AI and SF Writing

Windshadow

need more dried frog pills
Joined
Oct 25, 2015
Messages
3,156
"Deathworld by Harry Harrison"
Deathworld is a science fiction novel that follows the adventures of Jason dinAlt, a gambler with psionic abilities who gets involved with the colonists of Pyrrus, a planet where everything is trying to kill them.

The novel explores the themes of survival, adaptation, evolution and conflict between human cultures. It also features thrilling action scenes, exotic creatures and plants, and a mystery about the origin of Pyrrus' hostility.
The novel is well-written, fast-paced and engaging. The characters are believable and likable, especially Jason who is witty, resourceful and courageous. The plot is original and intriguing, with twists and surprises along the way. The setting is vividly described and creates a sense of danger and wonder.
Deathworld is a classic of science fiction that deserves to be read by fans of the genre. It is the first book in a trilogy that continues with Deathworld 2 (also known as The Ethical Engineer) and Deathworld 3 (also known as The Horse Barbarians).
Well, since no one bit at my lure above, I must out myself in reference to the Deathworld review I posted above. While it is not plagiarized it was also not written by myself.
I gave this instruction to the enhanced Bing ‘engine’
“Write me a short review of the novel Deathworld by Harry Harrison”.
That is all: and Bing’s output I posted verbatim
 
ChatGTP can write better than I think K-12 teachers will be able to detect if the student does a bit of light editing and tosses in a few misspelt words and vocabulary or grammar errors.

I would call this review high school sophomore or perhaps junior level based on my memory of how much I had to dumb down what I passed in to reduce the questioning about my real knowledge of the vocabulary I used. I did not have to fake my spelling errors or my misuse of commas which remains a bit haphazard to this day, as I am sure many here will be aware.


These are also generated with simple orders to the chatbot, for example:
Here is a short poem in the style of Rudyard Kipling about sail making:

The Sail Maker

He works with needle, thread, and cloth,
To make the sails that catch the wind;
He knows the art of every stitch,
The strength of every seam and hem.

He shapes the canvas to the mast.
He trims the edges neat and fine;
He makes the sails that speed the ships,
Across the ocean’s restless brine.

He loves his craft, and he loves his trade.
He loves the smell of salt and tar;
He loves to see his handiwork,
Unfurling 'neath the sun and stars.

He is the sailmaker, my son,
And you can learn from him a lot;
About the skill and pride of work,
And how to do what must be done.

Generating answers for you…


Here is a short poem in the style of Rudyard Kipling about candle making:


If you can melt the wax with a steady hand
And pour it in the moulds without a spill;
If you can scent it with the fragrant oils
And wick it with a cotton thread or quill;

If you can trim the edges smooth and neat
And wrap them in a paper spill;
If you can light them up and watch them burn
And see their flickering shadows on the soil;

If you can make them brighten up the dark
And warm the hearts of those who need a spark;
If you can sell them at a fair exchange
Or give them as a gift without remark;

Then yours is the craft of candle making, friend,
And - what is more - you'll be a candle maker till the end.
 
Ha, ha, you had me fooled. You are lucky I'm too lazy to tell ChatGTP to write a short rebuke for letting AI's do your reviewing for you.

What next though? Hey ChatGp-whatever...write me an interesting 500 page novel with dragons, elves and humans?

Skynet in it's infancy?

To be fair though, it read like every ebook blurb out there, so I suspect it is way more common than most people think.
 
What next though? Hey ChatGp-whatever...write me an interesting 500 page novel with dragons, elves and humans?

That's what GRRM is waiting for. As soon as the AI is good enough, he'll say "ChatGTP, finish ASOIAF for me."

Seriously, though, I have my doubts it will ever be good enough. I think there will be a limit on how good AIs can get at creative writing. While it does fine at writing essays, its stories are really lacking.
 
What next though? Hey ChatGp-whatever...write me an interesting 500 page novel with dragons, elves and humans?

It sounds ridiculous but that is exactly what things are going to become. We won't have to wait for anything we want - we will eventually be able to request anything in consumable digital form tailored to our exact requirements: novels, films, documentaries, cartoons, news - whatever - without any direct human involvment!

A lot of this is already available and in time the more complicated things such as films and full computer games will be only a few clicks or keywords away. It might sound far fetched but this is pretty much the beginning of the end of commercial creativity as we know it.
 
I recently read a NYT article that said that the editors of Clarkesworld, Asimov's and F&SF are currently deluged by chatbox submissions to the extent of hundreds (and in one case over a thousand) a month. The eds were also uniformly scornful of their quality, saying that they were both tripe and easily distinguishable from human output, HERE is the link if you can open NYT text.
 
We got Neil Clarke of Clarkesworld on BBC Radio 4 here interviewed and explaining the details. He didn't want to say what the tells were (obviously!) but he did say it was obvious what they were. Given time, though, who knows? :)
 
The eds were also uniformly scornful of their quality, saying that they were both tripe and easily distinguishable from human output

Currently, maybe, but give it a few months or a year and things will be very different. Everything kind of took a huge step up in quality and speed this week in AI and the trajectory seems to indicate exponential improvements in almost every area.
 
Currently, maybe, but give it a few months or a year and things will be very different. Everything kind of took a huge step up in quality and speed this week in AI and the trajectory seems to indicate exponential improvements in almost every area.

We have to be careful about extrapolating like that, though. We have had some exponential improvement recently, but that doesn't mean the improvement will keep being exponential in the future. There may be some fundamental limitation on what you can obtain with AI models of this kind. The fiction the AIs produce currently is very poor, and I have my doubts on whether they will ever produce long stories as well or better than humans, without those stories being extremely formulaic.

I mean, we thought self-driving cars would be a thing by now, but it looks like they have reached a wall, after making a lot of progress.

We'll see.
 
There may be some fundamental limitation on what you can obtain with AI models of this kind.

I don't think there is. I think some kinds of output are currently more easily generated than others, for example essays and pictures are comparatively easy, whereas short films are difficult. But the improvements are coming almost every day.

I'm an architect - I saw a video last week about a competition to design a luxury house between a group of architects and an AI. The results were put anonymously to a vote and the AI won with almost 2/3 of the votes. It had made some errors in the layout but in the grand scheme they were only small mistakes and were fairly easily rectified. It also did the work in a fraction of the time taken by the human architects too of course. So currently AI can get 90% of the way there and still win. I really can't see developers stopping trying to iron out a few imperfections to get to 100%. It will happen.
 
We got Neil Clarke of Clarkesworld on BBC Radio 4 here interviewed and explaining the details. He didn't want to say what the tells were (obviously!) but he did say it was obvious what they were. Given time, though, who knows? :)
Do you have a link to the interview?
Also do you think these posts should be moved to a thread of their own perhaps in the general forum? I did not mean to hijack this thread this way I was just reading about the latest version of the AI chat bots while having a reread of death world and for fun posted the review to see if any one would spot it. Then in an outpouring of guilt :oops:and shame o_OI outed myself:eek:
 
I did not mean to hijack this thread this way I was just reading about the latest version of the AI chat bots while having a reread of death world and for fun posted the review to see if any one would spot it.

To be fair, it was a subject very much on my mind the past few days. I agree that a separate thread about AI definitely makes sense on a SFF forum ;)
 
I like Sabine Hossenfelder's take on AI, particularly the conclusion that states AI may not need to be perfect to be able to wreak considerable havoc on our current economy. And when did the possibility of economic havoc stop us doing anything like that in the past anyway? :)

 
They aren't AI programs. They're auto-fill chat bots. Images and words are uploaded into the programs and they just copy and paste them together from entered prompts. A chatbot "architect" isn't building a blueprint on its own -- it's lifting blue prints from human architects. These programs mindlessly regurgitate our own content back to us -- content that their programmers largely stole. But right now tech companies are hunting for government money, so the marketing spin is truly weird.
 

Sponsors


We try to keep the forum as free of ads as possible, please consider supporting SFFWorld on Patreon


Your ad here.
Back
Top