Five Novel Openings: Creativity

One of these is extremely appealing to me. I bet you’ll be able to point to it right off. One of the others is generated. I did my very best to prompt ChatGPT to give me something that looked reasonably plausible. Below each entry is a moment to pause for thought. I’d insert a poll below each entry, but that doesn’t appear to be functional right at the moment, so just a pause for consideration.

***

1)

There were four of us.

Grandfather was an old fox, of perhaps eight or nine years. Gray ran along his narrow jaw and in a broad streak from his black nose to between his black-tipped ears; it frosted his pelt so that he seemed almost outlined in gray light. His joints stiffened on cold wet days, and he liked to doze in the spring sunlight when he could. He was missing a toe on one of his front paws. When I was little and first realized he didn’t have the same toes I had, I asked him why and he told me a tanuki-badger bit it off, but I think he was teasing. He was like that.

Mother was simple, even for a fox. My brother and I watched her sometimes catch and lose a mouse a half-dozen times before she remembered to bite it while she still had her paws on it. We were amazed sometimes that she survived long enough to bear us.

Fortunately, the place where we lived was thick with mice and chipmunks and other small prey. The grasses around our home were too long and dense for hawks, and the few humans who lived nearby chased off anything larger. Our only competition was a family of cats led by a black-and-white spotted female. They lived in a deserted outbuilding near the people, but they hunted in our range, and ignored us rigidly. The cats chased and lost mice, too. I think this was intentional for them, but who can understand cats? Even as a woman, I have never understood them.

***

Is this generated, or real human-written text? AND, how sure are you, on a scale of one to ten, with ten being positive it’s human-written.

I do, of course, know the answer. But I’m trying hard to pretend that I don’t so I can ask myself, if I didn’t know, what would I think? And what I THINK is — This is human-written, 10 on the surety.

I think what makes me feel this way is the creativity AND the style — both. This therefore abruptly becomes, in some measure, a post about creativity. As we all know (Do we know this?), ChatGPT is not great at doing anything other than lining up cliche after cliche. I don’t think that’s changed in ChatGPT-5 because how could it? If you say, “Let word follow word according to a frequency analysis,” then how can you get anything but regression toward cliches, even in principle?

Either way, I just think the above entry is fantastic. Tons of great details that ChatGPT is just not likely to come up with, strung together in a coherent style that ChatGPT is not likely to use. AND ON TOP OF THAT, a meaningful line. I mean the last line. “Even as a woman, I have never understood them.” This is the author inviting the reader into a joke or a common experience — ah, cats, so mysterious! — and also this is obviously foreshadowing. I don’t think this level of skill with setup and foreshadowing is something ChatGPT can manage. (Am I right about that? What do you think? Are you aware of any counterexample?) Because ChatGPT can’t plan ahead and can’t assess coherence — it can’t plan or assess anything at all, ever — I don’t believe it or any other text generator can even in principle insert a line like this into the opening of a novel and then have that line pay off later. I have not read a lot of putatively good fiction generated by ChatGPT-5, though.

Regardless, overall, it seems inconceivable to me that the above entry could have been generated. I’d bet a hundred dollars without hesitation and a thousand dollars without much hesitation. I think I would be REALLY sure even if I did not actually know.

***

2)

Curran watched the man whose life he required settle onto one of the faux leather couches scattered around the station’s reception module. The monitors showed him Amory Dane, spruce, tall, and fair. Dane made the perfect picture of someone prepared to wait patiently for an appointment. He was a radically different creature from the furtive Freers in the corner dickering over the delivery price for the wafer case that sat on the floor between them, or the gaggle of haggard mechanics who had put in one shift to many at the bar.

Curran wondered idly what they would do if he spoke up and announced what he was. Would they laugh, thinking it was a crazy engineer’s joke? Would they scramble for the wall and try to get at the computer system? Or would they just start running for the hatches?

He ran through each of the scenarios and decided that any would be amusing, but that the risk of being recorded on a hard medium was not worth it.

From his position of safety, Curran calmly overrode the inspection commands for the modules automatic systems. Then he ordered the hatches to cycle shut One of the mechanics, more sober than the others, jerked his head up as he heard the hatch seal.

Before anyone could make another move, Curran sent a single command to each of the three explosive charges his talent had laid against the module’s hull.

***

Is this generated, or real human-written text? AND, how sure are you, on a scale of one to ten, with ten being positive it’s human-written.

I’m trying to look at this as though I’d never seen it before. I don’t like it much, but I *think* I would come down on Human-written, about a 6 on the confidence.

To me, this seems kind of generic. This is partly because of the style, which seems bland and kind of cliched. What makes me think “cliched”? I think it’s the … adverbs! Wondered idly, calmly overrode. Even though I’m constantly declaring that adverbs are peachy and people should quit hating on adverbs!, I still sort of feel that way, and I wonder if that’s partly because I noticed extreme overuse of adverbs in fiction generated by ChatGPT-4, unless you prompted ChatGPT-4 toward a more literary style.

This also just sort of … feels … uninteresting to me. It’s action-up-front, situation-up-front, but … in a kind of uninteresting way?

I think if I came to this entry cold, I would not bet money one way or the other.

I should add, I’m pretty sure anybody could tell after the first CHAPTER. These very short excerpts exaggerate the uncertainty because there’s not time for ChatGPT to lose the thread of the story, which I expect it’s certain to do soon enough if you keep going.

***

3)

I always thought the VentureStar looked like a tombstone. When it was standing on end it was twice as tall as it was wide. It wasn’t very thick. It was round at the top. For a night launch it was illuminated by dozens of spotlights like an opening night in Hollywood It could have been the grave marker for a celebrity from some race of giant aliens. The stubby wings and tail seemed tacked on.

The VentureStar didn’t spend much time flying, which was just as well, because it flew about as well as your average skateboard. Sitting on the ground, it looked more like a building than an aircraft or a spaceship.

That’s okay. In about thirty seconds it would leave every airplane ever built in a wake of boiling smoke and fire.

“Manny, a Greyhound bus leaves Cocoa Beach every day for Tallahassee. Why don’t we go watch that some night? We could get a lot closer.”

That was my girlfriend, Kelly, trying to get my goat. Her point being that VStars left Canaveral once a day, too. Point taken.

***

Is this generated, or real human-written text? AND, how sure are you, on a scale of one to ten, with ten being positive it’s human-written.

I would say, This is human-written, 8 on the surety. The description seems unlikely to be generated. A spaceship that looks like a grave marker for a giant alien? Looked like it would fly about as well as a giant skateboard? It seems unlikely that ChatGPT would come up with that. But maybe it might? Not sure. That’s why 8 instead of more surety than that.

***

4)

The rain slicked the stones beneath my chair, whispering against the eaves like it meant to wear the house away. I kept my gaze on the garden gate, because looking at him would have been an invitation, and I’d sworn off invitations years ago.

“You’ve heard,” Mondrian said, stepping closer. His boots left no sound in the wet. “A new god has risen. He’s carving his name into the bones of the earth. We mean to stop him.”

I took a slow sip from my cup. “Then I wish you luck.”

“You’ve killed a god before.”

“That’s exactly why I won’t do it again.” I set the cup down, watching the rain thread silver lines between us. “The first one took three days to die. I’ve no desire to spend another lifetime listening to a second god scream.”

He didn’t blink. “The difference is, General, if you don’t, you won’t hear him scream at all. You’ll hear the whole world.”

***

Is this generated, or real human-written text? AND, how sure are you, on a scale of one to ten, with ten being positive it’s human-written.

This is the generated opening, 9 on the surety. Why do I feel that way? Because at this point I’ve read enough generated text that this seems to scream GENERATED in a loud voice. That’s what I think. The first sentence does it. I’m being very heavily influenced this post about where AI goes wrong when it tries to generate literary fiction. It anthropomorphizes inanimate things, such as rain. Everything snarls, hisses, and whispers.

Also, by lack of continuity. The first one took three days to die. He doesn’t want to spend another lifetime listening to a different one die. Three days is not what I would call “a lifetime.” This could work with adequate setup, but without setup, it’s nonsensical. Ditto for “If you don’t.” If you don’t what? Kill the new god? Or spend a lifetime listening to the god scream? Which? Because this is written, it ought to be the second, but that makes no sense.

And this would be so easy to fix! Say three years instead of three days! Put if you don’t help us kill this god, rather than if you don’t.

Also, I’m starting to think that “taking a sip of coffee” might be a tell for “generated fiction in a literary style.”

Agree / disagree? Do any of the previous entries seem more likely to be fake than this one? Why?

***

5)

Doctor Walden looked glumly at the forms she had to fill ink. At the top it said RISK ASSESSMENT.

She’d designed the forms herself, in a burst of optimism. They would have fewer accidents if people just stopped to think. It was an unfortunate truth that in the Venn diagram of “qualified to teach magic” and “still alive,” the overlap consisted almost entirely of people who had always been much too sensible to accidentally get eaten by a demon. Walden’s colleagues – in particular, those who were her responsibility, the loosely grouped Faculty of Magic here at Chetwood School – possessed, as a body, an admirable and well-judged lack of imagination. In the three years since she took the post as Director of Magic, she had had someone in her office once a term to weep on her shoulder and say, But why would anyone ever –

Some of these people had been teaching for years, and yet they still managed to be surprised by how bloody stupid the average teenager could be, given a group of friends to impress and a fifteen-second video about major invocation that they found on the internet somewhere.

***

Is this generated, or real human-written text? AND, how sure are you, on a scale of one to ten, with ten being positive it’s human-written.

Human-written, 10 on the surety. I might not be QUITE as sure about this one as about the first one, but I’m sure. Why? Because of the smooth, clever writing. That Venn diagram! That line about the fifteen-second video! I wonder if it’s possible for ChatGPT-5 to write with this kind of humor. I vote no, but I haven’t tried to get it to write with humor, either.

Also, the adverbs seem better to me here. More creative! More humorous! That’s what I mean by better. Looked glumly is funny and builds character in a way that, to me, the adverbs in #2 did not seem to do. Throwing “accidentally” in there is also funny.

Below, the non-big reveal because I feel like everyone’s going to vote for #4 as the fake.

***

1) The Fox Woman by Kij Johnson. This one has been on my physical TBR shelves for a long time, and this opening is so delightful that I have put it on the coffee table so it will be in front of me and hopefully I will read it.

2) Fool’s War by Sarah Zettel. This is a NYT notable book. Well, the beginning doesn’t impress me at all. I’m willing to believe that it gets better.

3) Red Thunder by John Varley. I’ve liked a lot of books and stories by Varley; near-future SF is a hard sell for me; I’m willing to bet that I will probably like this book if I ever get around to reading it.

4) ChatGPT-5

5) The Incandescent by Emily Tesh. Here’s what it says at Amazon:

Naomi Novik’s Scholomance series meets Plain Bad Heroines in this sapphic dark academia fantasy by instant national and international bestselling author Emily Tesh, winner of the Hugo and World Fantasy Awards.

That’s promising — I mean the Scholomance comparison. A commenter here recommended this to me, so I do look forward to reading it … someday … possibly even very soon, since I did decide to take off the rest of August to read books!

Fundamentally, I think smooth, fun creativity and actual humor are not possible for text generators, even though these things are practically effortless for a gifted writer. I might try asking ChatGPT-5 to write something in the style of Terry Pratchett and see what it does. I don’t think it can possibly do that, but rather than saying “No way!” and stopping, I should probably actually try it.

Please Feel Free to Share:

The post Five Novel Openings: Creativity appeared first on Rachel Neumeier.