Win At Business And Life In An AI World

The Secret World of AI: What Businesses Are Doing Behind Closed Doors (Episode 219)

Brad is the Founder & CEO of Jozu and a project lead for the open source http://Kitops.ml project, a toolset designed to increase the speed and safety of building, testing, and managing AI/ML models in production. 

‌This is Brad’s second startup, his first (Codenvy, the market’s first container-based developer environment) was sold to Red Hat in 2017. 

‌In his 25-year career in the developer tools and DevOps software market, he’s been the GM for Amazon’s API Gateway, and built open- and closed-source products that have been leaders in Gartner Magic Quadrants. 

‌In his free time he enjoys cycling, reading, and vintage cars.

‌What you will learn

  • Learn the importance of protecting intellectual property and sensitive data in AI.
  • Comprehend how AI is less deterministic and more human-like than traditional software.
  • Recognize the challenges companies face in integrating AI while keeping their data secure.
  • Explore the process and collaboration needed between data scientists and software engineers.
  • Examine a real-world example of using AI for inventory management in a global retailer.
  • Identify the problem of model drift and how to address it.
  • Distinguish between consumer AI applications and enterprise AI needs.
  • Appreciate the rapid pace of AI development and its unprecedented nature.

Transcript

Jeff Bullas

00:00:06 – 00:00:38

Hi, everyone and welcome to the Jeff Bullas Show. Today, I have with me Brad Micklea. Now, Brad is in an area where many people don’t tread, which is sort of software, machine learning. Brad is the Founder & CEO of Jozu and a project lead for the open source http://Kitops.ml project, a toolset designed to increase the speed and safety of building, testing, and managing AI/ML models in production.

‌This is Brad’s second startup, his first (Codenvy, the market’s first container-based developer environment) was sold to Red Hat in 2017.

In his 25 year career in the developer tools and DevOps software market, he’s been the GM for Amazon’s API Gateway, and built open- and closed-source products that have been leaders in Gartner Magic Quadrants.

In his free time he enjoys cycling, reading, and vintage cars.

Jeff Bullas

00:01:19 – 00:01:39

You and I are in the same tribe. I love cycling down the French Alps reading and vintage cars. So Brad, welcome to the show and we’re gonna get to the chase soon about what you really do apart from dark, deep science and, you know, create myths about people and companies. Anyway, Brad, welcome to the show mate. 

Brad Micklea 

00:01:40 – 00:01:42

Thank you very much for having me, Jeff. I’m looking forward to it. 

Jeff Bullas

00:01:43 – 00:02:30

So Brad, um I was gonna sum up for everyone really here. So we are all in love with chat GPT Well, most of us are, right. It’s amazing. It writes stories for you. It does creates images for us, it writes blog posts for us and that’s just the tip of the iceberg. But I’m on top of that, and this is something I’ve experienced with A I. It is essentially a chat, chat G BT, you know, the chat box essentially is scraping the intelligence of the planet and revealing it for everyone and then they organize it for you. So it’s pretty exciting times in that we feel like we’re plugged into the brain of the planet and the intelligence and the creativity. So, Brad, um 

Jeff Bullas

00:02:31 – 00:03:09

You provide essentially a tool that helps make sure that the intellectual property and the special things that make companies different from other companies is protected in a walled garden so that you make sure that they can access and use A I but not give it away for free. Because as we were discussing before, um what’s happening is that everyone can have access to chat GP TS information which means that your secrets could be revealed. So, Brad, welcome to the show. I hopefully explained it as best as possible and uh tell us about what you really do after I’ve made that story up. 

Brad Micklea 

00:03:10 – 00:03:55

Yeah. So uh you’re, you’re largely there. Uh The main thing that brought me to start this company is the excitement about A I. Um It is an incredibly powerful technology. I do think it’s going to change the way we interact with each other, the work we do. Um the businesses we build the interactions we have with our customers. Uh But it’s not quite as simple. There’s a more nuanced complexity to it than I think something like the internet or social media or mobile phones. Um Those are much more tools in the classical sense. They did a job if you use them the same way, every time, the result would be the same every time, very deterministic 

Brad Micklea 

00:03:56 – 00:04:34

A I is different. A I is a little less deterministic and in that way, it’s a little bit more human. If I ask you the same question 10 times over 10 years, I’ll probably get 10, slightly different answers. You’re human. Um If I ask a machine the same question over 10 years and it’s, you know, but not an A I, I will get exactly the same answer every time. That’s a pretty fundamental change when you really think about it and how we use and interact with computers. Now, what we’re trying to do is help companies that see this change and say great, I want to benefit from A I, 

Brad Micklea 

00:04:35 – 00:05:12

but I’m not comfortable taking all my sensitive data and all my secrets and all my competitive mojo and sending that up into an open A I out in the public realm and having it learn from me in a way that could benefit my competitors when they go and use open A I personally, that would make me very uncomfortable. It does make me very uncomfortable as a business owner, I wouldn’t do that. And so if you kind of follow that logic forward, you find that there are a number, an increasingly growing, rapidly growing number of enterprises that are recognizing this and saying we need to have our own, 

Brad Micklea 

00:05:12 – 00:05:40

admittedly smaller, um but still very powerful A I models inside our organization, we need to own them, we need to run them so that we know that they’re learning from our business and they’re not helping our competitors. We help those companies to get those A I models from kind of idea stage to in production, working with the business. 

Jeff Bullas

00:05:41 – 00:06:27

That is fascinating. And I think, you know, the pace of A I that, you know, the rise of chat GPT barely 18 months ago, that blew everyone away. Frankly, you know, 100 million users in eight weeks, signed up, fastest growing native application ever. I’ve been in the computer industry since the mid 19 eighties when the P CS started making their presence felt commercially, incredibly exciting times. But then we’ve had the web turn up in the mid 19 nineties, usable web and the internet which supports that. Then we have social media turning up in the early two thousands. I have never seen technology at this velocity before 

Brad Micklea 

00:06:29 – 00:07:14

nor have I nor have I it is unprecedented. And I think that there’s, there’s a bit of a lesson there and maybe we shouldn’t have been surprised because what makes a I so compelling is that human-ish aspect to it? Um I talked to it like I talked to you, I don’t need to learn a programming language. There’s no barrier like that. I don’t need to learn a way to touch the screen on my phone and what apps to click, to do what and how to slide this way and not that way or any of those kinds of user interface things, they’re not there. I just, I ask it a question, it gives me an answer. I tell it to do something, it tries to do it. Um And that really feels like that’s a lot of why I think it’s also why 

Brad Micklea 

00:07:14 – 00:07:30

you get so many people that are so disturbed or scared by a I, because there’s something, there’s almost nothing more scary than something that’s deeply familiar but deeply unfamiliar at the same time. And that, that really is what A I is. 

Jeff Bullas

00:07:30 – 00:08:20

Yes. And it’s, I’m both, I’m, I’m not scared of it, but I certainly think we need to make sure that as companies, as individuals, we try to use it in a way that enhances us and doesn’t take her away from us. And I think, you know, I think we’re heading towards what I call a hybrid world where A I is your assistant, it isn’t. You. Sometimes people will go in and want to write an article and they give it all to chat G BT, write me an article, don’t do editing, don’t do stories, don’t make it interesting and you can read it. You’re going, this has been written by a chat bot, right? Because, and then I think, you know, yeah, 

Jeff Bullas

00:08:21 – 00:09:18

Yuval, you know, Harari Naval, Yuval, Harari, sorry. He said that he’s really worried about the fact that the A I has now got the ability to write stories and the stories of what makes us human because it gets to the core of us. So what I’m intrigued by with what you’re doing is that it sounds complicated. So you how do you, what’s the process for you to provide this walled garden where you use a I within the company to actually amplify the company’s, um, intelligence, I suppose, make it more productive, um, increase the velocity of being able to do new things and test new things quickly. What’s the process behind this? I’m curious about where you start with a company? Right. Do they ask you, uh, a company, you know, basically 

Jeff Bullas

00:09:18 – 00:09:44

tech managers, you know, CTO S, are they the ones that are seeing the problem? Uh Is it the marketing department? Quite often the CEO wouldn’t even know what’s going on frankly in this space because it’s, they’re running a business, not running a technology. So what big problem are you solving that they come to you with? And then what’s the process? So there’s two parts to this and then there’s gonna be subparts. So let’s explore this. 

Brad Micklea 

00:09:45 – 00:10:35

Yeah, it’s a great question. I think in a lot of organizations right now they have software teams already, software engineering teams, they’re building services, maybe they’re building apps and not something that, that most organizations have been doing for 10 years, 15 years, 20 years, depending on how kind of uh old they are well established. And that, like you said, the CTO would run. Um Now comes a IA I requires a radically different skill set and it is much more in the scientific realm today. Um It looks more like experimentation because it is. And so you can’t go to your software developers and say, OK, just build me an A I model because they don’t know how there’s decades of research involved in that huge amount of complexity. So they hire data scientists. 

Brad Micklea 

00:10:35 – 00:11:20

ML engineers, folks who have spent their careers learning this highly specialized skill. Those folks typically because they’re so different in their makeup, in their background, in their, you know, experience, et cetera, they tend to get kind of shoved into a totally separate team. And so they’re in a separate little enclave, they’ve got their own little tools and they build, they build a model. Now, the problem is the model they build. And I, I grew up, my parents are scientists. So I say this with love. Um scientists are built to be experimenters and for them in a lot of cases, a job is kind of done when the experiment is proved out. But if you think about that, that’s really a prototype 

Brad Micklea 

00:11:21 – 00:11:47

That’s like, you know, the rolling test bed that they use for cars, like it’s not the thing you’re gonna sell on the showroom floor. It’s kind of an engineering mockup. Somebody, usually the software team has to take that model and then make a ton of changes to it to figure out how to make it run fast, reliably and safely, for four customers in a production environment. So there’s a kind of collaboration there. 

Brad Micklea 

00:11:48 – 00:12:14

The problem is that these two teams don’t speak the same language, they don’t use the same tools, and the data scientists have never worked. In many cases, businesses dealt with production issues, had to understand all the compliance regulations that go into software at a production scale. And the software engineers who know all that stuff don’t really understand models don’t understand how this A I stuff works. It is nondeterministic, it doesn’t look like what they’ve always dealt with. 

Brad Micklea 

00:12:15 – 00:12:59

So really, Josu is there in order to help, we have, we build tools to kind of help bridge that gap or or create those connections so that when a model is created, it can be packaged versioned, reproduced for people who are not A I experts. So they can just say ah OK, now I can run this, which means now I can integrate with my systems, which means now I can deploy it, which means now I can maintain it in production. So that’s kind of what the process looks like. And the and the fundamental challenge that is slowing people down right now is that gap that chasm that kind of exists between data scientists and really smart prototype models and production demands of, you know, 99% up time and 

Brad Micklea 

00:13:00 – 00:13:03

customers who are going to do wacky things with it, but it may not be expected. 

Jeff Bullas

00:13:04 – 00:13:12

Yes. So you’re basically trying to connect theoretical programmers with what we call business programmers. 

Brad Micklea 

00:13:13 – 00:13:15

Yeah. That’s right. That’s right. 

Jeff Bullas

00:13:16 – 00:14:02

So I’ve still got a big cloud. That’s IT I’m seeing a big cloud, right? Um Can you provide an example of some projects you’re doing or going to do that are quite practical in terms of OK, so we want to use A I, we want to solve this problem. Do we want to make our CRM better? I don’t know what CRM is actually a practical application of software. OK. Custom relationship management software, right? Uh Database software, whatever do you have, is it like because one part of it sounds to me like a big bang fluffy cloud of everything that is going to produce this something. Can we be specific about the sort of things that you work on? 

Brad Micklea 

00:14:02 – 00:14:49

Let’s start, let’s take an example. So um let’s talk about a global retailer. Um They’re one of the key pieces in their business is making sure that they’re holding only as much inventory as they can turn over in, let’s say a quarter a month, whatever the the right period is to do that, they’ve got to know that in different regions around the world, there are different preferences for certain products, for certain sizes, colors, styles, whatever. And so it’s not like you send the entire catalog out to every region in exactly the same way. Global retail sends a very tailored inventory all around the world in order to make sure that inventory management is as efficient as possible. 

Brad Micklea 

00:14:50 – 00:15:36

So this global retailer had built an A I and that A I was trained to understand the patterns of purchase throughout the year in all the different global locations and optimize what inventory was sent where fantastic, huge, huge benefit or a comp a large company like that. Um Excellent. So it’s created the data science team. Thumbs up awesome job guys that get handed over to software software have to make some tweaks to it. Um And then they go and deploy it. Now, the data science team kind of loses sight of that once it kind of hits that software team from their perspective, it’s kind of done. And now they’re moving on to a new problem because they’ve been asked to turn on to a new problem. 

Brad Micklea 

00:15:36 – 00:16:12

But the software team doesn’t really understand the model in depth. And so they deploy it out into production and it starts running now after a year, it’s made a great positive impact, everybody’s happy. But after two years, because models do something called drift where their learning gets kind of skewed a little bit, little bit over time. And so after two years, in fact, it’s not optimal, it slipped way back and is in fact fairly non optimal in the way that it’s telling people to distribute this inventory. 

Brad Micklea 

00:16:13 – 00:16:48

Unfortunately, the software developers don’t know what to do with that. They don’t really have the tools to figure out what’s going on or why or should we pull it out of production? Just not use it anymore? Is it that bad or do we need to tweak it if we need to tweak it, how do we tweak it? We don’t understand it. And the data scientists, like I said, they kind of lost sight of this in the rearview mirror. We solved that problem. We’re on to the next one. So that is a bad outcome to that company because it started off good and then could get worse and worse and worse. What we do, what our tools do help people to, 

Brad Micklea 

00:16:49 – 00:17:38

we basically simplify the way that you would um operationalize that model. So in production, you would better understand, hey, you’ve now crossed a threshold at which this model’s performance is bad enough that you need a completely new model. Or maybe before that, this model is drifting in the wrong direction. It needs a little bit of retraining. It’s a real reminder, um, or some updates to deal with maybe shifts in the way that regions are buying. We’re gonna help them see that we’re gonna help bring the model back. We’re gonna make sure that the versioning is clear. So people don’t lose track of what’s where and get them all reorganized. So that kind of you might see goes great, starts to go down, but now we can help it go back up again instead of goes great and then kind of plummets. 

Jeff Bullas

00:17:38 – 00:17:43

Are we using A I at this stage or we’re just talking about a software project that basically this, 

Brad Micklea 

00:17:43 – 00:17:59

This would be an A I project. Exactly. This is an A I model that makes these kinds of constant smart decisions about. Well, we need a little bit more, you know, blue t-shirts in size L in Canada, um you know, for the next month 

Jeff Bullas

00:17:59 – 00:18:03

and this was using A I before chat GPT Obviously, 

Brad Micklea 

00:18:04 – 00:18:20

This is not a chatbot. Um This is, I guess you would call more of a traditional kind of machine learning algorithm where it’s taking in huge quantities of data in real time and adapting based on patterns that it’s seen over a long history, 

Jeff Bullas

00:18:20 – 00:18:51

right? So the thing we talk about too is uh about stopping that learning using machine learning that’s not giving it out to the world is that part of what you’re doing as well, like it’s not out in the cloud, it’s actually just within the boundaries of the processes. And the data centered computers of the organization is because what we talked about before was he didn’t want to be able to, basically, we want to use A I but not teach the rest of the world too much. So we don’t give away the jewels to the kingdom, the keys to the kingdom. 

Brad Micklea 

00:18:51 – 00:19:34

That’s right. So that, and I’m glad you brought that up because I should have. Um But yes, this is a model that was owned and hosted and managed and operated within this global retailer. Because if they can create a hyper efficient inventory supply chain, that’s gonna benefit them versus their competitors. Um, they’re gonna be lower cost, they’re gonna be more capital efficient than their competitors. That’s great for, for their shareholders. Um, so, absolutely, this is something that they knew. They had the responsibility, sole responsibility for getting this right. Um, because they can’t put it out there and say, hey, Google, you figure it out or hey, open A I, you figure it out. Um, Otherwise it just helps everybody else. 

Jeff Bullas

00:19:36 – 00:20:10

So um we’ve moved. So A I is what we used to call A I in the past, like machine learning has evolved. Can you tell us the big difference between being a programmer and a machine learning engineer? So what has the chat box done for your industry? That machine learning is sort of out in the wild a bit more now? Isn’t it really? Instead of being in, 

Jeff Bullas

00:20:10 – 00:20:42

in, in the dark caverns of machine learning people that are actually just you, you, you put them in a dark room and they come out with a solution about three years later. So what’s, has there been a difference? We did talk about this before? A little bit in terms of what’s the difference between the A I that was being used in machine learning? In other words, let the machine learn about the patterns of stock that we have in different parts of the world. Has anything changed from that type of A I to today’s type of a I, and I’m being very simplistic. 

Brad Micklea 

00:20:43 – 00:21:29

Yeah, known a lot and a lot has, and I think, you know, some of it as with most things, some of those changes are good and some of those changes are less good. Um, I think that if you look more historically or at, you know, kind of more, I hesitate to say it but it’s simplistic, um, slightly simpler or smaller scale machine learning, um, models. One thing that was nice is that they, maybe because they were a little bit more constrained, tended to be a little bit more predictable. Um They would solve a task and do that task very well. Like inventory management. When you think of a, what’s called a large language model on LLM, something like open A I, you know, chat GP T 

Brad Micklea 

00:21:30 – 00:22:18

That’s very different. That is meant to be quite generalized. It is generative A I. It’s trying to create new, not, you know, pattern match and predict, you know, something that is more historically uh data driven. So I think there’s a good and a bad to that. Um It’s brought a lot of light to this area, a lot of attention which is fantastic. I think it’s important for people to learn about this area and, and I think it’s going to be a big part of our lives going forward. And so I think it just, it behooves all of us to get more educated. So I think that’s great on the downside. I think that there’s such a big focus on those large language models and the kind of chat style interaction that has resulted in a kind of 

Brad Micklea 

00:22:19 – 00:22:39

unintended devaluing of some of these other types of A I, these other types of machine learning, these other types of models which are incredibly valuable. Um but look a little bit more scientific or a little bit more mathematical than the chat bots. 

Jeff Bullas

00:22:41 – 00:23:36

So really we have Chat G BT which is like you said, a generalized use of A I. In other words, it’s basically a whole bunch of different things. But then we have what we call the vertical. So we can call chat G BT the horizontal landscape of A I as for want of a better phrase. I mean, people say, well, I want to create an inventory system that allows me to predict, you know, usage of summer in, you know, in Milan, right? For my OK. So this is more niche and vertical, isn’t it? So we’ve got, we’ve got the evolution of the large language models which is general and then a lot of people are going, wow, I can use this to develop vertical applications underneath this, which is fascinating. So um now the other thing that I’m intrigued by um also is 

Jeff Bullas

00:23:37 – 00:24:32

if you look at I wrote an article recently, which is about where’s the money going in A I? And it was fascinating um we’re looking at, you know, basically, you know, big data centers processes are hardware to actually do this and the cost of it is staggering. Then on top of that, you’ve got software which is the Chat G BT S and the vertical apps and so on. So, and then there’s the data that feeds this, you know, so, but what blew me away was the scale of investment going into, you know, the Fangs, the, the Facebook matters. Amazon’s Google’s, you know, Microsoft, one of the reasons that, you know, chat G BT teamed up with Microsoft. It couldn’t, it just needed big, big data centers, lots of processes. And 

Jeff Bullas

00:24:33 – 00:25:17

I did the numbers and we’re talking like half a trillion dollars of data centers and processors to run these monsters globally. So the question out of that for me was watching this and going in the past, we watched Little Scruffy start ups essentially go and win the world. That’s Twitter, that’s Facebook Matter. That was Google in the 19 9, late 19 nineties. I’d be interested in your thoughts as a scientist. Engineer. Are they, is it a lot harder for little guys as startups now to win in this big world of A I chat G BT? I’d be interested in your thoughts. 

Brad Micklea 

00:25:17 – 00:26:20

Yeah, I think, I think it is, it is harder, harder than it was with the internet or the social media or, or even the mobile booms because it does require a massive amount of capital to develop these huge large language models. Um However, I think one of the things that at least the way I look at it is something like an open A I that is essentially a consumer platform. And you always have had consumer solutions and enterprise solutions and business solutions. And they’ve always looked different. And usually the consumer versions have always been larger than the enterprise because the audience is so much larger. And so I don’t see a big difference there. I don’t think it’s very easy for a startup to come at this point and dethrone open A I 

Brad Micklea 

00:26:21 – 00:26:32

in terms of, you know, being the chat bot for everyday kind of consumer interactions. However, I don’t think that 

Brad Micklea 

00:26:34 – 00:27:21

I’m skeptical that any time soon we will get a chat GPT that is so good that it can actually do all the tasks that are required and specialized the way they are inside of a business because even the same task in multiple businesses as you know, is still quite specialized and, and can be quite different and messy. That’s right. All these things. And I think, you know, I know you’ve seen it, I’ve seen it, I think everybody’s seen it. You and, and these models are gonna get better. But even for now, the best models when you ask them to write, even just normal prose, not legal documents, not, you know, medical dissertations, not highly complex compliance regulations. 

Brad Micklea 

00:27:22 – 00:28:05

The results are ok. They’re not a bad starting point but you wouldn’t wanna just use them as is. Um because like you said, you can see that the blogs written by CE PTS are not amazing. The stories written by CG BT are stories they’re not in danger of winning the Pulitzer Prize. Um You know, so I think, I think the reality is that at a consumer level, all to all tolerance for quality is aish at an enterprise level. When somebody’s paying you a lot of money for something, their expectation of quality is extremely high. And I don’t think we’re quite there yet. Will it get there? I mean, probably uh at some point, but I don’t think that’s gonna be a tomorrow thing 

Jeff Bullas

00:28:07 – 00:28:47

in looking at it. Um because I’ve been in tech since the mid 19 eighties, which is a little bit scary but uh 40 years and as I mentioned before, the intro, the intro was, you know, the, the evolution of, of technology and it’s and humans use of it. And um uh I think just watching this is that companies still come to grips with a G using generalized chat G BT type models and platforms. But how can I make it specific? How can I develop something within an organization, let’s say, for example, I want to develop a um a training course, right? Or you know, 

Jeff Bullas

00:28:49 – 00:29:26

How do I do that? And it comes down to cobbling together a bunch of different A I tools, some or out of the box if you don’t want to be writing it from scratch, um, you can use chat GT for creating an overview, you know. But then how do I turn that into a presentation? How do I record it? Do I do a and create an avatar with my voice that can actually present it without me doing recording because we can do that now. Um So maybe the start ups are gonna specialize more in the vertical places to play such as specific applications. Would you think that would be correct? 

Brad Micklea 

00:29:26 – 00:30:07

Absolutely. Like I think that when people talk about A I agents, that’s really what they’re talking about. Um And I think increasingly when you look at what the big companies are doing with A I, that’s how they’re developing some of these solutions, which are very impressive is they’ll be a highly specialized A I agent that only does task A, you know, uh let’s take linkedin for an example. Um It’s only going to be able to scan and process all the different jobs that are posted on A I. Uh then on linkedin, another one is going to look at, it’s gonna be specialized in looking at your linkedin profile and understanding what your skills are ideally suited to. 

Brad Micklea 

00:30:08 – 00:30:50

Now to find out which jobs you’re ideally suited to, requires this agent to do its specialized job. And then talk to that agent about its specialized job and make that connection and you can kind of keep chaining these all the way along. So, yeah, I absolutely see a world in which a startup can innovate and excel at being one or multiple of those agents or within an organization. And I can imagine enterprises kind of picking and choosing saying, oh, that agent is perfect for us, you know, and that one is good, but I think we can find better. Let’s go look at this start up. Maybe they’ve got a better agent for doing that particular job in our industry and our vertical for our, you know, region, et cetera. 

Jeff Bullas

00:30:51 – 00:31:15

Yeah. Yeah, it’s because the pace has been so fast that um I think even I don’t know how you feel, uh you know, I’ve been in tech for a long time. I’ve, we live and breathe it. We write about it, especially for more marketing and, and communication and content, but I feel overwhelmed and it’s, I don’t know how you feel sometimes. Do you feel overwhelmed with the pace of it? 

Brad Micklea 

00:31:15 – 00:31:28

Oh, absolutely. It is very hard to keep up. I think you just to be honest, even I’m kind of at the point where I’m like, OK, there are just gonna be some areas of this that I’m not going to be able to keep up with like it’s just not possible. Um And 

Jeff Bullas

00:31:29 – 00:32:05

that would be about 99% for me frankly. But you might get to, you might get the 10%. Now, I’m gonna be about 1% you know, understanding. So I’ll just make stories up then, you know, and that’s comes down to, don’t ruin, don’t, don’t ruin a good story with the facts. Right. So, it’s, it’s really about that. So, um, yeah, that’s interesting to know. I, I, and I thought that may have been the case that you, you, uh, you might be overwhelmed by parts of it. You’re going to stop. I just need to get off the bus for a little bit. 

Brad Micklea 

00:32:06 – 00:32:39

Yeah, I think to some extent as the hype dies down, I think. Well, well, you know, because right now this is all so new that everybody’s talking about every little thing as the biggest thing, but at some point it’ll stop being quite so new and we’ll have a more reasonable threshold of what constitutes newsworthy. Right. We don’t talk about every little thing that happens in the automotive industry on front page news because we get cars, we get it, it’s cars. Um, and so you don’t need to talk about every little thing and it’s easier to keep up and right now that bar is not quite where it should be, 

Brad Micklea 

00:32:39 – 00:32:50

um, it’s quite low and so anything that clears even just a little bit of the bar is front page news that’s gonna start rising and we’ll get, we’ll get some, some separation of wheat and chaff pretty soon. 

Jeff Bullas

00:32:50 – 00:33:35

I think. I’m cool. I’m gonna get back to cycling rating and vintage cars because, uh, you’re actually, um, I’m not into vintage cars. I’m into cars. I love reading. In fact, half my day spent reading as Charlie Munger said back, Warren Buffett was all Warren Buffett does is Sydney’s arse all day reading. Ok. Not a bad life. But I still feel guilty that I actually spent that amount of time. But the reality is we now live in an information knowledge world and you just, you’re gonna be run over unless you’re doing a lot of reading. That’s my sense anyway. So, but we’ll get back and we’ll get the cycle in a minute. But I have one other area in the A I space which is not on topic for you necessarily, but you live and breathe this A I machine learning space. 

Jeff Bullas

00:33:35 – 00:34:26

We’re watching chat G BT has caught Google with its pants down even though it invented a lot of the actual A I technology. It brought deep mind, you know, Mustafa Suleiman, 800 million back in 2016. We now are watching, you know, Elon Musk, throw his toys out of the cot constantly, which is Elon Musk. Um He’s going, I hate you guys because you haven’t done the right thing with chat GP. T we’re gonna give it to the world and, and, uh, it’s gonna be really bad and dystopian. And guess what, next week he launches and raises $8 billion for his own chat bot. Right. So called Brock. And then we’ve got, you know, Amazon creating it. We’ve got Microsoft creating and it’s so we’ve got this, basically 

Jeff Bullas

00:34:27 – 00:34:38

fight for world domination for generalized chat bots. How do you think that’s gonna play out? Who’s gonna win here? I’d be interested in your take on this. 

Brad Micklea 

00:34:39 – 00:35:05

That is a, uh, that’s an impossible question. That is an impossible question. Um, I mean, it, uh, certainly at the moment, uh, there’s only a handful of companies that you can kind of see having a reasonable chance. I think. Um, I am not a Musk fan, so I’ll be honest, I’m probably overly dismissive of that one. 

Jeff Bullas

00:35:05 – 00:35:12

I used to be a Musk fan but I’m getting more dismayed as, uh, absolute power seems to corrupt, corrupt. Absolutely, frankly. So, I, 

Brad Micklea 

00:35:12 – 00:35:43

I think he’s a very good example of that. Yes, exactly. So, so, yeah, I mean, you know, I think what’s, what’s almost more interesting to me, Jeff is because that fight will happen and I think there, we, we kind of already know the players. Um, what’s more interesting to me is there is a whole raft of open source models, um, that are being created. Those I think are fascinating partly because I’ve got a strong open source meaning and background. Sure. That’s my bias. 

Jeff Bullas

00:35:43 – 00:35:47

So this is the, this is not the red hat was to programming 

Brad Micklea 

00:35:48 – 00:36:18

exactly was to was to operating systems. That’s right. Um You know, Linux kind of overthrew windows and it’s open source. This was a big part of that. My hope is that some of these open source models which are also growing in capability as rapidly as anything else that they will actually be uh a genuine player in this, in this market as we move forward. That’s my big hope because I think you need transparency 

Brad Micklea 

00:36:19 – 00:36:46

in these things. And transparency is not something that open A I or anybody frankly right now, other than some of those open source models is really bringing to the table. And that transparency I think will be increasingly as important as the power of these models grows. The more powerful something is, the more I think we should all want to be able to look under the hood, look inside the brain and see what’s going on. 

Jeff Bullas

00:36:47 – 00:37:38

The other thing I came across the other day, which I think is really interesting, is the challenge with a global chat. But chat G BT type model is that he homogenizes humanity because it and also the people that are creating it. A lot of them at the moment, America is the epicenter of A I. Um And on top of that, then you’ve got an authoritarian state which is China, which has also got the scale and data to make sense of it with its face recognition, data secure and it’s got so much data and that’s the more data you got the smarter it gets also, isn’t it? So, it’s really fascinating to watch what’s happening. And um and some of the likes of NVIDIA, which are encouraging it as well is what we call sovereign A I 

Jeff Bullas

00:37:40 – 00:38:04

where countries may need to say we are going to develop our own A I chatbot may be built on open A I. So we don’t lose our identity rather than being homogenized into America. Um basically large language models that homogenize everything. I thought it was quite fascinating actually looking at that and some countries are starting to lean into that. 

Brad Micklea 

00:38:05 – 00:38:33

Yeah, I find that fascinating as well. And I think it, it is, it is interesting when you think about an A I as being like a person. Um There’s a logic to that, there’s a logic to that. I mean, and I think you, you said it at the beginning, Jeff, what is different about these A is that they are categorizing and curating the information in a way that search didn’t search, didn’t originally curate anything. 

Jeff Bullas

00:38:34 – 00:38:36

It just made it visible. 

Brad Micklea 

00:38:36 – 00:39:26

That’s right, it made it visible and it, and more importantly, it made it visible based on the number of backlinks. So the number the the degree to which the general world population felt like that thing was valuable would raise its, it’s kind of, um, stature within the search. Um, it’s different when you’re talking about something that can be curated now because now that bias is built in and that’s something that should worry us. Not that biases are bad. We all have biases but the idea that they are, again, if they’re not transparent, if you can’t examine them, if you can’t say to yourself. Oh, well, I know this, I’s bias is ABC, then that gets worrying. A hidden bias is scary. An open bias is just something you need to be aware of. 

Jeff Bullas

00:39:26 – 00:40:28

Yeah, the other thing that fascinates me too is that let’s take two countries, China and America, one’s an authoritarian state. Um But I think essentially it is, well, meaning it’s almost like Singapore is a benevolent dictatorship. That’s how it grew up. In fact, if you look at Singapore, it’s amazing, right? That was driven by a benevolent dictatorship, which is quite authoritarian. We’ve got America which is messy and it’s all about independence and um giving everyone a voice. In other words, we have the rise of independence. Whereas in China, the philosophy is different. So this is where training models and bias becomes really fascinating in that China believes that its philosophy, which is, you know, millennia old, is that the state should come first. It’s almost like God. Whereas America says the individual comes first, that is a bias 

Brad Micklea 

00:40:30 – 00:41:00

exactly. Now, at that global level, that’s a transparent bias. And so that doesn’t scare me as much because you can say, well, I know this about the kind of Chinese approach to A I and I know this about the American approach to A I, it’s a little bit more worrying when it’s two things from America and you’re, you just don’t know or anywhere, it doesn’t have to be America, but from any location where you don’t know what kind of biases may have gotten in there. And so in some of the open 

Brad Micklea 

00:41:00 – 00:41:46

mo uh the open source models, they’re releasing what are called the weights along with the model, which tells you a little bit about how those kinds of biases if you will have been built in. And that’s, I think, really the direction that I would like to see more and more people go. And my hope is that as enterprises, as governments start to look at ways to build their own a is that they look towards those as the examples uh or the, or the foundation, let’s call it of what they build more than the foundation of those other options which are very closed off, which are very black box. Trust us, it’s really good. Um Versus here, check it out if you think it’s good. Um So 

Jeff Bullas

00:41:46 – 00:42:25

yeah, I totally agree because you only have to look at what was the tagline for Google when they started is the tagline was do no evil that has, that has since been removed. I’m not sure what it really is today. But, um with so much power comes great responsibility, the ancient philosopher said, I think it might have been Donald Duck or something. I can’t remember. But the reality watching Google and I’ve been watching them because we have very seo optimized content and we’ve played Google’s game and how, how, 

Jeff Bullas

00:42:25 – 00:43:10

how you, you just have to, there’s no other way, you know, this, this is the oracle, the God of search, right? So if you want to be visible, then you gotta play their game. Trouble is though that they’re slowly slowly taking the content themselves without links such as the new, you know A I reviews and basically saying this is our content. Well, everyone knows it isn’t, but then we all start to become invisible because we’re no longer discovered. Um, and visible. We are slow, you know, and we’re getting what’s interesting for me. I’m a little bit of a tangent here, but it, it does have implications for A I is that, um, 

Jeff Bullas

00:43:11 – 00:43:57

the data that feeds the beasts and data feeds the beasts of Google, data feeds the beasts of A I chat G BT and chat bots. Um, the reality is that, um, where the bigger players like news, I think Reddit and so on have struck deals with open A I, you know, 250 million here, I think with news, 60 million for Reddit or something. And that’s fun. Um, except it’s not that there are 100 million bloggers out there that actually are creating content and what’s their negotiating power? It’s so close to Zurich. Maybe we need to have a unionized blogging organization globally with a class action. 

Brad Micklea 

00:43:59 – 00:44:09

I mean, I think the world is going to change a lot. I, you know, you never know. Um Absolutely, the world is in the midst of a huge change. I think that 

Brad Micklea 

00:44:11 – 00:45:10

One thing that has not changed in my opinion though is that ultimately quality always, maybe not always, quality usually wins. Um When it comes to things like this, when it comes to knowledge work, usually quality wins. And so I think the ones most at risk are the ones who benefited from, despite lower quality content, benefited from exceptional seo tweaking or massaging. You know, I’m not super sad if, if that becomes less tenable, I would, you know, as long as the replacement is higher quality content. But again, that comes back to how good a job A I do in curating this massive amount of content because it’s not currently good at creating exceptional content. So, 

Jeff Bullas

00:45:10 – 00:45:18

yeah, and this is where I think the human makes the machine and comes together and creates something quite fantastic. Yeah, 

Brad Micklea 

00:45:18 – 00:45:18

That’s right. 

Jeff Bullas

00:45:19 – 00:45:45

So let’s go back to what you do because we talked about the fact that you’re trying to create um keep IP within an organization when you go in and do a project which uses A I um machine learning. How do you do that? In other words, are you basically using the IP? And they’re just using it to create vertical applications that really work well with A I? Is that what it does? So how do you, how do you create this walled garden? 

Brad Micklea 

00:45:46 – 00:46:25

Yeah. So it really just depends on, on what the, what the customer is trying to accomplish. Typically, these customers will have their own computing environments already. Um for their software assets, they’ll have created services, they’ll have created apps that are digital. So they’ve got that infrastructure there, but that infrastructure is not optimized for A I. You can think of it as um to, to create an overly simplistic uh kind of um analogy. Imagine you have a canning assembly line, it works really well for canning, but now you wanna get into making cars. 

Brad Micklea 

00:46:26 – 00:47:04

Yes, you’ve got an assembly line and yes, a car is made by an assembly line, but that is not the same assembly line. Like you’re gonna need to make a lot of changes for that assembly line to build a car or, or to get an assembly line, building a car when you’ve got expertise, building or, or doing canning. So it’s a little bit like that. These are companies that have that canning assembly line. Now, they’re looking at the A I and they need help to understand what we need to change? How can we reuse how much of our knowledge and processes will just work and where are we going to need to make those changes? So our tool kind of helps 

Brad Micklea 

00:47:05 – 00:47:18

people kind of uh update that assembly line to work with A I. And then we also work with them in some cases to give some amount of guidance on, on kind of what they can do from a process perspective as well. 

Jeff Bullas

00:47:18 – 00:47:35

All right. Cool. That’s great. So, um A I is gonna challenge, it’s gonna actually almost, it’s gonna not blow up, but it’s going to impact almost every corner of business and every corner of humanity, isn’t it? 

Brad Micklea 

00:47:36 – 00:48:18

I think so. Yeah, absolutely. I mean, you think about, uh, I mean, you’re old enough, I’m old enough to remember a time before the internet and it was a radically different time and there really isn’t any part of our lives today that has not been changed by this ubiquitous access to information, ubiquitous, access to each other. Um I think mostly for the better, there are dark corners, everything humans make but, but mostly for the better. And I see a I being the same thing, it is going to be another monumental shift. And it’s interesting to me that my kids who are teenagers now will inevitably be saying to their kids. Well, I remember a time when there was no A I 

Brad Micklea 

00:48:18 – 00:48:36

and that’s gonna blow their kids’ minds. What do you mean? No. A, I like how did you get anything done? Um, and so it’s going to be that kind of massive tectonic shift. It’s not gonna all happen overnight and it’s not all going to be perfect. Um, but I think we’re gonna figure it out. I’m an optimist in that perspective. 

Jeff Bullas

00:48:37 – 00:49:01

Yeah, it’s, um, I’m both in it and trying to stand back from it all at the same time and, uh, humans we struggle with change. So, but the pace of change here I think is actually our biggest challenge. Um And we’ve got a lot of information, but we need a lot more wisdom because information, data is not wisdom. 

Brad Micklea 

00:49:04 – 00:49:35

It’s true. And I think one thing that has always served me well in my career is, and it’s hard to do because it goes against human nature. But whenever I’m presented with a hard problem or an uncertain outcome, my instinct has always been of course to run away or to wait until everything becomes clear and then decide something. But I’ve always forced myself to instead say no, 

Brad Micklea 

00:49:35 – 00:50:12

When things are unclear, when things are uncertain, the best way to get to the correct outcome is to start working through it. Now, start early and iterate fast and you start early, you iterate fast and you get to the good outcome before everybody else before the people who are sitting back and waiting and going. Oh, that looks messy. Gonna wait till it cleans up a little bit, then I’m gonna jump in. And so I think that’s one of the things that’s happening right now is you have companies and leaders who are looking at this happening and it is very tempting at its heart. I mean, it’s going to be different for every company, 

Brad Micklea 

00:50:13 – 00:50:38

but I think some of them are going to look at that and say, you know what, we’re going to jump in, we’re gonna make some mistakes. It’s not going to be perfect, but the things we learn along the way and how far we get will ultimately be a massive competitive advantage in five years, 10 years. And that’s what we’re playing for. That’s how I look at the world. And I think those folks are the ones that are going to be the next set of real winners when this world changes because it will change. 

Jeff Bullas

00:50:38 – 00:50:46

Yeah. So in other words, just instead of sitting back and watching it going, oh, wait until it becomes clear, I go, you need to go from ideation to creation and then lean in, 

Brad Micklea 

00:50:47 – 00:50:51

you need to lean in, you need to start doing the work. Um And that’s where hopefully we can help, 

Jeff Bullas

00:50:52 – 00:51:23

right. So just to wrap it up here on that level and then we’re gonna have a little backchannel chat quickly about cycling if you’ve got time. Um, vintage cars and reading. Um you’re like, yeah, anyway, we’ll go there in a minute offline. But um in terms of uh what do you target customers now? It sounds to me like it’s more enterprise level. Um And do you specialize in certain verticals or do you just provide the tools to do any vertical? 

Brad Micklea 

00:51:24 – 00:51:57

Yeah, they’re not very vertical specific. I think the specificity is that um it kind of tends to be mid to large enterprises. It’s companies where they have an established software environment. Um, they have teams that are working on, you know, deploying production software, managing production software, building production software, testing, production software, they have all those things in place and they’re struggling with, how do we get A I connected into those systems? How do we build it into our digital DNA in a way? 

Jeff Bullas

00:51:57 – 00:51:59

How can you take the new and help the old? 

Brad Micklea 

00:52:00 – 00:52:37

That’s right. Exactly. And so the folks that we tend to find kind of get what we’re doing most are folks in that kind of CTO org folks. Um, you know, with more of a production responsibility, they’re the ones I think that correctly look at AI and correctly understand. Oh, this could be really scary once it hits production. Unless we’ve got our stuff figured out, we don’t have, it figured out this could get very ugly, very fast. The science side, like I said, I think they understand what we’re doing and I think they appreciate it, but it’s not the core of what they’re focused on, they’re focused on building the model and developing the model. 

Jeff Bullas

00:52:37 – 00:53:13

Yeah, thanks Brad. It’s been an absolutely fascinating chat. Um And um, we, I was gonna be on another call with someone else. They didn’t show up and this is like an accidental podcast interview that we were gonna do in about two weeks and you were gonna be somewhere, I was gonna be somewhere else and it was actually gonna get ugly, but we met accidentally and hit record. So it’s actually fantastic. So, um thank you very much and how can people find you um at Jozu? You’ve got yours, so I’m seeing, hearing Jozu and I’m seeing Kops dot Ml. So, yeah, 

Brad Micklea 

00:53:13 – 00:53:32

so I don’t want to overcomplicate things. The easiest thing is just to go to Jo zoo.com. Um Kops is our open source project. Um So for those who are more technical, they can look for that on, on github. Um But uh yeah, otherwise jozy.com, you can find me on linkedin. I’m the only Brad Mickley on linkedin to my knowledge. So it shouldn’t be too hard to find. 

Jeff Bullas

00:53:32 – 00:53:55

Right. That’s cool. So, uh you know, just a little side on that. It’s actually good having a different name. I’ve got a different name. So I got Jeff bullis.com uh back in the day. But um you know, so being called, you know, different, not very nice names in the schoolyard has actually worked out best in the long run. 

Brad Micklea 

00:53:55 – 00:54:00

It’s true. We lose back then we win later in life. That’s right. 

Jeff Bullas

00:54:01 – 00:54:23

Thanks Brad. And I look forward to catching up in real life one day in Canada or somewhere in the world or even in South Africa. You never know. So, gaming, uh, big reserved, big hunt. No, we won’t go hunting. That’s not allowed. That’s good. But thank you, Brad. It’s been an absolute pleasure, buddy. Uh, what do I use in Canada? We use, mate in Australia Americans use, buddy. What do you guys use or just Jeff 

Brad Micklea 

00:54:23 – 00:54:32

Brad? Jeff Brad? I don’t know. We use everything. We don’t run very picky, but it has been a pleasure. Thank you very much for having me on Jeff. I really enjoy it. 

Jeff Bullas

00:54:32 – 00:54:33

Thank you very much, Brad. 

Latest Shows