AI Common Mistakes We All Make When Using AI

What's up everybody? It's Mike with the HVAC Mentors Podcast, back with you to continue a discussion that we had started a few months back, and it was centered around AI. And I know now over the last several months, especially the controversy around AI data centers and AI-generated content has just, it's, it's really grown and it doesn't seem to be going away. But the thing is that we really need to put some critical thought into decisions Around our use of AI and what that means. AI is a useful tool. It does help us to be more efficient. It's very powerful software. And think about it, when the automobile was developed, when electricity first became mainstream, when computers began showing up in people's homes, and when the telephone was invented, all of these things were seen as the end of the world and every society is about to collapse. Everything is— Is going to end. And it never did. In fact, life got better. Think about the automobile and the telephone and the ability to communicate. You know, I do a lot of genealogy and looking back at my own family history. And when my great-grandfather came here from Poland, I mean, there were friends and family and people back in the old country that he knew he was never going to see again. And he didn't. His own dad eventually died just old age, you know, and nothing dramatic, but he didn't get to go back to Poland to bury his father. He just didn't. He was just here now. Think about, you know, the fact that we don't have to go through that anymore because of these inventions that were once seen as something to be afraid of that now are just everyday things that we use. And now, you know, you can be hundreds and thousands of miles away from your family and your loved ones. And you can communicate with them. You can see their face. You can hear their voice. That would have been unthinkable a few generations ago. So as we continue to kind of unpack what AI is, I just want you to think critically and think about, okay, well, sure, if unchecked, there are some possibilities of things going awry. But, you know, if we use this tool correctly, it can also do a lot of good. So You know, especially in the trades, and I certainly made some mistakes in using AI at first in my own job and as a service technician. But you know, I want to go over some things today and just talk about some common mistakes and things that we make and or things that we do that maybe we're misusing or misthinking about the way that AI is supposed to be used. So basically, basically just talking about you know the honest growth that takes place as we embrace a new technology. And a new tool. We've kind of spent the last 4 episodes trying to understand AI, what it is. And, you know, at its, at its base, I mean, fundamentally it's software. It, it does a good job of sounding conscious and sounding like it knows anything, but it's software. It's designed that way so that it makes it easy for you to use, but also so that you're comfortable using it. It doesn't make sense having a product out that is scary or weird Or awkward or cumbersome to use. Instead, they make it user-friendly, which part of that is making it conversational and making it seem like it's your friend. But we also want to make sure that we're aware of what it actually is. So we also talked about now what happens when we actually use it and how it is a very powerful tool, but truly it is easy to misuse. So today, again, I want to unpack some of the common mistakes that we make when we're using AI, which actually make it a little less useful or even potentially misleading. I was kind of teasing myself with AI that I've been kind of scrambling to try to get content published and scheduled and kind of a content pipeline, so to speak, moving so that again, between all the social media platforms, I can kind of keep the algorithms happy and keep posting and keep getting found because, you know, as much as I want it to be about you and about you all learning and growing and us growing together, you know, if we don't get more viewers and constantly grow, then, you know, it's, we become very limited and our growth gets stunted and we don't want that either. But I was joking with the AI, so kind of really joking with myself that that it's funny that I'm finally using this powerful tool for something other than creating funny memes and, and, you know, parody ads. And, you know, it's kind of true. We misuse this tool for this powerful tool that can help us in so many different ways. And it makes it seem almost more like a toy. That's not necessarily bad either. And we're going to find out why here in a little bit. So mistake number 1 is simply assuming that AI is always right because it simply repeating information that it has gathered in databases and websites and wikis and regurgitating it to you in a way that sounds understandable or logical. It can sound very confident. It's going to give you a very clear explanation, but that doesn't mean it's a correct explanation. It is still up to you to verify. So one thing that we've brought up many times in Shorts and other lessons I've done on the channel is trust but verify. So the, it may be going off of information that is just blatantly incorrect or maybe is outdated, but it's information that it has access to. So you got to make sure that you understand that it's right. If it generally tracks with what you already know to be true, you can probably trust it, but I'd still make sure that it is actually accurate. You can sound absolutely confident. In, in your incorrectness. Confidence does not equal that you're correct. It just means that you're confident. It may be giving you, in that regard, it may give you a very convincing answer based on incomplete information. So again, verify, verify, verify, especially the more crucial the solution, the more you need to verify. You as the service technician should stay in control. Have I used AI to help me in diagnostics? Sure. Because there's some good diagnostic literature. that it is pulling on to help me with information, but am I going to blindly follow it? Heck no. So I'll give you one example. I went on a service call. Of course, I, you all know I do HVAC refrigeration, but also commercial kitchen. And part of that is warewash. And I went on this big commercial dishwasher call and I found that you, it had a problem with, with the water supply. So basically the fill solenoid valve had failed and, and it was, it was opening partially, but not nearly enough. So it had low water pressure. I'm one that likes to photograph and document everything I do. And typically just kind of as a warehouse of photos and knowledge and to kind of summarize my service calls, I like to upload them to ChatGPT and sometimes it'll give advice when I didn't ask for it. In fact, Frequently it gives advice when I didn't ask for it, but I had by then tuned that out and just sort of like, okay, no, I just, you know, here are these pictures, just help me organize them later at the end of the day when I'm home. So the AI kept pointing out that there was excessive foam in the wash water. And because of that, it kept telling me that we had a problem with it dispensing too much foam. Well, I knew that that wasn't Or rather, excuse me, dispensing too much detergent into the water. I knew that wasn't the case. I knew in fact that the problem was that we weren't getting enough water. So finally, when I'm using AI to look for parts, which by the way, extremely helpful for that, it kept trying to steer me toward part numbers for the detergent dispensing parts. And I kept telling it, it has a failed solenoid valve. It's not a problem with detergent. I can't determine that until I get the right amount of water in. And so finally, I got the part number for the valve. It was fixed. Everything's fine. No excessive foam. The dishwasher worked perfectly after the proper repair. So after all was said and done, I was home kind of going through my text messages or my text messages, my service calls and pictures and going through my different threads for each service call. You know, I was reading through this one. I had to laugh because it was like, you know, here's this. Technology that it's not conscious, but it seems conscious and it's giving me feedback. And it, like a newer technician who is just hyper-focused on the wrong problem or a symptom, it's just going back to this over and over and over again. And it just really was caught up on the bubbles. So I jokingly in the threads said, okay, from here on out, your nickname is Bubbles because you were so caught up on these bubbles that weren't actually relevant that this is now I'm calling you Bubbles from now on. I don't think I actually did that because I moved on to other things, but I just thought it was funny in the moment because it's like, you know, hey Bubbles, you know, let's, let's, you know, misfocus on something else. The reason it's funny too is because having ADHD, I'm definitely one to hyperfocus on the wrong things sometimes. That's something that I've had to discipline myself on. But the AI wasn't being stupid or useless. It certainly wasn't Flawed. I mean, it recognized an actual symptom and a legitimate pattern, but it wasn't the actual problem and it wasn't analyzing the entire situation. Because I use my mind first and my observations and my tests and measures on a service call, I'm not necessarily going to give the AI all of the information because I don't need to. I'm going to give it what information I want it to keep track of so that I can stay a little more organized and a little more efficient. But, but it recognized legitimate, a legitimate pattern. And based on the context that it did have, and based on what literature it has, it recognized a potential issue. I had voltage measurements and, and open windings on the solenoid that it didn't know about. I had, I had further context that it didn't have. So it's not the AI's fault. It's just that it was reacting to what inputs it did receive. So you can certainly use AI as your own input, as one of many voices, as a way to kind of, you know, get down to the main issue. But it certainly should never be the only source of answers or the only source of knowledge. One other area that I remember, and this is, this is kind of going into our next point, is mistake number 2, treating AI like Google. It's not a search engine. Are search engines a component of AI, a conversational AI? Yes. Although its developers would say no, but it literally uses search engines along with, you know, along with other tools that it has at its disposal, but it is not in the sum of all of its parts a search engine. It is a language model, a context model engine. It uses multiple sources to construct the most relevant response. To the prompts that it's given. It can also generate information off of learned patterns about you and things that it understands about you. So, so if you're looking for technical information or schematics or what, how does this circuit work within this machine, I would trust AI to get me a wiring diagram, a service manual, technical information or literature on this machine. The nice thing is that you can take a picture of a schematic and put it into conversational AI. And it will help you narrow down test points and also potential, potential things that you need to zero in on. And, you know, stuff that you also maybe can dismiss. Because I, you know, again, in diagnosis, sometimes I miss the forest for the trees, but, but don't let it conclude for you a diagnosis. It can help you interpret All the evidence, kind of narrow things down. But ultimately you are the one interpreting. It is merely a helper. Mistake number 3, just as with the bubbles, not giving enough context. If you put into AI, my walk-in cooler is not working, what's wrong with it? That's not enough information. Now, if you give model, serial, symptoms, what you've already tested, what you've already done, what are your observations, electrical measurements, photos, The better information, the more useful AI becomes. So, you know, one example is, you know, I had a walk-in cooler that kept pumping down. It would not stop pumping down. And so eventually I found that it had a bad liquid line solenoid, but then I also found that it had a leak and it was basically running flat. Once I got the solenoid fixed, I found that's when I discovered the refrigerant issue. But, you know, what if I had just walked in and said, hey, this thing keeps pumping down. Do you think the compressor could be bad. I mean, it might tell you yes, it might tell you no, it's not the compressor, but you know, it's not gonna give you a complete picture. But that's why, again, we do not use AI to diagnose. We use it almost as a verbal notebook that talks back to us. And do we allow it to govern our steps in diagnosis? No. It's great for documenting them though, but it might also help us to maybe understand where we should think about checking, but by no means does it tell us how we should run a service call. But the better information we give it, the more useful it becomes. In fact, even, even to the point where I've had machines that had no visible data tag anymore, like it had just rubbed off over time and just took a, took pictures, you know, as many pictures walking around the machine as possible. And it was able to at least narrow down brand. Make and model, or yeah, make and model, approximate age. And from that, I was able to actually diagnose the machine. The other mistake, mistake number 4, is as we've kind of alluded to, so I'm not going to go into too much more deeply, is letting AI think for you. Especially for newer technicians, analyzing a data tag, sure. Finding, you know, manuals, schematics, helping you to interpret a schematic. Okay. But AI is not gonna be diagnosing for you, nor should it be. Observe, measure, understand the sequence of operations, form your own hypothesis, then bring in the AI. Now, AI for sanity checking, you know, so my, my unit's pumping down. Does it make sense that the compressor could be bad? You know, that's not a bad question for AI because it's simply a sanity check. It's not an actual diagnosis. Asking, what am I overlooking? Asking it to challenge your diagnosis. These are, these are the types of questions that we can use AI for to help us to think more deeply, which actually causes us to be stronger as technicians, not replacing us nor replacing our thinking. So think about a calculator, right? A calculator can be incredibly useful, but you still need to understand what multiplication means. You need to understand how mathematical formulas work. I mean, at least even the most basic formulas. The calculator does not do the math for you. You enter the digits and push the plus, minus, Multiply or divide sign, and then push the equal sign. And that's part of the reason too why we need to understand how to do our math problems manually with a pencil and a piece of paper, like when we're kids. That's part of the reason too why we need to understand how refrigerant charge works with our analog gauges before we switch to our wireless probes and our smart tools. You know, playing with AI and using it for funny little memes and, you know, parody ads is a fun way to learn about the capabilities of AI, but using it for things that it's not intended for is, at the end of the day, it's silly, somewhat reckless, and it also cheapens the power of a very real tool that could be quite beneficial to us. The biggest mistake I think we make too is, as technicians, and the first time, speaking of smart probes, the first time I bought a set of smart probes, you know, I kind of left them in the truck. I went out, I used my analogs. I didn't really, you know, pull them out that much. They stayed pretty much brand new for about a year. And then I finally started using them. And you know what it was is I was working in residential and we had a few apartment complexes that we did the HVAC systems for. And finally got tired of running up and down the stairs if I had a second or third floor apartment where, you know, I I needed to get temperatures upstairs and then while, while watching pressures downstairs, adjusting the thermostat upstairs, just going up and down the stairs. And, you know, I'm not, I'm way too spherical to be running up and down stairs all the time. So, you know, I finally just said, you know what, forget this. I got out my smart probes, my gauges hooked up my pipe clamps and my smart, my smart probes. To the outside condensing unit and opened my app and watched the refrigerant pressures and everything while I went upstairs and, you know, checked things up there, checked airflow, put my, put my psychrometers on the, the air returns and the supply. And it was beautiful. I was just like, why haven't I been using these? They're such a great tool. And, you know, I think a lot of times, be it for fear or a lack of understanding. Sometimes we just don't use the tool that we have. Just like my smart probes, AI has a lot of potential to make your life a lot easier. Even something like ChatGPT or Claude or, or the Google one. I can't think of the name of it, but, but they, they have a lot of potential to help you out. If nothing else, think about it this way. If you hate writing service notes, but your office is always asking you to be more detailed, ask To use this prompt, help me write a concise service note with short bullet points that describe this service call, and then talk through your service call and then hit the enter button. And it will give you a nice bulleted, clean set of point— key points of what you happened— what you did and what happened on the service call and what you found, what your readings were. I mean, just from that standpoint alone, it's a really helpful tool. Imagine having your own secretary in your passenger seat who would Who could, you know, kind of dictate your service notes for you. You know, take a picture of a data tag and ask it to find, you know, an inducer draft motor or a blower motor or a heating element or some other part. Just have it find a part number for you. I guarantee you, once you try it, you're going to love it because it saves you so much time. Ask it for service manuals, ask it for schematics. I guarantee you, give it a shot. Once you start playing with it, you're going to understand that It's been a, it's been sitting on your shelf for way too long and it's time you unpacked it. You know, it's very similar to those smart probes in a way, because I let them sit on the shelf for too long and could have been a lot more, a lot more efficient in my service work had I taken them off the shelf sooner. So, you know, that's just saying start experimenting with it, play with it. You don't have to be an AI champion or an AI expert, but you know, it's a tool that you can use, especially if you're a younger technician. Now is the time to start learning this technology. Because the bottom line is, guys, it's coming and it's, it's gonna be here and it's gonna be a huge part of our lives. We've already got AI, AI-based receptionists and scheduling. And, you know, if, if your office is using any kind of like electronic, you know, service calls and, you know, app-based tickets and whatnot, there's AI built into that already. And here's the thing too, as, as I've mentioned in previous episodes, if you have ever used your GPS on your car or ever Googled or used speech-to-text, you've used AI already. It's already been a part of your life and you may not have even known it. So for the next episode, we're going to kind of branch off of here and not just talk about the mistakes that we make with AI in terms of like overtrusting, underutilizing, but, you know, the idea of kind of Viewing AI in the wrong way, seeing it as, you know, too much of a friend and being too personal with it. And, you know, like basically some of the boundaries we need to make sure we maintain with it, because again, it can seem really human and really interactive and friendly. But at the end of the day, you're really talking to yourself and it's just analyzing the information you give it. So we're going to unpack a little bit more about AI guardrails and boundaries, what these do, and then After that, I'm really excited because we're going to have a very special guest on the podcast. We're going to actually have an interview with the, not really a man, but definitely the myth and the legend, ChatGPT. No, not the CEO of OpenAI, not an executive or any engineers with ChatGPT. We're going to talk to my own ChatGPT profile. And I'm going to ask it questions on the podcast about me and about my use of it and about our relationship and how we've grown. As both technician and tool. Imagine if your power drill or your manifold gauge set could tell stories about you. Well, guess what? I've got a notebook that's, that I can talk to and that talks back. And we're going to go into that. So we'll look forward to that in 2 episodes from now. So you all have a great day. Thank you very much for listening. Like and subscribe down below. And, you know, as always, if you have questions or suggestions for the show, feel free to also share those in the comments. And hey, if this is going to help someone, Feel free to share it. Have a great day and take care.