I am describing how video-to-video models are better at ‘change this one thing’ than ‘make up a whole scene.’ It’s not metaphysics. It’s CGI for dummies.
Local models run on the same power draw as a video game, and some can process ten seconds of footage every five seconds. The best use - because ‘change this one thing’ works better - is processing things humans made the usual way. E.g., real actors on cardboard sets, and other things I actually said.
You’re having a much more contentious conversation inside your head. Please stop projecting traits you’d rather be arguing against, when someone points out, it does the thing it’s for. It demonstrably functions. You could make a coherent moral argument about how it was made - but you haven’t. You’ve railed against an imaginary frothing psychopath, because someone politely described utility.
How it was made is addressable, by the way. It’s fixable. There will be vegan models made from bespoke, licensed, and public-domain data. Will that change your opinion in any way? If not, that complaint is decorative.
Okay, here’s reality from the recent past: some guy recreated GPT-2 for $20. Same size, similar training data, equal performance. The original required VC funding. This guy spent pocket change. That was a year ago. That’s how much efficiency has already improved, for training these models. These assholes only spend billions because it’s exclusionary, and they’re all caught in a dollar auction to see who can lose the gentlest. I’m sorry any hypotheticals about that are incompatible with your moral crusade.
Meanwhile, it does the thing.
That’s not going to change and you kind of have to deal with it. We now have programs that just do what you ask, for any output that’s text, images, audio, or video. They often fuck up in horrifying ways. But they’re usually about what you asked for. Especially if you asked for very little. That’s quite useful where small changes are wildly complex, like ‘make this guy look like another guy.’ The robot won’t do it as good as a team of human professionals, but I don’t have a million dollars to hire a team of human professionals, and I’m betting you don’t either. You can still consider projects that involve making one guy look like another.
Literally what are you talking about.
I am describing how video-to-video models are better at ‘change this one thing’ than ‘make up a whole scene.’ It’s not metaphysics. It’s CGI for dummies.
Local models run on the same power draw as a video game, and some can process ten seconds of footage every five seconds. The best use - because ‘change this one thing’ works better - is processing things humans made the usual way. E.g., real actors on cardboard sets, and other things I actually said.
Nope, this is a bullshitters tool for people with no talent who want to pretend there is a shortcut to making good art.
The tool you are obsessed with is just a way of convincing yourself you made something when it was stolen from other human artists.
Everybody else can see that but people who have drunk the Kool Aid of AI too hard to admit it to themselves.
You’re having a much more contentious conversation inside your head. Please stop projecting traits you’d rather be arguing against, when someone points out, it does the thing it’s for. It demonstrably functions. You could make a coherent moral argument about how it was made - but you haven’t. You’ve railed against an imaginary frothing psychopath, because someone politely described utility.
How it was made is addressable, by the way. It’s fixable. There will be vegan models made from bespoke, licensed, and public-domain data. Will that change your opinion in any way? If not, that complaint is decorative.
Stop referencing promises about the future to prove your point, you sound like a door-to-door salesperson.
Okay, here’s reality from the recent past: some guy recreated GPT-2 for $20. Same size, similar training data, equal performance. The original required VC funding. This guy spent pocket change. That was a year ago. That’s how much efficiency has already improved, for training these models. These assholes only spend billions because it’s exclusionary, and they’re all caught in a dollar auction to see who can lose the gentlest. I’m sorry any hypotheticals about that are incompatible with your moral crusade.
Meanwhile, it does the thing.
That’s not going to change and you kind of have to deal with it. We now have programs that just do what you ask, for any output that’s text, images, audio, or video. They often fuck up in horrifying ways. But they’re usually about what you asked for. Especially if you asked for very little. That’s quite useful where small changes are wildly complex, like ‘make this guy look like another guy.’ The robot won’t do it as good as a team of human professionals, but I don’t have a million dollars to hire a team of human professionals, and I’m betting you don’t either. You can still consider projects that involve making one guy look like another.
That utility is new and it’s not going anywhere.