I guess we'll see. But I remember how others described Sora when it was announced. To paraphrase:
"If we were really this close to GAI, or curing cancer, or achieving 10x productivity... we wouldn't be spending resources to make an AI Slop Titok. That's what you do when you need a quick buck, right now."
Also, as many observed: this was a doomed proposition from the start. People can generate videos on Sora and then... post it to Tiktok. This was a doomed concept from the beginning.
Some context... The Sora android app was developed (85% vibecoded) by a team of 4 people over less than a month. It's not an area they've invested a huge amount of time and money into, compared to coding, business productivity, medicine and image generation.
Which is to say that despite the initial training costs they didn't invest much into it beyond that. Video is expensive in terms of compute and there's a finite supply. To me it looks like scrapping their biggest loss leader to focus on areas with profit potential.
I think they kinda knew it was a doomed concept from the beginning. They were just locked into the idea that they should try to lead on all fronts, which they've recently rethought.
Eh, consider they probably made the iOS app first, doing it more by hand than the original Android release. Vibe coding the Android version makes sense when you have a working version from another platform to work from.
Also consider how much of the code is in the back end. Apps aren’t just the part you download. In Sora’s case the UI was pretty damn simple. The stuff running on the servers, even just the stuff specifically for Sora, is probably more lines of code than what’s in the clients.
Consider that once you’ve vibe coded an app and released it to millions of people you need to add a ton of people to the team to fix bugs, remove harmful content, develop new features, etc.
Don’t focus too hard on the one aspect that was made easy because AI is pretty helpful with porting working software between well understood platforms.
According to them the android app was built from scratch, and then maintained by the same team of 4 people. I tried using it once, months after release, it was a buggy mess. They don't seem to have devoted a lot of resources to it. Meanwhile it had to be bleeding money via inference on free usage. At least with coding and enterprise people pay them.
In any case I just use the android app as example of their lack of push with Sora. They also also never built any significant scaffolding or tooling around it to support professional workflows. It was more a proof of concept than a product.
To me it looks like scrapping their biggest loss leader to focus on areas with profit potential.
Maybe. But being able to reap the ads and influence of a potential social media platform isn't exactly small beans. It's just that the hard part of such a feat isn't the tech.
They are happy to prove me wrong, but it feels like OpenAI is in for a rough 2026 unless they get a very big deal from the government (and given the war the US threw itself into, even that may not be on the table anymore).
I mean ... they invested $15 million USD per day to keep it running. Money is fungible; they could've spent that building orbital datacentres, bribing politicians, or buying RAM futures. The opportunity cost was the investment, imo, and it seems like a practical loss?
Thought that was obvious from day 1. But it's clear things behind the scene broke down starting back from Nvidia getting cold feet. The bubble has clearly popped some time ago and these people would be the first ones to see it.
Here's to hoping that's exactly what happens. God knows it's been far too long since some good news hit.
Not that a bubble popping is likely to be all good; it'll bring a lot of economic damage with it, and the current U.S. admin is likely to respond in terrifyingly awful ways. But since a pop is inevitable, that'll happen regardless. The sooner the bubble bursts, the less collateral it'll inflict.
Would love to be a fly on the wall for this one. My money is on this being the first of many "wait what do you mean it doesn't do what the salesteam says" moments.
I thought Sora was actually pretty impressive, and definitely a long way ahead of anyone else when they first demoed it (although that was a relative lifetime ago and plenty of others have been catching up). Then again, I’m sure the sales team would find a way to over promise in an absurd way even if the product literally cured cancer, so you could well still be right there!
My bet’s more on risk/reward though - video’s computationally expensive and likely to get them sued by companies with deep pockets, I imagine the economics just weren’t workable for the relatively limited amount end users were willing to pay.
I think the issue for someone like Disney is how are you even going to use sora? There's SO much cleanup that has to be done on something like that to have a professional grade product. Sure it lets some schmuck like me vomit out a video, but that video has obvious issues, and it certainly doesn't translate into a multimillion $ grossing project.
It's the problem a lot of AI has in that you still need to tweak the output, and tweaking video output is basically as difficult if not more than "fuck it i'll make it myself."
The idea was, unironically, to add a "make your own disney video" section to Disney+.
They spent over a century lobbying copyright in their favor and they were willing to unleash the floodgates and let everyone build out AI sloppification of their most valuable IPs. It was mindboggling to behold
An unironic realization here that may have cut the deal off is that judges have consistently ruled AI being unable to be protected under copyright. I suppose even Disney's lawyers couldn't figure out how to trademark loophole around that one with confidence.
Yes, the copyright office's unchanged position on this has been the one bright spot of the era. You cannot copyright works generated by GenAI. That severely damages the potential profits you can monopolise from it, and makes it not worth it.
It wouldn't necessarily need to be for movies. I was thinking interactive in-person environments. Like maybe you can actually interact with an animated Elsa in real time in some new Disneyland ride or something.
Even then the difference between “oh that’s neat” and “good enough to go in THE theme park of the world” is way way way too high.
At the point it’s generating video , not just script, on demand you’re going to get all sorts of artifacts and oddities that are going to make it feel cheap, while also costing Disney a boatload of cash in the process.
Animation, unlike code or 2d art can’t be “close enough” and then just someone good goes in and cleans it up. It’s very hard to get everything to play nice as is, and that’s before you’re left with a flat scene as an output with no rigging or lighting
Except this is like investing in more freeways when what you need is a train.
AI will be useful in the movie industry but Sora wasn’t even headed in the right direction. Further I’ve no doubt Disney pulled the money because they’re not looking to gift it but expecting a product they can use
Personally compared to veo I didnt find Sora that technically impressive. However, I legitimately enjoyed using it on occasion. I dont generally like doomscrolling apps so what I liked about this is that every user was some sort of creator. I think there's something to lowering the barrier to content creation like this, AI or no AI.
I see a lot of people celebrating. I think my concern would be that this is not the end of AI video generation but rather the start of the technology continuing to be used by nefarious entities behind the scenes.
Say what you want about Sora but the fact that the common person could easily use the technology caused many to develop a familiarity with AI generated video and thus helped enable them to identify it in the wild. Open source models exist but they aren’t any as good as Sora to my knowledge and they certainly aren’t as easy for the common person to use.
I could be wrong. Video models are hard and expensive to train. Maybe open AI buckled under financial pressure, though I’m sure their juicy government deal will help with that.
From my understanding, research seems to indicate that video models are incredibly powerful for a large range of tasks beyond creating “slop”. I’ve seen some pretty bold claims about their potential. So it seems suspicious to me that open AI wouldn’t continue investing down that route. I suspect that research and development of this will continue but you and I won’t be able to use it easily.
I think it’s worth celebrating because it’s one of the AI tasks that take up the most resources and thus are the most impactful for the environment. It may not be stopping this kind of things from happening in the future but it’s at least slowing things down.
Best news all day. I'm not going naive enough to hope that this collapses the whole house of cards, but high profile failures like this might have a chance at spooking corps.
Can't figure out what I'm happier about: The news itself, or the fact that it's got such a positive reception here.
I'd felt for the last while like I'd be arguing myself blue in the face dealing with thread after thread seemingly filled with people crawling out of the woodwork to defend AI. Usually with regard to coding, but more generally as well. It got bad enough for my health that I filtered the tag and stopped reading Tildes logged out as I sometimes used to. I'm only here because I already heard the news elsewhere, and so temporarily skipped past my filter.
Needless to say, both this news and this thread are much-needed breaths of fresh air.
I find it sad that the companies in control of tech I find fascinating are such unmitigated arseholes that everyone's celebrating their downfall. And to be very clear here: I don't think anyone's wrong to be pleased, I just see it as yet another "this is why we can't have nice things" when the incentives favour organisations that justify that reaction.
I watched us go from text interfaces to blocky 16 colour sprites to Mario to Sonic to Jurassic Park to Toy Story to Tomb Raider to Skyrim to Avatar. Transformative jumps in quality happened every few years, and then went from huge studio projects to rendering in real time at home a few years later, and my jaw dropped every time. I had the same reaction to seeing a computer create photorealistic video from nothing but a rough description barely two years back, even if it does have the occasional imperfection.
To anyone who's followed or worked on computer generated imagery in the 90s and 2000s, it's a genuinely stunning technical achievement - doubly so considering it came only a decade after even getting a computer to recognise a bird in a still image was an unreasonable ask, let alone asking it to create video of one in flight from scratch.
But yeah. I'm starting to believe none of that really matters. We as individuals are seeing fuck all benefit from it, the astronomically wealthy are using it as yet another dice game with the economy, and their minions are using it as yet another avenue to spread misinformation and set us against each other.
I've said many times what incredible technical achievements these models are, and I stand by that. Some of the researchers involved have made enormous advancements in our understanding of information theory, statistical modelling, and both physical and computer science more broadly.
So I'm sad for them, and even more so for what it says about our society as a whole, that the incentive structure we've built is to use all of that for output that's legitimately called slop 99% of the time. I'm sad that we've so thoroughly given up on any advancements being used for the greater good that we celebrate their downfall, and that we're probably right to do so.
semsevfor | 21 hours ago
Is this the first domino? I really hope so. Let it all fall
post_below | 14 hours ago
They've been pretty public about refocusing on coding and enterprise, they're concerned about Anthropic overtaking them.
The bubble may pop, but I don't think this is part of it.
raze2012 | 14 hours ago
I guess we'll see. But I remember how others described Sora when it was announced. To paraphrase:
"If we were really this close to GAI, or curing cancer, or achieving 10x productivity... we wouldn't be spending resources to make an AI Slop Titok. That's what you do when you need a quick buck, right now."
Also, as many observed: this was a doomed proposition from the start. People can generate videos on Sora and then... post it to Tiktok. This was a doomed concept from the beginning.
post_below | 13 hours ago
Some context... The Sora android app was developed (85% vibecoded) by a team of 4 people over less than a month. It's not an area they've invested a huge amount of time and money into, compared to coding, business productivity, medicine and image generation.
Which is to say that despite the initial training costs they didn't invest much into it beyond that. Video is expensive in terms of compute and there's a finite supply. To me it looks like scrapping their biggest loss leader to focus on areas with profit potential.
I think they kinda knew it was a doomed concept from the beginning. They were just locked into the idea that they should try to lead on all fronts, which they've recently rethought.
teaearlgraycold | 10 hours ago
Eh, consider they probably made the iOS app first, doing it more by hand than the original Android release. Vibe coding the Android version makes sense when you have a working version from another platform to work from.
Also consider how much of the code is in the back end. Apps aren’t just the part you download. In Sora’s case the UI was pretty damn simple. The stuff running on the servers, even just the stuff specifically for Sora, is probably more lines of code than what’s in the clients.
Consider that once you’ve vibe coded an app and released it to millions of people you need to add a ton of people to the team to fix bugs, remove harmful content, develop new features, etc.
Don’t focus too hard on the one aspect that was made easy because AI is pretty helpful with porting working software between well understood platforms.
post_below | 8 hours ago
According to them the android app was built from scratch, and then maintained by the same team of 4 people. I tried using it once, months after release, it was a buggy mess. They don't seem to have devoted a lot of resources to it. Meanwhile it had to be bleeding money via inference on free usage. At least with coding and enterprise people pay them.
In any case I just use the android app as example of their lack of push with Sora. They also also never built any significant scaffolding or tooling around it to support professional workflows. It was more a proof of concept than a product.
raze2012 | 11 hours ago
Maybe. But being able to reap the ads and influence of a potential social media platform isn't exactly small beans. It's just that the hard part of such a feat isn't the tech.
They are happy to prove me wrong, but it feels like OpenAI is in for a rough 2026 unless they get a very big deal from the government (and given the war the US threw itself into, even that may not be on the table anymore).
kacey | 26 minutes ago
I mean ... they invested $15 million USD per day to keep it running. Money is fungible; they could've spent that building orbital datacentres, bribing politicians, or buying RAM futures. The opportunity cost was the investment, imo, and it seems like a practical loss?
redwall_hp | 16 hours ago
A Tom Morello song just popped into my head. Any loss for OpenAI is a win for the world.
Disney realized the salesman duct taped the door onto the car they were buying.
raze2012 | 14 hours ago
Thought that was obvious from day 1. But it's clear things behind the scene broke down starting back from Nvidia getting cold feet. The bubble has clearly popped some time ago and these people would be the first ones to see it.
Gourd | 20 hours ago
That was my first thought. Should be interesting regardless, because this is not a good look.
LukeZaz | 8 hours ago
Here's to hoping that's exactly what happens. God knows it's been far too long since some good news hit.
Not that a bubble popping is likely to be all good; it'll bring a lot of economic damage with it, and the current U.S. admin is likely to respond in terrifyingly awful ways. But since a pop is inevitable, that'll happen regardless. The sooner the bubble bursts, the less collateral it'll inflict.
Eji1700 | 21 hours ago
Would love to be a fly on the wall for this one. My money is on this being the first of many "wait what do you mean it doesn't do what the salesteam says" moments.
Greg | 20 hours ago
I thought Sora was actually pretty impressive, and definitely a long way ahead of anyone else when they first demoed it (although that was a relative lifetime ago and plenty of others have been catching up). Then again, I’m sure the sales team would find a way to over promise in an absurd way even if the product literally cured cancer, so you could well still be right there!
My bet’s more on risk/reward though - video’s computationally expensive and likely to get them sued by companies with deep pockets, I imagine the economics just weren’t workable for the relatively limited amount end users were willing to pay.
Eji1700 | 18 hours ago
I think the issue for someone like Disney is how are you even going to use sora? There's SO much cleanup that has to be done on something like that to have a professional grade product. Sure it lets some schmuck like me vomit out a video, but that video has obvious issues, and it certainly doesn't translate into a multimillion $ grossing project.
It's the problem a lot of AI has in that you still need to tweak the output, and tweaking video output is basically as difficult if not more than "fuck it i'll make it myself."
raze2012 | 13 hours ago
The idea was, unironically, to add a "make your own disney video" section to Disney+.
They spent over a century lobbying copyright in their favor and they were willing to unleash the floodgates and let everyone build out AI sloppification of their most valuable IPs. It was mindboggling to behold
https://www.hollywoodreporter.com/business/digital/disney-plus-gen-ai-user-generated-content-1236426135/
An unironic realization here that may have cut the deal off is that judges have consistently ruled AI being unable to be protected under copyright. I suppose even Disney's lawyers couldn't figure out how to trademark loophole around that one with confidence.
Raistlin | 12 hours ago
Yes, the copyright office's unchanged position on this has been the one bright spot of the era. You cannot copyright works generated by GenAI. That severely damages the potential profits you can monopolise from it, and makes it not worth it.
HiddenTig | 15 hours ago
It wouldn't necessarily need to be for movies. I was thinking interactive in-person environments. Like maybe you can actually interact with an animated Elsa in real time in some new Disneyland ride or something.
Eji1700 | 14 hours ago
Even then the difference between “oh that’s neat” and “good enough to go in THE theme park of the world” is way way way too high.
At the point it’s generating video , not just script, on demand you’re going to get all sorts of artifacts and oddities that are going to make it feel cheap, while also costing Disney a boatload of cash in the process.
Animation, unlike code or 2d art can’t be “close enough” and then just someone good goes in and cleans it up. It’s very hard to get everything to play nice as is, and that’s before you’re left with a flat scene as an output with no rigging or lighting
updawg | 13 hours ago
But that's why you invest in future technologies. To make current technology better.
Eji1700 | 4 hours ago
Except this is like investing in more freeways when what you need is a train.
AI will be useful in the movie industry but Sora wasn’t even headed in the right direction. Further I’ve no doubt Disney pulled the money because they’re not looking to gift it but expecting a product they can use
BroiledBraniac | 18 hours ago
Personally compared to veo I didnt find Sora that technically impressive. However, I legitimately enjoyed using it on occasion. I dont generally like doomscrolling apps so what I liked about this is that every user was some sort of creator. I think there's something to lowering the barrier to content creation like this, AI or no AI.
JesusShuttlesworth | 16 hours ago
I see a lot of people celebrating. I think my concern would be that this is not the end of AI video generation but rather the start of the technology continuing to be used by nefarious entities behind the scenes.
Say what you want about Sora but the fact that the common person could easily use the technology caused many to develop a familiarity with AI generated video and thus helped enable them to identify it in the wild. Open source models exist but they aren’t any as good as Sora to my knowledge and they certainly aren’t as easy for the common person to use.
I could be wrong. Video models are hard and expensive to train. Maybe open AI buckled under financial pressure, though I’m sure their juicy government deal will help with that.
From my understanding, research seems to indicate that video models are incredibly powerful for a large range of tasks beyond creating “slop”. I’ve seen some pretty bold claims about their potential. So it seems suspicious to me that open AI wouldn’t continue investing down that route. I suspect that research and development of this will continue but you and I won’t be able to use it easily.
Akir | 14 hours ago
I think it’s worth celebrating because it’s one of the AI tasks that take up the most resources and thus are the most impactful for the environment. It may not be stopping this kind of things from happening in the future but it’s at least slowing things down.
Raistlin | 19 hours ago
Best news all day. I'm not going naive enough to hope that this collapses the whole house of cards, but high profile failures like this might have a chance at spooking corps.
LukeZaz | 8 hours ago
Can't figure out what I'm happier about: The news itself, or the fact that it's got such a positive reception here.
I'd felt for the last while like I'd be arguing myself blue in the face dealing with thread after thread seemingly filled with people crawling out of the woodwork to defend AI. Usually with regard to coding, but more generally as well. It got bad enough for my health that I filtered the tag and stopped reading Tildes logged out as I sometimes used to. I'm only here because I already heard the news elsewhere, and so temporarily skipped past my filter.
Needless to say, both this news and this thread are much-needed breaths of fresh air.
Greg | 6 hours ago
I find it sad that the companies in control of tech I find fascinating are such unmitigated arseholes that everyone's celebrating their downfall. And to be very clear here: I don't think anyone's wrong to be pleased, I just see it as yet another "this is why we can't have nice things" when the incentives favour organisations that justify that reaction.
I watched us go from text interfaces to blocky 16 colour sprites to Mario to Sonic to Jurassic Park to Toy Story to Tomb Raider to Skyrim to Avatar. Transformative jumps in quality happened every few years, and then went from huge studio projects to rendering in real time at home a few years later, and my jaw dropped every time. I had the same reaction to seeing a computer create photorealistic video from nothing but a rough description barely two years back, even if it does have the occasional imperfection.
To anyone who's followed or worked on computer generated imagery in the 90s and 2000s, it's a genuinely stunning technical achievement - doubly so considering it came only a decade after even getting a computer to recognise a bird in a still image was an unreasonable ask, let alone asking it to create video of one in flight from scratch.
But yeah. I'm starting to believe none of that really matters. We as individuals are seeing fuck all benefit from it, the astronomically wealthy are using it as yet another dice game with the economy, and their minions are using it as yet another avenue to spread misinformation and set us against each other.
I've said many times what incredible technical achievements these models are, and I stand by that. Some of the researchers involved have made enormous advancements in our understanding of information theory, statistical modelling, and both physical and computer science more broadly.
So I'm sad for them, and even more so for what it says about our society as a whole, that the incentive structure we've built is to use all of that for output that's legitimately called slop 99% of the time. I'm sad that we've so thoroughly given up on any advancements being used for the greater good that we celebrate their downfall, and that we're probably right to do so.