
MVP
So OpenAI's release criteria for Sora, a model for generating moving pictures, did not include, uh... the thing being able to handle movement? How curious.
OpenAI has put its video generation tool, Sora, into the hands of ChatGPT Plus and Pro users. Sora will be familiar to anyone who has fed text into an AI to receive an image. Once magical, the technology has rapidly become ubiquitous. Earlier this year, OpenAI showed off Sora, which took it a step further with short videos …
You expect that they would have a quality standard on one of their products? I'm sure they're trying to fix this, but for something that will only be used as a toy, it really can't be the top thing on their priority list. I'm sure others are working on generative video systems that can do a better job, but those will be aiming their products at things like animation studios which can afford to put a lot of effort into verifying and regenerating until they get something that they accept because they already do that and more for manual processes. OpenAI, on the other hand, is trying to sell this to individuals who don't have as good a reason to use the tool and likely not the willingness to use it in the ways necessary to repair the effects of a brute force training process.
Youtube Shorts are about 20-30 seconds long and generate a lot of views thanks to their autoplay-when-scrolled nature, and I can imagine all those garbage channels that are already crapping out AI-voice narrated drivel by the bucketload will be loving this to get more traffic their way.
> spotting something generated by AI is relatively easy
But it won't always be like that. Early B&W movies were shaky, of varying speeds and without sound. But that all changed as the tech. improved.
So it will be with AI generated content.
We all know that what we read in the printed (mainstream) news media is of dubious origin, questionable honesty and intended as much to persuade as to inform us. Until now we have tended to believe photos as evidence of an actual event and film clips as being incontrovertible, just because we have always thought they were too difficult to fake.
We now know that cannot be relied upon, either.
No it won't.
The amount of power and money required to generate even the most shitty video, can banged out by a half competent amateur in a few hours with a standard VFX package.
It's shit, and will always be shit. They are running out if training material to steal and now it's ingesting AI generated slop.
AI is just another Metaverse / Bitcoin bullshit hype machine
Once the investors realise they are not getting any money back, they will pull out.
See how much money they are burning through, then compare that to revenue. It's unsustainable.
They have seen lots of gains by throwing more GPUs at the problem. I expect that their model will get a little better. Good enough that anyone can make unimpeachable video on a whim, no. Make something that doesn't get as many obvious things wrong, yes. They have money to burn to do that for at least a little longer. If they were thinking this through logically, they wouldn't have started this in the first place, so I doubt they'll choose to give up just because their product isn't very good. They won't be in a position to improve it quickly, but they can make some slow progress.
Look how many billions of dollars was thrown and the Metaverse and its associated ilk.
Is that still going strong? Or was it quietly dumped?
ChatGPT is kept alive by investors and freebies of Microsoft (server credits). MS sees ChatGPT as competition (there words not mine. So when the investors stop funding the money pit, or when the MS freebies run out, then what?
Where is the money coming from?
It'll quietly get absorbed by MS (they pretty much own the IP) and it'll die
With that dead, those dependent on it will either have to pay more money, or die.
Once that's gone, the others will quietly wind down the hype and it'll die a slow death. After all, this is Google we are talking about.
And where is "Apple Intelligence"? /crickets.
Did you reply to the wrong comment?
I am sure that I was responding to the author's point about AI video being easy to spot. Which it is now, but won't be in the future. And effect that will have on the verisimilitude of news.
Your reply just seems like a run'o'the mill comment about all these new technologies. Some of which will succeed and others fail.
Would be interesting (or revolting) to have Sora animate some of lyrics from the twelve days of Christmas viz
six geese a-laying
seven swans a-swimming
eight maids a-milking
nine ladies dancing
ten lords a-leaping
eleven pipers piping
twelve drummers drumming
With "unrealistic physics and struggling with complex actions" I am not sure that the Sora version of eight maids a-milking wouldn't require a strong stomach.
If AI (further) enshittifies movie production, live theatre might enjoy a renaissance. CGI didn't put Wallace and Gromit out of business and I can't see Ghibli embracing this excremental technology.
Sora works surprisingly well. Even janky movement can be remedied using the it's tools as you can mark and regenerate the sections that aren't right. A lot of scenes are almost impossible to tell from commercially made film. A little know.n amazing feature is you can make videos from existing still frames or existing videos. It does really well with that.
Science fiction stuff.