From The Writer's Chair Re. the AI discussion on this week's Pop-ups

[Adrian Bagley]

Fascinating discussion on this week's show about whether or not ChatGPT is going to put us all out of work. I ran a quick experiment a while back that really backs up what the guys were saying.

I asked ChatGPT to "write me a 1,000 word story in the style of Robert E Howard about a strong, agile barbarian hero called Conn who has to fight a kraken while sailing a small boat across an ocean". It came up with this, which I have to admit isn't dreadful. Honestly, the worst bits come from it trying to incorporate my original prompt.

"Conn was a barbarian hero who had sailed across many oceans and fought many monsters. He was strong and agile, with muscles that rippled under his skin like waves on the sea. One day, he set sail on a small boat across an ocean. He had heard rumors of a kraken that lived in the depths of the sea and he was determined to find it.

As he sailed further out to sea, the waves grew larger and more violent. The wind howled like a pack of wolves and the sky turned dark as night. Conn knew that he was getting closer to the kraken’s lair.

Suddenly, the boat shook violently and Conn was thrown off his feet. He looked over the side of the boat and saw the kraken rising out of the water. Its tentacles were as thick as tree trunks and its eyes glowed like embers in the dark.

Conn drew his sword and charged at the kraken. The monster lashed out with its tentacles, but Conn was too quick for it. He dodged and weaved between them, slicing them off one by one with his sword.

The kraken roared in anger and dived back into the water. Conn knew that it would be back soon, so he prepared himself for another attack. Sure enough, the kraken rose out of the water again. This time, it was even angrier than before. It lashed out with its tentacles and Conn fought back with all his might.
The battle raged on for hours until finally, Conn plunged his sword into the kraken’s heart. The monster let out a final roar before sinking back into the depths of the sea.

Conn sailed back to shore victorious, knowing that he had defeated one of the most fearsome monsters in all the oceans."

So, the plot isn't exactly sophisticated (to be fair, neither were Howard's as a rule) and there are some contextual issues - charging on a small boat? Finding the kraken's lair in the middle of an ocean? But the prose has a certain Conan-esque ring to it and the use of simile isn't half bad. Honestly, not nearly as basic as I was expecting.

So I tried again, this time angling after a bit of human interest...

Prompt: Write me a 1000-word story about John and Padraig, teenagers who fall in love in 1920s Ireland. Explore themes of forbidden love, substance abuse and what it feels like to be young and in love for the first time.

Response: "John and Padraig were two teenagers who lived in a small village in Ireland. They met each other at a local pub where they both worked part-time. John was a shy boy who had just moved to the village with his family. Padraig was outgoing and popular among the locals. They started talking and soon became friends.

As they spent more time together, John realized that he had feelings for Padraig. He didn’t know how to express his feelings as he knew that it was forbidden for two boys to be in love. Padraig also felt the same way but was afraid of what others would say.

One day, they decided to go for a walk in the countryside. They found a secluded spot by the river where they could be alone. They talked about their feelings for each other and decided to take their relationship to the next level.

As time passed, John and Padraig’s relationship became stronger. However, they faced many challenges along the way. Substance abuse was rampant in their village and many of their friends were addicted to alcohol and drugs. They tried to stay away from it but it was difficult as it was all around them.
Despite all the challenges, John and Padraig’s love for each other never faded. They knew that they had found something special and were determined to make it work.

Being young and in love for the first time is an unforgettable experience. It’s a time when everything seems possible and nothing can stand in your way. John and Padraig knew that their love was forbidden but they didn’t care. They were happy just being together."

Which I think we can all agree is absolute *****. AI is great at writing sentences, but anything a bit deeper, involving insight into the human condition, is a loooong way off.

What we're feeling right now is probably similar to how mathematicians felt when the calculator was invented. Machines can perform, almost instantly, tasks that would take us hours, and that's scary. But the mathematicians still have jobs, and I don't think we have much to worry about either. What will probably change is the way we work. If you've been banging your head against the desk for ten minutes straight trying to make a paragraph work, AI might just be able to help...

 

[Barbara]

Interesting experiments.

What we're feeling right now is probably similar to how mathematicians felt when the calculator was invented. Machines can perform, almost instantly, tasks that would take us hours, and that's scary. But the mathematicians still have jobs, and I don't think we have much to worry about either. What will probably change is the way we work. If you've been banging your head against the desk for ten minutes straight trying to make a paragraph work, AI might just be able to help.

I hope / think you're right.

A bit of optimism: I've been looking on jobsites for a new kind of day job and I've seen several writing jobs where the employers ask for writing samples alongside the application. A few of them (not many yet) say (and I'm paraphrasing) "don't even think about sending AI written work.".

One magazine offered a 2 month position and again "if youre intending to use AI for the job don't bother applying".

Hope this trend continues.

 

[AnnieSummerlee]

Chat GPT still has to learn how to show-don't-tell, so I think we're alright for now.

 

[Pamela Jo]

You know those new "colouring" books for kids where they dont have to colour. Instead they just put on stickers or wet the page. It depends on whether you think those are an advance or not. Kids never have to train to colour inside the lines or discover blue and yellow on top of each other make green. Parents never have to scrape melted crayons out of the carpet.

 

[Rachel Caldecott]

I have a few things to say about AI.
1. I have always hated the idea of it.
2. I discovered today that one of my mother's books is being used to train AI. (Moyra Caldecott, "A Breathless Pause").

https://authorsguild.org/news/you-just-found-out-your-book-was-used-to-train-ai-now-what/

3. As an experiment, purely in the name of science, I used AI to write a plot summary and tagline for my books. I admit, they were better than I could have done. I am ashamed to have done it, but impressed, nevertheless. But I'm going to keep writing my own, because it is a very good exercise.
4. I saw this article, about the WGA deal. The last paragraph is encouraging:

What does the end of the writers strike really mean?

We're breaking down the four biggest takeaways from the proposed WGA contract.

www.today.com

 

[AgentPete]

Chat GPT is quite good at summarising things, at least in my pretty limited tinkerings. It’s also strangely good at transposing from first person to third and vice versa, and also conforming passages to gender etc.

It has absolutely no conception (uh, let’s not grant it sentience… more precisely, it has absolutely no coding) to produce (a) long-form narrative content and (b) to make that content conform to dramatic structure, e.g. 3-act.

But… it could do. Simply a matter of coding, simply a matter of training it on the right material.

Thing is… is it worth the effort to do all that? What you’re going to get is derivative pulp with no voice (it can fake a writing style for a bit, but that’s not voice).

Whether all this is going to be worthwhile from the publishers’ / studios’ / readers’ point of view is a moot question. Do they all really hate writers to the extent that they’ll fund development of this weird offshoot, and then pay for ongoing costs (let’s not forget that AI comes with pretty astronomical costs). Don’t they (publishers & studios) have rather bigger issues to confront at the moment? And I don’t hear readers demanding more books from non-human entities very much at all.

I hope sanity prevails.

 

[Mel L]

Interesting experiment. Personally, I think we should all learn to work with it, tame the beast as it were. In the spirit of 'keep your friends close and your enemies closer.'

 

[Pamela Jo]

Ok Mel. Tame this beast. I just wanted to check spelling for haggard, which I had got wrong. So the algorithms decided I meant Hagrid from Hogwarts and would give me nothing else... until this happened.

What is another word for hagrid - WordHippo​

What is another word for hagrid? Need synonyms for hagrid? Here's a list of similar words from our thesaurus that you can use instead. "From then on the Hoo-Doo cowboy would hagride the night holding the horn of the lone green horse.".

 

[JohnBertel]

Ok Mel. Tame this beast. I just wanted to check spelling for haggard, which I had got wrong. So the algorithms decided I meant Hagrid from Hogwarts and would give me nothing else... until this happened.

What is another word for hagrid - WordHippo​

What is another word for hagrid? Need synonyms for hagrid? Here's a list of similar words from our thesaurus that you can use instead. "From then on the Hoo-Doo cowboy would hagride the night holding the horn of the lone green horse.".

I tried this:

The word "hagrid" as in "a hagrid ride" is probably misspelled. What is the correct spelling

And got this, which I think is helpful:

The word you're likely looking for is "haggard," not "hagrid." "Haggard" is used to describe someone who looks exhausted, worn, or gaunt, often due to fatigue or stress. So, the correct phrase would be "a haggard ride" if you're describing a ride that leaves someone looking tired or worn out.

 

[Pamela Jo]

I tried this:



And got this, which I think is helpful:

Yes that was the point. The sp I was checking was haggerd. But the powers that be decided I must mean Hagrid and we were off to the races.

 

[Adrian Bagley]

It has absolutely no conception (uh, let’s not grant it sentience… more precisely, it has absolutely no coding) to produce (a) long-form narrative content and (b) to make that content conform to dramatic structure, e.g. 3-act.

But… it could do. Simply a matter of coding, simply a matter of training it on the right material.

So far, its grasp of plot seems to be at about the same level as a very young child, minus the wild imagination (did anyone watch Axe Cop? Genius!) That's bound to improve, but a full length novel is an exponentially more difficult challenge. Possible, but as you say, I'm not sure it would be worth the sheer effort required to produce anything that wasn't utterly formulaic. It would certainly take a very long time (at least a generation IMO).

And even if it gets that far, plot is one thing, story and character arc quite another...

Whether all this is going to be worthwhile from the publishers’ / studios’ / readers’ point of view is a moot question. Do they all really hate writers to the extent that they’ll fund development of this weird offshoot, and then pay for ongoing costs (let’s not forget that AI comes with pretty astronomical costs). Don’t they (publishers & studios) have rather bigger issues to confront at the moment? And I don’t hear readers demanding more books from non-human entities very much at all.

I think they'd do it in a heartbeat if it made commercial sense. Cutting out the human probably isn't anyone's priority for a mere 10% of the take. But I believe digital-only contracts can award royalties of up to 50%? Which is far more worthwhile... Still, the results would be too derivative, as you say. No one would buy it.

Actually, I'll qualify that - it might make sense for ghost writing? I can imagine AI taking badly written celebrity scribblings (which nevertheless already contain the essential human/emotional element and have a real person's name on the cover) and turning them into professional-sounding, structured prose. And that's a sizeable chunk of the business, right?

Aside from ghost writing, I don't think AI is ever going to be up to the challenge of a full-length work of original fiction, unless or until it's so hyperintelligent that it can understand what it is to be human merely from analysing our online content. At which point, we're not so much out of a job as completely redundant...

 

[AgentPete]

Fron today's Publishers Lunch:

Authors Respond to OpenAI's Defense​

Attorney's for authors Paul Tremblay, Sarah Silverman, Christopher Golden, and Richard Kadrey have filed a new brief in their lawsuit against OpenAI, the makers of ChatGPT, refuting the company's initial defense.

Last month OpenAI moved to dismiss the plaintiffs' "ancillary" claims, hoping to focus on the core question of copyright infringement. In response, the plaintiffs ask the court to dismiss the motion, and "challenge OpenAI’s position that it should be allowed to train its generative AI products using anyone’s name and copyrighted literary works, without consent, for free, forever."

"In this case, the copyrighted works of a class of millions of individual creators—including the four representative Plaintiffs, who are accomplished book authors—have been misappropriated by OpenAI without consent, credit, or compensation, with the goal of enriching itself," the filing reads.

On the motion to dismiss, the plaintiffs say they have "adequately alleged," "state" and provided "alleged facts demonstrating" their claims. "OpenAI's motion to dismiss may be long on rhetoric, but it is inconsistent with basic pleading rules."

Based on OpenAI's initial answer, the filing asserts that the company is getting ready to defend their copying and use of the books as fair use. "OpenAI’s construction of fair use is at odds with settled precedent and if adopted here would overrule that precedent, entirely swallowing U.S. copyright," it reads.
Full filing

 

[Hannah F]

Fron today's Publishers Lunch:

Authors Respond to OpenAI's Defense​

Attorney's for authors Paul Tremblay, Sarah Silverman, Christopher Golden, and Richard Kadrey have filed a new brief in their lawsuit against OpenAI, the makers of ChatGPT, refuting the company's initial defense.

Last month OpenAI moved to dismiss the plaintiffs' "ancillary" claims, hoping to focus on the core question of copyright infringement. In response, the plaintiffs ask the court to dismiss the motion, and "challenge OpenAI’s position that it should be allowed to train its generative AI products using anyone’s name and copyrighted literary works, without consent, for free, forever."

"In this case, the copyrighted works of a class of millions of individual creators—including the four representative Plaintiffs, who are accomplished book authors—have been misappropriated by OpenAI without consent, credit, or compensation, with the goal of enriching itself," the filing reads.

On the motion to dismiss, the plaintiffs say they have "adequately alleged," "state" and provided "alleged facts demonstrating" their claims. "OpenAI's motion to dismiss may be long on rhetoric, but it is inconsistent with basic pleading rules."

Based on OpenAI's initial answer, the filing asserts that the company is getting ready to defend their copying and use of the books as fair use. "OpenAI’s construction of fair use is at odds with settled precedent and if adopted here would overrule that precedent, entirely swallowing U.S. copyright," it reads.
Full filing

I hope the authors win!