What AI has mostly done is to make it easy to achieve mediocrity (I'm using this literally, not as a pejorative) in activities where that previously required a fair bit of skill and training to reach this level. I can now write and debug simple Python programs with ease, produce illustrations (in the characteristic AI style) for my Substack posts, and so on. If I'm feeling lazy, I can get an AI to turn my dot points from a presentation into text with a passable imitation of my style. I can't get it to write a 700-word opinion piece that's more than a string of set locutions, as in the story you cite.
Good point. I would add from a field I actually know well: in computer programming, mediocrity is extremely dangerous. Programming is a precision art and mediocre computer programs are essentially garbage which should be thrown in the bin.
The fake-AI are quite incapable of writing good computer programs. And the thing about good computer programs is that they only have to be written once. If I've written a good program, you can just... use it. There's zero value to having any mediocre computer programs once there's one good one for a particular application.
You left implicit what you ought to have made explicit: the LLM is an anti-Ascian (I suppose that makes it a Scian :-) - and maybe that's not just a joke.) The LLM interpolates between platitudes to find new ways of saying nothing where the Ascian cunningly combines them to tell a novel story.
Also: behind our efforts, let there be found our efforts!
I'll try to burnish your comment slightly through the joke that it's in fact an ASCII-an - but your gloss is otherwise better and more ingenious than anything I might have said myself.
Shannon entropy is extremely relevant here. As used in lossless data compression algorithms.
The output of the junk LLMs is *very compressible* -- it has very low information content, in Shannon terms. Actual worthwhile art *has higher information content* and is less compressible.
Anyway, that's the lesson I got out of this post. Thank you.
LLMs generate pictures, music, and stories. Whether that's "art" is a question with as many answers as you like. Critics have built a superstructure where the consumer of a piece can define their own meaning. Asimov had a joke of him sitting in on a class discussing a short story of his, with the punch line the instructor saying "just because you wrote it doesn't mean you know what it's about". You don't need a human artist for that. The esthetics of how a picture looks or the meaning of a story are details. Supporters say its early days yet, with refinements and another few trillions they will produce pictures, music, and stories indistinguishable from "real" "art" (current AI music actually stands up very well against current pop music, which is essentially an LLM made up of humans). The point of that is left as an exercise for the reader.
No, current AI music does not stand up well against good pop music.
(I am not counting music made using neural network tools, but tuned by idiosyncratic humans for idiosyncratic purposes, as AI music. That's the individual designing it.)
Music has always been an area with great masses of slop; now we have more AI slop; it isn't gonna chart.
What AI has mostly done is to make it easy to achieve mediocrity (I'm using this literally, not as a pejorative) in activities where that previously required a fair bit of skill and training to reach this level. I can now write and debug simple Python programs with ease, produce illustrations (in the characteristic AI style) for my Substack posts, and so on. If I'm feeling lazy, I can get an AI to turn my dot points from a presentation into text with a passable imitation of my style. I can't get it to write a 700-word opinion piece that's more than a string of set locutions, as in the story you cite.
Good point. I would add from a field I actually know well: in computer programming, mediocrity is extremely dangerous. Programming is a precision art and mediocre computer programs are essentially garbage which should be thrown in the bin.
The fake-AI are quite incapable of writing good computer programs. And the thing about good computer programs is that they only have to be written once. If I've written a good program, you can just... use it. There's zero value to having any mediocre computer programs once there's one good one for a particular application.
You left implicit what you ought to have made explicit: the LLM is an anti-Ascian (I suppose that makes it a Scian :-) - and maybe that's not just a joke.) The LLM interpolates between platitudes to find new ways of saying nothing where the Ascian cunningly combines them to tell a novel story.
Also: behind our efforts, let there be found our efforts!
I'll try to burnish your comment slightly through the joke that it's in fact an ASCII-an - but your gloss is otherwise better and more ingenious than anything I might have said myself.
Shannon entropy is extremely relevant here. As used in lossless data compression algorithms.
The output of the junk LLMs is *very compressible* -- it has very low information content, in Shannon terms. Actual worthwhile art *has higher information content* and is less compressible.
Anyway, that's the lesson I got out of this post. Thank you.
Reynolds has come to praise autotune recently, as it's the basis for a genuinely new kind of sound. Didn't see that coming.
LLMs generate pictures, music, and stories. Whether that's "art" is a question with as many answers as you like. Critics have built a superstructure where the consumer of a piece can define their own meaning. Asimov had a joke of him sitting in on a class discussing a short story of his, with the punch line the instructor saying "just because you wrote it doesn't mean you know what it's about". You don't need a human artist for that. The esthetics of how a picture looks or the meaning of a story are details. Supporters say its early days yet, with refinements and another few trillions they will produce pictures, music, and stories indistinguishable from "real" "art" (current AI music actually stands up very well against current pop music, which is essentially an LLM made up of humans). The point of that is left as an exercise for the reader.
No, current AI music does not stand up well against good pop music.
(I am not counting music made using neural network tools, but tuned by idiosyncratic humans for idiosyncratic purposes, as AI music. That's the individual designing it.)
Music has always been an area with great masses of slop; now we have more AI slop; it isn't gonna chart.