Mediocre or Magnificent? What You Need to Know About ChatGPT with Frank Prendergast

SUMMARY KEYWORDS
gpt, prompt, writing, people, chat, creating, hallucinate, output, rewrite, bit, blog post, style, hallucinations, content, podcast, bland, input, copywriter, copywriting, talk

SPEAKERS
Alastair McDermott, Frank Prendergast

Alastair McDermott  00:03

Hello, and welcome to The Recognized Authority podcast. I’m your host, Alastair McDermott. And today my guest is Frank Prendergast, and Frank, somebody I’ve known for quite a long time. Actually, Frank, I think we we have been, I think we met first back in both our old days of word camp and WordPress.

Frank Prendergast  00:21

Yeah.

Alastair McDermott  00:23

That’s quite a while ago, that’s maybe that’s maybe 15 years ago.

Alastair McDermott  00:26

But today, we’re here to talk about something ultra modern, which is Chat GPT and AI. So we were discussing what to talk about here today. And for me, one of the things I think that’s, that’s important, too, to kind of put out is and get get in front of people is, there’s this kind of range of this range of opinions about it. A lot of people think Chat GPT and AI is amazingly good. And then a lot of people think they’re like, it’s absolutely awful in terms of what it’s creating mediocre to magnificent as the wording I use for the title here. So yeah, I just want to start by by just asking you, because I know you’ve written some great blog posts. And Mark Schaefer even posted some of your stuff up. Can you tell me a little bit about how long you’ve been using chat GPT for and how you’re finding it?

Frank Prendergast  01:18

Yeah, so I guess, so I guess late last year, when it kind of exploded on the scene, I started exploring it a little bit.

Frank Prendergast  01:27

But I, but at that time, I kind of I was just tinkering with it, and use it for some very basic tasks. Like, I used to brainstorm headlines for content that I’d written and things like that. But then this year, I wanted to kind of dig in a little bit more and kind of really figure out, you know, what’s it actually capable, capable of? And are there tasks that could genuinely kind of take off my plate on a more, you know, more consistent basis. So that’s what I’ve been playing with a lot now. Since since January, and just trying to come up with prompts that I can feed it that I can reuse reliably. And and yeah, I purposes, it’s best best used for.

Alastair McDermott  02:16

Yeah, and I think that’s, I think that’s one of the I think for me, the key thing is the prompts that you’re using with it, and then having a realistic understanding of what what it can actually do. Because, because when you when you get to understand that, like, the capabilities are pretty amazing. The output I can give you is incredible, in certain ways, and then it’s absolutely terrible in other ways, with hallucinations and things. And that’s something I want to talk about. And if you have if, if the people watching this listeners haven’t heard of hallucinations before, that’s when the AI actually just makes shit up. It just makes up stuff, it puts in fake names of books that authors have purportedly written. And, you know, and this is where it’s incredibly dangerous. Because if you publish something like that, in your website, that’s going to severely impact, you know, the trust the credibility. So, yeah, I think I think that’s one of the most important things to understand is when it can be dangerous to use it.

Frank Prendergast  03:16

Yeah, I mean, I love that the this is the actual technical term for when it makes stuff up is hallucinations. I love that. But it’s yeah, it’s like it’s a serious issue. And apparently, it’s an issue that is like fundamental to these types of large language models. So you had measure released their large language model, which was supposed to make like scientific research more accessible. And they had to shut it down after two days because it hallucinate it so badly that scientists are saying, Look, this is this is actually dangerous, because people think that they’re getting more accessible scientific information. But the the language model is actually just making stuff up here, guys. So they had to shut it down after two days, because of hallucinations.

Frank Prendergast  04:00

Then you have Google releasing are announcing that they were releasing Bard, and they which is, you know, similar enough to Chat GPT. And they did a promo video, and they had somebody looking for information to share with their child about a particular telescope. And again, nobody, nobody must have fact checked the video that they put out, because again, it was hallucinating facts about this telescope, and people noticed. And apparently, it wiped off like well over 100 billion in share price. When that was discovered. And similarly, chat, the technology behind Chat GPT has been you know, licensed to the Microsoft Bing search bar. And apparently their share price has tumbled because of the completely bizarre conversations that are occurring with people now.

Frank Prendergast  04:51

So, yeah, it’s this it’s actually a fundamental issue with the technology. So it’s crucial to real realize that if you’re like you have to be the subject matter expert in what you’re asking it because you have to be able to check. Is this actually true? And if you know if you if you’re not the subject matter experts, you need to be fact checking anything that tells you that you don’t know to be absolutely, certainly fact. I like I think a really good example of this is and why you have to be so careful is when I posted an article about the fact that hit hallucinates. And a woman commented, saying that she had been a guest on a podcast. And her subject matter is quite a sensitive subject matter. And the podcast host had used Chat GPT to create the show notes, and thankfully sent them to her before publishing them. Because she read the show notes. And she said, that’s actually not what I’m saying at all. And actually, that would run absolutely contrary to what I would believe or tell people. And because it’s a sensitive subject matter, you know, this is, this is a big deal. And of course, the you know, the only she could tell the nuance of No, that’s the wrong message.

Frank Prendergast  06:06

So yeah, and that’s, this is why, you know, with Chat GPT what I love using it for most is feeding it my own content, and then getting it to do certain tasks, like creating social media posts, or helping me come up with a title or you know, things like that, where it’s my content, so I can see immediately No, that’s not right. That’s not right. That’s not right. So yeah, that’s my kind of take on it at the moment.

Alastair McDermott  06:31

Yeah, I think the the thing is how convincing it is when it’s wrong. Because you see, you know, you’ll see that a list of eight books that are well known business author has written on all eight of those are correct. And then it’s got two in there at the end of the list that are totally wrong, have never have never been written. But it’s written those in this incredibly convincing way. And you think, well, everything else was right. So this must be right to. And I think that that’s the key for me is that these language models, what they’re really good at, is inventing very plausible output based on the input that they got. And sometimes what they create is, is wrong like that. And then other times what they create, what they invent, is brilliant. And that’s, that’s the other side of this, because we’re talking a bit about the negatives there, what the positives are, you can give it some some input, and then take your input that you’ve added and added to the, to the wealth of knowledge that it’s already gone in its database, and it creates this incredible output that can save you huge amounts of time. And undo a lot of kind of like the grunt work for you. In grading, first drafts are rewriting things or taking some simple input and turning that into something else. So I think that’s where that’s where it’s really amazing is it takes an inventor invents all this stuff for you. And if what invents is, is useful for you, then you’re going to think that this is one of the most valuable things ever created. If what it creates is something that makes you look bad, or you know, like in the case of the lady with the podcast, then you’re going to think that this technology is terrible. And and both of those are right.

Frank Prendergast  08:16

Yeah. Yeah, exactly. I actually, since since we spoke the other day, I came across a tweet from Sam Altman, the CEO of open AI that I had not come across before. And I thought it was really interesting. And I kind of feel like they could have been more open about this from the beginning. And it would have helped with this whole scale of people thinking it’s either like, absolutely fantastic. Or it’s absolute rubbish. Because in the tweet, which was in, I think, late December, he said, I’ll just read it out if that’s okay. He said, “Chat GPT is incredibly limited, but good enough at some things to create a misleading impression of greatness. It’s a mistake to be relying on it for anything important right now. It’s a preview of progress. We have lots of work to do on robustness and truthfulness.” And I feel like that message, you know, didn’t get out early enough to avoid some of the, the misleading or some of the misapprehensions about it at both ends of the scale. But I do think, you know, as he says, it’s great at certain things. So it’s just figuring out what those certain things are and using it for those things.

Alastair McDermott  09:28

Yeah, I, I guess I want to talk about some of the use cases. And I have an example that I want to tell you about. And I think that it’s important to understand, like for people to understand some of the capabilities of what it can actually do for you. So I was at training with my boxing coach and a training we were just having a chat. And I said, you know, Have you have you actually looked at this thing. And so he was asking me a bit about it. And so I asked her to give me a training plan for A boxer. And so I said, Give me a training plan for a boxer who wants to improve speed and power. And it gave me like a one month training plan. And I said, now give me that in the form of a five email sequence and write a subject line for each email. And so it just took the training plan that it had created. Again, it’s created this from its database of everything that it’s learned about, that has been written about boxing training that it has in its database, everything that was written up until 2021. And then it was able to it knows what a an email subject line looks like for an email. And notice what an email marketing email or an email that goes out a sequence might look like. And it took all of that it gives me out the entire sequence for that email. And I said that now give me give me an email opt in page that will ask that will incentivize people to opt into the sequence. And it was able to do all of that.

Alastair McDermott  10:54

So now if you are, let’s say you’re that boxing trainer, you want to do that for your website, it is getting you 90% of the way to having an amazing email sequence and opt in sequence for your email already done. The thing is, some of the information that it was suggested was it was okay, but it was a bit bland was a bit mediocre. And he was he wasn’t crazy about the the specific suggestions. But that’s okay, because that’s where we as human experts come in, and then go in and add our own take on it. But it’s done all the heavy lifting for us. I think that’s where the key is.

Alastair McDermott  11:30

Then the other thing that he thought was amazing was he said I showed him where it can actually write a story for you. And he said, Well, can it write a story with my daughter’s name, kind of write a story, where my daughter is a princess, and she’s she, she wants to get a unicorn. And so it’s this thing and that’s amazing that he said that he spent the next evening playing with his kids with Chat GPT and putting stories for them. So like those kinds of things, it is absolutely fabulous for

Frank Prendergast  11:59

And that is beautiful, because then you know, you’re taking full advantage of its tendency to hallucinate. Hallucinate unicorns, hallucinate rainbows.

Alastair McDermott  12:10

Yeah.

Frank Prendergast  12:11

That’s brilliant. Yeah, I mean, yeah, exactly. So another, you know, really solid example along the lines of the email you were talking about. So a practical use that I that I used it for, just late last week was, we had a series of videos about like five minute mark for a client, and I needed to get them up on YouTube, I needed to get some descriptions up on YouTube. And it would have taken me forever, because if sitting down with a blank, you know, with a blank canvas to watch, the videos have to decide what the main points are. So I used another AI tool D scripts to do the transcription of the videos, then I popped the transcription into chat GPT. And I asked it to summarize them and pick out the main points. It did a brilliant job of that. Now, as you say, it was really bland. To be fair, and I think we should talk about this because I think you have some great tips on this. To be fair, I didn’t push it to write in a particularly creative way. So what I got back was really bland. However, it was a brilliant starting point, it was all the key points, and it was stuff that would have just taken me forever, to even get going on, you know that that that creative ramp up that it takes. So it was a fantastic to cut down on that time. Thankfully, I’m also married to a copywriter. So I was able to send those to Marcy, and just say, Look, these are really bland. You know, can you can you work with these a little bit. And she added some flair and and some creativity to them that really lifted them. So I think that kind of thing is is really valuable just taking some of the taking some of the grunt work out of content creation.

Alastair McDermott  13:53

Yeah, absolutely. And before we talk about those prompts and things, I just want to say hello to hires Aladdin who’s listening or watching on LinkedIn. And Irene, I’m just going to show Iran’s comments. Irene says she’s seen this happen before where she asked Chat GPT to reread something that she had written and include information that she had not provided. I asked where it got certain facts that weren’t included in apologize and said that it assumed the additional information. So that’s from Irene color. So thanks, Irene.

Alastair McDermott  14:22

So this is this is the key thing, you have to go and check everything that you got it to write in case it does this and it assumes that. Now I personally would love if the if the way that it output information, if we could get it to markup in some way where it has created something or assumed something so that we know you know, maybe have it in a different font or a different background color or give you a note at the end, you know, something that tells us where it has made stuff up and where we, you know, we need to go back and fact check things as opposed to where something has just been created from the database. But then the other thing that’s key is, these have been seeded with this database of information. It’s terabytes of information, a huge amount of multiple 1000s of terabytes of information. But the thing is, if there was incorrect information in that seed, then what’s in there is going to be wrong. So, you know, that’s, that’s the other thing. If there’s if there’s misinformation in there, the output it’s gonna think is correct. It hasn’t hallucinate it. It’s just, it’s operating off. incorrect information to start with. So that’s the other thing we need to be careful of.

Frank Prendergast  15:36

I read the other day about a company, unfortunately, I forgotten the name. But I did read about a company who was trying to solve the hallucination problem by having AI fact check itself before it gives you the output. And apparently, they’ve you know, they’ve got the technology working, but the processing power that it requires means that it’s not actually feasible to roll out at the moment. But yeah, so I think people are definitely working on like, how how this problem can be addressed.

Alastair McDermott  16:06

So let’s talk about some of the other things. So actually, let’s talk about how to actually avoid hallucinations, or, you know, what can we do about those? And one of the things that I think both you and I have seen is that the shorter the input data that we give it, the less it tends to hallucinate. All right.

Frank Prendergast  16:24

Yeah, I’ve definitely, I’ve definitely found that. So I think as well, it’s probably worth mentioning at this point that there is a like, you can use chat up for free. But there is a monthly paid subscription. And I have found like since moving to the paid subscription, I get far fewer hallucinations. And if I feed it my own content, I’m much more unlikely to get hallucinations, whereas on the free version, it was hallucinating wildly, even from stuff that I had just given it. And, and I think as well, a bit like a bit like Irene said, like, it would actually lose the run of the conversation if I was doing iterative prompts. So I would put in a blog post about personal branding. And then suddenly, it would start talking to me as if we were discussing a post about mindfulness. Just completely random stuff.

Alastair McDermott  17:12

And just just sort of people who don’t use the system much or haven’t used it at all, maybe you’re wondering like what we’re talking about there. So there’s only a certain amount that you can actually pay stands, you’ve got a text prompt that you can, you can type your question or your prompt. And you can see, you can typically give it a prompt, like rewrite this with correct spelling and grammar, for example, and then you paste in some text, and it will then or you could say, summarize this for me and pull out the key points in a bullet list. So and then you then you give it some text, the amount of text that you can actually paste in there is limited by the system, it just won’t work if you go above a certain amount. And when you’ve got like AI, we’ve got this podcast, The Recognized Authority, and we’ve got episodes that are 45 to 60 minutes in length, it won’t accept an entire podcast transcript. Now what it will do, we’ll take about eight to 10 minutes at a time. And so you can input eight to 10 minutes of the conversation and ask it to summarize that. And the longer the block of text that you put in, I find that the more it’s going to hallucinate and make make mistakes.

Frank Prendergast  18:16

Yeah, exactly. I ransom. So I find I’m not sure what eight to 10 minutes would equate to but I find that about like 1000 to 1500 words is a good chunk that it can kind of manage. And then. So you and I have discussed previously, the this technique that I learned from URI blast, where you can input chunks of like 1000 words, but call them like part one. So you’ll say to chat GPT. Here’s part one of my blog post, just remember this and reply with Yes. When you’ve got it memorized. And you put in your 1000 words, and it just says yes, then you can say here’s part two. Again, please say yes, once you have it memorized, you type in you paste in your next 1000 words. And you can build up a much longer piece. That way by then at the end selling it Now please put together a part one plus Part Two plus part three, and call this the post. So that’s one way you can do that. But I have found that the more text you give it that way, the more likely it is to then hallucinate when you’re asking it to work with that text.

Alastair McDermott  19:24

Yeah, yeah, absolutely. And so that’s what we’re talking about, about the likelihood of hallucinations, being much more likely with longer text. And I haven’t really been satisfied with the output that I’ve gotten when I’ve used prompts like that where I’ve kind of concatenated So I typically now only stick to short text inputs that I’m giving it. And again, like you what I’m doing is I’m I’m asking it to a masking to work with the text that I’m providing myself. I’m rarely actually asking it to, to work on stuff, you know, to just make stuff up from its own database. Like for example, the The boxing training plan that I mentioned earlier. And what might be more useful in that case is if the boxing trainer comes up with five things, five, five workouts, and summarizes those five workouts and says, Here are five boxing workouts, please write me a list of emails based on this input. And that way, the output would be much, much more accurate, much more. It’s starting from its starting from that human input. It’s not just kind of taking this average mediocre amalgamation. That’s it’s it’s inventing or grading from its database. And that’s, that’s where it can be really useful is when you input like that.

Frank Prendergast  20:44

Yeah, I mean, I think that, you know, at the at this, at the end of the scale, where we talked about where people think that, you know, Chat GPT is just rubbish is where they try it out for the first time. And they say something, you know, let’s say they’re, if they’re a marketer, they go on, and they say, write me a blog post on marketing. And what comes back is bland, generalize, you know, but then the prompt was, was bland and generalized. So the more specific you can get with your prompts, the better the output you’re going to get. And for me, I find that like, using your own content, well, then that’s like the most specific you can get. So I totally agree with you. Yeah, that that starting early, as I said, I love working with with my own content, because that’s really specific. And it means that I can very quickly assess the quality of the output.

Alastair McDermott  21:33

Yeah, so I think we’re kind of getting into prompt engineering a little bit here, or prompt generation. So what that is, is, is, is about writing the instructions that you give the AI in this very specific way. And there are there are lots of ways that you can, you can write a really short, you know, a very short prompt, that’s only five or six words, it’s going to give you something that’s very generic as output. But the more that you add to your prompts, the more instruction that you give it, the more that it can modify the output that it’s gonna give you. And that’s, that’s what we’re talking about here is giving you can give you some really interesting output. So do you want to talk a little bit about some of the prompts that we use?

Frank Prendergast  22:15

Sure, yeah. And I think I think it’s interesting is also what I’ve, what I’ve been trying to do is like develop prompts, that I can just roll out really quickly and reuse that, or I would call them kind of like stable prompts, that they kind of generally speak and do what I expect them to do. And so, you know, and I’d love to talk to you actually a little bit about let me let me give an example, I guess, first. So one would be let’s see if I wanted to do like a standalone social media posts but based on a blog post I had written, I might say, please identify the top three most practical, useful and unusual standalone points made in this article, writer 75 words, social media posts, for each point, use the same style and tone as the original content, make the opening line, highly emotional and attention grabbing, please use the problem agitate solution copywriting framework for each social media post, and then I paste in the post. So that’s like, super specific.

Alastair McDermott  23:20

Yeah. And you’ve gotten with with, you know, with the amount of detail that you’re getting in there. So it has a lot to work with there.

Frank Prendergast  23:30

And one of the things I’d love to talk to you about is because I know you, you work on a kind of an iterative prompt basis. And what I’d love to talk to you about Jeff, you want to describe maybe that process first, and then I have a question for you on it.

Alastair McDermott  23:45

Yeah. So. So I’ll give it I’ll give it an instruction. So, you know, please, please. So first thing is, I typically will say, as an expert copywriter, rewrite this. And so I’m seeing with it with with the concept that is going to be writing in the style of a copywriter. So it’s going to use certain rules, by the way that it does that. So so I’ll say something like as an expert, marketing copywriter, rewrite this. And sometimes I’ll use in the form of bullet points, or sometimes I’ll be taking bullet points and asking it to turn it into a paragraph.

Alastair McDermott  24:21

So for example, when you serve that might be before this show before it before we got went live. I was I was going to get it to rewrite what we’re going to be talking about here, wholesales mcdorman. Friday guests, Frank pantograph discuss GPT in the possibilities of AI for content creators, I was actually going to ask you to give me some more for that. And then I was going to I was going to take that rewrite that and then use that to to create the event. Now as it turned out, Chat GPT actually went down this afternoon, even for for people on paid plan. So I wasn’t able to do that. But yeah, so typically I’ll give it as a as an expert marketing copy. Right, or sometimes as a social media expert, and get it to do that, when I’m when I’m using it for creation of content like that, and then it’ll give me some output. And then I will give it an iteration, I’ll say, rewrite that. But separate point two into two different bullets, and write and add a little bit more about x. And so I’ll get it to change something. And then I’ll look at the output. And I’ll say, You know what, it needs to do this. So I’ll say, you know, rewrite that would make point number three more impactful by adding some emotional language. And so I’ll get it to a look at the things that I think are not good about it. And I’ll get it to, to fix those things. Rather than fixing it myself, I’ll just ask it to do. And so I’ll go through like maybe 10 iterations on something. And the output that I have, then it’s not perfectly I still wouldn’t copy and paste into a blog post, for example. But it’s probably 90% of the way there instead of 50% of the way there. So that’s what I’m talking about when I’m talking about the kind of iterative iterative approach. So yeah, that’s, that’s how I do that.

Frank Prendergast  26:08

Cool. Yeah. And it makes sense. And I suppose each each piece of content is going to be, you know, kind of unique. So your your iterative prompts could be very much kind of contextual. And so because I was curious, when I was curious about was like, Do you ever, do you ever take successful prompts in your kind of iterations, and then try and like reverse engineer them back into your original prompt?

Alastair McDermott  26:34

Right, okay, I see what you mean, to create a an original kind of prompt that I could then use more generically. I don’t only do and I tend, I tend to try and remember, because I don’t actually yet have a good system for copy pasting prompts in and thinking about setting up a spreadsheet to put them in. And I’ve downloaded a lot of prompts from other people. But then I have, you know, have 2000 potential prompts on my desktop, and it’s like, well, I’m not going to do that. So I’m actually writing them by hand almost each time now. And then going through that iterative process, and I’m sure that I could get more efficient with that. But at the moment, it’s working well enough for me, so I’m okay.

Alastair McDermott  27:13

There is a lot of there’s a lot of things that I already know quite a bit about copywriting and things like that. So I’m able to make up some stuff on the fly that maybe I would recommend that people if you’re not familiar with copywriting that maybe you’d want to do, like, for example, I will quite often say, as a marketing copywriter, as an expert marketing copywriter, give me 15 headline options written in the style of a Copyblogger headline, give me some of those options, starting with the words how to, and keep them between five and eight words. And so that will be my specific prompt, and then then I’ll give the title for whatever it is. And so so if you’re not familiar, copy, copy blogger wrote the magnetic headline series, which was some great analysis on what was working as really effective headlines. They wrote this about 10 or 15 years ago, I think. But it’s a really great format to use for a headline that will get people to click on it. And it’s not too kind of clickbait either. So I really liked their style. So that’s giving it some very specific was very specific instructions about the headlines. And the reason I wanted to be short is because I want the headline to, to work on like, like, read a graphic for promoting something on mobile, I don’t want too many words in there. So So that’s, that’s the kind of a prompt that I would use for for headlines. So you can see I’m kind of I’m just putting together some stuff from my from my own knowledge. And maybe if you’re not into copywriting don’t have that, that same background, having having those provided to you will be really useful, because you can then use those.

Frank Prendergast  28:50

Brilliant, yeah, brilliant. And actually, it just might that might be a good segue into talking about, you know, getting Chat GPT away from the kind of more bland because it will it will kind of default to a bland, general style. And I remember listening to mark Shaffer’s podcast with Charlie Palmer. And he was saying that, you know, when you tried to first he tried, he asked her to write a blog post and he was quite specific about the topic about what it should include. But what he got back, he said, you know, he wouldn’t dream of publishing on his blog. But he then made a very simple modification. And he said, Write this in the style of Shelly Palmer and I’ve heard Mark Schaefer himself talk about this as well that when that he has, he has tried, he has he is asked to to write a book chapter on something very specific in the style of Mark Schaefer, but they have a massive advantage over me because I don’t have the body of work online. That the that the language model would be trained on up to 2021. But you you had come up you had you have developed some ways of getting charged up to be a bit more creative and it’s voice.

Alastair McDermott  29:58

Yeah, so um, One of those is so I, I was thinking about my own writing style. And I tend to be a little bit more informal, more conversational. I write very much like I speak. And I’m fairly succinct in my writing style as well. So what I said was my prompt was Chat GPT list, some business authors that you have been trained on, which was the key part at that have an informal, conversational, succinct writing style. Give me 20 options. You don’t have to go 25 options. So it gave me all of these authors. Sheryl Sandberg was one. And Malcolm Gladwell was another I don’t think that Malcolm Gladwell has the same kind of writing style that I do, but those are just some examples. And so what I did then was I took some, I took some bland chat GBT output, and I said rewrite this in the writing style of Sheryl Sandberg, for example, again, that’s just the one I happen to remember.

Frank Prendergast  31:00

Sure.

Alastair McDermott  31:01

So it then rewrote that it wasn’t perfect. It wasn’t perfectly me because I’m not terrible. But it was bloody good. It was really good. It was much more conversational, much more informal, I. When I say informal, Chat GPT sometimes interprets that as kind of folksy. And that’s not really me either. I have to look out for that specifically. But yeah, getting to say rewrite this in the writing style of x. And if you put in whatever you want for X. Now the other thing that I did that got me a bit of information as well was I took one of my blog posts and I said, analyze this writing. And give me a table and give me a table as a really useful prompt as well, you can ask it for bullet points, you can ask it for tables and other things. But I said, analyze this writing this text that I’ve inputted and give me in form of a table, breaking down what you can tell about the author’s writing style. And so what it did was it did an analysis of the texts that I had input, and it started calling out what it could see my writing style, because then I was able to use that to create a writing prompt in a writing style that is conversational, succinct, and whatever else. Those things were I can’t remember them all now. But I was able to get it to do that to start to kind of model my writing style. And again, none of this can be copy pasted. Exactly. But it can it can do a lot of grunt work for you. It can it can do the 80-20. And that’s that’s what I think, is where it really shines.

Frank Prendergast  32:37

Yeah, yeah, absolutely.

Alastair McDermott  32:44

So what about you, like how have you found the writing style thing? Is it that you’re rewriting things afterwards? Or are you trying to get it to to to learn your style,

Frank Prendergast  32:55

definitely rewriting things at the moment. And I’m going to, I’m going to try what you’re suggesting, I’m going to try and find people who might have a similar style and just see how that works. Because I have tried to get it to analyze my writing style. But I haven’t had much success in getting it. I haven’t had much success and getting it to write something that I would feel was was close enough. So yeah, absolutely. You know, right now, I’m definitely rewriting almost everything. What I find I do a lot of the time is I will get it to do things, I will get it to do any task at least twice, maybe three times. And I’ll copy them into Google Drive. And then I’ll cherry pick the best bits from each one. And then add a little bit of my own voice as well. So yeah, I’m still still doing, you know, still doing a fair bit of work. But it just takes, it just takes it’s like you said it’s the 80-20 it takes the the blank slate the blank canvas away, and it does some of the initial grunt work.

Alastair McDermott  33:59

Can we talk about the actual use cases there. Because I think it’s interesting to discuss, like what people can actually do with this because like, like, we’re not just talking about, you know, social media posts or blog posts and things like that. Can you talk a little bit about some of the things that you’ve used before?

Frank Prendergast  34:17

Yeah, so I mentioned earlier the the YouTube descriptions that was really useful. I have I’m trying. I’m trying to get it to write social media posts for blog posts so that I can repurpose old blog posts because for me one of the one of the things with repurposing is you go back to an old blog post and you’re trying to remember well, what were the main points I was making in this and you know, so I’m trying to come up with multiple posts that you can shedule in. But I’m still working on that. I’ve actually got it. I’ve got it, creating some really good content for that. I just haven’t started charging them in or actually using them yet. But I’m happy with where I’ve got to on that. Titles. By the time I finished writing a post a lot of the time, I’m like, I, I’m I, I’m not great at writing the title. So I will, you know, Marcy had me do an exercise. Last year when I, when I had written an article, she had me do an exercise, which was write 100 titles for it to find the best title. It was a fantastic exercise. But wow, like my brain nearly melted. So. So before Christmas, I instead of doing that exercise, I asked Chat GPT to write 100 headlines for me. And then I just went through, and I started picking out the best bits, the best ideas, it had angles that I wouldn’t have thought of using. And I again, I never I didn’t use anything like word for word. But I wouldn’t have gotten to the title I did if it hadn’t been for Chat GPT.

Frank Prendergast  35:52

What else have I used it for? Good question. We’ve used it to, we’ve used it to, to start blog posts, to be honest, most of our use and usage is social media, blog posts and titles, but never, never as is. I did also come across somebody the other day talking about you know, so you mentioned earlier, like you have a lot of copywriting knowledge. And I think I’ve learned a lot about copywriting over the last couple of years, and I’m married to a copywriter. Not everyone is you know, not everyone has those skills. And so I think Cha-Chat GPT could be really good for just quickly improving on content that you know, is a little bit lackluster. However, I will say as well that like I’m not 100% sure yet, how much copywriting skill do you need to then assess Chat GPT’s output.

Frank Prendergast  36:54

So the thing that I came across that was interesting was a copywriter who was working with clients, and who if they just needed a quick win, the copywriter would use chat GPT to improve say their LinkedIn about section. So they’re not having to pay the copywriter to go away and do a full rewrite on their about, but the copywriter is able to apply their skills to the Chat GPT output to just get them a quick win of have a bit of an improvement on where they’re at right now. So I thought that was interesting.

Alastair McDermott  37:27

Yeah, that’s, that’s an interesting one. And I’ve seen it, do that kind of summarize, rewriting, rewrite this sentence to add more emotion or more impact. And what you mentioned earlier on about asking for hundreds, I rarely actually ask for 100 in output because it just takes too long. But I have noticed that I’ve asked for five and I haven’t been happy. And then I asked him for 20 options. And then I don’t always take one of the 20 verbatim, but I might combine number seven and number 14, you know, and and I think that’s really effective. When you do do stuff like that. Giving giving options and things like that. And and it’s really great. Like for people who are not great communicators in the written form, it can be really great for taking emails or messages, they want to write and tidying them up. People who maybe English is not their first language, or they have dyslexia, things like that, it it’s got it’s gonna be amazing for for that.

Alastair McDermott  38:28

And I just want to mention because this is like we’re talking about the limitations and the mediocre I put and stuff like that the capabilities that it’s going to have in a few years time is going to be enormous it’s going to be because it’s going to improve exponentially. And I think that it’s important to remember that like, what we’re looking at here is going to is it’s going to continue to improve and it’s not going to improve in a kind of a straight line it’s going to improve, it’s just gonna get so much better. Even the amount of of data that they’ve got from from us from the way that we’re using it is going to feed in like they’re probably doing analysis on how we’re using with the system, you know, which reminds me, this was something else I was able to go to I have spreadsheets full of survey responses. And so I was able to take those survey responses, paste that data in and ask it to summarize the key points that people had put in when they when they when they responded. So I was able to take you know, 50 or 60 text inputs and ask you to summarize the key points. What were people saying? The survey was about Why had somebody not not started a website. And it was really interesting to get the reasons back as to why people hadn’t done it turns out that in some summary, I already knew a lot of the data but it was interesting to see what it had come back and it basically summarized it correctly, which is what I was trying to do with with Testa, but really…

Frank Prendergast  39:58

That’s too so that’s to it. You’ve given me that I’m gonna try out, when we finish this call one is the survey, one you just mentioned, because I have a survey that I need to analyze. So I’m definitely gonna try that. And the other one is the email one. That’s, you know, that’s really interesting, because I literally helped a client just today in, in, you know, writing a difficult email reply to get just to get the tone, right. And I’m going to, I’m going to, I’m going to put Chat GPT to the test, after this call, and just see what it would have said, to see if if, if it would have been an appropriate email to send, that will be fascinating.

Alastair McDermott  40:37

Yeah. And so that the scope of what you can do with it, you can ask it to, for example, you can ask it to give you a meal plan for two people, say, for a week, and you could say, create me a low carb, low carbohydrate, or high protein meal plan for two people for five days. And it’ll give you that, and then you can say, Can you give me that in the form of a shopping list, and it will do that it can do things like that really, really well. Now, it’s, again, it’s creating this from its database, so and so the meal plan that’s going to create is going to be because there’s a lot of meal plans out there on the internet, a lot of recipes out there, it’s pretty, pretty damn good. But it is only going to be as good as that whatever that input was. And it’s kind of averaging like, that’s what it’s doing is it’s averaging all of the meal plans that ever that ever encountered. And it’s taking the average of all of those and then giving you a suggestion. Like, we don’t know exactly how it randomizes or prioritize those things. But it’s, it’s, it’s, it’s, you know, using it like that, it’s it’s really useful. And a time saver, I think it’s going to be used for for stuff like that a lot. So that the uses are, there’s a lot of uses where it’s really incredibly good. And then the stuff that we talked about already where it can actually be dangerous, and you genuinely need to keep an eye on it to make make sure that it’s just not making stuff up. And, you know, that’s where being an expert comes in.

Frank Prendergast  42:12

Yeah, absolutely.

Alastair McDermott  42:14

So is there anything else that we that we need to discuss that we haven’t talked about yet that people need to look at when they’re when they’re doing this? I mean, I would suggest that, that everybody go and play around with it a bit, just to see. And, you know, and learn a little bit about the way that you write your prompts, because that will make a big difference in terms of what you’re going to get back in terms of output. Is there any any other recommendations that you’d have for people with this?

Frank Prendergast  42:46

I think, I think I think it is what you just said there really it is it is a case of just experimenting with it, playing with it, figuring out like, where where in your own processes, because as you said it, you know, the potential uses for it are very broad. So it’s figuring out like, well, what are the things that I’m doing right now, that Chat GPT could help me on, and everybody’s processes are going to be slightly different. So it’s going to be different for everyone. So it’s just that thing of figuring out Yeah, where in my process, could this speed me up? And I will say, you know, there isn’t like a word or word of warning, it will slow you down initially. Because you have to figure out, you know, how, how do I prompt it? How do I get it to give me what I want. That is that is going to be that is going to be a little bit of a time sink. Then it’s like so in my case, I’m just I’m saving any useful prompts into a Google Drive document. So that the next time I go to chat up, I can just copy paste and get a couple of social media posts or copy and paste and get a summary of a video that I’m happy with or copy and paste and get three really good quotes from a podcast episode that I was on or something or something like that.

Frank Prendergast  44:01

Because again, even with the even with prompting it for quotes from something, I found it would hallucinate. And the the prompt would not be exact, it would be kind of like Yeah, I think I said something like that. But I didn’t actually say that. Exactly. So I had to develop a prompt that said, find a quote, do not change it, do not edit it, do not modify it. Give it to me word for word from the transcript, you know. So yeah, there is a little bit of time in figuring those things out. And I mean, I share, I share a lot of the prompts that I find with people in the hopes that it’ll cut down on that time a little bit.

Alastair McDermott  44:39

Ask you to tell the viewer or the listener, because I know you had a really great way of describing kind of how you see chat GPT personified as the year. Can you just describe that for us?

Frank Prendergast  44:55

Yeah, so I called Chat GPT your new LSD. Loving Junior copywriter, because because there’s a good chance that can hallucinate wildly, but equally, you know, you will be you will be delighted with it with what it gives you some of the time, but but you’re always gonna have to check it. You’re always gonna check his work.

Alastair McDermott  45:19

Yeah, yeah, I think that’s a good, that’s a good place to leave it. Frank, I always ask people on the podcast, and I didn’t prompt you with this upfront. So sorry, what would be your number one tip for anybody who wants to build or increase their authority?

Frank Prendergast  45:36

Well, I guess the number one tip is start creating content, because I think it took me an awfully long time to start creating content. I mean, like you said, we’ve known each other for so long, but it’s, we’ve known each other for like, I think you said it was about 15 years. And I mean, I’ve been in the I’ve been in the space, like 25 years, I only started creating content, like properly in any way meaningfully in the last five years. I think it’s a huge hurdle for a lot of people to get over. And sometimes I think, you know, when you become a content creator, you’re in that filter bubble, and it feels like everybody is creating content. They are not. You know, there are so many business owners, there are so many solopreneurs that are not creating content. And I think if you you know, if you want to build authority, you have to raise your visibility, you have to start creating content, and start building your authority by sharing by sharing your expertise and sharing your knowledge.

Alastair McDermott  46:37

Yeah, yeah, I love it. And that’s, that’s like, I’m in the same boat. And I can I can say that it was around February or March of 2021. Because that’s when I started the podcast. So that’s when I truly in in I like I’ve been writing blog posts since 2007. But but that was the first time that I really became a content creator. was when I started the podcast. So So yeah, I’m 100% with you on that. Frank, where can people find you if they want to learn more?

Frank Prendergast  47:07

Yeah. Frankandmarci.com. Marci is M-A-R-C-I and connect me on LinkedIn. LinkedIn is where I just I love connecting with people chatting to people, building a community, building a tribe building my network. So yeah, please LIKE me on LinkedIn. And and let’s chat.

Alastair McDermott  47:25

Yeah, and I highly recommend following your posts on LinkedIn because they’re great. And Frank, thank you so much for being with us today.

Frank Prendergast  47:33

Thank you. This has been fantastic. Loved it. Thanks a million.