·
How can generative artists distinguish hand-coded art in the age of AI?
PP
Piter Pasma@piterpasmaabout 12 hours ago
This question was posted by Art Blocks on Instagram:
And I wrote a rather long comment below that post to answer the question, and I figured I should share it, so it doesn’t get lost.
How can generative artists distinguish hand-coded art in the age of AI?
It's always been an important reason why I've gone so hard on size coding. Sure, gas prices for on-chain art were also a nice excuse, but the real reason lies much deeper. It's a proof of hand crafting, artisanry, knowledge and understanding.
I didn't do that because I foresaw the rise of LLM-coding, but I did do it because I knew it was important. I believe that finding the shortest description of a certain algorithm or series of outputs proves/signifies a fundamentally deep understanding of the exact thing you're trying to generate. If I give you a list of 500 000 random digits, you'll struggle to write a succinct program to produce that exact sequence. But if I were to supply you the knowledge that this list of random-seeming digits are actually the digits of Pi if you skip the first million, writing a very short program for that would become much easier. The ability to write that program, becomes the proof of that knowledge. And don't understand me wrong I'm not saying LLMs can't size optimize, while the current crop is not really made for it, if it was important enough they totally could.
But it's not really the point. The best way to get a program small is to leave out the unnecessary and unimportant. This is something the LLM can't really do, you need to prompt it before it knows what is important about the task, and it can only infer that from what it's been trained on. And if your art is in any way original, what it's been trained on, might not be of much help, there ...
