Any of us can be fooled. I'm interested to know how it feels to fool. What joy or sense of gratification is there in submitting an AI generated poem and having it accepted? Conversely, the letdown in having been accepted for a poem one didn't write? In a way, it's like a rejection, but for something one didn't submit.
Yes! What's the point of having credit for work you didn't do? Doesn't it undermine your own sense of self and self-worth? Don't you have to work hard just to maintain the false front? Kinda reminds me of when I was raising my kids, not just the era of social media, but also the tendency to want to brag about this or that so the other kids will "like" you
Yeah, the writing would seem to be on the wall for us. Since GPT-4.5, at least, large models have been able to pass a standard Turing test: https://www.pnas.org/doi/10.1073/pnas.2524472123. While I’m sure that experienced editors could do better than those psych-study volunteers—23% pass rate for *ELIZA*, which is decades-old technology—AI poems do seem to be getting through. I won’t name names, having better things to do with my time than fight a libel lawsuit, and I won’t describe my methods, so that people who submit to Variant won’t know how I look (and please note that I am not such a rube about this stuff as to take Pangram’s opinion as gospel); but when a friend mentioned to me that a poem in a highly selective journal (one of Duotrope’s 25 Most Challenging) struck him as being written by AI, I looked up two other pieces by the same poet, in less selective but good journals, and reached the conclusion that all three did seem much more likely than not to have been written primarily by AI.
It’s depressing, and I don’t know what to do about it. Laugh, I guess, when I ask Fable to imitate me and the resulting text sounds like John Milton with a head injury and is rated by Pangram as 100% human.
Pangram doesn't seem quite there yet. Just saw early celebration that integration of Pangram tech into Substack will deter AI slop in this space anddd TBD.
I have wondered, too, how accurate Pangram is. I know it's favored as an AI detector, but.... And while people like me can at least avoid using AI and, yes, well, maybe laugh, we should also have empathy for editors (unlike some of the comments on the recent AI literary scandals). Since we have taught machines to imitate humans, and can train LLMs on the very best writing, what are editors to do? Ask the writers whose they want to publish for some kind of proof? What would that proof look like-- copies of various drafts (which could also be faked, right?) ? I hope that the impulse to create motivates most writers/submitters, but I am afraid that in a world where plagiarism (I count AI use as a form of plagiarism) is now so quick and easy, ambition may win out
That’s the thing with poetry. There’s no physical artifact, and the text is easy to memorize. Realistically, I think enforcement will have to be after publication :(.
I'll be honest, it makes me glad that poetry work is not my main work anymore. It's exhausting enough working in higher ed, but at least I understand why someone would consider AI when completing an assignment, for various reasons. I use it for coding, myself. But why prompt a poem when you can just...write one? Why gather accolades and publications? Is it a gotcha, like you described? Or is it intentionally short cut? What does a poem prompt even look like?
Exactly! Like, what's it for? What's the goal? Is it disruption? Is it cheating? Is it AI submitting poems on its own behalf? Did they make other AI writing friends? Can I join the workshop?
I like to think AI poems lack heart and emotional realness...so much that we could sense/smell/taste their fake-ness. That is what I like to think. AI creeps me out.
Exactly, we wish to believe this. Unfortunately... if someone uses halfway decent prompts and goes through a few iterations it is easy to make something out of nothing that mimics verisimilitude ... It's unsettling and uncanny valley but here we are
If you like works by AI and published them already you may be losing your vision of the journal.If we don't want to be a Cyborg why not have poets identify their work as not being AI generated before accepting them for the journal.
Not purposefully, Jamal. My point is that I likely unknowingly have already published work where AI was used. Work that felt a little suspicious. But it's hard to tell because common AI tropes are common for a reason— LLMs are built on human writing. AI imitates us and then iterates in ways that can differ because we don't retrieve information the same way. But it's really easy to cobble something together with AI collaboration. I'm not going to share techniques for fear of giving people ideas but it's alarmingly simple.
Any of us can be fooled. I'm interested to know how it feels to fool. What joy or sense of gratification is there in submitting an AI generated poem and having it accepted? Conversely, the letdown in having been accepted for a poem one didn't write? In a way, it's like a rejection, but for something one didn't submit.
Yes! What's the point of having credit for work you didn't do? Doesn't it undermine your own sense of self and self-worth? Don't you have to work hard just to maintain the false front? Kinda reminds me of when I was raising my kids, not just the era of social media, but also the tendency to want to brag about this or that so the other kids will "like" you
Yeah, the writing would seem to be on the wall for us. Since GPT-4.5, at least, large models have been able to pass a standard Turing test: https://www.pnas.org/doi/10.1073/pnas.2524472123. While I’m sure that experienced editors could do better than those psych-study volunteers—23% pass rate for *ELIZA*, which is decades-old technology—AI poems do seem to be getting through. I won’t name names, having better things to do with my time than fight a libel lawsuit, and I won’t describe my methods, so that people who submit to Variant won’t know how I look (and please note that I am not such a rube about this stuff as to take Pangram’s opinion as gospel); but when a friend mentioned to me that a poem in a highly selective journal (one of Duotrope’s 25 Most Challenging) struck him as being written by AI, I looked up two other pieces by the same poet, in less selective but good journals, and reached the conclusion that all three did seem much more likely than not to have been written primarily by AI.
It’s depressing, and I don’t know what to do about it. Laugh, I guess, when I ask Fable to imitate me and the resulting text sounds like John Milton with a head injury and is rated by Pangram as 100% human.
Oof.
Thanks for weighing in, David.
Pangram doesn't seem quite there yet. Just saw early celebration that integration of Pangram tech into Substack will deter AI slop in this space anddd TBD.
I have wondered, too, how accurate Pangram is. I know it's favored as an AI detector, but.... And while people like me can at least avoid using AI and, yes, well, maybe laugh, we should also have empathy for editors (unlike some of the comments on the recent AI literary scandals). Since we have taught machines to imitate humans, and can train LLMs on the very best writing, what are editors to do? Ask the writers whose they want to publish for some kind of proof? What would that proof look like-- copies of various drafts (which could also be faked, right?) ? I hope that the impulse to create motivates most writers/submitters, but I am afraid that in a world where plagiarism (I count AI use as a form of plagiarism) is now so quick and easy, ambition may win out
That’s the thing with poetry. There’s no physical artifact, and the text is easy to memorize. Realistically, I think enforcement will have to be after publication :(.
I'll be honest, it makes me glad that poetry work is not my main work anymore. It's exhausting enough working in higher ed, but at least I understand why someone would consider AI when completing an assignment, for various reasons. I use it for coding, myself. But why prompt a poem when you can just...write one? Why gather accolades and publications? Is it a gotcha, like you described? Or is it intentionally short cut? What does a poem prompt even look like?
"The Why" is certainly one of the questions. Especially when there is no $$ in poetry (relatively speaking).
Exactly! Like, what's it for? What's the goal? Is it disruption? Is it cheating? Is it AI submitting poems on its own behalf? Did they make other AI writing friends? Can I join the workshop?
Ha, indeed, so many questions!
I like to think AI poems lack heart and emotional realness...so much that we could sense/smell/taste their fake-ness. That is what I like to think. AI creeps me out.
Exactly, we wish to believe this. Unfortunately... if someone uses halfway decent prompts and goes through a few iterations it is easy to make something out of nothing that mimics verisimilitude ... It's unsettling and uncanny valley but here we are
Yes. Agree.
If you like works by AI and published them already you may be losing your vision of the journal.If we don't want to be a Cyborg why not have poets identify their work as not being AI generated before accepting them for the journal.
Not purposefully, Jamal. My point is that I likely unknowingly have already published work where AI was used. Work that felt a little suspicious. But it's hard to tell because common AI tropes are common for a reason— LLMs are built on human writing. AI imitates us and then iterates in ways that can differ because we don't retrieve information the same way. But it's really easy to cobble something together with AI collaboration. I'm not going to share techniques for fear of giving people ideas but it's alarmingly simple.