Resilience Sometimes Beats Prevention, and Six Ways to Get off the Starting Line with AI

Rob Collie (00:00): Hello, friends. In today's episode, Justin and I discuss two very different topics. First, we discussed my utter admiration for an invention known as The Green Biscuit. The Green Biscuit solved an impossible problem. How do you make a hockey puck that works on rough outdoor surfaces? I watched for years as manufacturers tried to solve this problem and they all failed. The fundamental problem is that an outdoor hockey puck will always catch its front edge on a bump and then flip. And once it starts flipping, it never goes back to gliding. Now, hockey equipment manufacturers knew that if they could crack this riddle, they'd sell a bajillion dollars worth of practice pucks, maybe even more practice pucks than actual pucks. And as a bonus, a practice puck that solved this problem is worth more money than a regular puck. So you can charge more per unit as well.

(00:55): So this untapped market, they all knew it, might actually be even bigger than the existing worldwide market for regular hockey pucks. So they all tried repeatedly and they all failed repeatedly to build a practice puck that reliably did not catch an edge on rough surfaces. And then sometime in the last 10 years or so, someone solved this problem when I wasn't paying attention. I believe this problem was not solvable, but they solved it by explicitly not trying to stop the puck from catching an edge, but instead they focused on self-correcting when it does happen. I'm not kidding. This puck self-corrects and that neatly parallels many things in life and work in my experience. Sometimes prevention is the right strategy. Sometimes it is the right approach to avoid having the problem in the first place, but in many cases you simply have to accept that there is chaos in the world and that responding to that chaos accounting for it is a necessary part of a successful strategy.

(01:58): We don't live in a clean room, and I think things like Power BI reflect that. Sometimes the data is noisy and your data solutions are going to keep catching on those rough edges unless you build in some noise tolerance, some resilience. And one of the things that differentiates Microsoft's current wave of tools from its previous tools and that differentiates it from many other vendors' current tools, is that it meets your reality where it is, not where some clean room computer scientists would prefer it to be. Now, the only thing in hindsight that I regret about this conversation, and I'm not kidding here, is that I may have gotten my Pac-Man backwards. I'd need some slow motion video to be sure, but the Pac-Man might happen on the trailing edge and not on the leading edge, and you haven't lived until you've kind of woken up the night going, oh no, the PAC-Man might be on the other edge. It's like a brand new sentence, right?

(02:51): Anyway, if that's confusing, it'll be cleared up shortly. Oh, and Justin walked us through a handful of ways that you can get off the starting with AI ways to safely dip your toe and start gaining confidence. And when I say you even, I'm really included in that population, knowing Justin as well as I do, I tried to anticipate all the examples he was going to share, and I scored less than 50% on that exercise. But I think we'll all score a hundred percent on understanding the examples you shared and also a hundred percent on the test of can we actually do it. So let's talk about Green Biscuits and Snackable AI, shall we?

Speaker 2 (03:30): Ladies and gentlemen, may I have your attention, please?

Speaker 3 (03:35): This is the Raw Data by P3 Adaptive Podcast with your host Rob Collie and your co-host Justin Mannhardt. Find out what the experts at P3 Adaptive can do for your business. Just go to P3adaptive.com. Raw Data by P3 Adaptive. Down-to-earth conversations about data, tech and biz impact.

Rob Collie (04:05): Well, hello there, Justin.

Justin Mannhardt (04:06): Hello, Rob.

Rob Collie (04:07): I've been meaning to bring this up with you for a while. I want to express my tremendous admiration for a very simple product known as the Green Biscuit. Well, I mentioned this to you. You're like, yeah, yeah, I totally know about the Green Biscuit, but to me, the Green Biscuit is a revolution. Back in my day, we didn't have hockey pucks that you could practice with outside. Asphalt and concrete are very unforgiving surfaces in terms of their coefficient of friction and their lack of regularity. So you take a regular rubber hockey puck and you try to slide that across something rough and it goes nowhere, because surprisingly, I know this is going to shock everybody, hockey pucks are meant to slide across ice, which is a famously slippery surface. Then there's those of us who play roller hockey, inline hockey. For those people we have the pucks we play with. They're a little different, but they've got these little nylon pads on both sides of them.

Justin Mannhardt (05:03): Like a furniture mover type of a surface.

Rob Collie (05:05): Exactly. It's that same super slick furniture mover material. And when you're playing on a rubber tile surface, like what we play on, those slide very well, they very closely simulate the feel of a regular hockey puck on ice. However, take those nylon furniture mover pucks outside onto asphalt or concrete, and they just don't go anywhere. As soon as you sling one, it goes like three feet and then catches an edge and starts flipping end over end or turns into a rolling on its edge sort of situation. There's no saucer sliding effect, and I just believe that this was not a solvable problem. We did not as a society have the technology to solve this issue. Now, there are many attempts. I saw pucks back in the day that had metal rollers in them.

Justin Mannhardt (05:50): Like ball bearings or something.

Rob Collie (05:51): And that seemed like a really good idea, but it turns out it didn't practice. They would go six feet instead three feet for giving up the glide. So I saw these Green Biscuit things advertised as outdoor practice pucks, and I'm like, those can't work. This is solved science, as they say. You cannot make a puck that slides on asphalt or concrete. It doesn't work. And then when I look at it has no rollers. It makes no attempt at being low friction. It's just two plastic disks and it's just the plastic disk that's making contact with the ground. I'm like, how the hell does this thing work? But then I saw one in practice, I saw someone using one. I'm just like, oh my God, that thing is amazing. It actually works. The way I am about things, I'm like, oh my, how does it work? Why does it work? And it is amazing how simple this invention is. Have you ever sat there and puzzled over how the hell does this thing work?

Justin Mannhardt (06:44): Not until you mentioned it. You asked me like, "Hey, do you know what a Green Biscuit is?" And I live in Minnesota, my boys play kids hockey. I'm like, yeah, of course I've got nine of them. And then you go off and you're like, "It's so amazing." And I'm like, I've never thought about it.

Rob Collie (06:58): Yeah, I mean, to me it feels like something that was invented yesterday. It's the same reaction that people get when they see Power Query for the first time. They go, what? Why haven't I been told about this before? You've been using this? Yeah, you have nine of them? It's amazing. It's just a puck where it's basically, imagine an Oreo cookie, but with no cream in the middle, it's just the two cookies. And they're only attached kind of loosely. You can actually grab the two sides of the Oreo cookie, the two discs, and you can pull them apart by about a quarter of an inch, half an inch, and they've got these bolts that allow the top and bottom to kind of move independently a little bit of each other. It's kind of the classic example of resilience over perfection. So perfection would be to build a puck that never catches a bump and goes end over end.

(07:49): No matter how you build these things, they still catch an edge. So the Green Biscuit takes the completely opposite approach and says, hey, catching bumps, catching edges is inevitable. What we want to do instead is essentially actively respond when we do catch an edge and recover. So what happens is that this thing's sliding along and the bottom Oreo cookie is the one making contact with the ground. Hey, that front edge is the one that's at risk. It's going to hit a little bump. When that happens, the backside of the puck is going to try to lift off the ground. It's going to try to hinge over and flip over that front edge. But because the top cookie is free floating, when that back edge starts to come up, it's going to encounter more weight in the form of the top cookie pushing back down on it.

(08:38): As that back edge starts to come up, there's going to be space opening up on the front edge between the two cookies. They're going to open a Pac-Man in the front, but that has the impact of pressing back down on that back edge, and it just re-stabilizes over and over and over again. So as it slides across the ground, it is hitting dozens, hundreds of bumps that would normally flip it. But by having this free-floating inertia of this upper cookie, if the thing flips over during normal use, well now the top and bottom cookies just reverse rolls. They're identical. It just blows my mind. There's something super, super, super elegant about this, but also a life lesson, something I tried to instill in my kids. Perfection's impossible, resilience is the thing you want. Making mistakes or finding yourself in difficult situations, you can't live life with the philosophy of avoiding that. You have to live life with the philosophy of dealing with it when it happens. And the Green Biscuit is perfect, it's just sliding around on the ground saying, see, this is that principle in action.

Justin Mannhardt (09:42): Green Biscuit, if you're looking to come on the show, maybe sponsor next week's episode, we'll do a giveaway.

Rob Collie (09:51): Green Biscuits for everybody.

Justin Mannhardt (09:52): They're commonly included with goody bags for the kids at the beginning of the season. It's like a little suspension system effectively. Our pad in the alley in front of our garage, it's terrible, and they're just out there passing this thing back and forth like nobody's business.

Rob Collie (10:06): It also has this tie in with Power BI, right? Power BI isn't built for clean room environments where all of your data is pristine and vacuum sealed and dust free. Power BI chews up real world dirty, noisy. It meets the real muddy world. The Green Biscuit to me is just like, as you can tell, I nerd out about common everyday things. This was an unsolvable problem and someone solved it.

Justin Mannhardt (10:33): And determined it was worth solving.

Rob Collie (10:35): They're just sitting there going there's got to be a way.

Justin Mannhardt (10:37): There's a way and a business. Good on them.

Rob Collie (10:40): All right, so that's my contribution to topic for today. You've hinted that you had something you'd like to talk about.

Justin Mannhardt (10:46): I was thinking about our conversation on the last episode you and I did together where we went through your PowerCore solution for the roster analyzer and fantasy football, and it's the first time where you and I got real down in the weeds about a AI use case. We put something out so people could follow along, and I have to remind myself all the time that there are a lot of people that are still behind me.

(11:12): There's a lot of people ahead of me. There's a lot of people way out in front of me, but there's a lot of people behind me in terms of their use of AI on a regular basis and how to get into it. And so I thought, well, maybe we could use today's episode just to talk about, if you're not using it regularly, what are some ways you could use it? How could you get into starting to use it more frequently? Because I've learned a lot over the last 18 months or so or whatever. It didn't just start one day, I arrived at the point I'm at. So I thought maybe it'd be fun just to sort of talk through some of those things and maybe it helps some people find a place to start.

Rob Collie (11:47): With a little correction of your humility first, yes, in terms of absolute numbers of people, probably a fairly sizable number of human beings who are farther down the road of blah, blah, blah but percentile wise, you're pretty up there. Just some of the things I've seen you doing, I'm just like, oh wow.

Justin Mannhardt (12:04): You're someone that's behind me and benefits when we talk about it.

Rob Collie (12:09): Absolutely.

Justin Mannhardt (12:10): So I reflected on this, and so I jotted down a handful of ideas over here of ways I'm using AI that are easy, simple. I'll tee these up and we can talk about them. You're going to ask me some questions.

Rob Collie (12:23): I think that's great. In hindsight, I'm going to give a score to how many of these I would've been able to guess.

Justin Mannhardt (12:28): Oh.

Rob Collie (12:29): I'm not going to go and make my list of what I think you're going to come up with, but I'll be honest about like, oh yeah, I would've come up with that.

Justin Mannhardt (12:35): Okay.

Rob Collie (12:35): Or not.

Justin Mannhardt (12:36): That's fun. So as macro context setting, before we get into specific examples, I think what I've realized is using AI tools is a habit you need to form. It's almost like going to the gym. You need to build that habit. You need to make it more normal, more natural for you. So that's why I think it's important to find here what's sort of the easy low-friction ways that I can start just engaging with the tools, because that's sort of my gateway to better and better use cases or more familiarity with them.

Rob Collie (13:12): One of the biggest obstacles to forming a habit, to building a habit is fear of the unknown. That's a very familiar feeling to me. It feels like walking the plank.

Justin Mannhardt (13:22): And so with that in mind, I mean you nailed it with the fear of the unknown, except that, A, if you're coming from zero or very little use. Except that you're going to roll up to ChatGPT or Copilot or whatever you decide to roll up to, and you're going to be like, what do I do? Do I say hi to it? Honestly, I should go find this. Just a quick aside, I think my first interaction with ChatGPT was having a discussion with it about simulation theory.

Rob Collie (13:48): About whether the universe is a simulation?

Justin Mannhardt (13:50): Right.

Rob Collie (13:51): Yeah.

Justin Mannhardt (13:52): I was trying to play tricks on it more than anything.

Rob Collie (13:54): Did it respond and say that information is restricted?

Justin Mannhardt (13:57): Oh, it's actually hilarious. It actually got to a point where it started to include things like even if life is a simulation, your life still has meaning.

Rob Collie (14:05): Yeah, you still have to play, right?

Justin Mannhardt (14:08): Yeah.

Rob Collie (14:08): If it's a sick game, it doesn't matter.

Justin Mannhardt (14:12): You'll roll up there and you won't be sure what to do. And B, you might not be wowed or thrilled with the first thing. You might be like, ah, that was lame, or that wasn't cool, or I don't understand all the hype. So we want to build habits first of all, but we also need to think about, well, how do I actually get better results from these things?

Rob Collie (14:34): Is this list, is it going to contain sort of technical and non-technical items?

Justin Mannhardt (14:38): Oh, yeah.

Rob Collie (14:39): Okay. This would be applicable to business leader types and maybe there's some that are specifically applicable to data practitioner types. Something for everyone.

Justin Mannhardt (14:48): So I might zig where you're expecting me to zag off the top here. If you have kids or even just friends, sit down and put ChatGPT up on your TV and just ask your kids to create images. Use it with your kids because it's fun. And I think for them, it's exposing them to this thing that's going to be a big part of their lives when they're adults. It won't be like it is today just to sit there and be like, hey, what do you want a picture of? I want a picture of a T-Rex wearing a motorcycle helmet, carrying squirt guns in a desert. Okay, let's make that thing.

Rob Collie (15:22): It's funny, I haven't used ChatGPT or DALL-E for image generation. I've only used Midjourney and I'm wondering if I'm kind of missing out because it looks like ChatGPT is much more natural, for instance, in terms of having an ongoing conversation and an iterative approach to an image. Whereas Midjourney, you've got to load everything into the one prompt, and so your intermediate steps are kind of worth nothing other than now you're constantly copying and pasting your prompt and adding things and removing things as opposed to, okay, that's pretty close, but change the nose. You really can't do that with Midjourney, and so I might need to just change gears as much as I've invested in learning how to use Midjourney, it just might be the wrong model.

Justin Mannhardt (16:04): I couldn't say for certain. And I think ChatGPT ultimately interfaces with DALL-E, I believe, but I think it has some additional layer of prompt enhancement between ChatGPT and DALL-E. Whereas if you go DALL-E directly, it's more reliant on the initial prompt. But I do know that it's fun. Low pressure, low friction, low stakes, doesn't impact your job. Just sit down and have a party with it, make some funny pictures.

Rob Collie (16:33): I like it. I'm in. It's a party game.

Justin Mannhardt (16:36): So with the rest of this list, again, going back to habit forming, my natural behavior, if I need to do something, there's a trigger that leads my mind to think, oh, I got X, Y, and Z. Oh, when I need to X, I should have Copilot up or I should have ChatGPT up. The thing I probably picked up on the earliest was anytime I have to write a document, do it with generative AI.

Rob Collie (17:02): So for instance, I reviewed two Word documents from you yesterday.

Justin Mannhardt (17:05): Yes.

Rob Collie (17:06): Was AI leveraged in the creation of those two Word documents?

Justin Mannhardt (17:10): It was.

Rob Collie (17:11): As I was reading them, knowing your workflow, and this is one that I would've guessed, I wouldn't have guessed the first one, but I would guess this one. Well, I thought it what's really funny about it was it like I'm thinking, okay, Justin probably leveraged AI to help him write these documents. They weren't super long, but at the same time, in my version of Office, Copilot is offering to summarize it, this is nuts, right? Yeah. It's like running a compression algorithm over and over and over again until you get to something that's a zero byte file, summarized enough. Well, what does that even mean though? I'm a writer. I have never done this. I know you do it all the time. I'm not taking a principled stand against this. It's so foreign to me. I sit down, look at the keyboard. I have a storytelling type of approach to a lot of the things that I write, even for business purposes. So how to allow it into my workflow would be almost like, hey, hey, hey, what are you doing here? Give me that steering wheel.

Justin Mannhardt (18:04): I'm so glad this came up because I am not anywhere close to being as affluent of a writer as you are. And the reason I think that's important is I'm actually more in on AI expanding people's utility than I am on it increasing their efficiency. So for example, for you, sitting down to write a story or a post or a script, when you have an idea, it flows out of you very naturally. And so the case with AI, one case people might try and make is, oh, well, if Rob works with AI, he'll get to his finished product faster, maybe.

Rob Collie (18:45): Sometimes, yeah.

Justin Mannhardt (18:46): But what it helps me do is it helps me do something at a level higher than I'm just naturally capable at. And yes, I am faster, but it's helping me do something that I don't do as naturally as someone else. Or your example with Midjourney, you have stories and you have ideas about how to tell those stories visually, but you can't draw.

Rob Collie (19:06): Totally true. Can't draw worth a damn, even my stick figures are bad.

Justin Mannhardt (19:11): You might not have the same level of enjoyment, fulfillment, assistance that I would on this specific type of a task. That's a real thing. So what does it look like for me? There's a few things I personally struggle with when it comes to writing a document. I find myself, I get stuck at the beginning a lot. My idea is there, but it's ill-formed and I kind of like, what do I start typing? So I struggle with that. You've seen this over the years. I spell things wrong all the time.

Rob Collie (19:42): The different versions of the words that are all pronounced your.

Justin Mannhardt (19:46): Stuff like that. So how do I use it when I need to write a document? All I do is I just try and explain to whether it's ChatGPT or co-pilot, I need to write a document about this. Here's some of my key thoughts that I want to make sure come through, tell it about my audience. I'm writing this for my peers or my team or a customer. I think this should be a page and then I let it go. Now, in the case of the documents that I wrote yesterday for you and Kellen, I actually ended up retaining very little of the initial output from ChatGPT.

Rob Collie (20:21): Okay, interesting.

Justin Mannhardt (20:22): One of my favorite use cases is to reach this point of like, oh no, not that, or it'll just help me get to a point of clarity. This is how to flesh out what I'm trying to say, but I wouldn't say it like that. Now I can get to the keyboard and my thoughts become more clear and more structured. The biggest benefit for me has been that getting off the blank page piece.

Rob Collie (20:43): And you've turned into a prolific writer over the last 18 months.

Justin Mannhardt (20:48): I know I have conviction and emotion and compelling ideas. I haven't always been great at being able to get those onto the page. If I could just go around and talk to everybody all the time, that'd be one thing, but I can't, so written media is an important part of my life. That's just a habit that's become natural. Sometimes I'm like, yeah, everything it gave me is exactly what I wanted to say. It sounds like me. I feel authentic. Sometimes I reuse very little of it. Anytime I got to write a document, I'm usually starting with a prompt.

Rob Collie (21:23): Can I make a humorous suggestion?

Justin Mannhardt (21:24): Yeah.

Rob Collie (21:25): The next time you sit down to do this when it's applicable, it's not always applicable, the next time you sit down and do something like this, tell it to make you a slide deck instead of a Word doc. See, here's the thing, right? This could unlock another superpower. There's something kind of neat about slides relative to documents. In that one slide at a time, you have the opportunity for a simplicity that a long scrolling word document doesn't give you. Even if you're not using any graphics at all.

(21:49): Just having the slide concept starts to leverage the little bit more visual corners of our brain to allow us to digest step by step, snack, snack, snack, snack, snack, that adds up to a meal. It's like tapas. Justin and I are always joking about how when given the same exact assignment in parallel, he comes up with a Word doc and I come up with a slide deck and my slide decks almost never have any graphics in them. So there you go. There's a leveling up or you could use AI to get you off of not just the blank page, but also the blank slides. The blank slides is still terrorizing you to this day. It's okay.

Justin Mannhardt (22:28): I know.

Rob Collie (22:29): Copilot's here to help.

Justin Mannhardt (22:30): Maybe. First iteration of Copilot for PowerPoint was.

Rob Collie (22:34): Well, I still think it leaves a lot to be desired. But I wonder if from a purely textual standpoint, if it is sort of the same exact experience that you've used with Word docs.

Justin Mannhardt (22:43): I haven't used it quite like that, but I have used it to do speaker notes. I have to present at meetings or present to customers, and I like to have a little something to anchor me and follow along with in the speaker notes, and so even just like, Hey, here's the slides and here's the headlines on the slides and some of the points, give me a talk track. I've done that, not as frequently as the document writing, but I have done that before.

Rob Collie (23:08): Very nice. Okay, so that was number two, document blank page, turning yourself into a better writer. In this case, the quality of your output goes up, but also the quantity of it. Sometimes it's both, a speed thing and a quality thing.

Justin Mannhardt (23:23): People that know me pretty well on how I run my days and weeks. I'm pretty strict on time boxing my life. If I got to work on something, I put it on my calendar, everything I'm going to do today is on my calendar somewhere. I've noticed that if I had to produce the type of documents I did yesterday, that's the type of thing where I might have blocked an hour for each of them. Now I'm like, I blocked 30 minutes for each of them, and so I felt those benefits of just even more confidence like I got to do this task. I feel more confident. I can do it in smaller increments of time. All right, next. Use an AI to take notes in your meetings and calls.

Rob Collie (24:05): Okay. When I'm familiar with the AI transcripts, the AI summary of transcripts for meetings, this sounds a little different. It sounds like taking notes.

Justin Mannhardt (24:17): So one of the things I struggle with is when I would try to take notes in meetings or calls, whether I was taking notes by typing or taking handwritten notes, I would become almost immediately distracted and disengaged to the conversation.

Rob Collie (24:36): You have to check out of the conversation in order to take notes. That is true.

Justin Mannhardt (24:40): And I probably never adopted a great note-taking system to begin with. I was actually degrading the value of the human-to-human interaction by constantly getting distracted, trying to write down things to remember. So having Copilot, yeah, it records the transcript and it can summarize, but it can pick up on what the action items were or what the key things were, so it simply allowed me to not have to worry about taking notes anymore. I just get to be fully present in the conversation that I'm having and not have to worry about my recallability later because something else is going to provide that value to me.

Rob Collie (25:19): Do you do anything explicit in the conversation? You're talking about something and you go, oh, right, okay. Do you stop for a moment and go, hey, Copilot, write that down, or remind me of that, or whatever? Do you put any explicit cues essentially into the soundtrack, or is it purely an after the fact, or is it a mixture?

Justin Mannhardt (25:39): Well, I don't think I do anything unique or special to instruct Copilot to make sure they remember something. I don't know if that I can attribute this to working with AI specifically, but I do state action items out loud and just naturally like, okay, yeah, Rob, I will call Joey and we'll schedule a meeting.

Rob Collie (26:01): That's pretty just natural human discourse.

Justin Mannhardt (26:03): Pretty natural. It picks up on those things, but sometimes people are like, yeah, I'll take care of the thing with Joey and say what was the thing. I would say, this is a great tip. It's not just struggle a little bit. I struggle a lot of bit. Rob, I was not listening to you for the last 60 seconds while I was writing things down.

Rob Collie (26:23): So basically you don't do anything explicitly different during the meeting to sort of tag things as these are my notes, so instead after the fact you ask it what are the action items?

Justin Mannhardt (26:35): That's right.

Rob Collie (26:36): This is kryptonite for this scenario. But I'm trying to think about all the different ways in which I would take notes. Sometimes I have an idea, like a thought, it doesn't make sense to share the thought like, ooh, I have an idea of how we might solve this, but you definitely don't want to sidetrack the meeting with it. That's something I would tend to write down.

Justin Mannhardt (26:53): Sure.

Rob Collie (26:54): But you don't want that in the soundtrack, so other than that, most notes are action items.

Justin Mannhardt (26:59): I'll catch myself doing something like that every now and then where you'll be talking and I'll want to circle back and ask you about something. In that conversation, my short-term memory is working real hard and it's highly effective right now, so I just can write down one word. If I was relying on that one word 24 hours later, I'd be like, what did goop mean? Shit.

Rob Collie (27:21): I get it. I don't know. I would've probably halfway have anticipated that one. I'm at one and a half out of three so far. What else we got?

Justin Mannhardt (27:28): We'll go into the developer crowd for now. These things are working for me. In my job, I don't write as nearly as much code as what our consultants do day to day, but I think these use cases have proved valued to me. Have ChatGPT, specifically ChatGPT at this point, format and comment your code. We talked about that last week, but just format and comment complicated DAX measures, format and comment your Power Query scripts. It does so good with that. It's such a gift to your future self or to anybody else that's going to have to collaborate on this thing.

Rob Collie (28:06): Yeah, I'm looking forward to, for reasons I will explain later, I am handing off ownership of the hockey dashboards to another member of the league. And so if I'm going to be helping him with this, I'm going to have to clean this up. The spaghetti bailing wire contraption that I built, part of it is just the fact that the data sucks. It's PDF and it's in different formats of PDF. When I talk about dirty real-world scenario, I didn't mean this. Yeah, we can handle some dirt, some noise, but this is sewage. Anyway, so I have to clean these things up. The Power Query scripts are awful, just awful. I'm really, really, really, really deadly curious as to what it's going to say. If I just say, hey, clean this up, comment it, change the names of the steps for me. I'm fully expecting it to kind of laugh at it. You fool, why would you do this?

Justin Mannhardt (29:08): One of the next evolutions is when the AI just tells you no.

Rob Collie (29:12): But legitimately that is something I'm absolutely going to do because I mean even I right now, trying to take redundant steps out and right now I've got these folders of hundreds and hundreds and hundreds of PDFs, right? One PDF per game. Every time I want to do a refresh, it's going back and rehydrating from games that happened seven years ago. I'm not going to do that anymore. To make it more manageable, I'm going to take the tables that are already loaded in Power BI, copy them out to Excel and say, here's the historical, and then only look at folders numbered this and higher for the future so that the refresh runs faster.

(29:52): Even just refreshing preview of a Query and Power Query is like you sit there, get up, get a cup of coffee, you can walk away. To do that, to risk that level of surgery, there are steps in there that are called remove column six, change data type nine. It's just like, oh, so gross, and I'm really looking forward to that process and just sort of seeing how much it intuits of intent. I'm sure it could comment things, but the naming of the steps, I'm giddy about getting the time to sit and do that.

Justin Mannhardt (30:23): Buyer be aware with everything, right? But it's funny, you think about people, the way they describe their development process and at the end we focus on documentation and it's like, nah, nobody does that. Nobody likes doing it. It's painstakingly tedious.

Rob Collie (30:39): And there's no dopamine reinforcement for documentation. When you're building something and you're making the impossible possible every step forward where like, oh, that metric didn't exist before, now it does, and now I can build that particular dashboard that I couldn't before and circulate it with the stakeholders and get their feedback, and all of that is right in our human coding, the way that we're made, those incentives count. We're not wired to be incentivized by documentation or cleanup, and that's why it never happens, so this is another one of those examples. It's like this isn't going to be, oh, you're going to be much more efficient about writing documentation or commenting your code or formatting it properly or whatever for the future. No, it's going to happen now.

Justin Mannhardt (31:26): Something that never happened.

Rob Collie (31:28): If you play the cycle to the end, what that does is it makes future work faster. Quality and speed are very often intertwined.

Justin Mannhardt (31:37): I'm glad you mentioned that cycle. Part of our jobs as consultants is we might encounter existing solutions that we are trying to improve or optimize or leverage in a different way. You got to invest time and just, I got to just understand which way is north here and just the ability to be like, hey, can you explain to me what this does.

Rob Collie (32:00): For my own curiosity, I copy a big M script from the advanced editor and I'll send it to ChatGPT and say, hey, rename this, comment it, et cetera. In your experience, has it ever come back to you in a form that doesn't work?

Justin Mannhardt (32:17): It's made mistakes, yeah.

Rob Collie (32:19): It's introduced bugs in the process? Can you tell it, "Don't change the code?"

Justin Mannhardt (32:23): The hallucinations that I've seen the most is it'll incorrectly restate the name of a function, Table.Buffer. It might make a mistake and it's just like Table Buffer.

Rob Collie (32:34): Those seem like the sorts of things that they'll iron out over time too.

Justin Mannhardt (32:38): I've had it miss commas before. I've had it where it's had a redundant in at the very end. Just stuff like that. The overall result has always been good enough where those things while annoying aren't, oh, this was complete and utterly useless.

Rob Collie (32:56): The other thing that I'm kind of cringing about is I can't help it, I'm like, moth the flame. I'm also going to try saying to it, here's the script. I want you to clean up the redundant trash steps in it. I'm going to let it try to consolidate all of my renaming into one step and all of my column removal into one step and all that kind of stuff. The way that the human brain produces things sequentially, I'm positive that I end up with 50% more steps than are necessary. We could shrink the length of it.

Justin Mannhardt (33:28): Yeah, because we realize we get 10 steps down. We're like, oh, I could have removed this column also, but I didn't, but I'll just remove it now.

Rob Collie (33:35): Yeah, I want it off my radar, right?

Justin Mannhardt (33:38): Yeah.

Rob Collie (33:39): I want to narrow my radar every chance I get, but as a result, I clog the other radar. Anyway, that's an experiment that, I've been saving that Christmas, that present. I'm saving that for later. I am so excited to start feeding it the hockey.

Justin Mannhardt (33:54): This would be a potential circle back of how well did it do?

Rob Collie (33:58): Did you ever see Trainspotting?

Justin Mannhardt (34:00): I don't think so.

Rob Collie (34:01): It's a really squalid movie. I mean, these people are all heroin addicts. It is a downer. It's also a little funny, but someone's has to go to the bathroom and it's in Scotland. The camera just shows the door to the bathroom, but it says toilet on the door and chalk script on the screen appears like this additional commentary, which is the dirtiest toilet in Scotland. They set your expectations before he goes in there and it is bad. And I feel that way about the M script for the hockey. The dirtiest M script in the world. This is it. I promise this thing could challenge for that.

Justin Mannhardt (34:51): This is something that I think is worthwhile for anyone that is doing Power BI work. Maybe more so our consultants to an extent is getting up to speed and familiar with a domain that you're not an expert in. For example, I am not a finance person by trade. I have never had a job in finance. I've had to understand financial statements or understand very basic things, but I've had to work on financial reporting, so being able to use ChatGPT to say, hey, I'm working on a dashboard for the CFO. Can we break down what are the top types of metrics that are typically included in this type of report?

(35:35): You get a little deeper and be like, hey, I'm not exactly sure how EBITDA is calculated. Can you explain that to me? Maybe you've not done a lot of sales dashboards. It's like, hey, we'd really like to understand our customer churn or a sales velocity, and you kind of get those things, but you're like, hey, can you explain how those things are typically calculated so I can get to an end? It can save you time. It can also save your stakeholder time instead of going to Rob, Hey Rob, can you explain to me how to calculate ROAS? I can sort of get up to speed and then I could fact check it later and be like, I think this is what we want.

Rob Collie (36:07): So that's an example of using generative AI and the place that we would traditionally just use search engines. I go looking for the article that explains this to me and it's shortcutting that. In essence already the LLM has already researched this in a weird way. It's already ingested all of this from everywhere. It's a lot more friendly than having to go search and then scroll through all the ads and everything and pop-ups and not knowing if the article you're reading is the best one. So this is a huge shortcut for all of that. However, it also exposes you to hallucination risk. Absolutely. There's always the risk that you're reading an article that is poorly written and is wrong.

(36:49): The classic example of this was the other day when I said, I don't want to Google, what was it? The density of fat. When you lose five pounds of fat, how much volume does that take off of your body? And it gave me a completely wildly incorrect number just straight up wrong, but it was delivered in the same authoritative fashion that it delivers all of its answers, and I went around and socialized that wrong answer because it kind of blew my mind and it should have blown my mind because it was wrong. Having it explain something to you like EBITDA, oh oh.

Justin Mannhardt (37:20): Again, buyer beware on generative AI in general all the time. The last example where I did something like this, we were working on something related to supply chain, and I was just like, man, it's been a while since I've worked in that domain. What are the things that supply chain teams are struggling with the most? What are they measuring to try and help with those things? If I could have just a little briefing on what's going on and eventually that put me in a better position to go out and research more. I didn't have any idea what I Google for.

Rob Collie (37:50): Yeah, that's another part of the shortcut is how to write the right Google query, right?

Justin Mannhardt (37:56): Yeah.

Rob Collie (37:57): We keep coming back to this. The fact that these GenAI tools truly for the first time are a conversational interface with any software period, is amazing. Even search engines. You don't interact with search engines in a natural way.

Justin Mannhardt (38:13): Putting things in quotes. Yeah.

Rob Collie (38:17): Right? Yeah. It's like we have no need as a species to learn. Prompt engineer is a thing these days. People who were good at Google searches were already prompt engineers. We were already learning to use the not keyword and the plus thing, and then, oh my God, the site operator. I remember when I used teach classes on Power BI, when I would use the site operator in a Google search, it was pretty clear to me half the room is going to think that was the most valuable thing they learned in those two days was the site operator.

Justin Mannhardt (38:47): If I were to be a consultant again right now, this is the type of habit that I would have going into almost any type of engagement where it's either been some time since I worked in that domain or maybe I'm encountering an industry that I'm not as familiar with.

Rob Collie (39:03): And just bring me up to speed. Unlike the density of fat hallucination example, this is a low risk thing. When you interact with people subsequently, you can still test the understanding that you've developed. And whether you're right or wrong, the confirmation process there is going to be so much more efficient than having them explain it to you from zero, and oftentimes people who are experts in a domain struggle to explain it to someone else because it's become so second nature to them.

Justin Mannhardt (39:33): Yeah, totally, and I'm not trying to get myself where I can show up to this conversation and fake it.

Rob Collie (39:41): Yeah, pretend like you've been a finance bro your whole life.

Justin Mannhardt (39:44): They're going to use words and acronyms and terms that I don't use all of the time.

Rob Collie (39:49): It's respectful preparation.

Justin Mannhardt (39:51): I even had a client that recently started working with us. They reached out to me the other day and they're like, Hey, is there anything we should be doing to get ready for this first engagement? First, you don't need to put any pressure on yourselves that you're going to learn all this stuff about Power BI. It's overwhelming. There's too much to know, but I said, here's a few simple articles that would be worthwhile reading. It's going to familiarize with some of the terms we're going to be using going through this process. You'll hear something and it won't be as scary or you'll understand the importance of something. Goes both ways, I think.

Rob Collie (40:23): Yeah, a hundred percent. Totally get it. So that's five. So far my score would be two and a half out of five.

Justin Mannhardt (40:31): Yeah,

Rob Collie (40:32): I'm batting 50% here, so this is it, for all the marbles.

Justin Mannhardt (40:35): The last thing, and I am getting a ton of value out of this is use AI as your sounding board and thinking buddy where you would normally have used another person.

Rob Collie (40:51): Okay.

Justin Mannhardt (40:52): For example, we try and solve all kinds of issues at our company all the time. We're trying to make things better. We're trying to fix a technical problem that came up in a system or some confusion, and these problems are not always easy or the opportunities we want to go out and pursue. It's not always clear. It requires thinking and conversation to distill down what are the actual things I want to do? As much as I like to pick up the phone and call you and have a conversation with you, I can't do that every single time that I need to think through a problem and find my way towards a solution. Let me give you an example. We are thinking about things like our marketing strategy. We're thinking about how to develop our people internally. These are complex problems that don't have obvious solutions, and so I use ChatGPT.

(41:46): My prompt will be like, hey, here's the issue I'm dealing with or the opportunity I want to pursue. Here's what I'm thinking about. How might I go about dealing with this? And I'll get some ideas and I'll be like, yeah, I like this one specifically and I can continue this conversation just like we might in a meeting. We'll have a discussion and we'll realize like, Hey, this is the right next step for us. That's one of my go-tos is anytime I'm stuck on a problem, I just fire up ChatGPT and talk to you about it.

Rob Collie (42:14): That blows my mind. There are just not many people doing that. I asked it questions one time that I knew that it wouldn't know the answers to just sort of see what it would do when it didn't have enough context. I asked it how much should we charge for our services at P3 Adaptive, just to sort of watch how it went from there. I think it just sort of gave me a bunch of general platitudes. It outlined a whole bunch of things, factors to consider, and I probably would've thought of most of those. There were a couple in there that maybe be like, oh, that's kind of interesting that this thing thought to suggest thinking about that. I'm just wondering how that plays out for you. Does it feel at all like bouncing it off of another human?

Justin Mannhardt (42:54): At times it does. There's pros and cons, so the pros are it's much faster because if I come to you, "Rob, I'm struggling with this problem." You might go, "Yeah, that is a problem. And now we're both going hmm.

Rob Collie (43:08): Good luck with that.

Justin Mannhardt (43:10): So it's much faster at at least trying to provide, here's the things to consider in making this decision, or here's a framework for how you might proceed with this. It doesn't get puzzled. It responds confidently and potentially wrong like you were saying earlier, but it just gets going, and so how that's played out for me because I'm thinking through problems at greater and greater levels of detail with an AI partner.

(43:37): When I show up to talk to you about something or to talk to a team member or a customer about something, I'm more prepared with my own thinking about it, and I think it is even helping in ways I'm not realizing in those conversations because it's like, "Hey, Rob, remember we talked about last week this thing we're trying to deal with. I thought through this, this and this. I really think we should focus on that." And now I'm giving you something much more specific to reflect on as opposed to like, oh, this big hairy problem, and you might still react to that and be like, no, that's not the thing. But now that you've said that, I think this other thing is the thing. I think it's just helping me accelerate through finding solutions and actions and moving forward. I find myself getting stuck in my own head far less frequently.

Rob Collie (44:21): Really interesting. I'm wondering about interpersonal situations, having a hard time explaining X or whatever. I'm going to go try to find some examples in my own life that I can try this out with.

Justin Mannhardt (44:34): Back to the top, right, trying to build habits, going into it knowing we might not know exactly what to say or how to say it to the AI. We might not be wowed initially, but if you pick one of these things that maybe the thought buddy thing is the thing just, hey, every time I'm stuck on a problem, I'm going to try and build a habit of using AI to help me think through that. Sometimes I go in there, I'm like, yeah, this isn't really working, what I really need is a meeting with the team, but most of the time I'm advancing what's going on.

Rob Collie (45:04): Do you ever use the voice interface for this stuff?

Justin Mannhardt (45:07): Very infrequently.

Rob Collie (45:09): Because it just seems like a lot of these things would feel much more natural if you were talking as opposed to writing.

Justin Mannhardt (45:16): People are doing that, right, especially on the mobile apps. I don't know. It's not something I've done. I used to have to drive around the cities for a job I had. This is ages ago, but I had a transcription app, so I could just talk out loud about what I'm thinking about after the meeting and at least get written down. There's something to what you're saying if just word vomit of everything that's on your mind, have the AI try and help you with clarity.

Rob Collie (45:41): I think I'm going to try all of these, all right. I'm going to try to try all of these anyway. How's that? And by the way, in the end, I was a two and a half out of six. Sub-50% in terms of anticipating these, that last one absolutely not, would not have seen that coming. It's so cool though.

Justin Mannhardt (46:01): Yeah. Anybody that's stuck on the starting line as we like to say, I think the biggest thing is just how can you build even just the muscle memory of like, oh, just opening it like, oh, I open ChatGPT. You do need to train your brain and yourself. When I encounter a situation, this is where this thing can help me, and so I'm going to open it and I'm going to go there. So get in and it won't take long before you're discovering your own little use cases that you really like and it'll become part of your daily routine.

Speaker 2 (46:30): Thanks for listening to the Raw Data by P3 Adaptive Podcast. Let the experts at P3 Adaptive help your business. Just go to P3adaptive.com. Have a data day.