Energy Bytes

George Danner, Founder of Business Laboratory, explore how artificial intelligence is transforming the energy sector. George shares insights into how AI is reshaping everything from operational efficiency to strategic decision-making, with a focus on business simulation and predictive modeling. The episode covers the challenges facing the energy industry, such as optimizing resources and reducing carbon footprints, and how AI offers innovative solutions. George also discusses the role of tech startups in driving change and offers a forward-looking perspective on the future of energy, emphasizing AI's potential to uncover new opportunities for sustainability and efficiency.

Join the conversation shaping the future of energy.
Collide is the community where oil & gas professionals connect, share insights, and solve real-world problems together. No noise. No fluff. Just the discussions that move our industry forward.
Apply today at collide.io

0:00 - Guest Introduction George Danner
1:46 - George Danner Tech Journey
9:58 - Unique Business Problems Analysis
12:12 - Industry Decision Making Insights
18:01 - Lessons from Elon Musk
19:55 - Counterintuitive Business Decisions
22:13 - Changing Business Practices
25:30 - Elon Musk Influence in Oil and Gas
33:00 - Artificial Intelligence Journey
33:48 - Understanding Language Models
35:50 - Knowledge Models Explained
39:34 - Artificial General Intelligence Discussion
42:09 - Heuristics in Problem Solving
46:09 - Experience-Based Learning
49:22 - Management Flight Simulators Overview
54:17 - AI Deployment for Practitioners
56:05 - Deploying AI Models Effectively
59:29 - Managing Cloud Costs
1:02:07 - Importance of Design in Tech
1:05:08 - Starting an AI Project Tips
1:10:49 - Vector Databases Introduction
1:11:14 - Exploring SymPy
1:12:01 - Social Media Influencers in Tech
1:13:10 - Google Cloud Platform Discussion
1:16:24 - Favorite Language Models
1:17:33 - AGI Development Timeline
1:24:22 - Finding George Danner Online
1:25:15 - Upcoming Energy Tech Events

https://twitter.com/collide_ai
https://www.tiktok.com/@collide.io
https://www.facebook.com/collide.io
https://www.instagram.com/collide.io
https://www.youtube.com/@collide_io
https://bsky.app/profile/collide-ai.bsky.social
https://www.linkedin.com/company/collideai

What is Energy Bytes?

Welcome to Energy Bytes with John Kalfayan and Bobby Neelon. Your essential guide to the intersection of data and energy. This podcast dives deep into the world of energy, shedding light on how data, AI, and technology are revolutionizing this sector. Each episode equips listeners with insights into the most efficient tools and resources, paving the way for a data-driven future in energy. From technical nuances to broader industry trends, Energy Bytes offers an unparalleled perspective on the evolution of the energy industry. Join us as we decode the algorithms of energy, one byte at a time.

Hey everybody, welcome welcome to a uh a new episode of of Energy Bites. I'm gonna mix up our our intro here. Rad John Calan is here with my faithful co-host Bobby Nean. What's going on? How you doing? Doing great. Survive Nape Week. I know you're I just went to I went went to one party. I went to Cororum's party. That was it. But I heard it. There were some pretty crazy parties from what I Yeah, I was hanging out with Jeremy Funk, so we kept kept it pretty pretty chill. Well, and then yeah, the Frack conference was earlier this week as well. Busy week uh in the energy space. Today we've got George Danner, founder of Business Laboratory. Thank you for uh for joining us today. Well, thank you for inviting me. It's good to be here. We were just talking about kind of how we have this weird, it's not weird, just happen stance. But one of the first people I met when I moved to Houston was ends up being George's daughter. And then I don't even know how many years later that was. Probably five, six years later, we end up Bobby and I when we were working together doing a consulting project looking for software developers. And the one of the first companies we met with was George. I was like, "Wait a minute. You are you happen to be uh related to Star." Yeah, it's my daughter, right? Small world. She's the famous one for uh for how big Houston is, right? It's still a very small community, but I'm glad to to have you here. So, that was the first time kind of we originally met back in the day. And it's I also noticed another coincidence about that is that same day uh we met. I vividly remember uh Bitcoin crashing and me being very disappointed and hard to uh focus on the meeting because I had my my Bitcoin chart up on the uh on the monitor today. Bitcoin is up. It just broke through. It's But even that that year was when it went up to Yeah, it went up to 20 by December. So that was another happening year and here we are back at another happening and prices keep going up. So that's another fun coincidence. But we'll get we'll get into it. Thanks for coming on. you've got a awesome kind of swath of experience, especially related on the energy and tech side. Um, but you know, as we we like to do, we like to ask people kind of just how did you you know, originally get into technology? Were you taking stuff apart? You know, you and I are both mechanical engineers. So, and I assume you've probably broken a few things uh in your in your lifetime. So yeah, tell everybody kind of how you got into that, how you got into, you know, the the tech space andor the energy space and kind of up to where you're at today with Yeah, sure. So, uh, I grew up not far from where we're recording this now down in Texas City, Texas on the on the coast. Uh, my dad was director of engineering at Monsanto, uh, one of the 19 petrochemical plants that, uh, that are around there. So, uh I think something that I breathed over the years of living down there maybe caused me to uh acclimate to the energy industry. But my my dad was an engineer and he imparted he and my mom both imparted on me this sort of fascination with with problem solving. I went uh went to school and became a mechanical engineer. And my first assignment was or first decade and a half of my career was as a factory automation engineer. And when when you do that role, you parachute into a factory and you size it up. You you walk it back and forth. You figure out all the machinery and the machines and the pipes and everything that connects that factory together. you figure out how it runs and then you try to figure out how to make it run better with software. Uh so I did that work for a long time. I loved it. I I I was I saw chemical plants, paper plants, semiconductor plants, diaper plants, uh just about everything that you can imagine that can be made. I I witnessed uh had a front row seat on that and loved that work. But uh that was probably an interesting time too as far as like with technology specifically, right? because that's kind of the early evolution. Yeah, I was going to say what what kind of technologies or you said software. I mean, what kind of stuff was being leveraged at that point? Yeah. So, that that's a really good observation. I graduated in '83 with my undergrad. So, this is sort of like early 80s through the mid 90s and uh we were very much into distributed control systems and factory automation. Um they used this term lights out factory back then. They don't use that as much anymore. Um but uh yeah uh it was it was a lot of programmable logic controllers, PLC's, which are of course are still around. Lots of sensors. They aren't as good as the sensors we have today. They weren't wireless either. Um so we we were we were dealing essentially with fairly crude technology. We we went back in time and go, "Oh, that's relative to now, I guess." But at the time was cutting edge. We thought it was amazing. But you were replacing what? somebody doing it by hand, clipboards and paper and and all of that. So, we thought it was a super exciting time. It was the age of the personal computer, right, as well, which started to come into play when we were looking at factory automation. So, it's very exciting time. We were we were essentially reinventing Yeah. manufacturing back then. uh lots of fun and that was sort of the the birth as well of uh methods like lean principles and six sigma that that conflated with the idea of using automation and technology. Um I absolutely love that work and in a sense I do some of that work today which I'll get to in a moment but uh I I then got the itch to go back to business school. Uh I went back to to get my my MBA. I went to MIT, Sloan School at MIT and um and it's a funny school. It they teach their business curriculum as if it were an engineering or scientific discipline. They they don't really make a distinction. Sure. Between business school and chemistry or physics. It's all problem solving. And I loved that. So, uh it was there I really honed skills in simulation modeling. really discovered th this idea that we can take a complicated system that we observe and recreate it in the computer. You know nowadays we have this catchy term called digital twins. Uh but back then it was just simulation just can we take a a a system that we want to understand and that could be a comp a company it could be the climate it could be a stock market it could be it could be large or small but we really want to get good insight as to how a particular system behaves by recreating it. So I learned at MIT all the tools and techniques for for doing that that digital recreation. So when I left MIT, I went to work at Arthur Anderson in their consulting side, not the audit side. And our job there again was to make businesses better and faster and cheaper by any means necessary. So they would come to us, these companies with a a a problem. Yeah. And they would say, should I merge with this company? Should I add this product line? Should I do this, that, or the other? And um our our team's answer was, "We don't know." uh but we'll find out. So, we would bring the company, if you will, into a laboratory like environment, and we would test it: what, let's try this, let's try that, let's try these two in combination with each other. And then, once we learned in that laboratory environment what would work, then we could layer that with confidence onto the real world. Uh so that was that was part of my work at at Arthur Anderson. Um and uh shortly before Arthur Anderson uh got embroiled in the Enron scandal which took the entire firm down. Very sad time. Uh I left and started my own company that was called Industrial Science. Okay. And I was determined to go out into the world and and just solve hard problems. And I did that for uh six years until my company was bought by an entrepreneur out of Austin who wanted to take some of my simulation software and fold it into his platform that he was building. And so I stayed on a little while to help them get that going. And then after that I started my current firm, Business Laboratory. And what we do day in and day out is sit by the phone, phone rings, and somebody on the other end of the line in some weird part of the world with some crazy exotic problem is telling us, "I got this problem, and I need you to help solve it." So, our reason for being is solving math problems. You know, our our our clients don't necessarily call them math problems. They just say, "I've got some business challenge or whatever." But at the end of the day um when you are figuring out how to make a company better through data you're solving math problems. Yeah. So we have tried to be as good as we can be in both building mathematical models that are helpful in that regard but also explaining the underlying truth in those models through visualization. So, we put a lot of work into visualizing our models. A lot of our models look like video games, quite frankly, and we use video game technology to do that. We could talk about that some more. Um, but um I have the best job in the world because I get to fly around the world and look at all these companies and and and help them with their most challenging problems and write about it. You know, I wrote two two books along the way. Profit From Science, I wrote in 2018, um, about the work that we do. And in excuse me, that was 2015 and in 2018, I wrote a second book called The executives's' How-To Guide to Automation. Automation is kind of the second cousin of, of analytics, um, so it's been a, it's been a fun journey. It's been fun writing about it. yeah uh it's been fun doing the work I still do the work I still write code today believe it or not awesome and it's just it's just the the best poss possible job in the world. I've got a question because I've I've had this hypothesis for a while as someone who's worked with

five six startups. Now, um I feel my hypothesis has always been that most people think their problems are unique, but generally speaking, regardless of the industry or the business, businesses have the same problems, right? they're trying to optimize for something or they're trying to increase revenue, increase profit, decrease cost or expense, etc. And so like you every industry has their own nuance and their own kind of constraints and variables that you have to play with. But I'm very curious because I know you have a variety of uh experience across a variety of industries. Do you think like would you say that that's generally true or do you see like uh do you see it being a lot more complex and dynamic in the actual real world as far as like obviously there are Nuances and limitations across every type, right? But like, for example, the last startup I work with was a edge computing company. Their initial strategy was wide and shallow to figure out which market was most interested, of course, edge being new and all of that. But you know, for example, energy in my opinion would is one of the hardest industries to kind of break into and also have a proven to do work in in that space. Understanding that it's not a plant where it's stationary and you have dedicated internet. You probably have your own dedicated fiber line. You have your own backup generators and power and all this stuff. Whereas, it's like no, it's a rig that can literally pick up and move every two weeks and may or may not have internet connectivity and has to be, you know, H2S environments and Class One Div One and all of these certifications. Or, you know, in a plant it's like there's a lot different; you know, you don't have all of those kind of communication restrictions, but there's other compliance issues and stuff. But even say in the plant, you probably have just more sensor. There's a lot there's a lot of nuances, right? But at the same time, you're still trying to automate a process or use IoT to pull data to then train a model that then ultimately automates or does something, right? Ultimate inform a decision then, right? Well, uh, not only are you correct in your your thesis there, John, I'll I'll give you a a story from my own experience that was very stark that brought this home. I was uh up in the Philadelphia area with a pharmaceutical firm and they were dealing with the subject of risk uh around their drug pipeline And they said, "Gosh, you know, we got these 300 compounds, you know and, they're they're at various stages in the FDA cycle, and you know, what money do we put in what? what compound? You don't always put put it into the leading horse; sometimes you put it into the horse that's, you know, three legs behind." um you know how do how do systematically deal with the risk in our uh drug portfolio pipeline. And so I I was helping them with build a model of their pipeline so that they could they could do that. The literally the next day, I flew back home to Houston, got uh into a conversation with somebody in a in a super major, um, who was dealing with um something like 1100 wells in a basin in North America and, um, you know, which wells do do we and in which sequence do we develop that different geologies, different permitting and and so forth. And I thought, "Oh my god, this is exactly the same problem." So what what you said there, John, was right there. There's these sort of like nuances, right, that get in the way that tell us, oh, this is a very different problem than what you would find in the pharmaceutical industry. And I'll I'll I'll lightly chastise the oil and gas industry a bit here. This may be the only time I'll take this. We need to We need it. We want it. I I just I think culturally this industry, oil and gas, doesn't reach out as as much as it should. That's exactly why I brought it up. Because, like, one of the things I love to do is, if I have a problem in my business or in whatever I'm doing, look outside of my bubble in my industry and see if that problem has been solved or exists in some kind of capacity outside of my bubble. And hey, if they have, what did they do? How did they do it? Is there a way I could potentially replicate that and do it successfully or at least be more informed about the risks that I'm going into because someone else has already done it? And it doesn't have to always be reaching out to Silicon Valley. I think that's where we get like this like it's it's it's black or white. It's oil and gas or Silicon Valley. And if you don't, but like no, there's like adjacent, you know, I mean like industries. You mentioned it earlier, right? Like my first internships were with in manufacturing. It was industrial engineering. We were building wind turbine blades, ironically enough, but that's what got me into the energy industry. When I started working there, I was like, man, I'd never thought about energy, right? Like energy is not going away, right? Like that's a great industry. I should probably think about this. But, you know, when I was on the plant for, you know, my internship projects and stuff were how do we shave minutes or seconds off of this process? How do we optimize everything? And then then my last summer before I graduated, I got an internship in the oil field. I loved it. I became a frack engineer, and I will never forget the first few months on a frack site and. We're sitting there you know, it's the middle of the night, it's raining, we're shut down for whatever reason and I look out the window, and there's like 12 sand trucks just parked, charging us to merge. And I'm like, well, that's dumb. Why didn't Why weren't they notified that we were going to be shut down so that we didn't order them? And now they're sitting here charging us god knows how many thousands of dollars an hour just to be parked and we're going to be down for 12 hours. Like, this is a simple problem that could have been. It's just a logistics problem, right? Right. And then you're like, man, this is super different than what I was used to. But the industry has this weird like cycle of when prices are when prices exceed a certain ceiling, the mentality essentially becomes get it out of the ground as fast as we can before it drops below that. I don't give a [ __ ] about anything else. Optimization, new technology. It's just make as much money as we possibly can during this time period because we know the only place for it to go is back down and then everything shifts, prices go down. Now people have to be more efficient and try and optimize, but then of course no one has the money and it's this weird like screwed up cycle that we force ourselves into. But yeah, that's the reason I bring this up is because there we think that our problems are so like unique and you know difficult in all of these different ways. And it's like no guys, like Hess had that article about adopting lean manufacturing in 2015 and I thought it was hilarious because it's like this was invented in the 70s guys. Like we've been doing this for a long time and oil and gas is essentially a manufacturing process in the worst possible way cuz it's all the money up front and then months to years before you actually start making money off it. Right. Like you know a great canary in the coal mine here is Elon Musk. Uh love him or hate him, he's teaching us a little bit about what it means to run companies in this new era. Yeah. And um one of the things that I find fascinating about what he did was he he's got Tesla, right? It's a it's a car business and he's chunking out cars as fast as he can. Just like you were saying, John, you know, there's a manufacturing focus and efficiency focus, shaving seconds off different processes. Then he's got SpaceX aerospace which is very requirementsoriented, very slow, regulated, uh highly regulated. So what he's done is he's taken sort of a car mentality, an automotive mentality, and put it into SpaceX. And now he's moved more payload, more people into space than anybody any other country combined. Right. Um so we we we need to take note of successes like that. And so he's taking u lessons he's learned from one industry and imparting it onto another industry with great and notable success. And I would love to see the oil and gas industry do that. It's changing. I'I'd say when I was younger, boy, it was really just parochial as could be. It's changing because of, you know, your generation is is thinking a little bit better and differently. So, that's good. I'm I'm I'm encouraged and I'm hopeful. We still got a long way to go to be a lot less insular about our industry. Even taking a step back from Tesla, right? He brought the Silicon Valley mindset to the car industry and then he took the car industry mindset to to space and like that's another thing too is like you know software is now in everything, right? So like if you're not keeping up with software and the cloud and data and machine learning like it literally spans every industry in everyone's life and it's not going anywhere. And so like if you're pushing back against it and oil and gas because that's how we've always done it, it's coming. It's like it's not it's here. It's here. Yeah. Like right. We're trying to figure it out as we speak. And so and and he's got a he's got a kind of a mantra that he uses a lot. It's not his direct quote is actually somebody who works for him. The machine that builds the machine. Yeah. Yeah, he's he's very much into the um there there's the product, the Tesla car, which is a depressive product, but there's the all of the software, all of the infrastructure, all of the digital workings that makes that happen that he's way more focused on and I think that that could we could draw lessons from that in oil and gas is like let's focus on um and you know, you mentioned Hess, you know, doing this thinking about lean manufacturing ing. Okay. They're thinking about Yeah. the the machine that builds the machine. Yeah. It's the how versus the what, right? Right. Well, and even like Tesla, for example, right, they're customdesigned GPU that runs all their computer vision stuff in the car, right? Like no auto manufacturer would have done anything remotely close to that had Tesla not done something like that, right? And then all the overtheair updates. Guess what? Now all the auto manufacturers are trying to do as much updates as they possibly can. And now all the Now all the auto manufacturers like we can't win this game. half them are pulling out of the EV business or whatever. You know, he's he's done so much counterintuitive stuff and that's why I think he's good to learn from. You know, the the these goldplated consultants that, you know, would tell a company to outsource this and outsource that and focus on your core business. He's he's just like swept all of that away. What's the PayPal model? And and has been very successful with it. He's vertically integrated, makes a lot of his own parts both in Tesla and in SpaceX. Uh, so there's just a lot to learn from from Elon Musk and his company. That was the wild thing about SpaceX. Like I think he went to Russia was trying to talk to them about the rockets forever and then like it was going to be way too expensive. So basically on the flight back he was sitting there with some pen and paper and Excel and figured out he's like we can do our own. Yeah. Why are we going to pay for someone for this? Yeah. And they use I'm sure with both Tesla and SpaceX I'm sure they're using 3D printed parts and material. Like, just we had the one of the ANCIS guys on a couple weeks ago and, uh you know that, that specific area, I think, is going to be fascinating uh moving forward as far as, like, part design and how, like we were talking about, you know, as engineers we're taught to a factor of safety and you design, overdesign the part for specific failure and all stuff. And it's like, well, now we can have a model design the part while optimizing the cost and the materials and the structure of the part and how it's actually manufactured while also optimizing for, you know, strength. And so, it's like the way that we were taught to design something is actually bad design now, because it's too robust and has too much material, or like even being able to change, you know the, the forces or the material and how it reacts to the forces across, you know, a longitude and a latitude while you manufacture those pieces with, you know, different types of uh composits and stuff. That's going to be nuts, right? We're we're completely rethinking Yeah. uh how business is done and a lot of these tenants that we've held dear for for the past couple of decades are just being sort of pushed away and it's not just Elon Musk and his companies there's lots of other great exemplars out there but the the I think the overall point to your audience is hey look you know conventional wisdom isn't always the right way to go we should we should be questioning everything yeah well I mean I think even just that whole PayPal mafia that zero to one mentality I mean that zero means you can start from kind of first principles and you can totally rethink how I would do this if I was just starting you know Let's say, even like say Boeing, like they got a 737; they have to maintain that, they have to build on that, and they have to kind of incrementally fix it. But like someone like Elon or someone can come into an industry like that, say, "No, this is how I would build a plane," or, "This is how I would do a rocket if I was going to start now." Like, even NASA has to build off of technology that they've, you know, kind of incrementally built upon, whereas he can come in and say, "No, this is how we're going to do it .and" yeah you know again. And sometimes it may flop, it may work, but I mean, you're going to actually change the industry by doing Well, I mean, I think that's the big thing, too, is it's like if you really want to make a lot of money, you really got to be able to do something that is a literal step change in the way that things currently work, whatever that may be. And so, it's like, yeah, most people normally run into that whole funding problem when they're trying to do that. But, you know, it we talk about this all the time, right? Like if you're a startup and you're starting today, where you're at is a completely different spot than you know an established incumbent Incumbent that has 20 years of data and 20 years of tech debt and 20 years of different ERPs and SAP and, you know, all these systems integrations and having, you know, maybe you acquired somebody and then you got all their crap that you had to figure out. Like, they're just very different scenarios, right? And, like, people don't take that into account a lot of times, right? Like, yeah, you're a new company, so you get to build it how you would today, but but that doesn't mean you're any better than the existing company because, you know, they actually have the robustness and the longevity and infrastructure. Yeah. It's, it's that, you know, the the difference in a startup and like an established company to me, one of the big ones is, you know, in an established company, you're optimizing like processes and you're proceduralizing a lot of things and you're trying to automate as many things as you can. in a startup it's you have you have so many problems that you're trying to solve. It's just: Which, what are the most important fires we're putting out today? And how can we try to process, you know, manufacturize, uh, these things, put them into processes, or automate them, or, you know, do as much as we can with what we have? And so, they're very different kind of things; they're all about optimizing in some way or another, which is why I think, as engineers, like, we love. Right, you...you mentioned it. We like to solve problems; like, at the end of the day, that's all we want to do is solve a problem. And so we talk about a lot too, like when it gets involved and, you know, they put up these barriers, and it's like if an engineer wants the data, they're going to find a way to get it, whether that means them writing scrapers or them, you know, doing all. going down in the basement once they figure out a way around it they're going to they're going to exploit that as much as they can if you're not going to work together with them which is a whole another that's a whole another topic conversation I hope I live long enough to see uh an Elon Musk or a Herb Keller or a Jeff Bezos come into oil and gas that would be and completely reinvent oil and gas. I mean some people have argued that George Mitchell kind of did that. George Mitchell was a a brilliant geoccientist. So so he kind of reinvented the industry from the the downhole kind of point of view. He wasn't a technologist but imagine if George Mitchell was a technologist what he could have could do to this industry. I mean, but even to that point, like he wasn't a technologist, but they tried a lot of new tech very early where a lot of I mean, that's that to me is one of the biggest knocks on our industry is: that we're one of the first people to look at tech, one of the last people to actually adopt it. right? And but it all boils down to, you know, that well, do I want to who's gonna be the one to put their neck out on the line and take the risk, or even just talk? Like, they were also really good about talking about it and sharing the information, like, just like Elon, right? Like, "Hey, we're going to open source this stuff so you guys can have it and help, help," you know? Uh, yeah. the evolution of the development like that's a whole another thing right our industry is so worried about locking down their data and it's all proprietary and all this stuff it's like at some point. All the rock is leased. Everyone knows who owns those leases. Like it's not new information. So like why can't we imagine what we could have done? We've used this example before with like frack hits or frack interference, right? We knew that was happening for years. Took them forever to figure out why. And then the best models that have come out of that are basic static dynamic models of I've got a pressure over here. There's this distributed load on a rock which is essentially my table that I'm pushing on. I vividly remember this from engineering class. You have, you know, a single point over here. How what are the forces on the two legs and it's like that's exactly what we were doing. But, you know, and it's this very basic simple physics based model. And it's like we took almost a decade to figure this out. And God knows how much money we not only spent, but how badly we screwed up our reservoirs because we over drilled and put too many wells or we overfracked them and we were just, you know, draining each other. Like there's no telling. No one will ever know that. But imagine if we had been sharing this data with each other for much longer and earlier and we were more transparent about it. This [ __ ] could have been fixed a lot faster and having a lot of really smart people work on the same problem. Yeah. Yeah. Well, that's another thing. All these big companies have brilliant people, you know, Kico, BP, whatever. I mean, but then you have the same brilliant people working on the same exact problem and no one's really sharing that information. It's really the the change needs to come sort of at the top. Sort of I'd love to see that pioneering spirit come back into the industry and and one day it will. I'm I'm optimistic that it will just that's not the current situation because to that end like my first thought was like say an EOG because I mean they're like almost basically an IT company that drills happens to drill wells right but at the same time they are like the the the tightest like you don't know anything about they can't talk to people like they don't do like so it's all kept close to vest so then it kind of blows up that whole like sharing that Yeah, no. I would say, uh, probably like the EQGs and Devans of the world would probably be some of the leaders in that kind of realm, I would say. But I agree with you, there's still lots of room for for improvement. But the problem with it is, again, most of those companies are 20 plus year, companies right? Like, they have all these legacy assets they've acquired god knows how much through all these acquisitions and devestaturures and stuff, and it's like they're not starting from zero. So they have a lot of these constraints. Essentially, that's where I was going originally. Like, people talk about these nuances and stuff; we're all math folks, like, it's literally calculus, right? Like, you have your constraints. You know what they are, right? The constraints just change depending on what industry you're in. Like that's all it is. Well, even like more simple than that, I mean, and this is where I kind of want to maybe pivot back to some of the work you're doing, but like you talked about solving math problems. I mean, like most of the problems that people have are still counting problems. Like, can you count this? Can you sum this? Like you mean like it starts there and then piece of data from there to here, right? A lot of the work that we do, I'd say 99% of the work that we do is basically algebra. Yeah. It's not uh a lot of really occasionally we get into some fairly sophisticated math. Yeah. But a lot of it's just algebra and a lot of the you you you peel away the layers of these business problems and you find kind of just an algebraic formula there. It's logic, right? Maybe a maybe a big formula, but it's algebra at the end of the day. Yeah. No, that's a that's a very good point, right? like so many of those processes is just moving data around or these nested logics of just basically algebra right but the thing that u the thing that arguably makes it complicated I think uh the what we've observed is the degrees of freedom right so if if you think about think about a simple problem okay I want to arrange 13 people around a dinner table for a wedding let's say okay here's here's aunt Mary over here and here's Uncle Jim over here and so forth. I got 13 people to arrange. So, how many combinations are there to arrange 13 people around a dinner table? Well, I I can tell you the answer. It's 13 factorial and that's 6.2 billion combinations. Still got it. And very good. Well, well done, John. So happy. I was going back. Is it a combination or permutation? Yeah, I couldn't. But but people think oh my 6.2 two billion combin. That's that's incredible. I we're just we're just doing putting 13 people around a table. How could it be that complicated? Well, it it is. That's the number of combinations. And and and so what do we as humans do as humans? We just kind of dumb the problem down. Okay. Well, I'm just going to get a piece of paper and I'm going to put Aunt Sally over here and Uncle Jim over here. It's like, I know they don't like each other, so we won't put them there. Right. They they effectively Yeah. There's your constraints. Dumb the problem down. Yeah. Exactly. and they use simple conraints. And you know what? We do the same thing in business. We take a a really complicated problem like that that doesn't look complicated on the surface and we dumb it down. And so we solve it intuitively. And I think that's the thing that that makes the work that we do fairly complex, because our models have to go out and search this vast solutions space to find the one perfect combination of those 13 people at the dinner table that nobody sort of gets into fights or throws wine at each other and that sort of thing. Uh but but that is the state of play today. I I think we have more degrees of freedom than we had 10 years ago. We have uh more people at the dinner table than we had right 10 years ago and yet we are solving them with the same sort of dumbed down principles and intuition and rules of thumb and that's just not going to be adequate going forward. We're going to have to use math. We're going to have to engage math, technology and software to solve these kind of problems going forward. I mean just at the simplest level, right? There's so much data now that literally as humans we can't there's no way we can comprehend. And then the flip side of that is a computer is really good at that if we can tell it how to be good at that, right? Like also we hate doing that generally speaking and computers don't care, right? Computers are great at repetition. So let's use them for repetition. Exactly. And not use them for necessarily the judgment aspect, right? Yeah. So, and I think this kind of kicks us into like the journey to to AI kind of where we're at now. And we can talk about where it's going, but even I kind of want to get your opinion like we we I think we've always had statistics, right? And then, you know, I'd say through the last decade and decade and a half, then there was statistics and data science, and there's obviously that overlap. But then there's kind of that blending of data science and machine learning, and then there was deep learning, and now we're getting into, like, the generative AI. And, like, is some of that stuff in the middle just going to be lost, or has it always existed? You know, it's kind of one of those things. And then, you know, where do we see see it going from where we're at now? Because it's advancing rapidly. And then, I mean, just really how much math is going on, you know, even within like the generative AI because? I I think it was always like deep learning there was a ton of math going on, but it was this black box, right? But... That's actually a very good point, and I think you're a good person to ask to frame this question in this way. Explain how some of these model, the new models, work generally speaking, but explain it to me like I'm a CF FO at a oil and gas company and I know nothing about any of that stuff, right? Like because I really do think that's one of the things that a lot of people get hung up on, right? Is it's like, well, how is this working? What is it doing? Can it think for itself? Like all there's all these questions around the AI stuff and the people that are making the decisions a lot of the time don't really know like what anybody is talking about. Even, you know, at the deepest level, I use GPT and I use a bunch of language models all of the time. that doesn't mean I actually have a true understanding of exactly how they work, what the bounds are, what their constraints are, and all that stuff. And as all of my math teachers have always taught me, make sure you know the bounds of of your conditions or of your um equation before you go and start using it. And so, I think that that would be a really good kind of snippet to get out of there, just to to, like, let the average even just tech focused person, but just that doesn't have that layer of of knowledge, to kind of understand, like, how they work, what they're actually doing, what they are not capable of doing, you know? What to look out for when people are promising you that they can do things that they actually can't really do, because there's a...I mean. the the creative marketing folks at some of these companies, especially in Silicon Valley that have lots of funding. Like, I'm getting really, really good ads about, you know, hey, here's this AI sales agent and he sounds like a real person and he can take all of your sales calls for you. Like, that sounds awesome. I have a I actually saw a video on on Real the other day where they said, "Okay, now write me a uh like they prompted it to write him to ask him to write him like a social media post." And it did. It just regurgitated it back to him. It's like, yeah, there's some there's some, you know, areas of of concern with some of those things, but Okay. Well, there's a lot there's a lot to unpack. Sorry. That's uh yeah, let me let me do the do the the best I can at all of this tying it together. First of all, let's let's just acknowledge that AI when you say AI, it's this giant umbrella and there's lots of sort of slices under that umbrella, just nestative statements, right? Yeah, exactly. Um, so the large language models, uh, is is one one thing that people sort of get excited about because it's been consumerized and and hey, I can get on chat GPT and I can I can type these things and I get this seemingly responsible answer. So the pivotal word uh in there is language large language models. It is a language model. It is not a knowledge model. So, for example, if you were to go to chat GPT and say, "Write a love letter to my wife. You know, we met 10 years ago and we have two children and yada yada yada." You give some uh salient facts about your relationship to your wife and it will write you the most beautiful love. It'll bring you to tears um and you think, "Ah, wow, that is really intelligent." But it's not. So, language models are trained on the internet. So it is responding to your ask of a love letter. So it's been trained on thousands and thousands thousands of things called love letters. But it doesn't understand the concept of love, relationships, marriage, etc. It's regurgitating to you. It's essentially tricking you into thinking it's intelligent. Now, does that mean it's not useful? No. I'm I'm not suggesting that. But do know the CFO of oil and gas firm know that um language models are good for production of narratives. So if you're talking about using it as a sales agent, you know, that's all that's all well and good, but the thing that is coming, it's kind of around the corner is actually a knowledge system. So you you type a question into something, it's giving you an answer based on actually knowing comprehension the answer. Now that's a that's a very different training uh scenario where you're curating knowledge. So uh say if I were to go into an oil and gas firm and there's um there's frack plug design which is like it's it's its own topic, right? People that dedicate their careers into designing frack plugs. Okay. And you always use a a a ball frack plug, you know, in this situation and another frack plug in that situation and so forth. I I don't really know the domain, so I'm talking out of my hat here. But um suffice to say, there's there's sort of factual knowledge that you can gather about that very niche topic and somebody who's intelligent about that can uh curate that into a system. And so when you ask about that topic, it can give you an actual answer. That's called a knowledge system. And the closest thing to a a public domain uh example of that is something called wolramal alpha. So if you want to go out to wolramalpha.com and and toy around with it, you'll you'll immediately understand the difference between that and a language model. It's really good for uh solving math problems. Yeah, I see that you've done that before. If you have any math problems that you're interested, right, or if your children, you know, helping them with homework, that that's ideal. Great parent tip. But but those two domains are starting to come together and and the two communities of scientists are sort of coming together. When we can actually have something that behaves like chat GPT but actually is the source of real curated knowledge then we've really got something. So that's coming is that I mean is that would you define that as AGI at that point or are we still well you're still the next step? Um it's still sort of hardcoded if you will. Yeah. Um, so in other words, some some human in in who's an expert in that domain is sitting down and and encoding that into a system. So I would not call that um AGI and that's artificial general intelligence. And for those of you in the audience that that don't know what that is, that's that's not AI. That's AI for AI. That it's learning how to learn. So that's systems with no presupposition that that can can go into a subject, learn it, and then come back out. Yeah. And and and that's AGI is like the next wave. It's fairly immature, but that's what a lot of people are sort of working on, which is which is this sort of AGI domain. And and I I'll give you a little example to sort of teach the audience what this means. a couple of professors from Cornell a few years back, they took a metal bar and hung it from a a hinge and swung it back and forth like a pendulum. And then they took a second metal bar and hooked it to that first metal bar. So now it's swinging back and forth in this sort of complicated wavy way. And they took a piece of software and they had that software just kind of sit there and observe that back and forth, back and forth, back and forth. and it ultimately deduced Newton's second law of motion. Now that is intelligence of another order entirely. Again, it's not there because it had no pri prior knowledge of that, right? No prior knowledge. No prior knowledge of phys. You we didn't sit down and teach it physics, right? It deduced that. So that's where a a lot of this is headed because if you think about chat GPT, you know, what was the diffic tomorrow? Well, the reason is because I don't have this massive training set uh to go out and train it with. And even if I did, you know, it's a lot of hard work that a lot of people at OpenAI put into this. Uh so so we're what we're trying to solve right now is the the training problem. Training is is difficult. It's manual. Once we solve that, then a you'll see AI go into lots of different domains. But back to the answer to your overall question is what is this AI thing? and you know how should people sort of practically think about that you know I think one of the areas that doesn't get a lot of mention which I think is the most valuable is something called heruristics now heruristics is just kind of this Latin derived term that means rules it's rules so when a human does something um you know they they function on a set of rules now they're not always conscious of those rules excuse me but they have rules tools by which they behave. And the most valuable application of AI these days is taking humans who are smart at some discipline and pulling those rules out of their heads and then codifying that in some system. I was just recently uh at a car plant in North America. It's for this particular very large car brand uh their second largest plant in North America. And it was fascinating to watch, you know, cars come off the line. They had one gentleman at one part of the line who um who does something called offlines. So a a car will come to him and it's got a little blemish. Yeah. And it may have a blemish over here on the door or on the hood or whatever it might be. And there's hundreds of different variations of this. And he could look at that and say, "Ah, okay. That was because robot number 32 was misaligned, blah blah blah blah." He could he could do that just intuitively because he's been doing it for 23 years. This gentleman's in his 60s, very nice guy, but he's not going to be around very long. How um do we capture and codify institution his knowledge knowledge your shop for men or your people call it tribal knowledge and put it into a system that's that's huristic. So um at some point we're going to sit this gentleman down. We're going to ask him lots and lots of questions and probably what's going to emerge out of that is this big decision tree. It's probably going to be a big old diagram, you know, size of a wall here in the studio. Um that uh that then we'll put into software. Now, let's turn the conversation back to oil and gas. I've been around oil and gas companies a lot and there are people like that particular gentleman. Yeah. All over. Yeah. Know a lot of things. They can hear the compressor and they know exactly what's wrong with them and and they're and they're not good at articulating how they do what they do necessarily. I mean, we call this the Michael Jordan paradox. You know, Michael Jordan is the greatest basketball player that's ever lived, and he's going toward the goal, and he's dribbling that basket, and he's looking at the goal, and he's moving side to side, and he makes this magical dunk. And you sit Michael Jordan down afterwards and you say, "Okay, Michael, exactly what was the physics that you used?" How did you know? Yeah. Why did you move left instead of right? And and he's going to really struggle at at describing that to you. So there's uh funny enough in in the AI world that we live in where we're trying to build these practical applications, there's a lot of sort of Dr. Phil moments we have where we have to use psychology and and and a little bit of art to draw people's knowledge out of their brain. Absolutely. And then put that into systems. And even when we would do that, they're not perfect. So we have to test them against the human and so forth. But um I you know I just want to express to your your audience that um large language models are great. They're interesting and they do some very useful things. But there's this whole other um area of application of AI that that people don't really talk about a lot that is just as powerful and I would claim is even more valuable. Yeah. Well, especially right like in an industry like this. I mean, we we're in the middle of the great crew change or whatever they call it, right? Where you've got that 20-y year generation gap. Every operator knows our knowledge base is basically a leaky bucket and they're trying to figure out all these ways of capturing that knowledge. And a lot of it to your point is it's not the heristics approach. It's the okay, well, can you, you know, write up a summary or something? Just, it's not specific enough and detailed enough, um, to really truly pass that on, because it's one of those it's learned knowledge, right? Like, the only reason those guys know that is because they've done it for 20 years, and that happened for six months in a row until they fi finally figured out what the issue was, and now they know forever that if that ever happens again, that was the problem, right? But, you know, I think to your point the, best the people that do that best do it where the user doesn't even know that it's happening. Capture is the example that I use all the time, right? Like it I didn't even know until a few years ago that that's what capture was, right? Capta is you training Google's uh computer vision model, right? It's a security feature, yes, but it's also you the user who most of the world has no idea that they're doing that training this model for them. And that's how software should be in my opinion. That's one of the biggest issues with a lot of oil and gas software is that it's older software. Therefore, they didn't have the technology that they have today. Okay, so it wasn't built understanding that, hey, as the user is tagging these things on a frack job, maybe I store that data and keep storing that data so that I can build a model off of that and then just do it automatically, or just like we...I... that's a specific example to me because I was a frag engineer and I've tagged thousands of stages and that software was like, you know 70, 80%industry do dominant software and they never did anything with that information, and you're just, "What is wrong with you?" Also on that topic to your point about you know the computers not understanding certain things that humans just immediately get. Bobby and I were trying to figure out how to pick start and stop of frack stages and then different parts of frack stages. And I was like you and I can look at this and tell but training an algorithmic model to go off of the data the hard numbers is going to be much harder. And so we actually B probably built a pretty successful model just off of a visually tagging the model as a human which then of course updated the data on the back side. But just little things like that, right? Like tagging models doesn't have to be horribly difficult if you are creative about how you allow the user to do that, right? Whether it's in production or whether it's just engineers in the back office that are just tagging models. But hey, adding that gooey step made it infinitely faster. And again, I don't know if it ever got there because I left. I mean, even it was one of those, like, say, at that point our account managers were supposed to be watching the stages like just as a, um, you know, as a QC for the data. That was something you come and do every day like, why not? While they're just doing it, just have, you know, brush it on the on the thing, and then that saves those outputs, and then you can use that to train. No, I I think that's a a big thing again with us being so private with our data and how we operate and how we practice. It's like look at the the leaps that happen when you allow the crowd when you crowdsourced your knowledge, right? Like look at how far GPT has come in a year just because they they gave it away for free. So now they had this huge user base that's coming in there and most people don't realize this, but you're training the model every time you use it. And so their model gets better and then they get a new version and they keep iterating and spinning stuff out. And that's how software development with AI should work. That's how it should work. But that's not how most of the industry does it, right? And so now we're having to play catch-up. I feel like almost. It's funny that you mentioned the the great crew change as well because that's another thing that the industry thinks it's the only ind oil and gas thinks it's the only industry that's dealing with that and it's not. No. Uh trucking, even airline pilots are are going through this. And um and what once again I think the answer is analytics and and data. The most encouraging thing that I've I've seen development that I've seen in trying to close that gap between the older generation leading the industry and the new generation coming up is these uh concepts called management flight simulators which is the idea uh just like an airline pilot. uh they have to climb into a machine twice a year and you know get trained and they're given various exercises. Okay, you you have an engine failure. What do you do? And they're supposed to follow a protocol. Um now we're starting to do that in business. We have simulation models that allow immersive experiences where okay congratulations you now run Devon today. You run this field today. uh you make all the decisions, all the uh technical decisions, the economic decisions, and you're probably going to fail. Yeah. Uh fine, no harm, no foul. Back up. Let's try a slightly different strategy. So, it's giving uh people an ability to practice and to fail and to then to correct. Um on on occasion, I've dealt with uh professional sports teams too. I built analytics for professional sports teams. This idea of practice is just so ingrained in them. And these are the best athletes in the world, right? And they practice seven hours a day, even in the offseason. And let's go into a company, a wellp performing company. And do they practice? Do they role play? Right. Generally, no, they don't. Um, but they have all these unforeseen circumstances sort of coming their way all the time. Why can't we use analytics, these immersive environments, these simulators, if you will, uh that allow us to roleplay certain exercises that have either actually happened or uh could theoretically happen to us. Um there's there's going to be some a role for AI in some of those systems. there's going to be a role for for expert humans to evaluate, you know, an expert pilot evaluating how the young pilot did, you know, when he encountered a bird strike. Uh that there's there's a role there. So other industries are uh taking this great crew change sort of problem on and then are closing that loop with things like analytical models, the management flight simulators and I'd love to see our industry do that too. Great. It's a great idea, right? Like we we have drilling simulators that the drilling engineers use to get certified and do all their training. Why don't we do that for I mean frack and everything else, right? An entire field. Yeah. The uh the whole process that especially like that's what always gets me right is our industry is so hyperfocused on risk as far as like leadership and management and that's one of the biggest things they weigh for every decision. But then it's like but then the riskiest things that we do are still the the the furthest behind tech-wise. They're still the most manual. They still have the most people involved, and like, we're shaving that down finally and stuff. But it's like, guys, you know, yes, like a a headset might be expensive, but if you've got a mechanic out there that has that and he's connected, and he can now do all of the maintenance without any questions or and has all the manuals and stuff with him, like that's a benefit that reduces your NPT. It reduces your downtime, which means you're going to have your wells on faster. You're going to start making money. There's all those like compounding things like we talk about in the manufacturing process they oil field so it's different. No, it's a manufacturing process. You have to do this before you can do this before you can do that. Right. And like then there's all these constraints and ties to them uh to each other. And so it's it's uh yeah that that's a a really good idea I feel like because there are so many hardest part about business is you don't know what's coming, right? Like if you knew like a a hurricane was coming just like it's like a hurricane versus tornado, right? tornado pops up in the middle of the night always. You can never see it and it's, you know, 2 am and the alarms are going off and you're like, "Oh, well [ __ ] I guess there's a tornado." Hurricane, you know, a week ahead of time it's coming. You can leave. You could prepare for it. But that's how most businesses they try and prepare for as much as they can. You just never know what you know is going to happen, right? Yeah. It's this is the most exciting and discouraging industry ever. And it's discouraging because we're not where we should be because we look around at other industries, we go, why why aren't we there? But it's exciting in that this is an industry that probably has the greatest potential to apply some of these things that we're talking about than any other industry around. Yeah. Yeah. The the benefits that they will get. I completely agree with you. So, speaking of application, I kind of want to maybe pivot a little bit because, like, we talked a lot about, like, the model. But, like, say as a practitioner, like, how can I even, like, deploy AI? Like, even just whether, it's is it cloud hosted, or what tools are out there? And even, it's just an evolving space. I mean, there's 20 different vector databases and that's actually all that kind of stuff. But like I mean, we're finally maybe kind of going back to where I was talking earlier like we're finally at the point where there's better tooling for say MLOps and maybe we have better processes around it. It's like, "Oh, we just got good at that now .like" is, is some of that obsolete, or can we take, you know...but now, how do we deploy these models? Or do we...are we only at the beck and call of Microsoft and Google to create it for us and we have to use that? You know, just kind of curious on your thoughts and how you're tackling that currently. Definitely an an aspect of this where it's like they don't know what they don't know. And so it's like as someone who, you know, I like to [ __ ] about this because it still pisses me off that Google did it, but when they came out with their Vertex Studio, it's like, oh, that's right up my alley, right? Like low code language model deployment and I can call it via API. I can do all that. Spent two days trying to get the stupid model deployed. Just stood up on their server after spending an hour googling. I find all these forums in the Google forum saying how Google none of the Google servers can support a Llama 2 model, even though Llama 2 is the first model that shows up in Vertex when you open it. And it's like, you know, how do people... people, like I said, they don't know what they don't know. So it's one of those things where it's like you've got, okay this engineer that has this really interesting model, and it's working on his computer for one brack spread, let's say, or drilling rig. It's like, "Okay, well, how the hell do I get that across the entire company?" And it's like, there's a lot of things that have to happen there that they have no idea about, as far as, like, "Oh well, it runs on my laptop; okay, well..." now we got to find a server for that to run on so that everybody This came up in a webinar I was in last week, too. That was like, now you take those federated, you know, learning, or you have like, um, say like Saudi or something where, like, the data can't leave there, but they want to leverage what was outside of it, but also with internal, but then, like, how do you blend those models together but? I we're getting deeper down the rabbit hole than I wanted to, but just, I'll let you kind of get. No, no, it's a legitimate question and. And you know I, I feel this pain because, you know, I'll write some code and I'll be so super proud of it, and I'll do it in record time, and then it comes to deployment, and then we got to deal with authentication, and you know, just all that junk, and so deployment is almost as hard. as as the and time timeconuming is writing the code. So it's really frustrating. Um I will say from our point of view when we build our models um you know we tend to kind of go toward a web stack that's sort of a almost like a roll your own easy web stack. Uh so Python is our primary language and we write a lot of our code in Python no matter whether it's a simulation model or a machine learning model. I mean we write it in Python. Use libraries of course but you know Python kind of glues it all together and then um we put it into we put our code into docker containers uh on Google cloud platform and we we have not used sometimes we're forced to use Microsoft we we have not used I don't have anything against Microsoft but Google cloud platform is is probably the easiest of sort of the big three that we found. That's I always heard for like developers. It's very developer friendly. It's just like that's why I like it. Pretty pretty plain vanilla. Uh so so Docker container on a goo Google Cloud Platform and then we just do kind of simple authentication stuff that that we do. I'm not an expert in that part of it, but um so our web stack is is just kind of very simple. Um and uh and we've we've used we've used Flash uh as sort of our application platform. Again, I'm I'm there's other people at our company that that kind of tend to specialize in that, but I I think the the overall answer to your question is we do sort of roll your own, number one, and number two, we keep it pretty simple. I yes, Microsoft and these others are coming out with these supposedly great tools and that's wonderful and I encourage that, but u it's like is the juice worth the squeeze with with those? It's just they they're and it can be it can be expensive. I think as soon as you start getting into any machine learning, deep learning and now into AI, I mean like just those things are constantly running and they're they need a lot. And that's another area of this where I absolutely am confident that they don't know what they don't know. And one of the things I learned selling trying to sell edge is that most of the energy companies are not paying much attention to their cloud bills because they're still running on all these credits that they got to migrate over to their cloud to begin with. And so, I am very interested to see how all this shakes out, especially with the cloud credit piece. Because yeah, that one Llama instance the thing I did like about it, like it had a good guey; it told me what my estimated cost was going to be per month, and it was going to be $30,000 just for the one Llama 2 instance. on a server that I couldn't get it spun up on. Um, so like who knows what it would actually cost, but that's something that yeah, like these companies have got to understand and think about. So, I'm also very kind of bullish, honestly, on hardware, and not just for everybody else, you know, using AI. But it's it's already to the point where it's actually cheaper to go buy and build your own server, even with the $10,000 GPUs that they require, than it would be to go spin up, and you get it back within a month or two. Like that ROI is super quick. The pendulum is kind of swung back. That's what we're talking about. Cloud ID got a surprise bill for um I mean for us was pretty big. I think it was $500 from Google Cloud Platform when it's usually like 20. Yeah. And it's because we put a we put a giant database onto Google Cloud Platform and boom, we got a hit hit with these huge license fees and uh Yeah. So, so I guess the message to your audience is kind of watch out for that. Be careful. Be, you know, kind of limited in what you do. Scale up things, you know, carefully. Um I mean, if you've got unlimited budget and you can write big checks to people like Microsoft, I mean, go ahead. But most most companies that I know of aren't in that position. Yeah. And it comes down to is it necessary too? I mean again like you said some people don't know what they don't know. So they're like oh I guess this is what it costs to do this and you know we're going to get an ROI so let's just pay it. But you know it doesn't you don't have to spend a lot to solve you know tough problems. You shouldn't have to. No. Well and you know we're we're still in the first of this game right? And so like as the longer you wait the more tools are going to be developed. I'm already starting to see I saw something on Instagram I think yesterday from a guy and it was uh it was this site that basically just gave you a real-time cost of all the different models on the different platforms. I was like, "That's actually great idea, right?" Like, and like you're going to start seeing more and more things that pop up that are, you know, insight into cost, or maybe even these brokers that I've seen people talking about where it's like it goes and finds the cheapest hardware to run it on you. So it's almost this weird distributed cloud kind of concept. Um and so like that's that is one benefit of waiting and being a lagard or a slower adopter is that you get to see some of the mistakes and right figure out which let other people figure out which vector database is the one that is going to be the best one. And like that's you know that's one of the things we've talked about a lot is you know because we're so early it's like you're damned if you do you're damned if you don't. Like the minute you pick a model that you're going to try and build off of, there's a better one that's faster or cheaper or whatever. And then the minute you you don't, you're being dinged by, you know, your uh your shareholders or whoever that you're not. And so it's it's this weird Yeah. And I I would say a good antidote to that, not a perfect, but a good antidote to to what you were just describing is good design. Yeah. If you have a good core design, uh it should be um not perfectly impervious to these kind of module module changes, but it you you're you're raising your odds there. Uh so when we do our work, no matter what we're we're building, it's a simulation model or whatever, um long before we write even a single piece of code or or bring in a library, whatever, um we spend a lot of time just designing this out on paper. Yeah. Yeah. Making sure that it's a good solid sound design. And then we kind of mentally stress tested. Okay. What if this happened? What if that happened? And um, I think there's a, there's a uh, for, for, for folks out there who are, are trying to build these systems and they're not necessarily technical themselves, you know, there's this kind of notion of, you know, where, when, when am I going to see the code? You know, let's write some code. Um and and I would caution you. Yeah. Yeah. To for the people that you're working with inside the company or outside the company, um spend a lot of time in design, you know, more than you think you should. Yeah, that's a very good point because, you know, like I said, we met as you were consulting with us, and in that process, you know, as the the customer, I'm getting pressure from my boss, like, "Hey, we need...we need to get something, we need to get it going." How much when is it going to happen? How much am I paying? I'm paying this. Where is the where's the output? Blah blah blah. And even as just a person, it's like, man, do we have to go sit through this? And like like I know how I want it to work, but the fact that I know how I want it to work doesn't mean that anyone else has any idea how I actually want it to work. And so like I absolutely agree with you. I like it. It's a process. It's mentally exhausting. It's time consuming, but the more you do up front, the better it will be when you do get it and the faster you'll probably end up getting there. Like I caution people if someone says they can spin out a pseudo complex app or code base in a short amount of time, there's probably something going on there. It's either going to be, you know, the classic developer, oh well, I checked the box that made it do what you wanted it to do. It's the most unintuitive way to possibly do it and your users will never know how to actually do it. But I did make it do the function that you asked me to do here in that in five years from now, nobody will ever know how he did what he did. So Uh, and so, yeah, that's a that's a big one. That, because again, it's management looks as it from, you know, the financial side; it's a cost and cost and time, which the worst things that that they never want to hear. And so, but, you know, if you don't do those things, you run into the the issues down the road. You're halfway through development; it's like, "Oh yeah, we forgot to think about that scenario where Billy couldn't push that button and he because he was on vacation or days off or whatever." It's like, "Well, [ __ ] now we have to refactor the entire thing, and now twice as far behind as you would have been if you would have just done it up front." right? And I think I think that's a good lesson for your audience too in that you know how do you start some of these sort of AI projects? I was at a conference just last week and with a very uh prominent you know technology provider and the first question that they asked at the conference is what could we do with generative AI and I thought oh my god that not only is that the wrong question uh but that's going to lead that you know whoever takes that advice down a path of building a very expensive science project yeah that doesn't actually provide value uh I think the best projects the one that I've seen succeed start with the math problem, the problem you're trying to solve, and then very carefully and thoroughly documents that problem on paper and then decides, okay, here's our problem. what what you know what pieces of technology what methodologies appropriately yeah apply to be a lot simpler than it needs and a lot simpler and you know there there's validity to asking the question of hey what class of problems does this particular technology solve I get that but that's not the way to start a project um so my my advice to your audience is look when you're trying to create value inside companies uh using the the, you know, quivers the arrow in your quiver. Arrows in your quiver. Yeah, if I say that right. Um the the best way to do that is to fer it out these problems which are often hidden in plain sight, but they're hidden. Elevate that in black and white with with diagrams, with documents. Yeah. But especially like like diagrams. Most people are diagram people. And and then work backwards from the business problem. anchor your work to a business problem and you'll create value every single iteration. Yeah. No, that's you cannot be specific enough when you're scoping and going through that process, right? Like I tell everybody that and you don't really understand that until you do it a time or two and you're like, "Oh [ __ ] I forgot I wasn't." Or I I also tell people now especially with tools like one of the things that again like we've talked about this whole time is AI is incredible right now at supplementing human activity in certain areas. And so, for example, I now tell people like if it has anything to do with the front end visually, mock it up. 100%mock it up every time if you have any kind of thing that that you're trying to get a dev to do because it can easily turn into, well, I did it. Well, okay, but this is how I wanted you to do it. Well, you didn't say that. And so, just having a mockup helps. But then you take it this next step. There's this tool that I've talked about before, TL Draw, and it's a basically just like paint. And so you can put a box and then you can put fields and squares and then you put text in it telling it what each thing does and then you click a button and GPT4 does its thing and builds you a interactive little spins up a little web app that you can share with people. So you can send it directly to the developer and he can see exactly how it works and you can just sit there and iterate on it. And it's like with tools like that like there's no reason anymore that we should have these issues of like scope creep or things that well I wanted it that way but you didn't inter you didn't know that I wanted it that way right well humans are always involved humans are terrible right about communication generally speaking but Yeah, and it was interesting too. I mean, just kind of think of you talk about the documentation and like solving the problem, but then it made me think, and it, it all starts to come full circle, I think, as you weave in some of this AI. But I mean, now, say with GitHub Copilot, I mean, like, and one thing I've heard people talk about, and I try to do, but you know I'm not great about it either, but would be like: before you write a line of code, write all your, write all your comments, right? And then fill in the code. a good idea actually but now with with GitHub co-pilot you can literally write your comments and it will write the code for those comments at least you know and you could actually achieve a lot by really outlining what you need to do first and without ever writing a line of code you know in this day and age right you can even you know take that further and write tests. Yeah. On the and then and then it'll work sort of backwards from the test to write the code. I mean, it's crazy what stuff like that will do. That's the only reason I can code as as much as I do these days is because I have gotten good enough at prompting GPT of like, okay, we're going to I need you to build me a Python script for Collab. So, it knows out of the gate exactly how to format it and which uh you know, type or what the format to use. And then I tell it exactly when I if I want to use a library, I'll put the git link in there and say use this library. And it does a pretty damn good job of it because I gave it the boundaries and the constraints of what I wanted it to do. But it's like if you just type in write me a Python code that, you know, scrapes this website, it's going to tell you a bunch of junk, right? Bad prompts give you bad answers. It goes back to the math, right? Data junk in equals junk out, guys. Yeah, it's a function, right? Inputs, outputs. No, that's uh man, we're definitely So, we've talked about this, but we're gonna have episodes in the future of like panel more panel type about a specific topic. So, we'd love to have you back for the the AI stuff. Let's come back. Yeah. Now, it's like we're coming like an hour 15. It's like we could we sit here and talk for three hours on this. Went by quick, but that's a sign that it was a good one, too, though. Um, you want to jump in speed around now, I guess. Yeah, we kind of do a little speed round at the end. kind of a this will be fun because we can make it we can make it kind of AI QA. Yeah. So, we just kind of pepper you with questions and short answers and stuff. Uh I'll keep it as short as I can. Yeah. What's your what's your what do you see as like the kind of leading um vector databases these days? Oh my gosh. You know, Oracle's new version, they just added vector capability. I'm not saying that's the best vector database, but it's telling that they added vector capability to this legacy uh database. I can't really give you an answer there. Uh because on the database side, on the repository side, I don't work in that very much. What about on the coding side? Do you have a favorite library or um model that you've been using? I assume you guys use PyTorch. So, we use a lot of Python language and within the Python language, there is a a uh simulation library called Simpai that works on Python's generator functions. It's a brilliant application of generator. Generator functions weren't created to to do simulation, but these very smart people put this library together and it is it is incredible. And so it's not just for discrete events. So if you're looking at any kind of process flow model, it's great for that. So it's called Senpai. Okay, cool. That's a good one. Um I Who's your uh favorite social media follower? Social media follower. uh follow like who do you follow? Like who do you like to who do you like to keep an eye out for their content or watch? Um I do follow Elon Musk. Yeah, believe it or not. Uh because I I I don't care about his sort of political statements and all that, but he says things on occasion about things he's doing at SpaceX, Tesla, the Boring Company, even Neurolink. Neurolink has has had some interesting news uh as of late. Uh, I like following him because then that directs me to do a little bit of research on stuff he says that I want to I want to understand better. You have any anybody in the AI space that you like to uh that puts out content that you like to read or watch? Um, yes, there's there's a lot of people trying, you know, gather Yeah. remember who they are, gather their their names. Um

um

we can come back to that one. Yeah, it'll it'll pop in right before you walk out the door, right? Like like my brain works. What is is GCP your preferred cloud provider? Yeah, it's just um it's probably not got as many bells and whistles as some of the others, but it's just easy. You can get in there, you can implement, and you're it's the easiest. It was the easiest to me to get just like I needed a quick database that I could connect to PowerBI. I could spin up a BigQuery instance very quickly. It was super well documented. There's lots of, you know, medium blogs and all kinds of articles about how to do all kinds of stuff with it. And I'd say generally reasonably priced as well. So, and also how you can authenticate with it. You know, all three of them like that's the beauty of Azure. You need you need like an Azure account to get into Azure. And then like AWS has this very bizarre IM like but like you get a Gmail account you can email and it knows that you're this human being. I use I use that quite a bit because, uh, I mean even just for like stuff within, right? Like I've said this before: I use appcript for a lot of just like kind of initial testing of code or of different processes and stuff, because it's just basically Java that I can now use Sheets as my database. I can use drive as another storage place that if I want to pull or push things from and uh yeah, it's wonderful because I don't have to authenticate because it's on my drive and on my sheets and you know, it's the click of the button and it it all works together really nicely. And I'd say generally speaking, we're kind of bought into the Google ecosystem. Yeah. And I haven't been unhappy. Yeah. Nothing you've wanted into that. Yeah. Just don't try and spin up a llama model. Don't promise I won't. Um, you got one more. Uh, what's your favorite favorite place to go on vacation? Sound like you've traveled a lot. Oh my gosh. Uh, so I love Cabo St. Lucas. It's kind of my happy place. Five on my bucket list. If you'll go, let me know cuz there's a place Rebecca told me to go to that was just awesome. You can do everything there. You can go uh out in the sort of desert-like conditions and do dune buggies. You can go fishing. You can go swimming. You can go sailing. Uh you can play golf. I mean pretty much anything you can think of to do short of snow skiing. Uh you can do there. I love Cabo St. Lucas. Easy to get to from Houston, too. Yeah. Perfect. Cabo was We had That was one of the last international trips we took ever since our kids started growing up, but it's a while ago, too. Yeah, it was a while ago. Uh but yeah, I agree. I was shocked when I landed in Cabo. Like the mountains and desert were just like right there. Then you're like, "Oh, and there's this beach." And then as you're driving to all the resorts, there's all these golf courses. Like what? Where am I? Yeah, it is. It's awesome though. Fantastic. Uh I think it's Fer Feron. Okay. El Feron. Have you ever heard of that restaurant? It's down like it's on the whatever that is northwest side of the arch. So it's around the tip kind of. There's a really fancy resort right there. But there's a restaurant there and it's literally built into like the coast and so you're like on the water. The water's splashing up onto the patio and so it's just really cool spot. You get to pick your meat and they cook it right there in front of us. My heart kind of belongs to Cabo because that's where we got married a year and a half ago, my wife. So, that's awesome. Fantastic place. Um, what is your favorite language model right now and then we'll wrap it up because I figured that'd be a fun question. Um,

I'll answer that by saying what what is going to be my favorite language model. You know, GPT5 was released in March, I believe, of last year. And I think chat GPT it's the most popular. It's most one-sided, but it is it probably is arguably the best. Yeah. Um now my clients that I work with are leerary of it because they're worried about putting their data on it. So they're looking for and there's a handful of them and I won't be able to like name off the top five, but there's a handful of them that keep your data private. Um, and so they're trained on public data, but but that portion they keep private. Um, so there's a lot of hesitation for my clients to to do industrial work with chat GPT. But for for my own just purposes of just goofing around or if I want to like write some code, I asked Chat GPT to write uh the solution to the quadratic equation. It did it very well. Uh, great for that. So it's probably the one I use the most. All right, last one. Bold prediction time. How far away are we from AGI?

Well, lots of really smart people are working on that and there's a lot of funding going into it. So, given that, I think we're we're less than 10 years away from practical applications of AGI. Yeah. Meaning that somebody will come up with a platform, right, that does AGI like things about 10 years. Yeah. I mean, when you were talking about the example you were using, I saw a very similar video where they put these two little, you know robots, little tiny ones in a little pin basically with a soccer ball and two soccer goals, and then they just showed them videos of soccer games. And like, that's what the training model was was: just this video of kids or whatever playing soccer. And it I don't know how many weeks it took, but sure enough, one of them finally scored a goal, right? And like and then they get they keep learning, right? Oh, scoring a goal is a good thing to do. Let me do more of that. Yeah, it's it's wild that we already have that. Well, I tell you a really interesting um application of of AGI in that regard is uh what Google has done with the Japanese game of go board game. You know, it's it's actually very simple rules, but but the unfolding of the game just like chessate super complicated, right? There's so many different combinations. So the Japanese game of Go is about 5,000 years old. And there are people even today who are go masters. So they start at a very young age and all they do every day all their lives is ma is master the game of go. And I'm I'm oversimplifying here, but generally speaking, the object of the game is sort of or the strategy is sort of capture the middle of the board. Okay? And if you can kind of capture the middle of the board, generally you're in good shape. You're probably going to win. And um so they they built um Alph Go. I think they called it a a synthetic system to play Go against Go Master human go masters. And um it started playing the conventional way and it lost. Played the conventional way and it lost again. Lost again. Well, finally it started playing around the edges of the board and it won. Won against professional go masters. What's significant about that is it sort of invented new knowledge. Um, a game that is 5,000 years old and nobody thought about, hey, well, let's play around the edges. Yeah. So, I think that that teaches us something. That teaches us that there's new knowledge out there that we don't know about. So, when we start to get into the the era of AGI 10 years away, um we're going to create new knowledge, new knowledge about how to how to drill wells and how to run them, how to run fields. We're going to have brand new knowledge that never nobody ever uh thought about. And that's that's kind of what keeps me sort of engaged in this game because it's just going to get more and more interesting and exciting as we go. Yeah. No, I I think that's uh like you said the the pendulum basically backing out Newton's second law, right? Like there is I'm very especially after the pandemic, I became uh very understanding of I feel like how much uh how overconfident the human race is and how much we know about pretty much anything and uh you know like even oil and gas, right? Like we're doing all this from thousands of miles away, and that's an incredible engineering feat. But we don't know what the rock looks like 10 feet outside of that; it's kind of this weird, "we're doing it in spite of a lack of knowledge," essentially, right? And so now we're going to have this opportunity where, you know, the computers are going to be able to see all these patterns that exist that there is literally not possible for a human to see, right? Because it's so much data or it's so connected and intertwined that the human would never make those connections, right? And that's that's awesome, right? It's awesome from the, you know, all the benefits that can come from that medical, like cancers, all kinds of treatment, all kinds of new stuff can come from that. Then it's also terrifying on the flip side, where it's like... weaponized yeah all this stuff is open source so like that's the thing like with AI once you let the genie out like especially as open source it's out the bad guys of course are going to do it right and and I I'm not uh worried too much about that because what I've seen is The answer to bad guys getting AI is more AI, right? Sure. Give an example. There's there the FAA is sort of looking into this, you know, automated airplane. So you you go onto a commercial airplane, there's no pilot being flown from the ground. Um, okay. So you have that uh and naturally people come up with, well, what happens if a bad guy, you know, hacks the airplane and crashes it somewhere? Well, what they've come up with, which I think is is pretty brilliant, is there's there's not one synthetic pilot on board. There's nine. And if one of them goes rogue or gets compromised, it it essentially gets outvoted. Okay. By the other eight, consensus makers. Yeah. Yeah. Exactly. Um that's why I'm a Bitcoin bull. Makes so much sense when you look at any kind of decision like that, right? Generally speaking, that's right. So, so there there are answers to these questions. So some people who should probably know better are saying well let's let's pause on AI because we have these sort of fears there and I I don't think that's that's the right way to go. I think we should um keep going on AI and develop these intelligent countermeasures using AI. That's that's the right way. You know the concerns, right? Like it's not like you're just going in blank. You've got it's just like any other science thing, right? You've got your hypothesis. Let's go test them. let's build it with those things that we know to be risk in mind and then let's make it flexible enough so that as new things pop up we can also fix those right It's uh, yeah, no, it's fascinating. Even I'm mostly worried on the hacking side around like quant and the actual compute power, right? Like you were talking about the 13 factorial, right? Like, that's why all the new passwords are 11, because 11 factorial is a [ __ ] ton of things to try and figure out the perfect combination to, and so it's like. you know that's Uh but now you throw a quant computer at it potentially and it's like even with all the crypto stuff, right? Yeah, that's super secure without quants, but apparently a quant could break that very quickly. Yeah. But then of course, how anyway that's a whole another there's lots of challenges in quant computing. So I'm I'm a little skeptical that it's going to come through as quickly as a lot of people are saying, but it's still a legitimate thing to talk about. You down the road, it's going to be viable probably at some point. At some point. Yeah. As with all tech. Well, George, this was a lot of fun. We really appreciate it. How can uh where do people find you? How can they get in touch with you? Well, uh they can certainly uh read my books that I mentioned earlier. They can come to our website, which is www.business-laboratory.com.

Uh they're welcome to reach out to me, George. ashlaboratory.com. I'm happy to I people write to me all all the time about interesting topics. I'm happy to engage with your audience about any of this. I love talking to people. Most active on like LinkedIn. Are you on Twitter? LinkedIn probably the most. Yeah. Cool. For sure. George, we appreciate it. Check out George's book. I've got a copy actually uh on my bookshelf in my desk or in my office from when it came out. Um appreciate you uh coming. Everybody check out George. Thank you, guys. Thanks for uh for watching. Um just a couple quick updates. We've got Empower coming up or no, we've got Energy Tech night coming up at the end of this month in February. Uh this one is in Midland. If you're out there and going to be in town, make sure you uh check that out, get some tickets. Um we also have Empower coming up at the end of March. That's our energy Bitcoin mining conference. Um it's looking really really good and exciting and uh if you guys are interested, make sure you uh check that out as well. Otherwise, we appreciate it. Thanks for coming again today, George. Thanks, Bobby.