Meet Your Hosts

Kathi Lipp

Well, hey, friends, welcome to writing at the Red House podcast, where we gather at the table to break bread and tell tales with some of our favorite writers and creators who share their wisdom to help us all share our story. And I am back here to talk not all things with AI. We’re going to be very specific today, but with my in house AI genius, it’s Roger Lipp. Hey, Rog.

Roger Lipp00:00:39 – 00:00:41

Hey. Good to be here.

Kathi Lipp00:00:41 – 00:00:50

Well, it’s, you know, it’s worked out really well for me that in your day job, you’re learning all about AI and can bring it home so that we can talk about this.

Roger Lipp00:00:50 – 00:00:56

And, guys, let me tell you, it’s working out for them, too, because I learned it for here and I use it there. Yeah.

Kathi Lipp00:00:56 – 00:01:29

Yes, yes, yes, yes. It’s so true. Right? And so I wanted to talk today about a very thin slice of AI that we’ve both recently discovered. Now, I use AI in my writing. I use it for outlining my speaking trainings and things like that. I use it for editing in my writing before I send it to a human editor. I use it for a lot of different things. And I have been using chat GPT for about, about a year now.

Kathi Lipp00:01:29 – 00:02:01

And that has all changed in the last week. And we’ve gone over to Claude, and I wanted to update our listeners to say, hey, here’s why we’re doing it. And if you’ve heard us talk about chat GPT, why you might want to consider Claw. So the first thing I want to talk about is we’re going to talk about why we’re using Claude, but just a couple of interesting things. There’s knowledge cutoff date, so could you explain what that means?

Roger Lipp00:02:01 – 00:02:51

Sure. When these language models, Claude or chatgpt, are created, they’re created based on a snapshot of data off various sources. And how current is that information? So chat GPT, for example, it’s 3.5 version, was cut off September 2021, somewhere around there, and then four. I think they started with that, but then pushed it a little bit later. Claude is pretty current. It’s up to date as of August of 2023. So you can ask Claude about events that happened prior to August of 2023, and it has some knowledge about them. So that’s kind of what they mean by the knowledge cutoff.

Roger Lipp00:02:51 – 00:03:02

Both of these tools can go out and search the web as well. You kind of have to tell it to do that, but it’s not quite as smart when it’s doing that. Yeah.

Kathi Lipp00:03:02 – 00:03:57

Okay. So that cutoff date is important because, say, like you wanted to talk about health and the royal family. It’s gonna know about something, you know, it’s gonna know about the death of the queen and things like that. I think. I can’t remember when she died, but it’s not gonna know about Princess Catherine’s cancer diagnosis and things like that. And so you want to make sure you know, but it will know about other things that have happened between 2021 and 2023 that chat GPT probably doesn’t know about. And so I don’t really use chat GPT or claude for, like, deep research. That’s my job.

Kathi Lipp00:03:57 – 00:04:10

But if you’re looking for outlines and places to go and things like that, I think that it’s really important to know how deep its knowledge is and where it’s getting its knowledge from. To me, that feels important.

Roger Lipp00:04:10 – 00:04:11

Yeah.

Kathi Lipp00:04:12 – 00:04:39

Also, I am learning today about something I had never really understood, constitutional AI. And so I’m seeing it defined as a set of tools, set of tools and techniques that ensure AI closely aligns with human values, ethics, and legal framework. So I don’t. Can you, can you just put that into. I’m a writer, not a techie kind of work words?

Roger Lipp00:04:39 – 00:05:17

Yeah. This is where chat GPT and Claude kind of diverged a little bit. How do you keep the AI from going off the rails and doing things that we, as society would say, okay, we don’t really want it to do that. We don’t want it to behave in that way. So for Claude, what they’ve done is created a constitution for it to follow. And it’s pretty funny how they created this thing. It’s really just a document that they always feed to Claude. Here, follow these rules just like it reads our prompts and our books.

Roger Lipp00:05:17 – 00:06:18

If we give it to them, constitutional AI means that this is a way of keeping the AI’s behaving, kind of how we want them to behave and not going off the rails and not answering bad questions like how to make nuclear bombs or something, all kinds of rules that we as a society would prefer that our AI operates by. And so what they’ve done to get started, because this is all new stuff, is all new field, right? They can’t start with a perfect answer. They just have to start with something that’s close and then improve upon it. It’s pretty funny. The history of Claude’s constitution. The rumor is that they started with documents like the UN charter, and they put that in there. They took the apple terms of service and put that in there. So they started with a number of different things.

Roger Lipp00:06:18 – 00:06:32

That talk about how we as people should behave. And they just collected them all together. And that was the starting point for Claude’s constitution in its constitutional AI. I’m sure it’s evolved since then, but.

Kathi Lipp00:06:32 – 00:07:04

That’S, and the reason why this is important for writers is that, you know, we’re kind of looking at Claude versus chat GPT here. Chat GPT was all the rage for a long time, and now we’re moving over to recommending Claude, and I’m kind of guessing six months from now, we may be back to chat GPT seven, or whatever it’s going to be. By all accounts, Claude is more ethical and less biased because of what it was trained on. Is that a good way of saying it?

Roger Lipp00:07:04 – 00:07:57

Yeah. Ethics were at the center of the team that built Claude. They were very, very concerned. In fact, they originally worked for OpenAI, the makers of chat GPT, and kind of did this. They spun off, they quit and created this company called Anthropic because they were concerned about ethics more so than they saw in the OpenAI group. So it’s like privacy is one of the big things. If you give, let’s say you cut and paste a chapter of your book into chat GPT without being careful, without setting things up in advance, chat GPT will train, it’s called training. Will learn that text that you fed it so it can use that the next time somebody else asks a question.

Roger Lipp00:07:57 – 00:08:17

And obviously that has privacy concerns and copyright concerns, those kinds of things. You can turn off the behavior, but you have to go to a setting, whereas in Claude, it’s just off. You can’t even turn it on. It’s not even an option. So Claude is more privacy focused out of the.

Kathi Lipp00:08:17 – 00:08:29

Okay, so we’ve talked about, I want to get down to the nitty gritty of why we’re using Claude. So talk about our little experiment with the ME project.

Roger Lipp00:08:30 – 00:08:36

Oh, my goodness. So we’ve recently bought back the rights to the ME project from the publisher.

Kathi Lipp00:08:36 – 00:08:40

That’s one of my books that Kathy wants to do ten years ago.

Roger Lipp00:08:40 – 00:09:24

It’s a great book. It’s actually one of your favorite books, I think, that you’ve written, and we want to do something, some kind of a revision to it, make some edits. I don’t know. We don’t know exactly what we want to do with it yet, but we’re just getting started. So this book is about 50,000 words, maybe 60,000 words. And because Claude can, it’s called a context. It is able to hold 150,000 words. At the same time, in its memory to kind of process, so it can see the connections from the beginning of your book to the end of the book, and it knows every word in between all at the same time.

Roger Lipp00:09:24 – 00:09:53

And this is very different than how chat GPT would handle this much data. What chat GPT would do is what’s called chunking. It would chop it up into little tiny sections. Each one is small. And if you ask a question, it can answer a question about this junk. This junk, this junk, this. But it can’t really see the overall flow. And as you start talking to it, after a while, it kind of loses its place because you’re dealing with so much data, and it forgets what it’s already said.

Roger Lipp00:09:54 – 00:10:36

It’s kind of like me. It feels like it’s getting old and tired. Whereas cloud can keep that conversation going for a longer time and not forget about things because it has this huge context window. So we handed it the whole text, the word document for the Me project, and asked it to do what three things. We asked it for an outline of the book. We asked it for key concepts from the book, and we asked it for its ideas on how to bring the book more up to date. And we did all that in one prompt, and it really did came back with wonderful answers.

Kathi Lipp00:10:36 – 00:11:25

I would say it was like handing it off to ten people who I trust and say, can you review this? Ask them all the same questions, and get probably 85% to 90% of that answer that Claude gave us. And, I mean, what a gift, right? So I still have to go write all that stuff. It’s not. I’m not asking it to write it for me, but I’m asking it, like, where are the holes for 2023? 2024? And of course, I’ll have more insight into that, but it’s given me such a place to start. It is the opposite of the blank page syndrome. It’s saying, go do this. You’re like, okay, I want to do nine out of ten of those things. That 10th one doesn’t really resonate with me.

Kathi Lipp00:11:25 – 00:12:37

But then there are a couple other things that, as I’m developing that content that I’m like, oh, yes, this would be good to put in there as well. A couple of other things that I just know I’m going to be using Claude for is to put in the text of my book and say, okay, I want to co write a Bible study to go with this book. Can you give me an outline for each chapter that a Bible study group could follow along with? If I was writing a work of fiction, I would say. I would like to write a book club guide to go with this. Can you give me discussion topics for each chapter? Because here’s what Claude does. It gives you an outsider’s perspective because it’s just encountering this for the first time and maybe the second time it’s not encountered. It hasn’t lived with the text like you have for what, three years? And so your, what’s that called? Not gaps of knowledge. Your, when you know things about your text in your brain even though they’re not on the page.

Kathi Lipp00:12:38 – 00:13:07

Oh, curse of knowledge. That’s it. The curse of knowledge. It, it doesn’t have the curse of knowledge. It’s not filling in things like your brain does. One other thing. I mean, and we can use this for a lot of different things, but one other thing that I like to do with Claude is when I write a, well, what was the last thing you asked it to do? You asked it for social media prompts about the book. Talk about that.

Roger Lipp00:13:07 – 00:13:21

Actually, yeah, we did three other prompts. We asked it for line edits for the book and it gave us some beautiful ones. We asked it for what’s the other term?

Kathi Lipp00:13:22 – 00:13:30

Some people would call them global edits, conceptual edits. Oh, gosh, I can’t think of the term right now.

Roger Lipp00:13:32 – 00:13:34

I’m pulling it up real quick.

Kathi Lipp00:13:34 – 00:14:03

It can do line by line edits, but it can also do a more holistic, like, does this make sense? Am I making progress on this? You know, in a way? Are there gaps of knowledge? Like you, you reference. Yes, thank you. You reference uncle Ted in chapter seven, but you never introduced him in chapter, chapter two. And so, like, what, what’s going on there? And that is a developmental edit.

Roger Lipp00:14:03 – 00:14:48

Right. And then we asked it to generate social media posts, promote promoting the book, and we asked it to generate those in Kathy’s voice as represented in the book. Did a pretty good job of that. You know, obviously we’re not going to cut and paste and put those directly out. We’re going to review those and make edits of our own, that kind of thing. But, yeah, did a really good job. Another thing we did was, hey, based on the writing you see in the book, could you generate instructions that our social media team can give to an AI to create copy using Kathy’s voice? And it did that. And it, yeah.

Roger Lipp00:14:48 – 00:15:11

And when, when you then go back later on and say, hey, I want to create a promotion for clutter free for life in using Kathy’s voice. Here’s our target audience. Here’s Kathy’s voice. Here are the details of the program. It does a really good job at getting a first pass of those social media guys.

Kathi Lipp00:15:11 – 00:15:43

We’re not cutting and pasting anything. We’re getting in cloth, but we are using it as the jumping off point or maybe making a few edits if it’s closer than what we expected. Roger, I want you to talk about the ray experience. This is an internal term. If you talk about it with other people, they’re going to have no idea what you’re talking about. But Roger and I know what it means. And talk about a project you’ve been working on recently for our team and how you’ve. Claude has simulated a ray experience for you.

Roger Lipp00:15:43 – 00:16:31

Yeah, yeah. So let me set the context here before I talk about Ray. The project we’re working on is migrating our team off of Trello onto Microsoft tools internally. That’s really where we live, is in the Microsoft world. But some of our team is still using Trello, and we love Trello, but for a team our size, Trello has more cost to it, and we just need to save those costs. So we have to figure out how to get off of Trello. And I have been banging my head against the wall trying to come up with a process that kind of looks like Trello, but isn’t. And I just wasn’t coming up with anything.

Roger Lipp00:16:31 – 00:17:33

And it occurred to me that my day job for a Fortune 500 company years ago, I worked with somebody named Ray. And Ray and I would bounce ideas off of each other constantly. We actually had them over to our house, and we would bounce ideas off of one another there, and we would always come out with something that was better than either of us came into a meeting with, because we would just feed off of each other and the quality of our ideas would improve. And I’ve recently missed that at my day job. And as I was working through this problem for our business, I found using Claude brought some of that feeling back. And the reason is that Claude has this huge context. Remember, we were talking about the context window, 150,000 words. I could keep a chat going for a long time, and Claude doesn’t suddenly become senile.

Roger Lipp00:17:33 – 00:18:20

It remembers the whole thread of the conversation. And so that allowed me to go back and forth for two weeks with Claude. I did start restart the conversation a couple of times, but for two weeks, I worked with Claude trying to come up with an idea of how to do this migration. And in the end, I think we’ve got an idea that is better than what I was bringing into the conversation, and it’s better than what Claude originally started with in the conversation. So we had, over that period of time, we went back and forth. I kept trying things and, okay, I don’t like this about that. And then go back, okay, what about this? And through that process, I feel like we’ve come up with something that’s really solid now. We’ll see how it goes.

Roger Lipp00:18:20 – 00:18:29

Proof is in the pudding, I suppose, but it was refreshing, and I found it enjoyable, actually, to go back and forth.

Kathi Lipp00:18:29 – 00:19:36

It’s so hard as writers, we’re often having to pull things out for social media, for how we run our businesses, for our writing out of thin air. And, I mean, I can collaborate with friends, but they’re writing their own stuff. They don’t necessarily have the time or the energy to be brainstorming with me. And so to have something to brainstorm with, to bounce off ideas off, or did you think about or what, you know, what a gift. So Claude is great for not only overcoming writer’s block, but creator’s block, you know, and that’s what you were having to come up with. It’s a great place to start with research and gather information. I was talking with Susie Florey, and she’s working on her doctorate, and she’s working with a very specific tool that is for academic research. And she said, while using this tool, it gave me ideas that maybe I would have had.

Kathi Lipp00:19:37 – 00:20:29

I’m thinking, I’m not quoting her accurately. She maybe could have come up with in six months, but now she can start working on. And because they were adjacent to what she was researching and what a gift and pointed her towards, it was more about the resources. Here are resources that she never would have found through, even like the academic Google that could start pointing her in directions that were surprising, but exactly what you need. I think of, like, of these tools as, you know, when you’re. When you’re googling, you’re looking for one answer. When you’re clotting or whatever we’re going to call it, it’s giving you different things to think about. I feel like it engages my brain so much more and ups my game in so many new ways.

Roger Lipp00:20:29 – 00:20:44

It becomes a dialogue where you can curate ideas, uh, that are collected through that dialogue. You don’t have to use all of them, you don’t have to use any of them, but, you know, it’s a very different experience.

Kathi Lipp00:20:44 – 00:21:28

So conducting research, collaboration and editing, you know, especially if you can’t afford to have some, a human edit right now, and you’re like, okay, this has got to be 90%. Get it to 90%. You can do that with these tools and that. It’s such a gift. Roger, anything else? We really want to emphasize that we are using AI as a tool to enhance, not replace, human writing skills. Because it’s your creativity. It’s your God given spark. It’s what the Holy Spirit has put inside of you.

Kathi Lipp00:21:28 – 00:22:02

I don’t care how far we get in AI, you’re never going to be replaced with that. And so we’re going to be. We’re going to be on here a lot, talking about AI in the writing game. You know, next time I want to talk about how I use AI to take care of some of the mundane tasks so we can get to the writing. But, so stick around. Share this episode with somebody who is also excited and interested about AI because we would love to hear your questions. If you have questions, please send them to us.

Roger Lipp00:22:02 – 00:22:34

So one thing I would like to just throw in as we’re wrapping up, Claude has a very competent free version. So you can get started with Claude for free. No, it doesn’t cost you. Eventually it’s free. And even at the free level, you still get this huge context window, the 150,000 words of context. So it’s really, really smart. Now, the paid version is smarter. That’s the difference between the two and the paid version.

Roger Lipp00:22:34 – 00:22:48

You don’t get blocked out because the systems are too busy. So two advantages of going to the paid it’s only like $20 a month. Well worth it. But I just wanted to point out, that’s beautiful. Or there is a free version of Claude.

Kathi Lipp00:22:48 – 00:23:00

Roger, thanks so much for hanging out with me today. And friends, thank you so much for hanging out. You’ve been listening to the writing at the Red House podcast. I’m Kathy Liff. Now go write your only you God story.