OpenAI is at war with its own Sora video testers following brief public leak

Group behind stunt says they're being used for "unpaid R&D" and "art washing."

See full article...

 

[ThatEffer]

A video of a dog riding a skateboard while wearing sunglasses. Will wonders never cease?

 

[Voo42]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

 

[Digitalclips]

I'm waiting on something like DiffusionBee that is for local generative video AI. Only a couple of years ago it was unimaginable that the likes of Stable Diffusion could run locally. With my Mac Studio M2 Ultra I can run generative AI image generation in seconds. I am pretty sure 10 second HD or even 4K video clips generated locally, are just around the corner without the need to pay, or work via some web site for cloud computing. Yes, the training of models will need massive computing power but once created they can be downloaded and used locally. Colab and the like are available for the grunt work.

 

[Fabermetrics]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

And here I am wishing Grinding Gear Games would exploit me with a Beta key for POE2.

 

[Psyborgue]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

Because it’s trolls. Bullshit “artists”.

 

[VaireWeaver]

This ridiculous hype train cannot crash soon enough.

 

[Psyborgue]

And here I am wishing Grinding Gear Games would exploit me with a Beta key for POE2.

Can I be please be exploited too? I feel like a Eurythmics song here.

 

[drmannevond]

I'm waiting on something like DiffusionBee that is for local generative video AI. Only a couple of years ago it was unimaginable that the likes of Stable Diffusion could run locally. With my Mac Studio M2 Ultra I can run generative AI image generation in seconds. I am pretty sure 10 second HD or even 4K video clips generated locally, are just around the corner without the need to pay, or work via some web site for cloud computing. Yes, the training of models will need massive computing power but once created they can be downloaded and used locally. Colab and the like are available for the grunt work.

You can already do this if you have the horsepower for it. Cogvideo and Mochi can both be run locally on consumer hardware.

 

[feersum]

A video of a dog riding a skateboard while wearing sunglasses. Will wonders never cease?

There's going to be an overwhelming amount of drivel and visual clutter "created" by these "tools." My sense of wonder is almost gone, and my expectations couldn't be lower.

 

[HamHands_]

Is this a viral marketing stunt?

 

[spinicist]

A video of a dog riding a skateboard while wearing sunglasses. Will wonders never cease?

Why does the baby fart?

 

[jaqrah]

Why does the baby fart?

I think a more important question is: how is that baby breathing?

 

[42Kodiak42]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

To piece together what's going on a little better, here's the correct order of events:

People, particularly artists, believe that OpenAI has exploited them by using the content of their work (posted online for general viewing) as a critical data component in the production of OpenAI products without compensation or permission.

These people who believe they have been exploited then volunteer for testing OpenAI products, I assume while withholding their grievances with OpenAI's actions from their volunteer applications.

These 'volunteers' with pre-existing grievances then deliberately leak Sora's results alongside an open-letter.

Said open letter also lists the volunteer work as exploitation*, among the grievances they had prior to the volunteer program.

It looks to me like a bunch of disgruntled artists decided to infiltrate OpenAI's volunteer outsourcing to deliberately sabotage OpenAI.


* I think it's worth going over what sorts of software testing can be involved, as most people are probably only familiar with user-beta testing (the software dev-technical definition of beta testing (tested on end user-machine in final environment), not the games-industry popular definition). There are three sorts of testing that they can be complaining about.

1. User-beta-testing: This is non-professional testing performed by releasing software to end-users and asking them politely to tell you if anything goes wrong. Given the lack of real expectations or requirements for the testers, this is typically unpaid. It's possible this is the level of testing they're complaining about, but unlikely as this sort of work is almost never part of a job or paid position.
2. Professional-development testing: You give someone the express job of figuring out where your software breaks. This is done in house and is paid work. This isn't the level of testing they're complaining about, you can't get unpaid volunteers to do a good job here.
3. Neural Network Testing/Grading: This isn't software testing, it's called testing, but you're not looking for bugs in the software. This is a specific part of Neural Network (Most cutting edge AI) development that serves the purpose of calibrating the AI for correctness. You look at AI results, and tell the AI if it's right or wrong. This can be paid work (It's either a duty of a wider development job, or it's performed on a 'mechanical turk' basis), but it's unskilled enough for volunteers to perform. I think it's likely that this is the sort of volunteer work they were tasked with: it's still a critical component of product development, but it's not a highly skilled professional position.

 

[danilluzin]

It looks to me like a bunch of disgruntled artists decided to infiltrate OpenAI's volunteer outsourcing to deliberately sabotage OpenAI.

"We are not against the use of AI technology as a tool for the arts (if we were, we probably wouldn't have been invited to this program)," PR Puppets writes.

To me those "artists" seem pretty ok with exploiting other artists whose work was used to train AI they just want their "efforts" in testing it to be rewarded

 

[MRL256]

I remember listening to the Prof G podcast about a year ago. They had on (sorry I do not remember whom) a guest who said that he thought the AI revolution would have all of us struggling to find human created content. His basic premise was that AI generated will be so fast and easy to create that it will quickly replace almost everything else.

I think this is just the latest sign that his vision is going to come to pass.

 

[rocky76]

The dog riding the skateboard is a small clip from the new teaser trailer for the live action "The Itchy & Scratchy & Poochie: The Movie" .

 

[MagicDot]

This type of technology, which is one of the few which probably deserves the AI description, is going to create a non-stop glut of video across all platforms. I feel sorry for people who are already addicted to TikTok and other platforms which exploit the weak...2025 may be your final days ever spent outdoors.

 

[planetguy]

If there's one thing OpenAI is known for, it's compensating the hardworking creative professionals whose efforts make their products possible /s

 

[kaibelf]

"I demand to be compensated for this thing that I did voluntarily because my true goal is actually money."

 

[Fatesrider]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

This is the confusing part to me:

But OpenAI says participation in the early alpha test is voluntary and has no requirements that testers provide feedback.

Isn't the POINT of testing to provide feedback? Or are they just testing the backend and needed more outside traffic to hit the servers? I'd personally expect the backend to be up and running and ready for an alpha, unless they're looking at the traditionally named stages of a release a lot differently than the rest of the industry.

I wouldn't expect people to be paid to participate in an alpha, or beta, for that matter. So yeah, very confusing messaging coming out here.

Maybe they had their latest AI do the press releases?

 

[kaibelf]

Can I be please be exploited too? I feel like a Eurythmics song here.

Waiting for the game remains a "Thorn in My Side," if I got "The Gift" of a code, I'd feel like a "Wide Eyed Girl" who is "17 Again," and my song would switch from "Sweet Dreams" to "Here Comes the Rain Again."

"Don't Ask Me Why" I chose to use my "Caveman Head" for such a blue-humor metaphor. I accept the "Shame" and you can "Put the Blame on Me." I promise, this is "The Last Time" I'll go down such a "Dark Road" of puns.

Sigh. "I Need a Man."

 

[Psyborgue]

You know, I respect that you're one of the few people here that isn't a full blown luddite. But I'll never understand why your beef with OSS is.

I have zero beef with open source. I have contributed to actual open source projects including at least one you assuredly use.

My issue with "open" models is:

1. It's open-washing. Binary blobs that take trillions of tokens to compile are in no way open source.
2. The "open" model community is toxic. It tends to be filled with people with a dislike of "alignment" for a variety of reasons.
   1. They don't like that the model won't generate erotica. This one I am sympathetic to.
   2. They don't like that the models refuse to generate hate speech.
3. As is the usual case, when Nazis are let into the bar, they take over progressively, infecting a community.
   Hugging Face in particular does not do a very good job of moderation. They allow:
   1. 4chan's waifu research department. Again. Fundamentally I have no issue here, but when it comes from 4chan, it comes with baggage. And it does come from 4chan's /g/ LMG (Local Models General) threads.
   2. They allow hateful datasets that exclude minorities. The resulting models are labeled "Uncensored" when they have just been aligned politically with the authors and can generate hate speech.

 

[Robin-3]

"We are not against the use of AI technology as a tool for the arts (if we were, we probably wouldn't have been invited to this program)," PR Puppets writes. "What we don't agree with is how this artist program has been rolled out and how the tool is shaping up ahead of a possible public release. We are sharing this to the world in the hopes that OpenAI becomes more open, more artist friendly and supports the arts beyond PR stunts."

"We are not against the use of human faces as an option for leopard nutritional supplementation (if we were, we probably wouldn't have agreed to be invited to this program,)" Leopard Leftovers writes. "What we don't agree with is how this program has been rolled out to the participating nutritional resources and how their engagement is being handled ahead of a possible public release. We are sharing this with the world in the hopes that OpenLeopard becomes more open, more human friendly, and supports the humane treatment of its nutritional resources beyond PR stunts."

 

[KjellRS]

(....) Neural Network Testing/Grading: This isn't software testing, it's called testing, but you're not looking for bugs in the software. This is a specific part of Neural Network (Most cutting edge AI) development that serves the purpose of calibrating the AI for correctness. You look at AI results, and tell the AI if it's right or wrong. This can be paid work (It's either a duty of a wider development job, or it's performed on a 'mechanical turk' basis), but it's unskilled enough for volunteers to perform. I think it's likely that this is the sort of volunteer work they were tasked with: it's still a critical component of product development, but it's not a highly skilled professional position.

For something like a video generator I doubt it's a matter of right or wrong, but there's a lot of work that goes into preference optimization - the most common being direct preference optimization (DPO). In short they're trying to figure out what it is users really want to see based on A/B testing of their generations, even though they're not articulating their expectations in the prompt and it's not showing up on the technical metrics.

 

[Celery Man]

A video of a dog riding a skateboard while wearing sunglasses. Will wonders never cease?

 

[Celery Man]

I remember listening to the Prof G podcast about a year ago. They had on (sorry I do not remember whom) a guest who said that he thought the AI revolution would have all of us struggling to find human created content. His basic premise was that AI generated will be so fast and easy to create that it will quickly replace almost everything else.

I think this is just the latest sign that his vision is going to come to pass.

I think it’ll replace content aimed at the lowest common denominator, like Hollywood blockbusters or Top 40 music, but that’s never been where the artistic side of human content has thrived.

There are still people releasing music on cassette tapes and vinyl records, people still painting with oils… there will always be a niche population of artists making human content, thankfully.

 

[TetsFR]

Am I understanding that right: They're voluntarily participating in an alpha release and are complaining that they're being "exploited"?

Very confusing.

Hum, not really confusing, just tactics. like when you want to win a point at badminton by deceiving your opponent I guess?

 

[A_Very_Tired_Geek]

To piece together what's going on a little better, here's the correct order of events:

People, particularly artists, believe that OpenAI has exploited them by using the content of their work (posted online for general viewing) as a critical data component in the production of OpenAI products without compensation or permission.

These people who believe they have been exploited then volunteer for testing OpenAI products, I assume while withholding their grievances with OpenAI's actions from their volunteer applications.

These 'volunteers' with pre-existing grievances then deliberately leak Sora's results alongside an open-letter.

Said open letter also lists the volunteer work as exploitation*, among the grievances they had prior to the volunteer program.

It looks to me like a bunch of disgruntled artists decided to infiltrate OpenAI's volunteer outsourcing to deliberately sabotage OpenAI.


* I think it's worth going over what sorts of software testing can be involved, as most people are probably only familiar with user-beta testing (the software dev-technical definition of beta testing (tested on end user-machine in final environment), not the games-industry popular definition). There are three sorts of testing that they can be complaining about.

1. User-beta-testing: This is non-professional testing performed by releasing software to end-users and asking them politely to tell you if anything goes wrong. Given the lack of real expectations or requirements for the testers, this is typically unpaid. It's possible this is the level of testing they're complaining about, but unlikely as this sort of work is almost never part of a job or paid position.
2. Professional-development testing: You give someone the express job of figuring out where your software breaks. This is done in house and is paid work. This isn't the level of testing they're complaining about, you can't get unpaid volunteers to do a good job here.
3. Neural Network Testing/Grading: This isn't software testing, it's called testing, but you're not looking for bugs in the software. This is a specific part of Neural Network (Most cutting edge AI) development that serves the purpose of calibrating the AI for correctness. You look at AI results, and tell the AI if it's right or wrong. This can be paid work (It's either a duty of a wider development job, or it's performed on a 'mechanical turk' basis), but it's unskilled enough for volunteers to perform. I think it's likely that this is the sort of volunteer work they were tasked with: it's still a critical component of product development, but it's not a highly skilled professional position.

I don't feel the disdain a lot of people are apparently feeling for the artists in this case. I think what's going on is this is group three, but what makes their contributions valuable is that they are artists and they're being asked to use those skills to train a neural network that is netting the corporation billions of dollars while not being compensated either in the first run up to the splashy releases a couple years ago (which is arguably infringement), nor are they being considered valuable enough to compensate now despite their contributions being necessary to the development process. It's not that it's easy (and no one should be duped into [thinking] being an artist is easy, cuz it's not) it's that they're necessary.

One can quibble over the methods, but it's very, very clear these AI corporations are exploitive and not at all after anything but their own ends, not the betterment of others. One should not attribute this to negligence at this point, it's deliberate exploitation of the not only the commons, but pre-commons works and exploitation of an arthritic copyright protection system that wasn't ready for what was coming.

 

[Geebs]

I think it’ll replace content aimed at the lowest common denominator, like Hollywood blockbusters or Top 40 music, but that’s never been where the artistic side of human content has thrived.

There are still people releasing music on cassette tapes and vinyl records, people still painting with oils… there will always be a niche population of artists making human content, thankfully.

Realistically, AI video in its current incarnation is only really going to impact short-form video content like vines/tiktoks. Piecing together ten second segments of unrelated hallucinated drek into anything longer would require actual creative effort, and if there’s one thing AI bros hate, it’s creativity.

 

[Murry Wilson]

were they under the impression OpenAI wanted to help them, empower their art, and boost their careers? Hilarious

 

[Murry Wilson]

"Realistically, AI video in its current incarnation is only really going to impact short-form video content like vines/tiktoks. Piecing together ten second segments of unrelated hallucinated drek into anything longer would require actual creative effort, and if there’s one thing AI bros hate, it’s creativity."

Nah. It's hitting everything already. It'll take over B roll in "documentary" type work.

I know lots of artists working on feature length projects as well, and a couple things scheduled for fests. It does take absolute shitloads of work, but then... anything somewhat interesting always does.

 

[Digitalclips]

You can already do this if you have the horsepower for it. Cogvideo and Mochi can both be run locally on consumer hardware.

Surely these are not run 100% locally on consumer (albeit powerful) computers, as DiffusionBee is, these are still using cloud computing as far as I can tell. Check out DiffusionBee and you will see what I mean. I will dig deeper on GitHub and apologies if I am wrong.

 

[Auie]

A video of a dog riding a skateboard while wearing sunglasses. Will wonders never cease?

Meh, 'seen it before; didn't even need a exploitative, multi-billion dollar, proprietary AI, either:

 

[ninjaneer]

I think it’ll replace content aimed at the lowest common denominator, like Hollywood blockbusters or Top 40 music, but that’s never been where the artistic side of human content has thrived.

That and the history channel will have an endless supply of poorly rendered supernova explosions and string theory representations for all of their pop science specials.

 

[otso]

I remember listening to the Prof G podcast about a year ago. They had on (sorry I do not remember whom) a guest who said that he thought the AI revolution would have all of us struggling to find human created content. His basic premise was that AI generated will be so fast and easy to create that it will quickly replace almost everything else.

I think this is just the latest sign that his vision is going to come to pass.

Good point! This is similar to what I experienced when trying to find new original music on SoundCloud. Most of it seems to be remixes (which are relatively easy to produce) and it was impossible to find original content or tell if it was original, so I gave up. So the problem exists already without AI, but with AI the problem will explode. This will definitely get out of hands. There should be clear labeling and search/filtering mechanisms, so it will be easy to find original non-AI generated content if you want.