#129 - ChatGPT & A.I. Expert - Kevin Williams

The Sprinkler Nerd Show

In this episode, Andy chats with Kevin Williams an expert in ChatGPT and A.I.

==========

References:

www.SprinklerNerd.com/inkworks

www.InkWorks.ai

==========

Kevin: You know, it's not going to be AI that replaces you as the employee or, or supplants you, your company. It's going to be a company that knows how to use AI or a person who knows how to use AI that's going to disrupt things.

Andy: Hello my friends. This is Andy. Welcome to episode 129. Of the Sprinkler Nerd Show, where it's my job to speak with world-class water and technology innovators from all walks of life so that it may inspire you and your business. My guest today is Kevin Williams.

Who is Kevin Williams?

Kevin has been featured in Inc.Magazine, The Wall Street Journal, Financial Times, and even as a Shark Tank business. Before starting his current company, www.inkworks.ai, Kevin was the former operating partner and CEO of www.balls.co. And before that, Kevin was the founder and CEO of Brush Hero, which is the product you may have seen on shark tank.

Our conversation today will be focused on AI tools like Chat GPT, and how you can implement these tools in your business. So with that, Kevin, welcome to the show.

Kevin: Thanks so much for having me, Andy.

Andy: I cannot wait to talk about AI and how service businesses, contractors, irrigators, and landscapers can learn a little bit from you, who has spent a lot of time, uh, really becoming an expert in this field. And I think that before we jump into that, I'd like to ask how you got your start in business and as an entrepreneur.

Kevin: It's, it's funny. I actually come from a family of entrepreneurs ever since I was about 10 or 12. My family was traveling all over the country with various business ideas and it was just part of the fabric of my life.

Kevin: Sadly, that story doesn't actually end particularly well. So sometimes I, I, I glib about it that I come from a family of failed entrepreneurs because in a period in my adolescence, my parents lost their business, they lost their house. They lost their marriage, like all of this horrible stuff. So young Kevin decides that a good idea is to not be an entrepreneur and instead go be a chemist.

Kevin: Well, fates have a way of, uh, of messing with plans like that. Um, I went on the straight and narrow path. I did a bunch of interesting stuff and I ended up at pretty good business school. And in business school, I entered a business plan competition just as a part of a, like an elective entrepreneurship class.

Kevin: And I won. And I won a bunch of money that came along with it for seed funding. Um, so I ended up starting my first business having done everything in my power not to be an entrepreneur. I was like, oh heck, here's an opportunity. I'm just going to take a left turn in my life and chase this now. Um, that business didn't necessarily go anywhere, but it introduced me to the angel and venture community in my town in Washington DC and uh I ended up operating businesses for a high net worth, uh, individuals for a bunch of years and my own entrepreneurial journey kicked in again, where I saw that there was just so much waste in a lot of startup companies that people really didn't know how to demonstrate.

Kevin: What we marketers would call product market fit, and instead they just dump bucket loads of money into things trying to prove a concept. Uh, and when I saw the rise of social media, I saw an opportunity to rapidly test concepts, um, without necessarily spending a lot of money. And that pivoted into a whole series of businesses where I would either license or buy intellectual property.

Kevin: And my dirty little secret was that when a patent was pitched to me, I could go out onto social media and test some concept around that product. I could throw a bunch of traffic at it, see if anybody cared. If people cared about the idea, then I would license the patent and then I would already know that I could get on to first base with the product.

Kevin: Was it going to be a home run? Who knows, but I could get on to first base. So that led to the Brush Hero product, which I had licensed. I'd licensed the underlying IP from a gentleman in the UK, um, and several other patents in homewares and kitchenwares. Uh, I sold, um, or I, yeah, I exited Brush Hero in about 2019.

Kevin: And, um, then I ended up running, uh, a large international brand. Usually I don't say, but yes, it was Balls. co.

Andy: Fuck it, you can say it on this channel, on this show.

Kevin: Yes, I was a manscaper. Um, so Balls was the largest, uh, manscaping company in Europe. Uh, you can probably already tell I'm not the guy who tells Balls jokes all day.

Kevin: So it was, it was pretty fun to dive into a brand like that. British sensibility, really cheeky humor. And, um, our goal was to drive it into, uh, the U S with that sort of humor. Um, the realities of running a UK European based business from the West coast of the U S not so great. A lot of early mornings, a lot of late nights.

Kevin: So, mm-hmm. , in part when I saw just the I I, I, I like to think that I immediately saw the opportunities that generative AI would represent when G P T launched in November of last year, and I left and dove feet first into generative AI and practical applications of it. Um, And I've been rooting around for business models in my M.

Kevin: O. You know, test some ideas, test a lot of different things, um, to see what might take root. And from there, ink works is one of several different products, projects that I'm working on, um, as well as doing executive coaching and executive coaching oriented around a I capacity development within organizations.

Kevin: Because one challenge of all of this Is that coming up with a one size fits all solution just isn't practical. So business leaders need to develop a framework around the way that they think about AI and how they're going to safely lever it in their business. Um, as opposed to just looking for a magic bullet type.

Kevin: Platform that they can just buy. That's going to solve all of their problems. Um, that's going to be very interesting to, to, to see how that develops. And it's been fun to, to, to work with other business leaders to try and identify how their particular business, be it, you know, in landscaping or direct consumer or.

Kevin: Business to business SAS type stuff. Well, how can they actually deploy this stuff right away to make changes in their business? Because, you know, the, the adage has become. You know, it's not going to be AI that, that replaces you as the employee or, or supplants you, your company. It's going to be a company that knows how to use AI or a person who knows how to use AI that's going to disrupt things.

Andy: I love that. So there's a couple takeaways. I'm going to start with the last thing you said, because it reminds me of a great expression that I can't remember who the author is, but I use it all the time. And that is the company that kills you will look nothing like you. So when you said AI may not replace the person, it's going to be a company that knows how to use AI that becomes your competitor.

Andy: That's a great example of another company that It looks nothing like you, but could end up killing your business and you were running balls. co and this is not the right time to talk about balls. co, but we don't actually talk about a lot of balls in this industry. We do talk about a lot of nipples though.

Andy: There are many different types of nipples in the irrigation industry, believe it or not. So I'm just going to, I'm just going to put that out there inside joke for those that are listening. We don't talk about balls, but we like to talk about nipples, talk about

Kevin: turf.

Andy: And turf. Yep. Totally. You can talk about turf.

Andy: There's a lot to, a lot to play with there. Not last week. It's been probably three weeks now. Kevin and I both went to a conference. I would say that's just for shark tank companies, just for those who have. been on Shark Tank, whether it aired or whether it was just taped, because we know that most of the businesses that tape don't actually go to air.

Andy: So we were both at the conference, and that's when I was learning about what you were doing in the AI space, because Kevin was actually presenting at the conference. And I thought this would be great, Kevin, to have you come and share some of your Uh, real practical world experience with AI, you know, and how you are coaching people to use it, some of the value that it has, and maybe even some of the best practices or things you should do first, second, third, or even how do you optimize the responses of, let's say, chat GPT versus a beginner that just goes in and asks it a basic question.

Andy: So very, very excited, and especially because this industry is tends to lag behind.

Kevin: So first, just to back up, I we were sort of operating on the assumption that everybody knows what this is, and I'm pretty sure everybody has at least heard of it at this point. That is this magic machine that can that you can talk to, and it can it can come up with responses.

Kevin: Um, but it is actually a success story. That's it's one of those overnight successes. That's eight years in the making that billions and billions of dollars has been poured into what are called neural networks that allow So Uh, highly abstract patterns to to interact with each other such that the magic machine can output based on a predictive model.

Kevin: What might come next from a thought? So that's essentially what it's doing. It's predicting from the sum of the human Internet knowledge. What? The next likely thought can be, and it is absolutely amazing what it can do, but the underlying fundamentals of neural networks have been around for a long time.

Kevin: The novelty and what was just completely mind blowing for most of us was the Interaction, the interactive effect. Like if you leave a bunch of wonky people together who are studying neural networks, they know how neural networks work. They don't need this chat functionality. What the chat functionality did is it made it much more accessible for we mere mortals to be able to lever these tools, um, on on even on a basic level, as opposed to going through a whole machine learning type process.

Kevin: So These are predictive models. They're taking the sum of human knowledge and they are outputting the next likely. So the first thing to understand about them is that They don't necessarily know or care if anything is particularly accurate. So, this is what you hear about in terms of hallucinations. And hallucinations are just wrong facts.

Kevin: Like, the AI is not particularly good with facts. It's very good at expressing A dubious fact in a very convincing way, which should be a giant red flag for most of us who produce any sort of content that particularly in a subject matter that's relatively technical like what you guys are talking about, um, it could easily.

Kevin: It could easily just lie to you. So the first thing that I tell people from a, from a mindset perspective is that you need to calibrate what you're doing with the AI based on who you are and what you know, so picture like a Venn diagram, you've got this. One circle, that's the size of my house, that is the sum total of human knowledge.

Kevin: And then you have this intersecting circle that's much smaller, which is the sum total of who you are and what you know and what you know about irrigation and, uh, and lawn care and everything else. Right? And the intersection of those two circles is where the power really lies. So If you, the farther you drift away from that, the more likely you are to get into dangerous territory.

Kevin: So, I know a lot about digital marketing. I know a lot about business operations and such. That is a core of who I am. But, if I drift away and I start talking with the AI about neuropsychology, I might get interesting results, but I have no way of calibrating whether or not those results are actually useful or, or practical or not.

Kevin: I'm just leaving it to the AI. So when you say

Andy: calibrate, what does, what does that mean? What does calibrate mean? So

Kevin: it's you know what you know. So imagine, you know, most of us have have businesses that are large enough that you have developing staff like there. There are other people that are involved in the business and you you take, let's just say a new sales guy and You, The way If you're the senior sales guy or you're the business owner, you might tell the sales guy to go off and do X, Y, and Z.

Kevin: And then you're going to look at the output and you're going to, you're going to coach them, you're going to push them towards an output that you know is going to work in because you have this expertise in the knowledge. It's the same as true for the AI. The AI doesn't necessarily know what it's talking about, but if you were to look at the output.

Kevin: Your art as a business person and just as an individual is being able to identify the value in that output. And if it's something you don't know anything about, that's going to be really hard to do. So if you're, if you're looking at, at creating something that's entirely new that you don't know anything about, there are ways to use AI that you can do that.

Kevin: But it's not as effective as Amplifying things that you already do know. So in a lot of organizations, let's just take a lawyer, for example, like you could you could call a lawyer and say, Hey, I need to set up a trust document and whatever. And right now the M. O. would be that lawyer would probably record the call or take notes on the call.

Kevin: They would go to their associate. Their associate would look through their templates about it. Trust. They'd adapt it to Wyoming. They put it back to the senior attorney who would then approve it, edit it, give it red lines, hand it back to the guy or gal and then process it and then finalize it and then send it out because that senior attorney really knows their stuff or you hope they really know their stuff.

Kevin: They can do that. That is their art. That is their job. That is their profession. But now you can bypass all of that, that associates job. Not so good for the associate, right? But you could output that document and be able to read it and have it done in 15 seconds, but you can't abdicate your professionalism and your art.

Kevin: You can't just trust it. You're going to get 80% of the way there in 15 seconds, but that last 20% of editing and clarifying and redlining, um, you still own that at least at the moment. So. The lawyer knows a lot about law. The business guy has actually read a ton of contracts, right? Like, I've probably read a thousand contracts in my career.

Kevin: I'm not a lawyer. I happen to be married to one. But, I... Not a lawyer, but if I need to create a new contract I can actually get 90% of the way there So let's just say 70% of the way there Because I know how contracts are written right and I can read it and I can interpret Okay, this indemnification clause makes sense to me The smart move is to then send it on to the lawyer, but I didn't have to spend the 500 for him to draft the first version.

Kevin: I just need to spend the 250 for him to take a pass at it at the other side, because I know enough to be dangerous. Now, if it were to get into case law, statutes, regulations, things like that, it could easily lie to you, and that's out of my realm. Like the lawyer might recognize that that case isn't a real case or that that statute isn't accurate, but dang, if the AI isn't going to be very, very compelling in its, it's sort of its defense of its own facts that it's putting forward, but that's the.

Andy: So, so would it be. Would it be safe to say that an attorney who uses Chet GPT, if Chet GPT or the AI can do the 70% as you describe, but because they're the expert in that field, they can review that 30% and get it right. So that if it's lying, they can correct it because they have the expert knowledge in that core business.

Kevin: Exactly, exactly. And this is where I like to focus when I'm talking to people about it. There's a lot of water is wet out there. Oh, you can just. Have it write a giant blog post for you. Okay. That's cool. You know, it's cool to watch it do its output. It's like, it's sort of mind blowing if you haven't seen it by all means, totally go sign up and see that because it's really cool, but that doesn't allow you to abdicate from your art and your expertise.

Kevin: So, you know, your audience knows a lot about lawn care and it like, like you can have it create a blog post about certain patterns of irrigation and you're going to be able to decide whether or not those are accurate or not. But if you want to reach into topics that you don't know much about, even if they're close to you, you can.

Kevin: But you have to have either some sort of validation mechanism such that you can determine whether it's accurate or not, um, or not care, so. Because

Andy: then if somebody who is an expert in that category reads it, they may think, Oh my gosh, what is Kevin talking about here? He doesn't know what he's talking about.

Andy: This is not accurate.

Kevin: Exactly. Like imagine, you know, I, as I understand it, that, uh, you know, grass varietals change by different continents and there's expertise in South America and they're, you know, sprinkler nerds in South America and like you pontificating about, you know, Argentine varieties of.

Kevin: Bermuda grass, like that person is going to be able to smell a rat because that's their, their expertise. And worse, this is sort of meta as there's an industrial scale opportunity for content production. If all of us. I'm not going to get noble about this, but like if all of us are out there producing bad content, the AIs will be trained on the bad content.

Kevin: So there is going to be value. Is that the

Andy: garbage in garbage out analogy? Yes.

Kevin: Garbage in garbage out. And at some point it all reverts to the mean. So from the segment of your audience that is out there and doing direct to consumer type marketing, don't be, don't be tempted to do just. Industrial scale output.

Kevin: Your art has to be producing new information from somewhere. But what AI can do is it can make some of that new information really accessible. Like there's a lot of geeky in this sort of field, right? And there's scholarly articles about soil density and all this other stuff. One really cool use of AI is to be able to contextualize something like a scholarly document and make it accessible to people who have expertise that can do something interesting to it.

Kevin: So, you know, somebody comes out with a paper from the university of Florida, as far as water absorption rates, whatever it is, and you can then use the AI to simplify that overly complicated document to a way that it falls into that zone of expertise and art. And then you can actually. Add to the corpus of information that's out there on the web in an additive way because that paper was never really going to get found.

Kevin: It was somebody's PhD thesis or whatever. But now you Andy can like actually make that accessible in a way that increases the store of human knowledge and from a Strategic perspective, I do suspect that, that, that brands, particularly in the internet who can truly add novel value are going to be rewarded by search engines, by advertising platforms, et cetera, and that those who simply put it out like high volume garbage are going to get severely punished.

Kevin: And, and I'm

Andy: thinking that likely the level one knowledge. Which may address the most frequently asked questions about lawn care on the internet will probably be garbage in garbage out and stuff that everybody talks about. I love what you said about finding a scholarly article and what came to my mind is that there actually are scholarly articles from, I believe, University of Florida on, you know, lawn care and let's say soil moisture sensor technology.

Andy: And my question would be, number one, Perhaps this would be a great training example for us to do live like, Hey, let's grab a couple articles and use that to produce some really awesome content using AI. And could we do that? You know, could we take an article of a research on soil moisture sensor, not right now about soil moisture sensor technology, real case studies and recreate it in a, in a, in a way that everybody could understand it simplified, but on a deep topic

Kevin: like that.

Kevin: So this is, let's walk through the practical example. The example is yes, that would be really cool, right? So first you're going to find the article and then let's just be practical. First, if you're not paying for GPT, pay for it. It's 20 bucks a month and it gives you access to GPT 4, but more importantly, it gives you advanced access to advanced processing.

Kevin: So the, the 3. 5 was the first model 4. 0 is where it is now. 4. 0 is roughly 10 times. It's more powerful as far as the level of connections. It's also slower, um, which can be a little bit agonizing in a demo because it writes really slowly. Um, but it allows you to contextualize. These are, these are all terms that are going to be so common in the next few years, but right now we're all kind of bending our heads around it that you have to set context.

Kevin: Like, uh, I like this particular example, like you go, you stand at the top of a, of a building at, uh, you know, Times Square, you stand in the middle of Times Square and you say, what should I read? And people are going to have all kinds of opinions. They're going to have like, Oh, you should read the Bible.

Kevin: You should read, you know, Tom Clancy. You should read, you know, the, the sprinkler digest of 2022. The Idiot's Guide to Landscaping. But that's because, yes, of course, you know, scintillating reading, right? But that's because nobody knows anything about you. So one of the first keys here is you have to set the context of the conversation such that you're narrowing, you're narrowing what you're after.

Kevin: So from a practical perspective in GPT 4, you can start out a conversation by saying, I'm a landscaping expert. Um, I'm interested in expanding my knowledge of lawn care practices using scholarly articles. And it's going to say something like, yay. Next you set the context for the conversation because you could just.

Kevin: Continue. And this is where it gets dangerous. Like, let's just chat about lawn care. And you're going to come up with all kinds of interesting stuff. In the back and forth that it's, it's, as you're, you're expanding. It knows who you are. It kind of knows what you're looking for. But now, you want to refine that context further.

Kevin: And the way that you do that is by contextualizing something like a scholarly article. And there are tools, they're called plug ins within GPT 4, that allow you to do that. And that's simply by Letting it ingest the PDF, and now this is what we're talking about. We're not talking about the body of the knowledge, of world knowledge about lawn care.

Kevin: We now have established minimal context that you have expertise in lawn care, and now specifically what we're going to talk about is this scholarly article. Like, use, oh mighty GPT, use your chat based functionality to make this easy, but this is what we're talking about. So now you've set the context for it.

Kevin: And you're going to do something like, let's ask for a summary, um, that would be applicable and interesting information for an audience that is focused in on lawn care science. And it's going to come up with a bunch of ideas. Okay, cool. Now. Let's say, oh, okay, I like idea number three, that, um, you know, I don't know, relative humidity and the impact of, uh, water absorption rates on whatever it is.

Kevin: Now let's dive into that, and let's put a marketing hat on. Okay, let's produce content, a blog post about this that, that, that incorporates interesting facts from the scholarly, the scholarly doc, document, and dresses it up with a little bit of marketing speak. Okay, cool. Now, because we're a marketer, we need to put headlines on it.

Kevin: So let's come up with 10 possible headlines for this. So now you have 10 possible headlines for the article. Now let's get a little bit wonky because this is a scholarly article. Scholarly articles often come along with data sets. Okay. So you could actually ingest a dataset using the, the, the code interpreter function within GPT and say something completely simple, like some giant dataset, and just say, help me visualize the data in this dataset in a few different ways that would be interesting to my audience, my audience.

Kevin: Like you've already defined who your audience is, right? It's another cool part. Like it has permanence. Um, So it's remembering

Andy: what you gave it earlier when you said, you know, my audience is homeowners, you know, interested in black, it, it,

Kevin: it, it stores that. So we've, we all, we all, I'm not going to say the name because it'll trigger, but the S device on a, on an Apple, if I were to say the name and I would say, Hey, what's, what's the weather in park city tomorrow?

Kevin: It's going to have an answer. And if I simply said, what's the weather on Saturday? It's going to say, what are you talking about? Because it has no permanence to it at all. You have to start over in that conversation. Permanence in GPT is so cool. So just a practical tip. You have chats. that maintain that context.

Kevin: And some of my chats are now hundreds of pages along because I'm chatting through specific business models and it knows that that's what we're talking about. It doesn't need to like remind itself. I can go back months later and bring something up and all that context is set and you get much better results once the context is set.

Kevin: So what you've done with that. So if you, if you have a

Andy: thought. Or you have another question, but it's really related to some other things you've already asked that you'll instead of starting a new conversation, if that's what it's called, you'll go back to your other one and add it into the dialogue.

Andy: Yep,

Kevin: exactly. Just write, don't even need to be like, do you remember what we're talking about? No, of course it remembers what it's talking about. It's a machine, right? Um, but imagine you've put out that blog post and, um, somebody now in the community has some insightful question and you're like, I don't know what the answer is.

Kevin: Dump the question in and say, this was a community conversation. Can you, can, can you come up with some sort of, can you help me answer this question? And it's going to use the context of your chat. The conversation you have, it's going to use the document that's been set as context. And it's going to try and answer that question within that much narrower context, um, than just the wild west of the internet.

Kevin: So taking it like just taking it to the logical conclusion again, as a marketer, you need visuals. So now we haven't talked about the visual tools at all. I've been very GPT focused and GPT is not the only language model out there. I just, it currently is the strongest, but. In my opinion, but there will be many, there's no real barriers to this except gazillions of dollars, which people like Mark Zuckerberg have.

Kevin: So you're gonna see a lot of different models and this is just, it's going, they're all gonna be out there. So people will choose their poison. Do you know,

top

Andy: of mind what a couple other models are that we could share list in the,

Kevin: so Google Bard is quite powerful, and it's not like Google wasn't working on this.

Kevin: They missed a tick. They have business model problems with this that are pretty obvious. You know, they make 160 billion a year off of advertising. And what does advertising mean if. Like you get the answer, it's not so great. So they

Andy: also not perusing the internet and clicking lots of times and visiting lots of pages and getting served.

Andy: Lots of visuals. We're

Kevin: in that space, right? Yeah. So it's, it's, it's going to be an interesting existential crisis for them. They seem confident about it. So I think they have a plan, but they're constrained. My, my worry is big brands like that get more constrained by reputational impact. We all have heard the stories of the New York times, I think.

Kevin: Kevin Roos, um, who like the AI tried to convince him to leave his wife and stuff. Um, like open AI, which is GPT can kind of get away with that with its. I'm a 10 billion startup thing, but Google has to worry about that. So naturally they've limited their model more. So there are all these instructions and there's a term that's, that's, that's.

Kevin: It may be permanent, but at the moment, I'm not quite sure it's called a constitution. And it's this idea that there's, there's an operating, we call the, the, the, the, anytime you type something into a LLM, it's a prompt, but there are all these hidden prompts that are behind the scenes. And those hidden prompts are, let

Andy: me catch you right there.

Andy: You, you, you mentioned a buzzword that I want to make sure everybody knows you said.

Kevin: Large language model. So a GPT is one of the large language models. Um, Lama from Meta and Facebook is another one. Um, Google has its own that, that underpins BARD. Um, these are all, they've all done the similar thing where they've subsumed.

Kevin: The Internet and are making these connections. Um, and then, yeah, that's a GPT is not the generalized term term. It's that it has that has to do with technical language transformation. So GPT is actually a technical term. Yeah. So anytime you put a prompt into these things, that's a set of instructions that the AI is then trying to follow.

Kevin: But there's a whole set of hidden prompts behind the scenes that are basically don't be psychotic, like Try not to say like racist stuff. Try not to like incite violence. Like, don't try, don't answer legal questions in a way that could be misleading. Like, it's, it's, it's like this whole giant set of things and, you know, building that constitution into the model, um, the, the, sort of the strength of that constitution ties into, this is slightly wonky, but it ties into the, the, the, how crazy the outputs can be.

Kevin: And there's another term in there that's called temperature. So the higher the temperature, the more likely it is to go batshit. That it's going to start making And what does a high

Andy: temperature mean? What,

Kevin: what is that? So it's a, it's like a continuum, like low temperature is cold. Just the facts, man. And stick to the fact Right.

Kevin: This is what Okay, I see. High temperature is, we're going to loosen up. That loose that that that neural network and you know, I'm being I'm trying to paraphrase a little bit But like it's gonna loosen up the neural network and allow the network to make kind of wilder connections between things

Andy: Okay, so it's called something that's extremely factual like one plus one is two would that be very very cold

Kevin: not factual So that you got to be super careful like it's pretty okay.

Kevin: All right high level of probability that 1 plus 1 is 2, but, but some of these models are very bad at math. Um, because they don't, that's not what they do. They're, they're predicting that one thought follows the golden rule. Okay, we have a lot of information about the golden rule. We, we are really, really comfortable that the golden rule is due unto others, right?

Kevin: So that low temperature, it's going to connect this extremely high temperature. It may come up with something like Hmm. Maybe that means something different. Hmm. Let's just like connect things. So it gets creative on its own? It's creative. Don't anthropomorphize, but it's, it's easy to do, but it is, um, it just gets looser in its neural connections and it can be very powerful in terms of being extremely creative.

Kevin: There's an example that's, uh, I like that. So a high

Andy: temperature means more creative.

Kevin: Yes. And most of the models by default operate at a relatively low temperature because, That's where the you should leave your life wife and marry me stuff comes in where it starts like It's crazy. Like, don't get me wrong.

Kevin: Like, researchers in this space, they anthropomorphize it because it's doing stuff that they don't understand.

Andy: Well, I'm just wondering, um, number one, I'll ask you, and you could answer it now or later, is this, is this regulated? Because if it's high temp and it's super creative, which means it may not be accurate, should there be a disclaimer in the response that must be included if you use the tool?

Andy: Because the answer may or may not be correct because it's high temp and where, where do we draw the line? Or is there a line being drawn on telling someone disclosing the use of the tool?

Kevin: So this is where Europe is heading. Europe is heading to a disclosure of AI, and I think we may see something similar.

Kevin: To this in the U S at some point, but it's a commons issue. Like that sort of disclosure is only as good as the compliance of the community and the enforcement mechanisms that make that happen. And I have doubts having lived in the digital marketing trends as long as I have that. If there's an edge to be had, people don't have to use these models.

Kevin: So that's, we haven't really talked

Andy: about that. And right now I can write a blog post about whatever I want, factual or not. It's up to someone else to actually decide if it's right or not. I don't have to, there's no disclosure I have to put on it currently.

Kevin: And you bury it in your terms anyway, in the bottom of the fine print somewhere.

Kevin: Yeah. It's like an affiliate disclosure that, um, it is possible that artificial intelligence was in some way used to construct this particular note, this particular route. Right. So maybe by

Andy: default, someone would have to trust the author, i. e. trust Kevin, trust Andy, trust the author. Then you'll trust the words, but don't trust the words.

Andy: All by themselves, unless you trust the source, which is essentially where we're at today anyway.

Kevin: Yep, exactly. Do you trust the source? So authority, you have a lot of authority in this space, because you have such a great community, and you know, there's a lot of energy and output and such. That is why search engines reward you, or I assume reward you for that output.

Kevin: Mm hmm. The same will be true in these models that, that, and that's the winners will be those who have a lot of authority and a lot of credibility and that will make it very hard for new entrants to, to batter through. In my opinion, there will always be shenanigans or tactics that are designed to like break through the model and try and get something that to get attention.

Kevin: Um, but I think it's going to be a lot harder than it has been with a search engine optimization SEO over the years.

Andy: Let me ask you a quick question. Do you think a service business? Landscape contractor, landscape maintenance, service business, irrigation, you know, they could write an article about, let's say, turf grass management, and they could write that article with the audience being the world, yet that, with the audience being the world, that pool is extremely competitive, which makes me think that they should use That's AI to write something more hyperlocal so that they're found with somebody in their service area, and that's what matters.

Andy: It's like lawn care maintenance in Peoria, Illinois, and being the expert there, but not Santa Fe, New Mexico.

Kevin: So I think, would that be the right way to think of it? I think it would. And you and I chatted briefly about Sunday. I mean, that might be a polarizing company in your world, but they do from a sales mechanism.

Kevin: They're very, very good about using satellite imagery and sort of loose connections about soil density and soil construction in order to get you into their marketing funnel. And they can do that because they were extremely well funded. And, you know, they, they spent a lot of money trying to figure this out.

Kevin: And the fact is there's really nothing right now preventing a Peoria, Illinois provider who knows about soil to be able to output like sort of micro geo content. Based on the information that they have to, to educate the Peoria, Illinois population about the very specific aspects of their soil on a level that someday can never touch because it's just microform.

Kevin: They're too

Andy: big, right? Exactly. They're the authority in this particular area based on their experience just in this area, which someone who lives in that area, they would want to hire someone. That knows a thing or two about that specific location.

Kevin: So imagine you have your scholarly article, let's put a few of them in there that are about like how to manage, you know, pH, whatever it is, and then you can put a data set or even even sort of qualitative information.

Kevin: Well, I happen to know that Be embarrassed to show you guys my lawn, but the, uh, like I happen to know I'm in a high clay area and like, I don't really know what that means, but like, you, if you feed all of this in, you could come up with a very practical micro guide that's very effective without necessarily having to do the whole lift of, of, of doing the brain dump.

Kevin: Of everything, you know, about high clay environments. Um, you could use the AI as your assistant to relatively rapidly output that information. So, I mean, honestly, that's, that's very practical. Like anybody listening to this. If you're that hypothetical Peoria, Illinois, like provider, you should totally do this.

Kevin: There's SEO value to that, like as far as having content that, that a local, like long care tips in Peoria, you know, right now you might actually find, if you were to type that in, there are all of the dynamically generated, like SEO articles about there that do that. But that's kind of crap content on the inside.

Kevin: If you have authority as somebody who's in that community. Plus, you know, being a professional organization there, plus offering this micro content that's useful to people. You wouldn't have bothered before this. Now you can do it. You can do it this in like a half an hour. Like this is, it's, it's way easier than it, than it would have been.

Kevin: And. Is it going to change your business, but is it going to, on the edges, allow you to build out just this, this corpus of credibility that, that in a very SEO sort of way can follow you over the years? Yeah, that's very plausible. Wow, this

Andy: is really hands on. I mean, I think we could, there could be a ton of value in actually doing a live workshop that could be recorded for people to see later, but take similar businesses, i.

Andy: e. irrigation, contracting companies that number one, all have to have a website. I'll know exactly the type of work that they do and all know exactly their, who their best customer ideally is create some content, you know, copy pasted onto their blog, you know, know what the traffic is now. And then over.

Andy: let's say three, six months, what happens after the end of six months? Could we get a cohort together that becomes like number one ranked in all of their local areas through a quick training demo seminar?

Kevin: There's value in that. I think that'd be

Andy: fun. Wow. Okay. Well, we can't show people actually how to use chat GPT today, but I love how you talked about some of the context because I feel like that's what I've had to learn the most about is not just asking it a simple question, but creating that frame.

Andy: Um, and believe it or not, I learned it from my son, who was apparently was taught how to use this in college, and he's a computer science major, and you know, he uses it actually to correct some of his code when it doesn't work, uh, among other things, but he was the one that taught me you got to basically, you know, tell it who it is, what its job is, all those sorts of things to frame it.

Andy: Which I had no idea about and I think that a lot of people may use again Just chat GPT and then say, you know, I tried it, but it didn't give me the results. So yeah,

Kevin: I'm done And so let's let's wrap a little bit of truth practical other practical ways that I think that everybody should be using it and cool, it's you know ranging from just dead simple to much more complicated but the Most simple bit is none of us have any excuse to have blank page syndrome again Like, some people are talented content creators.

Kevin: A lot of people aren't. I am not. I, I actually can write, but it is an agonizing process for me. I'm not that guy who just can hammer something. And I have blank page syndrome. I sit there, I look at the page, and I kind of play with some words, and I'm like, eh. We

Andy: don't have that. Then your mind starts to hurt, and you'll go, eh, I'll, I'll try it again tomorrow.

Andy: And then it's just repeat, repeat, repeat, and you never frickin do it.

Kevin: So now it's like I need to write a letter to X, Y, and Z client, and this is the sort of stuff that it can't contain, blah, blah, blah, blah, blah, and bam, you've got a draft. And then my art and my effort is spent on revising that draft and personalizing it and putting the me into that draft.

Kevin: But I'm already, I've gone from zero to eight, like in 15 seconds or, you know, it's spending a little bit of time. Going back and forth with it working on tone, you know, bringing that down a little bit. So that's one It's just you you don't need blank page syndrome again Like you just start with something and then work them in related is a brainstorming partner is like Trying to isolate good ideas.

Kevin: What's a bad omnipotent Personal assistant, um, next to you who knows everything about everything and occasionally lies to you, but they're very enthusiastic about it. So pretty cool to have like this brilliant thing that you can like bounce ideas off of and none of it's perfect, but boy, does it come up with some just interesting things, particularly if you, if it's like come 10 actionable headlines for this topic, like, oh, that's kind of interesting.

Kevin: Like, Oh, that's neat. Let's explore this a little bit more. Um, also related. Um, you know, I think a lot of small companies struggle with creative design and creative development and are frankly beholden to a lot of creative like people and agencies out there that charge a lot of money for it. Being able to use this as sort of a creative designer assistant, again, you're not going to get to eight or nine in this case, if you're not already creative, but being able to use it to get to, you know, six, seven, eight, and like maybe script out a video or a piece of content and have an idea of what that's going to take to get it done.

Kevin: Oh, you know, what sort of camera angles might I use? And then you walk into the conversation. with that creative partner, and you're way more equipped. You have a good idea of what the storyboard looks like and what it feels like. And then their art is layered on top of that because they don't know your business.

Kevin: Like as much as we all love the idea of having a creative agency that knows everything about us, they're busy and they might know something about your business, but you will always know your business better than they will. So if you can kind of skip that phase and get to the, the, the creative production part.

Kevin: That can be super, super useful, um, correcting documents, um, or assessing documents rapidly. Just, just being able to absorb information like in your guy's world, like every time a new regulation comes out or if, uh, you know, scholarly articles, if you're a real geek, you know, whatever it is, being able to rapidly ingest that information in a way that, that you just, you have it in your to do list.

Kevin: Okay. Like I have family members who love to send me Atlantic articles. Like they're always like. 15 pages long and like, I just can't, I like don't have the bandwidth to read it. So I'll put the article in GPT. I'll summarize it. I'll frame who I am, uh, and like be able to come up with a summary that, uh, that, that, that is appropriate for who I am, that, and then I can decide if I'm going to engage in whatever, wow.

Kevin: The first thing

Andy: I think of when you say that is, could we take, let's say the national plumbing code and use it to help understand what the requirements are for irrigators as it relates to the national plumbing code on what you can and can't do and what the laws and regulations are.

Kevin: Theoretically. Yeah, you could.

Kevin: Wow. Your art. Don't forget your art. Like, you know, there's a code inspectors, like even you guys probably don't totally know, like, Where that line is, and it would make me a little nervous, um, to do that, but I bet you'd get some pretty meaningful output from it. Um, it would be an interesting test actually.

Kevin: And just maybe

Andy: a summary format, like, Hey, can you summarize the national plumbing code and what irrigation companies should know are responsible for, you know, in a

Kevin: summary. So I literally did something like this. It was an OSHA regulation for scissor lift safety. And, um, I put in the OSHA reg and I asked it questions about like, this is not something I was doing as a hobby.

Kevin: There's, I have a client who is in this world, just to be clear that I, yeah. You

Andy: weren't just going to rent a, yeah,

Kevin: a lift and go out there. But I was, I was, I was. Clean your windows. About like, you know, I have this situation, like there's a two 20, um, junction right here. Like how far away does the scissor lift need to be?

Kevin: And it did an ad, it did a very good job. And I think let's think ahead a little bit. Well, what, like this is all cool. Right. But. Man, this is going to be powerful. Like you imagine that you're an onsite contractor and you've run into X, Y, and Z scenario, like right on your phone, like, Hey, I've run into this.

Kevin: You know, there's a T junction of whatever flow rate. And how does this apply to, um, you know, the code of. Peoria, Illinois and flow rates that will yeah, no doubt about it. Right.

Andy: Exactly. So today, uh, we've got something called friction loss, uh, friction loss charts. So those listening likely know what a friction loss chart is.

Andy: It'll tell you what the PSI loss is per hundred feet based on a specific flow rate and a specific pipe. And if somebody were out in the field today, it's very hard to have all of those things memorized. It's actually. probably impossible to have them all recalled in your head, but just to be able to ask, ask it, Hey, what's the PSI loss on 2.

Andy: 5 inch PVC pipe or a hundred feet. Cause you want to make a change to your design. You need some quick engineering facts.

Kevin: Sounds interesting. That's going to be a thing. It will, it may already be like, that's, this is changing. So, so fast. I mean, I literally was in the middle of the presentation and when Andy and I were in Vegas and a new feature was released in GPT.

Kevin: As I refreshed my screen, and I was like, Oh, that's new. Yeah, it's, it's very hard to keep up with. Um, and the possibilities are virtually endless. So, yeah. So what other things? So business businesses, you all have a bunch of customers. You could dump your customer data into it and ask it to visualize it and visualize where the clusters, you know, you have sales staff that are out and they're, they're covering things.

Kevin: Um, It's a bit of a lift to do some sort of geographic sales analysis of how effective you're being. But if you dump the data in and you can tell it to visualize, um, where all your customers are, like almost do like word cloud type type. Deals, it'll do that. You can see that, you know, 85022, like you're doing really well in that zip code.

Kevin: So kudos to that salesperson. But these other zip codes, you know, they're not, or maybe they have a lot of customers, but revenue is lower. Identify

Andy: opportunities. That was my next thought is I think that. Contractors probably have revenue per customer because that's the account. They may not have profitability per customer because they may not job cost down to that level of detail.

Andy: But could you, you know, do what you just said? Say, show me geographically where, you know, a majority of our revenue comes from or where our profit comes from.

Kevin: You absolutely could, and easily, easily you could. Major caveat. You're putting your data out there, and you've got to decide if you care. Um, I have a very practical, uh, attitude about this.

Kevin: That... There. The, the LLMs are not in the business of yielding your data, but there is evidence that they are porous at the moment. So if you put highly sensitive data in there that you know, the formula for Coke or something like that, it is possible that the MO model. maybe training itself on that. So if somebody somewhere then asks for the formula for Coke, um, since you've put it in there, it can connect the dots, but we're not talking about,

Andy: so you might not want to disclose the name or the address, but maybe just the zip code.

Andy: And that might be good enough. The first column was a zip code.

Kevin: So, but I'm, I'm not too stressed about it. You can turn some of these tracking features off and the training features off. But somebody would have to. Like the data would have to be meaningful, right? That, that, that somebody would be interested in it and like be able to put it together and whatnot.

Kevin: So know that and you're going to start to see micro LLMs develop. Um, I don't think in this scale of business so much, but in medium sized businesses, you're going to start to see captive this is just Kevin pontificating, but you're going to start to see captive LLMs such that they are walled. Such that the organization can play with the LLM, but it's not necessarily getting out into the corpus of the world.

Andy: I mean, I think it, uh, you're right, is we don't know how the data could be used now, but if, if the engine, if that's what you call it, the, the machine ends up with, if everybody uploads all of their sales data by zip code, then potentially the machine knows where people are spending money on outdoor, you know, services.

Andy: So if we had a new, business we wanted to sell into a brand new greenfield market. It's a startup and we could ask it, show me the areas that spend the most on irrigation systems and it could provide that to us. Then we would have a target on how to go sell holiday lights or ponds or landscape lighting or something else, patios.

Kevin: Yeah, we could. And it's pulling from all kinds of different data sources as well. So, um, so other practical things. So let's talk about images.

Andy: I was gonna say, um, that's what I wanted to get into next, just briefly, because we are kind of running out of time. But could we talk a little bit about images?

Andy: Sure,

Kevin: uh, so In much the same way that the language models can predict what word comes next, image models do the same thing on a pixel level. So they're predicting, based on their neural network, what could actually come out next. And this can allow you to enter prompts, in a similar way as GPT, into a model like MidJourney.

Kevin: is, uh, the one that I prefer, but there's also Dolly, uh, stable diffusion. There's a few, a few others that are out there that allow you to visualize things. So here you've done your blog. Um, now you need, uh, an image to go along with the blog. You could go to Getty or one of the other image provider things and find a dude squatting next to a sprinkler.

Kevin: Or you could ask the The image generator to come up with, you know, middle aged guy working on a sprinkler in their yard mountains in the background and come up with a plausible image that you can use very, very quickly. That's adapted to what you need on a professional and you and I did

Andy: this. Briefly, like with a five, a five minute, you know, demo.

Andy: And I'm, I'm curious, do you still have the copy of that image that we created with

Kevin: AI? Which one that does the discord image? Yeah, I do.

Andy: Yeah. Cause maybe what we could do is if you could email that to me, I will, you know what, I may use it as the cover of this article. I'm not article, but of this podcast, uh, on sprinkler nerds so that you guys can see an example of an AI generated image that Kevin made with me with a couple prompts.

Andy: So if we still have it, let's, let's

Kevin: pull it up on my screen. So it's, uh, yeah, cool. Um, so it's, it's like a guy, I think, what's the, what's the prompt? Let me read the prompt. It was, uh, If you're

Andy: listening to this on Apple podcast or Spotify or something to see the cover art, I think you will need to go to this episode on sprinklernerd.

Andy: com. That's where you'll see the actual graphic that Kevin's talking about.

Kevin: So the prompt is 30 year old energetic man checking a sprinkler in a deep green lawn. Nikon photorealistic and the trigger there is I'm trying to get it. It knows what a Nikon photorealistic image should look like, so it's not going to be some wild cartoon like, you know, psychedelic type thing.

Kevin: It's trying to get it to be as real as possible. And sure enough, there's a guy squatting next to a sprinkler that is pretty well unusable. Now, from a processing perspective, you know, just let's just talk more work a day like you don't know what's going on. I, I've always advised, so I guest lecture on the stuff, um, entrepreneurship in general.

Kevin: And I've always told my classes that, you know, you need to know basic Photoshop if you want to be a, a group by base level entrepreneur, because if you're not a creative person, you're going to be beholden to those agencies and it takes a long time, even if you're outsourced. So you're waiting for the student.

Kevin: So things like practical things, like I need to remove a background. So, you know, I see Andy's logo behind him. I need, I need a transparency of this logo. Like there's an app for that, that, you know, for basically nothing. You can go to remove BG and it's going to pull out the background or image correction or image resizing.

Kevin: And what you're going to see is a lot of these tools are going to be baked into the image processing software, like. Photoshop and Illustrator. Uh, I highly recommend if you're graphically oriented that you check out Adobe Firefly, uh, because it is magic. Like. I want a picture of a deer. Okay, now let's put the deer in an alley.

Kevin: Oh, let's make the alley dark and add a sign over this door. And it's just on the fly creating all of this stuff. Which should make any graphic design oriented person tremble in their boots because... The most graphic designers make their, most of their income off of the stupid little stuff. The image correction and things.

Kevin: It's not the big creative projects. And you're going to see that's going to be an industry that's going to be highly disrupted as a result of this. But yeah, even

Andy: Canva today is really disruptive, but not nearly what you're talking about. But Canva

Kevin: will implement this stuff too. So you can also do video voiceover.

Kevin: I mean, be very afraid about voiceover and deep fake potential. Like we're not going to get political, but the next few years in this country should be very, very interesting. That way it's an election cycle and we're going to see all kinds of crazy stuff. And just to get, you know, philosophical for a second, we're going to end up in a place where you can't trust things and that's not a good place to be at all.

Kevin: But just know that you can replicate your own voice in 15 minutes. Like I do a lot of podcasts, my voice is out there. So I had this bit of an epiphany and I called my, you know, 83 year old mother and I said, look, it is entirely possible that somebody could call you with my voice and try and get access to your bank accounts.

Kevin: Like. That is actually possible right now and I gave her a safe word, like, you know, if you ever feel weirded out, whether or not it's actually me, um, just ask and, um, you know, we can verify, right? Don't say your safe

Andy: word. Don't say it.

Andy: That's a great tip, actually. I think I'll, I'll, uh, with my family, come up with a safe word for all of us in the event that somebody does this. I think that's a really good

Kevin: tip. And it's awful, but people, it's already happening where people will get calls from their kids. I've been kidnapped. You need to send, you know, 10 grand right now.

Kevin: Like that is happening. Now the positive side is like you're very soon you're going to have the ability to have these virtual customer service agents that can actually talk to people. Um, it's also terrifying, but. Like that's, let's just stay on the positive, right? That these are, these are, these, we are going to be able to offer such a personalized experience to our customers that we are going to just be able to blow them away.

Kevin: Like when you, right now you're busy, you're running around, you got all your crews, you maybe have one person answering your phone, maybe you have nobody answering your phone. The phone can be answered. Chats can be responded to like, this is an whole aspect of practical applications that you, you all should be thinking about that.

Kevin: How can I, how can I create a better experience on a creepy experience, but a better experience for my customer using some of these tools to get them to what they need instead of the endless frustrating, like back and forth. So, so

Andy: this might be a good point. Two are a good time for me to mention. I would like to run an experiment.

Andy: So if you've made it to the end of the episode here, I would like to run an experiment based on what Kevin just said about personalization. And I think I would like to I'm kind of just spitballing this as as I go. I'd like to put a form on my website. So let's just say I'm going to go with sprinkler.

Andy: com forward slash Ink Works, I N K W O R K S. Ink Works. I'm gonna put a form there with a few questions, including your address and when you fill out the form, I'm gonna use one of Kevin's projects, ink works.ai to send you a personalized letter handwritten from me based on the input, personalized based on the inputs that you enter in the form.

Andy: How's that

Kevin: sound? That sounds awesome. So, and that's, that's my, and

Andy: I'm going to pay for it. It comes with a fee and that's Kevin's project right now, inkworks. ai. So I'd like to actually test it in real time with you guys listening and, um, you know, give you, give you a taste of what Kevin's

Kevin: working on. So we do, we're using, uh, LLM technology to interpret messages.

Kevin: And then we're using pen wielding robots to handwrite notes. So, let's imagine you did a big landscaping project for a customer. Like, you know you should send them a thank you note. Or a Christmas card, or whatever it is. But you never get around to it, because it's, it's, it's time consuming. Um, using Inkworks, you can produce that letter.

Kevin: And it comes out handwritten, absolutely unique. Um, I of course have them piled around here. They look like they're written by... And, um, it's remarkable efficacy and very ironic that I'm using multiple layers of AI to create something that's so highly personalized specifically because people are craving that personalization that we're all bombarded by all of this information constantly with emails and SMS and all this stuff and people just ignore it and it's just going to get worse as AI continues to advance.

Kevin: Um, so. Ironically that my, one of the first toeholds I have is doing something analog with something amazingly complex.

Andy: So great. So great. Can't wait to run this experiment. Uh, on that note, you know, Kevin does, uh, coach businesses in this field. If you would like to, uh, hire Kevin to, you know, help you with your business, coach your employees, give you tips.

Andy: How can somebody reach out to you, Kevin?

Kevin: Yeah, the easiest is, uh, Kevin at www. inkworks. ai. Um. Or I'm relatively easy to find on, on LinkedIn. Um, yeah, I'm, I'm, I'm out there.

Andy: Very cool. Very cool. And hopefully we can maybe find a time to do a little online training as well. And again, visit sprinkler. com forward slash inkworks and let's test out Kevin's software.

Andy: I'm really excited to do that. And. You know, Kevin, I think that from all the people who I have met that are into AI and use the tool, I don't think I've met someone as knowledgeable as yourself, and I really appreciate you sharing

Kevin: this with us today. Thank you. I'm clearly passionate about it. This is the future, guys.

Kevin: Okay,

Andy: well, until our next AI conversation. Thanks so much, Kevin. Have a great one.