About a year ago, there was a lot of fear. I did one creative event that I spoke at. They actually had to provide security for me because of some comments people made online. And I would say in the last six months, that fear has really turned into curiosity. What you hear as the meme is, why don't these models do the things that I hate, like washing the dishes, and let me write the poetry and let me make the art? But the reality is, is within every genre of work, there is a subset that is the laundry and the dishes of that role. Welcome to the Expert Intelligence Podcast. Today's guest Mark Heaps is a professional creative, turned AI executive, who's gone from designing for giants like Apple and Google to shaping AI's futures, VP of brand and creative with Kroc. If you're wondering how to bridge the gap between creative thinking and emerging technology, you'll enjoy this conversation. His mission to keep the soul of humanity present in an AI world, couldn't be more relevant in today's business landscape. Mark, welcome to the show. Thanks so much for having me, glad to be here. Before we get started, we're gonna talk about Kroc and your transition from creative to an AI executive, but you had an aha moment with AI. I think everyone has one of those moments where you're like, hey, I'm working on something and it's big and it becomes really personal. You're on stage at Imagine AI Live and you shared a story. I just wanted to start there because it was a personal story that I think is helping you think through navigating all this. Yeah, yeah, it was an experience traveling with my family. We were in Puerto Rico. The short version is we were swimming at a fairly remote beach about a mile away from a road. You actually had to hike through like a mangrove forest to get there. And this was a area of swimming where the beach was very small. You couldn't really get there by boat because of the coral reef. Just very isolated. And my son and I were swimming alone out there about 150 yards offshore and I heard him scream and it turns out that he got stung by a lionfish. We actually had two of the tines in his toes. And so had to swim to shore with him. All the panic sets in. It was a very scary moment. And realize I'm in a place where if I call emergency, they're probably gonna speak to me in Spanish and I don't speak Spanish. And so we got to the shore and he was still in a great deal of pain and screaming. And as a parent, you're freaking out and realize how am I gonna get him through the forest? The Coast Guard can't get to us. The helicopter can't get to us. Pretty risky situation. And at first I called his grandfather who's a retired Coast Guard member and a stress and rescue diver. And he lives in Hawaii. So I'm like, he'll know what to do. And when I called him the first thing, I told him the situation, the first thing he said was, hold on, let me Google it. And I realized in that moment that I was gonna hang up on him. And what do I do in that moment? I don't know if I should take him out of the water. So I actually pulled up an LLM on my phone that Grock had deployed. It was actually Lama at that time. And I asked it, I described the entire situation and I asked it what it should do or what I should do. And it gave me a series of steps that were spot on and got my son a stable situation, got him to where he was comfortable for us to move him. And by that night, he was in a good place. But I think first I wouldn't tell people to trust AI as a medical practice. But what it did for me was the, calmed me as a parent to the state of mind. And that's the power of information and access to information. It made me realize we're gonna be okay, but I've just got to get him comfortable and safe. And it gave me those steps, that guidance, in fractions of a second. And that's a really, really big deal if you think about what AI is gonna do in the world of emergency services, people in the military, educators and more, you have support instantly. It's one of the things that I've noticed, I've stopped Googling things as much because I end up getting lost and not getting the answer. And even if it's as simple as not saving a life from a lionfish sting, but just getting a recipe has become easier when you're able to get that information and very specific. Well, I'll give you a counter one. This was not as intense or serious, but I play in an old man's basketball league. And it's like pickup session. And I went the other night and all the guys started arguing about who should be on whose team. And we have this app that we built called AppGen that people can use for free. And I literally there on the sideline, described into my phone, build me an app that I can tell you the names of players and the position they play and organize that into groups of five. And for any number that is non-divisible by five, say needs to pick up more players. It wrote the app for me in about three seconds, had an interface, I put in people's names and their roles, clicked a mix and immediately I had lineups for teams. Like this is the world, right? Like where we're gonna be able to build the tools we need and have access to what we need in an instant. And my kids would agree with you by the way, they think Googling things is stupid. They absolutely hate it. They think like, why would I search something that gives me more work to do? Right? So tell me a little bit, let's not talk about Grock. Tell me a little bit about, you know, Grock, I know you guys are in a highly competitive space of AI chips and there's been a lot of news, we'll talk about that in a minute. But tell me about Grock and the problem that the company is looking to solve in the AI space. Yeah, so a quick overview. So the company was formed in 2016 by Jonathan Ross, one of our founders. And he was originally a member of Google. He was actually at X over there at Google. And he was one of the creators of the TPU, which is the chip that they used famously to win the AlphaGo competition and handle a number of other AI tasks. And what he realized after leaving Google was there was an opportunity in the market and a problem to solve. One, the problem to solve was related to inference. And so for folks that don't know what that is, inference is when you take something unstructured and put it into an AI model and you hope to get some sort of result out of it, right? Ultimately when you deploy in production. And there is a challenge related to legacy architectures like GPUs where they hit a certain level of ceiling when it comes to inference for low latency. So you might be able to do a lot of AI processing in large batches and that's very, very helpful, right? There's a lot of great research that's done with that methodology, but it's not gonna give you the instant result that you would need, like my kids would want when they're trying to do their homework or when you're talking to an Alexa or one of these other devices. So as the world becomes more of a smart assistant, you need really, really low latency for real-time AI solutions. He recognized that there was an opportunity there to solve that problem. At the time, he was way ahead of the curve, everybody was training models, but not a lot of people were deploying them in production. They weren't using them in the real world. They were sort of still a research experiment in academics and more. And so he basically said, look, how do we solve this problem? Well, he'd already done it at Google with the TPU. So this opened up the opportunity to do a new chip with some different thinking and really execute that in novel ways, like starting with the software first where we built the compiler. That's not normally done in the chip world. What's happened since then though, is that's the first part of the process in solving that problem. The second part is you wanna have access to AI for all, democratizing AI is kind of what you hear everybody saying as a buzz term. And that meant once you had the chip, you have to be able to deploy this, build a lot of infrastructure, create these systems that everyone can access. And today that's normally done via a cloud. So that's kind of where we started in 2016 was, let's build the chip. And what we've been doing in the last year is providing access to everyone via a cloud, which we call GROC cloud for free. And this means now we can unlock the potential of these developers that had ideas about apps they wanted to build, but really couldn't build it because of that latency ceiling that they get from GPU providers. Again, the simplest version of an app that people could realize with this is just conversational AI. If you want it to be fluid in the sense it's a natural rhythmic conversation, but also fluent to the degree of intelligence that it understands what you say and you understand what it says. That's just one of those examples where real-time AI is something that we're solving. What's interesting is there's now some changes in this space that's making this even more valuable. And what changes are you seeing specifically around inference? Because what makes sense to me is, I have this situation where every week these folks come to my house and I need to be able to communicate with them. They speak Spanish, I speak English, and I pull out my Google translate and it is literally the most painful conversation. And it works and it really helps us communicate in a real human way and get things done. But the app and typing it in and the time it takes is just awkward. Yeah, and that's a great example of where latency really hurts, right? Where you want that to actually still be a natural conversation. There were some hackers at a hackathon building on Grock recently in Canada that we were at. And they took your example to another level. They actually had a model using computer vision that was recognizing British sign language that they were doing. And then on the screen in real-time was actually translating and giving them the sentences. And then they could say something back and it would actually generate video which is not something Grock does today, but their application was using layers of different function calls. And it would generate a video of someone signing based on what they had said into the app. And that was really, really interesting as well. The one thing that I would say that we're seeing right now is applications are becoming more capable. And most people may have experienced like chat GPT or a similar LLM like Llama or others where you put in a prompt almost like you're Googling and you ask it something and it gives you some kind of answer back, right? It's not quite prompt engineering but that's a single shot experience that you have with that LLM. Now in the last year and a half, everyone's really said, okay, this is good, but I'm so dependent on the training quality of the model. And also the model can only have knowledge up to the date that it was trained in. So now they start looking at the application saying, well, how do I add layers to this application to make it more useful to me? Can it go scrape the latest news? Can it actually go look at the latest sports scores, the latest stock scores? And so you add all these other components to your application that makes it more useful. But when you introduce all those layers, you're also introducing greater latency. So the more accurate, the more valuable, the more informative and assistant you want that application to be, the slower you're making the app. So with inference and inference speed specifically, the more margin of speed you give yourself, the more complicated you can make an application and not feel that slowing down as an experience, as the end user. So that started early last year where everyone was building RAG demos where you have this retrieval capability of a database where you put your specific information in that database. That means it's got good quality of information available to people to answer questions. But then we started getting into where we're now hearing everyone talk about agentic workflows. And this is where you have a function of calling an agent that has its own, for simplification, I'll just say persona, right? You can say, hey, are you an editor? Are you a doctor? What are you? And then you're gonna do lots of calls to lots of agents. In fact, I was talking to someone yesterday about an app they're building that calls more than a thousand agents. And each has a different role and a different persona. And when you put your request in, they literally debate each other and come out with a probability of what would be most useful to you. Every one of those agents and that call to those agents slows down your app. But you want that quality. So when you have the sort of speed that we're providing and the sort of quality that we get out of all of these open source models, we're finding AIs finally becoming a quality and usefulness that people always hoped for. But we are still at the earliest stage of what that could look like. It's easy to get caught up in the public views of, oh, hey, AI is doing these great things. Then when you actually get your hands dirty and start building these things and trying to deliver them as end user experiences, I still talk to Alexa and Alexa the long way to go. Very long way to go, absolutely. But yeah, so when the rubber hits the road, I think you're right in that. We're in the early innings of a very long ball game. That's right, that's right. Yeah, and people don't understand that these capabilities would not have been unlocked unless people started providing more inference. And so we see that companies like NVIDIA have advanced their architectures to try to solve some of these problems. There's some pretty unique ways that they do that. And they are very good at training in some of these other technologies where Grock and some of these other companies, more of the incumbents, separate is when you talk about building systems that do this at scale. So people talk about buying a single chip. Well, Grock doesn't sell a single chip. So we distribute large distributed systems so that people's applications can serve massive workloads and massive user bases. So no one's gonna buy our language processing unit, the LPU, and stick that in a desktop and have a little research station that they can experiment with. But you can do that with some of the incumbent technology and they're great for that. So this is really where you start seeing the separation of if you wanna put one chip in a little toy robot to answer questions in your house, that's not gonna be us. But if you want that to call to the cloud like an Alexa does and you want it to be a more natural experience and much more intelligent, that would be powered by us. You give a couple of examples of stuff people are building and you guys talk about community a lot. They are the developers that are out there really trying to build useful and helpful things, whether it's an assistant or an application. What are some of the trends that you're seeing from your community? As far as what problems they're trying to solve and what they're building? Yeah, I think there's a number of different applications out there, whether that is a more human-like experience in their applications. So we've seen a real rise in the implementation of conversational AI. Scott Belsky, one of the heads over at Adobe said, we're going into the control era. And one of the examples of the control era is with any application, how do I actually interact with this thing versus a keyboard, a mouse, even on your phone, you're swiping. And the concept for many folks is we want to have a language user interface. We want to have a Louis. And this is where you just talk to the thing and it does the thing. And we're starting to see now engineers who talk to coding applications and it writes the code. And so a lot of the developers are saying, hey, I built an app that does a function, but what I haven't implemented yet is this capability of having an interface that allows me ultimate control and freedom. And one of our engineers the other day said, I went for a walk at lunch and completed my code snippet while I was walking for lunch. He was just describing it, right? So I think- It's interesting because I look at how my daughters interact with technology. They don't type things. They literally, because they have access to GPT and the other models and they don't type in questions. They literally go and talk to it. And you know who else does that? My dad. My dad talks to his phone. And so there's this kind of interesting thing when we talk about how people communicate with technology that it's already happening today. That's right. That's right. Yeah. And I think getting down to some of the brass tacks, there's ideas of creating greater access to any kind of content or media. So one developer that we featured, when we have this program where we celebrate developers, we try to showcase their work more than our own, one developer, they actually built the capability to take a video from like an NBA game. And then in the side where you would see like a chat window for a community, it was actually acting like a narrator or a commentator of the game. And you can give that commentator a personality, but it was giving summaries. Oh, Steph Curry just took this shot. He's been making five shots in a row, blah, blah, blah, blah. You know, that changes how people are going to interact and have access with that media. One of the other things was, can you create, we had another hackathon where you, instead of having an answer engine where you ask a question, it just gives you an answer, what if you had this mentor type agent where on the screen, your kids would get the answer, but then you have this secondary agent that says, you know, seeing as you looked that up, you should probably think about this and you may wanna know about these other three things. Well, how do people search for things that they never know about? They don't. So this is where we see some of the devs are trying to solve problems about bringing greater context and mentorship through the technology to the end user. You know, it's interesting. One of the things that I've started to realize as I used AI conversational technologies, I have to work on how I think. Yes. Now everybody gets caught in, you know, the battles of prompt engineer and all sorts of stuff. If you strip it all back and critical thinking and chain of thought and those other things, those have been around for a long time. Those are not new concepts that you need to learn. I'd imagine and could argue they'd go back 100 years or so. Yeah, you just needed the technology to be able to execute it, right? People have known how to fly rockets to the moon for a hundred years. They just didn't have the technology to do it. And you always read the quotes about, you know, it's not what you know, it's the questions you know you gotta ask, right? And that concept. And so it is a lot easier to be able to talk your thoughts out than type your thoughts out and it removes some friction. But it's interesting because it's one of the things that I've really started to lean into is how do I change my thought process? How do I interact with this technology? I worked at Microsoft and it was interesting, you know, why does Microsoft have a mouse and keyboard business? Right. But they created that business because it was literally the way that they had to teach people to type and click to interact with that technology. And I think now going into this era, it's going to be how you wire your brain to think, which in many ways is a lot harder than clicking a mouse. Yeah, completely. And you know, I tell my kids all the time, what matters most and both of them will tell you the old grandma philosophy. It's not what you say, but how you say it. And it's this same idea. My kids have learned that they have to think about the statement first, how would they frame it? And then make that in a way that gives them the experience they want from the technology. Now, I would say that we are on a fast track that in the next year or two, that will become less specific. In fact, some companies are already doing this, right? Where they've done such a level of prompt engineering that at the front end of the app alone, even before you get into unstructured or structured data, at the front end alone, you have agents that say, huh, they asked this, I think that probably meant X, Y, and Z. I'm going to rewrite what they said so that it makes sense to the actual application and then put that in. This is really powerful because one of the challenges that we've seen over the last few years, with any kind of NLP is what happens if you have a Scottish accent? That's not another language per se, it's an accent. And so we see all the time where people have complaints about this app didn't understand me, it didn't understand the slang that I used. Well, I got teenagers, they use a lot of slang. I'm not asking AI agents about Riz, but they might. And so the thing is, they shouldn't have to worry about their mannerisms or their slang, right? They should just be who they are, and the agent should do the work to get the value out of what they say into the application. That's really one of those fast tracks that many people are trying to solve. I want to pivot a little bit from the actual technology in AI. You do something that I think is brave. You're an AI executive, works at GROC, and yet you have your creative chops, play music and you're creative, but you walk into Adobe Max and get on a stage in front of a bunch of creatives, representing AI. And to a lot of creatives, it's unsettling, it's disruptive, it's change, and it will impact jobs. There'll be less people needed, and those will transform, and there's a belief that, hey, in the future, it'll be abundant. But in the immediate, when you walk into that room, you represent something that is relatively disruptive to their day-to-day. So what are you hearing when you walk into that room and get on the stage and talk about AI, specifically to the creative community, who I feel in a lot of ways, it feels more acute to them? Yeah, I think we've been in this radical shift in the last 18 months in regards to this. 18 months ago, I would walk into a room and talk about AI and people would laugh. They'd go, oh, you mean the app that makes everyone's fingers look like hot dogs? Ha ha ha, you'll never replace us, right? That will never replace me. And then you got to about a year ago, maybe a little less than a year ago, and there was a lot of fear. I did one creative event that I spoke at that was filled with designers and photographers and similar, and they actually had to provide security for me because of some comments people made online. That was that level of fear. And that didn't bother me. I respected the fear. I was like, okay, here's some people that don't get to see what I get to see on a regular basis, and access usually helps. And I would say in the last six months, that fear has really turned into curiosity. The thing I love about the creative community that I've been a part of my entire life is they're ultimately all problem solvers. That's what creative thinking really is. Expression, yes, but problem solvers. And so suddenly problems that have always been challenging to solve now have a new tool and methodology for solving. And that has inspired a lot of people. I get more people today from the design community, the artist community, messaging me saying, hey, hey, wait, wait, wait, can I do this? Because this would literally let me triple what I produce in a week. That will make me more valuable. And so, I'm seeing a shift in what they're concerned with. What you hear as the meme is, why don't these models do the things that I hate, like washing the dishes and folding the laundry, and let me write the poetry and let me make the art? But the reality is, is within every genre of work, there is a subset that is the laundry and the dishes of that role. When I first started my career as a creative, I was in 1920, I got paid weekly to cut photos out. This was before there were transparencies and all kinds of things in graphic today. You say remove background in an AI app and it removes it. But I literally would cut out 30 images in a week and pay my rent. I mean, I even remember going to put them on the paper and Xeroxing them. Yeah, yeah, yeah, absolutely. That was, if you wanted to make a flyer, that was gonna be put on somebody's car to go buy something. It was tape, it was scissors. Yeah, you had to do mock-ups, you had to do physical paystuffs, right? And I did those things too. And so I think people are getting excited about what it's capable of. They're realizing now it can do things that surpass their individual abilities. But look, I spoke at a photography event last weekend and the one thing I reassured people is the technology isn't blocking you from doing something you love. And that's an important detail. If you have a craft that you love, you like painting, you like drawing, you like taking photos, whatever it is, writing music, the technology isn't stopping you from doing that. It does shift maybe how you operate as a business, but this is where you separate artisan from business. Now, if you don't wanna cut out photos because that's the dirty laundry part of the job, then don't use the AI to do that and do the part that you like. And what I think we're gonna see in the next few years is where schools previously had a huge part of their curriculum to teach you the craft skills, the technical skills of designing. We're not gonna have that. We're gonna have every person is gonna graduate as if they are a producer or an art director. Your value now is gonna be your creativity, your opinion and your awareness of trend and style and the thinking part of it. Now, what's interesting to me is you said something earlier about there will be less of these jobs. I remember looking into the data at one point when automation became a big factor in the automotive industry. What's interesting is once upon a time, the automotive industry was completely filled with craft artisans. You were able to shape metal. You were able to do these custom paint work. You did stitching and there are some companies like Bugatti and Maserati and Ferrari and these others. They still do that. They do it by hand, Rolls Royce. Their lab and studio is amazing, right? It's all humans. But for the most part, those folks skills in the automotive industry went away through automation and technology. However, more people today work in automotive using technology than we ever had in history. So what happens is the humans have moved to a different place in their role and they've let the labor part be taken on by the technology. And that's really what we're trying to get to. Now, the challenges as a society, I think, this is my personal opinion, not Grox, there's a lot of people that don't wanna elevate their activity in a workforce. They're happy working on the warehouse floor and pushing the cart around, but now they're replaced by a robot. So I think the challenge is how do we figure out helping those people? And one of Grox core beliefs is everyone should have access to AI so they can learn it, so they can use it. So this is why we provide access for free. We don't charge for people to use chat. We don't charge for developers to build their apps. They can have free access and you can learn and you can experiment and you can remove that barrier of entry to play with AI. And this is why we've got 850,000 registered users in our system, because people are trying it. So that's how you help is you give them access to the education, the tools, the resources and the means. And you hope that as they elevate up, all ships rise in the tide and that supports the business. We talked about people in careers, but we both have kids. Yeah, sure. And for them, part of being a parent is helping guide them to where they need to go, right? Giving them advice, like, hey, what should I study in college, dad? Or, hey, what, you know, how do you think, because in many ways this changes or impacts the advice my father gave me. My father gave me different advice. His son, go work for a good company, be a good employee, work for people that are ethical, that's a good manager, and do good work and work hard. And I think that advice, while good at the time, is not the same advice that I'd give my daughters. From your perspective, based on what you're seeing, what advice are you giving your kids? Well, the first thing, and I get this call all the time from friends, families, because they're concerned about, hey, I don't know AI well enough to be able to guide my kids. So I don't know what their risks and opportunities are. And the one thing that I say is there's a lot of roles out there that humans do best when supporting humans. You know, my daughter, for example, is doing a role shadowing next week of a physical therapy office, because we're probably a pretty long way off before AI can do anything related to physical therapy. You know, plumbing, AI is a great tool if you wanna go in and say, how do I fix this type of pipe? But it's not gonna fix the pipe. You know, the fidelity, the electricity, the resources to get technology to do that, pretty massive. So I start looking at where do humans have the most capability and the most value. Our greatest value is critical thinking. And AI isn't inspired. AI doesn't walk down a street, hear the sounds and suddenly say, I'm gonna write a song, right? It also doesn't walk down the street and see someone have a problem, you know, maybe with a cane or an object and go, I can invent a solution for that. It has no motivation to do that. It's not incentivized. It doesn't need money. It doesn't need to be fed. It has no incentive to do it. So I think, you know, the core thing that I talk to my kids about is, how do you practice critical thinking? Looking at the world in a way where you can look for problems to be solved. How do you learn to develop an opinion? How do you learn to ideate? Ultimately, how do you practice being inspired? You know, one of my favorite things is getting in a room with other humans and someone says, okay, we don't know how to solve this. And we jump on the whiteboard and we pull out the papers and we ideate, we ideate, and we ideate. But I'll say, do we exclude AI from that room? Not at all. We actually will talk to AI, to LLMs and say, this is our problem, what do you think? But it can only give us ideas based upon what it's been trained on, which is human ideas. Right. So that's where our value is, is humanity is gonna have to level up to be able to really have a secure future. But there's a famous quote that's gone around the internet. I forget who said it. And I'm gonna butcher this, so I apologize to everyone listening, if they know the source. But in essence, it was, AI won't replace humans, humans who know how to use AI will. And it really should be a tool of human agency. I often joke that, you know, we're moving into an ubiquity of AI. And it's kind of like Luke Skywalker in Star Wars having R2D2 and C3PO. Those are two droids, which we could replace with as agents. One of them was a language droid, C3PO could speak every language of the galaxy. But R2D2 couldn't, he beeped. He was a logistics operations kind of engineering droid. And that helped him with the very technical problems. That's the world we're moving into. You as a human should still be the hero, but you need these technologies to help you accomplish massive, powerful, great things. Mark, thanks a ton. Great insights. I enjoy your content, your stage talks and your articles, because it comes to technology from a very human centric, I think you call it human plus perspective, which I think a lot of people miss because if you look at social media, it's the latest LLM or today it's Deepseeker or whatever it is. And we forget that it is a human innovation tool. And a lot of ways we need to figure out, or we're all trying to figure out how best to interact with it, how to get the value from it and how to stay sane along the journey. So I appreciate your time today. If somebody wants to get in touch with you or follow your content, what's the best way to do that? Yeah, I'm on the same handle on all social media. It's at life by pixels. I'm on X and a number of other platforms, but they can always find me there. They can find me on GROC channels as well. With that in the show notes. Mark, thanks a ton for your time and to everyone out there. Stay curious. Take care. Keep that in mind, that is, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really, really,
