4 ms·

I think it’s clear that what’s happening is (mostly) legal, but the better question is whether it should continue to be. It will be a bad outcome if a few peopl

I think it’s clear that what’s happening is (mostly) legal, but the better question is whether it should continue to be. It will be a bad outcome if a few people profit off of a huge group of people’s labor without their consent and without compensating them, in order to put that large group out of work. Technological advances change the labor market all the time, but it’s absolutely different to exploit your own labor in this kind of way to put you out of a job. I think if someone wants to build a research ML model, it should be fine to use whatever data you can get your hands on. But if you want to make a commercial model, you should have to have the content creators’ permission to use their data to train your product.

visarga 4y ago
It's not just the model publishers that profit from generative models, the users gain something as well. And the access barrier being lowered, now many more people can benefit. Even the artists can benefit. Let's not forget the contribution of the prompt to the final result. The users have a finger in there too. That means they are directly involved in the creative process, they deserve a claim to the copyright.
- 6gvONxR4sf7o 4y ago
  If the artists benefiting too is a reason to allow this, then the artists having a say in how their data is used shouldn’t be a problem.
  - visarga 4y ago
    Training neural nets is a novel transformative process - it only takes gradients for each training example, it doesn't save any image, just take the gradients and sums them up overlapping the influence of all the examples, mixing up everything in a way we can't tell which feature comes from which training example. More like inverting the dataset into abstract principles and then generating top-down novel examples. It's like learning, but differs in that humans can't possibly see so much data. It's not like copying, but it can generate copies of training examples as an edge case. It feels like stealing, but you can't assign who is the victim, maybe it draws a bit from multiple sources of inspiration. It's very creative and inspires all sorts of idea juxtapositions, but some people consider it's only parroting known styles. In the end I think we will want to use the new toys so we will allow training on anything. Maybe a "Do-not-use-for-training-neural-nets" flag will be observed by the conscientious labs.
ChadNauseam 4y ago
Cat’s out of the bag. Stable-diffusion runs on an m1 mac, is competitive with dall-e, and will soon be open source. Midjourney was made independently from dall-e and is arguably better. This is the most revolutionary artistic tool to ever exist and anyone with a few thousand bucks can download a script from github and train it themselves. My advice to creatives is to figure out how to thrive in this new environment because you aren’t going to be able to hold back progress or stop time. (Don't worry, us programmers will be next. Language models can already generate training data in the form of tests for code-authoring neural nets.)
- fariszr 4y ago
  I mean CoPilot is already similar, no?
- quitit 4y ago
  The creative process, like many kinds of fields applicable to business is about problem-solving. AI text-to-image generation doesn't replace that function, however it does form an excellent tool, especially when it comes to rapid conceptualisation. This will allow more people to be creative problem solvers without needing to possess technical skills in image creation. Much in the same way that graphics apps allowed people to make image without needing to learn studio or art skills. Or DTP tools allowed more people to publish without the tedium and high set up costs. I will still be hiring illustrators and designers, and this may be one of their tools, and it would be their responsibility to be experts in it, but it doesn't replace them - it makes them better illustrators and better artists. The right way to think about this is not that it shrinks the field, rather it opens it and accelerates it - for that it's a very welcome addition. No creative is scared of this - they're looking forward to the next-generation approach, and it's clear that 2D images are not the end point. Soon we'll have 3D (already in progress), soon we'll have music, soon we'll have this for animation and programming. One can easily imagine all sorts of ways this technology can be implemented, take a simple example of a video game: Imagine a horror genre that hones into what scares the viewer and delivers further upon that.

visarga 4y ago

It's not just the model publishers that profit from generative models, the users gain something as well. And the access barrier being lowered, now many more people can benefit. Even the artists can benefit. Let's not forget the contribution of the prompt to the final result. The users have a finger in there too. That means they are directly involved in the creative process, they deserve a claim to the copyright.

6gvONxR4sf7o 4y ago
If the artists benefiting too is a reason to allow this, then the artists having a say in how their data is used shouldn’t be a problem.
- visarga 4y ago
  Training neural nets is a novel transformative process - it only takes gradients for each training example, it doesn't save any image, just take the gradients and sums them up overlapping the influence of all the examples, mixing up everything in a way we can't tell which feature comes from which training example. More like inverting the dataset into abstract principles and then generating top-down novel examples. It's like learning, but differs in that humans can't possibly see so much data. It's not like copying, but it can generate copies of training examples as an edge case. It feels like stealing, but you can't assign who is the victim, maybe it draws a bit from multiple sources of inspiration. It's very creative and inspires all sorts of idea juxtapositions, but some people consider it's only parroting known styles. In the end I think we will want to use the new toys so we will allow training on anything. Maybe a "Do-not-use-for-training-neural-nets" flag will be observed by the conscientious labs.

ChadNauseam 4y ago

Cat’s out of the bag. Stable-diffusion runs on an m1 mac, is competitive with dall-e, and will soon be open source. Midjourney was made independently from dall-e and is arguably better. This is the most revolutionary artistic tool to ever exist and anyone with a few thousand bucks can download a script from github and train it themselves. My advice to creatives is to figure out how to thrive in this new environment because you aren’t going to be able to hold back progress or stop time. (Don't worry, us programmers will be next. Language models can already generate training data in the form of tests for code-authoring neural nets.)

fariszr 4y ago
I mean CoPilot is already similar, no?
quitit 4y ago
The creative process, like many kinds of fields applicable to business is about problem-solving. AI text-to-image generation doesn't replace that function, however it does form an excellent tool, especially when it comes to rapid conceptualisation. This will allow more people to be creative problem solvers without needing to possess technical skills in image creation. Much in the same way that graphics apps allowed people to make image without needing to learn studio or art skills. Or DTP tools allowed more people to publish without the tedium and high set up costs. I will still be hiring illustrators and designers, and this may be one of their tools, and it would be their responsibility to be experts in it, but it doesn't replace them - it makes them better illustrators and better artists. The right way to think about this is not that it shrinks the field, rather it opens it and accelerates it - for that it's a very welcome addition. No creative is scared of this - they're looking forward to the next-generation approach, and it's clear that 2D images are not the end point. Soon we'll have 3D (already in progress), soon we'll have music, soon we'll have this for animation and programming. One can easily imagine all sorts of ways this technology can be implemented, take a simple example of a video game: Imagine a horror genre that hones into what scares the viewer and delivers further upon that.