In a study published in the journal Judgment and Decision Making, 1,682 adults were asked to read one of six short stories, three of which were written by humans and three by ChatGPT. Each AI story had a similar theme to one of the human-authored works.
The team told participants whether their story was written by a human or AI, but this information was not always correct. The researchers then asked participants to rate how absorbing and engaging they found the story, and its quality.
Participants who read an AI-generated story rated it as more absorbing and of higher quality than those who read a story written by a human. However, participants gave higher ratings to stories they had been told were written by people.
AI is more competent than most. At most things too. It’s not – so far at least – better than the best.
Also note that the human preference still seems to be for the human production. Seems fair, given human desires etc.
The secret of a good short story is efficiency : a concise narrative that gets the point across.
A lot of sci fi writers are good at this sort of structure; John Wyndham’s stories were sometimes better than his novels.
The master of course is PG Wodehouse.
So really given 10 or 15,000 or so words, it is not surprising that AI could knock off a decent effort. Feed it a plot and character and let it trawl a host of examples. A 100,000 word novel would rapidly degenerate into repetition.
I’ve seen pro musicians admit that AI is in equal measures scaring and impressing them in its ability to add phrasing and extra expression to music that they had assumed was the preserve of human skill driven by human emotion.
I have no musical ability but would imagine AIs pattern matching and prediction would excel at that.
Especially in adding a word or two. Does AI work in teams? With one version being the copilot? That’s a poor outlook for us humans, and must surely be the power point of it all.One note of course isn’t a symphony, but perhaps AI has the edge.Thak heavens we all have access to it.
Damn Microsoft!
Bravo, sir, bravo
Does it really though? I like synthesis. Which is recognising that solutions in one sphere may work equally well in totally different spheres. Which is pattern matching. But to do that, you need a deep understanding of both spheres to detect the patterns.. Can AI do that? It’s not as if it actually understands things.
I’ve listened to someone playing live vs. AI jazz piano trio music and the AI was extremely convincing until you listened hard. More than good enough for Muzak and restaurants, though, and that’s going to be the use case.
That and Boy George using it to knock up his pro-Jewish song, presumably because he couldn’t get his woke mates to play on it.
If I were still churning out vids I’d certainly use AI for the music pad. No copyright in it, and therefore no need to struggle with trying to get sync rights.
This has made me think of that Bob Newhart sketch where he has an infinite number of monkeys at typewriters trying to write Shekespeare :
Hold on… I think Monkey 3758 is onto something here.
To be or not to be.
That is the zpryqul
That would be the famous Arabic playwright Sheik Speare, would it not?
A book of verses underneath the bough.
or, since we’re being vaguely medieval and therefore Robin Hood-y, “…underneath the bow”…
Incidentally, most of what Shakespeare did was taking existing stories and sharpening them up. He looked at them, and figured out how to make them better.
And then Terry Pratchett sharpened them further….
It was William’s non binary sibling.
Or should that be Xekespeare ?
‘spear’ is still too patriarchal.
Remember that the prompt to generate the AI story is written by a human.
My own ventures into creative writing have not been met by great success. There are good human writers and there writers who should stick to their day jobs, which I incidentally did. The three human writers in the study were perhaps not of the same caliber as the greats. So we’re comparing the best of AI with three not spectacularly gifted hacks.
So far. But presumably it’s possible for a human to generate the prompt which generates prompts to tell the other AI bots what to write.
Revealed preferences maybe? If you ask people they will tell you that human is best, that AI can’t be a real author… Because that is the right thing to say, right? But none the less they will consume AI generated stuff and judge it good. Rather like UPF etc.
I’m seeing a distinct kneejerk prejudice against generative AI, yes. Sure, there’s an awful lot of terrible slop around, and I’m sympathetic to people irritated with running into such slop so often. And yes, the whole training data issue is an open wound.
But I’ve seen people break out the pitchforks over decent, good, and remarkable AI creations. Which seems Luddite.
Hmmm. For this to be informative, all the writing would have to be of the same high quality. So what’s the repeatable agreed metric for quality of writing?
Oops.
There isn’t one.
So exactly what this useless exercise is showing is somewhat unclear….
The problem is that it ain’t going to write The Martian, Murder on the Orient Express or The Age of Innocence.
And most humans can’t do that thing of writing a really clever or meaningful story, either. Even most bestsellers, most booker winners are not mind-blowing books. I remember reading Wolf Hall and I couldn’t get what the fuss was about.
It’s like ai music sounds like album tracks from a local band. But writing Yesterday or Dancing Queen are next level. It’s why songwriters can be stinking rich. Writing a banger is exceptionally rare.
And this is what entirely disproves the Labour Theory of Value. Elton John writes a global hit in half an hour; some bloke takes seven years to make a 12′ Eiffel Tower out of matchsticks. Which is the more valuable?
A composer’s assistant knocks up most of the underscore for a film, as my wife did for Joe Hisaishi for Laputa, Castle in the Clouds. Or nowadays, Joe knocks up a few prompts and gets an AI to do it. Which is the more valuable?
‘Course, Joe will keep the main theme and its orchestration for himself. He’s not stupid.
Another example. I’ve recently rediscovered Eric Malpass, an author almost entirely unknown in England but Big In Germany. I think he’s fabulous, thoroughly recommend him, and to my mind most of his books are equally good, but most of his income came from just one. All took around the same time to write. What does the Labour Theory of Value have to say about this?
From wiki:
The fucking Krauts – completely mad.
Isn’t this roughly what you’d expect? I think I’ve got this right. The AI is working from a database of short stories no doubt rated for popular acceptability. So it emulates the writing styles, characterisations, descriptions, themes etc This is exactly the same as most human writers are doing. So is it a surprise it does it better. I think Ottokrings reference to sci-fi (although the SF I prefer is spec-fi because most sci-is cowboys & indians with space ships – highly derivative) is a good one. An AI is never going to do a William Gibson who pretty well created cyberpunk. The first page of Neuromancer describes the “television sky” over Tokyo Bay. The screen of CRT TV tuned between channels might not mean much to people now but it did in ’84. In a couple of words it gave the concept of both industrial & light pollution added to a Western Character down & out in an Oriental city & produced the World City concept. In a single page the reader is transported across time & space to a different & new world.
Even the title of the novel is descriptive.
Incidentally, Gibson is a writer who does short stories in the long form. His novels have high information density right the way through. And high information density is the key to writing good short stories.
It will be interesting to see Apples’ take on Neuromancer as a TV series coming out next year.
I’ve had a look at the original article, and it’s *exactly* as you’d expect.
They’ve taken 1000 word short-shorts from actual publications **.
And then asked AI *very specifically* to create a 1000 word short-short based on the same concepts and plotline, with a different Hook. ( it’s in the Methods, very clear for a change….)
Given what “AI” actually is, and the *extremely limited* boundaries in which the story had to be generated you *would* get an output that would be near-indistinguishable from the original.
There’s only so much you can do with 1000 words, after all…
So small wonder people had trouble distinguishing the stories.. There was hardly anything to distinguish…
What depresses *me* is the fact that 1000 words is apparently a “5 minute read”….
I’m not *exactly* a speed-reader, but I can go through stuff at a decent clip, and 1000 words shouldn’t take more than 30 seconds for a competent reader..
** The stories themselves reflect the Woke/Arthouse/Literary genre…
I pity the poor subjects… the $3 compensation for those 5 minutes seems *way* too low to be subjected to such torture…
What depresses *me* is the fact that 1000 words is apparently a “5 minute read”….
I’m not *exactly* a speed-reader, but I can go through stuff at a decent clip, and 1000 words shouldn’t take more than 30 seconds for a competent reader..
I think it depends on how you read. It sounds like you read like I do. I don’t read every word consecutively. I read in blocks of text, picking out the significant words. Some blocks I don’t even really read because there’s few significant words in them.
How many other people read like that? I think most people read more consecutively.
I do both. Like you when I want info; slowly and consecutively – sometimes repetitively – when I want to savour the imagery or humour.
Yeah. I was thinking about what happened last night. Girlfriend came home in a panic thinking she’d bricked her phone. She’d accidentally turned it off & tried restarting, but the volume key rather than power. Not understanding what came up she ended up with a fresh install.
A search for the model produced an entire para from which I got three words without reading the rest UP*VOLUME*RESTORE. Gave her back a working fone 30 secs later.
And a reference back to yesterday’s on YT. First thing came up on search was a YT vid. The audio track was a very poor voice translation with matching sub titles. Utterly incomprehensible.
Anything on YT is either trying to sell you something, an attempt to moneterise YT, or someone wanting to boost their own self esteem. I wouldn’t trust anything on it.
I find some of the car repair (and washing machine repair, etc.) videos very useful.
Yes, a lot of them are, as you say, appallingly lacking in information density. But you can tell that just by how long they are.
The well-done ones can show me what to to a lot better and (for me, being an ignoramus about this sort of thing) a lot quicker than trying to work out from a text description which bit I have to twiddle which way to get it to come out without breaking all the clips.
Anything that, after the youtube video, still looks too difficult (or liable for me to seriously bugger it up) still does to the mechanic. But youtube and ebay spare parts between them have saved me a lot in repairman fees.
I’d say be very very careful. One of the vids I saw, a cambelt replacement on a car engine. If you’d followed the instructions, you’d a very high probability of a 2 grand engine rebuild. One presumes it’s a real repair being done on a real car. Could just as easily be a clunker had the engine hosed over & the ‘new’ cambelt, the old one bootpolished & turned round. A vid made just to get views & some dosh.
If you had access to a wreckers yard you make stream of vids like that in a day. Take about 10 mins per
Here’s another thought.. Most good writers have a concept in their heads they want to share. The writing is the sharing. The better the writer the better the sharing.
How many people who write just want to write? To be a writer? They have to struggle for concepts. An AI is the latter but it doesn’t have concepts..
Lots of people want to write, but they don’t have a story to tell. And I don’t mean the surface level stuff, but the human story underneath it. Or don’t do anything interesting plot-wise.
I think a lot of women write because they could write nice paragraphs at school and got talked up as “very good at writing”.
I read Wolf Hall. Huge book and at the end I was like “and?” It didn’t leave a mark on me. It doesn’t really have anything to say about the human condition.
The extraordinary fraud chap at Cambridge explains his plagiarism by saying that his autism means that he has to learn things by mimicking them.
I suppose he’ll later expand that to say that learns things just like an AI package.
That’ll also explain the hallucinations. I’m mildly surprised that he hasn’t hit on this excuse yet. Maybe he’s a bit dim. Though not half as dim as all the stupid buggers who’ve spoken up for him.
I suppose he’ll later expand that to say that learns things just like an AI package.
That’s because he hasn’t got round to inventing AI yet. Give a few more days.
He’s now resigned.
I wonder if he’ll be prosecuted for fraud. (lol, I don’t really).
Don’t know why I should care if a product is produced by machines or people. Gamecock will decide what he likes; the marketplace will decide what succeeds.
Journalists, who are professional writers, write junk. Advantage: AI.
I’ve been using AI to write fiction for about 18 months now, mainly sci-fi and fantasy. I’ve had numerous ideas over the years but lack the time and skills to write by myself.
The premises and concepts are mine. I use the AI (mainly Claude, but also ChatGPT to brainstorm ideas and do a second pair of eyes review) to work out the structure and key details (esp in a sci-fi story to check the maths, timing, consistency, etc) and then generate the body text. I then review and edit where needed.
I view the relationship as akin to when an established author would come up with an idea, then get an up and coming author to do the grunt work (Arthur C Clarke did this a lot in his latter years).
To date I’ve done a novel, a novella and numerous short stories ranging from 1500 to 12k words. If you want to have a look check out my Royal Road page.
I view the relationship the same as stuffing sausage, or using a machine to help.
The world fellowship of writers seems to think they should be shielded from mechanization.
A. They can’t stop it;
B. Their product isn’t all that good in the first place.
Consumers can expect BETTER product!
Rather like the great renaissance artists, who got their apprentices to do most of the picture, then the Great Master fiddled around with the face and hands.