"But because they hate AI, a lot of these friends refuse to use it... This is a good way to end up badly mistaken about AI."
My goodness, this divergence is so real in my life. I'm a heavy user, paying for multiple AI services and getting great value from them. But I sometimes bump into people who are very dismissive and negative, and it's hard to even have a conversation with them about AI. They clearly have no idea what they are talking about, and they criticize how it works in ways that clearly demonstrate they are not trying to use it in ways that would help them.
So strange. It will be fun to watch this sort itself out in the years ahead. They are in for a heck of a surprise, I expect.
I think the Commonwealth Foundation isn't particularly interested in bring intellectually honest or rigorous, they just want to hand out what are effectively DEI prizes to briefly assuage the insatiable White guilt of their donors. Having to acknowledge and contend with AI trickery is irrelevant or even counterproductive to that purpose.
20 years ago I was an english major. I took creative writing classes. I thought at the time that it was bullshit dripping with pretension. AI is an excellent bullshit generator and thus an excellent creative writer. for the minority of folks who even care, I wouldn’t be shocked if they prefer the creative writing from AI.
Note that even if pangram is great at detecting AI generated writing now there’s no guarantee that it will be in the near future. Realistically if you wanted to enforce human authenticity, the only reliable anti cheat detection is to require human beings to attend an in person proctored test environment where they have a blue book. anyone who leaves to even go to the bathroom is automatic disqualification.
But i doubt anyone will care enough to even attempt that. But even then what many people will do is have the AI generate the text before the test and memorize that.
Yeah, what this article shows is that in May 2026, it is still true that if you prefer AI writing, either your taste is bad or the human competition is bad. Who knows how long that will last. But for now, admitting to accepting AI winners for a literary prize still means admitting your prize sucks.
With all of the descriptions of AI writing quirks that are out there (they are voluminous, and they generally point to the same quirks), it does make me wonder why these same quirks keep showing up in AI writing. Is it some feature of transformer-based LLMs that they're going to keep producing uncanny valley prose? Or is it merely a feature of the RL training that LLMs get that leads to these quirks?
I think it's pretty clear that the profit motive of companies producing frontier models are not likely to lead them to make AIs that are great at winning literary prizes. There's already very little money in that world, and what little pockets do have money require having a living human being embedded in contexts with other human beings (the LLM can't yet go to a cocktail party, or give a reading in front of a live audience). So it wouldn't surprise me if we just don't know yet if LLMs can produce literary artifacts that can make it through Pangram-style screens. Why would Anthropic or OpenAI or the Gemini team spend time on that now?
There's also no "correct" target to hit for what the next great literary work is going to look like. LLMs that can loop over and over with different instances of itself can catch errors in math and code logic much easier, because there's often a correct answer in what works. It seems possible that building a mechanism to iterate on prose to try to get to a superior work is just not a problem that an LLM (*as currently architected*) can "solve" in the same way.
"But because they hate AI, a lot of these friends refuse to use it... This is a good way to end up badly mistaken about AI."
My goodness, this divergence is so real in my life. I'm a heavy user, paying for multiple AI services and getting great value from them. But I sometimes bump into people who are very dismissive and negative, and it's hard to even have a conversation with them about AI. They clearly have no idea what they are talking about, and they criticize how it works in ways that clearly demonstrate they are not trying to use it in ways that would help them.
So strange. It will be fun to watch this sort itself out in the years ahead. They are in for a heck of a surprise, I expect.
addendum - some of the actual things I have used AI for in the past month:
- college research for my kids
- debugging problems with my car not starting
- investment strategies
- technical design questions (coding related)
- how does lucid dreaming work
- brainstorming on political policy impacts (past and future)
- kind of as a therapist, or at least exploring ideas related to therapy
- discussing workouts and form queues, e.g. "I did this exercise but I felt it more HERE instead of THERE, an I doing it wrong?"
- prepper research
- instructions for how to install local LLMs on my computer
“You’re right — I was hand waving, those metaphors were not load-bearing”
I think the Commonwealth Foundation isn't particularly interested in bring intellectually honest or rigorous, they just want to hand out what are effectively DEI prizes to briefly assuage the insatiable White guilt of their donors. Having to acknowledge and contend with AI trickery is irrelevant or even counterproductive to that purpose.
20 years ago I was an english major. I took creative writing classes. I thought at the time that it was bullshit dripping with pretension. AI is an excellent bullshit generator and thus an excellent creative writer. for the minority of folks who even care, I wouldn’t be shocked if they prefer the creative writing from AI.
Note that even if pangram is great at detecting AI generated writing now there’s no guarantee that it will be in the near future. Realistically if you wanted to enforce human authenticity, the only reliable anti cheat detection is to require human beings to attend an in person proctored test environment where they have a blue book. anyone who leaves to even go to the bathroom is automatic disqualification.
But i doubt anyone will care enough to even attempt that. But even then what many people will do is have the AI generate the text before the test and memorize that.
Those phrases seem silly to me. Perhaps another issue is that the contest runners chose to cheat by using AI in grading entries...
Yeah, what this article shows is that in May 2026, it is still true that if you prefer AI writing, either your taste is bad or the human competition is bad. Who knows how long that will last. But for now, admitting to accepting AI winners for a literary prize still means admitting your prize sucks.
"
One would hope that the poorly written metaphors would, on their own, have been sufficient to disqualify some of these submissions.
"
THIS
With all of the descriptions of AI writing quirks that are out there (they are voluminous, and they generally point to the same quirks), it does make me wonder why these same quirks keep showing up in AI writing. Is it some feature of transformer-based LLMs that they're going to keep producing uncanny valley prose? Or is it merely a feature of the RL training that LLMs get that leads to these quirks?
I think it's pretty clear that the profit motive of companies producing frontier models are not likely to lead them to make AIs that are great at winning literary prizes. There's already very little money in that world, and what little pockets do have money require having a living human being embedded in contexts with other human beings (the LLM can't yet go to a cocktail party, or give a reading in front of a live audience). So it wouldn't surprise me if we just don't know yet if LLMs can produce literary artifacts that can make it through Pangram-style screens. Why would Anthropic or OpenAI or the Gemini team spend time on that now?
There's also no "correct" target to hit for what the next great literary work is going to look like. LLMs that can loop over and over with different instances of itself can catch errors in math and code logic much easier, because there's often a correct answer in what works. It seems possible that building a mechanism to iterate on prose to try to get to a superior work is just not a problem that an LLM (*as currently architected*) can "solve" in the same way.