Writers, composers, musicians, and creators make deliberate choices. These choices carry their emotions and incite specific ones in us, mere mortals lucky enough to enjoy their creation. You said it best: "No matter how advanced the models become, they will always remain just that: models. Representations. Replicas. Forgeries. Frauds." Inference is linear algebra, not original thought. There is nothing underneath, just weights and biases.
The NYT quiz's ending note is not only insulting but stupid. Absolutely braindead. Referring to McCarthy's "As well ask men what they think of stone" as evidence of an "aversion to punctuation"??!! I guess someone at NYT thinks that the same meaning would be conveyed by "As well, ask men what they think of stone", even though the latter would be grammatically nonsensical and convey nothing close to the same meaning.
I like to write, so for me, using AI would defeat the entire purpose.
I’m also intensely turned off by finding out some writing is AI-assisted. Like in chess, we want to see the human voice. Nobody wants to watch bots play, even if they can outplay all humans.
For the love of the game!! Joy is always a good argument. Seth Rogan recently made a similar point at Cannes: “if you don’t want to go through the process, you shouldn’t be a writer. The idea of a tool that makes me write less is not appealing to me, because I like writing.” To your point about watching bots play, I wonder if the oral tradition will make a comeback 😂 Maybe people will memorize and regurgitate what their bot wrote for them, but 1) sincerity is more easily assessed face-to-face, and 2) it’s a lot more cumbersome, not to mention embarrassing.
The samples in the quiz are just too short to have a proper opinion I think. Other than the two examples you quoted in your article I didn't have a strong feeling about the passages either way. What I always notice with AI written content is the tendency to use words like decorations, string them together to sound nice but they're not saying much and there's no underlying meaning. I just feel like that mostly becomes apparent in longer texts though. Anyway, English isn't my native language so that might have also influenced me. (i chose 2/5 AI written passages and it bugged me)
Yeah, in the first case I was thinking the McCarthy text was an LLM trying to cloak some pseudo-profound nonsense in competent language, whereas the second text wasn't trying so hard. The author here faults it for the mismatch between the rapid description and the meaning intended, but that could also be an intentional revelation of character, understatement or resignation. It's not a beautiful passage on its own but all the analysis of the McCarthy text didn't wow me either. Either way it's a guy (or two) talking, and most people don't speak in beautiful sentences even in "literature." A longer passage would reveal much more.
I can't help but feel the poetry question was deliberately made difficult. I would have read both passages as parts of novels, and then Bishop reads much worse. The AI passage is meaningless, but since it is about death and I assumed there was more context after the quote, the meaninglessness had a point. Poems work very differently but I'm not sure the NYT knows the difference between prose and poetry.
Certainly, LLM 'writing' has its tics that make it obvious to a reasonably literate person, but even if it were indistinguishable from real (read: human) writing, it still wouldn't BE real writing; all the tech bros have done to produce it is feed a gigantic load of real writing into a probability engine. Without the genuine article to work from, the LLM can't produce anything at all.
Great article! I read this before taking the quiz, but I still got 50/50 (on the first 4). Because I was just trying to sus out the AI, I was not following the Times instructions to “Choose the passage you like best, regardless of how it may have been written.” But I wonder if on some level in taking tests like these it is unavoidable to favor or disfavor certain choices of words, or even the content of passages. Maybe the reason why I chose the AI passage over the Carl Sagan passage, for instance, was because I am not a fan of the word/concept of ‘spirituality’. Of course, I tried to consciously steel myself against such biases in my analysis (especially this one, since it was quite transparent to me), but there’s only so much that can do. If something doesn’t resonate with me, it’s going to be harder for me to connect with it and recognize it as human… right? Or am I just rationalizing my failure? Probably a bit both!!
All such thinking is fundamentally flawed. Unlike human writers, AI is in its infancy and progressing so rapidly the government stepped in to halt progress. LLMs are not even 5 years old. If allowed to develop another 5 years they surely would produce writing that is as good or better than all but a tiny handful of humans. When computers became capable of beating any human on earth at chess, the same sort of people as the author exclaimed ah, but they will never beat us at Go, a deeply human game that cannot be brute forced like chess. A decade later AI now beats all humans at Go and transformed the game by discovering vast new approaches to gameplay strategy that no human had thought of in 2,000 years. Examining a snapshot of current AI writing and drawing universal conclusions is a rudimentary intellectual error that Claude or Gemini wouldn’t make.
This is really well-written. The writing itself is evidence for the value of human thoughts. And the viewpoint is easy to follow and persuasive.
I’ve mostly landed in a spot where I don’t care if something is written by AI or a human so long as it’s good (through a wholly subjective lens). For many of the reasons you state or allude to, that’s far more likely to be fully generated by a human. I’m just not interested in the witch hunt.
Thank you for this incredible essay. I learned and laughed as I read it. I haven’t taken the quiz, but preferred the human written passages you put here precisely because the AI ones said nothing. You articulated beautifully why.
Lovely essay. I especially enjoyed you putting into specifics things I notice more a vagaries. I feel a draft in the room. You know it’s the shitty insulation around the window frame. Now I know it’s the shitty insulation too. And the is a gift. Thanks! Much appreciated.
Writers, composers, musicians, and creators make deliberate choices. These choices carry their emotions and incite specific ones in us, mere mortals lucky enough to enjoy their creation. You said it best: "No matter how advanced the models become, they will always remain just that: models. Representations. Replicas. Forgeries. Frauds." Inference is linear algebra, not original thought. There is nothing underneath, just weights and biases.
I read this piece twice already.
Thank you so much for reading so deeply ❤️ it’s the greatest gift you can give to this essay
The NYT quiz's ending note is not only insulting but stupid. Absolutely braindead. Referring to McCarthy's "As well ask men what they think of stone" as evidence of an "aversion to punctuation"??!! I guess someone at NYT thinks that the same meaning would be conveyed by "As well, ask men what they think of stone", even though the latter would be grammatically nonsensical and convey nothing close to the same meaning.
BookTok and its consequences have been a disaster not just to reading, but to literacy.
I like to write, so for me, using AI would defeat the entire purpose.
I’m also intensely turned off by finding out some writing is AI-assisted. Like in chess, we want to see the human voice. Nobody wants to watch bots play, even if they can outplay all humans.
For the love of the game!! Joy is always a good argument. Seth Rogan recently made a similar point at Cannes: “if you don’t want to go through the process, you shouldn’t be a writer. The idea of a tool that makes me write less is not appealing to me, because I like writing.” To your point about watching bots play, I wonder if the oral tradition will make a comeback 😂 Maybe people will memorize and regurgitate what their bot wrote for them, but 1) sincerity is more easily assessed face-to-face, and 2) it’s a lot more cumbersome, not to mention embarrassing.
The samples in the quiz are just too short to have a proper opinion I think. Other than the two examples you quoted in your article I didn't have a strong feeling about the passages either way. What I always notice with AI written content is the tendency to use words like decorations, string them together to sound nice but they're not saying much and there's no underlying meaning. I just feel like that mostly becomes apparent in longer texts though. Anyway, English isn't my native language so that might have also influenced me. (i chose 2/5 AI written passages and it bugged me)
Yeah, in the first case I was thinking the McCarthy text was an LLM trying to cloak some pseudo-profound nonsense in competent language, whereas the second text wasn't trying so hard. The author here faults it for the mismatch between the rapid description and the meaning intended, but that could also be an intentional revelation of character, understatement or resignation. It's not a beautiful passage on its own but all the analysis of the McCarthy text didn't wow me either. Either way it's a guy (or two) talking, and most people don't speak in beautiful sentences even in "literature." A longer passage would reveal much more.
After I read this article, I finally did the NY times quiz with “does it actually say something” as the benchmark, and what a filter it was.
I can't help but feel the poetry question was deliberately made difficult. I would have read both passages as parts of novels, and then Bishop reads much worse. The AI passage is meaningless, but since it is about death and I assumed there was more context after the quote, the meaninglessness had a point. Poems work very differently but I'm not sure the NYT knows the difference between prose and poetry.
Certainly, LLM 'writing' has its tics that make it obvious to a reasonably literate person, but even if it were indistinguishable from real (read: human) writing, it still wouldn't BE real writing; all the tech bros have done to produce it is feed a gigantic load of real writing into a probability engine. Without the genuine article to work from, the LLM can't produce anything at all.
Great article! I read this before taking the quiz, but I still got 50/50 (on the first 4). Because I was just trying to sus out the AI, I was not following the Times instructions to “Choose the passage you like best, regardless of how it may have been written.” But I wonder if on some level in taking tests like these it is unavoidable to favor or disfavor certain choices of words, or even the content of passages. Maybe the reason why I chose the AI passage over the Carl Sagan passage, for instance, was because I am not a fan of the word/concept of ‘spirituality’. Of course, I tried to consciously steel myself against such biases in my analysis (especially this one, since it was quite transparent to me), but there’s only so much that can do. If something doesn’t resonate with me, it’s going to be harder for me to connect with it and recognize it as human… right? Or am I just rationalizing my failure? Probably a bit both!!
The second just sounds like Hemingway.
I'm sure that most readers in the world do deep semantic and structural analysis of texts. Wouldn't be lovely if all was so simple?
We are taking snapshots of AI development for an specific time - kind of a photo that soon will fade into memory.
I'm very sure that next year AI capabilities will be the same - an we can say "Humans win, I told you so."
All such thinking is fundamentally flawed. Unlike human writers, AI is in its infancy and progressing so rapidly the government stepped in to halt progress. LLMs are not even 5 years old. If allowed to develop another 5 years they surely would produce writing that is as good or better than all but a tiny handful of humans. When computers became capable of beating any human on earth at chess, the same sort of people as the author exclaimed ah, but they will never beat us at Go, a deeply human game that cannot be brute forced like chess. A decade later AI now beats all humans at Go and transformed the game by discovering vast new approaches to gameplay strategy that no human had thought of in 2,000 years. Examining a snapshot of current AI writing and drawing universal conclusions is a rudimentary intellectual error that Claude or Gemini wouldn’t make.
This is really well-written. The writing itself is evidence for the value of human thoughts. And the viewpoint is easy to follow and persuasive.
I’ve mostly landed in a spot where I don’t care if something is written by AI or a human so long as it’s good (through a wholly subjective lens). For many of the reasons you state or allude to, that’s far more likely to be fully generated by a human. I’m just not interested in the witch hunt.
https://substack.com/@jacekhoffman/note/c-278347843
I instantly knew it was claude
Thank you for this incredible essay. I learned and laughed as I read it. I haven’t taken the quiz, but preferred the human written passages you put here precisely because the AI ones said nothing. You articulated beautifully why.
Lovely essay. I especially enjoyed you putting into specifics things I notice more a vagaries. I feel a draft in the room. You know it’s the shitty insulation around the window frame. Now I know it’s the shitty insulation too. And the is a gift. Thanks! Much appreciated.