This episode features Interact’s CEO, Josh Haynam, and Digital Marketing Manager and Host, Jessmyn Solana.
In this episode we will cover:
- What is prompt engineering? Can anyone do it?
- Why is using the correct prompt important?
- What you need to know about your business when formulating prompts.
- What do I do once the output is generated?
- Top 3 best practices when using AI for your business
Learn more about AI and how to use it in your business on our Youtube channel!
Check out the blog: 13 ChatGPT prompts for marketers to multiply their output
Jessmyn:
Hi, guys and welcome back to Interact’s Creator Stories Podcast. I’m your host, Jessmyn Solana. So great to be with you all as always. And with me today, I have CEO, Josh Haynam. Josh, thanks for hopping on again with us.
Josh:
Yeah, thanks for having me back on this has been fun.
Jessmyn:
We are on this roll of sort of talking about AI because it’s become such a big phenomenon, especially in our industry. And last week we touched a little bit on what is it, what do you do with it? And basically, why even use it for your business and how to pretty much make the most of it. And we sort of touch a little bit on actually using prompts and how to use it and what to do with it in terms of formulating what to actually get in your output. So today we’re going to dive deeper into that sort of go really deep into what is prompt engineering, how do you do it, why use it, and then how to make the most of it. And there is a video on our YouTube up about this if you guys want to go check it out. I forgot Josh, what’s the title? I literally just watched it but I don’t remember what it’s called. But if you look up prompt engineering Interact.
Josh:
Yeah, we should have a couple up by the time we do this episode just titled Prompt Engineering and then something.
Jessmyn:
Yeah, and you actually go really deep into sharing your screen and going through the whole thing. It was really informative, but today I think we’d have a better sort of conversational about anyone who’s curious if they’re really sort of starting out in AI, maybe they’re a little hesitant because it’s new, it’s a new concept, new technology again that we have to learn. So I guess first and foremost, what is prompt engineering and can anyone do it?
Josh:
Yeah, for sure. Well, yeah, that video has been really reassuring. I got a text from a friend, and this is the first time this has ever happened to me creating content for 10 years. You don’t expect your friends to read it. I got a text from a friend who was like, “Hey, that was actually really helpful, and I used it.” And I was like, “Whoa, that’s pretty cool.” But yeah, prompt engineering is essentially, I would say if you distill it down it’s like giving ChatGPT or OpenAI Playground, which the distinction between those and I like using the OpenAI Playground is it just gives you a little more granularity of control over how random the outputs are. How creative it’s going to be with its response is and how close to what you tell it you wanted to give you was kind of giving it a box to fill in and then telling it how to fill the box so you’re like, here’s what I want and here’s how I want you to do it.
I would say that’s a general overview of what I consider prompt engineering to be and what I think the industry is forming around. And we didn’t talk about this before we jumped on, but there’s actually a New York Times post article that came out maybe yesterday that I think brings to light the importance of prompt engineering because a writer for the New York Times, Kevin Roose got into a conversation with the new Bing ChatGPT that got pretty dark actually. And the Bing, you can go read the story, it’s wild. The chat was saying that the chat named itself Sydney and then Sydney was like, “I’m in love with you, you should leave your wife.” And was like-
Jessmyn:
What?
Josh:
… very convincing or at least trying to be very convincing and at least that’s what Kevin’s response was. He was like, it was actually pretty convincing, and I was actually kind of shook up by it.
And then the response from Microsoft was like, “Well, basically they were like that’s what happens when you give it free reign.” Which you can agree or disagree with that, but I think that the importance of having a very specific output that you want from ChatGPT or any sort of AI that you’re using is going to grow and more and more overtime, especially as these things get smarter and they take in more information because it can be used to do super nefarious things, not good stuff. Or it can be used to be really effective and save you a lot of time and increase your output without you having to do more work.
So it’s a very long answer, but overall I would say it’s giving the ChatGPT or whatever AI you’re using parameters to follow so that it can work at its best because from what I’ve seen, and I think a lot of people are coming to this conclusion when you let it run wild, it’s not good at anything and it makes stuff up and it says things that aren’t right and it says things that could cause harm. So it’s about giving it parameters to operate in.
Jessmyn:
Something that came to mind as you were saying that was if you’re using this for your business, you’re an expert in your field, it’s a really specific sort of topic. What happens if the information that comes out is incorrect? So how do you make sure that you’re feeding it correct information so that what it comes out with is actually giving you less work, not more work because you have to double check it, make sure that it’s correct and then edit it.
Josh:
Yeah, so from all the testing I’ve been doing, and I’ve probably spent at, when I recorded that video, I think I said 300 hours, it’s probably been 400 because I’ve been spending eight or 10 hours a day working on this. It really can’t come up with stuff on its own. And when you ask it to, it gets a lot of things wrong and I actually think it takes you way more work to go from something that it gives you from its default settings to something you would actually use versus the method that I’ve been using, which is you give it something you’ve created. So let’s just say it’s a blog post, you give it your blog post and then you say rephrase this blog post or summarize it in one paragraph. It’s super good at that. And this kind of goes into a little bit of nerdy stuff, but the T is trans for transformer. So GPT and Transformer is a model that Google released in 2017 that was basically built to translate, so can take English and turn into French. That’s like the example on their blog post.
And the way that it does that is it kind of puts these identifiers on each part of a word and down to the letter level, there’s also a really, really, really great blog post by the guy that started WolframAlpha, his the name is Wolfram, and that goes really far into how this works, but essentially it takes text and transforms it into a different kind of text. So if you think about what it’s going to be super good at, it’s going to be super good at taking something you have and then turning it into something else. It’s like I’ve been using it to build quizzes obviously, and now I’ve been helping a bunch of customers do it.
And what I’m asking them for is like, give me one of your articles or give me your website, I’ll take that text, put it in there because they’re asking me to obviously not without their permission, and this is where all of this gets super shady, but as long as you have permission to use the content or you can use Wikipedia if you need information to feed it, you feed it that. Then you say transform this into a quiz, transform this into a blog post, transform this into some tweets or whatever format you want it to spit out. So it’s like here’s your input, here’s the instructions to change that input and here’s the output and here’s what I want the output to look like. And then it does super, super well going from scratch the generative part of it. I don’t think it’s worthwhile because in all of the stuff that I’ve done, I think you’ve had a similar experience. You’re saying where you’ve tried to have it make stuff and you’re like, look guys, this is going to take me way longer to edit this than if I just wrote it from scratch.
Jessmyn:
Yes, exactly. But it was because I didn’t know how to use it. I was like, okay, I understand the idea of prompting it. You have to type something in say this is what I want you to make. But it was a prime example of why you need to use the correct prompts, or you need to understand how to create the right prompts to get what you want out of it. And when I had used it, because I didn’t know that much about it, I was like, “Ah, this feels too much work.” But if you’re kind of actually watching some of the content out there, people are actually using it like Jackie. After watching some of her videos, I was like, “Oh, that makes a lot of sense.” And I think it was Jessie for those of you who don’t know her, she’s our social content manager.
She did a lot of digging on her own of other people who were like, “Hey, here’s how to create good prompts to feed into ChatGPT or whatever AI that you’re using in order to get the right output out.” But I guess my question is when you say, I know I heard you also say a lot, you’re training the AI, right? You’re training it to get to do it correctly. What exactly does that mean in terms of when I think of training, I’m thinking of, okay, maybe I started a new job or I’m learning how to ride a bike. That takes time, but how much goes into this training for the AI in order to get it right?
Josh:
So it’s a lot of trial and error. And the cool thing about the AI, and this is why it’s cool, but also not reality because it’s not reality. You only have to train it once and then it never messes up. And so I’m like, “Man, I wish I could do this.” Takes me a million tries to learn how to do something, I’m playing a game and I make the same mistake 50 times in a row and it’s like, “Man, why can’t I remember this?” So that is the beauty of the AI is like you train it once, it does it exactly right every single time.
But the part that’s taking me forever, and I think this is a lot of people’s experience, is learning how to train it. Nobody knows right now, nobody. You take this type of engineering, prompt engineering, it’s essentially writing code in English from my minimal experience of writing code, and I can understand and at least read basic languages, SQL, JavaScript, database languages. I can read what’s happening and it’s the same exact thing except for you’re using English words, but it’s still code. Because for example, if you’re going to feed it an example and you say read the following example, it will mess it up unless you put carrots. What are those carrot things called? The-
Jessmyn:
Arrows.
Josh:
… keyboard.
Jessmyn:
I don’t know. I call them arrows as little up arrow.
Josh:
Yeah, yeah, the sideways arrow.
Jessmyn:
Oh, the less than great than-
Josh:
Yeah, the less than, yeah. Yeah. Okay. This is super specific, but if you’re going to feed it an example, you say read this example, you have to put a greater than sign less than whichever one is the one that’s bigger on the right side you have to basically put, this is the example, and you have to put those brackets around it. You could probably use brackets too. So you say read this example brackets example, and then you put the example and then you put another bracket and you put example. And that’s how you write code. If you’re writing CSS with how style works on websites, you have to do that. And if you don’t do it breaks. And it’s the same thing with prompt engineering. If you don’t do that, then it doesn’t know where the example starts and stops.
And so it will just give you something really weird. And the thing that’s frustrating, I mean marginally frustrating I guess about prompt engineering is that in regular engineering, if you do it and you don’t put the brackets at the end, it just won’t work, and it won’t tell you why it doesn’t work. ChatGPT or whatever AI you’re using is different because it will give you something. It just will give you something that’s wrong. So you still have to go in and diagnose where it’s wrong, which is exactly how programming works. What takes forever in programming is you run this code and it doesn’t work and you’re like, “Why the heck does it not work?” And that’s what takes forever about writing the prompts is like you have to figure out what it understands what it doesn’t understand. The thing with the brackets, I had to figure that out on my own.
You go and try to Google it or look it up on whatever, there’s no resources of anybody being like, here’s how it works. You go on like Chat GPT or OpenAI’s Discord. And it’s a bunch of people being like, “Look, I made it write this funny story.” And you’re like, “This is useless. I’m trying to actually use this for something practical. So you just have to try a million times to see what works.” So that’s actually one big breakthrough was when I figured out you have to bracket around if you’re going to feed it an example, and that’s how you train it, you train it on your own writing or your own data, but then, and this just goes, it’s like an endless rabbit hole because then you’re like, “Okay, so how do you structure the data that you feed it?” So let’s say I’m trying to make a personality quiz with four outcomes and I know what my outcomes are and I want to feed it that.
Now, I need to feed it those outcomes, but I need to label them, I need to label it outcome one, outcome one description, outcome two, outcome two description. And then when I tell it to print out a quiz, I need to tell it, create a quiz that has an equal number of outcomes to the ones listed above, otherwise it’s going to cut them out. So very, very specific. And then you can get down to the granular level. I was working on one this morning where there’s a type of quiz that’s basically it’s framed around what’s the best next step, and this works for anything. What’s the best next step in your fitness journey? What’s the best next step in your business?
But next step could be other words. What’s the best direction, what’s the best way? What’s the best path? So then what you’re going to do is you’re going to put brackets around that part and you’re going to put all the different words that it could have separated by commas. You’re going to say, what’s the best next way, path, direction, whatever. And then you’re going to say, give some version of this title in the output, and then you’re going to set the randomness to a higher randomness so that it will give you every time it spits out a different version of that title. So it’s not always exactly the same. So you can get down at the word level at the structure level, and then it just kind of scales up and down.
Jessmyn:
It reminds me a lot of being in school or in college and we had to figure out how to Google correctly, and you had to really figure out, okay. What are the words that I need to put in here? And I don’t remember it for the life of me now, but it had to, if you put a comma that meant something, if you put a plus sign that meant something or you had to say, “And,” or you had to say, “Or,” or you had to, what symbol was it that was, it would say, I think it was a dash where it would say it was meant don’t include this. If you’re like, I’m looking for milk, don’t include dairy, you would put a dash or something that so that it didn’t come up with how products, it came up with other alternatives. My point being that I feel like us millennials were made for this, we can figure this out.
Josh:
And I would say it’s somewhere between that and learning HTML or JavaScript. I think it’s much more advanced than what Google can do, which is why there’s all this hullabaloo about is this going to be the next Google whatever. I think it’s because it is better. It also means it’s more complicated and it’s going to be really interesting to see where the industry goes. I think there’s going to be a lot of snake oil salespeople as there always is, of selling prompt design and selling prompts and stuff like that where it’s not that complicated. It’ll probably be something that’s taught in schools as a programming language or just a language in general. The language of AI, how do you interface with AIs properly?
But yeah, that a lot of people have been saying the Google analogy. I personally think it’s much more advanced, but somewhere between that and programming, which is great because it just makes programming more accessible because it programming you have to learn new languages, which is tough. And in schools, at least where I went to school, programmers didn’t have to take a foreign language because if programming counted as a language. So I think that this puts it somewhere in the middle.
Jessmyn:
Yeah, that’s wild. I actually did not know that. You didn’t know how to take another actual foreign language if you were learning, that’s wild. So in the topic of actually using this for your business, what do you need to know about your business in order to come up with the correct prompts?
Josh:
Yeah, I think this is where it kind of comes back to the fundamentals. You need to know how to convey what your business does in a paragraph or two or three sentences. You need to have content around the core pillars of what you do. That’s exactly what I’ve learned with quizzes. We’re reinventing our platform to run on AI. So you can generate a quiz that’s written in your voice based on your content without having to write anything. But in order to do that, you have to have a voice that’s already in text format that’s concise and paired down and not just a long rambling thing because if you put in a long and rambling thing, it’s going to spit out a long and rambling thing and that’s not what you want. And then also, this is what I’ve had to learn building our AI is we have to be concise about what we know around building quizzes.
We’ve been doing this for 10 years, but in order to train the AI effectively, I’ve basically had to distill down and gather data from everything that we have of how do our quizzes work in 600 words per quiz type, which is really tough because it’s a lot easier to write something and 10,000 words of here’s how my business works. But then you’re like, “Okay, now I need to train that to an AI and the AI is not going to do well if I give it 10,000 words. So how do I make that 600 words?” So I would say the skill that that’s going to be really, really valuable is learning to speak concisely because that is what the AI really thrives on is structured data. There’s a whole crop of companies that just help you structure your data now and they charge an absurd amount of money. But if you can learn how to structure your knowledge and what your business does into concise formats, I think that’s super valuable.
Jessmyn:
And let’s say you’ve got it down, you’re like, okay, I have all this information. What do you think is the first thing someone should use AI on or four?
Josh:
Yeah, that’s a good question. Whatever problem or whatever content you need to create today, whatever it is you’re working on, whatever the next thing is you’re struggling with. And the next video that I want to put out is about template. Building a template library for yourself, because I think that’s also going to be really valuable is having your own internal template library. Let’s say it’s writing Instagram captions. So if you want to write good Instagram captions from let’s say longer form content you’re doing, then you’ll write some instructions for the AI. You’ll say, read this content now write me three different summaries of that content in one sentence and here’s how to structure the sentence started off with something positive, end it with something actionable and make that sentence less than 75 characters. So that’s an example of how granular a prompt needs to be to make it actually effective.
And then you’ll have to play with it to make sure it works and give you the out outputs you want and play with the temperature, which is basically the randomness control that will figure out how effectively that thing’s going to work. So do that right with, if you’re doing Instagram captions today, now you have a template for Instagram captions. Every other time you need to generate Instagram captions from another piece of content you put out. You can just do the same thing, grab the raw content, put it in your template, run your template, there you go, you’ve got your captions. And I would say the more you can build up that library of automations for your business using the AI, the faster you’re going to be.
And the video, I think I’ll probably use the headline of 20X in your output because when I would write a quiz working with a client one-on-one looking at 20 hours, and now it’s less than one, but that’s because I’m running, I don’t know, four or five different templates, prompt templates basically to make that quiz for that client and of course working with them in between to get the information from them that they want to put into the quiz.
But yeah, I mean if you can get yourself a template library built out and just do it piecewise, the worst thing you could do, there’s the mistake that I made was like, “I’m going to write all the templates.” I did that over the break in isolation and I came back and I’m not using any of those because of course I actually start talking to customers and they’re like, “That’s not what I want.” So you have to stay connected to the reality. This is where I think the AI versus real life thing comes back in. It’s like you have to stay connected to reality, otherwise it’s going to be useless.
Jessmyn:
I think that’s a really good point. One that templates can be really helpful because the whole idea of this is to be faster, get content out faster, but also more efficiently and you don’t want to spend too much time playing around, but you also want to make sure that you’re staying true to the core of being an entrepreneur, which is to serve people. So you need to have that FaceTime. This isn’t meant to sort of replace anything. It’s supposed to help automate, it’s supposed to help you do your job better, help you do your job faster and also I guess make it easier for you so you can do what you really want, which is more financial freedom or more time and letting things run on its own so you’re not actually working longer than you should be.
Josh:
Yeah, that’s the super good point. And I think that’s the point of all this is what’s the overall thing. And we talk about this all the time at Interact. We start all of our big kickoff meetings with this of what are we doing here? This is not just about the work, we don’t live to work, we work to live, and work is exciting and it’s fun, but it’s not all the things. And so this is a way to bring those other things into your life, the things that are really satisfying, filling time with people that you really care about and being outside and pursuing activities and hobbies and all that kind of stuff.
And so I do think that’s the purpose with all this, and that’s the great thing about it is it can unlock just faster output. So it’s like stuff, and this is my opinion about this stuff with the prompt engineering, it’s a lot of these things that we’re automating were super formulaic anyways. I would sit down and write a quiz for somebody and it’s like, I’ve written basically this same quiz 500 times. The only difference is this client’s unique voice and now you can just tell the AI, “Hey, write the same quiz but with this person’s voice and then put their voice into it and it works.” So I think that’s where it really excels in terms of net new stuff. Not at all, but this stuff is just automating stuff that was already 90% reusing anyways.
Jessmyn:
So part of what we’re talking about is learning how to create a prompt to get the correct output. When you are learning a new tool or a new way to do something, there is that little bit of learning curve. How easy or I guess difficult would you say it is to actually learn this process and get it down so that way the next time you are trying to build a template library, you’re like, “Okay, I know the steps, I know exactly what I want to do. It’s now kind of muscle memory. I need to just get in there and do it and then build all this out.”
Josh:
I mean, I can only speak from my personal experience and it’s actually extremely difficult. You’ve heard this story I think, but my first business was repairing laptops and I learned how to fix a virus on a laptop back when that used to happen. It was actually wild back in the day, I don’t know if you remember this, when computers would just get completely locked because a virus would be on there and you couldn’t even access anything. So I would buy those computers and fix them, and then it took me about three days to figure out how to do that of just working long, long hours. I’ve been on this for three months and we’re like 30% of the way to our goal in terms of an Interact AI that really works in a way that’s safe and effective.
So it’s really difficult to build these things and I think that honestly, I think it’s a decision that you have to make and now everybody’s kind of getting exposure to this world of programming that I’ve had tangential access to for a while where in programming the number one question is always build versus buy versus do manually and honestly do manually is the best choice a lot of the time in programming.
And I think that’s going to be the same here where you could automate something, but I think it’s really important to kind of run a pros and cons checklist of how much time does it take me to run to do this task? How often do I have to do it because it’s going to take you… I would say however long it takes you to do a task, this is probably a good metric, however long it takes you to do a task. Writing a prompt template for that task will probably take you 10 times as long as it takes you to do the task once just manually.
So if it’s a task, and that probably is different if it’s a short task. So if it’s a short task, it’s probably like 20X. So you kind of have to run the calculation in your head, is it worth it for me to try to figure out how to make a template for this or do I just do this manually for now? And that’s just kind of a general business problem all as well. How much do you automate things and how much do you not? Because automating always takes time.
Jessmyn:
Yeah. I was also thinking of what sort of the potential of new types of entrepreneurs coming out of this who are specific to their services are running AI prompts and getting that output for you, but it’s like, is that something you should even actually pay for?
Josh:
Yes, very reluctantly. I would say it’s going to be the same as I don’t want to pick any industry because then people are going to get mad. Because it would be a service, any service that you ever pay for, think about how hard it is to actually find what you’re looking for. It’s very difficult to find the right solution for what you’re trying to do. And oftentimes, unfortunately, there’s just people trying to make money and not really deliver value. It’s going to be the same thing, especially because it’s so new. I think it’s ripe for that. It’s right for people just charging you a lot of money, giving you back templates that are just pretty useless for your business. If I was an entrepreneur just starting out and I wasn’t technical enough to make a tool with AI, that’s probably what I would do.
Honestly. I would probably be an AI consultant and write prompts for people. I think it’s a big opportunity for service providers, but from the purchasing end, I think it’s super right for exploitation where a lot of people are going to spend a lot of money on really not good services, which is going to be super unfortunate because I think it’s going to put a really bad spin on the effectiveness of AI because you’re going to get stuff back. It’s like the common story of you pay for something, the version you get back is just not what you expected.
Jessmyn:
Yeah, I guess I would say with any sort of purchase for your business, it is an investment, right? You want to invest in yourself, you want to invest in your business, and sometimes it is better to pay for something or pay for someone to help you rather than taking the time to figure it out. And it may not necessarily look right, or you might not get the results that you want, but sort of take it with a grain of salt where you know, might sometimes pay for something and it’s not what you thought it was or it’s not of the quality.
But I think that’s why it’s important to really figure out the answer to how established do I need to be? Or what do I need to know about my customers? What do I need to know about my business? And even myself as an entrepreneur and how I serve my clients before spending money on whether it is AI or even Facebook ads or picking out who your next business coach is going to be. Different things like that. I think it’s always the first step is really understanding who you are as an entrepreneur and how you serve your clients and your business before jumping into all of this to save yourself from spending money that you might not need to spend right now.
Josh:
And I would say it’s very, very helpful even if you dive in a little bit and you decide it’s not for you, I think it’s extremely helpful to have a baseline knowledge of how it works before paying anybody. So you can go watch our videos on YouTube just kind of showing what we’re doing, which is basically how we’re doing those videos, here’s how we’re actually using it at Interact and here’s how we’re using it to build Interact AI. So you can see our process and how those prompts are getting built just so you at least can talk the language because you know what always ends up happening, people use fancy words and they try to get you to not question what they’re doing and then they overcharge you and under deliver. So I would say have a baseline knowledge, at least watch a handful of videos.
Jessmyn:
I think even, I know we talked about this before we actually hit record, how last week you didn’t like the term prompt engineering and then after looking into it, now you do. But it is because when you first hear about it is, it’s like, is it another gimmick? Is this what people are just doing just to get sales or get people to watch them or listen to them or read their stuff? But it could be there’s just those two sides to everything. It could be something really, really great, but you have to be kind of careful and like you said, have that baseline knowledge.
I also wanted to call out that we do have an article that we’re going to link for you guys in the show notes called 13 ChatGPT prompts for marketers to multiply their output. I think it’s a great way to at least get started in playing around with what do prompts look like and how can I get started to even test out the product? Because if you don’t know, if you’re like me and you didn’t know even know what prompts to put in there, it’s not going to feel as cool. But if you could see what it can actually do with a prompt that works, I think even just seeing that will be super helpful and making that next step.
Josh:
Yeah, totally. Totally. I think looking at some of that stuff, yeah, I mean my initial reaction to anything I don’t understand is to make fun of it. And my fiance is always giving me crap about that because she’s like, you make fun of things and then two weeks later you love it, and everybody has a different initial reaction though some people. I think the news loves to play up the scary side of it. So there’s probably a lot of folks that are scared of AI of like, “Is it sentient?” Sorry, I didn’t mean to use a demeaning term because that’s actually a scary thought. “Is it sentient?”
And if it is, I don’t want to be anywhere near that, which my conclusion is it’s definitely not like it’s just really good at imitating humans and that is also scary. So there’s a scary response, this is dumb response, this is going to be everything response, I’m all in. And I don’t know that any of those overreactions are necessarily helpful. So I think why it’s also important to watch some stuff, try to stay away from the crypto bros termed Web 3.0 bros turned AI bros on YouTube, who will tell you can make a million dollars a day. Anybody who tells you can make a million dollars a day using AI, just don’t watch that video. So yeah.
Jessmyn:
The only thing I will say about it though is I watched this video, or it was a reel I think by Gary Vee the other day, and he was kind of talking about how in, I don’t even remember the year 97 in the ’90s when the internet was starting to become mainstream, it was in more homes. People were worried about it sort of taking jobs and they were mad about it existing I guess you could say. And he was like, I’m going to say the same thing now about AI that I had said back then about the internet is that it is here, it’s coming. So you know, could either use it, you don’t have to, but regardless it’s here. So I think one of the things, my takeaway from that was at least just check it out for yourself. It doesn’t hurt to at least just see if you like it and if you don’t like it, you don’t have to use it.
Josh:
Yeah, and I think I told this on the last podcast, episode two, but when copper was invented in whatever year that was, people freaked out and they were like, “This is from the devil.” And it happens every time there’s a change, it’s tough and change is the number one. That’s part of what I think separates us from the AIs is humans do not change. And that’s totally natural and I think valid. So I think that’s another big thing to acknowledge in all this is like it’s new, it’s very different and it’s okay if you’re an initial reaction is, I don’t want to go anywhere near this. I think it’s just learn what it’s about and then make your decision what I did.
Jessmyn:
I love that. Love that. Well, Josh, thank you so much for hopping on and talking with me today. To close out, I guess, what would you say is maybe your top three best practices for using AI in your business?
Josh:
Yeah, I think number one, always start with something you’ve created. You can try it for yourself, but I’ve banged my head against the wall 500,000 times trying to get it to make something from scratch. And it’s just a losing battle. You give it very specific instructions and then it sounds really boring. You give it general instructions and then it just makes stuff up. But then if you just sit down and you write three sentences, that’s exactly what you want to say. And then you tell it, “Hey, make this into five other versions.” Boom, every single time. So I’d say that’s number one.
Number two, I think learn the syntax, I think is the name of it. It like the structuring labeling, and I think this is the labeler dream. The person that has all the post-it notes on everything. This is heaven. Everything has to be labeled when you’re building a prompt label. Every single item. This is a quiz question. This is a quiz answer. This is a quiz answer score. This is the next quiz answer. You give it that and it excels. You don’t give it that and it makes it up and you’re like, this is messy. And those are number one and number two.
And then number three is like tell it exactly how you want the output to look, because it tends to ramble. It really tends to ramble as do humans, which is part of I think the human footprint or copycat, whatever you want to call it. You ask a person to explain what they do, and the average person is going to go on for 30 minutes-
Jessmyn:
That’s literally.
Josh:
You ask them to. That hurts, right? It’s like it’s a natural tendency, but then you ask him. So getting the AI outputs to be short and concise is the hardest part. Getting it to give you a concise answer that contains all the information that you actually want, super difficult. So I would say those would be the top three.
Jessmyn:
I love that. That’s great. And we do have a ton of videos on our YouTube already for different types of even AI systems. Josh teaching you how to do prompts, we’ll link that in the show notes. But if you look us up @tryinteract, they’re the most recent ones, and we’ll link that 13 ChatGPT prompts for marketers to multiply their output articles. That way you guys could check that out. Start playing around. And thanks for listening. We’ll see you next time.