Anduril launches Copperhead AUV and Seabed Sentry at world's largest naval trade show

market's melting down. It's a little chaotic, but uh a lot to talk about and a lot of interesting uh opinions flying around on both sides. Hopefully things are doing better for you. Uh massive launch today. Congratulations. Can you break it down for us and maybe introduce yourself as well? Thanks a lot. Yeah.

Uh so we're at the world's biggest Navy trade show. Cool. Here in DC, so the sea air and space conference and we're in the dive bar. I know. Our marketing guys are epic. So I love it. Yeah. So we've got in the dive bar. So my name is Shane Arnut.

I'm the senior vice president for maritime business line here in uh at Andrew. I'll say yeah really really really big conference for us and um two brand new products. Seabed Century Copperhead going public today. So yeah it's pretty cool. Can you uh maybe let's start with Seabed Century.

Uh what what was the genesis of the project? What's the need? Uh, and then what is the what's the product actually do just for like a layman? Can you hear me? Yeah. So, Seabird Century is about undersea awareness, if you will. Um, so hopefully am I coming through okay for you? Yeah, you're good. Yes. Yep. Great.

Um, so it's pretty crazy. Like if you zoom out, like as as a species, we know less about the seafloor than we do about the moon or Mars. Like, uh, some 70 to 80% of the seafloor is uncharted, which is crazy. So, it's a good place to hide if you're a submariner. Um, it's also a good place to hide if you're the threat.

Um, so we kind of created this system, if you will, to be able to bring light to the deep dark sea, if you will. So you look at kind of vast amounts of infrastructure. So uh 80% oh sorry 90% plus of uh internet traffic. So this particular podcast would be running across subc cables. Yeah.

Many nations have their gas lines under the uh the waves. These are broadly unprotected unwatched right now. So the greatest vulnerabilities are below the waves and no one's really talking about it or working on it apart from some super secret things. And we kind of think that's wrong.

So Seabed Century was kind of came from that place as you know Andrew we can invest our own money put points towards things that we care about and we believe this is a big issue that people should be working on say and seabed century is seabed century uh do you expect you know DoD our allies etc to be the the main buyers or is this the kind of thing if you are in the business of laying undersea cables would would uh you know th those types companies be customers as well or is it purely defense uh applications?

No, we we definitely see it as dual use like anyone who's got energy infrastructure. One of the key things I will uh lay out is this is a cableless system. So, it's completely wireless. So, all the all the systems as you guys probably know are cold war era.

You need big cable laying ships to put these uh systems on the sea floor. uh almost impossible to upgrade. These are completely wireless um fully electric and you can place them in the ocean.

Um specifically, we can use some of our autonomous submarines so you can place them covertly or clandestinely, which isn't a bad idea if you're setting up security cameras and the bad guys know exactly where all your surveillance systems are. Kind of defeats the purpose a little bit.

So, this approach is is different thinking, if you will, that we can place them anywhere on the planet. we can get one of our submarines and you can place it with all sorts of different payloads in um in the body of it for long range surveillance for listening for other actions that you might be taking as well.

So can you talk about the broad risk of uh undersea drones?

You know, I imagine it's the same type of dynamic as, you know, uh, traditional, you know, airbased drones and that you could send a $10,000 drone to destroy, you know, a pipeline that that could cost a billion dollars or or cause a billion dollars of economic damage.

And so um the uh the whole ratio is pretty pretty thrown off and and uh but but I'd be curious like is it something that's not getting enough attention because it feels like right now kamicazi drones and and uh everything um in Ukraine is sort of like dominating the narrative but clearly you guys are um have found this important enough to to put a lot of R&D into it.

Yeah, the maritime environment's been kind of slow to the punch to be honest. like you you look at the proliferation and the success of robots in the Ukraine conflict and everyone looks back now and goes, "Yeah, of course, uh this was a better idea.

" Then you also see kind of crazy things like Patriots being used to drop um UAVs that are a tenth or 100th of the cost. The the unit economics of that exchange just doesn't make any sense. So y maritime domain just on the military side exactly the same.

There is a proliferation of unmanned underwater vehicles primarily from the Chinese and the Russians, also unmanned surface vehicles as well.

And if you're going to utilize, and this is probably a good segue to the copperhead, which is the the next thing that we announced, which is a fast, flexible payload um AUV, which could be uh made into a munition. Um so a torpedo, if you will.

um that particular system is built to be at a fraction of the cost of say like a Mark 48 or a Mark 54. These are very prevalent um uh torpedoes for for your listeners uh who aren't familiar with that. But when you're launching a Mark 48 against a USB that's like a tenth of the cost, the economics don't sort of work.

So we've specifically built these things to be at a price point that's a fraction of. We can also build them at thousands per year whereas kind of right now you may be lucky to get 100 200 uh torpedoes out of a supply chain as it exists today.

And again that that exchange ratio just does not work against kind of the fight that we're basing into. So the Asia Pacific fight with uh with China clearly that's a water-based um conflict that's coming but also people don't talk a lot about the high north. So, and that's where Russia is dominating.

So, uh you know, there's a lot of great stuff um sitting under the ice that's now retracting. Everyone wants a piece of that. You know, some of the discussion around Greenland and others is the basis of why the high north's kind of the next focus, if you will.

So, having these capabilities that are fully electric so we can get them under the ice caps, for instance, is of interest to folks. um the ability to do it at on mass affordably in order to kind of deal with a very wide waterfront is kind of the basis why we invest in these capabilities.

Can you talk a little bit of a little bit more about the trade show that you're at and exactly what type of buyer are you interfacing with? What kind of uh points are you trying to make to them beyond just hey there's this new thing it's important and cool? Yeah.

So as as I said um and you know what's happening around me at the moment world's biggest is everything from frigide to you know weapon systems whatever else um that's kind of happening. So I think a big part of what we're trying to paint if you will is the ability to mix match.

So we've got our autonomous underwater submarines um that are in effect mother ships for delivering these things plus the seabed sentry plus the copper head and just to be able to give the creativity back and it's not an obvious thing.

So enabling you know within the dive XL which is our big uh underwater submarine or autonomous submarine I should say we can put dozens of copper heads in. We can put a dozen seabed centuries. So the ability that you can kind of mix and match and give more creativity to the commander and that's not immediately obvious.

So just kind of we've got virtual reality um experiences for people to kind of ah here's how I would do this as well as coming into our our dive bar that we have here to you know have the conversation. But the it's not obvious.

A lot of this is second order kind of understanding for what's incredibly wicked difficult problem um that the subc warfare environment kind of presents if you will. So that's what it's about being able to come here start those conversations.

Uh I I I always notice that both the dive submarines and the new copper head are square. Is this is are there advantages to that?

Does that mean that you're less compatible with old torpedo tubes or is there some sort of backwards compat compatibility or does it just not matter and it's better to stack in, you know, a shipping container as you ship these around the world? Yeah. Yeah, great question.

So, truth be told, it all started with uh with Palmer. Okay. So, Palmer Lucky, our founder, and um you know, I was kicking around and said, "All right, you know, boss, we're looking at doing a a torpedo. " and and he said, "Ah, but make sure it's square. " And I like, "Well, why?

" So, I took it away and the guys did the numbers and interestingly for what we're doing, it's actually faster. Wow. So, given we've got the the software magic to fly these things, it actually flies more stable and quicker for the application that we want. Secondary is then the um uh the manufacturing.

it's a lot easier to manufacture something that's like you know a bathtub as opposed to a big spherical pressure vessel thing. Um so that was very significant as well. So those two kind of together is why that's the case. Um now it is a similar diameter to existing uh aectors like the Mark 48 so they could be used.

It's not our intent at this stage. So, our intent is for this to be launched off an autonomous to be launched off an autonomous submarine. Um, and that it makes the integration and the safety case a lot easier when you're both sides of that equation.

So, when you're the archer and the arrow, um, you can just go a lot quicker when it comes to changing the threat, changing the seekers, changing the capability, you can pivot a lot quicker. So, that was the basis of the intent. But, um, certainly a lot of customers have asked the question that you've asked.

So, we'll look into it.

Uh maybe last question for me just curious uh does Anderoll have scuba divers on the on the payroll or do you guys observe the R&D and and the testing process using other is it is it fully autonomous end to end you just sort of launch them and and say goodbye and then and then watch from from on board or I'm curious whatever you can share I think it's be fascinating.

Yeah, I mean part of a dev process is things go wrong. So I'm a scuba diver. Um so when things go wrong, I'll stick the tanks on and jump in as will the team. So I mean it's it's the iterative process.

So uh it doesn't work all the time, but we're able to spin the wheel fast enough to get to good from, you know, bad fairly quickly. But uh yeah, certainly all the team, everyone's big enthusiast of the uh the water space. It's uh not terrible thing.

group because all the tests are horrible places like Florida and you know Sydney or um you know California or whatever. Previously I did fast jet stuff and the the test ranges for the Air Force guys are in these awful dusty places. So it's not a terrible thing doing Navy projects nowadays.

Uh I mean last question for me and then we'll let you go get back to the important work you do. Get back to the dive bar. Get back to the dive bar. Uh, I would love to hear about some of the challenges of communication in the underwater domain. Obviously, you're flying a drone around.

You might be able to talk to Starlink. There's tons of satellites and other drones up. When you're underwater, signals don't travel as far. That feels like it makes autonomy more important being able to operate even with if the connection gets lost.

But can you talk about how you're thinking about uh communication and signal transmitting across the different fleet? Yeah, a thousand%. So underwater is the worst domain. So com suck. There's no GPS. All the things that you rely on for an autonomous system. So we it is the masterclass for autonomy.

So not space, not air, not surface, not land, it is subsurface. It's the most difficult. Right. So, um, we've spent a lot of time and I think, uh, you know, Andrew, this is kind of our, uh, what we're good at.

Uh, but this is certainly I've got probably the best autonomy engineers on my team because you've got such thin pipes or non-existent pipes. So, you're in effect GPS denied. You're a whole mission. Um, you can only pass very, very small amounts of information between the assets because they all do team together.

M so in doing so like you need to be very judicious about what you pass and when in what configurations etc.

So um it it's the most difficult domain and and one of the highest reasons for autonomy and I think where we're getting the success is we've come with this software first thinking for a particularly difficult problem which is this I think everyone can build you know a a sub a submarine autonomous submarine that's not difficult having it survive communicate act still be on a string from a human particularly when you're taking kinetic action um to make sure all the approvals are done in the right order is that's that's the big challenge.

So, you hit the nail on the head there. Sir, I I I did hear one interesting thing in the uh Seabed Sentry announcement. I think Palmer mentioned that it can detect biological creatures too. Is does that have a benefit or is that just like a fun fact? Yeah. Yeah. I mean, anything below the waves that kind of makes noise.

And in fact, uh, you know, one of the big things we want to do is actually detect when the biologics are, you know, seasonally coming through because you want to be careful when you're actually putting energy into the water because it can be harmful for sea life.

So, we all of our systems are collecting the pattern of life, if you will.

So we make sure that if there's some humpbacks, you know, cruising through a surveillance area that we're tuning down our sound system, if you will, to make sure that there's no damage to the u uh to the people who live in or the animals that live in that space. So that's just, you know, good practice. Um so yeah.

Yeah, it does all those things. You you mentioned that uh the underwater domain is the hardest from a comm's perspective, but Palmer's been talking a big game about doing subterranean warfare and tunneling.

Uh do you think he's going to give you for a run for your money with an even harder domain or you think you're going to maintain? It is a more hard domain, but we'll we'll see what systems actually get employed. So for right now, yeah, I think it's subc, but he's he's not wrong. He's always a step ahead. So yeah, drill.

It's going to be eventually it'll be time to drill. It's time to drill. Drill, baby, drill. Palmer and Vander, uh, well, thanks so much for taking the time out of your out of your day to come chat with us about this launch. Congratulations. Uh, it looks awesome.

The design's fantastic and I'm looking forward to learning. Yeah, I imagine I imagine you're quite busy because I doubt anybody was creative enough to come up with the dive bar at that at that conference. I'm assuming everyone's coming by, I'm sure. Yeah. Awesome. Yeah, we had to shut it down just for this.

So, everyone's very upset. They're they're lined up outside. I'm not kidding. So, Oh, really? Okay. We'll get back to it. Thank you so much for calling in. We really appreciate it. We'll talk to you soon. Cheers. Well, we shut down the biggest naval warfare conference. Yeah, we set back the America America's military.

It's actually brutal. Imagine Imagine your your Loheed Martin. You probably spent a hundred times more than Anderl on the booth and then nobody wants to come by if you can't. Xander is like, "Yeah, come by the dive bar and hang out. " Yeah. you actually get to touch the the tech. Ridiculous.

Uh should we should we go back to the timeline and turmoil? Do a little bit more timeline and then we will I I mean actually we could move on to uh some of the more optimistic news. Uh I don't know if you saw Llama 4 launched. That's very exciting. And of course we got Zuck with a front-facing camera video.

Hello everyone. It's Llama 4 day. Uh Llama dropped a of course the open- source LLM from uh Meta. Uh now they have a 10 million context window which I believe is 10 times what Google was previously offer offering with Gemini and uh the ELO is insane. There's a bunch of debate about how much of an advancement this is.

Uh are are the benchmarks getting saturated? We should get into all of that but I think it's just cool that um what they've clearly recognized is that you need to be superlative in something when you're launching a new product.

And there's a lesson for there for that in even if you're launching like a CPG product, you know, be Celsius. It's the It's the energy drink with the most caffeine of all of them, more than Monster and Red Bull, right?

And so you you can say a lot of things about it, but like you can d sum it down into what is llama for good at? Well, they're the best at the biggest context context. And then there's a lot of other stuff that's interesting. Uh but it's funny. Uh Austin Red posted a screenshot from Zuck on threads.

Uh somebody said, "Why did you drop Llama 4 on a Saturday, bro? " And Zuck responded, which is awesome. normally like the coms people would be like, "Oh, like the person's using bro, like shouldn't engage with that person. Like they're not serious. You should only take questions from like really serious people.

" Uh, but Zuck just comes in and says, "That's when it was ready. That's when it was ready. It was ready on Saturday, so we launched it on Saturday. " And I love that. And Austin says, "Bullish. " I agree. Very bullish. And Zuck's photo on threads is him, an AI image of him with like 25 gold chains.

And that's actually wild. It's such a good such a good meme that he's just like, "Yeah, I'm I'm I'm rocking the 25 chains. " Give I remember when he's got a piece on in the video, too. Yeah. Yeah. Yeah. He's not holding back. He's he's hanging out.

He's says, "If Satia and Zuck are shipping on a Saturday, what's your excuse? " Satcha posted, "Thrilled to bring Meta Lama for Scout and Maverick to Foundry today as we continue to make sure Azure is the platform of choice for the world's most advanced AI models. " LLM indifferent. They just want to rent the space.

Sachio wants to rent. He wants to vend. He wants to deliver models all over the place. Uh this is interesting. There's an article uh a post from uh less wrong uh says recent AI model progress feels mostly like BS. And uh we won't get into the whole article. You should just go read it um from zeropath. com.

Uh this was posted by LC on on Less Wrong. Um, but it tells the story of someone building a uh penetration testing tool driven by AI.

So, uh, you know, penetration testing of course is you you build a a a software application, you put it up on a server on the internet, and then, uh, are there different ways you can get around the security, figure out if there's a different way to to break into the server, get root access?

You you hear about this like WordPress gets hacked every once in a while and people figure out how to get into the admin page and kind of change your blog posts.

Um but of course pentesting is very important especially at the higher level and the more money is at stake the more the bigger the business the bigger the prize for the hackers.

Um and so this this person was building a an AIdriven uh pentesting thing says about nine months ago I and three friends decided that AI had gotten good enough to monitor large code bases autonomously for security problems.

We started a company around this trying to leverage the latest AI models to create a tool that could replace at least a good chunk of the value of human pen testers. They've been working on this project since June 2024.

And what he goes on to say is that there's all this hype on on on X and in AI circles about a new model that oh now it can, you know, do IMO gold medals and stuff, but for some reason he he Yeah, not yet, but soon.

um but but but he keeps he keeps switching to the new models and getting results that are quantitatively better but not qualitatively better and so it's not really solving his problem and this is this question of you know we've seen with GPT 4.

5 maybe pre-training is maxing out maybe scale is all you need but you got to apply that scale to the right algorithm maybe now it's reasoning models and you got to scale those up maybe it's else program synthesis is basically write have the model write code that does the problem that you want it to do.

Um there's also the just needing new ideas new ideas and and it's crazy I I need to dig this up because I think this would be a very viral clip but George Hots went on Lex Freiedman after GPT3 dropped. So this was maybe back in 2021 and he said that uh you know AGI or ASI will not be GPT9.

it will not just be scaled up of next token because that is not the algorithm for intelligence.

That is one thing that we do as humans but there are a lot of other factors that go into intelligence and you and you think about that where you know even someone like uh isn't it Daniel Conaman with thinking fast and slow this idea that you have two different frames of thinking.

You can you can be very reactive and think fast but then you can also step back and and there's all these different subsections of the brain. We barely understand the brain, but we understand that there are long-term memory, short-term memory.

Uh, you know, there's probably some humor segment that we haven't even tried to build in AI yet. Uh, and all these different the funny bone. The funny bone. Yeah, the funny bone has not even been tried to be put into silicon.

Um, no, but Mike from Arc Prize and Zapier aligns with what he was saying, which is like, hey, we need we still need new ideas. Yep. Which is part of what gets me excited about SSI. Totally trying to take a new approach. Yeah. I mean, yeah.

I mean, I siter seemed to be not only the person that decided that that that read the attention is all you need transformer paper and realized the impact of it because we look back on that and there's this survivorship bias where it's like, yeah, of course, like all these Google brain people put out this amazing paper called attention is all you need.

They describe the transformer architecture. Of course, that was like going to be groundbreaking from day one.

But you forget that on that day there were probably a hundred papers that dropped and the next day there were another hundred papers that dropped and even that Google brain team was publishing other papers and others other research trees and other paths and even things like symbol manipulation were popular at the time and all sorts of different algorithms were being tried.

Um and Ilia had the foresight to realize that that was the architecture that you could throw a billion dollars of compute at and get a really magical result. He did it. Open AI released chat GBT.

it becomes this amazing like you know LLM we know about that they implement transformer architecture in a ton of different stuff even Tesla's autopilot has transformers baked into it now it's all over the place this architecture uh then he realizes that reinforcement learning kind of the the alph uh you put that on top of an LLM have it selfplay have it talk to itself reasoning tokens that will be the next uh the next amazing thing originally called Qstar very scary.

There's all this chaos. Uh then it becomes uh you know what was it? Strawberry. Then eventually it becomes 03 01 03 high 03 mini high R1 from DeepSeek. All these different reasoning models come out. They clearly deep research. They clearly advance the ball towards something that's more usable.

You see the coding agents like uh cognition and uh and Devon doing something similar where there's there's a reasoning layer on top of a pre-training LLM.

uh but uh the question is what's next and I think everyone is kind in a and seems to be coming to the conclusion that there will be a next thing in order to unlock the next big breakthrough. Um and we're excited to watch it.

But as the model as the race for just general LLM training and scale heats up and all the news about you know Grock is in the game now and and OpenAI is scaling up to 4. 5. There's GPT5 coming at a certain point. There's uh there's DeepSeek. Uh Meta has been continually uh pushing the ball down the field.

They have a lot of infrastructure as a hyperscaler. But there was some uh controversy on the timeline from Nick. He says breaking meta meta AI internally caught cooking benchmarks to hide poor poor results. Absolute state.

Um, I don't know how how ex uh how um how real this is, but they are alleging some benchmark fraud, which you hate to see. Um, let's read it. It says, "The original post is in Chinese.

" Um, but the translation here says, "Despite repeating traded efforts, the internal model's performance still falls short of open source state-of-the-art benchmarks, lagging significantly behind.

Company leadership suggested blending test sets from various benchmarks during the post-training process, aiming to meet the targets across various metrics and produce a presentable result. Failure to achieve this goal by the end of April deadline would lead to dire consequences following yesterday's release of Llama 4.

Many users on X and Reddit had already reported extremely poor real world test results.

There was a debate about whether or not you could even run this on uh you know I'm sure people will like play with the models and we we talked to some of the open AI folks about like yes there are all these eval but at the end of the day you kind of just got to talk to the model and play around with it and see if it's useful and then also you need to implement into your system or your codebase and see you know is this solving a real problem.

Maybe it is, maybe it isn't, but it kind of depends on your your the economics of your business and the and and the application that you're building forward. So, uh interesting to see where this goes. There's been allegations of benchmark fraud and and eval saturation all the time.

Some of it is just completely accidental just because, you know, someone posts a a benchmark, then all the results go onto the internet, that goes in the next scrape, and then of course it gets baked in. And so, uh, you know, AP bio, for example, every AP bio question is just on the internet now.

And so, it goes into the pre-training data. And so, it's basically just memorized every AP bio question and answer. Is it really good at biology? Like, it's kind of hard to say. It's certainly good at looking up the answers, which is great, and that's useful. Memory is all you need.

And it's a it's a great Yeah, it's a great Is it intelligence? These things are certainly extremely high on knowledge, and they are amazing knowledge engines. Isn't that the term you use? Knowledge engine. Answer. Answer engine. Like they're fantastic at that.

And what I love about Llama 4 is that it is funny to think about you could have if if you took a human who had was sort of lowest on intelligence and then you gave them all the knowledge in the world. In so many situations they could fake intelligence.

Like they could you sort of ask them something, they give you an answer. You go, "Wow, so smart. " do do Jeopardy or do do uh trivia at a bar, you'll be like, "That guy's so smart. He knows exactly who won the World Series in 1982. " But it's like, is that intelligence? It's unclear.

Anyway, I like David Holes, founder of Midjourney. He says, "I desperately want Mark Zuckerberg to pose with four llamas for the Llama 4 release. I'll even say please. We need aesthetic moments in tech, and this llama branding is good for one thing and one thing only. " Great picture. Great picture.

Clearly generated from MidJourney. Looks fantastic. But Zuck has the resources to do it and I think he has the the character and the and the personality to pull it off and have fun with it. And so I would I would hire drop the chain, pick up the llamas. Yeah. Do get get four llamas together.

Do some do some llamas llama photos. Uh anyway, Toby Lkey has uh uh a leaked memo that's been floating around that's pretty interesting all about how uh spot uh Shopify is using AI. Uh he says AI usage is now a baseline expectation.

Uh we are entering a time where more merchants and entrepreneurs could be created than any than in any other than than any other in history. We often talk about bringing down the complexity curve to allow more people to choose this as as a career. Shopify's whole business is making e-commerce easy.

So they want to make it as easy as possible to set up an online store. They've done it very very well in my opinion, but they want to continue. Each step along the entrepreneurial path is rife with decisions requiring skill. judgment and knowledge.

Having AI alongside the journey and increasingly not just doing the consultation but also doing the work for our merchants is a mindblowing step function change here. Our task at Shopify is to make our software unquestionably the best canvas on which to develop the best businesses of the future.

We do this by keeping everyone on cutting on the cutting edge and bringing all the best tools to bear so our merchants can be more successful than they themselves used to imagine for what we des for what we need to be absolutely ahead. And he uh outlines this in an email to every Shopify employee.

Reflexive AI usage is now a baseline expectation. Maybe you're already there and find this memo puzzling. In that case, you are you already use AI as a thought partner, deep researcher, critic, tutor, and pair programmer. I use it all the time.

And so, he's telling folks, you got to be using AI and we got to uh build it into all of our products. So, just interesting to hear from a big public company CEO in founder mode. Definitely pushing towards more AI. Yeah. I'm gonna go out on a limb and say I don't think this was leaking. Oh, you think it's recruiting?

I think uh Oh, no. My my memo leaked. my extremely well-written, kind, visionary memo. I think it probably was leaked, but he did post it now and it's good content. No, I'm sure it would have leaked. It's just it's just you kind of write these things with leaks in mind.

Anyway, uh we got our next guest coming in to the studio. Welcome to the show, Logan. What's up, guys? Uh we wanted to bring on the