Stuff like this just bugs me, though.
I like Frank a lot, but I just can’t groove to the suggestion that the output on the right has anything meaningful to say when compared to the artwork on the left. It’s not just that the AI version is meaning-neutral. It’s like it deliberately tried to ignore everything meaningful in the original, solely to produce the optical illusion of art.
So it’s horrible, and I loathe it, and I am deeply suspicious of any argument that says this is the future. People make art to tell stories, and the story didn’t make the transition from Rockwell to whatever video game made the other output. And all the manifestos and proclamations in the world will do nothing to get AI past that gap.
Okay. Shaking-fist-at-cloud time done.
https://x.com/nullJP/status/1890189489458590073/photo/1
I suspect there’s a line somewhere between art and memes… but I am well aware that I could be very wrong.
Mew
It’s hard for me to draw the line, but subtle manipulations of an initial work seems to be more insidious than a newly generated work based on prompts.
If, as in both of the examples of AI art (Moe’s and the one I linked) an original piece is given to the AI as a prompt.. “make me something like X art but with Y changes” … it’s going to get something like this.
It’s going to get worse….
Mew
So far, the reasoning ability for AI is remarkably _limited_ (to the degree it exists). Context has no real meaning. This is literally an ontological problem.
This shows up in many, many ways, but art is a great example of it. There is an attempt to pass by imitation, but without understanding the meaning of the pattern it is imitating.
There are areas where people’s impression of the results of AI is a Rorschach Ink Blot test. People see patterns and meaning in it, but there is no actually depth or plan there. A lot of jobs that depended on basic symbolic manipulation in regular patterns are going to be completely replaced by this.
There are other areas where the pattern matching is ok, and is likely to improve. Merging AI code analysis and recommendations with static code analysis to identify bad patterns and exploitable code is going to get better. (Note, this is at the IDE level for writing code, and not a recommendation to let AI write the code for you! This generation of the technology is not ready for reliably, IMHO.)
I remain skeptical that LLM are somehow creating reasoning systems. I know that trying to improving the reasoning of the AI systems is a critical goal for almost everyone working on this technology. Get improved reasoning and data modeling engines with a more sophisticated LLM on top of it, and you might have something that will really help in a lot of fields.
Is that around the corner? Or is that “about 10 years in the future” like fusion power has always been? I think you are a fool to bet everything on one or the other.
Eh. It is itself.
And that’s all that it is.
It’s a mildly interesting transformation of something good.
By itself, it’s meh. It’s the context we’re bringing into it that gives it any interest.
But it was human agency that injected that context on both ends.
If I required art to have *meaning*, I’d hate pop music.
Most of the time, art is just occupying an otherwise blank space with something more colorful to break up the monotony.
One thing is certain: AI art beats the hell out of Jackson Pollock.