The Year AI Became Militarized: Shelly Palmer on Government, Defense, and $3 Trillion Stacked

Hey, everybody. We just wrapped an incredible episode of the AIXR podcast with the great Shelly Palmer. He is a consultant to all of the big media companies, knows more about technology than I will ever learn. And he really holds forth and gives a point of view about AI, what's happened in the past year, and what's coming up. This episode of the AIXR podcast is brought to you by Zapper, the folks behind MatterCraft, the leading visual development environment for building immersive 3D web experiences for mobile headsets and desktop. MatterCraft combines the power of a game engine with the flexibility of the web and now features an AI assistant that helps you design, code, and debug in real time right in your browser. Whether you're a dev, designer, or just getting started, MatterCraft speeds up your workflow and brings your 3D ideas to life faster than ever. Start building smarter at mattercraft.io. Good morning, everybody. I'm Charlie Fink with Ted Chilowitz and Rony Abovitz for the AIXR podcast. Today is Friday, December 26, 2025, our last show of the year. Good morning, guys. Morning, gentlemen. Morning, guys. So how was your Christmas? Good so far. I'm in Utah. Unfortunately, it's a light snow year, but we escaped the rain in LA, so we're just having fun with friends up in Utah. Yeah, it's been raining here for several days. It always rains here at Christmas. Well, we were hoping that rain would have turned into snow in Utah, but it didn't, unfortunately. I think it's snowing. I'll be back for Sundance is the end of January. Well, down here in Florida, we're suffering from absolutely perfect blue skies, 70 degree Fahrenheit, just perfect weather. I don't know what to say about it. It's horrible. So this is the show where we do the year 2025 in review. And we're being joined by Shelley Palmer, who is always comes on. We're trying to make a tradition. I guess this is the second or third year, I guess, he's come on at the end of the year and helped us recap. Shelley is, for those of you who don't know who Shelley is, I guess it's kind of a he's a celebrity to those of us inside of the entertainment and technology industries. industries. He consults with all of the big media companies on what's happening with technology and generally with markets and a super insightful guy. I learned something every time I talked to him. So I don't think we would call him a pundit. I think that's the word. He's a pundit. Well, he's a guy like you guys who knows a lot about a lot. so Shelley has been you know on the cutting edge of digital media since desktop publishing in fact I think he started out in the music business writing jingles so not I think he wrote a very famous jingle for CBS News I think I can remember he told us that one year that he wrote the like the CBS News theme that we've heard our whole lives I think he wrote that pretty sure I can confirm it today when he comes on. So anyway, it's a great opportunity for everybody to hear from Shelly. He's got, by the way, a fantastic daily news email that I would highly recommend people sign up for. It obviously does a little news recap and Shelly's unique editorial point of view is all over it. So we'll look forward to getting some of that in person. Wasn't a big news week this week. Obviously, it's the Christmas week. So mostly we're going to talk today about 2025, which is in our rearview mirror. We were just talking about stock rallies at the end of the year, the so-called Santa Claus rally. We were also talking about people making moves like my sale of my Snap stock so I can harvest some losses and then I'm going to rebuy it after January 1st. There was an interesting article in the LA Times this week. I mean, Snap has a billion, a billion users and they're not profitable, which is kind of shocking. Here is Shelly Palmer. Great to see you, Shelly. Hey, guys. Happy Christmas, everybody. It's all that. How are you guys doing? Hey, Shelly, before we even get started, we need to confirm, because I think you told us one year that you wrote the iconic CBS News theme. Is that correct, the music for CBS News? Depends on what you mean by iconic. That's it. That's iconic. I did about 9,000 arrangements of that, yeah. it's actually in the universe called the Palmer News Package, not the CBS News thing, but it's based on something that was written for the Chicago affiliate. And it literally has over a thousand cuts of audio. It just, it's on all over the place. And so much fun doing that over the years. I did a lot of news themes. I did one called Brave New World. I did AP's news theme. I wrote the original music for TNT when it went on the air, the sounder package for that. AP Radio. Well, if you've got any spare time on your hands and you want to write one for the AIXR podcast, we'd be honored. Or maybe you just got one sitting in the closet. In your copious free time, could you please write a free theme song for us? You know, there is something on the order of 5,000 pieces of music in my ASCAP catalog. And in the piano bench, and Ted, this is probably not any exaggeration. There's probably another 3,000 just tone poems and tone fragments and things that never became songs that were just, I write all day. Like, there's stuff going on in my head all the time. And every once in a while, I sit down and actually write it down. God knows what's going on in this ridiculous head of mine when I'm not writing stuff down. But every once in a while, you kind of go, wow, that's good. I should write that down. So I don't forget it. Well, we'll take anyone you want to give us. Oh, thanks. Yeah, you know, look, mom and dad met at Juilliard. They own music stores. And they were both music educators. They taught in the New York City school system as music educators. And then after I was born, their teacher salaries did not support childcare, like they couldn't afford childcare. So they have to figure out how to make extra money. And what they did was they got together with a couple of their friends. And they opened a little music studio where they were teaching after school. So they teach all day till three o'clock and then they'd teach after and then my dad would play gigs nights and weekends. But back then it was mostly Friday, Saturday nights and a Sunday wedding or two, right? So, you know, he made a little supplemental income there. Then the music stores did okay. The music store did okay. And he bought out his partners. And they ended up buying a couple more retail stores. I think at the height of his thing, he had like five or six retail stores and a mail order business, which was interesting, and some manufacturing. My brother still runs, such that it is, the U.S. regulation bugle business, which my dad bought years and years and years and years ago. Really? So it's a small little thing. It's basically military and Boy Scout bugles, if you will. You know, and whoever, who needs a bugle nowadays? But yeah, so anyway, there was a lot of music in my household growing up. That's an adorable story about the bugle store. It reminds me of there was this classic SNL skit. I don't even think you can find it anymore online. About something called the Scotch House. And all they sold was Scotch tape. Remember that? I know, that was a Saturday Night Live sketch from 40 years ago. people would go in and they'd be like do you have duct tape no we only have scotch tape masking tape no it was great it's no no yeah no just bugles it's true though that's i mean it's it's a wholesale i think he said it's genius he sells he sells on amazon and online you know i love it it'd be funny if there was a physical store but like i think there are two skus you can get a brass bugle or a silver bugle which is really a brass bugle that's silver plated that's pretty much that okay you can buy the bag or the mouth with ai i think with ai you should go ahead and make him the physical bugle store and put it out that'd be funny and yeah fondly remember especially this time of year it's fun to think back on yeah it's great on my parents and what they what they did but yeah a lot of music i wrote i and i still do i mean for the first 40 years of my life pretty much all i did was write music you know every day and made a living out of it. And then the tech thing got really out of hand. I mean, I just, I started, I learned my technology by learning to be a better producer, just being more efficient in the production tools. And then one thing led to another, sold a few patents around 2000 and actually earlier than that, 95, 97, 2000. And then I just got super involved. The big transition was when Disney bought one of my patents and they're a pretty serious organization when it comes to technology and they were going to reduce it to practice. That was enhanced television at ABC. And that patent, I didn't want to sell it to them unless they were really going to use it because I was thinking about doing a thing that it teaches this concept that ended up being named OCAP, the Open Cable Application Protocol, which was a way to turn set-top boxes and servers into client server networks for data. And it predates the internet. It's a 1993 patent that predates the World Wide Web. It doesn't predate the internet. And it was fascinating that they wanted to use this technology. It allowed you to basically play Who Wants to Be a Millionaire on your desktop computer or your laptop while watching the show in real time. It powered a lot more, but that was the overlap that Disney was interested in. They had Celebrity Mole and they had Who Wants to Be a Millionaire and they had Monday Night Football, the fantasy football, and they wanted to involve people with the technology. And once they agreed to let me consult, whatever that means, I got a three-year contract and I leased them and ultimately sold them the patent. And what an amazing opportunity to learn from some of the just best engineers in the world and brilliant executives. And it really was a transition for me from something I genuinely knew how to do my whole life, which is write and produce music, to stuff I do nothing about, which is like how to reduce a patent to practice. That was fascinating. So I learned an awful lot during that time, watching the engineers do their thing. And I was pretty good with the soldering iron. I'm not going to say I wasn't, but it's a different level, you know, what I was doing, what they were doing. So, yeah, that was the real start, like, 25 years ago of the transition. Anyway, trip to memory lane for no reason. So it's 10 days-ish before our annual trek to CES. What are you excited about? What are you looking forward to? What's your take on what AI's presence will be at the show? Well, wait, wait. This is the year-in-review show. This is not the preview of the 2026 show. Well, let him do a little bit of predictive analysis. Okay. All right. Let's go for it. Look, last year, you could smell the paint drying on the letters AI on every exhibit. Like they had just, everybody just painted it on the exhibit. Hi, we got AI. It's like, there was AI toothpicks last year. There was AI toaster ovens. Like everything had AI on it. This year, it'll be worse. And we're going to be drowning. My prediction is we're going to be drowning in knockoff AI smart glasses. Drowning. Look, at the end of the day, no predictions required. This is really easy to understand. This is the year that AI is deployed. We are going to see everybody. There were two schools of thought, really. There was, you can't make a living wrapping a foundational model because you're going to lose to open AI. You're going to lose to Anthropic. You're going to lose to Gemini. Then there was a school of thought that said you could. You could build a wrapper and you're going to do fine. Now, as it turned out, everybody was right. There are some people who have been able to create vertical subject matter expertise where they have built, they've done post-training and they've also done embeddings and created databases that are subject matter expert. They've created evals. They've created entire workflows that are self-improving workflows inside subject matter expertise. Some people have gone a different way. they put together groups of different models, small language models and large language models, mixture of experts, technology, techniques. So we've seen a lot of companies that have done both. We've seen some that have become vertical experts in either medicine or research or protein folding or something, marketing, what have you. And there are people who've done some wrappers and they've done them in a sloppy way and they're going to go out of business. They're not, they're parlor tricks. this year is going to separate the miniature golf from big boy golf right we're going to know who's on the pga tour metaphorically and who is basically playing goofy golf so i think we'll see that at ces in a way that i don't believe people are prepared for because i i don't think you can pose it's because you peel away one layer of the onion and it's instantly obvious something bad is happening So I think we'll see that at CES. And I'm excited. Also, CES has added a couple of things I know nothing about yet. NVIDIA is sponsoring these two days, Thursday and Friday, that are going to be all AI at the Fountain Blue. So they're giving free AI training opportunities to people and stuff. And I don't really know what that's going to be yet. I've read a little bit about it. And I've talked to the people at CTA about it. They're all super excited. Jury's out. We'll see. It's first year. First year, you'd ever know. But how exciting that they're going to have two whole days dedicated to AI. So I think that's a big plus. And look, the guys at CTA, you've got Kinsey and John Kelly. Gary's going to be on the show next week. Look, they've done a good job of adapting. They've done a good job of trying to stay relevant. And the world has so dramatically changed. That is a very big deal. Samsung's not going to be on the floor. They're taking, I think, 120,000 square feet at the win. So that's going to be interesting. That's big. Well, they've never not been in Central Hall, I mean, in the recent memory of Man or Beast. And Sony is not bringing any electronics. They're simply bringing the Honda Afila car. They're not bringing cameras or TVs or any other electronics or experiences. So we're going to see a very different kind of CES in Central and North Hall. They have a lot of robotics in North. We've got a fair amount of Chinese companies, but we don't know what the climate is for the Chinese companies based on the current administration. And so we don't know if they're going to be there selling or just displaying or haven't really got a deep handle on that. TCL will be there for sure. I mean, they're all, they're places on the floor on the maps, but last year was a kind of a different year. You know, they were under a little bit of pressure last year. There's a lot more pressure this year, tariffs being what they are. So I don't know, I'm not going to get into politics, but whatever those politics are will have an impact on the way that CES presents this year. Then you've got C-Space. They've done a really nice job there. We're going to do our Shelley Palmer Innovation Series Breakfast presented by 3C Ventures and the Association of National Advertisers Wednesday morning as an official CES event This is the first time we been an official CES event in probably a decade We been kind of independent with the breakfast This will be the 30th year of the breakfast. So it's sort of always been my favorite part of CES because it's the thing that we have control over. CES has always been very kind and gracious to us about it. We were We're taking the Encore Theater and, you know, they were, you know, on a Wednesday morning. They were very supportive, although it wasn't an official event. This year, we're going to be in Mariposa. We'll be at the Aria. It's a smaller room. I think it only holds four or five hundred people. So it'll be a harder ticket to get to. Really nice guest list. Shakir Moen from Coca-Cola. And we've got Susie Deering coming in and a bunch of other really, really extraordinary guests. Michael Kasson will have a minute. and I'll do a keynote like I normally do. It'll be fun. And then we're doing, for our biggest clients, our private tours, as we always do, of the floor. We're not doing publicly available tours this year. We didn't do them for the last five years. We never sort of went back after COVID. We never went back to public touring. It's just our private clients. So I'm really looking forward to CES. And as I always do, it's from the Palmer Group's perspective, it's our Super Bowl. I've got a really nice opportunity to go on stage for Omnicom. They've invited me to go on stage and do a 20-minute little mini presentation about the way I'm seeing AI. And that's always a fantastic audience. There are a couple of dinners that I'm looking forward to that I will not promote because they're not mine. And I don't know that the people are that excited to have other people know they're having them. But there's a couple of not-so-super-secret handshake dinners with marketers and tech executives that I'll be attending. I'm always looking forward to CES. By the way, we've got some big choices now on the marketing and tech side, right? I mean, CES is certainly the kickoff of the year, but possible in April. Very important conference. It's turned out to be really something special. and of course can has taken on a completely new like i don't know that's not the film festival but but it is no marketing festival that goes on next to it yeah can lions it's it's really something so those are the three big ones i super look forward to and ces is always you you don't get a great christmas new year's vacation free ces a little busy but uh i'm very much looking forward to it So yeah, at 10, it's going to be all AI all the time. This is going to be my 35th CES. Get out of here, really? Wow. Do you know at 25, I tell this story too much. At 25, they give you a little special badge. It's like 25th anniversary or whatever. But then everybody you talk to wants to talk about your 25th anniversary, which I am not there to talk about me. I'm there to learn things and let other people talk. So eventually I went after like a day and said, you got to give me a normal badge. This is really embarrassing. Also, you know, I mean, how old and how much of a nerd do you have to be to actually get one of those? So both you and I used to go to CES when it was comedy. Yes. And that's all anyone needs to know. It's just that simple. It was one hall in the convention center, which was much smaller than anyway, and then a tent out back with guys selling porn videos out of cardboard boxes, which I guess now they have their own whole thing in Las Vegas. set four years as everybody knows the adult video news awards the avian awards took place saturday night and at the venetian and walking through the lobby of the venetian as that pregame was going on was was one of the funnest things you could do just it was just hysterical i mean they just came dressed they themed dressed for their award show yeah all of them oh yeah yeah and it was it was it was very funny look ces is a long history it vegas has changed so much oh my gosh i mean it's a different it's a different city um it used to be run and i would say like i don't know how other people are going to feel about this i probably kind of get in trouble saying it when the mob ran las vegas i liked i liked las vegas better it was the safest place on earth they knew how to take care of high rollers they knew how to take care of entertainers and the the restaurants that were at the casino level did not have individual pnls and therefore they were there for the guests they understood if they fed you and watered you if you were properly fed and watered you would sit at the tables all night it was it was i don't know kinder gentler mob run world okay so that's that's a preview of I would also add better comedians. I would also add You know what? You should start Mava. Make Vegas great again. Or Mavka. Make Vegas Mavka. Make Vegas Mavka. Yeah. Shelly, you might be getting calls from people you may not want to get calls from after this podcast. Again, it was you literally I'll never forget this. I was at, I won't mention the casino. I was in a poker game. That was a, because I'm in the old days, before video poker, you actually could sit down at a poker table in Vegas and not be faced off with a 24-year-old with a still plastic brain who'd seen 5 million hands of poker from playing video poker and could just wipe the table with you because the skill level and pattern matching was unbeatable. But back in the day when you were playing people who were just human, who had not ever had video poker experience, you could actually have a nice night and win some money. And it was commonplace, commonplace, to leave your chips right where they were, get up, go smoke a cigar with somebody, take a break, come back, nothing would be gone. like you were it was the safest place in the universe to do anything okay and you could walk on the streets at any time of the day or night never fearing anything ever regardless of what the police were doing that you were in a different place it was a different vibe it just was and look i i don't want to harp on the old days vegas is what vegas is now but i'm just saying if you were asking me like which vegas i prefer if we're reflecting on memory lane it's like they definitely Medicare business. So obviously you'll have no time to attend. No, I was going to plug on the AIXR podcast dinner, which is eight years old, although it's our sixth annual because we missed two years because of the pandemic. So Ted and I were doing this dinner before there was a podcast. We did it twice before we started the podcast during the pandemic. So if you want an invite to that very exclusive podcast party, shoot me an email. Hit me up on the socials. I'm the easiest person to find there in the world. So I'd be happy to shoot you an invite. If you're a listener, we would love to meet you in person. So let's pivot now to 2025 in our rearview mirror. I'm sorry, Ted, go ahead before we get started. I was just going to tell Shelly, if you want to get yourself a throwback Vegas moment, go see Wayne Newton at the Flamingo. Wayne Newton, not a young man, still performing. and you would think, let me tell you something, you would think this is not something you want to see. I guarantee you, you will be charmed and amazed and wondered. It's in a tiny little showroom at the Flamingo and it is absolutely worth seeing. It will take you all the way back to the heyday of Vegas. I cannot stress it hard enough. You, because of your opinion of Vegas and the old Vegas and what you want out of Vegas, we should all actually go. We should go. I bid with a very close friend of mine. It was a magical experience. We should go. Trust me. So Wayne Newton figured prominently in that Albert Brooks movie Lost in America from 1985. So just to give you a sense of Wayne Newton, he was a big Vegas star in 1985. That's 40 years ago. Wait, wait. Hang on a second. Hang on. He has an iconic moment in Vegas Vacation, Chevy Chase's Vegas Vacation. iconic that he talks about in his show. Which I just saw, by the way. But by the way, the funniest thing I ever saw in Vegas, I was only there a couple times, unlike you guys. During MAKO, we had a huge conference there, one of the ortho conferences, full of orthopedic surgery nerds. And I just wandered over to, I think, the Flamingo and got to see Jackie Mason by accident. So if anyone, one of the classic, extremely rude, every kind of bad word you can imagine he was like the funniest two three hours of my entire life i think i just was on the floor probably unable to not like just like convulse the whole time he was like it was like the greatest it was like an old school that's what you were saying shelly the old school vegas you know kind of a grungy you know grungy hotel at the time jackie's like you know kind of already peaked it's like uh you know this was like in the mid mid to late 90s yeah any anything in Vegas that he did is post his peak. He was a Borscht Belt comedian. I, fun fact, opened for Jackie when I was music directing. You opened for Jackie? A wonderfully young female singer who was his opening act, and I music directed those performances. is um he actually had a very he he taught me one thing that or gave me a line that i really feel like is uh maybe the most important thing i've ever taken away as a stage performer or like you know guys i you know i do like 60 speaking engagements a year it's like a more than one a week um so he's we're we are in the poconos at a the beautiful mount airy lodge or whatever they called it back then and it's two shows a night there's an eight o'clock and there's an 11 o'clock and we open we're the opening act for both so of course you sit through the whole thing because you do it's no you don't just leave so the four nights in a row where it was a five night gig we're there through the fourth night every set was precisely and exactly the same from him right down to the beat right down to the timing it was stunning to me how accurately he replicated his performance so we're at the pool him and i daytime saturday because our gig ends sunday night and I won't say we're friendly, but we are quite, you know, we're talking because we're sort of doing the same thing and we're sitting there ordering whatever drinks we're ordering. And I said, Jackie, I got to say, I'm so impressed. I'm actually stunned by the accuracy and the quality, everything I just told you. And he takes a pregnant pause, a good theatrical beat. and in that in the in the thickest jackie mason accent he could possibly muster which i will not attempt to do now because hey no i'll do it well but nobody will understand what i'm going to say he said he looks me straight in the eye he goes shelly it is way easier to find a new audience than new material so truer words have never been said in the history of the world that's really funny that's really brilliant and i i've never forgotten i will never forget it i and every time i use the same joke in a speech or every time i reuse a slide i just think to that one little moment by the pool and jackie's sipping a mai tai and he's well it's what much easier i just exactly right it is much easier to find a new audience the new material may he rest in peace he was an amazing amazing performer so what do you want to talk about 20 yeah so tell us tell us What surprised you about 2025? I think what surprised me the most about 2025 was everyone's willingness to say they were playing with AI as opposed to their willingness to roll up their sleeves and put hands on keyboard. People were posing and talking more than they were doing, or they were dabbling without going deep and talking as if they'd gone deep because they didn't understand the difference between using it as a parlor trick and using it to shift a paradigm. and this was so pervasive that it actually sat me back in my chair this year. Everyone was talking like they were expert at it when they had done nothing or they would have a deep opinion about it. And when you peel away one layer of the onion, not two, just one, you realized they were completely full of it. Like they had literally done nothing. And I'm not indicting any one person. I'm indicting like every executive. There are people that would sit in meetings and shake their head and smile as if they knew. And then you would ask a question and you'd get a like, no, no answer. And I got to the point where I stopped asking questions in public forums and I would carve someone away from the pack and say, hey, you know, you said something earlier. I just want to go over that with you. And it would be clear. They had no idea what they were talking about. And this is the first time in my professional career where that can get ratted out by somebody asking an innocent question. Not even a technical question, an innocent question. And also, I've never seen this kind of nonsense this pervasive. And it's really bad. Because people are ascribing all kinds of capabilities. They're also willing to condemn the technology. Well, it hallucinates. Well, it does this. It's bad for society. It's going to kill everyone. Parroting back nonsense. And it is the parroting back of nonsense as opposed to the, hey, you know what, guys? All you've got to do is sit down and roll up your sleeves. And if you don't know what you're doing, ask the model. Just ask it. Hey, I want to do this. How do you do it? And you're going to get somewhere way, way farther along than reading somebody's blog post or listening to some YouTuber who knows nothing tell you stuff that they're like copying from another YouTuber who knows nothing. The game of telephone stunned me this year. And I've never thought, look, I've been doing this 45 years. I've never seen anything like it. To be fair, Charlie, I just never have. Have you seen anything on the other side, like breakout successes where the opposite has happened to you, where you were going in thinking these guys are all smoke and mirrors and it turns out they really have it going on. They figured it out. They've start to use these tools in the right way and they're going to be successful businesses. Have you seen the other side of that coin? Can you give us some examples of who's doing it right? Because of the NDAs I'm under, I cannot call out the companies by name, but I can tell you with absolute surety that several of my clients are exactly the opposite of that They are rolled up sleeves hands on keyboard Here the recipe for success goes like this First, leadership believes. The actual senior leaders believe this is real and they use the technology. When the leaders use the technology, everybody else does too. You get a senior leader walking into a leadership meeting, into an LT meeting, and they say, hey, I made this app over the weekend, or I built this agent that does this workflow. And everybody in the room goes, oh my God, if the boss can do it, I have to do it. So that leading from the front is the first thing that every one of them has in common. The second thing they have in common is that not only do they believe they put their money where their mouth is, there's Tech Tuesdays or Tech Thursdays, where there's an hour or two every Tuesday or Thursday, and it's, hey, everybody gets, The winner gets a Starbucks card. We're going to automate this workflow or we're going to rethink this workflow or we're going to come up with some blue sky. We're going to have a blue sky meeting and try to do something where AI will amplify our skills or our capabilities. They literally do blue sky meetings. They create PRDs and they go ahead and they try to make stuff happen and they test, fail and learn. But they're doing it on paid time where senior leadership says, we believe in this so much. We're going to pay you to sit at your desk and do this. And the third recipe for success is they figure out that this is a cross-discipline issue, and they all know that SecOps has to be involved, IT needs to be involved, risk and compliance needs to be involved, the legal team, the general counsel's office is going to be involved, and they're paving the way. like a new model comes out and instead of having to go through like 14 weeks of testing by every different group, they have some statement of principles or manifesto in place where this can be reduced to practice relatively quickly and they can learn from the mistakes of others. And they don't react to the advent of new technology, which happens every morning. They've built a mechanism to respond. And when you have those things in place, you are seeing real success and real progress and honestly where they're not in place you see chaos absolute chaos and 300 projects that are just all over the place and everybody's fighting in it it's it's you choose your metaphor knives out or game of thrones the home game whichever one you like it's one of those if if you don't have a process you are not um if you build it everyone dies mentality clearly no no humans kill humans machines don't kill humans 100 everyone may die but everyone's gonna everyone may die but they're not going to die because the machine woke up in the morning said you're all supposed to be dead we'll fight over resources like electricity or oil or water we'll fight over capability make no mistake this is a militarized project this is the Manhattan Project of our lifetime. There's $3 trillion stacked up behind this. No one's ever seen $3 trillion stacked up behind anything. This is insane money. Some people are calling it a bubble. I'm sorry. It would be a bubble if the United States government wasn't saying out loud in writing and every way they could scream it to the rafters, we, the US, will have to beat the Chinese to AGI. Undefined as it is, artificial general intelligence is a concept that people believe they understand. And the government has made it very plain. We're beating the Chinese to artificial dental intelligence, whatever that means. Okay, that may get us killed, but that's not going to be the AI killing us. Look, if you saw at the Beijing Olympics, the Chinese put, I don't know, 20, 30, 50,000 drones in the air and did a light show. That wasn't a light show. That was them demonstrating they could put 20, 30, 50,000 drones in the air, swap them out in real time so that it looks like they're continuous, regardless of their battery life. And you know what that's for? Let me put 10,000 drones over a battlefield and eliminate the enemy combatants. Let me put 100,000 drones in the air over the Gerald R. Ford class battle group and take it out. No conventional weapon could take it out. You can get within 500 miles of that battle group with a conventional weapon, but with 100,000 drones, you'll overwhelm anything that it has to offer unless they start doing crazy EMP pulses, you know, like that are directed. I mean, we'll probably have some countermeasures for that. But at the moment, what the Chinese were saying is, you know, you guys proved you could put a man on the moon. What you're really proving is that you had intercontinental ballistic missile capability. That's what you told us with an Atlas V rocket. You could put a big payload up in the air and ballistically land it anywhere on Earth within 20 minutes. Okay, we got it. Nuclear age has begun. The age of intercontinental ballistic missiles has begun. Beijing, they announced flat out, we have the AI to put the drones in the air. We're not going to stop till We can put them everywhere. And you see what's going on in Ukraine. Like, this is militarized. So if AI is going to kill us, it's going to be through the technologies of war that we are so focused on right now. And with $3 trillion lined up to take over the world, most of which is not militarized, but it's public record. They sent a couple hundred million dollars to each of the foundational model builders from the Pentagon. What do you think that was for? It's for cyber warfare and robotics. So, Shelley, are you saying that, like, if we look back at 25, to stay with our theme, was 25 the year that AI became hardcore institutionalized in the Department of War, Department of Defense? Like, it's become the thing now versus, like, maybe a thing. I think a better way to describe it is that it became publicly understood. I don't think that anyone who is knowledgeable about the defense community and the 17 three-letter agencies could be naive enough to believe that AI wasn't a massive component. But AI itself took a turn after the advent and consumerization of generative pre-trained transformers that I don't believe anybody could have expected. To be fair, we've been talking about AI in our practice vigorously since 2013, vigorously. First neural networks went online. We're starting to use them for media mix optimization. We're starting to use them for simulations. We're starting to use them for optimizations of every kind. And then you had the early transformers like BERT, WIF or Google Translate. We started to use those to try and get more context. When in 2017, after the generative pre-trained transformer, the attention is all you need, white paper. Remember, at that time, you're coming off of the 2016 AlphaGo, AlphaZero timing, right? So that big transition was AlphaGo was trained on a few hundred thousand gameplay transcripts. They transcribe Go games the way they transcribe chess games, right? They write them all down move by move. So they taught AlphaGo how to play Go by looking at hundreds of thousands of previously played games. AlphaZero, according to Demis Asabes, was trained with rules. And then it was taught to play itself. and in about four hours learned to play better than AlphaGo. It also learned Soji in a couple hours and chess in a couple hours. In all three of those games, AlphaZero played better than anything that it ever played before. So it learned a lot in that 2016-17 timeframe. That's when attention is all you need. That white paper came out. So all of a sudden, general pre-trade Transformers become something that can be realized. We don't see Chatubiti as a public product until November 30th, 2022. that was a fairly significant time period where people in the business knew this was going on. Certainly people in the military knew it was going on. But your average consumer thought we were still in the AI winter. But what was really happening? All the data exhaust from all of the laptops and desktops, all the data exhaust from mobile, all the location data. Like all of a sudden you had these massive data sets we really never had before. And I have an old phrase I've used in every speech probably for 25 years, which is the velocity of data is increasing and will always increase. And if you think about that, it's true. Like there's no way that tomorrow there's any less data than there is today. We're on a crazy exponential trajectory for data creation. Well, to that level, that's what AI feeds on, better data. And, you know, when they get it to a point where, you know, Jan LeCun just resigned from Meta as their chief scientist, and he just raised on a $5 billion valuation to start his new startup. And what is he going to do? He's going to try to build a model or group of models or a platform based on a concept called world modeling or worldview modeling. very different from large language models or reasoning engines in that rather than it train on vast amounts of giant corpora of data where you just overwhelm it with patterns and it learns these patterns and figures out how to do what it figures out how to do here they're going to try to train it on the physical world this is more for robotics than anything else but with the thesis that unless ai can model the world it can't experience what we experience or understand how to experience what we experience. So he's going to do that. And maybe it will fail, maybe it will succeed. But DeepMind's on the same path. And Zuck knows it too. NVIDIA totally understands it. And everybody, and so does Elon. And everybody's thinking humanoid robots and other robotics are what will benefit from this modeling of the physical world. We don't know where that's taken us. Nobody knows where that's taken us. So I would argue back to your question that it's not that 25 became the year that AI hits front and center. It's that consumers all of a sudden have a pathway to understand some of this. And that's good and bad. The good news is everyone gets to do it. And if you're listening, I encourage you, I beg you, open up your laptop, take your favorite model, Gemini, Claude, any of the chat GPT models. Ask it some questions and just go deep. And the way to do this is to choose a personal project, something you really care about that you don't think you're schooled in the art or skilled enough to do, and have the AI help you do it to your satisfaction where you know the output and it gets you there. And once you've had that experience, once you've gone from beginning, middle, and ending, rising action, climax, falling action, you put your hands on keyboard, then and only then will you have an emotional attachment to your point of view, grabbed it in real life or lived experience. I think we all need that. We owe it to ourselves. We owe it to the politicians. We owe it to our kids. When someone says, oh, they only use AI to cheat or, you know, no one's going to be writing original stuff. Whenever anyone starts to pontificate about what they think the future of AI is, you will have a personally lived experience, fully grounded in your own two hands, understanding of what AI can do for you today, more importantly, what it can't do, which is maybe the thing that's missing from everybody. We're ascribing these capabilities that literally do not yet exist. They may, they will, but they don't right now. So giving this thing the license to say it's going to kill us and or cure cancer, it's like those are two sides of the same question. Come on, guys. It's neither curing cancer tomorrow nor killing us, it will be able to do both at a certain point. Where are you on that trajectory? You need a personal point of view for that. That only comes from you putting your hands on keyboard. That is true. It's also a way to understand what it can't do, right? And there are some nuances that unless you're using it like a power user, you won't discover the limitation of context windows. The fact that they don't let you track your tokens, which is incredibly annoying because if you're in the middle of a large project, you have to wait for it to start to break down for you to figure out, oh, wait, I've used up my context window. It's not going to go anymore. And then you have to figure out a way to back out the data that's important from that interaction and redeploy it in a new window. So there's a lot of nuance to using these models that unless you're spending hours and hours, you will never get there. It's true. Like this past week, OpenAI said that they were going to adopt the skills methodology that Anthropic made popular a few months back. Skills are different from custom GPTs. They're different from evals. They're different from pre-prompts. They're different from context files. Skills are instructions that agents will read in a context window before doing their agent thing, whatever they've been sent to do. So one of the most interesting refactors is to now start to build a set of skills that will allow you to do things like, hey, when you understand that you're coming to the end of the token window, build yourself a continuation file that like writing those kinds of skills. So I have spent the last three months trying to get everybody to understand what a pre-prompt is, what a prompt is, what an eval is, what a context profile or a context file is. And of course, most importantly, what an agent is and how a skill empowers an agent. Just knowing those things exist will help you. But literally, what you just described, Charlie, is so important. When you run out of tokens and you haven't prepared to run out of tokens, you are for all intents and purposes. You're starting over. It's so painful. Or if you want to try to use Gemini for something where it has no memory, context memory whatsoever, but you have context memory fully turned on in chat GPT and you have it moderately turned on in Claude, you cannot interchange those windows mid-project because they don't have the context that each other has. And you end up literally starting everything from scratch, which locks you in to a given model for a full given project. But if you're capable of building the appropriate tool set that's portable to make yourself portable, then you can set up an environment like I have here where the model literally in an adversarial way determines my little what we call TPGPT, the pomegranate GPT. What it does is it goes and tries about a half a dozen different models of different sizes with a version of the prompt you're doing. Yes, it takes a minute because it's thinking, but it's finding the model with the best answer for the specific problem. So, you know, you could say, well, that's expensive and time consuming. Depends on the project you're doing. If you're doing a project that's a production-grade bulletproof project that you're going to either use day-to-day or you're thinking about selling to someone and you're not an actual coder and you're like vibe coding, or you are having this thing write English for you, but you wanted to write a proposal at a very high level, or you wanted to write some kind of report at a very high level or do research at a high level, having this thing with real constraints, it's like this is critically important. So if it takes 90 seconds and you don't have your instant gratification gene satisfied and you have to sit there and go get a cup of coffee while it's going to wait for five minutes to give you an answer. But if it the right answer at the right price you didn just spend getting the answer You spent and it happens to be the right answer in tokens Oh and there some tokens left over Oh and the context window you in is still fully functional Like a different level of AI, right? And you could say, well, I can go use a model with a million token context window. It's like, yeah, you can. But at some point it gets overloaded too and needs to, you need some knowledge base and some system for continuous improvement. They don't come stock out of the box like that. You need to build these components for yourself that are tuned to your workflows and also your tolerances for good and bad. Everyone's got different tolerance levels for good and bad and everyone's got different risk profiles. So I can't give you my stuff. It's not one size fits all. I've been tuning the stuff I use for myself personally for over four years, like well before ChatGPT, we've been using this stuff. And there's also stuff that I've been working on that just is absolutely not portable. And there's stuff I'm working on that's so portable that it's going to be a feature of OpenAI's stuff like at some point, or Claude's going to just bring it out. If you don't notice that every morning, you must have noticed every morning these guys come out with new stuff, like subtle stuff, a new slash command they built in. Someone's got some new Git repo with some stuff. Somebody really smart said, you know what? I've had this problem. Let me solve it with blank. You got to be on top. If you need to be on top of it, it's there for you. One of the tools we have, Charlie, you really like, goes out every day, literally every morning. Actually, it goes out three times a day, 6.30 a.m., 2.30 p.m., and 6.30 p.m. And looks for stuff I haven't seen yet today. So I don't have to wait for it to tomorrow. Because it knows what you know. It knows what I know, what I don't know, and it knows what I'm looking for. It's got very, very strict criteria. There's an eval that we constantly use for self-improvement. this model and every day I have a scorecard and when it delivers the brief, I will say good, not good, good, not good. And it now knows what I'm, it's very focused on what I do. And that thing's been building a knowledge base in, in, um, there's a little vector database and oddly it's a combo. Uh, it's an older technology that people, some people like, and some people don't. It's Postgres, which is a SQL structure database. And then Chroma, which is a plugin, a vector plugin for it. So it kind of does both. It makes it embeddable. And it also has a structured, you know, completely searchable with SQL queries way to go in there and find it. So you can troubleshoot. It's not just black box crazy. There are many other ways to do this. This is just, I started it so long ago, that was the technology that was available, but it's become quite powerful. And there's no reason to change it. Like it does its job. It goes out every day and it gives me three briefs a day where it goes ding, ding, ding, and it sends notifications via text and in Slack. Why did I build that? Because it's constantly looking and I'm not. Like I go into client meetings. It doesn't go to client meetings. It looks on the internet all day. So Rony, I'm interested in your perspective because of course you are building a special and unique model on the kinds of things that Shelley and I have been talking about with regard to context windows and training GPTs. I'll only say this about it. I'd love to get your take on it, Shelley. I think of the field as an ecosystem of diverse species. And don't get too used to thinking one species is the only thing that's happening. Because right now I think people are trained to think that, let's say we use the word animal, and animal is AI. And you think there's only one kind of animal, which is a very large cow. And everything looks like a cow. The kind of stuff I'm working on does not look like a cow at all. It doesn't have the same traits or behaviors, but it's in the world of computing intelligence. And I think we're going to see a lot more diversity because a lot of money has gone into one paradigm, which has limits. And I think people that see the limits, whether it's Jan LeCum and others, are going to go in all these different directions to take the good and leave behind the bad. I think some people put like hundreds of billions of dollars into like one thing and maxed it. And you're like, oh, it does have issues. It's got some real power, but it does have issues. And I think the next, this prediction on 26 and beyond, we're going to see a diversification, like a Cambrian explosion, of many pursuits to use intelligence from the underlying silicon to do things in ways that may look different from all the money that was spent before. The cure-all answer is not necessarily from OpenAI or Gemini or Anthropic. It's one animal in the ecosystem. Love to get your take on that. Look, you're right, and your metaphors are good. What I do know is that there are a very, very, very small set of organizations that will bring this to you. The clear winner with no competition of any kind is Google. They own their own data centers. They own their own fiber. They own their own TPUs. They own their literally own compute, which they build themselves. Everyone you know in the world has been logged into Google for more than 25 years. They have 11 products that have more than half a billion users a day. And Chrome has three and a half billion users a day, which is two thirds of the users of the internet every day. No one has that kind of data. No one has that kind of infrastructure. They cannot, no matter how many times they make a mistake, no matter how badly they are managed, no matter what faux pas or screw up, nothing will stop these guys. And now that the U.S. government needs to beat the Chinese to AI, they're not going to be torturing Google. They're going to be encouraging them to win and win at all costs because everyone believes national security is at risk. So Google will clearly win. Whether OpenAI ever figures out how to make a living or not, they have demonstrated that they are some of the best researchers in the world and some of the biggest brands in the world. And while they have lost a lot of talent to startups and they've lost some key players, it is very clear that these guys are serious and amazing and they have captured the hearts and minds. Plus, they own the brand name. Nobody Geminis. Nobody Clauds. They chat. And what they mean by chat is chat, GPT. The guys at Amstropic are my heroes. They are the most profitable on a pure play of AI because I've got four Claude code windows open on my desktop right here, right now, grinding tokens. I am a couple 300 bucks a day into Claude every single day because it's never not running, ever. I pay OpenAI 200 bucks a month for their max plan. And apparently I'm in the 0.1% according to their like wrapped thing of people who use it. But I use it a fraction as much as I use Cloud Code. And in true reality, 97%, and this is their words, not mine, of OpenAI's users are on the free tier. Only 3% pay. So there's no, and it costs them a fortune for every query. So they have no, they're the most valuable startup valuation wise. They say not a half a billion, sorry, half a trillion dollars that has ever happened. But I don't know that they have a path to profitability, which they may not need, by the way, because the investment community is exuberant. So you look at Anthropic, they are making money. They're going to make money. They are brilliant researchers. They are doing brilliant research. And coders are just going token after token. It's grinding all day long. Every engineer I know is grinding on cloud code. They're making coin. Open AI, don't know if they're ever going to make money. but they have the hearts and minds of the consumers and Google just wins. So to your point, you know, is it going to be that way? Any minute, any day, any second, we could have two things that change the world. We could have enough qubits for quantum computing to take that class of mathematical problem and make it doable by anybody, which means all of the harvest and hold encrypted stolen data that's useless now becomes useful. Forgetting about the data that's already post-quantum encrypted because that's safe. And all the current data is, for all the big institutions, mostly safe. But all that harvested data that no one's been able to crack gets cracked on that day. And it will be analyzed by AI because there's no other way to analyze it. And large language models are totally fine for that. So when we talk about world-changing things, that is a world-changing thing that could happen now, next week, next year, or never. Like, we don't know when that technology happens, but when it does, everyone sits back. The system will be shocked in a way that no one can predict, but shocked it will be. And the other thing you don't know about is on the same timeline, which is tonight, tomorrow, next week, next month, never, this idea of world modeling or this different, the way it was just described, different species, these different flavors, different approaches to the problem set, which, or some combination of things we yet know. I think some nine-year-old who's going to Dartmouth early decision for mathematics and is going to be the first self-made billionaire. And I literally think it's going to be a nine-year-old sitting at AI and not even typing, talking to it. You know, something like super brain. Maybe it's a 14-year-old, but you know what it is? An old guy like me. It's going to be like some kid with a plastic brain and like who's grown up in this world and feel super comfortable and has some way to, some vision of the future we don't have. And I'm encouraged by that. I'm not scared by it, but I know that where we are is not where we're going to be. Everybody uses this analogy. It's an imperfect metaphor. But in 1998, like when FrontPage from Microsoft was out and you had to connect some kind of MySQL or SQL database and you didn't, there was no CSS or it was just beginning this concept of cascading style sheets, just regular HTML and was really primitive. And you had to build your database and you had to connect it by hand and you had to write your own HTML code and like every line of it. JavaScript was very infrequently used. Everything was by hand, as I remember back to those days. Nothing was standardized. Nav bars weren't even standardized. And look where we are today. You can vibe code a website in 90 seconds. And it just is. All right. Whenever AI gets to where you say to the model, hey, spin me up, an expert that does this, without asking for anything else. Like, how far are we from that? And what technology will build it? I don't know. But it's got to be going there. It has to, like, there's no other place it can go. And if you want to call that agents or agentic tools or agentic, like, I don't know what you want to call it. It doesn't have a name yet. But you have to believe the end state is a notification shows up in your smartphone or your Alexa says, ding, ding, ding, hey, I have a notification for you, and it's telling you, yeah, I just renewed the car lease at a favorable rate. And we're getting somewhere close to there. I don't know when, but that's where it's going. We're over an hour now, so we've got to wrap up the show. AI bubble or no AI bubble? Shelly's laughing because it's probably the question everybody asks him first. Can I give you my answer, Charlie, first? Please go ahead. It's a qubit. It's both. There's a bubble that will splat, and there's also going to be massive success at the same time. We live in a quantum world. Both things will happen. Love to get Shelley's take, though. I love that. I mean, I just love that as a metaphoric look at it, whether it's a qubit or whether, well, here's what's going to happen. Like people always try to time the market. I know exactly what the market's going to do tomorrow. I know it more accurately than anyone else can ever predict. It's either going to go up or it's going to go down. It won't be the same. That much I know. And so there are people who made financial investments that are incredibly stupid because deploying capital is a game of being, you know, right 51% of the time, not 80% of the time or 90% of the time. And so there are people deploying capital in places they shouldn't deploy it. And there are people who are willing to take that capital who are complete charlatans. And valuations for those kinds of things are insane. If that's what you call a bubble, and that's a good definition for one, there's no way that doesn't burst. It will burst. What's not going to happen? The U.S. military is not going to back off. The administration is not going to back off. Say, you know what? We're just kidding. The Chinese can get to AGI first. Not going to happen. Not happen. So as long as that's not true, this is the Manhattan Project. And the internet wasn't understood to be the Manhattan Project when it started. Although DARPA started it. They knew it was important. And the World Wide Web, when it happened in browsers, certainly no one made the leap that it was a military project, that cybersecurity and cyber warfare were going to be a thing. Now, no one's that naive. We understand, everybody, that you don't need to be technical to understand that AI can be militarized and it can be done both in information warfare and in actual warfare. So it's like physical warfare and cyber warfare will both be dramatically impacted by AI. So whatever bubble you think is happening, which may absolutely burst in some Wall Street kind of way for the retail investors, if you're making drone control AI, if you're making robots that go into battlefields instead of people, if you're like doing stuff where, you know, it sorts out DDoS attacks in nanoseconds and makes them go away. Like if you're building that, there ain't no bubble in your world. There's no defense bubble. It's not happening. And so, yeah, there's some things that are similar to the 1999, 2000 demise of the World Wide Web first, you know, the first dot com bubble burst. But, Charlie, it's not. It's different now. I think that's different pressures and different stuff. I think this is a really great answer and a great way to end the show, which is it's both. Both things are true. It is not absolute. So there are going to be some incredible home runs in AI that are going to change the world. And as you said, Shelley, there are lots of charlatans running around who don't know what they're talking about and certainly are going to ultimately explode. You know, the round tripping, the other stuff we've been seeing is obviously not totally legit, but it doesn't mean that the technology is not changing the world at the same time. Right. Both things. Exactly. Right. So that's our show, everybody. We'll be back here next Friday with Gary Shapiro, the head of the CTA, which puts on CES. Gary also, like Shelley, has become an annual tradition for us. So we hope you tune in. Come see us at the AIXR podcast party. Shelley, thank you again for coming. I will wave to you on the stage next Wednesday or the Wednesday, I should say two Wednesdays from now when you do your annual CES breakfast, which was terrific last year. So anybody who can get a ticket to that is going to one of the highlights of CES. It is lucky to get in there. So I feel very privileged and I look forward to seeing you and everybody else in the coming year. Have a great new year, everybody. And we'll see you back here. you