Low-Tech Industries and High-Tech AI: a Great Match Explained Through Examples

Rob Collie (00:00): Hello, friends. By chance, I reconnected with an old friend today that I worked with for many years back at Microsoft, and in fact, I met him on my very first day at Microsoft in 1996. He's an extremely competent software engineer and developer, but he's just as savvy and capable in the human side of business as well. He's on a very short list of people whose opinions I trust almost without reservation. Those people are really rare.

(00:24): And we were talking about AI, as one does, and even though he's been very active in the tech world lately, he confessed to having not done much poking around with AI so far. And since we were kicking around the idea of collaborating on something again, even just for old time's sake, I told him not to worry that he hadn't been thinking much about AI, and the metaphor I gave him was that he's standing on the starting line, and I'm only about 100 feet down the track.

(00:48): Now, it's just that those first 100 feet were a really difficult slog. It wasn't like a regular track for that first 100 feet. It was more like the nastiest of quicksand swamps, but not really because it was hard work to make that first 100 feet of progress, that first beginning progress. It was actually because it's hard to get acclimated in the AI space to what's real and what isn't. You got to get your brain into a different place, and to do that, you have to sift through a lot of noise.

(01:17): And the good news here, which is why I use the 100-foot metaphor, is that it's not hard to catch up. As I told my friend, it's been a hard slog through the swamp, but it was like I was installing a wooden boardwalk in the swamp as I went, so now, he can just stroll right on up and join me. Even better, on the other side of the swamp, the ground is now more solid and the real work can begin. The contents of what I've learned, the results of what I've learned over the last, let's say, year and a half, are actually pretty simple and therefore easy to share, even though it was hard work to acquire that clarity.

(01:54): I'd like you to think of this episode, and really, most of our AI-related episodes, in that sort of vibe. We're sifting through the noise and then sharing our progress in easily digestible form. And this episode really lives up to that theme, because it's all about, quote-unquote, low-tech uses for AI.

(02:13): When we hear about AI, I think we all have a subconscious tendency to immediately think about Silicon Valley startups and big tech industries, and that subconscious industry drags our thinking off-course from the beginning, which of course makes sifting through all of the noise that much more difficult.

(02:32): So the antidote to that tendency is to talk about low-tech uses, and even in, quote-unquote, low-tech traditional industries, grounding our thinking about AI in simple and familiar human workflows, because those are the thing. That's where business lives: in those workflows. So we're enforcing a no-jargon zone, except, of course, the places where we make fun of jargon, which is a time-honored tradition around here.

(03:01): Oh. And by the way, I mention another old friend from Microsoft in this episode, Brian Jones, who's been on the show before. That's a different friend from who I was talking to this morning, and come to think of it, we really should get him, the other friend, on the show, too. Mental note. All righty, then. Let's get into some low-tech use cases for AI, shall we?

Speaker 2 (03:21): Ladies and gentlemen, may I have your attention please?

Speaker 3 (03:26): 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.

(03:43): Raw Data by P3 Adaptive. Down-to-Earth conversations about data, tech, and biz impact.

Rob Collie (03:56): Well, hello there again, Justin.

Justin Mannhardt (03:57): Hello, Rob.

Rob Collie (03:58): Back from the tropics, back from your cruise?

Justin Mannhardt (04:01): I am.

Rob Collie (04:02): Now, back to the other thing. The thing that you enjoy second-most to being on vacation is this podcast, I'm sure.

Justin Mannhardt (04:10): There's maybe a few things in between vacation and the podcast, but the podcast is very high on the list.

Rob Collie (04:16): We should rebrand the podcast that. We should call it A Little Vacation in Your Week.

Justin Mannhardt (04:23): It is like that. For the few hours that we sit and record, I get to close all of the other things that have to do with my job and work, and hang out.

Rob Collie (04:32): It also is this reserved oasis where what we do is we think about bigger stuff-

Justin Mannhardt (04:37): That's right.

Rob Collie (04:38): ... as opposed to just the day-to-day tactical grind of our jobs, and stick our heads up and do some of that, a little bit more longer-term, a little bit more reflective thinking. So I wake up this morning and you had an idea.

Justin Mannhardt (04:49): I did.

Rob Collie (04:50): I used your idea and had a slightly different inspiration, so we'll play a little bit of tug-of-war over your idea.

Justin Mannhardt (04:56): Oh. Right on.

Rob Collie (04:57): Why don't you present what you think this episode is about, and then I'll tell you what it's about?

Justin Mannhardt (05:01): Okay, great. I've actually been thinking about this for more than just this morning. We've been talking on the show, behind the scenes, how AI is coming. It's going to change the nature of work, it's going to change the way projects get done. It's going to change the way businesses operate.

(05:18): And so my idea was how do we think that's going to play out in reality inside of companies? But specifically, what's the human impact of this wave going to be? And I think it's going to be net-positive. There's a lot of fear or don't worry about it. We know enough now to speculate what that's going to feel like, so that was my idea for our chat today. So tell me what it's really about.

Rob Collie (05:44): I think it actually might be about what you think it's about. We should just keep repeating this. AI is not nothing, at least on its current trajectory, it is not the end of the world, but you are going to have to change.

(05:59): Going back to the Mike Tomlin interview that I like so much. He trained himself to be comfortable being uncomfortable. You have to do a little bit of that. Not a lot. You and I have now reached the point. I really do think we're there. I think we've kind of crossed through the phases where our subconscious tries to reject things, and the truth on the other side can be calm, but it is not nothing.

(06:20): When we talk about AI, I think everyone just immediately thinks of really high-tech scenarios. The question of how is AI going to be used by us subconsciously kind of morphs into how do we become this ridiculously high-tech company that doesn't seem like it's going to be a natural fit for us? It seems like a completely different thing, maybe even a completely different industry.

(06:40): If you're a construction operation, the real world kind of forces you to be low-tech. You're out there in the gritty world, in a way. You're not Meta. You're not Google, you're not Facebook, right? And you don't want to be.

(06:53): But coming back to the low-tech world of people is a great place to ground things, and so keeping this lens on, it probably doesn't really change the output of our thinking that much. If we start to list the places in which AI is going to be used and all that, it probably isn't that different than if you left the filter off and applied it across all industries.

(07:16): But I do think that our brains get kind of waylaid. I was just reading an article about it turns out there's this one advertising firm that, well, maybe, yeah, we have been listening over your microphone, your phone microphone, when you didn't know it, you know? The thing that everyone's been afraid of forever. "No, we don't do that." And finally, someone says, "Well, yeah."

Justin Mannhardt (07:33): Yeah, we do.

Rob Collie (07:34): Yeah. We've been doing that, and it's on page six of your user agreement that you don't read. It's just so easy to get distracted back into these dystopian or gleaming Silicon Valley scenarios. As a result, you miss the real stuff.

(07:53): So I think keeping the industrial gritty filter in place, the real-world filter in place, kind of tricks us in a way into reaching the right answers, even probably for other industries that are traditionally more techie, or finance, or whatever.

Justin Mannhardt (08:08): A lot of the examples and stories you're probably seeing in your feeds, the most obvious, I think, AI's impact on software development, right? Well, not every company makes software. Or where the output of the work, its main vehicle is a digital medium, like marketing content, or financial report, or something like that, right?

(08:34): And so it obfuscates in a way the idea that this technology is also going to be really, really good for the work that isn't like that, and it's going to have a positive human impact on the people that do that work as well.

Rob Collie (08:50): Absolutely. Even that first example you gave. Writing code, you think we're not a software business. Well, yeah. But you use a lot of software. And imagine if you had more software customized to you that did exactly the things that your business needs?

(09:07): And Power BI is an example of that. There's so much in these modern platforms that absolutely make your, quote-unquote, lower-tech business, your real-world business, will make it much, much, much more effective, and it's just been out of reach for so long.

(09:24): Even before AI is now within reach in a huge way. But now that AI's coming along, we've got these gen AI tools that you and I both believe they're going to get there. Even the things that have only recently become more accessible are going to become even more accessible. If you're a business leader listening to this, you yourself are going to reach the point where you can use the gen AI Copilot interface to build Power BI data models to write data transformation scripts or whatever. You might not even want to, right?

(09:54): But the point is it's going to get faster. The ROI of this is going to go up because it's going to be quicker to get there, less cost-prohibitive. So the fact that writing code, you might look at it and say, "Well, we're a manufacturing company."

(10:09): Well, writing code is going to help you. AI writing code. You are in that, business because you have electronic systems, you have data, you have all kinds of things, both read and write, that you need to do all the time. And having customized, built really just for you, middleware software is a huge game-changer, and you've been locked out of that game for so long.

(10:31): The three categories that I've been breaking things down into are productivity, and then that subdivides into personal productivity and team productivity, so there's different flavors of it. Then I think my favorite might be the enhanced BI category. I'm not saying it's the most impactful necessarily overall, but it's the one that I get the most excited about, and the application of the human interface. Every time I sit down to play with ChatGPT, I'm like, "Oh, right. This thing is amazing at understanding the intent behind my sloppily constructed question, my sloppily constructed sentences that tell it the background of what I'm talking about."

(11:12): I don't have to think so hard. It's able to read the nuance, and I'm able to tell it, "No, slide the dimmer a little bit farther this way or that way," and it does it. That interface against a data model, and if there isn't a dashboard that's already been built for you, or even if there is, maybe you don't know where it is, just being able to walk up to the system with a chatbot interface and say, "Hey, sage wizard, tell me, do we sell more or less of Product X? Is there any durable trend there?" And, boom. Not only are you getting texts answered, but you're going to be getting essentially dashboards built on the fly by this thing to answer your question so that you can see it.

(11:54): That is going to be a huge change. As it turns out, we've gotten to the point now where the bottleneck, the biggest bottleneck on Power BI, other than getting started and getting into the game and understanding that it actually can be done for you, but once you're down that road, the biggest bottleneck actually becomes how well can you anticipate all the questions that are going to come up? The person who builds the dashboards, are you going to be in the room every time the questions come up? Because you're probably not.

Justin Mannhardt (12:20): Yeah. It's the end user's utility. That's a brand new frontier. And the technology is already doing the thing you're describing. Maybe it's not doing a great job of it yet, but that's, I think, the belief we've both adopted. Pretty soon, it's going to get figured out.

Rob Collie (12:35): 100% figure-out-able. And it's figure-out-able to the level that it's not going to be perfect.

Justin Mannhardt (12:42): Yeah. What's compelling to me about the technology is we all value outside expertise, outside perspective. I value your thoughts about something, you value my thoughts about something.

(12:57): These tools possess such a wide and deep range of information that you can get a thought partner about all these different types of issues. I use these things so much in that get me off the blank page. I need to deal with this thing. What are some ideas? And so you take that same experience and you put it in the context of enhanced BI. What am I trying to do? I'm trying to be productive at pursuing answers to questions and insights and generate ideas.

(13:28): I don't know how many times I've been either in a room or on a call with customers, or ourselves internally, you end up at a point where you're like, "Hey, this was great. We did learn a lot. The Power BI stuff is awesome, but we needed to know something else." And it's like, "Okay. We're going to go off and do that and come back." That's going to happen less. A lot less.

Rob Collie (13:49): Let's say you've developed your Power BI investment to a certain point, a lot of good data models, and again, I want emphasize: those things can be built and delivering mind-blowing results for you in very short periods of time.

(14:03): And once you get off the starting line, you've moved from group A to group B, you see things completely differently. Get those investments made, some of them just a little bit, and then see what's going to happen with the generative AI chatbot and user interfaces to those models that you've built. I really do think that's going to change the world.

Justin Mannhardt (14:20): The amount of information that these models, that they're able to consider as a prompt, the amount of information it can take in to process and respond to is getting larger and larger and larger and larger.

(14:35): An example of this is, I would predict this will translate into tools like Power BI and Fabric, one thing end users struggle with that seems so simple is they don't know where to go find the things they're trying to find in the system. Not just on a report.

(14:50): And so it's one thing when you have a search mechanism where it can look at names and titles and properties and like, "Oh, you might want to look at one of these reports." But when it's able to not only consider that, but all of the metadata about the models, query the models, look at the reports, understand the report, and then it's how much more effective everybody's going to be able to be at leveraging their business data to be more educated, more informed, more actionable, more innovative. All of the above.

Rob Collie (15:18): I remember my second year at Microsoft, two years into my first big software project, and sitting back one day at lunch with a group of my colleagues, "Why does it work that way?" We're talking about our own software. The people who made these decisions were sitting around the table, and we didn't remember why we decided something we did 18 months ago. The volume of decisions we make every single day when building a huge software project, we were anywhere from 50 to 100 decisions a day that got ultimately put on disk.

(15:50): So here we are. We're all the human beings that have made these decisions. We have no idea, and I started laughing. I'm like, "You know what? We need a dedicated historian. They need to be part of the software team." And what they do is they attend the meetings, they follow up. Their job is just to soak it all in, and then at moments like this go, "Well, okay. Here's what happened. Rob thought this, and then Malcolm said that. And we said, well, what about this? And then Ben said this. And we actually tried it the way that we're talking about right now, and it didn't work, and it didn't work because of the Windows API, blah, blah."

(16:23): All of that was available somewhere. Like the late '90s, so much of what we were doing was captured in free-form text in various systems, like check-in notes, email exchanges. Hell, we didn't even have instant messenger back then. We didn't have a chat interface. Okay. There were phones, damn it. The phones were a leak. Nothing was being recorded there.

(16:47): The bug tracking system, an AI system that's just trained up on everything that you've done, just like your own internal information, both hardcore data, like data models and things like that, that we're talking about, Power BI data models, but also the softer stuff: specs, meeting notes, email exchanges, right?

(17:08): I guess that's kind of personal productivity meets enhanced BI. I think of it more as personal productivity though because of the way that people approach it. It's the way that they're experiencing it. They're going to experience it as personal productivity.

Justin Mannhardt (17:21): Productivity is underpinning so much of this just because of both the speed, but also the increased agency as a person or a team. Some of the things I'm doing, I couldn't do all by myself before, or certainly not in the time that I'm doing them, or maybe I didn't have all the expertise to think through a problem on my own, right?

(17:42): You mentioned the historian. You me think of something. I was trying to find a specific email from Kellan, but it's been some time since he sent it to me. You know how frustrating it is to go in and search, and has attachment.

(17:55): I ask Copilot, "Hey, Kellan sent me an email about this. Can you find it?" It found it. The ability and the opportunity for it to be able to scan your email, your Teams, your OneDrive, and just summarize and find. Holy cow.

Rob Collie (18:12): I was talking to Brian Jones the other day. We were talking about all the same stuff. We used our short meeting to talk about this. It's really neat to see what the internal lingo at Microsoft is sounding like these days in an organization just full of nerds, with nerd history and a nerd culture. You end up on words that you wouldn't end up on in other places, right? They might be accurate words, they might be good words, but it's a little different.

(18:38): So he was making this distinction between AI tools like ChatGPT that have access to the overall general corpus of all human information. And then there's the other things that have access to the graph. I'm like, "Okay, whoa, whoa, whoa. You're going to have to translate that for me. What does access to the graph mean?"

(18:59): "Well, the internal." I was thinking he meant the office object model, like the APIs of office, right? It could do something in Excel for you. And he's like, "No, no, no, no. The internal, quote-unquote, graph of your organization's information." Access to the graph. That's how they're talking about it behind the scenes, right?

Justin Mannhardt (19:16): They've gained access to the graph. I forget the movie, you might remember this, but the opening scene, there's a bunch of hackers in this dark room. They're speaking in a foreign language, and the subtitles go, "Attack their databases with SQL!" Access to the graph sounds way cooler.

Rob Collie (19:37): It does, right? But this is how over and over and over again in that culture that I used to be a very willing participant. You can sound really cool, you can sound really smart, but also at the same time, ensure that no one else is ever going to know what you're talking about.

(19:49): You can go and build really amazing things and then still end up with an interface. When you go to change the number of decimal places of precision in Excel, you don't know which button to click. You just flip a coin. It's a little bit different type of humanity building these tools, and it's never quite 100% mammalian. It's just a little off.

(20:08): Anyway, so the graph. Access to your internal information, again, structured and unstructured, which is why they call it the graph, right?

Justin Mannhardt (20:14): Yeah.

Rob Collie (20:15): It's a type of data structure.

Justin Mannhardt (20:16): Graph as in graph database?

Rob Collie (20:19): Yes.

Justin Mannhardt (20:20): Graphical, nodal?

Rob Collie (20:21): Yes.

Justin Mannhardt (20:22): Yeah, I got you.

Rob Collie (20:23): So I know, Justin, I've told you this story, but I don't know if it's in the context of this podcast or just some other conversation, but the story about when I went to the doctor feeling lightheaded, and he's interviewing me. Do we know if I've done that here?

Justin Mannhardt (20:36): I know you've told me. I don't know if it's been on the show, but if we wired up AI to all of our transcripts, we could find out real fast.

Rob Collie (20:45): Well, we just clearly haven't connected to the graph yet, have we?

Justin Mannhardt (20:49): Let's jump in.

Rob Collie (20:51): How awesome is that? We just stumbled into that just now. Okay. So perfect example. Well, guess what? We haven't graphed it up.

(20:57): So I'm feeling dizzy. I've been feeling dizzy for a couple of weeks. Finally, I go to the doctor, and he's holding his laptop the whole time, which I'm used to. He's a laptop-wielding doctor. He almost apologizes for having the laptop. It's listening to our conversation and taking notes. I'm like, "Oh, cool. Cool. No big deal. That sounds nice."

(21:15): So we had probably about a 15-to-20-minute conversation about my symptoms and everything, and then he takes me over to the exam table, and he does all these physical tests on me. And then at one point he says, "Okay. Close your eyes, stand up, put your hands out in front of you." I forget. He pushed on my hands or something to try to make me lose my balance, but he didn't tell me he was going to. He says, "Okay. No problem there either."

(21:37): And then later, I said, "Okay. I want to look at the notes." We go and look at the notes, and at that point in the conversation where he was doing that test, it put the name of the test in the notes, which he never mentioned. It figured out from the context of, "Okay, put your hands out in front of you, blah, blah, blah," the [inaudible 00:21:54] test. I'm just like, "Okay, that's cool."

(21:57): And it was a very concise summary. We think about meeting summaries. I think meeting summaries are kind of a sneaky, good thing. For the person who's just been in the meeting, I have no interest in reading the summary. None. I've never gone and read a meeting summary for a meeting I was just in. I might read a meeting summary for one that I wasn't there for.

Justin Mannhardt (22:17): Oh, I'm doing that all the time. Copilot in Teams, it keeps getting better and better and better, and so it's a quick way to get what was talked about, what were the action items. It's doing a very good job of this.

(22:29): And it's cool in Teams too because it summarizes it. There's maybe two sentences. It's like, "Rob and Justin talked about, blah, blah, blah, blah, blah." If you click on that, it takes you to the point in the recording if you want to watch the discussion or read the full transcript. It's so cool.

Rob Collie (22:42): Oh, yeah. So something that you were saying there, how you use meeting summaries already today of meetings you don't attend. This has become, for you, an indispensable tool for you in your job.

Justin Mannhardt (22:57): Yeah. I would not get rid of it for sure.

Rob Collie (23:01): Remote company. It's inherently knowledge work that we're doing-

Justin Mannhardt (23:05): That's right.

Rob Collie (23:06): ... and it's a big team now.

Justin Mannhardt (23:08): Big team.

Rob Collie (23:09): There's a middle-management layer, and you can't be in all of the meetings.

Justin Mannhardt (23:14): Yeah. Couldn't physically be in all the meetings, don't need to be in all the meetings, either.

Rob Collie (23:18): You'd also change the meetings if you were there.

Justin Mannhardt (23:20): You would change the fundamental dynamics in a bad way, sometimes. Think about the social dynamics in an in-person office setting, how fluid it can be to just pop into an office, or word travels more naturally between humans that are in proximity.

(23:38): Technology makes it easy for us to connect. I mean, we're thousands of miles away from each other, and we can talk to people anytime we want. But at the same time, that also becomes a point of constriction on information flowing freely.

(23:51): And so for me, it's allowed me to understand and appreciate what's on people's minds, what they're talking about, what they're concerned about, but also to be proactive when I feel it's needed for everyone's benefit. I noticed y'all were talking about X, Y, and Z, I have some thoughts about this, and hopefully I'm able to be more helpful because I'm actually able to be in tune with what's being talked about.

Rob Collie (24:13): I can't believe this is happening again. There's another parallel in the Mike Tomlin interview, where he talks about how he shows up at the facility in the morning before everybody else, gets the things he has to do as an executive out of the way, like your paperwork and stuff, whatever, answering the owner's questions and stuff like that, but his real job is the performance of the team. He makes sure that he gets all that other stuff out of the way.

(24:37): He has this routine where he walks around the building every morning with his coffee, just saying hi to everyone, just milling through the crowd. And the way he describes it is exactly the way you just did. He doesn't know who the team needs him to be that day.

Justin Mannhardt (24:51): That's a cool dude.

Rob Collie (24:52): What a leader. He's not walking around saying, "I need to find all the places where people are doing the wrong thing." He's saying, "I need to know who I need to be for the team."

(25:05): There are some things that are going wrong. The way he looks at it is he needs to know who he needs to be, who they need him to be that day.

Justin Mannhardt (25:12): Wow.

Rob Collie (25:14): And it's sincere, too. It's his genius and it's sincere.

Justin Mannhardt (25:17): You just gave me this flashback. Before P3, I always was in an office-type setting, and you get so much information when you see somebody, and so this technology in a way is sort of filling that gap in a remote, digital world. What's on people's minds? What's bothering them?

Rob Collie (25:35): It's a whole category. Justin, you're from the future. The future of management bears some resemblance to what you're already doing. This is the kind of tech that I'm sure people listen to this are going like, "Oh my God. This is Big Brother coming for all of us."

(25:51): The thing is, it depends upon whose hands it's in. If it's in Mike Tomlin's hands, it's not Big Brother. It's actually someone who cares, and is truly going to be helpful, and has the right attitude. There's nothing sinister about him and there's nothing sinister about you, but yes. It's going to matter where you work.

Justin Mannhardt (26:11): Yeah. The Big Brother stuff, where the bad applications of this are things like, okay, yeah. Invading people's privacy. So I'm not trying to spy on anybody. I'm just trying to get information from business conversations that are happening within the organization. For example, I'm not reading summaries of people's one-on-ones with their managers, as an example. That's, no. I would never do that.

(26:32): And then the other thing is the clock-watchers, like, "Oh, we can't have a remote workforce because we need to make sure we can see you at your desk, doing your job, typing on your keyboard." Like, "Well, no. I'm not going to use AI to make sure Rob's mouse was moving all day."

Rob Collie (26:45): Well, I need an AI-powered mouse-mover.

Justin Mannhardt (26:46): Yeah. An AI-powered mouse-mover.

Rob Collie (26:48): These mouse-movers that you can buy that move your mouse for you randomly all day, they're detectable by AI, because the algorithm that's moving the mouse is obviously random. But if you had an AI controller physically moving the mouse all day, now it's an AI arms race.

Justin Mannhardt (27:06): So you know when you sign up for something or log into a website, sometimes the reCAPTCHA is just a click of a checkbox? And so apparently, the way this works is there's an algorithm that judges the movement of your cursor to that box, and says, "Human, not human."

Rob Collie (27:22): Which, again, it's just so easy to exploit. These walls matter for an eye blink in the course of human history. It all keeps coming back to the attacker always has the advantage, because the attacker knows where they're going to hit, and the shield has to be everywhere. I don't know what to do about CAPTCHAs. They're already past the point where the machines might be better at it.

(27:46): And it's always so disappointing when you click that checkbox, the one that just says, "I'm not a human," and you're expecting to click it and you're done. You click it and then it gives you the stoplight pictures. The last time I got one of those, I failed. And I don't even know if I failed, right? I selected some boxes and said, "Okay." And it's like, "Okay. Try this one." And I'd do it again, and then I'd do it again.

Justin Mannhardt (28:05): My favorite is the type the letters you see in this image, and I'm like, "I don't know if that's an F or an H."

Rob Collie (28:11): Have you ever tried listening to one of the audio CAPTCHAs? You have to have the audio CAPTCHA because you can't just lock vision-impaired people out. Holy hell. Next time you get one, listen to the audio CAPTCHA. It really shows you hopefully how much more developed people's hearing comprehension is in response to having reduced eyesight, because there ain't no way I'm picking out what's happening in that audio CAPTCHA. The visual CAPTCHA is one thing, but the audio is 5X difficult.

(28:42): One more little vignette before we go. I've been making a presentation, a PowerPoint deck. For those of you who know me, if you've ever seen me speak anywhere, which I haven't really done much of that lately, very, very visually oriented. Always got visuals. Visuals help. Even just holding people's attention, they help. Explaining concepts, they help.

(29:05): And so Midjourney, the generative AI interface for one of the many for generating images has hit an inflection point in its capabilities lately where now it is actually a productive tool for me. Maybe it kind of was, let's say, six to nine months ago, and I just didn't know how to use it well enough yet, but it has gotten better to a degree where I can now interface with it, and it's amazing.

(29:30): This slide deck I've been working on, it's an internal deck, I've probably got seven or eight images in this thing that were all produced by Midjourney, but you're not going to know it unless I tell you. You're going to be very curious where I came up with these images. "Rob, how did you come up with the cartoon version of the Mandalorian looking in a mirror? That was just out there on a clip art site and in the same visual style as a giant wave crashing across the rocks beneath the castle? The same visual style for both of those? You have one artist that did this?"

(30:01): It so perfectly parallels the whole writing code. I have a visual story to tell. Midjourney does not have that story. I can't get Midjourney to tell that story. I can't get ChatGPT to tell that story. That's my story. The thing where I say knowing what formula to write is much more important than knowing how to write it, this is a perfect example of that. I don't know how to draw. I can't at all. By the way, I had this argument with my father-in-law when I was doing this.

(30:26): I'm not putting any artist out of work here by doing this because I wouldn't have had time to hire an artist for this. No. My presentation would've just sucked more. That's what would've happened, and it wouldn't have been as effective. This one will help more.

(30:40): I'm asking it for pictures of unhappy data analysts, and they look too happy, right? I'm like, "No, that's not working." Or I'm asking it to keep the same visual style. Just be a stick figure, damn it, so that we don't know if this is a male or a female figure, and it just keeps insisting on giving me dudes, fully-formed dudes, and I'm trying to trick it into not giving me dudes.

(31:00): And so eventually, I'll be like, "Oh, I know. I'll just ask for one of each," right? I'll ask for a male and a female, and I outflank it, and I'm not getting sort of the tone that I want, so eventually, I give it a completely different set of instructions. I give up on the whole show me an underappreciated data analyst. I know how to make it underappreciated. Show them alone in their cubicle, and it nails that. Or I notice that when I get the Mandalorian looking in the mirror, it looks perfect, but I'm like, "Wait a second. His arm's missing in the mirror," right?

(31:29): Or other examples, like I get pictures back that are perfect and I want to use them, but there's extra stick figures in the background, or someone smiling when they should be frowning. I just open up a simple graphics tool and erase the extra stick figures or change the squiggle line, someone's smile. I tune it just like I would tune code. You see all of the parallels there. I have a story to tell. I know the goal. I know what I'm trying to build, and there's so much nuance in that, that the machine isn't going to do that. I need to be the architect of it.

(31:57): Also, I'm the referee of what looks good and what doesn't, and furthermore, I'm tweaking some of the output and not using it exactly as-is. It's my first truly productive use of these graphics interfaces. It's really, really cool. It's going to change not dramatically, but it's going to change the arc of my career.

Justin Mannhardt (32:17): You have a greater range of utility as a person.

Rob Collie (32:21): Yeah. And I do have a core talent for visual storytelling, but I can't draw. I needed to be born five years ago instead of when I was, right? My whole life would be different.

(32:32): Also, my wife Jocelyn's in this class for her master's degree, and we're not using ChatGPT to cheat. We're using it as a tool. We're sitting there brainstorming fake company names for this fake client of hers in this class, this group project.

(32:46): Normally, I go to Thesaurus.com and I enter words that I know that are close to the core of the theme, and I start getting other words, and I did that, and like, "This isn't really going anywhere." So I'm like, "Well, I'll just describe this to ChatGPT and see what they come up with," right?

(32:59): And I don't know if we'll end up using any of these names, but oh my God, did it get us farther down the line. It's actually also terrifying how good it was.

Justin Mannhardt (33:06): I posted this on LinkedIn a couple months ago. Honestly, one of my favorite outcomes of using these tools is when I go, "No, not that." I know I need an idea or I need to think of something, I write something, or whatever it is. And so I get this system like, "That's not what it is, but that might be st the thing I need to get my brain onto the thing I really want."

Rob Collie (33:27): And you were hinting at this earlier without calling out the specific reference, but you sent me something from Professor Scott Galloway talking about getting advice. Most of the time when you get advice, you're hiring some fancy-schmancy consulting firm, or more often than not, not getting advice is what happens. You don't get advice.

(33:44): But if you get advice, you're getting advice not just from a system that wants to give you advice, but they have their own incentives that are sometimes not the same as yours. These are the human beings you're talking to. They can kind of figure out that you've already decided what you want, and even though deep down they think you're wrong, they'll tell you that you're right anyway so that you hire them again, or they'll give you the advice that they know makes you more dependent on them going forward. You can't trust that agency risk.

(34:12): But there's also his point in this post was like, "You don't have to worry about that now. It truly doesn't have any ulterior motives." Now, it kind of does. The damn thing was trained on motives, so when it, quote-unquote, hallucinates and gives you something that's misleading, it's a similar version, but it's much, much reduced compared to this other thing where it's almost built into the system that they're going to be manipulating you in some way.

(34:34): The same thing you were talking about there. When AI gives you something and you say, "No, not that," if it came from a human, when you say, "No, not that," you have to be cognizant of the person's feelings. If I had sent an artist off to draw me these images, and I threw away eight for everyone I kept, probably more than eight, it's probably more like 16-to-1, you just can't do that to someone else. It took this machine 15 seconds, and it doesn't care. Every time we talk about this stuff, it gets clearer to me.

Justin Mannhardt (35:04): I agree. In the last 18 months or so, the way my habits have changed, and almost honestly in ways that I don't really realize, because it's been this slow, subtle accumulation over time, and then I think, "What am I going to be like three years out from today?"

Rob Collie (35:22): And the inspiration I want to give people listening to this is that for a while, I've seen you using these tools. It's been a year-plus. You gave these things a hug really early, and I wasn't, and I was deep down getting that sort of left-behind feeling.

(35:37): And I think some people listening to this are probably feeling exactly that right now. The thing I want people listening to feel better about and be inspired about is that I am not Justin. I don't lean in on things. You really just need to get started.

Justin Mannhardt (35:53): This stat's, I think, a bit dated. Only 10%, 20% of working adults have even tried ChatGPT. Just go, get a free trial, sign up for it. Check it out if your organization has Copilot.

(36:07): My wife, she works for the hospital system here, they have Office Copilot. She's a nurse by trade, she's a manager. She is not a technical person whatsoever, but she comes home and she's like, "I love this thing."

Rob Collie (36:21): I actually can perceive that the tools have gotten just better enough since you started leaning into them that they don't repel me anymore. They repulsed me in the beginning. They sort of reached this tipping point for me, where using them has gone from scary to fun. I don't think that's just me acclimating.

(36:40): I think, for example, Midjourney has gotten much better. The Midjourney that I first got a subscription to a while back couldn't do the things that it's doing now. It couldn't do text. It can do text now. It can write text on the image that you tell it to. It couldn't do that before.

Justin Mannhardt (36:54): It would make lines that looked like text, but-

Rob Collie (36:58): That's right. It would give you this hallucination, fever dream of what text might look like, and it was really unsettling.

(37:07): Now, it just gives you the text that you want, and it's so clearly like a secondary model that goes in after the fact. The thing that generates the image is different from the thing that generates the text. But it fits. It chooses fonts that match. I don't know how to do that. I'll never know how to choose a font that matches a visual style.

(37:23): Anyway, so every time we sit down and talk about this, things get clearer for me. I hope that they get clearer for the people who are listening.

Justin Mannhardt (37:29): You know what we need to do next, is we need to send AI a transcript of this session and ask it what we learned.

Rob Collie (37:37): And then we'll have a podcast episode where we talk about what the AI said. The content churn. Perpetual content. We're going to get ourselves into an AI feedback loop. The podcast is going to get sucked into the singularity.

Justin Mannhardt (37:52): No. We're going to access the graph.

Rob Collie (37:54): We will become one with the graph.

Speaker 3 (37:56): 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.