How can we use AI tools to enhance and improve our creative process? How can we double down on being human by writing what we are passionate about, while still using generative AI to help fulfil our creative vision? Rachelle Ayala gives her thoughts in this episode.

<figure class="aligncenter"></figure>

Today's show is sponsored by my patrons! Join my community and get access to extra videos on writing craft, author business, AI and behind the scenes info, plus an extra Q&A show a month where I answer Patron questions. It's about the same as a black coffee a month! Join the community at Patreon.com/thecreativepenn

<figure class="aligncenter size-large is-resized"></figure>

Rachelle Ayala is the multi-award-winning USA Today bestselling author of playful and passionate romances with a twist. She also has a series of books for authors, including Write with AI, An AI Author's Journal, and AI Fiction Mastery.

You can listen above or on your favorite podcast app or read the notes and links below. Here are the highlights and the full transcript is below. 

Show Notes

• Understanding generative AI tools as a non-technical person





• How the creative process can work with AI tools and why it's always changing





• Using AI tools as a collaborative discovery process, and why it's all about your creative vision and author voice. For more on this, check out my AI-Assisted Artisan Author episode





• Aspects of copyright





• Staying focused on writing as new AI technology emerges, and why you need to double down on being human

You can find Rachelle at RachelleAyala.net.

Transcript of Interview with Rachelle Ayala

Joanna: Rachelle Ayala is the multi-award-winning USA Today bestselling author of playful and passionate romances with a twist. She also has a series of books for authors, including Write with AI, An AI Author's Journal, and AI Fiction Mastery. So welcome to the show, Rachelle.

Rachelle: Thank you, Joanna. Thank you for having me.

Joanna: I'm super excited to talk to you. As I was telling you, I have the ebook and the print edition of AI Fiction Mastery because I think you put things so well in your writing. Before we get into it—

Tell us a bit more about you and your background in technology and writing.

Rachelle: Okay, sure. I was a math major, and I actually have a PhD in applied math. So you would think that's kind of the farthest thing from writing.

I got into parallel computing back in the 80s. Then in the early 90s, neural networks, where we were basically trying to recognize handwritten characters between zero to nine. So that was quite interesting and fascinating.

So I basically worked in software development and network management until 2011. Then I got into writing. So romance writing was my gig, and I liked dealing with feelings and happy endings.

Joanna: Well, I love that, going from maths and neural networks into romance. You do explain a lot of the stuff behind AI in your books, which I think is really good. You're used to writing for normal people, so I don't find your writing technical at all.

Do you think people who are not very technical are struggling with this AI world at the moment?

Rachelle: I don't even think you need to be technical to understand AI because—well, there's different types of AI, but we're talking about large language models for writing.

So there's other AI systems like expert systems, machine learning, and people have been using that. They don't even know it, but they've been using it under the hood.

The AI we're talking about, large language models, ChatGPT was one of the first ones that most people became aware of. So GPT is a Generative Pretrained Transformer.

You could think of it as a word slot machine, where you could think of all these slots. So when you write a prompt, then the AI will look at the words that are in there, and then try to predict the best word that comes after.

Let's say, we say Monday, Wednesday, and most people will say Friday because that's the next word that you think of. Or if you say Monday, Tuesday, most people will say Wednesday.

So what the AI does is it was trained on reading, I think somebody said between half a trillion or trillion pieces of text. When an AI is trained, it's not reading a book like we do, where we read it from beginning to end.

So think of if you cut a piece of newspaper into a strip or a square, and then it's got all these words that are in there, and it's looking for words, associations, and patterns. So it'll say, oh, this word goes with that word, and those words go together.

So it could take a word like, say, “bark.” If it sees dog in the other slots, it's going to most likely come out with “woof,” but if it sees trees in the other slot, then it might say, “the bark is wrinkly or hard,” and it's thinking of a tree bark.

So that's how it is able to create words, and that's why you think it's intelligent, because it understands the context. It does so with these huge, huge context windows. So I don't want to get too technical, but a context window is how many words can it keep in its memory.

So it can look at all these associations and how those words go together, so it can best predict the next word that comes out of this word slot machine, so to say. It doesn't remember anything.

Joanna: It's interesting. You mentioned words there, like associations and patterns. I feel like the big misunderstanding with large language models is that some authors think that it's more like a database, where all these “stolen” books are sitting in a big database.

Then if you query it, it will pull out exact chunks from other people's books and use them. So you're always going to plagiarize or you're always going to be “stealing.” Like you and I hear these words a lot from authors who are really just starting out.

Can you explain why it's not a database?

Rachelle: Well, databases are storage. So if you query at a database, it pulls out exactly what's in there. I mean, this is like your social security number. It's not going to get it wrong, it's going to pull it right out. Your birthdate, if it's entered in correctly, it will pull it out.

Everybody knows that AI doesn't get things correct, or it doesn't get things exact. If you prompt it twice with the same prompt, like, say, “Write me a story about a road runner who is sick,” or something, it's going to write you something different.

Even that, if you think about how they trained AI, they trained it by inputting all these words that are associated together. Then they adjusted the weights of how these words are more likely to be with those words.

They're not retaining the words, the words are thrown away. The only thing it keeps is the weight.

So sort of like when you read a book, unless you have a photographic memory, you cannot recall that book, but you can recall the concepts because you have made associations between what you read and it communicated to you these ideas.

In fact, people say our memories are not like videos, our memories are actually assembled whenever we're recalling something. So we are making things up on the fly, based on all the associations that we've had in our lives. Similarly, that's how AI LLM really is making up things.

So when people say it lies to you, it's like, no. It's actually just making things up. You gave it a prompt that said, like, “Say happy birthday to me,” and it just keeps going with that.

There's also something called a temperature knob where you could basically increase the randomness, beca