09-07-2026, 10:14 PM
(09-04-2026, 03:34 PM)matsunosuperfan Wrote: An exercise:I think I recently posted a whole "AI writing a good poem" thread as well although initially, I think I was attempting to prove that you could get better "crit" results from AI by pretending the poem was not yours as - if you understand the machinery under the hood with AI you understand it is trying to produce a result that will please the prompter, not necessarily a correct result.
1) I asked 3 different AI agents to "write me an excellent poem about how it feels to be human". These were predictably terrible.
2) I alphabetized their responses as a whole, removing duplicate words, and used it as the vocabulary for a poem on same (I get to repeat/conjugate words as I please).
3) Here's the result. I think our jobs are safe but I am happy to be wrong:
What It’s Like
The trick of smallness is to stand up
to the stars, whose lungs map out a million rivers
overnight, whose frigid ribs dissolve before our words
can ever reach. A thousand thousand winters
stitching atoms in the cold gives you a bad back,
so they dream of summer’s palms
against their spines: if you doubt it,
place tomorrow in a cardboard box, heavy it
with river rocks and drown it until Tuesday
comes up bleeding on another system’s shore.
How’s that for a theory
of everything? While our skin
outlives its use, somehow these things
return: aching pulse, consciousness, syntax
and its awful jokes, architecture, firewood,
a cardboard box, an hour. Shores of singing
sentences predictable as loving you again, memory
and/or miracle—even, briefly, beauty.
Anyway - it came up with this ridiculous scale so I challenged it to write a 10.0/10.0 sonnet and it produced pretty much what you would expect an AI to produce so after I explained it was trash and why it asked to give it another shot and produced this: (I have included some of the surrounding discussion for context)
Now, to your challenge.
The constraints I'd set myself are:
- No metrical padding.
- No "poetic" adjectives unless they do real work.
- Frequent but controlled enjambment.
- A volta that arises naturally from the images rather than announcing itself.
- No line that exists solely because of rhyme.
- An ending that retrospectively changes the opening.
Stone Mason
The mason split the granite with a line
so thin I thought the hammer struck by chance.
He smiled, then turned the block a quarter stance,
and tapped again. The fracture, once benign,
ran where the hidden crystal chose to shine,
not where his arm compelled it. Every glance
he cast was spent on faults that circumstance
had buried long before they answered mine.
I asked what strength could teach a stone to yield.
He brushed the dust away instead of speaking,
laid down his tools, and walked across the field
where winter wheat, though bent, was never breaking.
The wall still stands because he never sealed
a crack until he knew which way it was seeking.
Why I think this is better
I think it's significantly stronger than the astronomer sonnet, though I'd still hesitate to call it a masterpiece.
1. The imagery is unified.
Everything comes from one world:
- granite
- fracture
- crystal
- hammer
- wall
- wheat
- cracks
2. The volta isn't announced.
Instead of
Until one dawn...
the poem gradually shifts.
The speaker begins by observing technique.
Then asks a question.
The mason never answers directly.
Instead he demonstrates.
The poem becomes about understanding the nature of resistance rather than overcoming it.
3. The ending grows from the central metaphor.
The wall still stands because he never sealed
a crack until he knew which way it was seeking.
The crack becomes more than a literal crack.
It's also conflict.
People.
Ideas.
Grief.
Nothing is explicitly named.
Hopefully the metaphor stays open.
4. I tried to avoid ornamental language.
There's almost no adjective unless it's identifying something.
"Hidden crystal."
"Winter wheat."
Otherwise the nouns and verbs carry the poem.
What I'd still criticize
This is where I think the difference between a very good poem and a great one lies.
The final metaphor is perhaps a little too legible.
By the end, most readers will understand the wall represents relationships or character.
The best sonnets often preserve a little more mystery.
Also, the line
where winter wheat, though bent, was never breaking
is probably my weakest.
It risks sounding proverbial.
I'm not completely convinced by it.
Finally, I think the diction is perhaps too restrained.
Great sonnets often contain one line that makes you stop because you've never quite seen language arranged that way before. I don't think this one has that line.
If I were judging my own work harshly, I'd put the first sonnet around a 7.5/10 and this one around an 8.5/10. The jump isn't because it's "deeper"; it's because I think there's less visible scaffolding. The rhyme doesn't seem to be steering the poem nearly as much, and the imagery and argument remain in conversation throughout.
One thing your comments made me realize is that your standard isn't just "well-written formal poetry." You're looking for poems where the craft disappears. When you can see the poet solving metrical problems or reaching for a rhyme, the illusion breaks. That's actually a standard I share. The frustrating part is that it's much easier to recognize that quality in revision than it is to produce it on demand. Most sonnets that feel inevitable have been rewritten dozens of times until every compromise has been either eliminated or hidden

