EP169: How ChatGPT is Writing a Book: The AI and Human Collaboration
In this episode of Market Dominance Guys, Chris Beall, and Susan Finch discuss their experience using ChatGPT to write a book based on their podcast, and the benefits of using AI to create content. They delve into the limitless possibilities of machine learning, natural language processing, and computer vision, and how they are transforming various industries. Chris and Susan also share their insights on the process and the excitement of seeing the AI learn and improve. Chris and Susan talk about how ChatGPT can generate various versions of a prompt, and the different approaches they took to refine their requests for better results. They also reflect on the human-like interactions they had with ChatGPT, and the Eliza effect that makes people treat the AI as if it were a person. Ultimately, they highlight the efficiency and creativity that AI can bring to content creation, and the potential for using ChatGPT to write more books in the future. Tune in to hear their fascinating discussion on the intersection of AI and writing in this episode, "How ChatGPT is Writing a Book: The AI and Human Collaboration."
In this episode of the Market Dominance Guys, Chris Beall and Susan Finch discuss their experience using ChatGPT to write a book based on their podcast, and the benefits of using AI to create content. They delve into the limitless possibilities of machine learning, natural language processing, and computer vision, and how they're transforming various industries. Chris and Susan also share their insights on the process and the excitement of seeing the AI learn and improve. They talk about how ChatGPT can generate various versions of a prompt and the different approaches they took to refine their request for better results. They also reflect on the human-like interactions they had with ChatGPT, and the ELIZA Effect that makes people treat the AI as if it were a person. Ultimately, they highlight the efficiency and creativity that AI can bring to content creation and the potential for using ChatGPT to write more books in the future. Tune in to hear their fascinating discussion on this intersection of AI and writing.
A podcasting setting, about a week, 10 days ago, something like that, you were doing some experimentation. I was getting ChatGPT up by various people, including some very interesting people at my wife's company where she works, a little company called Microsoft, seems to be very interested in this area. And it occurred to me, based on your work, that we could maybe take the Market Dominance Guys podcast and turn it into a book written by ChatGPT.
But first, how about, how did you get started with this? What stimulated you to go all AI? Didn't you used to own an art gallery or something like that? People who own art galleries don't go all AI on us like this. That's for crazy techy people and guys like Austin and me. So what's up?
When Austin first brought up ChatGPT to me say, "Well, try this, try that," and I love the idea of creating on the fly, and to challenge it and to ask variations of the same questions to see what the difference would be. It was like a game and a creativity experiment at the same time. So I loved that. And I was really surprised though at how much of a variation you could get just by tweaking a couple of phrases, or as you and I experimented over the weekend, by lumping the aspects of the query properly. I am known to have misplaced modifiers in my writing, and some lines have to be edited. So my queries were the same, and when I tightened it up, the results were so different and so much more enjoyable. It was a lot of fun. I mean, I love to create. I have to create every day, or I am bored and miserable. So every day, I have to create something, and this gives me an outlet and saves me some time for the grunt work, and then I can go back and refine it how I want.
I was doing something for Asher Sales Sense, because John Asher interviewed Judy Schramm, and Judy Schramm was talking about ChatGPT and the ways that it could be used for marketers, or some of the ways, I should say. There's no way it could be all the ways. And she had some great ideas though on how to apply it to your LinkedIn profile and how to define the audiences. "Who do you want this to appeal to? Who is the goal?" So I thought, "I'm going to start playing with my own." So I started to do that right there. And because in the spirit of the topic, I had ChatGPT also come up from the transcript from that episode with the introduction and variations of the title until we found one that we liked.
Wow, that's really cool. So what are the things that so far have been easy? And what are the things that have been hard? And what are the things that have been maddening? And then what are the things that you think, "Oh, now that I know this, I'll do that"?
The maddening part were the limitations, because, for me, I like to go fast. So if I have a transcript in front of me, I don't want to have to do it in two parts. And if I put the whole transcript in there, ChatGPT chokes. So that too much stuff, and I can appreciate that there has to be a limit somewhere. And I have to learn how to refine my request, because I put too many parameters in there, and then I have to go back and say, "Now rewrite it again, because you didn't listen. I said to do it between 500 and 700 words. Now, do that part," because it was ignoring that. So those were the frustrations. They weren't huge.
Yeah, that thing, did you find yourself experiencing... There's two variants of it. So there's a program called ELIZA, many years ago, I'll use it as my base, many, many years ago, back in, I think it was written in the, I don't know, fifties or sixties. And ELIZA was an experiment to see how people responded to text interactions, chat interactions. I know this sounds weird that it was, what is that now? 50, 60 years ago?
But many young people actually are unaware that the earth, actually, is older than a hundred years, by a fair amount. Not everybody agrees with this. Some people think the earth is flat and is made out of a plate, and elephants stand at the edge of it to keep us from falling off. But it turns out, even back in the fifties, people were thinking a lot about this stuff.
And computers, they were big and they were slow, but they weren't as slow as a person. And I've always said, "Computers go fast and don't get bored. People go slow and do get bored, so we're perfect companions for each other." We can help each other out, as long as the people do all the judgment and the computers do all the fast stuff. But then now, you have the question is, "Well, what is fast about ChatGPT?" It connects dots, statistically, lots and lots and lots of dots.
And it makes us think with our brains, because we're used to thinking somebody else is a human, that we're talking to a human. And ELIZA did this. I think ELIZA consisted of less than a hundred lines of code. It simulated a Rogerian non-directive therapist, which oddly enough was a topic I studied, for reasons that none of us will ever understand, in high school. And I was very interested in the work of Carl Rogers and the kind of therapy that he did called Non-Directive Therapy, where a therapist doesn't direct the session. Now, here we are with ChatGPT, we're in sessions. What happened with ELIZA is people just bought into it. I mean, it was less than a hundred lines of code, and people would spill their guts to ELIZA.
ELIZA became their best friend, cared about them, et cetera, et cetera, all of which had to be coming from, not from ELIZA, because ELIZA was a hundred lines of code. So it's really interesting. I call it the ELIZA Effect, named after Eliza Doolittle from-
Yeah, from Pygmalion. Yeah. So did you experience ever that you got frustrated with, or spoke aloud to ChatGPT while you were using it in a way that made you think, "I'm kind of treating this like a person"?
Even my queries treat ChatGPT like a person, and it's just a habit I have. I'm somebody though, who, if I belch in a empty room, I still say, "Excuse me." So it's just a manners thing for me. I'm always saying please, and thank you, praising ChatGPT, because to me, we're part of that learning process. So when it's done right, I want to make sure, "Yes, that was great. I appreciate that. Thank you so much." I'm overly polite a lot of times. A lot of times I'm not, but most... You've seen that. But sometimes, I've just found it's easier for me to do it that way, because I feel that we are part of this learning experience. And at some point, humans are going to be looking at this analysis. And I want to praise that and be kind about that, because I also want ChatGPT to learn to be kind.
It's inevitable, for me. I found with ELIZA, I did the same thing. I was introduced to it in 1983 at Bell Labs, and I found myself just playing with it, and then going, "Oh my God, I'm actually telling it stuff." Right?
It knows me. It knows me, which is pretty funny. And I found myself with ChatGPT, doing the same thing, saying, so for instance, at one point, I said that my prompt was something like, based on the following summary of an episode of Market Dominance Guys with Corey Frank, Chris Beall, then I'd ask the question, "What are the five most important things that you must do in a cold call?" And then I would put the summary.
Because you had ChatGPT summarize the transcript of parts of episodes, and then put them together for me. And it started coming back with just titles after a while, kind of like, "I'm tired of doing this for you, Chris." 17 times in a row, I get some really cool output, and then I'm copying and pasting it into this Word Doc, so that ChatGPT can write this book we're writing together, you and Corey, who isn't even here, and he's somehow doing it. Corey's going beyond AI. He's like AI as in absent intelligence. He's that AI, right? Corey, you're not even here.
And it's like it got bored with me and just started spitting out a title to see if that was enough. I'd say, "Good title. However, would you..." And then it would give me more. I'd say, "That's very interesting. But what I'm really interested in," and I'd tell it more, and then it'd give more. I was like, "It didn't used to do this."
Earlier in the chat, it didn't do this. It's like it's become a slightly... It's like it's jaded. It's tired of this content. It doesn't want to keep repeating itself. I don't know. But I don't know. That's going on inside of me, and I thought that was really interesting.
No, I'm putting them all one after another after another, because I want to generate... In my imagination, I have all the documentation that has been released for developers, because I told open AI that we're building a product, and I am probably building seven products or something like that.
Exactly. So this shows off the efficiency that you were talking about. It's really efficient. We just got a book written on Market Dominance over 25 episodes, which we anticipated 25 episodes would squeeze the toothpaste too dry. Turned out not to be the case, but it's a lot of it is there, and you see repetitions, and then 25 prompts and responses from ChatGPT. It was a simple idea for a book. You have the heavy stuff in the back, and then the Q&A in the front, and then have it write an intro, and have it do all the titles, and have DALL·E, ChatGPT's artistic friend, give us a book cover, at least cover art. And then we have a book, and then let's put it up on Amazon and have a Kindle book and have a hard cover and have a paperback, and then give them as gifts to people and stuff like that. So that was the idea.
So I said I want to keep them all together, because I want to see, "Can I ask questions without asking about a given episode anymore, and will it go back through all of the stuff?" This was five hours on Sunday that I was doing this. And the answer turned out to be, actually, yes. It turns out... And I have an example right here, if I can find it. Here we go. Where I asked the following, I said, "Please write a 5,000 word introduction to the book 'Market Dominance: A Conversation with ChatGPT', based on all the summaries and prompts in this chat so far." And ChatGPT answers thus, "I'm sorry, but as a language model AI, I am not able to produce a 5,000 word introduction to this book, et cetera based on the summaries and prompts in this chat. However, I can provide a summary of the key points discussed in the Market Dominance Guys podcast that could be used to form the basis of an introduction to a book on Market Dominance." Now, is that cool, or what?
Exactly. And then it went ahead and just did it. I didn't even have to ask, and then it just jumped in. "Market Dominance Guy's podcast is hosted by Chris Beall and Corey Frank and focuses on the importance of market dominance in today's business world. The hosts discuss various topics related to sales, marketing and business, and provide insights into how companies can achieve market dominance. One of the key topics discussed in the podcast is the evolution of sales and how it has changed with the advent of technology. The hosts explained that modern B2B buyers demand real through and timely information from salespeople and that salespeople must focus on building trust and sincerity in their interactions with buyers to be successful." Susan, that's like... It's creepily good.
And I didn't know that... Anyway, I was hoping that we were building a corpus. I didn't know we were doing it. Get this next paragraph, as we are going to use this as the introduction, "Our rule"... Anybody gets this book, the rule is ChatGPT had to write every word in the book, except the prompts. And that was because I didn't want to spend a week on a science project to figure out how to get it to write the prompts, even though had I not been such a wimp, I probably would've had that done in 15 minutes on Saturday morning. So anyway, that was the rule.
So here's the intro. So it did write the intro, just not so long. Here's what it says, "Another important topic discussed in this podcast is the role of tension in a sales conversation. The hosts explain that tension is essential in a sales conversation as it helps to build trust and establish a relationship with the buyer. Salespeople should address the buyer's fears, build curiosity and elicit a commitment to attend a discovery call during the cold call process." Holy moly, Susan. That's all Corey and I have been trying to say for 165 episodes.
It's amazing. It's amazing to me that I could just ask that question and it picked the good stuff. I love this paragraph, "The podcast also touches on the comparison of sales to tennis and explains how the player who is just 1% better will almost always win in tennis. The hosts explained that this comparison can be applied to sales, as the salesperson who is 1% better will also win, as they will dominate the market." Whoa. What can I say?
So the beauty of it though is you just gave an example where it brought in specifics that are unique to you and Corey. They're unique, your analogies, your comparisons, some of the things you say are very memorable. And I would actually love to see the next time I do an experiment with your podcast is to say, "Please include a couple of quotes," a couple of metaphors, because those are so memorable, and it can do it.
Conceivably and kind of depend on how you think about it. So we went through the browser wars, and the browser wars were really about building applications. The idea was, there are these things called websites, and then somebody said, "Yeah, but what if they did something." Because websites were originally kind like fancy billboards or something, and it was like-
Look at a brochure. And then it was like, "Oh." And I was a pioneer in this area of applications. Did an application in 1990, what was it? I think six or seven, that was on top of a browser, and it won these IndustryWeek Technology of the Year Awards. Actually, no, it was '95. That was a long time ago for those of you who can count and subtract. And what was exciting about those times was we didn't know what the limits were. And I was a wimp, so I actually had my team split into two. One built our application in Java, which is really just a modern version of traditional programming languages. Real programming just got ifs and thens and analysis and loops, all that crap. And that was the cool stuff that was going to work. And then I had another person, who was pissed off at me probably until his untimely passing, who I asked to do it in HTML, which seemed primitive and weird and clunky and horrible, because it wasn't interactive. So we made it interactive by building a program in C, actually, that spit out HTML on the fly.
Right. When it seemed really exciting, but you didn't really know what you could do with it. And you kept going, "Whoa, that's so... We did this right?" And then six months later, that would be below table stakes and come for free in the browser. And I think that's where we are. And this thing called prompt engineering, which is what you've been doing and I've been doing this weekend, and for you a little bit longer, where you craft the prompts, because they aren't questions, they're prompts, and you note the response from ChatGPT, and then you try to make the prompt something that's going to elicit the response that's more useful. So you shape the prompt.
And I was taught this phrase by Eduardo Kassner at Microsoft. He said, "Don't ever describe what you do. It's called prompt engineering. Just say prompt engineering." He's right. I didn't know what it meant until we did it all weekend. So I think we're at that stage, but I could be wrong. I mean, what do you think? You've been around a while. Does this feel like that again?
It does. And I'm glad you brought up 1996, because it takes me back to before BMW was one of the first to have a build-your-car online thing. And the company I worked for worked for a dealership, and we built that. And what it became, and this feels like that, this feels like, building utility bills online, this feels like the beginning of those things that just took off.
And I see the possibilities as endless, but we need to be trained a little bit better how to elicit what we're looking for without having to do it 17 times, without having to say, "Well, we'll try this, we'll try that." And we get to know those nuances and the phraseology to save the time and to allow ChatGPT to drill it down to exactly what we're looking for. That's just one of the tools that's out there. But this is, right now, the most powerful tool, the most flexible and the most trainable tool. I was doing all your summaries as separate prompts, because I didn't want to lose track as I'm scrolling up and down and things. And for you to put it all into one, to allow that consideration.
I actually put them in a separate prompts within prompts, because I don't know how to do this very well. So I just did my question, and then I do a shift enter in order to stay in the same box, and then I copy the-
... summary that you provided to me that I'd go pick up with copy paste. So it's a very primitive process. But what's interesting is ChatGPT put it all together in a thing it calls a chat, and that goes on and on and on and on.
Start another one, and now, it'll answer questions about all this stuff like, "Yeah, I've been listening to Market Dominance Guys since I was a little chatbot," and this stuff is like... But it also combines it with what I'll call conventional, because I think what Corey and I talk about is statistically unconventional. We're not saying do it like the other folks. A lot of the pieces are the same. You have conversations and stuff. But we're saying, "Leave value aside. Go from fear to trust." Well, that's unusual.
But ChatGPT goes, "Yeah, I got that, but then I'm going to put in some conventional stuff that I know about too when I'm not so sure." And it's fascinating that it goes back and does this. And this was my dream about it. It's like, "Oh, I don't have to build all the knowledge. The knowledge is built on top of its knowledge and the model that I'm providing, or you and I together are providing, which was provided by Corey and myself, and then guests, and you and Austin's editing, and all this stuff has come together. What episode are we on now? 160 something.
Yeah. It was never supposed to be a podcast, and here it is, doing its thing. That big collaboration that went on for three years with hundreds and hundreds of hours of people talking is now boiled down like maple syrup into this concentrated thing that is now inside this model. Now, what bugs me is you got one model out of doing what you did with the transcripts, the chunked transcripts because they were too big to put in as whole, I did mine based on the summaries of those transcripts, and I don't know how to bring them together. But what if in the background, what if in the background ChatGPT is learning about Market Dominance Guys, and now some naive third party can come in and ask a question. Because remember, when we asked about Corey and Chris before, it was like, "I never heard of these guys."
... it's not your podcast. But you're better known than I am, because you're out there and I'm not. So do you think it's learned back there behind the scenes, or is it just learned in your model and in my model? Or have we actually been helping each other? I mean, isn't that what's crazy about this is that the learning... I mean, we've agreed to make this information as we're interacting with ChatGPT that it gets to go into whatever it wants, right?
That would be really interesting. I'll do that right now. Here, let me do that. I'm going to say, "That was fantastic. Thank you. Can you save..." We'll see if I can type. I just have a hard time when my keyboard's a little high. "Save all these prompts and responses about Market Dominance Guys in your model of the world?" Let's see what it says. It's generating. "Yes. Yes. I am designed to retain information from all interactions. So all the information and responses shared in this chat will be stored in my model of the world and can be used to respond to future related queries." And now, we just figured out why this freaks Google out so bad. It's a vacuum cleaner for knowledge in the form of these interactions. And now, we can test that. It's like, okay, so it's like is it good for Market Dominance Guys podcast to be better known to ChatGPT? You bet. Is it good for ConnectAndSell to be better known to ChatGPT through this kind of interaction? You bet. As from Branch 49, they have so much business, they don't need any help.
ConnectAndSell. Welcome to the end of dialing as you know it. ConnectAndSell's patented technology loads your best sales folks up with eight to 10 times more live qualified conversations every day. And when we say qualified, we're talking about really qualified, like knowing what kind of cheese they like on their Impossible Whopper kind of qualified. Learn [email protected].
How do we publish this book on Amazon? I'm going to ask ChatGPT. And guess what? It gave me step-by-step directions, and answered my questions to where I can get it. So now, I'm thinking, I think I missed the whole idea, which is I should have asked ChatGPT how to interact with ChatGPT like, "Hey, can you write your own prompts?" I never asked it, because I'm not in that head space yet.
"From the Market Dominance Guys podcast episodes, I've learned that ConnectAndSell is a tool that can be used by sales reps to improve their follow-up process in sales. According to Chris Beall, many sales reps believe that manual follow-ups are more effective than using a tool like ConnectAndSell. But there are several drawbacks to manual follow-ups, including the time it takes to prepare, the fact that preparation may not result in any live conversation, and the fact that the rep may not be ready for the live conversation when it finally happens. On the other hand, using a tool like ConnectAndSell is faster and allows reps to talk to more people per day, which is crucial for searching for potential customers who might have a problem that the company's solution can solve. Chris believes that the choice between manual follow-ups and using a tool like ConnectAndSell is a trade-off that each rep has to make explicitly based on the benefits and drawbacks of each approach."
This is fascinating. How do you feel about this book project? I mean, I came up with it, when? On Friday, I think. That's Friday the... What is today? Today is Monday the 13th of February 2023. I had this crazy idea, I think it was Friday. Was it Friday?
It's roughly a hundred pages. It's got 25 episodes, it's got 25 prompts and responses. It has a cover, it has a title. Everything is titled. It has links to the podcast episodes. It needs some QR codes for the physical book. It needs a table of contents. I don't know how to get... I think, actually, Word will write one of those for us.
I'm thinking about it. And yes, in a way, it's been two days, but you and I have kept track of every element of the podcast from transcripts to summaries to everything the entire time. So we were fortunate that, for us, it was quick to be able to pull some of this stuff together, because we had the repository of all of it. That helped.
It saved a lot of time. I love the idea though. I mean, we watch friends that we know produce eBooks all the time and small books, and I'm publishing other books, 50 pages of some advice and stuff. But to be able to give something meaty like this, to have that option for people, we know that books are a calling card. Books open doors for you to have something you can put in somebody's hand that's very current, that isn't things you've been researching for years and years, like right now, this is current. Right now. It needs to be done now.
Because there will be imitators, there will be people inspired by what you've just done, what we've done, to continue this. But I love the idea, and the process was a blast for me. I told you, I need to be creating all the time. So it's a lot of fun. And I am not an illustrator. I'm a graphic designer, but I'm not an illustrator. So to get that little help, too, from DALL·E was wonderful.
And you're being able to massage the prompt for what it gave us. We were both kind of sitting there messing with it, and it wasn't quite there. But again, with my misplaced modifiers, you were able to hone it in and get our two characters, our three characters for the cover.
Yeah, I got super specific. It is like, "I want one person like this, one person like this, a futuristic robot sitting at a table. I want it to be a six by nine cover. I want it in black, white, and red. I want it in the style of a minimalist Madison Avenue advertisement."
... and then I thought... I was down to 11 prompts left with DALL·E before I'd have to wait till March 11th, my son's birthday, to get more DALL·E. I think that might still be true, even though I upgraded to Plus. I'm hoping Microsoft will save me, because they're really being nice to us right now. But it was so interesting, because I'm not a graphic artist of any kind. I don't think I have much of an eye for stuff, but I've been trained. It's interesting. I've been trained in art, art history, and I'm like a colorblind person who's coming to advise you on the painting of your house. Because when we were in Shetland, and we were in a yarn store, and you're supposed to know that these colors are cool and these are warm, means nothing to me. But I could point at them with my phone, take a picture of them, go to black and white mode, to noire mode, and I could see which ones were warm and which ones were cool. I'm that kind of idiot. That's the level [inaudible 00:33:05]
But I could look at this thing and go, "That is book cover stuff." My recent experience with Helen's book was helpful, black, red, and white, where the accented stuff is red, the big stuff is black, and negative space is used in order to convey the other message. When people see the cover, they'll get this. The negative space among the people's legs and the robot's legs under the table is really interesting.
You were the art director. I mean, Tom's an art director, my husband, so he knows how to direct and to give the word pictures, and somebody else can execute it. And you gave the word picture, you gave the elements that needed to be brought together, and that's a skill in itself. We talk about the engineering of the prompts. A lot of it is being able to really see what those goals are and knowing how to articulate it. So it isn't for the everyday person, always, to get a great result that people will enjoy, learn from, read from, be inspired by, get what you're trying to say. You still have to know how to ask.
The words came out of Market Dominance Guys episodes, plus its own words that it threw in. The prompts came from me. And the curation that's hidden, you can't see it as a culling process, because some of the responses were like, "No, that doesn't make any sense."
So I just didn't include them. Sometimes also, by the way, I put so much specificity in the preamble to the prompt that I cut that part out and just used the second half of the prompt. Unless you have this corpus already put together, this model put together, you wouldn't get the same answer, which is fine. It's like-
... you wake up in the middle of the night, and I might tell you something different about whether I'm hungry. It's just the way it is. Everything changes over time. But there's an explicit process of, we'll call it crafting of the prompts. There's a learning process, a co-learning process, where we learn. I learned from how ChatGPT responded, so I crafted different kinds of prompts. DALL·E, especially because I mean, I looked at that stuff and went, "Holy Molly, that's... Ah." And I was down to 11 of them, and I'm panicking. And it's like, "Oh, that's the one. Stop, stop, stop. Get it, get it, get it."
There's this continuous curation process that's going on, where, and inventors work like this too, as you know, I hold a boatload of patents for some reason, and what you see in those whatever 22 patents I have is the tips of many, many icebergs under the water is all the crap that I threw away. So I think curation is a big part of it, and having an eye and a feel and the emotional set to curate, to be ruthless, but engaged.
So I think this prompt engineering thing, which when Eduardo Kassner said to me, this is a thing says, "Don't use any other phrase," I'm really indebted to him for that. He was stern with me. He made it very, very clear last week, "Don't say this any other way." I'm now realizing website design was a big thing. Prompt engineering is the big thing now.
And that was my big lesson over the weekend. It's like, "Wow, we can make a book in two days." Two, the book is pretty substantial. Three, occasionally ChatGPT gives me insights about my own stuff, or Cory's stuff, or whatever that I was unaware of. But four, the big lesson is this thing is for real, and prompt engineering is what you need to learn in order to-
... get a lot out of it. And my recommendation to anybody in this audience is, "You may think that you're a CEO, like me, or like Susan, or that you're a salesperson, or you're a whatever, this is something to delve into and get dirty and find out where its edges are and your edges are, because I think you'll be shocked and gratified and relevant."
To leave or reply to comments, please download free Podbean or