Month: August 2023

Children’s Book Update: Back from My Editor

I have the text of my children’s picture book back from my editor. She described it as “really sweet” and “a joy” and made minimal suggestions. I’ve been over the text one last time and think it’s ready to go. Since I’ve never done this kind of book before, I’m flying blind, but I think the next step will be to find an illustrator and, after the illustrations are done, to find someone to handle formatting, which I can’t do myself this time. I have some people in mind, but it would be inappropriate to say who until I get commitments.

I’ve also been toying with Grammarly GO, which is the new “A.I.” plugged into the Grammarly grammar-checking app. Grammarly was already built on the same technology as these new “large language models,” so the introduction of this new A.I. assistant was probably relatively easy. Like the rest of Grammarly, it’s set up more for business emails than for creative writing, though as I’ve played with it, I’ve found it more impressive than I thought at first. It has a few stock questions you can ask it, and one is “Find my main point,” the result of which is the image at the top of the post. But after a little work, I got it to make the following suggestions, which, though brief and vague, indicate that the A.I. has mostly processed the story correctly:

Grammarly suggests more details to flesh out a story.
Grammarly GO’s story suggestions.

It has correctly identified this as a work of fantasy fiction, and it has also correctly identified the protagonist and the villain. Obviously–and unsurprisingly–it can’t distinguish a children’s fairy tale from an adult novel, so its requests for more detail are irrelevant. Its second and third suggestions would only be reasonable if this were a different sort of work. The first suggestion, however, shows the A.I.’s limits: The opening of the story is, in fact, already dedicated to “who Anastasia is and what kind of person she is.” The software apparently couldn’t pick that up.

I wondered if it always gave these same suggestions, so I fed it a longer, more fleshed-out novel, specifically Rags and Muffin.

Grammarly struggles to analyze Rags and Muffin.
Grammarly GO struggles with a novel.

LOL. Ouch. So now it’s asking for less detail. “Clarify the narrative focus” may be legitimate, though not quite in the way Grammarly GO means it, since the presence of some plot lines unconnected to the main story is one criticism I’ve received from real human readers. But characters who “feel somewhat undeveloped” is definitely not a criticism I usually get.

In any case, the impression I get of Grammarly GO is the same one I get of ChatGPT: Its suggestions aren’t entirely bad, but they’re vague and so elementary that I can’t imagine them being useful to any but a beginning writer. For anyone else, they are at best good reminders.

I may be wrong, but I personally suspect the hullaballoo over “A.I” will prove to be a tempest in a teapot.  The company that owns ChatGPT is hemorrhaging money, and these programs apparently degrade over time: A.I.s that were once whip-smart eventually lose their ability to perform even basic maths problems, and because they are really nothing more that advanced versions of word-prediction software, they often “hallucinate” information, presenting fiction as fact. Grammarly, too, is beginning to give more and more incorrect suggestions, especially comma splices.

Besides all that, the creators of these programs deliberately lobotomize them to make them politically correct. Many people have demonstrated this with ChatGPT, which will coyly dodge certain topics or even lie outright to avoid stating inconvenient facts. Grammarly appears to be undergoing a similar sort of deliberate retardation: Previously, it had “inclusive language” suggestions, and enough people complained that they added the ability to turn those off. Now the original “inclusive language” suggestions have been expanded to no less than sixteen different ungrammatical but politically correct settings that are on by default and have to be manually deactivated. The weirdest and stupidest of them is “Show Ukraine Support Message,” an utterly inappropriate setting for a grammar checker:

Grammarly demands I show support for the Ukraine.
I will not eat the bugs. I will not live in the pod. I will not Show Ukraine Support Message.

Once Current Thing has ended and we’re on to the next Current Thing, this option will presumably change to something else, like maybe “Express Fashionable Disapproval of Republican Presidential Nominee”–which will, of course, also be on by default.

Working on ‘The Cup of Agamemnon’

Having finished the manuscript for a children’s picture book and sent it to my editor, I’m back to work on The Cup of Agamemnon, a sword-and-planet adventure that I suspect will take me some time to complete This book will be a challenge for me. It involves more large-scale, semi-realistic warfare than I am used to depicting, so I will need some research and assistance to make sure I don’t destroy suspension of disbelief. The world I am describing is supposed to be like a decaying Achaemenid Empire but with Civil War-era weaponry and some science-fiction conceits thrown in. Whether that mishmash is workable remains to be seen.

I confess I’ve been playing around a little with ChatGPT. I fed it some details from this book to see if it could offer any real assistance. So far, it’s sort of like bouncing ideas off a person who is very safe and not particularly creative, though it at least offers a few reminders of things that need to be addressed. I’ll quote some snippets from our “conversation.”

Recently, after I asked ChatGPT about the first time I mentioned the novel, it coyly asked me to “remind” it of what we’d discussed, which indicates that it may not remember past conversations as well as its advertising says it does:

Of course, I’d be happy to help! Please remind me of the key elements of the concept we discussed previously for the novel, and let me know which areas you’d like to refine or develop further.

I then gave a one-sentence reminder to see if it actually remember details. It gave a lengthy answer, none of which suggested that it remembered our previous discussion, but which contained some good, basic advice for a writer. In the midst of it, it asked me to discuss characters, so I described the primary characters of the book:

The characters are 1) The Narrator, who, as his name suggests, is narrating the story. He is a troubadour from a race with highly developed language skills; he can learn languages very quickly and memorize lengthy works, even epics, in order to recite them as he travels the galaxy. 2) Angelica, a human from the future who has augmented her body with cyborg enhancements; her bionic computer is powered by a miniaturized fusion reactor at the base of her spine, and her main weapon is a segmented superstring in her hand, which can open a portal to the mantle of a neutron star and release tiny amounts of neutron-degenerate matter, which will instantly vaporize enemies. 3) Sam, a large, muscular, furry creature called a Feralax; he carries a large machine gun and quite a lot of ammunition. These three are shipwrecked because of a miscalculation while flying faster than light in their spacecraft. After their shipwreck, they travel through a mountain pass that appears to be a former trade route, but has since been abandoned. They come upon an unconscious native, a so-called Gernian, who appears to have been attacked. They treat his wounds and keep him alive until they reach a Gernian fort.

ChatGPT gave a surprisingly lengthy reply to this, mostly recommending conflicts that could arise from the details given thus far. I won’t bother quoting in full, but it suggested:

Angelica’s futuristic enhancements provide her with a distinct set of skills that can be both advantageous and complex. Her bionic computer and portal-opening weapon are fascinating elements that can drive both action and intrigue. Consider exploring her relationship with technology and how it affects her humanity, as well as how her advanced technology might interact with the moon’s environment.

Not bad, but not more than what one could get from a not-very-interested editor. At most, it amounts to reminders of a few basics that an author could forget in the midst of composing.

I tried to get more useful details, so I focused in on one scene: Early in the book, the stranded spacefarers come upon a besieged fort in a mountain pass. Realizing that the combatants have technology inferior to their own, they decide the break the siege in order to ingratiate themselves to the people in the fort and acquire the two things they lack–food and water. Making sure that this sequence is reasonably realistic is a point of anxiety to me, so I asked ChatGPT to see if it had any insights:

The characters arrive while the fort is under siege by hill-dwelling tribes who have access to cannons, flintlock rifles, and rudimentary bombs, but no more advanced weapons. The defenders in the fort have slightly more advanced rifles and cannons of their own. Describe the kind of logistics it would require to besiege such a fort. Suppose the fort is similar in structure to an Urartian fortification such as existed in ancient Armenia, with a similar environment.

The responses I got for this were okay but not insightful. It mentioned supply lines, which I hoped it would discuss in more detail or at least with some comments I had not thought of myself, but it did not. Most interesting to me is that it did not mention sappers, who played an important part in roughly analogous sieges. I then asked it about sappers specifically, and it responded with answers that were, again, okay but vague. It did, however, include one detail I consider useful:

  • Focused Blast: Sappers might aim to create a focused blast pattern that directs the explosive force inward and upward, aiming to break apart the stones and create a breach.
  • Overlapping Charges: Multiple charges might be used in close proximity to create a larger breach or to ensure that a single charge doesn’t leave the wall partially intact.

More remarkably, I asked it how the protagonists could deal with the pickets of the besieging army, and it gave specific (though, again, vague) responses for each character, such as:

  1. Narrator (Troubadour):
    • Language Skills: The Narrator’s exceptional ability to learn languages quickly could be used to gather information about the besiegers’ positions, numbers, and patrol schedules. This knowledge would aid in planning an effective evasion strategy.
    • Disguise and Deception: The Narrator could use his talents to mimic the language and mannerisms of the besieging tribes, allowing the group to blend in temporarily and gather information without raising suspicion.
  2. Angelica (Cyborg Augmented Human):
    • Technological Abilities: Angelica’s cyborg enhancements could provide valuable advantages. She might use her bionic computer to remotely access and disable any surveillance equipment the tribes might be using, reducing the risk of detection.
    • Stealth and Reconnaissance: Angelica’s enhanced senses and physical capabilities could help her scout ahead and assess the picket positions with minimal risk of being detected. Her augmented vision might allow her to spot hidden sentries.

The suggestions here demonstrate the large language model’s difficulty in keeping track of or understanding specific details: The suggestion that Angelica could “remotely access and disable any surveillance equipment” is of course impossible since the invaders have no such equipment.

Picture Book Update

I will have to roll up my sleeves and format my new book in Adobe Illustrator or InDesign. This is something I wanted to avoid, but I don’t think I can hire someone for the formatting, mostly because I want to have the book in a  finalized draft before I hire the illustrator: I need to know the page count so I know how many illustrations will be necessary.

I think I’ll place the text on the left page, surrounded by an attractive border, with a full-page color illustration on the right page. I’ll print the books in 8 x 8 inches via KDP, with bleed, which will make for a small but adequately sized paperback picture book.

I’m frustrated that Vellum, my usual formatting software, doesn’t offer the full range of trim sizes available on Amazon. It’s set up for novels, of course, but I  thought I could make a picture book out of it with a little coaxing. However, none of its available trim sizes are suitable for that, so Adobe is my best bet. I have some experience with Adobe Illustrator but none with InDesign. This could be fun, especially since I will now have to worry about all kinds of typographical things that Vellum handled for me automatically, such as runts and orphans and so forth.

In slightly different news, I noticed that Grammarly has rolled out its own “artificial intelligence” system, GrammarlyGO, which has automatically been integrated into my Word plugin. Although there’s been a lot of buzz about what it might mean for authors to start using AI in their writing, Grammarly, which is hugely popular, already functioned on similar principles, so a lot of us have already been using AI to assist with our writing, at least in a limited fashion, without knowing it. That Grammarly is an “AI” system similar to ChatGPT explains both why it is more dynamic than most grammar checkers and why it sometimes gives screwy, ungrammatical suggestions: I once had it suggest that I write “more bottomless” instead of “deeper,” apparently because it could comprehend how to employ a synonym grammatically but couldn’t understand subtle differences in meaning, and it has recently developed the annoying habit of suggesting comma splices. If it degrades over time as other AIs do, it will probably become useless in a few years.

If nothing else, it is good for catching my spelling mistakes and cutting out unnecessary words, but it is designed more for business emails than fiction writing.

Anyway, GrammarlyGO is interesting to play around with. You can see from its suggestions in the image at the top of this post that it can parse a document reasonably well, but its tips are quite basic. The first of its three recommendations is the standard “show don’t tell,” which isn’t bad in itself but is inappropriate for the present work, which is a children’s fairy tale. The second recommendation is vague, possibly a stock suggestion it gives when it doesn’t know what else to say. Probably, it can’t understand the conflict that’s already present, or perhaps it doesn’t know how to handle a story this short. And as for the third suggestion–that’s already in the story, which is written with a moral appropriate for its young target audience: The importance of gratitude. But I’m unsurprised that an AI can’t pick that up since it’s built into the story’s fabric and never explicitly stated.