How AI Tools Are Transforming Business: Insights From Luke Arrigoni

How AI Tools Are Transforming Business: Insights From Luke Arrigoni

Sean Si: Hey, guys. Welcome back to the show. It’s me again, Mr. CEO at 22, Sean Si, this is going to be the first episode I am recording this year, and I am so glad that we have someone who is an expert. Over the last 15 years, this guy has worked with some of the top brands in the US building machine learning and data science programs. Even before those terms were popular or even before they were invented, he worked with companies such as UPS, J&J, Getty, AT&T, Stryker, Goldman Sachs, Sephora and a host of other companies out there. His name is Luke Arrigoni, and we are honored for him to be here and to learn from him today. Luke, thank you so much for being here on the show.

Luke Arrigoni: I’m really excited to be here. Thanks for having me.

Sean Si: All right. So in the pre-show, we talked about crypto. There’s so many things we can talk about and I want to start this year actually with discussing AI, crypto still it’s still a bear market. There’s so many things happening right now. It’s actually more like a tabloid or gossip than real news, you know, real actionable news. Right? So let’s talk about AI. Since ChatGPT came out, there’s so many use cases like they’re saying it’s going to be used in school to teach toddlers all the way to college students. Sure, they’re using a 2021 dataset. There are things that it cannot do, but there are so many things it can do. But what really is for those who are listening, who have no idea what AI is or how AI is built and made, what is an AI and how are so many companies coming up with their own AI’s right now? Is it really even AI? Should we even call it AI?

Luke Arrigoni: So this is a great place to start, right? Especially if you’re not familiar at all with AI. Like it would be helpful for us. Set kind of the groundwork, right? AI is a big term. It’s an umbrella term. And underneath this umbrella you have things like machine learning, which is what most of us care about inside of machine learning. It’s like supervised learning. Like it gets really, really specific and then you have other things like I could make an, if then statement like 50,000, if then statements and technically it’s AI, right? So just to give you this broad sense, when people say AI, they’re using the most generic term, right? If you hear machine learning or neural networks, those are real specific forms of AI. They’re underneath the umbrella. They’re like strategies you would pick that are also AI. So now that we kind of have the taxonomy down, when you hear big tech companies talking about AI, what they’re really talking about is the machine learning or neural nets, especially like ChatGPT that’s machine learning, that’s a neural network. You have really small plans where some people will build, you know, we call it A.I. Now I’ll actually start using the more specific language for all the listeners. We’ll build machine learning that might do something like predict when someone wants some other item in their cart, or we’ll predict when a user should get a newsletter sent to them to upsell them on some product. We know that they were on our side and they looked at a couple of things and they bought one thing six months ago. And we know that when people fit this pattern, they’ll also buy this new item we have. Machine learning is this extrapolation of data that we have to try and predict what we think will happen in the future. So yeah, it’s usually a very simple task. The idea of having one AI that does everything is kind of more distant future type work. But right now it’s just one simple task. And if you think, well, you know, my car drives itself and that’s like a million tests really. Those Teslas are loaded up with like 45 different neural networks. They have little machine learning projects, one to detect a stop sign, one to detect if there’s, like, a deer in front of your car. Like there’s they have the little tiny models for everything. So even the most complicated things end up being boiled down to just small little projects.

Sean Si: Yep. And all of these things working together is like what makes a good A.I. good. There’s a broad range of AI now, right? I mean, the digital marketing industry, there’s so many tools out there that pop up and they’re saying like, we’re using AI or AI backed. And I go ahead and try to use some of these software’s and honestly, they suck.

Luke Arrigoni: If someone is like a small business owner, right? Like you have a company and you’re trying to vet vendors, you probably are in the middle of 20 different conversations with people. They all say they have AI. I mean, the person that’s saying that they’re going to supply janitorial supplies to your company is probably telling you they have AI to dump the trash bags like we’re all bombarded with it. It’s like a sales pitch, right? And so as a decision maker, you probably are wondering, how do I sift through all of this? What’s appropriate for me to ask, like, hey, pull back the curtain. I need to know magic in almost every conversation. It’s okay to ask that if someone from your new sanitation company comes and says we can dump trash bags 20% faster with AI, you should be asking how. Don’t pretend. Oh, well, that’s above my intellect or my. Standing. Make them prove it to you, because there’s a lot of people that are peddling snake oil like kind of, and they’re calling it AI. And people are kind of tepid about questioning these vendors, right? Like, well, I don’t really know if I’d understand. Make them help you understand because it’s not outside your understanding. And also, you should ask for things or feel comfortable asking about accuracy, precision and recall. And I could talk real quickly about what those are, but those are the three big metrics. There’s others, but those are the three big metrics. We measure the success of any AI or any machine learning model. The difference is between these three. If you have an auditorium with 100 people in it and five people are murderers, if I said I built a machine learning model to predict that there are no murderers in this building, it’s 95% accurate, but it’s totally worthless, right? Like 95 of the 100 people are, in fact, not murderers. It’s 95% accurate. But that’s actually not useful to anyone, right? We actually want to find the five murderers. So if I built another model and I said, Hey, I found three of the five murderers, that’s precision. That means, hey, I guess three times and three times I was correct. Now, if I said, here’s eight people that I think are the murderers, obviously three of them are definitely at least three are wrong, maybe more that’s recall. That’s saying it’s more important to get even good people out of the building. And so as you go through this, depending on your business and of course, you will know your business better than I do. Right. And go ahead and pause, rewind the kind of process that if you think about, hey, this vendor is trying to pitch me on something, I should ask them about recall because that’s important to that project. You will wow your vendor and you will get answers, right? So, yeah, don’t be afraid to question the magic. It’s not magic. It’s science. It’s math. And it’s not above anyone’s understanding.

Sean Si: Yeah. And so the question now begs, why are all these companies saying AI and they’re selling. Is it the new sexy word in software?

Luke Arrigoni: Absolutely. It’s something that everyone is hearing about, right? And I think a lot of companies have realized how attainable it is. So like the bar for it is probably if we had had an extra 30 minutes, I’d make you a quote AI right? and we all laugh about it. And then you would be like, I have AI now right? like it’s actually incredibly accessible, which is great. But at the same time it means that the combination is bad, right? Its high accessibility and high buzz Word means you smash them together and you see it everywhere.

Sean Si: Yeah, yeah. I feel like. I feel like it’s dumbed down to just being generic and vague. That’s it.

Luke Arrigoni: Right? Like, it doesn’t even. That’s why I really want to encourage your listeners to question when people say that’s because a lot of times it’s not.

Sean Si: Yeah. And so if you’re a small business, let’s just say you’re starting with e commerce right now and a lot of people are starting with e commerce, especially with the accessibility of being able to start up with Dropshipping on an affiliate marketing, maybe on TikTok, Facebook, Instagram, you use Shopify, e commerce. What are some of the good and legitimate AI softwares out there that you hear a lot about and you feel like, Oh yeah, that’s a pretty good AI tool to use?

Luke Arrigoni: I don’t know the exact brands, but I can tell you the problems and then I’ll tell you, like these are the sets of problems that are very doable and you should believe your vendors when they say they have AI. One of the first big ones is recommendation engines. If you are in a conversation and your vendor is saying hey, we can recommend the perfect package for the perfect size like shipping logistics type predictions, or also recommendations for what each audience would want to see. That’s really well vetted. There’s like ten years of history in that space. You don’t have to question it too much. Obviously, make sure the vendor is legit and they have it, but that is very solid. That’s not so much magic anymore. Anything around the email list that’s actually pretty solid as well. You start to veer off into kind of like no man’s land. When people promise specific stuff, I’m like, I can find the very specific person, specific person that wants to buy from you, or like, and how would you find that one person or the people that say they’ll do any level of like really specific tracing on a demographic of people? It’s like that’s actually really hard to get done. So the more it sounds like some kind of dragnet surveillance program is required to make this thing work, it probably doesn’t work. Yeah. And you see marking claims like that. So those are the two big extremes you have recommendations, emails, networking, stuff that actually is very well vetted. And then you have the hyper focused trace promises that aren’t probably bad. Then you have the third and final category, which is ChatGPT, GPT three, which is you hear people saying we have AI that can help write our newsletters. That’s here, that’s now, and that’s working. It’s incredible. Actually. I’ve known a lot of people that will actually start developing. A ton of blogs and content on their site for SEO and other reasons. That’s all original, which is amazing for their brand. So that is something also that you should believe in. ChatGPT and GPT3 are pretty legitimate enterprises.

Sean Si: Yeah. What do you think about that? I mean Google on the other hand is saying hey, we don’t encourage AI produced content and we’re going to crack down on it someday. And then there are these big banks in the US right now that are using ChatGPT to produce at least the base layer of their content. When I say base layer, it means the outline, it means the initial draft before they have a human edit and publish it. What do you think about this? Is it something that’s sustainable? I mean, we see that ChatGPT is having outages sometimes because so many people are using it.

Luke Arrigoni: This is the fascinating part. Obviously, Google and other search engines, they’re going to want to play a cat and mouse game with this right where they’re going to say, we’re going to eventually find out if something is AI generated and then shut it down. But I think that they’re going to come to regret that decision because at the end of the day, people are looking for information. And if someone has generated information that solves a legitimate problem, whether it’s AI generated or not, like whether that content is there, that’s something that people are going to search for and look for. So I think at a certain point I get why Google is saying things like, don’t overwhelm our system with all this AI generated content, but there will come an inflection point where that is what people want. People are going to want to figure out how to search across all the possible knowledge, even AI generated knowledge, to find what is real and true. So it’s kind of a tossup, I think, for now advice, do it. You know, I’m no one’s marketing strategist, but I really actually think it’ll be very difficult for Google to detect what is human, what is A.I. It comes to the written word. There’s a lot of stuff around the media, like deep fakes. Those are easier to detect when it comes to written words, though, that’s going to be much harder because you have at least in native English speakers, there’s a huge range of capabilities for writing, right? Like I went to high school with some people that like they didn’t know how to write and they were like 15 right. Like there are some people that just like you could not say, hey, this is poorly written its AI. And you also couldn’t say this is so well written and say it’s AI. And so I think that there’s going to be a huge challenge there. And I think Google is largely bluffing when they say they’re going to be able to detect it. I just don’t think they’ll be able to. And I think they’re going to put the resources in to do it because once again, people are going to want information. And if ChatGPT comes up with it versus a human, I don’t think anyone really cares.

Sean Si: Yep. Yep. Good insight, I think. Well, I think if people just copy paste like whatever ChatGPT regurgitates right. And puts it up and Google sees like oh this content is in like ten other websites, then it will be able to detect that, oh, this is not really AI generated content but duplicate content, right? It can only detect. Yeah, but GPT doesn’t duplicate.

Luke Arrigoni: That’s what’s incredible. What you see is actually there are some duplications around like their disclaimers, but like the content that comes out of it is original every single time.

Sean Si: Oh wow.

Luke Arrigoni: Yeah, it’s, it’s pretty incredible. But at the same time, you should know that most people, most operators for content, take the AI generated content and then they scan it as a human. So once again, it doesn’t replace their job. It just makes them better, like makes them more efficient. So yeah, if you were to copy and paste, you probably might get caught by some kind of future detector. Right. But yeah, if you use it, how most people use it where you get the first 1000 words of your blog pre-written and you go through and tweak a word here or there or fix a grammatical error and like, I just don’t know if there would be a differentiation.

Sean Si: Yeah. Yeah. Good point. Good point.