Hey, hey, Geekly Gang! Kyra Kyle here with another writing brain dump. I’ll begin with a quick update with my progress. It is Ber season after all. I’ve reached 1,000 to 1,200 words per day. At this rate, I’ll easily reach my goal (between 50,000 and 70,000 words) by the middle of next month (November 2026). Who knows? If I finish early, I may write some of the prewritten Geekly posts for next year. If I get done really early, I may write some extra scenes for Gods and Dolls. Yay! I’ll hold off on editing until 2027. Gotta make sure the manuscript sits for at least a month, so I can look at it again with fresh eyes. If you begin editing too soon, you’ll find fewer flaws. You don’t want to be too close to the project.
With my personal update out of the way, let’s talk about something that may discourage writers from writing: Artificial Intelligence. That’s right. I’m talking about AI. If you’ve followed this site for a while, you’ll know that I don’t have a lot of nice things to say about AI. I’ve deleted more than a few board game and video game entries from our weekly round up of new releases (Geek News on Sundays) because the project used AI art. In short, AI art isn’t art. Art is an expression of the human condition. Artificial Intelligence is incapable of that because it isn’t human. With the exception of a handful of artists who give their permission for large language models (LLMs) to use their work, generative AI art is plagiarism.
One bit of good news: AI writing can’t replace human writers. AI writing is inferior to human writing, and I don’t see that fact changing any time soon. To better explain what I mean, let’s discuss how to spot AI writing and why the programs that produce it behave the way they do.
Em Dashes
First off, I’m going to stop saying AI writing. Artificial Intelligence has no intelligence. What tech bros call AI is a Large Language Model (or LLM). LLMs scrape as much content as it can from as many sources as it can, the more the merrier, and splices that content together into something that sounds original, but isn’t. This means LLMs behave in certain ways, ways that will always make human writing superior to text vomited by an LLM. The most documented way people spot LLM text is the model’s misuse and overuse of em dashes (–).
Technically, there’s nothing wrong with using an em dash, but you need to know how to use the punctuation. This goes for all forms of punctuation. Em dashes are used for abrupt changes of thought. The most common ways LLMs will misuse em dashes is replacing proper punctuation with an em dash. For instance:
I loaded my backpack–writing log, pens every color of the rainbow, and a granola bar for a snack.
In this example, an em dash was used instead of a colon. LLMs use em dashes whenever they don’t know the exact punctuation to use. Current LLM builds will misuse and overuse em dashes, but the punctuation the LLM will overuse and misuse may, and probably will, change in the future. Humans have figured out this tell. The next punctuation LLMs might overuse could be commas, semi-colons, colons, quotation marks, parentheses, or something else entirely. Who knows? But humans can learn which punctuation to use. Style books come in handy. Personally, I like Strunk and White’s Elements of Style, but any style book will work.
LLMs aren’t human. They can’t learn; they can only copy.
The Rule of Three
Three is a magic number. I use the rule of three in my writing. Heck, I have a short story collection of literary triptychs, the same short scene shown through three different lenses. The Rule of Three is a powerful literary tool. One or two might be too few. More than three may be too many. Three is just right. Just ask Goldilocks.
Despite the power of three, it can be abused. LLMs abuse the power of three aAll the time. A couple of months ago, I googled the definition of “business” to see what I’d get as a result. You see, Google added a Gemini AI assistant to its search engine. The website places its “AI” overview at the top of every search result. Basically, Google’s LLM scours the internet and compiles answers for what you searched against, so you don’t have to read an article a human wrote on a website. Who wants to do that? How do you think the Gemini AI assistant defined business?
A business is a corporation, organization, or workplace that produces a product, services, or professional services.
You might not need “workplace,” and Gemini said “services” twice. Wouldn’t professional services fall under the blanket category of services? Gemini did this because it had to have three items. There is no one or two or four.
LLMs will never be able to fix this issue. The only logical way to address this would create more problems than it would fix. Theoretically, LLMs could use a random number generator that picks a number between 1-5, weighing the time each number can be rolled (like on a twenty-sided die) with three being the most common number rolled. Again, LLMs don’t think; it’s nothing but programming. With this hypothetical “fix,” what happens when the LLM rolls a five twice? Would the Gemini AI assistant redefine a business like this?
A business is a corporation, organization, workplace, enterprise, or association that produces a product, services, professional services, individual services, and special services.
The above would be far worse than the previous LLM text vomit. The best course of action for LLMs is to stick with The Rule of Three.
Humans can sense when they need one or two or three or four or even thirty-something items. Tim O’Brien listed around thirty-something items that Vietnam War soldiers stored in their backpacks in his short story, “The Things They Carried.” Listing those thirty-something items served an artist function. Humans can write better stories and sentences. Humans can make art. LLMs can’t make art.
Transition Words
LLMs love transition words like however, consequently, and furthermore. (I had to list three transition word examples because three.) Humans will use transition words, too, but we don’t use them that much outside of technical essays. LLMs LOVE transition words. Large Language Models are trying to find connections between words and exploit patterns. Transition words can help with forming connections when used correctly. Of course, LLMs seldom use transition words correctly. Consider the following:
San Diego is a city in southern California. Furthermore, it’s on the Pacific Ocean.
The above example sounds correct, because both sentences state facts about San Diego. But “furthermore” is used to support or add to a previous statement. Those two facts above have little to do with each other. The following would make more sense if you needed to use the word, furthermore.
San Diego is a city in southern California. Furthermore, it is the farthest south of any of California’s major cities.
In this example, you’re building on San Diego and how south it is with the second statement. Many human writers wouldn’t bother with using “furthermore” in the instance above; we’d rewrite it. LLMs will always want to use transition words because they’re trying to create connections according to its programing, but LLMs will fail to use these transition words correctly because they don’t know why a human writer would use these words. Either don’t use these words or use them correctly. Style books are your friend.
Not X, But Y
At their core, LLMs are computer programs. While computer language has changed in the last several decades, “if/then” statements remain at their core. (“If/then” statements are used in programming like so, if X is true, then go to line Y or perform the Y function.) Or at least some variant of “if/then” statements remain in computer programming today. “Not X, But Y” statements are similar in structure to “if/then” statements, so this removes the language barrier between a computer and humans. As a result, LLMs overuse “Not X, But Y” statements. You may not know what I’m talking about with “Not X, But Y” statements, but you’ve definitely read them or heard them. I’ll give you a common, overused example of a “Not X, But Y” statement.
It’s not a job, but a calling.
When you see this tell, you won’t be able to not see it in LLM text vomit. LLMs love this structure. Here’s a writing sample I saw from a published work that was totally not written by an LLM.
It’s not a job, but a calling. It’s not a workplace, but a family. It’s not a desk, but a throne.
Three! They used three in a row. Three is the magic number. That series of three “Not X, But Y” statements almost makes sense. I love it when LLMs go off on a tangent, like the following:
It’s not a job, but a roast beef sandwich.
Was the statement above written by an LLM, or was someone off their meds, or was a human trying to sound like an LLM? Hey! Three more lists of three. Oh, yeah! Three!
LLMs use “Not X, But Y” statements because it mirrors “if/then” statements, and these can sound profound. “It’s not a job, but a calling” may have sounded deep when it was first uttered, but LLMs will spam this crap. Don’t spam “Not X, But Y” statements.
Closing Thoughts
Human writers matter. Keep writing. You’re better than any LLM now and unless corporations perform some zero gravity moon leaps with their programming, you’ll be a better writer than any LLM in the foreseeable future. Your writing has value. You have value.
Thank you for reading, and wherever you are, I hope you’re having a great day.
