The rhyming dictionary is a dangerous toy. Metrical lines are also easy to turn into statistics problems. But - note the absence of any real ideas, and the fallback to "people are special" instead of actually understanding and representing the argument about taste. It's a program that makes something that "feels" like a poem.
Just for laughs, I had it do one in the style of Charles Bukowski. Same sentimentality, plus whiskey and swearing.
I like that you went from taste to "gut feeling," because another thing a computer lacks is a brain/body relationship. How much of human consciousness and our gut feelings come from our body's physical interaction with the world? The concept of "muscle memory" is largely debunked -- a professional boxer doesn't throw rapid combinations under pressure because their arms and torso have "learned" the moves, but the mind and body have used their experiences and practice to allow the brain to direct action faster than consciousness can keep track of. So the body matters. The gut matters, as a gut, not just a metaphor.
100%, and another way in which AI is useful, but nothing like thought, or even reasoning. I took up the hungry judge problem in a different post - the way sentences are harsher before lunch than after - and also the Ramadan judge problem - sentences are more lenient during fasting, because we value being thought of as pious even more than we're moved by hunger. All of which is to say, so-called reason is shot through not just with body knowledge, but memory and emotion. And yet, much of the AI community seems to think that if they build something that builds logical rules around language, it will create a machine that reasons. I doubt it.
I've noticed the same issues when asking for poems from the LLMs. Asking for a poem in the style and tone of [poet name] helps, but again, it's clearly lacking the actual taste and selection those poets would take care in choosing the substance.
If you dig into the reasons for the banality, it's kind of interesting. LLMs are basically working through the collective wisdom, which is to say the average view, which is to say the cliche. Each individual view of a poet is interesting, but put them together and you get... Bukowski is booze and swearing; Eliot is austerity and distance; Dickinson is dashes and oblique statements. Whereas a real encounter with art is between the work and a particular individual, and that resonance can't aggregate.
OK, that poem is freaking me out a little. But you are right, it isn't memorable.
The rhyming dictionary is a dangerous toy. Metrical lines are also easy to turn into statistics problems. But - note the absence of any real ideas, and the fallback to "people are special" instead of actually understanding and representing the argument about taste. It's a program that makes something that "feels" like a poem.
Just for laughs, I had it do one in the style of Charles Bukowski. Same sentimentality, plus whiskey and swearing.
I like that you went from taste to "gut feeling," because another thing a computer lacks is a brain/body relationship. How much of human consciousness and our gut feelings come from our body's physical interaction with the world? The concept of "muscle memory" is largely debunked -- a professional boxer doesn't throw rapid combinations under pressure because their arms and torso have "learned" the moves, but the mind and body have used their experiences and practice to allow the brain to direct action faster than consciousness can keep track of. So the body matters. The gut matters, as a gut, not just a metaphor.
100%, and another way in which AI is useful, but nothing like thought, or even reasoning. I took up the hungry judge problem in a different post - the way sentences are harsher before lunch than after - and also the Ramadan judge problem - sentences are more lenient during fasting, because we value being thought of as pious even more than we're moved by hunger. All of which is to say, so-called reason is shot through not just with body knowledge, but memory and emotion. And yet, much of the AI community seems to think that if they build something that builds logical rules around language, it will create a machine that reasons. I doubt it.
I also immediately corrected "can't" to "cannot" in the last line. And what's with all the commas?
Loved this, and also want to see the Bukowski version. But also want to write it myself.
In order to do that you'd have to lack talent.
I've noticed the same issues when asking for poems from the LLMs. Asking for a poem in the style and tone of [poet name] helps, but again, it's clearly lacking the actual taste and selection those poets would take care in choosing the substance.
If you dig into the reasons for the banality, it's kind of interesting. LLMs are basically working through the collective wisdom, which is to say the average view, which is to say the cliche. Each individual view of a poet is interesting, but put them together and you get... Bukowski is booze and swearing; Eliot is austerity and distance; Dickinson is dashes and oblique statements. Whereas a real encounter with art is between the work and a particular individual, and that resonance can't aggregate.
This ties into the art question too. I've yet to see any image generated by AI that summoned a meaningful feeling.
Ever heard of Kolmar & Melamid? You're gonna love this: https://www.artsy.net/article/artsy-editorial-komar-melamid-americans-painting-thought-wanted
Please tune in later today for our next installment...