Recently my friend Himadri got ChatGPT to “improve” his sentences (as a joke), which turned out to be shite, so I tested ChatGPT with my blog post “What makes a good screen adaptation?”.
This is the comment from ChatGPT:
“Your piece already has strong ideas and a clear critical voice. The main improvements needed are concision, rhythm, and sentence structure—many of the original sentences are overly long and repetitive. I’ve rewritten it to keep your tone and arguments while making the prose sharper, more fluid, and more polished.”
Hello? The repetition is the point!
Here is one paragraph from my blog post:
“The 2022 Persuasion also seems to be an odd thing that is neither approach: perhaps I shouldn’t comment as I haven’t seen the whole film, but from what I have seen, it is neither a faithful adaptation, depicting the Regency era, nor an independent film with the story moved to the modern era; instead, Carrie Cracknell has characters of different skin colours wearing Regency costumes but speaking modern slang, and changes the character of Anne Elliot beyond recognition (it is perhaps aimed at the audience of Bridgerton).”
ChatGPT’s version:
“The 2022 adaptation of Persuasion seems caught in a similarly awkward middle ground. I should admit that I have not seen the entire film, but from what I have seen, it appears undecided about what it wants to be. It is neither a faithful Regency adaptation nor a complete modern reimagining. Instead, Carrie Cracknell presents characters in Regency dress speaking contemporary slang, while Anne Elliot is altered almost beyond recognition. The film perhaps aims for the audience of Bridgerton, but the result feels conceptually muddled.”
The parenthesis is part of my style! I don’t even think I have a particularly strong authorial voice, but the new version is bland.
Here’s my final paragraph:
“And this is something lots of people don’t seem to understand: whenever someone criticises a film adaptation for misrepresenting or betraying the text, some people just say fidelity is unimportant and the film is its own work of art, but most of the time it isn’t—most of the time it doesn’t have enough strengths and originality, most of the time it doesn’t transcend its source material—all we’ve got is just a poor film that doesn’t quite transfer a great work of art onto the screen.”
This is ChatGPT’s version:
“And this is what many people fail to understand when discussions about adaptations arise. Whenever someone criticises a film for betraying or misrepresenting its source, others immediately respond that fidelity does not matter because a film should stand as its own work of art. But most adaptations do not succeed as independent works either. Most lack the originality, vision, or artistic strength necessary to transcend their source material. Too often, what remains is simply an inferior film that fails to bring a great work of art convincingly to the screen.”
Where is the individual voice? The repetition is the point. The repetition is part of my style.
I also fed ChatGPT my blog post “Jane Austen’s 250”.
“In her four masterpieces—Pride and Prejudice, Mansfield Park, Emma, Persuasion—Jane Austen explores different kinds of love, different aspects, different angles. She adopts different tones. She uses different techniques. Pride and Prejudice for example is light, bright, and sparkling, with lots of dialogue. Mansfield Park is sombre, and she uses more layers and metaphors. Emma is where she masters the free indirect speech, blending the voice of the third-person narrator with the voice of the protagonist, colouring your perspective of the scene. Persuasion is autumnal, her warmest and most romantic novel. I say Austen is narrow—and in some sense, she is—but these four novels are all quite different. She is wonderful.”
What ChatGPT spat out:
“In her four masterpieces—Pride and Prejudice, Mansfield Park, Emma, and Persuasion—Austen explores different forms and dimensions of love. Each novel has its own tone, techniques, and emotional texture. Pride and Prejudice is light, sparkling, and full of lively dialogue. Mansfield Park is sombre, layered, and rich in symbolism. In Emma, Austen perfects free indirect discourse, blending the narrator’s voice with Emma’s own and subtly shaping the reader’s perception of events. Persuasion feels autumnal: her warmest, most wistful, and perhaps most romantic novel.
I sometimes say Austen is narrow—and in one sense, perhaps she is. Yet these four novels are remarkably different from one another in mood, structure, and vision. She is extraordinary.”
Again, ChatGPT removes the repetition, removes the thing that makes me me.
Over the past few months, I’ve been seeing more and more tweets that reek of ChatGPT (or whatever AI that people are using). The same blandness. The same soullessness. AI junkies don’t seem to understand that delegating writing to AI not only produces writings stripped of any individuality but also turns their brains to mush—for writing is thinking—and makes them gradually adopt that blandness and sound like an automaton even when they sit down and write something themselves.
If anything, this experiment makes me want to spend more time and effort cultivating my own voice.
Very interesting experiment!
ReplyDeleteAs you say, removing the repetition of "most the time it isn't/doesn't/doesn't" in the first passage weakens your expression in addition to homogenizing your voice.
A couple of further points to note in its rewriting of the first passage. First, it removed "of different skin colors"; perhaps (to anthropomorphize) it felt that this was politically problematic. Second, it changed your unambiguous "modern slang" to the ambiguous "contemporary slang" (contemporary with us, or with the Regency?) Why it thought that your "misrepresenting or betraying" should be reversed is anyone's guess.
I do actually agree with ChatGPT that there is no loss in force and a small gain in clarity if you break the single long sentence with its five independent clauses into four or five sentences; but that's a matter of taste.
A couple of additional comments:
ReplyDeleteAs you say, your repetition in the last passage of the word "different" and of the sentence structures "Austen explores ..", "She adopts" "She uses..." and "Pride and Prejudice is..." "Mansfield Park is", "Emma is", "Persuasion is", give force to your expression; and rewriting it for the sake of "variety" doesn't add anything, though it's what the handbooks recommend. As an editor, ChatGPT has no business adding "wistful" to your description of Persuasion or adding "in mood, structure, and vision" in the second-to-last sentence. A lot of its changes -- "your perspective of the scene" to "the reader's perception of events"; "quite different" to "remarkably different"; "wonderful" to "extraordinary", "more layers and metaphors" to "layered and rich in symbolism" seem completely arbitrary. ChatGPT does not seem to have any idea of the first rule of copy-editing; Unless the author's text has a problem, or unless you have some really inspired improvement to suggest, leave the author's text alone!
By the way, many of us consider "Sense and Sensibility" to be a masterpiece; and I know a few who consider "Northanger Abbey" to be as good in its way as any of the others.
Welcome, Ernest. I don't think you've commented before.
ReplyDeleteA friend tried this with Grok, twice. The whole posts are too long for comments, so here are some excerpts:
"Not every transplant succeeds. Bride and Prejudice has some clever ideas in moving Pride and Prejudice to modern India, but it loses the novel’s sharper edge around marriage and social standing. The characters feel lighter, the tone more obvious."
I don't know why it removes my specific examples. The examples are to back up my arguments.
"Clueless is another quiet success. It’s very much a product of its time, yet it catches something of the generous spirit in Austen’s Emma that several more serious versions overlook."
That's an AI tell: "quiet".
"What I’ve noticed is that when people say a loose adaptation is “its own thing,” they sometimes give it too much credit. For it to stand on its own, it needs real vision and coherence. Otherwise it just feels like a diluted version of the original.
In the end, the better adaptations tend to do one of two things: they show a clear, thoughtful understanding of the book, or they use it as ground for something new that holds together in its own right. The rest leave you wondering why they were made at all."
Like ChatGPT, Grok flattens everything, removes my anaphora.
Grok also adds something I would never say:
"What about you? Which adaptations made you fall in love with a story all over again—or left you wishing they’d tried something different?"
But of course, this kind of questions I have seen a lot on book blogs.