Can AI tech work in live television broadcasts? Carl and Richard talk to Tim Huckaby about his work building Duff Foreman, an AI-powered avatar for live golf programs. Tim talks about using AI to build AI, using M365 Copilot to architect and ultimately implement the AI avatar, pulling from golf statistics to provide a digital commentator for the sport. The conversation explores the stack of tools needed to build Duff, some of which Tim selected and some that were already in play for the project. And then there was that first live broadcast - Duff "sitting" with two human presenters during a minor PGA event. Lots of great stories and questions about the future of this tech!
Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.
Transcript
From the show. Tap a time to play from there.
Speaker 1How'd you like to listen to. NET Rocks with no ads? Easy. Become a patron. For just $ 5 a month, you get access to a private RSS feed where all the shows have no ads. $ 20 a month will get you that and a special. NET Rocks patron mug. Sign up now at patreon.dotnetrocks.com. Hey, and welcome back to. NET Rocks. I'm
Speaker 1Carl Franklin.
Speaker 2And I'm Richard Kemp.
Speaker 1We're here for another hour of silliness. Probably silliness because Tim Huckabee's here.
Speaker 2I strongly suspect, yes.
Speaker 1But all good meaty stuff, too, and some good stories, which Tim always brings. But before we do that, let's talk about this being episode 2024. Let's talk about what happened a couple of years ago.
Speaker 2Yeah, it's only literally a couple of years ago. I've kind of gotten to the end of this. Almost to the end. Where do you want to start? I know what you're going to start with.
Speaker 1Well, I mean, if you look back at the top 10 events, they're pretty historical events, right? And we just don't know it yet because we're only two years out. But first of all, Donald Trump won the 2024 U.S. president election, right?
Speaker 2That's historic.
Speaker 1And that's after Joe Biden withdrew from the presidential race and Kamala Harris was appointed to, you know, the actual running mate. The Israel-Gaza war expanded into a broader Middle East conflict. The Russia-Ukraine war enters its third year and Russia gains some ground.
Speaker 2Second year, but yeah. It's 2022, it's 2024.
Speaker 1Okay. I would argue the more- But it did enter the third year in 2024.
Speaker 2I think the more important thing was this was the point where Ukraine actually took Russian territory. They entered the Kursk area and the Western restrictions on I realize in hindsight now that when they went into Kursk and nothing bad happened, really. I mean, it's a war. It's terrible. But, you know, the Russians always said that was a red line. That red line was nothing. That was the point where the Western nation said, okay, yeah, you can launch our missiles into Russia now. That's not a big deal. And really did change the vibe.
Speaker 1And North Korea brought in a bunch of troops and promptly died.
Speaker 2They all died. But first they got on the internet and spent their whole time with cell phones on porn.
Speaker 3Yeah.
Speaker 1For the first time.
Speaker 2Exactly.
Speaker 1This is good. I want to stay here. Boom.
Speaker 3Boom.
Speaker 2If you turn on your phone, you get bombed.
Speaker 1Bashar al-Assad's regime collapsed in Syria.
Speaker 2Yeah. Took long enough.
Speaker 1Kind of a surprise.
Speaker 2We really haven't heard much from Syria since then either. They're just trying to pull the pieces back together.
Speaker 1Donald Trump had an attempt on his life on July 13th.
Speaker 2Crazy.
Speaker 3Yeah.
Speaker 1Butler, Pennsylvania.
Speaker 3Yeah.
Speaker 1Let's see. Historic global election year.
Speaker 2Yeah.
Speaker 1Produces an anti-incumbent wave. Roughly half the world's population lived in countries holding national elections in 2024. And governments or governing parties suffered major setbacks in the US, UK, France, India, Japan, South Africa, other places. Inflation being the cost of living is a major factor there.
Speaker 2It's also the Olympics, Paris.
Speaker 1Mm-hmm.
Speaker 3Yeah.
Speaker 1The Francis Scott Key Bridge collapsed in Baltimore.
Speaker 2That was amazing.
Speaker 1Because the boat didn't put on the brakes.
Speaker 2Well, it had a whole bunch of problems.
Speaker 1Of course, I'm being facetious, but boat plowed into the bridge, basically.
Speaker 2Crazy.
Speaker 1Hurricane Helene, just absolutely devastating in the southeastern United States.
Speaker 3Yeah.
Speaker 1That was in September. And South Korea had a martial law crisis. You remember that?
Speaker 3Yeah.
Speaker 1December 3rd. And parliament rapidly voted to overturn it, but it was like, I hereby declare martial law. Everybody's like, what?
Speaker 2No, you don't.
Speaker 3Yeah.
Speaker 2How to lose, to not be a leader in a big old hurry.
Speaker 1Yeah.
Speaker 2Haiti collapsed. Again.
Speaker 1Into escalating gang violence.
Speaker 3Yeah.
Speaker 1We don't know what to do there. And the Venezuela election crisis. There's more, I'm sure, we could talk about, but pretty historic year.
Speaker 2All such good news.
Speaker 3Yeah.
Speaker 2Yeah.
Speaker 1That's the bad news. What was the good news in space and.
Speaker 2Tech, Richard? Um, so the 261 space flights in 2024 up from 223 in 2023, the numbers of course are huge. Like this is crazy growth and SpaceX is 138 of them, including their four Starship test flights. So more than half of all the flights in the world in 2024 come out of SpaceX. China's the next most numerous at 64, but you know, SpaceX doubles that too. Um, Various events in January, JAXA, that's the Japanese space program, lands their
Speaker 2slim lander on the moon. That will make them only the fifth nation to have a soft landing on the moon. It's soft, but the vehicle's also upside down, so it doesn't last very long. It tips over and ends up inverted. Also in January, ULA, the United Launch Alliance, launches their Vulcan Centaur after many, many delays and problems. They finally have a maiden flight, and it is successful. This is their replacement for Atlas V and Delta IV. The last Delta flight ever will be in April. Their payload is
Speaker 2the astrobiotic Peregrine lunar lander, which fails in orbit before it even gets to the moon. And so nothing wrong with the Vulcan Centaur, although there are problems with the Vulcan Centaur. It hasn't flown much since. But yeah, the lander was a failure. And also in January, the Ingenuity helicopter on Mars. This was a little experimental vehicle on the– The Perseverance rover that was only supposed to do five flights, it did 72. Yeah. And on that 72 flight, it tipped over as it was landing and damaged the
Speaker 2rotor so it couldn't fly again. And that was the end of it. But it definitely set the path for flying vehicles on Mars going forward.
Speaker 1Yep.
Speaker 2In February, Intuitive Machines has a soft landing on the moon. The first time an American vehicle will soft land on the moon since 1972, since Gene Cernan and Apollo 17. In March– We have Starship Test Flight 3, one of the more successful ones. This is the largest vehicle ever flown in an orbital-class rocket, nine meters wide, unbelievably huge. And it's starting to work, these first few flights. I would argue most impressive space mission in space. 2024 is in June. It's China's. It's the Chang'e 6. So this
Speaker 2is their sixth version of a moon lander. This is the one that does sample returns. So it not only lands on the far side of the moon using the telemetry satellite to be able to communicate there, but then its rover picks up. or the arm picks up some samples, loads it into a capsule that launches back into orbit, returns to the Earth and soft lands. Incredible mission.
Speaker 1Wow.
Speaker 2Contrast that also in June to Starliner, Boeing's alternative to the Crew Dragon that's now wildly behind schedule. Its first initial test flights had both failed for various reasons, so why not put people on it? Yeah. Sonny Williams and Butch Wilmore managed to get to the space station and At the time, there was obviously problems, but we didn't understand the full scope of it. We do now. Their lives were absolutely in danger. They do manage to dock to the space station with its maneuvering systems largely crippled and are supposed to be there for a week and
Speaker 2take Starliner back down. NASA decides it is not safe and insists that they stay. They end up staying for nine months.
Speaker 1You know, the funny thing about that was everybody felt sorry for them and felt bad for them. And they're like, are you kidding? We want to stay forever.
Speaker 2Exactly. They're astronauts. Astronauts are not normal humans. They are delighted. And eventually they were delighted to stay. Yeah. Crew Dragon mission is changed to leave two seats available for them so they can come back down. Starliner returns empty. successfully, by the way. But yeah, it'll take some time before they'll finally recognize that it's a Class A mishap, that the vehicle was absolutely dangerous. And you will note, it has not flown since.
Speaker 3Yeah.
Speaker 2Although there are rumors they may attempt again later this year. In September, a Crew Dragon privately rented by Jarek Eisenman, Jared Eisenman, the current NASA administrator, does the Polaris Dawn mission. This is the highest orbiting vehicle, manned vehicle since Apollo. And he'll do the first commercial EVA ever. Just another experiment with what the Crew Dragon can do. Two missions in October. Both are to the elsewhere in the solar system. In October, we have Europa Clipper. This is an ESA
Speaker 2and NASA mission. Solar-powered mission to Jupiter, which is crazy because it's like 1 25th of the solar power available out of Jupiter compared to around the Earth. And its job is to evaluate life around Europa. It's not going to go into orbit around Europa. The radiation levels are just too high around Jupiter. So it does this big looping orbits in and out of the radiation fields of Jupiter. But it'll do 59 close flybys of Europa, although that will not happen until 2030.
Speaker 1Hmm.
Speaker 2And also in October, the HERA mission is launched. This is the follow-up mission to the DART mission. DART was the mission to redirect an asteroid. So this is Dimorphos, the asteroid that they deliberately impacted. So this will actually arrive later this year in November 26. The HERA mission will be to evaluate the fact that they changed the orbit of Dimorphos far more than they expected. They were expecting just a few seconds to change, like 70 seconds, and it ended up being 30 minutes. So they're starting to understand what it would take to modify asteroid orbits,
Speaker 2you know, because the reason the dinosaurs died, no space program.
Speaker 1That's right.
Speaker 2Yeah. And that's what I got for space for 2024. Should we talk a bit about compute?
Speaker 3Yeah.
Speaker 2I mean, I don't want to make it all about AI because clearly it is all about AI 2024 for better or worse. But I would bring up like in February, Apple launches the Vision Pro just to prove that not every new product that Apple makes is a wild success.
Speaker 1Well, okay. How about the Apple glasses? That's just what we're talking about. Yeah. That was 2024? That's 2024.
Speaker 2Yeah.
Speaker 1Is that when you became a glasshole? No, no.
Speaker 2That's Google Glasses. Oh, that was Google Glass. Okay. Yeah, yeah, yeah, yeah.
Speaker 1That's why I was confused.
Speaker 2But this is Apple. And I did not... I considered buying this. They're really expensive, like $ 3, 500. But I ended up not getting one. But... Yeah, not exactly. It doesn't look like there's going to be a follow-up to it.
Speaker 1I didn't buy either of them, so you can understand my confusion.
Speaker 2You're fine. Also in February, Blue Sky launches. This is the time when Twitter is being mangled by a certain tech billionaire and gets a million users on the first day.
Speaker 3Yeah.
Speaker 2You can argue how good or bad it is. It's its own thing, but there it is. I like Blue Sky. I've been there since the beginning. I like Blue Sky too. Yeah. Yeah, I've been there since day one. All right. Now, for better or worse, now we get into May and it becomes all AI all the time. So OpenAI in May ships GPT-4.0. That's the multimodal version.
Speaker 1Right.
Speaker 2So text and images and audio. Also in May, we have Microsoft's Build event, which is where they launch that huge GitHub Copilot focus. Now you can submit issues. You can create issues in GitHub and assign them to Copilot.
Speaker 3Right.
Speaker 2And it'll submit PRs to make your changes.
Speaker 1This is where we learned about MCPs, right?
Speaker 2MCPs had already been around for a while at that point, but this is the first time I think Microsoft's AI plan was more coherent for developers. Like, I mean, GitHub Copilot had been around for a couple of years at this point, but now they really much are tuning up what we today would call a harness. They weren't even using that term yet. Also at Build, although they actually become available in June, are the Copilot Plus PCs.
Speaker 3Right.
Speaker 2I question that, you know, over on Windows Weekly, we talk about how messed up this launch was in so many ways, because this was actually about the Qualcomm Snapdragon X processor. So bringing back ARM to Windows again, although I think with much more success. But they tie it in with this AI story in a way that I think really cripples it.
Speaker 3Yeah.
Speaker 2Because to be a co-pilot plus PC, you have to have an NPU. That NPU needs to be at least 40 tops. But hey, my RTX 5090 with 2,000 tops doesn't count.
Speaker 3Yeah.
Speaker 1I mean, and I think they misjudged that people wanted to use– well, we're going to use these PCs. They wanted to use them for running models and stuff, not chatting with– your local co-pilot. You know what I mean?
Speaker 2Well, they, that was the whole idea, but at that point they really hadn't got anywhere. The launch product, the thing that they were really focused on was a thing called recall. Do you remember this? It's only two years.
Speaker 1Oh yes, I do.
Speaker 3Yeah.
Speaker 1And we were concerned that it was going to be on by default.
Speaker 2It isn't.
Speaker 1No. But it basically takes a screenshot of everything that you do and saves it somewhere. And so you can go back and.
Speaker 2Say, hey– So you can search it.
Speaker 3Yeah.
Speaker 1But, of course, you know, the security concern is, well, if I can search it, so can everybody else. Right.
Speaker 2And they did– you know, I don't know that it was as big a security concern as it was, but, boy, it was a botched launch. And the product still exists, by the way.
Speaker 3Yeah.
Speaker 2Like they– They quietly went and stopped talking about it for six months, ostensibly to improve security. I don't know they made any changes at all. They just never talked about it since. And now, you know, we question what Copilot Plus PCs even are anymore because there are Intel and AMD versions as well that all qualify and is next generation. At some point, these are just going to be PCs.
Speaker 1Well, I kind of think Microsoft got tired of coming late to the party and decided to be really early. And by being really early, they, you know, They had premature products, I think. Yeah.
Speaker 2Well, but I also like that the ARM PCs built on ARM is a good idea, especially with these new processors. Sure. And of course, for us as devs, it's not a big deal because we've always been able to operate on ARM. But over on the run-ass world, as sysadmins, corporations aren't buying these machines. They're expensive for what they are. And the features just haven't been emphasized enough. And there hasn't been the wave of, of senior leadership demand. no sysadmin wanted an iPad in the office. The problem was that
Speaker 2the CEO showed up with one and says, I'll be using this from now on. Figure it out.
Speaker 3Right.
Speaker 2And so we were sort of forced to. There's just never been that energy for ARM. BYOD. Yeah. That was the way it was, right? Yeah. One more thing in May, at the Google I.O.
Speaker 1Event.
Speaker 2So this is crazy. This is when Google sort of gets their act together with AI. Of course, they were Google brain. They were the originators there. Why OpenAI exists. They did BARD. It was kind of a disaster. but this is where they really get their sort of Gemini act together. And they talk about Gemini 1.5 pro. This is where the AI overview started appearing in Google search, right? Which we've, you know, it's been a couple of years of that now. They also talk about their Astra universal AI system, but I don't know that it really landed all that well. And their response to Sora, which was announced in February, although we don't get it till December
Speaker 2from open AI is a thing called VO. And speaking of Apple, Their WWDC in June, right after all this, is where they do the most smoke and mirrors thing I've seen Apple ever do in decades.
Speaker 1Apple intelligence?
Speaker 2The Apple intelligence. Oh, boy. And I mean, I feel for them because they do one show a year. And everybody's talking about AI. And they're not a hyperscaler. They're not a research base. They're not those guys. So they feel like they have to do this thing. And it's like, I'm not used to Apple blinking. They're usually pretty darn confident. But they had also shipped the Vision Pro that year. So they're struggling a little. Right. And Siri's got serious problems.
Speaker 1And they had to say something.
Speaker 2They had to do something. So they made this demo. And like I said, it was like one of the first real smoke and mirrors things I've ever seen them do in really a long time.
Speaker 1And two years later, the most I can think of for Apple intelligence on my iPhone, which isn't really intelligent, Is that the Siri connected through CarPlay has a new voice.
Speaker 3Yeah.
Speaker 2Well.
Speaker 1And I can ask Siri to ask ChatGPT and give it a prompt and get something substantial back rather than just, you know.
Speaker 2I think it's finally two years later coming true.
Speaker 1Yeah. Which is normal.
Speaker 2Like Apple doesn't. experiment they do they deliver finished product like i was when they said apple tells you like cool name show me something amazing because that's what you guys do yeah didn't happen for a while yeah uh one more thing in june which i think is one of the most fundamental things about ai in 2024 which was when claude 3.5 sonnet shipped oh yeah because the later that year they shifted in june but they did major tweaks in october and october seems to be october 2024 is when the debate around AI and development kind
Speaker 2of changed. We saw it on the show, right? That all of a sudden, there's no longer, is this useful? It's, okay, what's the right way? Yeah. Anthropic changed the tone on this. It started in June, although it takes a little while from that. Also in June is when NVIDIA's market cap becomes number one, exceeds the rest of the world with the HP200. And it's like the video game card company suddenly becomes dominant.
Speaker 3Mm-hmm.
Speaker 2It's kind of insane.
Speaker 1Kind of.
Speaker 2July, CrowdStrike. Damn. I've never been able to do a follow-up show on this thing. So this is a product that runs as part of the boot-up sequence in Windows that is supposed to be for security. They botched the deployment, which they had done before, that literally causes machines to blue screen.
Speaker 3Yeah. Yeah.
Speaker 2And because it's mostly running on servers, these servers are now in boot loops. They realize within minutes they've made this mistake and push out a fix. And if you boot looped enough times, sometimes before you could crash, you could load the updated version, sometimes not. So the outage actually goes on for a while.
Speaker 1I knew there was a problem because I was in the airport and all the terminals were blue screen.
Speaker 2All of them.
Speaker 1Yeah.
Speaker 2Not a good sign.
Speaker 1You know, the ones that show you like where you should go.
Speaker 3No.
Speaker 1What baggage claim you're at?
Speaker 2Because a lot of airlines ran this security product. And the reality, of course, is these companies write into Ring Zero so they can crash Windows. And the answer is don't write to Ring Zero anymore. And their response was, okay, Microsoft, we won't write to Ring Zero if you won't. And Microsoft's not going to do that. So near as I can tell, the outcome of CrowdStrike was be more careful next time.
Speaker 1Right.
Speaker 2Which is a hell of a strategy, isn't it?
Speaker 1Yeah. Don't do that.
Speaker 2Yeah. And in November, we get. NET 9. This is one of the standard support versions on a long-term support versions with C Sharp 13. And our second version of Aspire called 9 because numbers are hard. But also, you know, Aspire sort of grows up.
Speaker 3Yeah.
Speaker 2Yeah.. NET 9 time where we really get a good scaffolding for building cloud-native stuff in.
Speaker 1NET.
Speaker 2One last thing in the quantum world, the Google Willow processor, we've been talking quantum off and on through this time. What's interesting about Willow is 105 qubits, but they're starting to do real integrated error corrections. This is a change from trying to push the qubit count up into making more reliable qubits. One would argue that Willow is the beginning of what we call fault-tolerant quantum. and the continuation of that race. And that's what I got. Okay.
Speaker 1Tim, before we introduce you, we got a couple of things to do here, but you could jump in at any time. Starting with Better Know Framework.
Speaker 2Awesome.
Speaker 1All right, man. What do you got? Simon Kropp's been busy since we last recorded a show.
Speaker 2Simon Kropp is always busy, man. That's not a thing.
Speaker 1He built a tool called Build Monitor, which is a build CI monitor that runs in the system tray on Windows, Mac OS, and Linux. It pulls the CI services a developer cares about, shows one row per pipeline with a progress bar and countdown for running builds, links straight to the build, the branch, and the pull request, and can retry or cancel a build and copies the log of a failed one. And that's just the first paragraph. It's
Speaker 1pretty cool. The other thing that he mentions is, and I'll include this link as well, That build monitor MCP runs a model context protocol server over standard IO. So a local AI assistant can read the same builds the tray shows and act on them. It starts the tray if it's not running and talks over the local port. So just talking to the MCP server, you can rerun a build or fix a bug or that kind of thing.
Speaker 2It's pretty cool. Awesome. It doesn't surprise me. He's always thinking about better ways to see stuff. And let's face it, there's more builds happening than ever now with these tools.
Speaker 1Yeah, that's right. So build monitor, know it, learn it, love it. And who's talking to us.
Speaker 2Today, Richard? Grabbed a comment off the show 2017, the one we did with our friend Billy Hollis, which I ultimately just named very simply, arguing about AI. Because we got a ton of comments on the show, which Billy responded to many of them. This one comes from Vale, who says, I love the discussion, notably standing out to see of them these days, but somewhere in the 50, 55 minute Mark about trades, the tech evangelist in me was ringing as the transition of software development into a trade has been my path. And in a world of athletes that didn't do great in school and went to work with their hands, uh, I find some connection to
Speaker 2all of this. Uh, Still trying to figure out who said this, but I suspect it was me. He says, we always talk about the trades when we talk to schools. I thought this is a perfect transition back into software on a software podcast. It was like you all joked about this. But starting in 2025 until my graduation in 2024, he came through the Orlando Seminole State's previously named Computer Programming and Analysis Program, which was both system design and a programming degree. And it skipped over the unnecessary com-sci
Speaker 2calc and physics parts. It was a more practical path in the workforce. So while the foundations are in Java, it was about data structures and algorithms and data sets and development and languages and that sort of thing to actually build software. So when people are talking about steering kids into trades because of AI, I also think that developers could also be more of a trade because most of us aren't building frontier AI systems. We're integrating software, moving data around, maintaining applications, solving business problems, and that's all skilled work with craftsmanship, even apprenticeships, and focus on
Speaker 2maintainability and scalability. I find it really interesting in the relationships and market for these skills and letting schools like UCF focus on aerospace, physical systems, and research, while community colleges produce software developers that are ready to communicate and be stewards of business systems. That seems like a healthy ecosystem. Also, I appreciated the historical perspective throughout the episode. Billy made some great points. You all did a good job of pushing back on that context. And for the record, I really hope the story about the philosopher being mad when the papyrus came along was true. That is true. It was written down at the time. Because humans don't change in 2,000 years. And that doesn't surprise me at all.
Speaker 2Thanks for the great episode.
Speaker 3Yeah. Yeah.
Speaker 1Yeah.
Speaker 2That's what I think too.
Speaker 3Uh, fail.
Speaker 2Thank you so much for your comment and a copy of music go by is on its way to you. And if you'd like a copy of music, go by, write a comment on the website at.net rocks.com or on the Facebook. So you publish every show there. And if you comment there and I read it on the show, we'll send you a copy of music to go by.
Speaker 1And if you don't want to wait for a published comment to get music to go by, you can just go to music to code by.net and get the collection in MP3 wave or flack formats. And if you don't know what it is, it's music took, code by keeps you in a state of focus right all right so now it's your turn tim finally after well we've spent half the show talking about stupid.
Speaker 3Well i i gotta tell you not chiming in listening to you to talk about 2024 was one of the hardest things.
Speaker 2I've ever done i know you should have.
Speaker 3You know, bringing us down memory lane two years ago, but doesn't it seem like that was yesterday?
Speaker 1It does seem like it was yesterday.
Speaker 2Yeah, but also forever ago, too.
Speaker 3Yeah. Okay. I mean, you know, I'm not the only one out of the three of us with gray hair, and it just seems like time is flying by.
Speaker 2Yeah.
Speaker 1So, I got to just say, you sent us two bios. One was the short bio, and one was the long bio. The short bio is probably the longest bio anyone has asked me to read on this show ever.
Speaker 3I'm not asking you to read it. In fact, we can skip it completely.
Speaker 1Well, you know, let's skip to the important parts. Like, first of all, your history with Microsoft and Windows NT and all that stuff and your Bill Gates demos in there. Internology was your first big worldwide custom application development company.
Speaker 3Yeah, the first company I started, yeah, a long time ago.
Speaker 1And then Actus.
Speaker 3And what I'm doing now is actually came from an internetology customer from 20 years ago that just tracked me down and said, hey, what are you doing?
Speaker 1So you're into television and AI. Do we want to brag a little bit about the CNN wall of voting?
Speaker 3Well, only because … Voting map. Only because I have a little bit of broadcast history with the CNN magic wall.
Speaker 1That's it, magic wall.
Speaker 3But unlike you guys, I've always been in front of the camera. Not that I was on CNN, but we do these Bill Gates keynotes. And there are 250 people dressed in black running around screaming at each other behind the curtain. And I always thought like, oh, what a waste of money those people are. What do those people do? Now that I had to live on the production side of a broadcast, my God, there are so many moving
Speaker 3parts and so much points of failure, which makes software so reliable. Now, I used to think like, oh my God, 100% uptime, 100% SLAs on live TV, that's impossible. Well, it's a lot harder in broadcast when you're you know, that one production company's in Nova Scotia and the other one's in Florida and then we're all in New Mexico in the studio and they're syncing voices together and all that.
Speaker 3Oh my God, it was complicated.
Speaker 1So you're talking about Gopher Hole Productions? Is this your newest venture?
Speaker 3Yeah, sorry, I skipped forward. So yeah, I've had a long, fun career but I just, before we go there, I just got, I was laughing at you guys talking about Billy Hollis because About 20 years ago, I told Billy Hollis and Uthal and all those guys, you know, within a decade, you guys are going to be obsolete.
Speaker 2Nice.
Speaker 3They, of course, reacted violently. I was a little off. It was about 20 years that they became obsolete, these ivory tower software architect people. Yeah.
Speaker 1Well, I mean, I brought it up on the show once that Pretty someday soon, and I was still in the old studio, so this was probably 15 years ago, maybe. Someday soon, you're going to be able to just walk up to a computer, have a conversation with it, and out will spit an enterprise application. And I remember Richard goes, nah, that's a long way.
Speaker 3Off, dude. Yeah, yeah. And then, holy crap. Even Richard is wrong every once in a while.
Speaker 2Yeah. Seems closer now, but depends on how you define it.
Speaker 3Well, that's another part of my story, which we should talk about, is stumbling bass-ackwards into M365 Copilot Pro, as opposed to going straight to Claude for coding this thing. But so, we should probably back up. You know, I retired for the second time. The queen said, you need to get back to work. And then randomly, this amazing guy, Paul Sidlow, this creative ex-internology customer, I think he built the ATM software. Not ATM, the fast food kiosk software.
Speaker 1Oh, yeah.
Speaker 3Everybody, McDonald's. This is a long time ago, too. We built that for him. He's won 16 Emmys. He's won Academy Awards. He's worked five Super Bowls. for Olympics. He is, you know, a talented, creative guy. And he, he tracked me down through LinkedIn, I think, ironically, and I say ironically, because I'm overwhelmed by LinkedIn messages, and it just doesn't seem very functional. But he tracked me down and says, Hey, what are you doing? I got this idea to do an AI avatar in a live golf broadcast. My first
Speaker 3reaction was, boy, that sounds stupid. Who would want that? You know, Jim Nance, I don't know how much you know about golf, but Jim Nance is a god in the broadcast world and you're not going to replace Jim Nance or any of these these great golf announcers um but the more i got to listening to him the more i thought how you know how do i turn this guy down you know his resume is just amazing and he gave me a good deal too and and um.
Speaker 1I'm like, what the hell?
Speaker 3I'll jump back into technology in a really hard way, like coding again, which I haven't done in a long time because I've always been the boss. And yeah, so cut to the chase. We did a live broadcast, the live avatar tech that I built by myself, if you can believe that. Well, with tremendous help from M365 Copilot Pro, which is the story in itself.
Speaker 1Wow.
Speaker 3It was wildly successful. The software was wildly successful. The negative reaction of the people who love golf and are traditional was unbelievable. Over 800 million media impressions. Yeah, I don't even know what a media impression is, but that's a lot. Yeah, and mostly were negative from these golf people. Yeah, wow. over what it was over a hundred live tv um
Speaker 3shots of the thing uh it was on fox and prime time you know all this stuff because it was so controversially and then one of the big guns in the company is really good at this social media thing i barely understand yeah so um well.
Speaker 1I don't know if you know this but most americans are really down on a high At least the ones that aren't using it, that's for sure.
Speaker 3Yeah, yeah.
Speaker 1And down on data centers, and it's a big, hot topic.
Speaker 3Yeah, and then there's the anthropomorphism thing that we keep, we as a humanity, not me, God help me, but keep attaching humanistic traits. So when anybody calls this avatar he or things like that, I always correct it. you know, the machine or it.
Speaker 1Yeah.
Speaker 3And really all I built was a tool that is kind of like the Wikipedia of golf. It just happens to have an avatar and that the broadcasters and the live broadcasters were big guns to note the brigade. The third, who's the current, um, Ben Crenshaw award winner, which is the biggest award in golf. He's a pro golfer, you know, Tiger Woods, roommate, Stanford. So amazing guy. And then Ryan Burr, who is ESPN and golf channel. So, they used this avatar. You know, if they didn't know the answer, they'd go
Speaker 3the avatar. And then.
Speaker 1So, tell me this. Wait, I got to understand how this works.
Speaker 2Okay.
Speaker 1So, who, can anybody ask the avatar a question interactively online and then it will display the answer? Like, who uses it?
Speaker 3Yeah, yeah. I could, if you want, I think. It was broken this morning because I broke it. I think I could demo it live if you want. You can ask it whatever you want.
Speaker 1Well, it's, we're not video, but. But is that basically how it works? Is that the people watching can ask it questions?
Speaker 3No, no, no. This is a broadcast partner of the two people doing the broadcast.
Speaker 1Okay.
Speaker 3So you see three of them on live broadcast.
Speaker 2Okay.
Speaker 3Well, you know, the production people bring up the Avatar broadcast. when, when the broad, the real humans talk to it.
Speaker 1I see.
Speaker 3And then they interact, they interact with it back and forth, you know, like.
Speaker 1Hey, okay.
Speaker 3What is the course record on this course? You know, and that.
Speaker 1Right. So if the two guys don't know the answer to something, the avatar will pop up and set them straight.
Speaker 2So to speak.
Speaker 3That's basically what I thought the purpose of the thing would be.
Speaker 2Color commentary.
Speaker 3What it, what it, yeah. Expert level commentary that, you know, a human just couldn't memorize all that golf. You know, there's no Richards out there in golf.
Speaker 1That's your next job, Richard.
Speaker 3But honestly, because Nota Beghe had to play in this tournament, the other broadcaster, Ryan Burr from ESPN, had to go by himself with the avatar for an entire round of golf. And he ended up just using it as a broadcast partner. Right. You know, just say, hey, what do you think of that shot? You know, blah, blah, blah. And honestly, those people are very talented, those broadcaster people.
Speaker 2Yeah.
Speaker 3I could never, ever do that. I'm just not talented enough to do that.
Speaker 2Just to fill the air, just to have a comment, to watch the game and interact like that. It's not simple.
Speaker 3So as long as we're still talking about the solution, the most interesting problem we had was when Ryan Burr asked the Avatar, a question that it didn't know.
Speaker 2Oh, wow.
Speaker 3And, and, you know, I'm back with all the production people. And I've built this producer console software. So you could, you know, live interjecting with data and all this type of stuff.
Speaker 2So you can see the stream of the back end of this thing?
Speaker 3Oh, yeah, yeah, I had to build all that set the transcript, because, you know, in the human language, we have all this, these weird phonetics problems that we don't even realize, like, You guys are not made to know this, but one of the most famous golfers in the world is named Jack Nicklaus. That's how he says his name, Jack Nicklaus.
Speaker 1Yeah, sure.
Speaker 2Never heard of him.
Speaker 3Well, it's spelled Jack Nicklaus.
Speaker 2Right.
Speaker 3And the avatar doesn't know that.
Speaker 1Yeah, yeah.
Speaker 3So you ask him, hey, tell me about Jack Nicklaus on this course. And he'll go, I don't know who you're talking about. They'll say it in a funny way. You know, all that, the prompt engineering is intense, really intense.
Speaker 1You want to hear a funny story? I tried calling Rocky Latka. I tried calling Rocky Latka with Siri. Siri, call Rocky Latka. I don't have a... Yeah, yeah, yeah, yeah.
Speaker 3Because we have all these phonetics. So I had to build a phonetics feature that basically maps. Anyways, it asked this question. The avatar said something funny like, you know, I'm sorry, I don't know. And it was a relevant question. It's like, what's the course record here at Turtleback Mountain Golf Course? And it said, I'm sorry, I don't have that access in my database. Give me a few minutes and I'll figure it out. And I'm scratching my head like, that is such an obvious question because the RAG pipeline is tight. The guard railing is so tight. Most of
Speaker 3its information is resident in the knowledge chunks. It rarely has to go to an LLM for an answer. So I'm looking and it went out to the LLM and it didn't get an answer. Well, guess what? It's not recorded anywhere on the internet.
Speaker 1Oh man.
Speaker 3Wow.
Speaker 2Randomly.
Speaker 3So the producer comes running into the room, points at me and says, run with me, get a piece of paper. And as we're running about 200 yards, we're going to the pro shop where the only human on the planet has the answer to this question. And it's this guy that works in the pro shop. He's worked there for like 35 years. I essentially have a napkin and a pencil.
Speaker 1And he's like- The name is Krusty.
Speaker 3You know, Jimmy Smith, 1973 with a score of 59 at a PGA blah, blah, blah. And I run back in there, type it into the broadcast intelligence feature, the hot thing that I built. They come back from commercial and Ryan, the broadcaster guy, goes right back to the Avatar. Hey! And he's trying to mock the Avatar. Hey, did you figure it out? Yes, I did. It was at 61 by Jimmy Smith in 19... So that I found super interesting.
Speaker 3Everything we want to know isn't on the interweb. And guess what? We have hallucinations for a reason because there's a lot of shit that is not right on the internet. Yeah.
Speaker 1Yeah.
Speaker 2Yeah.
Speaker 1You got to find the old guy in the pro shop who knows everything.
Speaker 2But also the, the tool is built to always answer even with low confidence. So. Right.
Speaker 3Yeah. And, and I can't, you know, I casually mentioned that, you know, this thing is contained. I use a gem. I can't remember the, the version of Gemma I use, which gets back to another interesting story, but that's the fastest one. And I knew that, you know, I didn't need something really rich in an LLM because- Yeah, yeah.
Speaker 2The scope's narrow.
Speaker 3You know, everything is so contained. And, you know, the golf avatar who has a name called Duff Foreman- You know, if you ask it, you know, hey, why do you hate Catholics? You know, it's just, it's not going to answer that question.
Speaker 2Right.
Speaker 3You know, it's always going to.
Speaker 2Because you're basically, this is live, right? Like you're maybe a seven second delay.
Speaker 3No, yeah, it's live. It's live. And if it answers that question, why it hates Catholics, I can say that because I'm a Catholic. Then, you know, it's over. You're dead. You're doomed. You're never working in this business again.
Speaker 2But an interesting point that you've got to resist the way these LLMs were designed in the first place, which is don't answer stuff you don't know.
Speaker 3Yeah, exactly.
Speaker 2Or it's out of scope.
Speaker 3So, which leads to, you know, the way this grows into an enterprise app for different sports and things like that, which makes it more and more and more complicated, you know. But it's still super fun work.
Speaker 2Absolutely.
Speaker 1Hey, this seems like a good place to take a break. We'll be right back after these very important messages.
Speaker 2And we're back. It's. NET Rocks! I'm Richard Campbell. Let's call Frank.
Speaker 1Yo.
Speaker 2Talking to our friend Tim Huckabee about Dove Foreman. Dove Foreman, what a name. So you said Gemini, like what's behind this thing? How'd you build this?
Speaker 3Well, That's fairly interesting because I mentioned Paul Sidlow and the genius creative that he is. And now I joke with him like, you know, I just love working with creatives until I don't. You've been there in your career, right? These are, they're such out of the box thinkers and like, oh my, no, we can't do that. No, that's, you know. So I inherited the avatar technology that he built the
Speaker 3prototype on. There's this company, and there's lots of, not lots, there's a handful of good companies that do avatar tech. The one I inherited is called Lemon Slice. And they have an amazing prototyping capability. And I thought, ah, well, just, go live with this prototype and I won't have to do anything. But as I got to building software, it became evident we needed a piece of software. Anyway, so I inherited the avatar, its voice, what it looks like.
Speaker 3Everyone thinks it's dressed up in a Scottish way.
Speaker 2It's very Scottish looking, my goodness, yes.
Speaker 3And there's the joke because there's no Scottish language tech. available it's all english so it's speaking a version of you know england and everyone thinks it's scottish so a lot of the critique that that we got was oh you're you know you're mocking scottish people well guess what it's it's an english it's an english accent.
Speaker 1We're separated by a common language Yeah.
Speaker 2Especially Scottish. Goodness knows.
Speaker 3Oh, my God. And here's where, as it relates to. NET and. NET Rocks, here's where it gets interesting. These lemon slice people have become an amazing partner and they have awesome tech. And I was going back and forth with them and they said, well, you know, we have a sample app. It's similar to something you're going to do. You know, no one's ever done live broadcast with our tech. It's mostly, you know, tech support avatars, right? This is one of those companies you talk to all the time
Speaker 3when you do tech support at some website. So they gave me this sample app, and I started staring at it, and it looked like Latin. It wasn't. NET MAUI. It was freaking Python. Oh, wow.
Speaker 2Of course it was.
Speaker 3And there's this other language in there, and I couldn't tell what the hell it was. It's called TypeScript.
Speaker 1Oh, no. It kind of looks like JavaScript, but not really.
Speaker 3My first reaction is, holy shit, I got to convert this to. NET because otherwise I'm doomed. And that was my plan. You know, I'll just run it through Clot or whatever and get it converted and work from there. But I only had like four weeks from, you know, this prototype to production.
Speaker 2Sure.
Speaker 3And I didn't think I was going to make it. Thank God I did. So I just started running with Copa with, um, um, um, um, um, Claude code.
Speaker 2No.
Speaker 3Um, with, uh, uh, the language I don't know, which I'm drawing Python, Python, the old guy's moment right there. Yeah. So I started running with Python, but, um, I had talked to Andrew Bruss and I assume you've had Andrew on the show.
Speaker 1Yeah.
Speaker 2And he's a data guy. So of course he has to breathe Python.
Speaker 3Yeah. But he knows a lot about LLMs and I don't, you know, I use the one I use and that's what it was. And he started lecturing me on the different ways, different LLMs and their versions. You know, some questions are good for some, some questions are good for another. And I got to talk to him and it was overwhelming, but it was wildly valuable. And I should also mention Matthew Renzi. And if you've not had him on the show, you need to have him on the show. He's the
Speaker 3one that taught me about open claw and avatars and self-deterministic agents. And this is not that, right? This doesn't run amok by itself. It's strictly guided. So I put two and two together. And at the same time, Microsoft Copilot Pro, which is the fancy version that I get, came out with model routing. And they didn't do a big fanfare. In fact, I don't even think they announced it.
Speaker 3But if you go in the configuration of Copilot right now, you can actually direct it to Claude. Or GPT or some others. Or you can say, just decide by yourself where's the best place to get this answer.
Speaker 1Wow, that's interesting.
Speaker 3Model routing tech. You know, I've been hearing about it for years from Microsoft, but I'd never seen it implemented. It was always like you were talking about before. It was always smoke and mirrors, you know, demo tech.
Speaker 2Yeah.
Speaker 1So I thought, what the hell?
Speaker 3Let me run with this in Copilot. And... You know, it was obvious that when I had a question about code, it was going off the clock, you know, and it was generating code and I would be asking it. But what was most interesting, for me at least, this is probably old half for you guys, but what was most interesting is I was more concerned about the software architecture of the thing.
Speaker 2Sure.
Speaker 3You know, because I have no experience in this, especially on the Python side of things. And it became like a consultant to me. I mean, a legitimate consultant. And it wasn't Microsoft jaded. Every once in a while, you would say something like, you should use the Microsoft blah, blah, blah, in so many words. But it was wildly honest about what I should do in software architecture with the assumptions that I gave it. Hey, I got the avatar's lemon slice,
Speaker 3so we need to work backwards from there. Okay, so that went wildly well. Not perfect. ironically, a co-pilot has a limitation on how long the chat can be.
Speaker 2Right. Right.
Speaker 3Or you knew that? I didn't know that.
Speaker 1Context window.
Speaker 3I got screwed by that. You know, I'm hot and heavy, you know, in the middle of the project and it says, Oh, we're done. Start a new conversation.
Speaker 1Yeah. Like, wait, what?
Speaker 3Are you going to remember the last two weeks?
Speaker 1The number of tokens in your context. And you can, you can adjust it on some things. I'm, I'm running a local model. And I have 128K tokens in my context window. And that's working really well.
Speaker 3Okay, well, I got burned by that. So there's that weakness. And there's some annoying just app-level UI behaviors of Copilot. Like the way it scrolls and the way you ask it a question, it'll jump like a mile up the thing. But the more I kept going, the more I kept saying to myself, I'm going to pull this off in Copilot. And I did. I freaking did. You know, it wasn't perfect. Probably would have been a lot easier in Cloud. Or GitHub Copilot, in all honesty. But I did, and I
Speaker 3found that wildly interesting.
Speaker 2Yeah, I know. Because, of course, I presume you're reusing GitHub Copilot just because that's the dev tool.
Speaker 3Yeah, yeah, yeah. And Visual Studio integration, all that type of stuff. Yeah.
Speaker 1I also did this little trick, Tim, which is I created a repo, a private repo on GitHub that contains memories that I want Copilot to refer to later. And I know it does some of that itself in compaction and all of that, but I basically have in my copilot instructions md file which is sort of the system prompt when you're using copilot i tell it to check that and that anytime uh we're doing a task before you do a task write it and you know publish
Speaker 1it up there so you know what you're working on so if something happens and the context blows up or i got to start over again i can just say you know it's right there in the Copilot instructions and it can figure it out. Okay.
Speaker 3You're talking about GitHub Copilot, right?
Speaker 1Copilot CLI. GitHub Copilot.
Speaker 3I'm talking about M365 Copilot. M365 Copilot. Oh.
Speaker 1Yes.
Speaker 2That's a different story.
Speaker 3If I could fix. Never mind.
Speaker 1No, Doug.
Speaker 3The way you fixed it, if I could fix it in M365 Copilot, or now that I'm done and I proved it, just move over to M365 or move over to GitHub Copilot.
Speaker 2Right, right, right.
Speaker 3Which is probably the logical thing to do.
Speaker 2Yeah. Yeah.
Speaker 3So the, the other funny thing about this old guy with gray hair is, you know, I was really worried about, you know, putting the stuff, you know, into production, just normal DevOps stuff. Sure. It seems so complicated and so riddled with failure. You know, I do, I do pulls and pushes and, and sometimes it wouldn't work and, you know, I couldn't understand. And then the, the, the, M365 co-pilot's
Speaker 3trying to explain it and I don't get it. It just seems so fragile. So, and I had so little time that, you know, I was starting to really to panic about just getting it into production. And at one point I said to the M365 co-pilot, like, I'm afraid to push this into production because, you know, we're an hour to broadcast and I just can't break the whole thing. And it comes back and essentially says, oh, don't worry,
Speaker 3it's just Python that we made changes to. Okay. And I'm like, what? What do you mean? And it's like, well, Python's interpreted. And I'm like, oh, my God, I forgot. I totally forgot. We're not even compiling. Like, what a world. What a world this is. That Moore's Law has gone like this. We don't even compile anymore.
Speaker 1It's VB all over again.
Speaker 2Yeah.
Speaker 3Or Fox.
Speaker 1What?
Speaker 3Fox Base. We're Fox Base again. Yeah. Which I found so amusing that we've got so much power in our CPUs that we just don't need to compile them. anymore you know we can run these interpretive languages that are so high level and so forgiving um so it's it's been a long.
Speaker 1Road oh it worked out okay for javascript yeah so it's.
Speaker 3Been a long road since you know working on you know a server product at microsoft where there's 2500 people on a team to an old guy by himself building broadcast level software in in six weeks what a world huh Yeah, what a world. Yeah.
Speaker 2What a world indeed. But now, let's cut back to the beginning of the story, which is, and then it ships and people hate it? Yeah, well... Like, it sounds like the announcers liked it, like it was a useful tool, it was a foil, right?
Speaker 3But there, you know, that Payne Stewart award is an award in traditions of golf. So, Nota put his neck out. Nota Begay put his neck out to do this.
Speaker 2Sure.
Speaker 3But, you know, once... You know, it's like almost anything AI. And, you know, I'm not a trendy AI guy, right? You guys know this. I've been building AI apps for almost 20 years. This is just a tool. I keep telling people, this is just a tool. We're not replacing broadcasters. It's just like the cancer tools we used to build at Internology, you know, that looked at an x-ray.
Speaker 2Sure. I mean, this could have been a person reading a screen, too.
Speaker 1Yeah.
Speaker 3Yeah. So the funny thing is golf has a tradition of breaking tradition. Interesting. Like over the last hundred years, I read this impassioned speech by one of the founders of this company about all the things that happened in golf over the last hundred years. One of them is Callaway came out with the big Burford driver and it was this huge thing. And everyone said that would be a fad. And everyone hated it because they said it was cheating.
Speaker 1Well, guess what?
Speaker 3We use that. Every driver is huge right now, right? Or they went from persimmon to steel shafts back in the 30s or 40s. And everyone said that was cheating. And one of the golfers, who was it? Sam Snead or somebody like that. invented the pitching wedge. Believe it or not, they were playing golf without a pitching wedge or a sand wedge, which is the.
Speaker 2Really- The really high angle, yeah.
Speaker 3Right? And everyone said that was cheating. Well, now those are the dominant clubs in a professional golfer's bag.
Speaker 1It wasn't breaking tradition. Wasn't it Tiger Woods who broke tradition over his wife's head, I think? Was he the guy?
Speaker 2I don't remember.
Speaker 3He's not in the news anymore. Phil Mickelson is for all- For very similar reasons.
Speaker 2Yeah.
Speaker 3Yeah. So, it's just, you know, humans are humans. And, you know, there's so much hate on the internet. Some of it is really amusing. The ignorance is spectacular.
Speaker 2Right. But you wonder if this does this stick.
Speaker 3Well, if you're asking me to predict, not only golf, but, you know, this is going to stick throughout sports. We already have AI actors and actresses, which I'm not so I'm not really pop culture, so I don't really feel like I have expertise to weigh in.
Speaker 2Yeah, I guess the question is, does the interaction get better? Like, do people enjoy the show more because they have access to more information about the game?
Speaker 3Well, if history proves true, then this will become the predominant tech in a golf broadcast. By the way, we don't just do the avatar. All that swing technology with the really high-speed cameras that analyze the swing, that's all part of it. Um, one of the tricky things in golf is doing the, you know, pretend like there's a, um, a drone following the ball above the golfer.
Speaker 1Yeah.
Speaker 3Well with AI, you know, you could go all the way around the golfer on the course where it appears that way.
Speaker 2Right.
Speaker 3In AI and analyze the swing from a 360 view. So we do a lot of that super high-tech stuff too.
Speaker 1Speaking of that, I was at a football game on Sunday in Philadelphia, and I noticed Go Eagles! Yeah, I noticed there was a drone, but it was not making noise and it was just sitting there. And it turns out that it's connected to four different cables at You know, the four corners of the stadium.
Speaker 2That's a normal football technology.
Speaker 1Yeah, I had no idea because I'm not a.
Speaker 3That's prevalent these days.
Speaker 2Yeah, yeah.
Speaker 1And guess what? The software to control it was written in Delphi.
Speaker 2Hilarious.
Speaker 3Oh, that's awesome.
Speaker 2You know it's been around for a while.
Speaker 3That's awesome. Did you tell Lino and... And all our other buddies that work on Delphi. Yeah, right.
Speaker 1Mark Miller.
Speaker 2But when they use that camera for a kickoff and they chase the ball, like that effect is so powerful. And I was just reading that. They're going to use this 3D modeling software to be able to do a follow cam on a golf ball. Wow. So they simulate as if a drone could fly fast enough and precise enough to chase behind a golf ball in flight.
Speaker 3Which brings me to another resistance of golf. When they first broadcast television live in golf, everyone hated it because the cameras were so huge.
Speaker 2Right.
Speaker 3That they basically had to stay at the first hole.
Speaker 2Yeah.
Speaker 3It's useless. They couldn't follow the ball in flight or any of this stuff you're talking about now. Yeah. Yeah. So, I mean, there's an argument that the demographic that isn't us- loves this stuff yeah just loves it and it's the get off my lawn old guy demographic that hates this new fangled yeah new fangled.
Speaker 2Yeah that moment it's like we didn't build it for you like part of this is keeping the sport relevant to a different generation.
Speaker 1I have a relevant story here it's quick um no go for it I have a friend who I met in junior high school and now he lives like literally around the corner from me okay And he's a musician. Well, not by trade, but he's an amateur musician. And he wrote a bunch of songs and uploaded demos to some AI. And I don't know which one it was. But basically, it was something where he just took him playing a guitar and singing, uploaded it, and told, with prompts, the
Speaker 1AI to create polished, production-ready songs with singers and guitar riffs and bass and drums and steel guitar and all this stuff. He comes to one of my gigs and he says, I've got these songs and they sound good, but I want to redo the whole album with real musicians. At first, I'm thinking, that's going to be very expensive.
Speaker 1He's not a rich guy. He's not going to drop 20 grand to make an album, which is probably what it would cost from my time and musicians and everything. So, and his reasoning, the reason that he didn't want to just publish what he had, which by the way, sounds pretty freaking good. It sounds good, but the kickback on the, you know, as soon as anybody knows that it's AI, They don't want anything to do with it, even if it's actually good.
Speaker 3Yeah.
Speaker 1Isn't that interesting?
Speaker 3It is. I'm not so sure media has helped in this anti-AI thing, you know, because AI is still saving lives.
Speaker 2And the industry hasn't helped itself. This marketing strategy through catastrophization is not helping.
Speaker 3But you mentioned the money thing, and I think that's important to point out in this. This was just like a proof of concept to do it at a PGA sectional tournament, not like the Masters, not like the US Open. And before the PGA picks it up full-time in the big tournaments, Really, if you think it through, the right way to do this is on the edge with $ 2 million of NVIDIA hardware.
Speaker 1Right.
Speaker 2Right.
Speaker 3You know, as opposed to that trip through the satellite dish up to the internet back. Because there's latencies. When you do that, remember we're talking about the laws of physics earlier before we came on here, you know, the avatar, the avatar tech can be lightning quick. If it has that I built, if it has a really nice internet connection.
Speaker 2Mm-hmm.
Speaker 3If not, there's that lag. It's almost like that CNN.
Speaker 2Yeah, yeah. Well, and I was wondering, for broadcast, like, were you really depending on the internet? Shouldn't you try to go all local model?
Speaker 3Exactly. Yeah.
Speaker 1Sounds like you have a local model.
Speaker 3That's my– well, yeah, just the model. But the future of this thing, to do it right, is on the edge. And, you know, we talked about that before we came on air, too.
Speaker 2Mm-hmm.
Speaker 3It's just going to take a lot of money of Nvidia to pull that off.
Speaker 2Right. I think it'll get cheaper too.
Speaker 3You know, and and Nvidia is lined up as a partner of ours. They don't typically give away $ 2 million for free, but, um, you know, that, that's the right way to do this. Um, uh, it just, this is new for production people too. So the whole way they set it up, and of course they wouldn't listen to me or us with the Nova Scotia to, uh, Florida back to New Mexico thing. It just didn't make sense at all, you know, and then stitching it all together, you know, God knows where they did that and then
Speaker 3sending it back to the feed. Um, so, you know, there's, there's a lot of room for improvement, you know, but, but the funny thing was that the software ran great.
Speaker 2Yeah.
Speaker 3Like, like a hundred percent uptime in an interpreted language. Who knew? Yeah.
Speaker 2Yeah, that wasn't the limiting factor.
Speaker 3Right.
Speaker 2But one would argue if you had more time, there is more optimal code approaches.
Speaker 3Oh, God, yeah. There's so many. The feature backlog list I've built for this thing. You know, I've got to obviously add real developers to this company in short order.
Speaker 2Yeah.
Speaker 3And the feature bank.
Speaker 2I'm used to you building prototypes, just not by yourself, Tim. Like that's weird.
Speaker 3I wasn't by myself. I'm like a Microsoft commercial. I wasn't by myself. I had M365 Copilot.
Speaker 1So, uh, seems like we're coming to the pointy end of the story. Is there anything else that you wanted to say before we hang it up?
Speaker 3Well, you know, I think it, As I look back on the way it went down, it was just so fun to be a developer again. It's still such a great occupation. And I just, you know, the world's telling us that developers are going to be eliminated. I just don't see it. I just don't see it. This thing is intricate. intricate list of steps to pull this off
Speaker 3you know it's not just the avatar there's the orchestrator which is live kit you know that that orchestrates the whole thing it's not a one-to-one you could run 200 these avatars at a time which which means you have to build enterprise features right for it um The text-to-speech, speech-to-text engine, I chose Cartesia just because I couldn't find a better voice to match. But this is all built so you could swap in, in and out tech. It's
Speaker 3built like enterprise software. And the takeaway should not be for me that, hey, I built this in six weeks. It's that You really still have to pay attention to building enterprise-level software.
Speaker 2Yep. Sure.
Speaker 3You cannot hack this type of stuff and expect it not to fail. Right. Yeah, which was honestly super fun and very challenging.
Speaker 2This also speaks to this is an immature market. There's so many products. Like, everybody's experimenting.
Speaker 1Yeah.
Speaker 2You've got to kind of assemble from the tool suite. There's no clear path forward on any of this yet.
Speaker 3Yeah, yeah, yeah.
Speaker 1And also, there's a whole rash of new companies that are– that are AI on the, you know, in the backend. And it seems to me just run by one or two people, you know, that promise all these great things. And the ads for these companies are showing up on TV and streaming platforms.
Speaker 3But don't you feel like in the 20 or 30 years we've been doing this, that we still have the same problem that the Silicon Valley and other venture capital or financing companies have? are still throwing a lot of money at stuff they shouldn't. Yeah. Don't you feel like that?
Speaker 2Mm-hmm.
Speaker 3Yeah. Some of these companies that get funded for these gazillions of dollars, I scratch my head like, what?
Speaker 1Right. It's an outdated model.
Speaker 3It is.
Speaker 2It is. Yeah. And it's definitely this new wave has shown how outdated the model really is.
Speaker 3Yeah.
Speaker 2It makes me wonder what the next one's going to look like, just because I don't think we're going to do it this way again.
Speaker 3Hysteria.
Speaker 2Yeah. This crazy hype is not useful.
Speaker 3Yeah.
Speaker 2Well, that and Microsoft naming everything Copilot.
Speaker 1Oh, God.
Speaker 2Can we stop with that? How about that? That would be a good one to not do again.
Speaker 3How many times have they done that, though?
Speaker 1Everything's Copilot.
Speaker 2But this time, I think it really burnt them. Like that's the only thing I would say is like more.
Speaker 3Well, Carl and I during.
Speaker 1Yeah. Look at the confusion. Right.
Speaker 2Totally. A hundred percent. That's why I bring it up.
Speaker 1I heard them three 65 copilot earlier in the show. And then you just started saying copilot copilot. And in my head, I'm thinking the CLI. Yeah.
Speaker 3You know, and if you really stay on the Microsoft stack, not that I totally do anymore, but all the copilots, you know, are different. Like as your copilot is terrible, in my opinion, in my personal opinion, If you're in the Azure portal and use that copilot thing, it's useless.
Speaker 2Yeah, yeah, it's bad.
Speaker 3It's absolutely, and now I'm talking.
Speaker 2The SRE copilot is impressive, impressive enough they've stopped calling it a copilot. Like they're now called the SRE agent. Like you can see the different teams are abandoning the brand because it's so tainted.
Speaker 3The Excel copilot is.
Speaker 2It's amazing. It's astonishing.
Speaker 3Awesome.
Speaker 1It is.
Speaker 3I will never hack a spreadsheet again.
Speaker 2I think it's Claude Cowork under the hood, which is why it's so good. And once again, this is why the name is so bad.
Speaker 3What a world we live in.
Speaker 2What a world, friend. What a world.
Speaker 1My wife uses Windows 11 Copilot.
Speaker 3I don't even know what that is.
Speaker 1It's just an agent that's built into Windows that can help you with stuff.
Speaker 3Like, I can't print? I can't print. The internet is down? Is that what you give it?
Speaker 1She uses it to extract information from PDF files and turn them into Excel spreadsheets, that.
Speaker 2Kind of thing.
Speaker 3Fair enough.
Speaker 1Yeah.
Speaker 3So. Say hi to Kelly. Yeah.
Speaker 1And, and to your Kelly as well. Thank you very much, Tim Huckabee from.
Speaker 3Super fun boys.
Speaker 1Go for whole productions now.
Speaker 3Yeah.
Speaker 1Good luck in the future.
Speaker 3Yeah. The next one should be interesting.
Speaker 2Hockey.
Speaker 3Okay. Right?
Speaker 1We still have to build that Star Trek holodeck.
Speaker 2There's that.
Speaker 3There's that, too, that we've been talking about for 30 years. Yep. All right, man.
Speaker 1Thanks very much.
Speaker 3Later, boys.
Speaker 1Always a pleasure.
Speaker 2Me, too.
Speaker 1And we'll talk to you, dear listener, next time on.
Speaker 2NET Rocks!
Speaker 1NET Rocks!. NET Rocks is brought to you by Franklin's Net and produced by Plop Studios, a full-service audio, video, and post-production facility located physically in New London, Connecticut, and,
Speaker 1of course, in the cloud. Online at pwop.com. Visit our website at dotnetrocks.com for RSS feeds, downloads, mobile apps, comments, and access to the full archives going back to show number one. recorded in September 2002. And make sure you check out our sponsors. They keep us in business. Now go write some code. See you next time.
How does ASP.NET Core help you make secure web apps? Carl and Richard talk to Anders Abel about the OWASP Top Ten and how ASP.NET Core addresses many of the vulnerability elements - at least where it can. Anders talks...
What is mentorship like today? Carl and Richard talk to Ethan Winer about his experiences helping people grow in their careers. Ethan talks about growing Crescent Software back in the QuickBasic and Visual Basic days -...
Whatever happened to the future we were promised? Time for a Geek Out! To celebrate the end of the opening history bits, Richard pulls together all the most popular topics of "things we were supposed to get in the...
It's Tray-Tober! Carl and Richard talk to Joe Finney about his latest project - building small, useful Windows apps that live in the system tray. Joe talks about his original project inspiration, Radio Tray - a way to...
OpenSilver evolves! Carl and Richard talk to Giovanni Albani about how OpenSilver continues to help folks migrate Silverlight apps into a sustainable model not dependent on the plug-in - but now the focus is on WPF!...