Salv8 (sam)

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

the filmora community looks like it'll be no-mora after this.

(oh shut up someone would make it)
 
anyways, they technically advertised it on their website and since they have taken it away it could be a case of lying to the consumer. if anyone where to oh i don't know, complain to their country's advertising standard's bureau, filmora could get into a bit of trouble and i wouldn't be surprised if they had to honor the old licenses (begrudgingly at least)

the filmora community looks like it'll be no-mora after this.

(oh shut up someone would make it)
 
anyways, they technically advertised it on their website and since they have taken it away it could be a case of lying to the consumer. if anyone where to oh i don't know, complain to their country's advertising standard's bureau, filmora could get into a bit of trouble and i wouldn't be surprised if they had to honor the old licenses (begrudgingly at least)

the filmora community looks like it'll be no-mora after this.

(oh shut up someone would make it)
 
anyways, they technically advertised it on their website and since they have taken it away it could be a case of lying to the consumer. if anyone where to oh i don't know, complain to their country's advertising standard's bureau, filmora could get into a bit of trouble and i wouldn't be surprised if they had to honor the old licenses (begrudgingly at least)

the filmora community looks like it'll be no-mora after this.

(oh shut up someone would make it)
 
anyways, they technically advertised it on their website and since they have taken it away it could be a case of lying to the consumer. if anyone where to oh i don't know, complain to their country's advertising standard's bureau, filmora could get into a bit of trouble and i wouldn't be surprised if they had to honor the old licenses (begrudgingly at least)

the filmora community looks like it'll be no-mora after this.

(oh shut up someone would make it)
 
anyways, they technically advertised it on their website and since they have taken it away it could be a case of lying to the consumer. if anyone where to oh i don't know, complain to their country's advertising standard's bureau, filmora could get into a bit of trouble and i wouldn't be surprised if they had to honor the old licenses (begrudgingly at least)

Maybe not removable like an old Nokia 3310. But at least making it easier to disassemble the device. Have parts easily available, standard bit sizes, charts/guides/instructions on taking it apart and putting it back together, a way to return the old battery or a designated place to drop it off for proper disposal - stuff like that. I suppose in this day and age, most people wouldn't think too much about replacing the battery any more though, and would simply opt to buy a new one. "Oh it's already almost 3 years old, I'll just upgrade!" I think the market has been permanently ruined in that way. No one is gonna want to go through the effort of sourcing, buying, and actually replacing the battery. Hell most people just live with a cracked screen these days. The exception to the rule are people like us, in forums like this, because we think of and are into these sorts of things.

This is "steal" in the same context of software, video and music piracy, and you'll likely find the same people who rage about AI art generation are the same people who have no bones about watching pirated video or reading pirated comics.
 
Let's put aside "this" project for a moment and put all the cards on the table, because nearly everyone tries to understand AI art generation at a kindergartener level, and people fight about it in the same way Metallica vs Napster does. With the idea that the tools are expressly designed for piracy. Some people even seem to equate this with NFT's.
 
LAION and CLIP is the process that actually goes out and indexes, NOT stores images. When some entity like stable diffusion creates a corpus for training their AI models, they use these indexes (which stable diffusion , dall-e, and likely everything derived from it) to retrieve the images, resize them to 512x512, and then generate the word pairs. If it found your copyrighted material at this point, you should check the terms and conditions of where you hosted the image, because you likely gave the rights away.
 
Under US Copyright law, LAION and CLIP would be fair use since they're used for research purposes and do not devalue the commercial use of the original. They are indexes in the same way that Google is an index. LAION is the image-to-keyword index, CLIP is the identification of what that image is in English.
 
There is still a language model the AI has to learn. That is where things can be "censored", by intentionally, or accidently, where many of these AI projects are less alike. https://arstechnica.com/information-technology/2022/12/stability-ai-plans-to-let-artists-opt-out-of-stable-diffusion-3-image-training/ , Stability AI plans to let artists opt out, and my theory here is when they next generate a model, they will simply delete LAION entries that have been "opt'd out", which will degrade CLIP's ability to identify "art-theft", and people who re-uploaded artwork, but it is what it is.
 
So with Dall-E, Stable Diffusion, and things derived from it, we get to the "throw spaghetti at the wall and compare it to artwork", there've been plenty of videos on how this actually works. Suffice it to say "the actual artwork" is not, and has never been in the model. 2TB of images does not compress down to a 4GB model, no matter how you try to frame it. What is stored in the model is "weights", so if something over-fits (because it shows up repeatedly) that's a problem, but not an unsurmountable one.  There is no magic prompt that will re-generate any training image. So this is a largely overblown argument except when it comes to trying to generate artwork in a specific style.
 
Which leads us to what kicked off this "artists hate AI art generation and have behave as luddites every time someone has brought it up", Yes artists, specifically "concept artists", have screamed loudly on twitter for this stuff to die. They're not interested in learning how it works, have no interest in using it, and believe nobody should have access to it, in the off chance it leads to one less job for them. Where have I heard that argument before?
 
https://www.riaa.com/reports/the-true-cost-of-sound-recording-piracy-to-the-u-s-economy/
Yes, the RIAA, and it's fictitious "facts" about piracy where it equates every unlicensed use with a full CD sale. Even when it's 2 seconds of someone else listening to it on a stream or the background of someone's video shot at a birthday party.
 
That is what this looks like. Despite the exasperated arguments from artists, most people do not want AI generated artwork to begin with. Why would you settle for an 8-bit bit-crushed cover of a music track that takes 10 minutes to generate, when you can just go and buy the legit music off itunes and listen to the the actual thing. If people need artwork, they will continue to do what they've always done and either hire people to do it, or do it themselves.
 
All AI art generation is doing at this point is bringing "do it themselves" closer to scratch data that can take hours or days to find a "prompt" that gives the user anything remotely resembling what they want. And all the cherry-picked stuff you see on twitter and in news articles is rarely "prompts", but rather image2image.
 
image2image, is the style transfer of one image to another. And the technology isn't even new
https://affinelayer.com/pixsrv/ is from 2017.
 
All ML AI stuff really is at this point is "auto-complete", and the conflation of the image2image with "all AI art generation" is what makes this a mess, because the image2image style transfer is the only case where you can in fact "steal an artists style", but that is still limited to what the AI understands, and the AI understands nothing about composition. This is why these AI art generators universally suck at rendering fingers, text, and eye-directions/colors. People don't label those things way back in the LAION dataset. Hell the LAION dataset is biased towards actual photographic images and not "artwork" in the first place. This is why asking the AI with a prompt for something that only became a fad or popular in 2021 or 2022 will produce at best garbage, and at worst, nothing.
 
The models will constantly have to be retrained on new data, otherwise they will be stuck in the past.
 
Which brings us to the other half of the argument. Consent and "why does it not just use Public Domain data"
 
Because if you train an AI only on public domain data, you will get an AI that only produces things with the understanding of a 19th century scholar, at best. All the racist, sexist, and ableist language that has gone out of style will be in there. Ideally one would attempt to get consent to include specialized datasets so that the model, could generate celebrities likeness, but that's unlikely to ever be the case, and waiting a century for them to die and be irrelevant is an even worse argument. So I believe Stability Diffusion is correct, they do not need the consent to use the images if they've been posted to the internet, because they can lean on the research part of fair use to do so.
 
That said, "making money" off the research dataset? That violates the spirit of the fair use laws. They should always release the production model  to the public if they want to use the research argument as a means of generating it.
 
And before anyone chimes in and goes "what about (imagen, midjourney, etc)", no it's very likely they have used LAION too, or have done something similar to it. 
 
AI research is very "reuse old data forever" as far as datasets are used. Other AI research like ASR/TTS and Text(GPT), it's far easier to curate what you put on the input side of these, because you can train a TTS from just 500 samples, or about 10 minutes of one person and get something sounding reasonable, but incapable of saying any zoomer language, because the training data is still only public domain books, and has to rely on a pronunciation dictionary of english words to fill in the blanks.
 
There will always be blindspots, and you have to re-train the model to fix them.
 
Now... does THIS project have merit?
 
https://updates.kickstarter.com/ai-current-thinking/
 
Alright, so what was the intent?
https://kotaku.com/kickstarter-ai-art-image-porn-unstable-diffusion-nsfw-1849921325
I think the title speaks for itself.
 
The reason Kickstarter suspended the project, was because it could "be used to create porn of people without their consent", which seems to be the goal. This seems more like a grift than it does an actual serious attempt to address the problems created by the first model released by Stable Diffusion.
 
At least Stable Diffusion got (possibly unfairly) ripped to shreds for how it presented it being able to copy an art style. Reminding me of all those crappy "emulator console's" who's entire selling point is the piracy angle.
 
You know, who doesn't get ripped to shreds? When people talk about emulators. The same arguments about "is an emulator a good or a bad thing" can be applied to "is an AI Art Generator a good or a bad thing" because people are going to pirate games even if you tell them not to, and people are going to generate AI art, even AI generated porn, even if you tell them not to. 
 
All the yelling in the world is not going to stop people from working on AI art generators, but what it will do is make people who are working on better ones, less willing to share their progress, because they don't want to be villainized. Meanwhile, corporate AI art generators end up becoming established, and profit off these "research" models.
 
And yes, I've seen it happen several times. An open source project stops being developed and the people who were working on it are now part of some commercial enterprise, doing the same thing the open source project did, but now has a fancy UI.
 

Has absolutely nothing to do with NSFW vs SFW...
 
The image set that these AI models are trained on (LAION) is filled with billions of copyrighted images/artworks. The technology is built on the backs of people who never gave consent to steal their work for profit.
 
Your freedom of software doesn't include freedom to steal other people's work. These AI models steal from millions upon millions of people.

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

https://mastodon.social/@elonjet

Ooooh goody, a new benchmark!

They should call it Prime of the Tomb Raider.

So Lara is now an employee at a warehouse and you spend 20 hours trying to get out of there alive. It'll have a stealth mission side quest that will involve getting to the toilet without Bezos noticing you.
 
To be honest, it's not surprising that there will be a new Tomb Raider game. The recent sale of the western Square Enix studios to Embracer Group along with all the IP (seriously, what was Square thinking...) pretty much cemented that these franchises wouldn't go anywhere. Though I'd imagine Crystal Dynamics fans were hoping for a new Legacy of Kain game first. But hey, Lara is a cash cow and buying studios left and right requires some income. It's interesting that Amazon Games are publishing it, though. Embracer already own a bunch of publishers and I would've thought consolidating all those development houses would've also incentivized the parent company to publish their studios' stuff. So this deal will probably involve a lot of money from Amazon.

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

About damn time.

*sniff sniff*
more powerful then the 5000 series?
yea i call bullshit, they'll skew the results to make it look like it's more powerful but when reputable sources get their hands on them it'll be a blood bath.

What will be interesting is when YTers get their hands on a few, and security analysis is run. For a 1st try, I expect nothing less than Piledriver-levels of performance...
 

so they didn't optimise shit, the PS5 has 16 GB of RAM, why do they recommend double that on PC? they are relying on the brute force method instead of putting in effort.

Whatever is necessary for next frontier of gaming hardware.

They are seeking to block the acquisition, Pointing out in their complaint that Bethesda titles have been made exclusives despite assurances to European antitrust that there was no incentive to withold games from rival consoles and mentioning the Series S and X are one of only two types of high performance game consoles. I will say I have problems with the FTC post like calling Activision Blizard a video game developer rather then publisher and don't know if I agree with the statement of high-quality video games like we aren't talking about Valve, Nintendo or Supergiant this is Activision.

The FTC hasn't yet made the administrative complaint available explaining why it has reason to believe that it needs to intervene and intervening is in the public interest.

Gotta love the pr speech
Hey Apple: what about China?