A.I. For Audio Visual Systems

[00:00:00] Speaker A: Foreign. Welcome to this broadcast to post episode of AI for Audio Visual Systems. I'm still your host, Michael Kammes. Now we're hosting this episode because Key Code Media works very closely with corporate education, government and enterprise teams as a systems integrator, helping them to design smarter, more intuitive and more reliable AV environments across conference rooms, classrooms, training centers, theaters and other collaborative spaces. We're trusted to support organizations like UC Davis, Google, Oklahoma State University, Amazon, and a ton more. Now we're kicking things off with a short explainer on the top AI solutions for AV systems, looking at some of the practical ways that AI is already improving cameras, audio, meeting, capture, room intelligence, support, and much more. Then we'll move into our keynote interview with Christopher Jaynes, CTO of Q Sys, to talk about where AI fits into the future of AV system design, deployment and user experience. Let's get to it. AI in AV is moving from concept to practical deployment. In classrooms, conference rooms, training centers, lecture halls and theaters, we're already seeing AI and automation take on real jobs like auto tracking cameras, intelligent source switching, speaker detection, transcription monitoring, and of course, room analytics. So for this segment, we put together our top seven AI solutions for AV systems based on what's available today and and where the technology is headed next. Number one Auto tracking cameras and speaker framing One of the most visible ways AI is improving AV is making cameras follow the conversation instead of forcing someone to babysit the shot in conference rooms, classrooms and lecture spaces. That means the active speaker gets framed automatically and the far end viewer gets a better experience and the room feels more polished without adding constant operator attention. Examples of this approach include systems like Q Sys Vision Suite and the Crestron one. Beyond Solution number two Smarter switching with less manual intervention. A lot of rooms don't need a human making every camera or source decision. AI assisted workflows can choose shots or trigger actions based on who's speaking, how the room is being used, and what kind of session's happening. That creates a much more reliable experience for hybrid meetings, traffic training sessions in classrooms while reducing the chance that a live production gets slowed down by manual switching. Number three Audio Aware rooms that react to people, not just presets. Another huge shift is moving from rigid room zones to systems that understand when [00:03:02] Speaker B: someone is actually speaking. [00:03:04] Speaker A: That matters. In spaces where people move around, shift between presenter and audience, or change roles during a session, the benefit is pretty simple. The room becomes easier to use because it responds to the discussion instead of forcing the discussion to fit into a fixed layout. Q Sys Vision suite, speaker Spotlight is built around that kind of voice aware behavior number 4 AI for clearer audio Another major area where AI is helping AV is in Audio systems today can reduce background noise, cleanup, reverb and balance voices automatically so remote participants hear clear speech and in room users don't have to fight the room. In some spaces, that means better intelligibility, fewer complaints and a much smoother experience for transcription and hybrid collaboration. Examples like sure, IntelliMix, but also Biamp's AI noise reduction systems and Crestron's DM NACS intelligent audio are awesome choices. Number five monitoring and analytics that help teams make better decisions. AI is also showing up in the backend where it helps teams see what's happening across their rooms as well as devices. Instead of waiting for users to complain, operators can monitor room health, receive alerts, and look at usage trends to understand which spaces are working and which ones still need attention. That means fewer surprises, less downtime and better planning. Examples include Q Sys, Reflect and Crestron Xio Cloud. Number six Transcription, Records and Review Another growing benefit of AI in AV is making meetings and classes more useful after the sessions end. Transcription and content capture can improve accessibility, make sessions searchable, and give management or instructors a way to review what happened later. For training centers and lecture halls, this can turn a one time session into a reusable resource. This is part of the broader AI powered collaboration direction being discussed across the industry. A good example of this is Shure's IntelliMix room kit. Number seven scheduling, room status and Automation all right, the next one is not technically AI, but it sure does make your life a heck of a lot easier. Room scheduling, wayfinding status indicators and automated control can reduce friction before meetings even start. The value is that people can find a room faster using the technology with less confusion. Extron's room scheduling ecosystem is a great example of how these workflow improvements are being packaged today. That's our list for now. So how did we do? Is there anything that you would add? What AI solutions for AV are you most excited about? Let us know in the comments. And if you are planning a classroom, conference room, lecture hall or theater and want help figuring out where AI actually makes sense, Keycode Media can help design the right mix of cameras, control, monitoring and automation for your space. In case you haven't heard, AI is quickly becoming part of the AV conversation. But for most customers, the real question isn't whether AI AV sounds impressive, it's what AI actually does inside the classrooms, conference rooms, training spaces and business facilities. In many cases, that means Making rooms easier to use, improving how people are seen and heard remotely, and reducing the friction that keeps AV teams tied up in manual tasks. We're joined today by Chris Jaynes, CTO of Q Sys, to unpack what AI really means for AV systems today, where it's already being used, and how Vision Suite fits into the broader shift, so towards smarter, more automated, collaborative spaces. Chris, thanks so much for being with us today. And let's dive right in because we've [00:07:02] Speaker B: got a lot of ground to cover. [00:07:04] Speaker A: So when people in AV say AI, what are the main categories we should actually be talking about? [00:07:10] Speaker C: Okay, that's an awesome question. You know, it's funny because there's a big range of what people actually mean, and I've seen everything from what I'd call an algorithm running on a cable, output of a cable, all the way to an LLM running in the cloud. So there's a. A variety of things. What I don't mean when I talk about AI is automation. So automation is really the mapping of a sense state to an outcome, right? Like, hey, a thermostat is automatically changing the temperature of the room based on sensed parameters in that space. It's too hot. Change the thermostat. That is clearly not AI. When you get into the realm of pattern recognition insights, predictive analytics, you start to step into the realm of AI and, and the world of AI obviously has changed over the last four or five years, and AV is now poised to really take advantage of it in ways that I just find super exciting and really compelling, honestly. [00:08:04] Speaker A: Well, let's pull on that thread a little bit. What is the most practical use case [00:08:08] Speaker B: for AI right now for av? [00:08:11] Speaker A: So what AI driven capabilities are customers [00:08:13] Speaker B: really actually using today in classrooms, conference rooms and training spaces? [00:08:19] Speaker A: But also, what are they not asking [00:08:21] Speaker B: for that you've kind of found that they actually do want? [00:08:24] Speaker C: Okay, that's awesome. Well, av, you know, audio, video. So I'll give you an example of both. In the audio world, you may not know it, but AI has already arrived in things like noise cancellation, pattern recognition, like I mentioned, for things like auto transcription. So when you're seeing Copilot transcribing a meeting, that is AI doing that work. So that's already arrived in the AV space in a couple of ways. You can have microphones that can detect and remove noise in the space build right into the microphone on the algorithm that's running it. AI actually, in our case, we actually have an algorithm to detect direction of arrival of audio on microphones based on AI. So that sounds like an algorithm, not an AI like I, you know, I gave you that spectrum earlier. But if you, if you cast the problem as a, I don't know, I can't estimate direction of arrival, but I hear lots of audio and I can train an AI to learn how to do that and then put it on that micro. You basically have leapfrogged years of hardware development using software. So that's a, that's a great example of AI today that's quite practical and already having impact. And I want to emphasize that change because I talked to my own engineering teams and we have a group called the Applied Research Team that's stock full of AI researchers and things. And I actually talk to them about the end of the algorithmic era, because in the old days, you know, I, I did my research in AI. And so the old days you would be building algorithms by hand, right? Coding things or thinking about how you want to design this particular thing. That, that era is somewhat ended for certain categories of problems. Because if you have a pattern and you have an outcome you want, you can train an AI to map to that and everything in between inside of that big LLM. All those millions of parameters you're training is the algorithm itself. You just didn't have to go to the trouble of actually designing it yourself. You let the AI training process do it. So that's a big change. And what it means for us practically is it'll accelerate development in a huge way. On the video side, I'll give you an example. AI that does things like camera preset recall. That's a common problem in the higher ed and in corporate, where I've got a camera that wants to see a room and I want to aim at this speaker when they talk, and then I want to aim at this speaker when they talk, right? So I can auto crop it. In the old days it would be like, well, the integrator would mount your camera, they might make some measurements in the room, they would write code to program, you know, have auto AV control and maybe pan tilt the camera when somebody's sitting in that seat. And then have a even worse case in the old days, a button on the control panel to change the camera to aim somewhere else so that users had to be involved. And then AI arrived and you could do face tracking in real time. You can understand where people are moving around in the space. And now you've got a CPR auto camera preset recall happening automatically because of AI has arrived. Now where that goes, what people aren't Asking for, but they are now, because we sort of launched it as a product, is the ability to do that with lots and lots of cameras at once. Because if you can automate one camera, couldn't you have 10, 12 cameras in a space that track users as they stand up even, or walk to a whiteboard or sit back down? Can that AI understand that? I approached a whiteboard and auto zoom in on the whiteboard so that remote learners can see it, or if I approach a podium, zoom out because we're kind of going into lecture mode. Those are the things that AI is taking us into in a really fast way. So awesome. It's an awesome time to be part of av. Honestly. [00:11:54] Speaker A: That brings up another good point because [00:11:56] Speaker B: from a marketing perspective, we've been fed that we kind of need this orchestration layer that is controlling both computer vision, any audio processing that's going on. But also you find that an orchestrator [00:12:07] Speaker A: to do all this has to be fairly large. [00:12:09] Speaker B: And sometimes, hey, you need to put that in the cloud. But then we have the, well, why don't we have agentic AI on prem, something on camera, something on microphone that can be run with lesser horsepower? And with the supply chain constraints right now there's a big discussion on do we want to pay for something big and on prem, or are we just going to go with many little models and then being controlled by an orchestrator? [00:12:30] Speaker A: So what's kind of your thought process [00:12:32] Speaker B: on how to approach that situation? [00:12:34] Speaker C: Oh, that's an insightful question. You know, and it's, it's. I don't know if there's a complete answer to it yet. I think the debate is still ongoing, but you do have to have flexible systems that can support both. So our approach has always been to if you can do compute at the edge, absolutely do it. If you can do compute even farther from the edge, at the peripheral, do it. So that's already happening on things like, you know, video conferencing, bars that can auto frame and do onboard processing or color correction algorithmically or even with AI. As you move up the stack and you get to that edge, compute, you have more flexibility because obviously you can get some inference compute there. It's quite expensive. But inference per dollar is not as expensive when you get to the large LLMs. Okay. And that's where things get very pricey. I tend to think about. And if you've heard me talk about AV in the past, I talk about these grand challenges that have survived decades in AV that we still haven't quite solved the LLMs are going to have a play there. So some of those giant grand challenges like the 10 minute meeting start problem, I think I can, you know, bring an LLM into play and we have some ideas about how to do that in the future. That will be really exciting. But if I'm using an LLM to aim a camera in a room, then that's just not architected appropriately. You need an architecture that supports all three. So often I talk about full stack av and that's exactly what I mean. You need smart peripherals attached to Edge Compute, that's doing stuff in the room at low latency. And then you need that attached to basically a flexible cloud infrastructure that's, that can flex when needed to solve the bigger problems. [00:14:12] Speaker A: It almost sounds a little bit like [00:14:13] Speaker B: the discussion we had on broadcast. [00:14:15] Speaker C: Right. [00:14:15] Speaker B: We still need to have the traditional [00:14:17] Speaker A: baseband SDI, but video over Ethernet, you [00:14:20] Speaker B: know, video over IP21, 10 NDI, that [00:14:23] Speaker A: makes us flexible as well. So we have to have kind of [00:14:25] Speaker B: a little bit of the left and a little bit of the right at the same time. [00:14:27] Speaker C: Totally, totally. Yeah. That's a great analogy. You do, you need a little bit of both. And I think when you marry yourself as an end user, if you marry yourself to architectures that force you down one path, that's a little bit risky. So it's better to have sort of that full stack approach. When you think about deployment, you mentioned [00:14:44] Speaker A: at the onset kind of the difference [00:14:45] Speaker B: between automation and AI, those terms kind of get conflated quite a bit. [00:14:51] Speaker A: So where does automation make the biggest [00:14:54] Speaker B: difference for the end user for the AV team as opposed to AI? [00:14:58] Speaker C: Oh, oh, interesting. Well, whenever you have predictable state that needs to drive control, then you should use automation because it's a cost effective approach. Right. So if I had. Let's, let's use an example out of higher ed. If I had a classroom that I know is scheduled and I've got booking data and I can auto enable the lighting in that space and do a, you know, an auto check on the mics, build that algorithm, do it. That's a highly valuable thing for the AV integrator consultants to think through. But if I want to report on the predictable arrival time of the students in that space because I've observed it for a year, that's an AI question. You aren't. That's not a traditional AV question even anymore. You're starting to ask things about patterns of arrival because you want to know, did I book the right size room for these classes and can I be More optimal. As I think about my real estate utilization going forward. That's a clearly a different category of problems that we now have access to as an AV community, which is really exciting. I mean, I really encourage us as both end users and integrators and vendors building the technology in this space as a community, as a whole, to really be thinking about how we define av. Because as AI arrives, if you think about it, we are the last hop to the ears and the eyes of people, right? That is what we do. We have a rightful seat at the table to help define how that's going to change our lives for the better as AI arrives. So for me, that's like a social mission. Like, how do I make a space more embracing and better for people as they show up for a meeting? Can I ensure that the temperature has been set based on user preferences, not just based on automation? Like, what's the, you know, at best I could maybe do a web hook to read what's the weather outside and what's the, you know, current temp in the room. But now we have access to a way broader set of capabilities. Like, if I can know the volume of that space and I know how long it takes for me to heat it, and I can predict the weather because I have all those patterns and I can feed that into an LLM and say, what's the best setting for this temperature for this room at this size? Go do it. That's perfect. That's like AV control. We could have only dreamed about, you know, 10 years ago. I knew a few consultants that were dreaming about it and designing very complicated systems, but now it's more accessible to everyone. [00:17:30] Speaker A: There's another kind of more advanced term [00:17:33] Speaker B: that we've been talking quite a bit about with clients, and that's the concept of an mcp, right? A model control protocol. And at a very high level, it allows you to use an LLM to interact with an application programmatically, but it's doing it programmatically in the background via API. Normally you're just typing in things, and [00:17:51] Speaker A: where do you find that that comes [00:17:53] Speaker B: into play as a bridge between automation and AI? [00:17:56] Speaker A: Because now you don't have to train [00:17:58] Speaker B: someone on how to use an automation system or how to use a fancy dashboard that Claude has written for you. But how are you actually using MCP [00:18:06] Speaker A: to, say, communicate in the way you [00:18:08] Speaker B: want to communicate to get that outcome? [00:18:10] Speaker C: Yeah, there's a couple of views of an MCP server that we should talk about. One is the what. What does it. How does it Replace the old API approach to things. And I think that will change our industry quite a bit. You know, in the, you know, even eight years ago, if I wanted to control a third party microphone, for example, I would have to go meet with that company, ask them, do you have APIs, negotiate a licensing deal, talk to them over time and then babe, maybe we'd become partners and we'd build an integration pipeline. Right? And end users are like pulling their hair out the whole time, being like, why is this stuff not all work together? Well that's part of the problem with mcp. We basically up leveled that conversation, said just publish your capabilities and give access rights to agents that can now query that server and say, what actions can I take here? Can I read, you know, the microphone levels from this room? Because I have my use case that I need that allows us to move really fast and I think that already will have a dramatic impact now on the agents being able to access things, to do automation. That's where I get really excited because end user outcomes is kind of where I get jazzed. So now an end user that comes to a room and says why isn't the, you know, the TV monitor on and gets frustrated and has to call facilities or file a ticket. Why wouldn't we, right, right now building be building agents that could automatically turn on the camera in the room, look at the tv, turn the TV on through edid know or through, you know, the control protocols and then if it doesn't show up, send a ticket to ServiceNow. That what I just said probably could be done by a clever developer in a basically an hour. You know, that used to be a. Whoa, what a futuristic view. That's why I say AV is in a really unique position to really take advantage of some of these changes that have happened. [00:20:04] Speaker A: So this came up during our pre [00:20:05] Speaker B: call and it really piqued my interest. A lot of companies are now using note taker apps to generate meeting summaries and some of these tools are starting to add analysis features that help managers understand trends in employee meetings. [00:20:19] Speaker A: Is that the kind of, the same [00:20:20] Speaker B: kind of monitoring and analytics now coming to classrooms and conference rooms? [00:20:24] Speaker C: Yeah, oh yeah, for sure. You know, I get such a kick out of it because I remember meeting with a university professor at Harvard who was thinking through av and I was there with an integrator and this was a long time ago, probably 11 years ago. And he said, you know what the dream is is a real time indicator of student engagement over time. Right. Okay, so fast forward today I, I couldn't have built it for him before, but if you think about it now, if we have a, you know, like a vision suite system that's got eight, nine cameras in the room that can see the everybody's face, their kinematics, right, how they're gesturing, where they're sitting. I could easily build a insightful real time graph that shows student engagement of a large tiered classroom to the professor live as they're presenting. I used to be a professor, so I would have loved that actually. I'd be like, this slide's boring, people go on to the next one, you know, and then that analytics data goes to the cloud. Here's where it gets interesting. That data, that data is like, it's like oil. When cars got invented, it just became way more valuable. Yeah, it's gold. Now. It used to be interesting, but you had to hire, you know, data scientists and teams to go dig into it and use really clever techniques. Now you can just hand it to AIs for context and start to get some really interesting insights. So imagine I have that kind of engagement data, not just for the end user professor, because that's already valuable, but I can now correlate it with things like what equipment was in that room, what was the mean temperature, where is that room located on campus? Can I start to see patterns or ask for patterns around that data to say, well, well, Classroom Building 3 is actually underperforming and we might know why. I mean, that would be a dream for, for a provost at a university to start to understand how they can better become more efficient, deliver better outcomes to their students. So that's where I get excited. [00:22:26] Speaker B: I think some of the other things that plays right into is that we've had that YouTube analytics right for years that tells us when we're losing viewers, when we're losing interest. And we're now applying that same methodology. And what I also like is we're all aware of the changes in learning in every subsequent generation and being able to know that, hey, maybe I'm teaching in a way that was great in the early aughts, but maybe hasn't translated to this new generation. Maybe it means I need to start [00:22:55] Speaker A: speaking in sound bites or I need [00:22:57] Speaker B: to have more flashing images on the [00:22:59] Speaker A: screen or just things to tailor what [00:23:01] Speaker B: your presentation is to the audience. So you are retaining it. [00:23:04] Speaker A: So again, it almost feels like you're [00:23:06] Speaker B: tweaking a YouTube video. [00:23:08] Speaker C: Yeah, that's a really good insight. You're right. It's very similar actually. But the, but the closed loop tweak could include everything from real estate planners and facilities management to the professor themselves as they think through how do I design my coursework? My pedigree. Yeah, that's great. [00:23:25] Speaker A: So you mentioned Vision Suite and we've [00:23:28] Speaker B: obviously heard a lot of buzz about it, especially coming out of Infocomm. [00:23:31] Speaker A: So for folks who are just sharing [00:23:32] Speaker B: that in for the first time, what is Vision Suite today? [00:23:35] Speaker C: Yeah, Vision Suite is a fairly aspirational product that is delivering on that vision of AI enabled large conference room classroom experiences through camera control and direction. So in the, you know, imagine like a human operator that's operating, you know, a tiered classroom at a premium university for broadcast and remote students. You would have a control booth and a user there and a professional director thing and zoom in on that guy. Uh oh, a student's asking a question at the back of the room. Camera six switch, you know, well, this system does all that automatically and it does it in ways that really mimic the smooth experience that a human operator would have created, your director. So imagine this intelligent AI taking like I'm actually sitting in one of those rooms right now, six of those cameras, and be able to zoom in on me when I'm talking. But as I stand up in my chair and say, I meant that over there, you know, that it's able to pan out automatically and see that. And then you can set up trigger zones, which I find really interesting. So there's certain things in control based on where I am in the 3D space that you can trigger. So as I walk to a whiteboard, I can switch on the document camera that's looking right at that whiteboard and then broadcast it out. Or I can create exclusion zones and say, don't look at this. You know, students that walk into this door, if they're making noise because they're always arriving late. So I'm going to go ahead and you know, cancel that space out. So the whole vision of this is to build an experience that's much more similar to as though you're there now, there's work to be done to even go farther to this true vision of telepresence. But this is on the path for that. And I get really excited about that because in academics I used to work in an area that was in the telepresence field and I felt like it was still 50 years away. And with this product, I feel like we're on the path now, which is really exciting. And it's deployable and manageable, just. But through a bunch of software tools that let you drop it into a space and get it configured and deployed. [00:25:42] Speaker A: So if we take a step back, [00:25:44] Speaker B: and I'm sure you've had a lot of folks ask you this, I want to start introducing AI, not automation, but AI into our classrooms, into our lecture spaces. What are the number one things that they need to think about before they deploy or just try and plot a checkbook? [00:26:01] Speaker C: Yeah, I think that's such a good question. We hinted at the answer to that earlier. I think you have to be careful to adopt a full stack platform approach and not an AI product, or even worse, AI as a feature kind of approach. You know, you just can't get attracted by the shiny. This thing can do AI too, and I need it because you might end up with disconnected systems again that aren't truly, you know, driving these kind of outcomes. The other thing to keep in mind is that as AI emerges and more data arrives, right as the cameras are able to track student position, or I want to build an auto attendance engine on top of that to detect motion at the door or whatever it might be, those things should be flexible enough to be able to respond to those needs as they arrive. So that means software based, it means full stack, peripherals, edge, cloud, all involved in working in concert as a single system. And it means innovative companies that you latch yourself to. I mean, that's how I would view it if I was thinking through how I'm going to be responsible for the next generation of a classroom or a conference room. You'd have to really think those principles through. [00:27:12] Speaker B: In talking to clients, there's often the expectation that because you're using AI, things will be delivered faster and things will be delivered less expensively. And I kind of want to get your take. Does AI change the installation process, the design process, or the training required for AIAV teams? [00:27:29] Speaker A: And is it less expensive? [00:27:32] Speaker C: That's. Those are. Those are like maybe three questions in there. That's good stuff. Does it change the deployment? Let me start there. It will. I think it absolutely should. I think the vendors need to be responsible and accountable for delivering AI systems that focus on that part of the value chain. So it'll get very tempting to build like AI based features that wow the end user because all that means is, oh, top line sales growth. I got an end user excited. But what we should be responsible for is a lot of the value is in the deployment challenges that our AV integrators have. Right? Rolling a truck to a site, getting equipment in the right place, turning it on, making sure it's maintained, making sure, it's operating. That should be the focus of agentic AI honestly agents that monitor that process and assist as the deployment. So I think that's already going to happen. But you asked about cost. You need a company that's responsible in thinking about AI and not just token spend means more value to the end user because that's going to end us up in not a great place. I will say, I'll go on record as saying I think token value is enough for everyone. And I don't believe. I know I have some technologist friends that believe that we may be over leveraged on token spend and they're being under over subsidized by the. I don't believe in that. I think that there's enough value there for everyone to benefit it. That being said, you have to be very, very cautious about how you deploy your architecture. So I can't necessarily say I'm going to, you know, just access our cognitive cloud that's constantly spinning on LLMs to do things like, you know, book your next best meeting room. That's, that's not a great idea when you can do that on prem through the booking panels themselves because they've got a little chip in there that by the way could be, you know, through a couple signals, you know, given a small, very, very fine tuned model. This goes back to that agentic model architecture that you hinted at. You have an orchestrator, you've got what we call stewards in our architecture and then you have probes. Those probes are actually little AIs but they can only do one thing or learn one thing about a space. How many people are there? [00:29:37] Speaker A: No Sherpas. No Sherpas in there. [00:29:39] Speaker C: Yeah, I would love that. Sherpas that help move the data. But when you do that you're actually being very cost effective. So that helps. And then it remains to be seen how this all gets commercialized over time. The obvious choices are things like subscription models, one time contract buys all that. I think the market will tell us how that works. The good news is consumptive models are feasible in a token based architecture that's using AI on the back end because that's a per query kind of thing. And you can be careful about how you do that and be responsible as an end user as well. [00:30:14] Speaker A: You bring up some good points, but [00:30:16] Speaker B: I know we've been bid at this at key code which is hey, we're going to buy an enterprise AI plan and then oh, we've used up all our tokens in the first few weeks Right. So how are the economics for token consumption versus maybe something like OAuth where it's just, hey, you get to use X amount per month. [00:30:33] Speaker A: Are you seeing the economics kind of [00:30:34] Speaker B: different in the AV space compared to other verticals? [00:30:39] Speaker C: Yeah, well, I think so. I mean in AV there's a, you know, the, the market has been set up around historically around hardware based sale selling. Right. So it's really, you know, one time buy, you now own it, you got a maintenance contract on the back end. We have evolved of course, the Avit Convergence into SaaS models, managed services, et cetera. I think that the token spends will get probably embedded inside of managed services and SaaS models over time. But users have to be very, you know, hold us accountable to that and make sure that there's value there. You know my statement about there's enough value in the token for everyone, that's like a rallying cry that I have. We have to prove that by delivering true value. And I, I hinted at some. I won't go into the details because it's in our roadmap. I'm sitting in, in Zurich in one of our AI labs and the team upstairs is building something that is super interesting, but I can't talk about it. But I will say if it materializes, users will know because we're addressing some of the grand challenges in av, things that didn't seem that feasible for AV to even address. It needed maybe some new innovation from a massive tech company. The reality is we are massive tech companies in this space and we are sitting in one of the most important parts of the value chain, human experience. So we are able to build really compelling solutions that do some pretty insightful things and I think it'll really change the way we manage, deploy, understand and even experience our built space. So I'm pretty excited about that. [00:32:12] Speaker B: I would almost counter to some extent with a topic we covered here on the podcast called DMF Dynamic Media Facilities. And it's meant for broadcast, but it means when there are times where you need more computer, you have the compute there and when you have other times when it's not being used, well, it can be repurposed for something else. [00:32:30] Speaker A: It's not just purpose built. [00:32:31] Speaker B: And I would argue that given if we take the supply chain issues and the cost kind of table that for a moment that, you know, we're able to run decently powered 30 billion parameter models on prem. [00:32:44] Speaker A: And as you pointed out earlier, we [00:32:46] Speaker B: don't need the most expensive model to schedule booking a room or to change the temperature. So I'm optimistic that we'll be able to deploy reasonably powered models on prem that you could potentially unplug the Ethernet cable from at least the one that gets a public ip and it can run everything locally without having to rely on the cloud. But that's obviously very optimistic. [00:33:10] Speaker C: Oh, I like that optimism. I think it's real. I think it's real. I think that that's a balanced view of how technology actually works. Right. Like, I've got a lot of compute sitting right in front of me on my laptop right now. You didn't have even close to that 15 years ago. It's here. And it's for sure there's valuable models that I can run right on this laptop that do really compelling things. Can I do everything that the large frontier models do? Of course not. Those are way too big. But we don't need them to address every problem in av. Certainly not. So the vendors that build in this space, us included, have to be very aware of that and careful in how we architecture. [00:33:50] Speaker A: So final question for you. [00:33:52] Speaker B: When folks are coming to you to talk about, again, automation and AI, is [00:33:57] Speaker A: there kind of a checklist of things [00:33:59] Speaker B: that you have in the back of your mind that you say that you would give a potential client and say, here are a dozen questions I'd like you to answer. Here are half a dozen questions. These are things I need you to think about before we even start whiteboarding this. [00:34:11] Speaker A: So are there a handful of questions [00:34:14] Speaker C: that really need to be answered? There are. That's a good question. There's a couple obvious ones. Right. There's the how are you thinking about security and privacy when you're considering deploying? Because those are two different things. Security, obviously is really, really important. Privacy is very important in all spaces, I think. And you have to think about that social contract as you build these systems, are you providing value to the users that are producing the data? That seems like an obvious question, but it's the philosophy by which I've built a lot of products. If I built a company that collected data for someone else and the end user doesn't see any value whatsoever, I think you might have some issues with the privacy problem there. But if an end user becomes aware that they get value out of it, then there's a social contract in using the system because they get value. It's sort of why I'm willing for Uber to know where I am on my phone all the time. Right? Because I might want to call an Uber. But if all Uber did was sell my location to ad agencies or something. I would probably deinstall the Uber app. I just wouldn't want it. So I think there's that dialogue you should have. What your privacy and security policies should think through too. What your goals are, what problems are you trying to address. Like you can't get too attracted to the here's the technology I want to use before you know what problem you're trying to solve. That's super important. And those are the best customers that can will probably be the most successful is here are the key challenges we've had right Like I can no longer deploy, you know, my remote learning classrooms because they're too expensive. Okay, problem how do we solve that? Can we do that with automation and AI? Yes. Let's talk about how that would look and how you get that deployed. Those are the challenges that I think that, you know, users should be thinking through in light of AI and not the other way around. What's AI and how can I use it? I think really just start to make your laundry list of challenges you want to overcome and then ask can AI assist? [00:36:14] Speaker B: Chris, this has been wonderfully insightful and I appreciate your time on the podcast today. [00:36:19] Speaker C: Great. Yeah, a lot of fun. Thanks for having me. [00:36:22] Speaker B: Thanks for watching Broadcast to post don't forget to follow Keycode Media on social and contact us about your [email protected].