I know how it works, but you clearly don’t. I’ve watched the videos and read the papers. (Have you read any research yourself, or just watched YouTube videos of other peoples’ opinions?) I educate myself about the latest technologies.
Just because it’s original work doesn’t mean it isn’t derivative. Legally, work produced by AI is considered derivative work. You can’t use an AI-generated image in a commercial context, because legally you’re violating the copyright of anything in the data set. You can’t use AI generated code in a project, because you’re violating the copyright of the data set. Are you reading the things I’m linking? Or are you just demanding I look at your videos? Because this isn’t me saying this, it’s the law.
If you aren’t going to acknowledge my sources, why in the world would I acknowledge yours?
Sorry, but that doesn’t make sense at all. When an artist feels sad when creating something and the people experiencing it feel happy, it wouldn’t be art?
I did not read that link because its about AICopilot. We are not talking about AICopilot we are talking about DALLE-2. And you are making quite a claim that all of its content will be copyright and not be able to be used commercially… We shall see, in a few months when it gets released widely. I don’t know if you are right or not but time will tell. As far as I can tell it will not be breaching copyright because it does not work by stealing images from its data set or mashing them together.
Not sure whether you are trolling, but that is an extremely narrow definition of art. And more importantly in this context, you define art as purely human created and even then, only in certain contexts when you feel like it…
You know nothing about how copyright law works. You also know nothing about how AI works, if you’re discrediting CoPilot’s relevance to this discussion because it doesn’t have DALLE-2 in the name. I was hoping this could be an intelligent conversation, but given your refusal to acknowledge any perspectives outside of your entirely-YouTube based research (seriously, have you ever read a single technical paper about how AI works? Have you ever done ANY programmatic work with an AI? Because I have) sadly makes that impossible
Youtube videos are just as informative as papers if not more so, papers arent a special form of learning. I do know a bit about AI and about DALLE-2 although i have not nor would i claim im any kind of expert on it, im just a casual observer. Now we shall wait and see if your claim on copyright is true or not, like i said it will only be a few months, and this post will still be here. And since you are talking about papers, here is the 2 minute papers episode on it which i obviously also watched.
There is not really a definition of what art is. As I had a conversation with you, I was interested in your view and why you believe it is impossible that AI can create art. I would have no problem if you had stated that only things created by humans can be considered art or anything else. I was confused by your answers and that’s why I asked for clarifications.
Can’t say anything bad about this one, it got both the idea and style perfectly, you can easily imagine this selling on photostocks and some online article about food using it.
My prediction is that OpenAI will be quite eager to make DALLE-2 available for commercial use, as that will be nearly the only way in which they can secure subscriptions and generate revenue/funding to do their research which is very expansive and their mission is to stay ahead of the curve to find out if these technologies will be dangerous or not and how they can mitigate any damage.
For now DALLE-2 is in some kind of private exclusive beta and cannot be used commercially. But if you look at GPT-3 it is used commercially and its user base of paid subscribers generates OpenAI’s research funding.
Forgot to mention, there will also be huge pressure from millions of everyday users and business people to use DALLE-2 for commercial gain, so the pressure will definitely be on! Seriously they cant keep this stuff non-commercial forever, to do so would be ridiculous.
It is one thing to give the AI a sentence and for it to return a collage of concepts. If there needs to be any specificity about style or content this will fail. Over time AIs may become better - make it fantastical, like LOTR - but at a certain point translating all of the desired details will be as consuming as sitting with an artist to do it. You and the artist will share some amount of context that will enrich the result. Will the system be able to find and blend those contexts to get a result? Will it be a good result or the best result?
So, yes, a future system could conceivably create images in a style, but how much effort will it take to create a specific image based specific details of the objects in a specific style? What I think AI will do is take over more mass-produced activities in areas where the patron does not care about specifics, and just wants something “nice”.
For one thing, having AI generate the imagery you need could end up being a lot cheaper than a subscription to iStockPhoto or other royalty-free image shops. From a 3D perspective, having copyright-free textures made in an instant for the art pieces in your interior could also save a lot of time as well.
Sure, some jobs will be lost, but the best-case scenario is that everything is fully automated, so the cost of most items become quite low (even buildings) because of how cheaply they can be produced. It would be exciting if the advances can soon be applied to general robotics and we get a world like this.
It creates images in the exact style the user specifies, it is built on GPT-3 which understands the real world. Check the video on DALLE-2 about 1000 robots, look at the free ebook.
One of the first things i thought about with DALLE-2 was can they make a robot or machine that paints walls in any design you generate both inside and outside walls. No more brutalism grey concrete just artworks everywhere.
Artists are still going to be needed to produce the inputs to a system like this one. It will be “yet another artist’s power-tool.” It will eliminate a certain amount of artistic drudgery, but it will never replace the skill of the artist, who knows how and where to employ it in order to produce a compelling result by deadline.
Actually i think there are enough images already in existence to give any future model enough training data to produce any image. The limit at the moment is storage space on their super computers.