AI crap in compo entries?
category: general [glöplog]
Quote:
you released an objectively bad prod
Now it's getting interesting! :-)
@bifat:
OK, you tell me. Rate your prod on a scale 0-10.
It’s gonna be better than to looking as you paint yourself into something awful and cruel which I’m sure you are not.
OK, you tell me. Rate your prod on a scale 0-10.
It’s gonna be better than to looking as you paint yourself into something awful and cruel which I’m sure you are not.
No, you tell us. I'm keeping the rate to myself. Please elaborate extensively on the objectivity of your assessment. :-)
I won’t. The prod is so good that I refuse to discuss it further. I said what I had to say before. Krill accused me of “reiterating” when I tried to give constructive criticism. So you can either roll back and look. Or not. But c’mon are you so insecure that you can’t rate your own prod for us all?
Quote:
Just please read this thread from the start to end. After you’ve read it, if you still have a need to say what you’ve said here, then please do not answer my posts.
lol
Quote:
lol
Aha, that’s what I thought.
Current state of AI, AI is a very, very complex sample interpolator.
So "new" stuff generated with it is always based on a giant and very complex interpolation of real world stuff.
IMHO AI can be used in the demoscene, but since is something so different, new and disruptive, it should be clear when it was used.
We didn't had this problem in the past, saying the code was made by a compiler or handmade in assembly was not that relevant, or if a demo is made in a tool or just pure C code, but this time is different, the generated content uses all the samples and combine them with the "human" as a director.
So it something like a movie director that says "I need a more sci-fi look here", he didn't did the work, but the idea is from him.
In the triangle analogy it's like choosing a inside point position in the triangle, the 3 samples are not from him, but the "director" choose the xy position in the triangle.
Also in the triangle analogy the director would never be able to create a "black" point, that is something really new.
Current AI does not know how to do black.
So "new" stuff generated with it is always based on a giant and very complex interpolation of real world stuff.
IMHO AI can be used in the demoscene, but since is something so different, new and disruptive, it should be clear when it was used.
We didn't had this problem in the past, saying the code was made by a compiler or handmade in assembly was not that relevant, or if a demo is made in a tool or just pure C code, but this time is different, the generated content uses all the samples and combine them with the "human" as a director.
So it something like a movie director that says "I need a more sci-fi look here", he didn't did the work, but the idea is from him.
In the triangle analogy it's like choosing a inside point position in the triangle, the 3 samples are not from him, but the "director" choose the xy position in the triangle.
Also in the triangle analogy the director would never be able to create a "black" point, that is something really new.
Current AI does not know how to do black.
@PauloFalcao
And the only way new samples can enter the dataset is if a human makes them. Which would mean if humans go crazy with LLM prompt-to-image generatots and furtherly saturate the dataset from which the machine learns (the internet) with AI generated imagery, we get a closed loop. The machine feeds itself and after extreme averaging of results all we get is increasing level of noise. Right?
And the only way new samples can enter the dataset is if a human makes them. Which would mean if humans go crazy with LLM prompt-to-image generatots and furtherly saturate the dataset from which the machine learns (the internet) with AI generated imagery, we get a closed loop. The machine feeds itself and after extreme averaging of results all we get is increasing level of noise. Right?
Quote:
@PauloFalcao
The machine feeds itself and after extreme averaging of results all we get is increasing level of noise. Right?
True, but not noise, more like every thing becomes more like "average", just like we can now see that something is AI generated, the content will start to converge, to something more "mainstream" more like things everyone likes.
It's my opinion, that this will NOT change in a near future, a new sample generated by a human has (or can have...) something very different from a sample generated by a "sample interpolator" it has expressed in the sample the "needs" of the creator, what the creator like, what the creator cares, the life experience of the creator, and what is beautiful (if the creator wants to represent that) for the creator.
@PauloFalcao
What I’ve read happens when you pollute the LLMs learning dataset with the output of the machine itself to a certain degree is something they call MAD (Model Autophagy Disorder) - which causes the machine to progressively deteriorate.
What I’ve read happens when you pollute the LLMs learning dataset with the output of the machine itself to a certain degree is something they call MAD (Model Autophagy Disorder) - which causes the machine to progressively deteriorate.
Quote:
But c’mon are you so insecure that you can’t rate your own prod for us all?
Yes, I'm very insecure. But that's besides the point: I did not pose that it's objectively bad. That was you.
And I find the very idea of objectivity here so interesting that I'd love to hear your defense.
Quote:
It's my opinion, that this will NOT change in a near future, a new sample generated by a human has (or can have...) something very different from a sample generated by a "sample interpolator" it has expressed in the sample the "needs" of the creator, what the creator like, what the creator cares, the life experience of the creator, and what is beautiful (if the creator wants to represent that) for the creator.
I agree. Current AI technology has many limitations (especially when it comes to real understanding and creativity). I recommend two books to illustrate this: one is "Artificial Intelligence, A Guide to Human Thinking" by Melanie Mitchell. The other is "Natural General Intelligence" by Christopher Summerfield. These are two of the best books on AI I have read in recent years.
@4gentE like doing a photocopy of a photocopy of a photocopy... it starts to deteriorate, because on each iteration, or in this case with more AI generated content, the errors start to acumulate, true.
Quote:
I'd love to hear your defense.
Defense? Are you mad? Should I get a lawyer? Get a load of this folks!
No seriously, I retract everything I might have said about the prod. OK? Screw the prod. More interesting stuff is going on here.
Correction: I mean "Artificial Intelligence, A Guide for Thinking Humans".
(I used that stupid Google Translator to check the grammar of a sentence and it changed completely the title of the book, even if I put it inside quotes, thinking it wouldn't be changed... Well, this illustrates even better my point.)
(I used that stupid Google Translator to check the grammar of a sentence and it changed completely the title of the book, even if I put it inside quotes, thinking it wouldn't be changed... Well, this illustrates even better my point.)
Quote:
I don’t understand why are you so eager to take from people what they love
Is anything being taken away though? Just continue doing what you enjoy?
@4gentE:
I bet bifat does not care at all about the content of your opinion about his prod or any other prod. He is, however, very interested about your criteria for objectivity in art. Or at least that's my impression.
Quote:
Quote:
you released an objectively bad prod
Now it's getting interesting! :-)
Quote:
I retract everything I might have said about the prod. OK? Screw the prod.
I bet bifat does not care at all about the content of your opinion about his prod or any other prod. He is, however, very interested about your criteria for objectivity in art. Or at least that's my impression.
Quote:
Is anything being taken away though? Just continue doing what you enjoy?
Before
Quote:
I don’t understand why are you so eager to take from people what they love
This was written:
Quote:
I can see no evil in machines making work unnecessary. AI is entering the toolboxes of today's illustrators, and a future illustrator's job description will look differently from today's.
Insinuating this change in the whole concept of working in “art” field is good and should be welcomed, or that’s how I understood it.
Meaning, bit by bit we will set the needed speed of delivery and prefered aesthetic in such a manner that doing things the old way will cease to be feasible.
Quote:
He is, however, very interested about your criteria for objectivity in art.
I believe he doesn’t consider his prod to be art.
I agree with bifat. This new way (not a change, just another technique for the toolbox) is good and it should be welcome.
Why do you think the old way will cease to exist?
In a few years, new techniques will render current AI techniques obsolete.
Still, many people will continue using current AI techniques if they want.
And so on, and so on...
Why do you think the old way will cease to exist?
In a few years, new techniques will render current AI techniques obsolete.
Still, many people will continue using current AI techniques if they want.
And so on, and so on...
Quote:
I believe he doesn’t consider his prod to be art.
That is a very strange believe. Based on what evidence?
Dear ham and 4gentE: I don't know. I wouldn't totally rule it out, but it would be crazy presumptious to say so.
As a safe bet, I would say: Maybe some glimmers of art may shine through here and there.
But isn't that the case with basically every demoscene production? Including the "objectively" crappy ones!
The big problem with demos is their extremely multi-faceted nature, which spans different departments, and in doing so, usually a couple of persons - with totally different mindsets and ideas.
As a safe bet, I would say: Maybe some glimmers of art may shine through here and there.
But isn't that the case with basically every demoscene production? Including the "objectively" crappy ones!
The big problem with demos is their extremely multi-faceted nature, which spans different departments, and in doing so, usually a couple of persons - with totally different mindsets and ideas.
@ham
Let me get this straight: You think that we should remove pencils, pads and graphic tablets from the illustratots hands and make them write prompts?
The talent and craft they posess will be used to do minor corrections to what comes out of the machine. Like finger count up until these days. Until we perfect the machine so that no manual postproduction is needed. Then, just typing.
And you see this as natural. From stick in the sand to coal on paper to colors on canvas to airbrush on wood to Photoshop brush on digital canvas. To this. Typing text in a prompt.
Let me get this straight: You think that we should remove pencils, pads and graphic tablets from the illustratots hands and make them write prompts?
The talent and craft they posess will be used to do minor corrections to what comes out of the machine. Like finger count up until these days. Until we perfect the machine so that no manual postproduction is needed. Then, just typing.
And you see this as natural. From stick in the sand to coal on paper to colors on canvas to airbrush on wood to Photoshop brush on digital canvas. To this. Typing text in a prompt.
Quote:
Based on what evidence?
Based on what he just said.
You know, first 3D software was also prompt based. It was a long time to WYSIWYG. And now, let’s go back. Interesting.