My view on the AI thing for now will always be, if an AI is created that could solve the entire field of art / graphics, then theres no reason that an AI could not be created that also solves the entire field of Science or engineering, or maths, or services or any other field… And if we had that we’d actually have the potential to transition to a post scarcity society where all goods were nearly free and people spend their time in the metaverse or something similar…
And no i don’t think AI is anywhere near that level yet or sentient…
1 Like
that’s why it’s 203 posts already and not really going anywhere
So you would people exclude from such a discussion because they don’t understand how gradient descent works, instead of explaining the overall idea behind it?
I am confused 
You had some critique regarding the usage of analogies, but I couldn’t find examples where that was obviously an issue in the most recent discussions.
Even though the videos you linked are good, they are still not useful in regards to some of the recent points that have been made (the ones where it was about whether the training dataset can be viewed as reference or not).
Hello all,
Recently as I am sure that most of us knows, AI generated programs have really made a big progress and many people are using it now. I have even seen an add-on made for Blender.
I personally have big reserve of their use as I see it as corporations are trying hard to kill the needs for artists. I thought that I only formed part of a few who saw it as a threat but after I saw a recent YouTube video I do see that really many see what is behind it.
Would encourage everyone that would read that thread to view this video and if you want share your thoughts after please do.
1 Like
Disclaimers: IANAL/TINLA, and I read the transcript, I didn’t watch the 48 minute vid . . .
Edited to add: This popped up in my Unread Topics saying it had a new reply. I clicked the link, saw the latest post, and posted the above. I didn’t realize that the notification was in error, prior to this post the last activity seems to be over a week ago. Sorry!
It did not take long, but this video shows once again the natural superiority of closed-source commercial software when the company involved does proper R&D.
In other words, proprietary tech. is once again leading the way while FOSS plays catch-up. It really depends on whether it is worth adding yet another monthly subscription bill to manage (even though Stable Diffusion remains not that shabby at the moment).
I took the time to actually listen to the whole thing, though there is not much more value in it than what’s in the conclusion to be honest. (All quotes are from the video/transcript)
This analogy simply doesn’t make sense to me:
Huge amounts of it are copyrighted images that, to be clear, you certainly wouldn’t be able to copy and paste onto even your personal blog without incurring some legal risk.
So how is it that they can include them in these models?
When someone uses the actual image, it is obviously a copy. However, the models (the trained neural networks) don’t store the actual image. It is some abstract representation of the image and especially when using large datasets, it usually isn’t possible to replicate the training examples. Though, there are certainly exceptions.
That’s quite easy to try out. I picked something random from the dataset like this (first example):
I used a bunch of different seeds, trying to get close to that example and I wasn’t successful at all. All the produced results are very far away from it, even though, during the training, it was trained with this example.
To me, that’s one of the core arguments that should be discussed. Should it be considered copying or not?
Later in the video:
The AI can perfectly achieve, through digital reproduction, memorization, and overfitting any effect it is sufficiently trained on. So while a human’s most vehement efforts to recreate the best parts of another’s work might result in acceptable deficiencies and pleasant surprises, the AI can achieve perfect theft on mass.
Neural networks can produce perfect thefts, that’s correct. But to me it is very dishonest to point it out in this way. First, it is highly unlikely that it happens (as pointed out above). And everyone who trains neural networks tries to make sure that no memorization/overfitting or as little as possible takes place, because this usually has a very negative effect on the overall quality.
A human is also ennobled by their attempts to replicate and copy, at least when doing it through hand skills rather than copy pasting parts of an artwork digitally.
The copy pasting claim is not uncommon, but I haven’t seen anyone actually demonstrating that it happens on a large scale. Sure, it can happen because of overfitting, but as mentioned, that is not desired and usually doesn’t happen.
1 Like
Except you have to compose in photography, it’s a whole fundamental of it and what makes it an art form in my opinion, To quote legendary cinematographer Vittorio Storaro “The composition of Images should be influenced by a knowledge of Painting. If we look at the world in front of us, we see around 180 degrees, and we call this view “Reality.” Only when we select one portion of this world, defining a specific space, with a particular composition, does that image become “Art”, which in Latin means “ability”. In this way, we can express ourselves in visual art” https://www.fdtimes.com/2015/12/07/the-calling-of-vittorio-storaro/
1 Like
This is not related to ART exactly, but is very interesting to mention it.
Joe Rogan interviews Steve Jobs. An interview with all dialogues and voices generated with AI.
1 Like
Stable Diffusion 2.0 has been released. At first, it looks like a surprisingly big step forward to me:
They are using a improved text encoder, incorporate depth information and a new text guided inpainting.
https://www.reddit.com/r/midjourney/
People test some bizarre ideas. Some fake movies people created:
Midjourney still has problems with anatomy, especially fingers.
2 Likes
Cory Doctorow Wants You to Know What Computers Can and Can’t Do
That link’s to his tumblr post, which includes a link to the New Yorker article (semi-paywalled and with a potentially epilepsy-triggering graphic).
“I think that the problems of A.I. are not its ability to do things well but its ability to do things badly, and our reliance on it nevertheless. So the problem isn’t that A.I. is going to displace all of our truck drivers. The fact that we’re using A.I. decision-making at scale to do things like lending, and deciding who is picked for child-protective services, and deciding where police patrols go, and deciding whether or not to use a drone strike to kill someone, because we think they’re a probable terrorist based on a machine-learning algorithm—the fact that A.I. algorithms don’t work doesn’t make that not dangerous. In fact, it arguably makes it more dangerous. The reason we stick A.I. in there is not just to lower our wage bill so that, rather than having child-protective-services workers go out and check on all the children who are thought to be in danger, you lay them all off and replace them with an algorithm. That’s part of the impetus. The other impetus is to do it faster—to do it so fast that there isn’t time to have a human in the loop. With no humans in the loop, then you have these systems that are often perceived to be neutral and empirical.”
1 Like
Anybody else tried chatgpt to make blender scripts? It’s not bad actually, might be a good learning resource as well. Or a tool for lazy folks to add comments to code.
4 Likes
I don’t see it replacing developers or doc writers any time soon, but I got to admit it’s pretty handy.
Who’s gonna write essays they’re not interested in anymore? 
Also, I don’t know a lot of languages but it’s pretty fluent in German and Hungarian.
It’s just a test. I’m not using it to write a Christmas card to my mom. I’m not a monster.
2 Likes
To me, providing artists this kind of tool is one of the best use cases of large language models. To heavily increase the value, it would need to know what is already in the Blender file to perform very specific operations as well.
I hope those kinds of models become available in the coming years, such that they can easily be fine tuned for Blender. If that happens, it might be valuable to add a text representation of shaders and geometry nodes, to allow them to be generated by the language model as well.
2 Likes
We may be further into that as we thought I don’t see this being geo nodes specific but hey…
After trying it it definitely could use some finetuning.
I try not to get overly excited about these things, but I used it just yesterday to create an addon with a handful of task specific functions that saved me a bunch of time, create a few maya scripts, help me with solid argumentation when talking to department production, as well as help me out with a bit of vex.
4 Likes