 8, 2024 20:42
|
|
