AI and Writing - A Series of Looks Inside: Part One

Oh my goshness! Hello, good people of the blogosphere. I'm going to discuss something that's been discussed to death: AI and writers using AI. We're specifically talking generative AI here, but there are a ton of writing helpers that have been using one form of AI or another for a long time. We'll get into those as we go along, but today, I'll try to do some explaining on how LLMs work. If you already know, feel free to skip to the next post (if it's up). If it's not, wait two days, and I'll have moved on to examining the tools.

When AI first hit the market, I was as flummoxed as everyone else. I had trepidation, but I stood back and watched rather than freak out, and I learned as much as humanly possible about the way it works. I watched videos, did shallow dives, and did deep dives, which I emerged from in a much more confused state than when I went in.

I know, right? UGH. BUT THEN!

I was on the phone with one of my children (one that happens to be a computer genius, and I don't use that word lightly) when I lamented my lack of understanding. She proceeded to explain it to me. Only, she was using computer terminology.

Me: "Okay. You're adorable, but I can't understand a single word coming out of your mouth."

Her: "Hmmm... Well, how would you understand it?"

Me: "Pretend I'm a small child with only a basic understanding of the English language and know nothing about computers."

Her: *thinks for a minute then laughs like a loon* "I think I know how to help you understand."

Me: "Hit me with it."

And that's how I came to understand how a Large Language Model (LLM) works. Let me see if I can explain it the same way. If you're confused when I'm done, please drop me a comment below, and I'll do my level best to get you articles that might help.

Imagine you're a small child learning how to speak. You tune in to your parents, the TV, books, and every other thing that'll talk to you. As time goes on and you consume more information, you learn how to construct a basic sentence.

Example:

Mom always asks if you want more food.

Eventually, you understand to say, "I want more food." You understand this by hearing Dad say those same words in the same order. Mom rewards him with more food. However, the possessive I is something you learn from listening to TV and being corrected. Most babies would say, "Me want more food." Right? It's only through repetition of hearing it in the correct order when you begin to understand it's not me; it's I. Make sense?

Now, imagine you have a vast network of text to learn from because you can't hear (not speaking about the hearing impaired, they learn in other ways). You learn word B comes after word A in a typical sentence because after examining literally thousands of sentences, that word is the one that comes next most often. Word C comes after word B most often, so that's the one you use next. Then, you use word D, putting it all together to convey your idea.

I want more food.

On a much larger scale, this is how LLMs work. They're not smart, and they're not aware; they just have a vast amount of text that tells them what word most commonly follows the first in any conversation. Some of them pull from websites and give you a summary using the exact same method, but they're not regurgitating lines word for word, nor do they steal your words and give them to someone else.

It's highly unlikely you're constructing sentences in the exact way hundreds of other people are, so the likelihood your exact structure and prose would be used by an LLM is about a billion to one.

I can hear you being skeptical. Let's go for one more example.

I create a small language model using text from Bob, Lia, Dot, and Steve. Here's what they give me when I ask, "How do you get ready in the morning?"

Bob: I start by getting out of bed, having coffee, and taking a shower. Then, I brush my teeth and get dressed for the day.

Lia: My mornings start with me having coffee and watching TV. I tend to rush around feeding the animals and sit down to play a game or two after about half an hour. Once I'm relaxed, I get dressed and start to sort out what's for dinner.

Dot: I get up, shower, eat breakfast, brush my teeth, and get dressed.

Steve: I roll out of bed and fall into some clothes before rushing out the door and driving an hour to the office.

From those few responses, my language model would "probably" return something along the lines of: I get out of bed, brush my teeth, and get dressed.

Because those are the most common denominators in the text. Hell, it might not even be able to answer the question based on those limited responses and it not having enough reference material to use.

See how it's not the same but contains the general elements of the other items?

That is how an LLM works, folks. It's not smart enough to pull sentences out of thin air. So now, when you see people freaking out about their words being used to train an LLM, you can just smile and move on.

So, for this series! We'll be looking at the following apps/helpers, and I'll do my best to ask them all the same things/insert the same prompts to see how each one responds.

• ChatGPT
• Sudowrite
• Jasper AI
• Grammarly
• Squibler
• Hyperwrite
• Rytr
• Copy.ai
• Type.ai
• ProWritingAid
• Claude
• Shortly AI
• Anyword
• Quillbot

I'll pepper in some posts about other things as we go to break up the monotony of a series, and those programs that don't have a free test feature won't be studied. This is just what I could find when I did a quick search. These will be updated with links to the individual posts.

Anyone who uses AI to 100% write a book and does no editing is an ID10T because of the very thing I explained above. However, they can sometimes pull someone out of a funk or help with a sticky spot in a plot (which is how most authors I've interacted, who use AI, use it). 

What AI can't do is generate large amounts of text and it be original. It'll sound dull and robotic (because it is). You can't just plug in a plot and get a good book. Period. So, let's see what these programs can do. If you don't want or need to use any of these, that's fine. Please just move along. :)

One of my biggest challenges will be with the prompts. Authors I know who use AI at all spend literal hours on their prompts then edit the hell out of the text. It's a helper, and IF you're going to use it, that's how it should be used.

I hope you're all looking forward to this. I'm interested to try them! Maybe I'll be disappointed. Maybe you'll find a tool that does exactly what you're looking for (which isn't writing an entire novel for you). We'll see!

Fingers crossed you all enjoyed this post and are stoked about trying stuff.

That's all for today, folks!

Until next time, WRITE ON!

Jo