Overly fawning and positive tone when the terrible examples speak for themselves. I did laugh loudly at the “black man eating spaghetti” when he goes to take a second bite which transforms mid-flight into a wine glass and smooshes his face. As an unintentional comedy generator, B+. As a useful tool that isn’t built on the stolen labor of millions of youtubers, F.
The machine does a half-decent job of ‘basketball ghosts on a moving train’ and people still go ‘yeah but it’s not flawless.’
Last year this tech produced goop.
Two years ago it was impossible.
Your cynicism is dull and performative.
‘Oh it’s only stealing existing footage.’ Show me one million cats. Let me see that video. Even the failures are mostly some continuity fuckups between photorealistic frames of basically what someone asked for. In one sentence, by the way. We are still talking about whole-ass videos rendered from two dozen English words vaguely describing an arbitrary, nonsensical, and often deliberately stupid scene. Meanwhile even static-image prompts often involve twiddling numbers and keywords to tease out what the person wanted, because otherwise you get issues like these on the first thing that came out.
Right now is the worst this tech will ever be.
All these blatant issues are the seven-fingered hands you don’t really see anymore. Remember when everyone scoffed they’d be able to spot AI, because of those issues? Well… go check. Eighteen months is a long time when things move this fast.
Exactly, exactly this. At best, AI is “so bad it’s good.” But like Hollywood, generally it’s just bad and boring. I guess if it’s funny enough maybe it’s worth sharing the silly thing it did, but let’s not act like it did something cool - it fucked up in a way so surprising that it’s funny. That’s all.
Right now, sure. But remember that 10 years ago, neural net generated images were putting eyes everywhere, and wouldn’t create anything close to a believable photo. I wouldn’t be surprised if 10 years from now, video’s will have made a similar leap.
On the other hand, I do hope that between now and then, some laws will have been put in place to only train on ethically sourced datasets - which will slow down progress, but is more fair to the creators.
On the other hand, I do hope that between now and then, some laws will have been put in place to only train on ethically sourced datasets - which will slow down progress, but is more fair to the creators.
I don’t care what published works a neural network gets trained on. How else are we supposed to make one? We tried all the clever ways and they don’t work.
Nothing as miserable as copyright should prevent the obviously transformative act of grinding the entire internet into a couple gigabytes of linear algebra. The more stuff goes in, the less any single piece matters. If the robot can reproduce more than a vague resemblance to particular inputs then that’s a failure called overfitting. A network that can spit out Man Of Steel frame-by-frame won’t be good at much else. We want it to know who Superman is and how capes work. You can’t get that by scanning the same DVD over and over.
Hard, because the countries are not going to agree and do not want to limit themselves with what could end up harming them in the future.
At least that the international copyright treaties are updated (a titanic task to be fulfilled), but it does not ensure that all countries, organizations, companies or individuals follow and respect the agreements made, especially in developing countries.
It’s definitely cool from that standpoint. Don’t get me wrong. But the output from image and video AI usually doesn’t impress me all that much. It’s technically impressive but the output doesn’t surpass what a competent human can do in terms of composition, creativity, and depth. And I know the fear is that eventually it will, but I adamantly believe it never will.
It will depend on how much investment they put into the development and research they do with AI. Because this is a race and for now nobody wants to step on the brakes until there is another winter for AI.
I think it will be missing whatever you want to call the human spirit for a long time. It’s missing in writing and it’s missing in pictures. The technical ability is advancing at an impressive pace, but at the end of the day I find things generated by AI to just feel empty and dull.
I still use them and enjoy them and want to see how far they can go, but I think there is an upper bound that is below human journeyman level until we create a vastly different sort of AI that might surpass sentience and perhaps even be considered sapient. And we’re currently far short of sentient.
Pictures have an easy fix because text is the worst possible input. It’s a demo gone feral. Image-to-image can fill content and apply style over anything, which is how you get those detailed medieval landscapes where you squint and it’s Rick Astley.
The near-future use of this tech is rendering blocky CGI (or even just finger-painting keyframes) and having the machine tweak that to hell and back.
Also if you actually use AI image gens over a long period you’ll find that it creates the same thing over and over. Same poses, same angles.
It all gets very samey, which makes sense
Hope that goalpost is easy to carry.
It’s all like a bad dream
