IoT Coffee Talk
Welcome to IoT Coffee Talk where we talk tech and all things related to the Internet of Things. This program was started by a group of friends who just happen to be thought leaders, experienced practitioners and business leaders in the technology and communications services industries and more. This pod will feature weekly recordings of our casual conversations on the tech topics of the day with the hopes of delivering engaging and thought-provoking insights to our audience. We hope you enjoy program.
IoT Coffee Talk
#322 Let's Get Physical
Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.
Usually play that perfectly. Oh well.
SPEAKER_03It's a nervous to really go in front of the coffee talk audience, right? No. The millions of people here getting staged fright. I can feel it. Yeah. Absolutely.
SPEAKER_02Yeah. For all the people who are talking crap about my guitar playing, you do that. You embarrass yourself in front of the entire internet.
SPEAKER_01You're the shredmaster. You're the shredmaster.
SPEAKER_02Yeah, a couple of beers in me, man. I'm I'll I'll friggin' I'll friggin' shred. There you go.
SPEAKER_03Oh my gosh. So welcome to my T Community Talk. What if we didn't drink coffee? I have coffee. I already drank my coffee. Yeah.
SPEAKER_02Oh yeah. Yeah. So before we get started, remember um we're all about communities, elevating communities, and there's a lot of crazy crap going on right now. Like in San Diego, we had like 95 degree weather. So we got womped. Um hey, prayers to the folks in Hill Country and Texas who are getting flooded. You know, you heard of Texas floods?
SPEAKER_03Well, Stevie Ray Vaughn. Yeah, they're real. And I think he told us about that years ago.
SPEAKER_01He gave us a heads up.
SPEAKER_03He gave us a heads up. That's right.
SPEAKER_01Yes.
SPEAKER_02We're still working on the details on uh elevate communities, but we hope that everyone gets involved because you know at some point we have to care about the planet, um, which people seem to not care too much about. Uh, but you know, you'll start to care again because it impacts communities. The Earth impacts communities.
SPEAKER_03Yeah, you know, I think a lot of people, we always have to remind, you know, IoT coffee talk, we've been going on since 2020. It's actually a charity. That's that's what this is. We're having fun. We talk about tech every week, but it it's all about raising money. You know, we did elevator kids during COVID, and now we're doing communities, and yeah, we're gonna we're gonna go after a lot of these things we're seeing around the world and and do really hands-on help.
SPEAKER_02Yeah. And uh Rob has a book that provides the recipes for all of that, so it's really awesome. And remember to take it seriously at your own risk. We highly recommend that you take out insurance if you do.
SPEAKER_03Are you saying they shouldn't take any stock advice from us or anything like that?
SPEAKER_02Yes, especially from Dev. And uh, yeah, we're just gonna have fun and um, you know, so you mean like when I say things like the NASDAQ sell off today?
SPEAKER_01Double down on SpaceX. Now's the time. Now's the time. Buy on the dip, buy on the dip.
SPEAKER_03So we have a special guest today. We've got Devin Young here with us. Woo! Welcome to the show. Welcome to the line, man.
SPEAKER_00No, that thank you for having me. Hopefully, I'll be more a permanent installation in your little coffee talk or big coffee talk.
SPEAKER_03Absolutely.
SPEAKER_00I see some cool models back there.
SPEAKER_03Look at those planes.
unknownYeah.
SPEAKER_00Well, my wife used to be a flight attendant for Cathay Pacific, so I've always been passionate by aircraft, so I've had these there for over a decade. Wow.
SPEAKER_02Really? You know, I used to fly Cathay all the time when I was traveling to uh Hong Kong every other week. Uh yeah, great surviving pandemic. Yeah, and that was like when first class they actually had a first class. Now they like squish you in together, right? Premium classic. Yeah.
SPEAKER_03What? You mean you're not lying flat? What's wrong with you, man?
SPEAKER_02Really, Leonard. What are you doing? Do you do remember first class back in the day? It was like crazy. You you were up on that second level of the uh 7.7, yeah, and you had your own space. And then in the middle, uh on Cathay, they had like a bar, open bar, and they did uh tableside um salad service and meal service. It was that was like real first class, and catha was one of the best. They're good, man.
SPEAKER_01Delta one's not bad, but Delta's pretty good too.
SPEAKER_03Yeah. So Devin, why don't you introduce yourself to the audience and kind of what you do, your background, all that kind of stuff.
SPEAKER_00Yeah, absolutely. No, thanks, thanks you for having me and inviting me. I'm the global lead of IoT for NTT. Uh my background is I have been a consultant for almost 25, 30 years. Um Accenture, PWC back in the day, you know, when it was telemetry and machine to machine, um helped a lot of companies do a lot of their IoT strategies. You know, I I think it's just always been something that you just fell into. And then, you know, everything of IoT deployments in the jungles of Columbia to the basements of hospitals. Um, you know, it's just been a fun ride. So, you know, thank you for having me. And I think I've known Rob for at least a decade or so through you know Chetin's events starting off and just running and sharing things.
SPEAKER_03Yeah, I mean you guys all know Chet and Sharma. Sharma, yeah. Yeah, yeah, he and Devin, man, they're buds. Yeah, I'm gonna do it. I'd still wait for him to invite me to an event, but oh, oh, do you want to go to Mobile Future Forward?
SPEAKER_02Oh, that's a great game, man. That's his show. I don't uh It's a good event, man.
SPEAKER_03No, I'm good.
SPEAKER_02Uh no, I totally get it, but it's his show. Do you know what I'm saying?
SPEAKER_03So it's he always has it up there at the Freddie Couples golf course there in you know in Bellevue. That's uh that's always a great show for sure.
SPEAKER_02Yeah, no, Chetin's awesome.
SPEAKER_03And uh, serious players show that thing.
SPEAKER_00I think anybody beats Chetin more for slides per second.
SPEAKER_03Yeah, you're right. You're right. He's moved. I just remember when 5G came out and he felt the need to dress up like a surgeon and do remote surgery using 5G. I was like, all right, somebody had to do it.
SPEAKER_01Somebody had to do it.
SPEAKER_03Oh my god. Um, so hey man, we got a lot of news going on in the world today. You know, uh some of us have been talking back channel about Kimi 3. Yeah that just dropped on the planet, and it's outperforming all of the models or pretty, you know, or equal or better. Uh the developers like it more. Um and that price per token per million tokens is like three bucks input, 30 output, whatever. Um, does this totally destroy the whole tokenomics and all our financial modeling for all these players in the AI space? Like pretty much. Pretty much. I know that's a that's a I know that's a provocative statement, but it it's real.
SPEAKER_02Yeah, I mean, I'd love to get Devin's take on it because you know, um, you know, I have my my observations, not even a just an opinion, but uh what are you guys seeing at NTT in terms of like these IoT models? You know, how's how's how are LLMs and MOEs and all this generative stuff uh trickling down into what you guys are saying? I mean, um it'd be cool to compare notes. You know what I'm saying?
SPEAKER_00No, I I think you you you're absolutely right. One of the biggest challenges, especially when you're trying to build these models, is the the lack of compute availability. Yeah. You know, for example, let's say Cosmos. Um, you know, just the infrastructure alone needed to work on that, most people don't have access to. But you know, this tokenomics is really impacting. I think you see it in the news at all all the time.
SPEAKER_02Are you are you talking about wait, wait, wait, are you talking about NVIDIA Cosmos, the world model? Sorry, yes, NVIDIA Cosmos, etc. Yeah.
SPEAKER_00I sorry, I'm jumping all over the place.
SPEAKER_02Not everyone is like you, dude, you know, knows everything, okay?
SPEAKER_00You have to assume that right, but I but I think you you know, one of the big there's this big dream of how what we can do, and everyone's excited to do something, and then when you realize how much it's actually costing as far as compute power and things, and that's where I see, you know, I I think the Eastern world is you know, because we in the West have are are fat and happy because we have all this compute power available to us, all the GPUs. Yeah. So we weren't really focused on efficiency, we're just focused on the at least it's my opinion, on creating cool things and getting everyone started. Now everyone's jumping on board, but they're jumping and using AI for things that are just um rudimentary. And so it's sucking up a lot of the compute power. But uh in the East, because they have well supposedly they haven't had the access to you know a plethora of GPU compute, they've had to be efficient. And so now we're starting to get to this disillusionment that there is no ROI in a lot of these models and these use cases that we're building, because it just you know needs so much compute and tokens to do it. And then I think the next phase is you know, if you don't move to a new you know, um model such as photonics, you know, and things to make things more efficient, it has to be in the models. Yeah, yeah. So do you have to reinvent the wheel every time you build a model? Are there things that can be learned? Uh you know, do I need to train, for example, a computer vision model every time to recognize a you know a car versus a truck, or are those libraries just available to where I don't have to you know rebuild the model from scratch? And I think economics and efficiency is gonna be the next phase of where we go. And it sorry, not to dominate, but again, processing edge you know, the edge the edge has become so important. You know, companies back then like Fog Ford and others, of saying, Do I need to capture every sensor reading and store it in the cloud? Um you know, you you you don't. I think you may know. Um someone told me, you know, we don't use 80 to 90 percent of the data that we collect, but the storage of that data in the data centers, the energy used to store all that data exceeds that of the commercial airline industry, keeping with the theme of airplanes. Yeah. Um and if you think about sustainability, Rob, and all that, that is kind of a sin to just have all that waste.
SPEAKER_03Yeah, right. I agree. I know that NTT is in the data center business.
SPEAKER_00We do have quite a you know, we do have that little business. No, it's a huge business. Um, you know, I think we're one of the top five data center providers. You know, I think a lot of people don't realize a lot of things NTT does as far as you know, photonics, data centers, submarine cables, um, things that we really enable behind the scenes.
SPEAKER_02Are you talking about silicon photonics when you say photonics, or is it a product?
SPEAKER_00Uh right, photonics all the way down to the chip. So the ion talk about this in a future. Right, but the old ion and all you know using light for you know, that's really where I think everything's gonna change. You know, first the you know the intersection of quantum whenever we get that uh working and AI is gonna really change a lot of things, yeah. But uh it's not scalable, so you know, photonics is gonna be another area where we we see things becoming more readily affordable for to you know to the masses.
SPEAKER_03Yeah. So P. Pylonics is one of the hype it's gonna be in a couple of years.
SPEAKER_00Right. Well, don't stop believing.
SPEAKER_04You know, we have the biggest don't stop believing.
SPEAKER_01Yeah, right. Yeah, I was gonna say about Kimi 3, though, is uh, you know, I mean Devin's right, you know, after you sort of uh get it working, as we all know, then you try to commercialize it, and that's when cost comes into play. And I think that's where we're hitting in the the larger kind of generative AI cycle, right? Is now the commercialization and the economics need to start to work. And so, you know, people are really looking at their P's and Q's about like what is the token cost? We've seen this huge backlash now where you know, remember we had that kind of 18-month run of like use as many tokens as you can. There was like contests to see how many tokens you can use. Yeah, then at some point someone said, God, that is a dumb fucking idea. Like, and so stop doing that. And um, it's expensive. Shit is expensive, and um efficiency is a new currency, right? So uh efficient models are gonna work. The Kimi 3 though is interesting though, because it's like a 2.8 trillion parameter model. So uh it's not it's not running on it's it's it requires like a lot of heavy metal to run, right? It's not running on edge devices. For edge devices, you need billion parameter models, right? It's about a gigabyte of RAM per uh billion parameters or so. Yeah, so it's not there yet, but you know, you know, my my hypothesis obviously the gravitational pulse toward the edge, and uh eventually it uh that's where the efficiency is. And you use the cloud when you need to, but the cloud's kind of gonna be pretty expensive token costs for probably the foreseeable future.
SPEAKER_02But here's the thing that's really disrup uh gonna be disruptive, and I think this is what's gonna we're gonna see in the next probably six months, and it it's going to cause ripples across the US AI industry, not the Chinese one, because the Chinese uh have already baked a lot of this stuff into their approach. You know, the thing that's really startling is a 2.8 trillion parameter model, and this is what we were kind of like talking about uh you know offline. Sparsity is a dense, densely sparse um MOE or a mixture of experts where you have only 50 billion active parameters. So we're talking about less than 1%. So just for reference, a year ago, a year and a half ago, Llama was the top open um model, right? Um Llama 3. When it came out, it was at mixture of experts 40% sparsity. Then Deep Seat came out, brought that down to 5%, meaning out of all of the parameters for any uh input, only five five percent of those parameters are active. Now we're at below 1%. It's like half a per half a percent. This is what the Chinese have done. And the thing that's incredible is the efficiency and the efficacy of their model architecture, right? The MOE architecture, which is kind of like an agentic framework within a model. Um and the crazy part, I think I don't think there you're gonna be seeing this thing run at scale except for certain hyperscale, like sort of supercomputing applications like drug discovery support. It's not gonna do the drug discovery itself because that requires traditional, you know, high precision um uh supercomputing that you know largely runs off of CPUs. And then people don't know that. Supercomputing, traditional stuff, the stuff that quantum is gonna take over, that's a different category. It's not large language models, you know, running at like FP4. So I think this large model is just going to be a m you know, like the teacher model for a bunch of smaller, distilled models along the lines of what you were talking about just uh just prior, Pete. These smaller models that'll be used for application-specific purposes. Yeah, and they're just they're gonna be great, they'll have 90% of the capabilities of the mother or that mama, big mama model. But you know, most of us don't need that capability. Developers, like what you're saying, Devin, they don't give a crap about the latest. If they built on a model, it's good enough, they don't want that thing to change. And if it's being delivered as a service, they certainly don't want the service provider, the AI as a service guy, cutting them off. You know, that pisses them off more than anything. So, and then you have the economics associated with that. And I I think it's it's just and you know, this whole mythos thing, the the anthropic moat that everyone was talking about, it's gone. There is no moat. There is that was two weeks ago, Leonard.
SPEAKER_03Two weeks ago, where were you? Yeah, crazy, dude. That was June.
SPEAKER_01That was June.
SPEAKER_03Isn't that interesting how anthropic keeps extending the usage of Fable 5 to all their pay customers? They're like, oh, you're only gonna get it, and then we're cut off. And it's like they keep extending it.
SPEAKER_01It's like they might the other thing is too, when you talk about sparsity and research, you know, we work with a lot of academic institutions, and all a lot so much research over the past few years has all been about model optimization and sparsity and you know, doing more with less, right? Because ultimately that's the economic impact, that's the scale. And as you mentioned, outside the US, outside of our Ford F-150s or whatever we have here, yeah, where everyone else is working with you know, less maybe less resources in some cases.
SPEAKER_02Like a fiat 300.
SPEAKER_01Yeah, yeah. That's cool. Yeah, so everyone's trying to be more sovereign with their AI and have more control over it, and they don't want to write a check to AWS, although we love AWS, they're on a board of directors now, so peace. But you know, people don't want to have to write a check back to these US hyperscalers all the time just to run AI. It doesn't make any sense. Yeah, yeah.
SPEAKER_03Interesting.
SPEAKER_01Very exciting and interesting to see. I saw that Apple is now gonna use Quen in China for their Apple intelligence, right? So they're and Quen is probably one of the most popular models out there right now.
SPEAKER_02Well, yeah, that that's the thing. They're they're uh in IoT, in consumer IoT, everyone's using it. You talk to any of the Korean companies that are coming up with like goofy robotics and physical AI stuff, physical AI stuff.
SPEAKER_03What's this physical AI you're talking about? Physical AI?
SPEAKER_02I think Devin needs to tell us. Oh, Devin or Mr. Mr. Qualcomm over here.
SPEAKER_03What's your take, Devin? What's your take on physical AI? Are you guys having to do all that stuff now? Are you guys getting physical?
SPEAKER_00Yeah, I think we've had several conversations I've had with many analysts too. Obviously, it's a packaging, it's the hot word right now. And people are, you know, Jensen said a couple times, and now everyone's jumping on what is physical AI. And people define it differently. You know, the way we look at it is you know, the interaction with the real physical world. But if you think about AI, I think Rob, you and I had this conversation. Most of what you see is just an evolution of other models that have been around. So, for example, all this you know, agentic AI we said used to be called robotic process automation. Now it's you know, robotic process automation on steroids. So this whole physical AI is you know, limited to a few things that we're doing with you know like digital twins and stuff, but it's now like how do the robots and AGVs interact with the physical environment. But some people aren't saying it, you know, it's become this pot new thing, but it really isn't. Yeah.
SPEAKER_01It's been around a long time. I'll just point to um we had uh this event in London, I think I mentioned last time, and one of the keynotes was from Max Versace from he's the VP of Emergent AI from Analog Devices, which is a really fascinating company, by the way. Been around forever out of Boston. Everyone's heard of them. And he did a really cool keynote on what they're working on in terms of um uh touch, uh dexterity and touch and really actually detection. Basically, they're building a neuromorphic AI processor chip into the fingertip of each uh gripper. And that thing can actually detect like, am I touching leather or wood or cement? You know, based on all and and then then the gripper can like he was showing a demo of like how do you want to like spool cable, you know, like have that that motion of spooling cable and things. Like that, being able to use you know AI in each kind of gripper tip to detect the touch and stuff. So for me, that's really physical AI, right? That's really that reality digital boundary. And so you see companies like analog devices getting involved. And they're working with a company called Inoterra out of Delft that does the neuromorphic chip. You've probably heard of them. And but it was a really cool like uh it's more than just like you know, warehouse robots spinning around and stuff like that. It's really like how do you really make these digital things interact with the physical world, which is a very, as you know, messy, uh weird, unpredictable world. So I think I think there's a lot of cool stuff happening in the space. So I would give I would give a you know, I would say when I think of physical AI like that, I think that is a new thing. I think that is some new capabilities that we didn't do five years ago, not even close. Five years ago, we were still like had cameras like we were like, is that a human? I don't know.
SPEAKER_03Yeah, no, but that's that's great because you know we've you know when I spent all that time doing IoT and agriculture, and I saw startups trying to do robots picking apples and things like that, and they kept squishing them or they'd bruise them.
SPEAKER_01Yeah, yeah, yeah. Picking strawberries, right? You can pick strawberry, mango, apple, like any of this agricultural stuff requires the this dexterity and intelligence.
SPEAKER_03The stuff I saw in the past, the robot I'd say flashing light, it's like with computer vision. It would burst this light at the apple, they'd grab it, and yeah, it was like X percentage of them were successfully picked. Right. And then too many of them were squished or bruised.
SPEAKER_01Nobody buys apples with bruises on them.
SPEAKER_03No, you can't sell them. You can't sell them. And you know what those become? Apple sauce.
SPEAKER_02Yeah.
SPEAKER_03Yeah. Yeah.
SPEAKER_02So they're just really good. They have a thing called strawberry sauce. That's jam, right?
SPEAKER_01Jam preserved. That's right. Preserves. Yeah. Yeah, we call it strawberry sauce.
SPEAKER_02Oh, hey guys, I just have to tell you, all these GMO strawberries taste terrible. My wife's been buying this crap from uh Costco and watermelons. Wow, what the hell is happening to food? Anyways, I just wanted to grow your own strawberries. That's my recommendation. Yeah, this is what I want to say. Uh, you know, and you know, kudos to analog devices, the only guys that show up at uh NAMShow now. Um Quacon was there last year and they they kind of bailed out. I think that was a bad move because people were really interested in Snapdragon X Elite. They didn't even know that it existed.
SPEAKER_01Then that then the suggestion box, Mark.
SPEAKER_02Yeah. But the the tactile stuff, it it's it's already there in um, you know, digital instruments like uh interfaces and stuff, right? Whether it's a keyboard, you have like the velocity tracking, and but the the here's here's what makes the human uh uh incredible is the whole tactile thing is bi-directional. You have to be able to sense, right? But then also you have to be able to actuate, and and so it goes both ways. And when that tactile stuff becomes the next hype, the technology is already there. People have these researchers have already been struggling with the math. So this is the problem with the hype, right? It it's classic. People are not aware of a technology, they go overboard, and then you know, they get disappointed, right? And um and that that's that's what's gonna happen there as well. The physical AI stuff, a lot of things that you know, physics models, you know, simulation, um, a lot of this uh stuff that uh has already existed for a long time, like um what ANSIS does, the the thermal, all these things that are used in um you know, like engineering, um they don't they've already existed, you know what I'm saying? And and so this is I think really the problem. People get excited about stuff before really doing their homework and understanding um where the technology has uh been and where it's going, they overshoot where they think it's going. That's usually complete hyperbole. And this is the the thing that I think is kind of comical about physical AI and Devin's point. You ask somebody, oh, what's physical AI, they can't explain it. Then they revert to robotics and it's like, what well, robotics has been around for a long time. Sure. And we go to these shows with these robots doing fancy shit. There's a dude you know behind a corner with like peepholes, yeah, with a remote control doing like Tekken shit to make it look like it's doing fancy stuff.
SPEAKER_01Dancing robots. That's the new red flag. That's a red flag. Anyone has a dancing robot? We've actually banned dancing robots from our events. We're not allowed to have a dancing robot. What about the dogs? You're gonna get kicked out.
SPEAKER_00Oh, that was the dogs that sort of randomly kick this way and but I think a more fundamental question rather than what is what is physical AI is what is intelligence? And is AI the AI that we have actually intelligent, or is it just a calculator, or you know, is it just running a program, a prediction? I mean, what what defines intelligence and are we there yet? Thoughts? Yeah, yeah, definitely not there yet.
SPEAKER_01Well, you know, do you know many intelligent shape or forms? Human or mechanical. What was that, Mark? Yeah. Do you know many intelligent people then? Or not here.
SPEAKER_03Well, but you know but when you're when you're talking about the intelligence, is it really intelligence? Is it just doing pattern matching? Is it just is it am I just doing select star against a vector database? You know, um uh same, whatever, the the guy who had a deep sea, uh not deep sea, deep mind at Google, you know. I love when he remember he came, oh my god, on my phone. As soon as I said that, Jim and I came on. Wow. Um but he said, all right, how are we gonna test it? We're gonna have the when we have the best model that we think we have, and it's been, you know, and it has data that goes up to like 1901. And if it can discover the theory of relativity, if it can figure out how to build an atomic bomb on its own, then yes, we actually have AGI. If it can't do that, if it can't come up with that stuff on its own, then it's just a pattern matching engine, and that's all it is. Yeah, pretty much. And he's running, he's running Deep Mind, you know, and so Yeah.
SPEAKER_01There's a there's a uh interesting I saw well, first of all, I saw an article, I think it was in the New York Times, about how a lot of AI companies are now hiring philosophers, right? So philosophers, I went to school with my one of my roommates, had a philosophy degree. We used to give him a hard time. He was gonna go open a philosophy store when he graduated, but but they're hiring philosophers now to sort of kind of ponder these questions about intelligence. But I don't know if you know this, you've heard of Eliza, right? Remember the program Eliza back in the 70s and 80s, this guy Joseph Weisenbaum had this, he wrote a whole book, had his whole thesis around how humans we over uh we kind of over-rotate on intelligence of mechanical things, like we we have a bias to add to ascribe too much intelligence to mechanical and digital things. Like that's just the way we are as humans, yeah. And uh, which kind of explains our fascination with robots and anthropomorphic technology and stuff. But if you do some research on him, he wrote a book about that, and uh, and it's true. And so Eliza was his program that he wrote that did a very simple pattern matching, it would kind of repeat phrases back to you, and it fooled a lot of people. They thought, oh wow, this is an incredible machine that knows what I'm talking about, and that's kind of what's happening today, too. Like we get these chat bots that are doing mimicking and doing this kind of repetitive stuff, and people go gaga over the intelligence of these things. Um, yeah.
SPEAKER_02Well, you know, the problem is those are all distractions. There's some really cool stuff that can happen, like like for instance, um coming out of sensors converge, uh perception edge, right? The idea of perception, and it's not just a sensor, it's just all how you can architect as like a perception application or or um architecture. There's a lot of new possibilities, and these are like real things. This isn't like it doesn't look that sexy, right? And it's not LLM based, and it's not LLMs down at the sensor level, it's probably a little bit further up. It might be on uh edge uh edge uh infrastructure, you know, appliance or something like that, but it's it's um you know taking uh information off of sensors and then contextualizing them, doing the translation much closer um uh to the premise. And then whatever inside, I mean, this is like the classical stuff that we were talking about like 10 years ago about I, you know, IoT, right? Industrial IoT. Well, you know, I don't think that I industrial IoT vision is going to necessarily happen, but there's a lot of new capabilities that you can deploy on premise, which kind of flips the script on how we were thinking about industrial IoT back in the day, right? And then that everyone's gonna be sending you know the data up to the cloud. No, you we're starting to build a really compelling case for uh compute and quote unquote perception intelligence at the edge. And there can be a lot of benefit, but you know, this question that you're asking, Devin, uh, you know, what is it is there intelligence? That that may not be the relevant question to ask. It's what can we do with whatever intelligent kind of technology to um change the way we do things uh in our industrial environments, our enterprise, right? And those are the questions that are quite frankly haven't been asked because most people can't answer that question. They're distracted by chat bots and agentic nonsense. AGI and all this AGI, which nobody talks about anymore, right?
SPEAKER_01What the Chinese are doing, it's not AGI. That was back in May we were talking about it.
SPEAKER_00What do you think, Kevin? Well, you know, this is this is our struggle every day. So the two things, whenever you have an IoT deployment, one number one is for security. What vulnerabilities are you going to introduce to my environment? But the bigger challenge is this ROI. So especially with AI, I'm replacing someone running around with a clipboard or a master mechanic that can just slap his hand on a pump and just say, hmm, Betsy's gonna have about 10 more months of life on them. And so the key question from this, you know, it's the CFOs who have to write the check is not only what's the cost of implementation, but what's the total cost of ownership? And when I look at the hourly rate of what I'm paying some of these skills, does AI make sense or IoT with AI, physical AI make sense of replacing them? So, you know, that's the hardest thing right now because if you start pulling the bill of material together for any of these things, aside from the sensor, then the platform, then the cloud, and the edge compute, then you throw an AI model on top of it, and then maybe a layer of security, you know, the license costs, the you know, what's it gonna cost to run, and they run it against someone that is you know doing that that actual work. So I'm just gonna keep my hourly employee. Right. We are running against a labor shortage of skilled labor, but a lot of these AI things require a whole new set of support of people that understand robotics or drones and things, and those skills are harder to find than you know than you know, hiring Bob to run you know with cloak or just meter reading.
SPEAKER_01Yeah, though, it's like uh it's like it's um I agree with you, there's a labor shortage in the physical space, right? And there's um and actually, I don't know if you know David Randall from AWS is really articulate on this. But he you know, when you talk about physical AI, sometimes it's not about replacing, but it's extending. So imagine if you had earth moving, uh, you know, someone who's running an earth mover and doing some construction. What if that was you know automated and it was running during the night? Right. So now you've got two shifts going. You've got the robotic shift at night, and nobody's gonna do that at two in the morning, and then you've got human labor doing it during the day. So now you've kind of you know doubled your efficiency, right, for the equipment on site. The equipment is now on site half the time because you're now running it at night. So a lot of times you're gonna see physical AI and AI being able to sort of extend the capabilities of an existing workforce as opposed to replace workers. But yeah, absolutely.
SPEAKER_03Oh, that's good. Hey, you know what? You're learning other news also from China this week. I read a blurb about how the Chinese government's trying to crack down on Chinese people, you know, having relationships and girlfriends and boyfriends with the chatbots because they need to increase the birth rate in China. The birth rate is so low, and they're and so I don't know what they're gonna do or whatever to prevent you from having a new AI girlfriend, um, and therefore not have babies, but it's a it's a thing. It wrote it bubbled up to you know at national level.
SPEAKER_02I I I I think that's like the wrong strategy, man.
unknownI don't know.
SPEAKER_01I don't know. Well, they've had a series of wrong strategies on birth.
SPEAKER_02Yeah, just lean into depopulation, just lean into it. We've talked about it, right? Just come up with an economic that benefits from less less people.
SPEAKER_00But it's a slippery slope, though. So, first relationships, and next, you know, all of us on this coffee talk is going to be replaced by chatbots.
SPEAKER_03Yeah, you're probably right. You're probably right. We'll get ready.
SPEAKER_01You've probably seen this. We got we get a meeting going, and then someone's AI note taker shows up, but they don't show up. Yeah. So I'm like, hey, what's the deal? You can't send your note taker and not show up at the meeting. So the next step is the the uh, like you said, Devin, the AI avatar will show up instead of the human, who'll be like, Yes.
SPEAKER_03You know what that sounds like, Pete? Do you remember that movie Back to School with Ronnie Dangerfield?
SPEAKER_04Yeah.
SPEAKER_03She goes back to college. Do you remember how they showed throughout the semester? And that one auditorium classroom, everybody's there, then a few people with tape recorders, and then by the end of the semester, every seat was a tape recorder recording the professor. Nobody was there.
SPEAKER_00But getting back to what was said about AI being able to extend. Yeah. Also, you know, I heard I read this article out of China that you know they're able to take your loved one and kind of download their personality and voice and looks so that after they pass away, yeah, you could still interact with your loved one. Is that you know is that creepy?
SPEAKER_02It's like it's you remember Twilight Zone, they had an episode like that, uh, sort of like that, where um actually it wasn't like that.
SPEAKER_03You know, it might feel pretty initially pretty close. I I I think it's gonna be pervasive.
SPEAKER_01I think so too. I agree.
SPEAKER_03I think it's gonna be totally pervasive, it'll seem creepy. A good example happening right in the US of A is a whole bunch of people raised money to build a presidential library for Theodore Roosevelt, who didn't have one. Uh I guess they didn't used to have presidential libraries back in the old days. That was more of a recent phenomenon. Yeah, and so they built it in the Badlands in North Dakota, and it's a beautiful place. It doesn't look like any presidential library I've ever seen, but they've got an avatar in the world.
SPEAKER_01Yeah, you can talk to him.
SPEAKER_03And you can and it was on the it was on TV last week uh on a couple of shows I've seen where they're talking to Theodore Roosevelt, and he's actually it's because they're using AI, and they've trained him. You know, it's a pre-trained model, and then they fine-tuned his avatar on everything he ever said, wrote, did some president or you know, remember in Disney they had those uh the hall yeah, Tomaton. Yeah, iconic yeah, that's right. It's a small month.
SPEAKER_01It's definitely gonna happen. I mean, we used to think taking pictures of food was weird, and now everyone does that. So it's inevitable that we will have this will be a service. Every funeral home is now like, hmm. Let's get the intern working on that project to like extension services.
SPEAKER_03We're gonna increase our product lines at an external home, yeah, you know, for an extra 10,000.
SPEAKER_01As a service, you know, loved one as a service, I think it's you know the part.
SPEAKER_02You know what's really weird though? It hasn't taken off. I mean, that was like one of the first use cases for a generative AI, like LLMs, you know, it's gonna take plugging. I know. I mean, and and the question is is like, okay, what kind of societal what what kind of impact will it have on mental health?
SPEAKER_03I think it's great. I'm and I want to announce right now I'm launching a new startup called Tupac.ai. Are you serious? That's we're are we using that revolutionary. All gonna be branded. Well, you Blackman. Holograms are extra. Yeah.
SPEAKER_01Hologram, pre-trained, chatbot is silver level, gold level is hologram.
SPEAKER_03Yeah, that's good point.
SPEAKER_01Loved one as a service. Let's do it.
SPEAKER_03Loved one as a service. Yes. Yeah.
SPEAKER_01Hey, can I do a little PSA? I wanted to mention the um, I'm gonna show my little sticker. I don't know if you can see that. Yeah, ML and Systems Rising Stars. So there's this program uh every year. They've been running it with AMD and NVIDIA and ML Commons, and it's like 38 PhD students that go in a cohort for the year, and they're from around the world, and uh basically helping support their kind of research. So we're now a sponsor of that, and we made these really cool holographic stickers. Wow. But if you go to the ML Commons or look up ML and Systems Rising Stars, incredible students from around the world, uh Cornell, George Institute, uh, you know, ETH in Switzerland. So I'm gonna be down at AMD on the end of July at their have a big event down there as kind of kick off with the cohort for the year. But you know, talking about investing in the future, um, you know, we need to train people who are passionate about this space so that you know we can keep making improvements. So we're pretty psyched to be sponsors now.
SPEAKER_02Are these like forward deployed engineers that we're talking about here?
SPEAKER_01Yeah, right. No, not exactly.
SPEAKER_02Devin and I were formerly uh forward deployed engineers, right, Devin? Right? Come on, dude. Is that like the control? Yeah.
SPEAKER_01FAEs, right?
SPEAKER_02Don't tell me that you didn't get all those miles and not being a good idea.
SPEAKER_01I was an FAE for a few years, I was an FAE for Phoenix Technologies for the BIOS companies.
SPEAKER_03I was, you know, I was I was chatting back and forth with a former Microsoft colleague, and we were talking about how Microsoft is, you know, we're gonna lay off a bunch of people, but then we'll create a whole new company of forward-deployed engineers um to get them using Copilot, maybe or whatever. And we talked about, if you remember, there was a time we had these uh CSAs, cloud solution architects, and at Microsoft, and the whole point of them was even though our Salesforce could get people to sign up for Azure, and of course, you know, because we had that enterprise agreement that most companies use, they've got lots of free Azure hours. But our problem was people weren't consuming the hours, the cloud at all. And so it's like, and they heard that AWS was doing something similar already had, because obviously AWS came before all of us. And so we created that Cloud Solution Architect deal to literally put these guys' bodies in with the customers and kind of like fold a gun to the customer's head and say, consume Azure, consume Azure, and show them. But it but it was the same kind of thing, you know, it was like people with skills, I want you, you're not doing AI enough because that's been the big thing. We're spending all this money on building AI infrastructure, and it's all really cool, and there's a lot of good slop out there and a lot of good research, but enterprises are just not adopting it the way we thought they would or should. And so they think before deployed engineer stuff that that you know was originally created by Alex Carp and all his buddies at Palantir, you know, a long, long time ago. Just embed them in your customers, I guess for free. You can't I don't know if they're charging for them uh hourly rate or whatever. I don't know how that works.
SPEAKER_01This concept's been around since the beginning of time. You know, we used to call them field engineers, and you'd go sit there at Samsung and help them with this product he just sold them and help them deploy it and commercialize it and train them on it so that they bought more of it.
SPEAKER_03That's a good point.
SPEAKER_01Yeah, speaking of nothing new, I mean that's the way we could tech forever, right?
SPEAKER_03You know, I remember you're right, it's not anything new. I remember having the discussion. Let me do a name drop here. No one will know that I'm doing a name drop. So, my good friend, Rod Canyon, who invented Compaq Computer way back when, um, he talked about when after he left Compaq in the early 90s, he he said that was a big Problem he he thought was we built PCs and we built all this stuff, but he thought that people weren't making good enough use of this new computer technology, like not even close. And he I remember having a discussion with him, and it's like we really need to get out there and really show these customers how to use these new computers that we're giving them, laptops. I thought I thought it was interesting because I felt like we were doing just fine. But he thought we were barely scratching the surface of the value we could get from a PC.
unknownYeah.
SPEAKER_01We used to joke that you know you don't want to have to ship an engineer in the box with your product. Right. But I guess now that's the new thing. It's like, oh, this product comes with an engineer in the box. You open the box and the sky pops out in the bottom box. I'm here. Where do I sit?
SPEAKER_02Yeah. Well, you know, um, I mean, that's the thing. Um like we've said before, it it was sold as easy and it's not. Um, and then, you know, going back to your little story about compact, Rob, um, you know, it it's not just about the hardware, it's about the applications. But then you have to have a sense of what the solution is and then what the supporting application would look like. And that's pretty much you know, non-existent because again, it's the distractions. There's things you could do. You have to change your mindset. You have to look at like sort of boring looking stuff that's not that sexy that can actually have huge quote unquote exponential impact. But the thing is, is everyone and it's not gonna have a huge market cap and it's not gonna make you a trillionaire. Yeah, it's gonna be beneficial for organizations, you know, and people's people just uh can't see it.
SPEAKER_03You know, Pete, you just kind of you know, we gave them PCs, we give we give software, we've done this technology, and you said the word self-serve. Word processing and said self-service, and so we assumed that the customers would just figure it out, and they're self-serving and you do this thing. And I'm having a flashback to, I don't know, maybe 10,000 episodes of IoT Coffee Talk ago, uh, where we uh when Mark Post brought on one of his friends from Spain um who had a company, and remember he uh gosh, I forgot what his name is, and they talked about how their company was almost ready to close down their company. It was like they were doing okay, but they were never breaking out another IoT platform company and all that kind of stuff. And I remember he said they did like some kind of offsite with their executives somewhere in Spain, and they came back. Are we gonna shut down the company? And then someone said, Maybe we need to do services, maybe we need to have our people go in in bed with the customer and hold their hand and show them how to do this IoT stuff. Yeah, and none of them really wanted to do it, you know. I I always there's always that tension, you know. I'm a per product company, I don't want to be a consultant, I'm not billable. Here's the product figured out.
SPEAKER_01Well, it's low margin business, you know.
SPEAKER_03Yeah, but it turned out they did they decided they decided we're gonna try the professional services and embed people, and then they had super success, and that changed everything for them. And so it turns out also a lot of people are really busy. If you ever show up at some company and I'm I I'm here to you're gonna do this digital transformation thing, and the guy's like, Well, you know, I'm heads down doing my day job, right? Yeah, and so sometimes you just need to come in and do the thing. Come do the thing. It's like I always talk about these IoT platforms. I don't just want the insight. I'm paying a lot of money for this. I want you to do the thing.
SPEAKER_02Yeah.
SPEAKER_03Automation, whatever.
SPEAKER_02It's like metaverse, industrial metaverse, dude.
SPEAKER_00Well, we you know, we also used to always say, you know, it's a people process technology. Unfortunately, everyone just focuses on technology. But you know, if I swapped out my mom's iPhone for an Android, she just sits there staring at me, like, what did you just do? So it's that user experience that is so important. I get the buy-in of the people that are going to be adopting this technology. I think we all have stories about how you know change is hard, and when you throw something in front of someone, they just look at it like I'm not gonna use it.
SPEAKER_04Uh right.
SPEAKER_00If you change the process, you get the people bought in, that's when you start seeing the ROI of the technology.
SPEAKER_02Yeah, isn't that weird? After all these years, we're still asking the same questions. It's the same thing.
SPEAKER_03These are like PowerPoints from like you're no, you're right. You you you put something different in front of them, and they're like, What is this? You know, why don't we get Steve Sanofsky on the phone and talk about Windows 8?
SPEAKER_01The start button. Where'd the start button go?
SPEAKER_03I don't know what to do. We got the feedback when we deployed Windows 8, and we got feedback from our enterprise customers that we're that every one of them said, we're gonna have to send our employees to five-day training classes to teach them that they have to swipe to find the printer and all this stuff. And and I remember arrogant Microsoft people like, This is you're so stupid, it's so easy, just figure it out. What's wrong with you? Well, you know what? You can be arrogant and go out of business real fast.
SPEAKER_02Well, you know, and and so, you know, speaking about um uh uh solutions and getting to applications that actually matter, I mean, it it just turns out the worst people to get advice from or look for guidance to are the likes of Dario and Sam. They're the worst because they don't know anything about this stuff. Uh they don't. And you know, it might sound like like a shocker, but they don't. That was never part of the fabric of their experience, right? I mean, they are focused on building models or doing startups. There they haven't been in the business of doing like the hard work of actually uh instituting or helping uh uh uh organization go through change, right? That might be driven off of technologies and the you know, sort of the theoretical benefits of applying that that technology there because you know what, and they say it here's the thing people don't listen to these guys, they are the first ones to say, well, you know, we're gonna just push this shit out there and someone's gonna figure it out. Developers, developers, developers, developers, right? They have no clue. So why do we all look toward these guys for the answers? They don't have the answers, and they told you they don't have the answers. That is the weird, weird thing about these guys because people have deified them to the extent where they think that these guys are are know-it-alls, right? They have the answers when they clearly tell you that they don't, whether they intended to or not.
SPEAKER_03Don't you find it interesting? Yeah, there's the people who are in the AI business that are actually doing the thing, and there's what they're saying, and then there's everyone else who is racing forward into the future trying to take advantage, you know, analyst firms, consulting firms, product are jumping on this AI wave, and they're like, We're in an AI super cycle, and this can be the greatest thing ever, and da da da and all this stuff, and so they're full speed ahead. Meanwhile, the people who are actually doing the thing are going, Yeah, we might, you know, it might be an extinction event, or I don't know what's gonna happen. And I'm like, weird. Yeah, I mean it's weird. It's like, who do you listen to? You know, but you know, people are kind of self-serving a lot of times, yeah.
SPEAKER_02Like, don't ask the economist, ask the AI guy who had doesn't even have a economics sub major, you know, uh or minor. It it just doesn't make any sense.
SPEAKER_03It's you know, Dario, Dario's clearly a Doomer, right? All 24-7. He's he's one of the Doomers, and and that's fine. You can be a Doomer. Um, you know, Sam is is he really? Sam is all in. No, Dario, every time he gets a chance, please regulate us. We're gonna end the planet.
SPEAKER_01Did you see the latest model? Did you see the latest anthropic TV ads or ads they had on YouTube? It was like apocalyptic visions of AI, and you know, sort of positioning themselves as like we're we're gonna help keep this from happening. But they're the good guys, it was a horrible, horrible. You look it up on YouTube. Insane. Like whoever's in charge of their marketing, that is disaster. But yeah, so they have this, so he is yeah, they have this whole doom thing, and of course, Sam is probably more like all about abundance, you know, abundance, abundance, and Dario's all about doom and gloom, but uh yeah, somewhere in between. The truth is somewhere in between, right? Yeah, no, that's it. A little bit of doom, a little bit of abundance.
SPEAKER_02Hey man, I think that perception stuff has a lot of potential. I mean, again, it's super boring, but it has huge potential impact if we can get people paying attention to it. You know, but look at sensors converge is not the hottest uh conference in the world. Maybe it should be or edge. Well, you guys are doing crazy stuff.
SPEAKER_01Yeah, we do have the hottest conferences.
SPEAKER_02Right, yeah, you do. I think you're like freaking killing it, bro.
SPEAKER_01Thank you, man. See you in Singapore, Singapore, October 20th. Oh, you bastard.
SPEAKER_02You're gonna be there, Devin? Singapore. Yeah, dude. You're come on, dude. Yeah, Devin, come to Singapore.
SPEAKER_01We have we're doing Singapore and Taipei for the week.
SPEAKER_02End of October. Oh my god.
SPEAKER_03But Taipei, just remember they've got three weeks of power if they get cut off by the Chinese. When we talk about Taipei, we talk about TSMC, there's always that discussion around you know, it always pops in my head, you know, that the Chinese love to do drills about surrounding the island and everything. And um, you know, I'd only heard just recently that they all, you know, uh Taiwan is totally powered by you know liquefied natural gas, and they have three weeks of energy. And so if if it gets cut off, I was listening to I guess I was listening to all in and they were talking to Pat Gelsinger, and they were he's just like he goes, if they get cut off and the fabs get shut down, he goes, it takes 90 days to restart a fab. Yeah, it's not like that, right?
SPEAKER_02The uh that's why TM TSMC upped their investments in the US, I guess. But it looks like, by the way, people are not too crazy about AI ran. Just want to throw that out there. AI ran?
SPEAKER_01AI ran.
SPEAKER_02Oh jeez. I know. No, it's just another example of like hyperbole and ridiculousness that makes no traction. It's like, come on, focus on focus on valuable things, and if you can't do that, then I mean Well, you know what?
SPEAKER_03Why don't we get Chetin on the show and he can tell you why AI ran is a good thing.
SPEAKER_00Yeah, I actually sat down with you know Ono San and Chetin prior to MWC, and we had a long discussion on AI RAN, but I'll keep my mouth shut for my friends at NVIDIA that are trying to push it.
SPEAKER_01NVIDIA, you know, you gotta give them some they need a little help. NVIDIA needs a little help. Yeah, yeah.
SPEAKER_02Yeah. Um, anyways, uh yeah, I'm totally bought into my theory of uh disaggregated AI RAN. So, anyways, uh hey, um, I think uh we all need to get back to work. So you want to take us up? Hey, Devin, thanks for coming in, dude. Oh, thanks, thanks for having me. Really, really great having you.
SPEAKER_03It's awesome. Absolutely, yeah. I really appreciate coming. Anytime. Anytime. Yeah, so thanks everybody for joining us on another fun episode of IoT Coffee Talk. This was a lot more serious this time, I feel like. We've got some serious deep AI conversations. There's a lot going on. There's always a lot going on every week, but there's some serious stuff going on uh that involves economics and tokenomics and token maxing or TJ Maxxing, if you will. It just depends, you know. Uh, but thanks for joining us. Check us out every week. Uh, and and definitely look to uh our our main charity, Elevate Communities. Uh think about donating. And of course, we have some great merch. Look at Leonard. You can get your own cool IoT calling.
SPEAKER_02Kevin, pick it up, dude. Okay, your whole whole family, and you know, you're gonna look cool, man.
SPEAKER_03It looks cool. It looks cool.
SPEAKER_02And people will tap you when you go to conferences and go, oh, IoT coffee.
SPEAKER_03It's one of the coffee talk people. Yeah, exactly. That's yeah, that's who we are. So have a great weekend, and we'll see you on the other side.
SPEAKER_00Hey, Tom.
