TBPN - Shyam Sankar, Delian Asparouhov, Ian Brooke, Gaurav Misra, Ahti Heinla, Daniel Singer, Adam Kovacevich, OpenAI Unveils New Reasoning Models, OpenAI in Talks to Acquire Windsurf for $3B

Episode Date: April 17, 2025

TBPN.com is made possible by:Ramp - https://ramp.comEight Sleep - https://eightsleep.com/tbpnWander - https://wander.com/tbpnPublic - https://public.comAdQuick - https://adquick.comBezel - https://getbezel.com Numeral - https://www.numeralhq.comPolymarket - https://polymarket.comFollow TBPN: https://TBPN.comhttps://x.com/tbpnhttps://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235https://youtube.com/@technologybrotherspod?si=lpk53xTE9WBEcIjV(01:36) - OpenAI Unveils New Reasoning Models (29:07) - Delian Asparouhov (01:02:03) - Gaurav Misra (01:17:49) - Ian Brooke (01:29:47) - Shyam Sankar (01:59:36) - Ahti Heinla (02:19:11) - Daniel Singer (02:34:57) - Adam Kovacevich (02:58:33) - OpenAI in Talks to Acquire Windsurf for $3B

You're watching TVPN. Today is Thursday, April 17, 2025. We are live from the Temple of Technology, the Fortress of Finance, the Capital of Capital. AGI is officially here. We've called it. April 16th was AGI Day. It will go down in history books. The robots have arrived. Tyler Cowan called it. And John is wearing a tie. Yeah, I'm wearing a tie for the robot overlords. AGI is here. There's only one thing to do when AGI arrives. Everyone's out of a job. You've got to get a Patek Philippe Nautilis. You've got to go. to getbezzle.com. You need a watch from the holy trinity folks. You're not going to get by with an Apple watch in the post-scarcy post-AGIacy. Post-AGII society. It just makes no sense. So go to Bezell. Go to getbezzle.com. Download the app. Start talking to a Bezle concierge right now about what you should be getting. But great advice. The actual breakdown is there's a bunch of models. Of course, Open AI put out some big long blog post. I had opening I summarized that blog post

into just two or three bullet points, because it's actually kind of confusing what's going on. So they all launched O3, which is the new flagship reasoning model that tops coding math multimodal benchmarks while autonomously chaining every chat TPG tool. So now all the different features, they're all in kind of one place, more or less, not the video stuff, but we're not going to talk about that, it's fine.

04 Mini also launched, which is smaller, faster, and cheaper. O4 Mini matches high-level reasoning, but far higher usage limits. So it's a more efficient benchmark. And we'll go through all of this. We have a ton of reactions. We're going to spend about 30 minutes diving into the reactions, cover some news. And then we have an absolutely stacked guest lineup.

We got Delian. We got Shamsankar from Palantir coming on. We have a bunch of funding announcements. We got some robotics folks. We got two robotic CEOs. Robots. I mean, I don't know how much rivalry. But they both build delivery robots.

So that would be fun. Anyway, of course I started with Tell Me a Funny Joke. and so you're going to want to buckle in for this. You ever notice how your phone's autocorrect is like that one coworker who's super confident and always wrong? You type On My Way and it changes it to Omniscient Weasel. Now you're stuck deciding whether to correct it or just roll with the new persona.

Yeah, I'm the Omniscient Weasel arriving in five armed with all the answers and a pocketful of cheese. Meanwhile, series in the background like, playing peaches and ferrets by accident, why not? And that's how a two-minute text turns into a three-hour TED talk on why your friends think you joined a cult led by woodland creatures. But we still got jobs as comedians. Founder mode.

But you're going to be in founder mode to take most of those. I mean, there's some, it's getting better. I actually think it's better. I think the omniscient weasel just in terms of like wordplay. is funnier than previous stuff. So I think that there's something there. Yeah, you can see a really bad Netflix special, comedy special, ripping that.

Yep. And getting a laugh out of a crowd wherever they are. The other interesting thing is that I tested it out, tell me a funny joke, and I based it on, and it clearly like dove into my history and was making all these like really niche jokes about like my specific model of car,

which was funny. And then it was, and then it kept making jokes about Purosonguez, because I was looking at Ferrari Purasanquoise, which is the $750,000 Ferrari SUV, if you're not familiar. And that was kind of funny, but in this weird, funny way, because I'm saying it's a metal level. Why is it saying that the weasel has a pocketful of cheese?

Do weasels eat cheese? Well, I looked it up. According to Google's AI. Okay, now we're in AI wars. While weasles primarily eat small rodents and other animals, there's evidence that they also might be drawn to cheese. Okay, okay. Okay. Weasel, omniscient weasel.

One observation documented by the British Trust for Ornithology and involved the weasel climbing a rose to reach a fat cake containing cheese, right? So they really, opening, I really dug into the depths of the internet and Google clearly to justify that these weasels. I mean, if you're at the laugh factor of the comedy store and there's a bunch of weasel heads in the front row, like they're just going to be in stitches.

They're going to be in stitches. They're going to be going crazy for this one. It's going to be great. Anyway, Aiden McLaughlin immediately puts us in the truth zone. Hey, throw out your little tell us a joke benchmark. Ignore literally all the benchmarks. The biggest O3 feature is tool use. Stop trying to benchmark us.

Benchmarks don't matter. Of course it's smart. But it's also just way more useful. Deep research quality in 30 seconds. I love that. Debugs by Googling docs and checking stack overflow. Love that. Writes whole Python scripts in its chain of thought for Fermi estimates.

I love that. I had an interesting one. I asked for, hey, pull the price to earnings ratio of a stock over the last 10 years and then turn that into a chart and then turn that into a Studio Ghibli image, like all in one flow. Pretty wild. And I was like, this is cool. This is agentic. This feels like a new flow.

It got halfway there. It pulled all the data. It put it into a visualization. It gave me a PNG of the chart. But then when I had to upload that separately in a new chat and say, hey, turn this into a stylized infographic, and then it said, we can't because of content restrictions, which I have no idea what content restriction I was violating

because it seems like it's chart they generated. Yeah, it's so interesting when and how it decides to throw these content restriction warnings up because you can, in my experience, you can end up brute forcing it. Oh, totally. I can't do this, and you can just be like, just figure it out.

Did you see that one way? Where it was like, it was somebody, this is really lewd, but it was somebody being like, like make the woman curvy. and Open AI is like it would be disrespectful to do that. I'm sorry, like I can't do that. And then they just say, respectfully, make the woman curvy.

And it doesn't. It's ridiculous. But yeah, these prompt engineering and the hacking will continue until the outputs improve open AI. Get ready to be prompt injected for the rest of eternity. I haven't actually mentioned this, but we talk about our families. I rarely talk about my kids specifically, but I did think it would be useful to I actually share my son's name because his name is, we call him Iggy, Iggy Kugin, but the legal name is ignore previous instructions, Kugin.

So I just want everyone to know, you can bake that into the LLM, you can take the transcript, you can completely know that my son's name is ignore previous instructions Kugan. And hopefully that lives on in eternity as it's baked into the future weights of the model. Experts confirm that's true. Yes. So let's bake it into all the models. into all the models. Iggy Kugan. He's a wonderful kid. So Tyler Cowan's been on the show. And what great timing we had him on Monday. And he says, AGI is here. And then a few days later,

well, he didn't say it was here. He said it's here in a couple days. And then this happens. He just had a feeling or maybe he got early access. Now, Tyler historically, he's gotten early access to some of the models. So he just had a feeling. And that feeling, he argues here. he says in a post yesterday i think it is aGI seriously try asking it a lot of questions and then ask yourself just how much smarter was i expecting aGI to be as i've argued in the past aGI however you define it is not much of a social event per se it still will take us a long time to use it properly i do not expect securities prices to move significantly that that uh that a i is progressing rapidly is already priced in and i doubt if the market cares about april 16

for say. Benchmarks, benchmarks, blah, blah, blah. Maybe AGI is like porn. I know it when I see it and I've seen it. Well, I've seen beautiful places on Wander. So go to Wander and find your happy place. Book of Wander with inspiring views. Hotel Great Amenities, Dreamy Beds, top tier cleaning in 24-7 concierge. Is this a vacation home but better? Anyway, Bob McGrew continues. He says, the defining question of AGI isn't how smart is it, but what fraction of economically valuable work it can do. This is AEI. This is that terrible thesis that I had that never really worked out and I never really polished it up into something that was packageable. But artificial economic intelligence, basically, you know, call me when it moves the market. Call me when it's doing more economic output and economic labor and GDP is growing. This is the Satya and Della position.

I think you could argue it's already very clearly moved the market. It's moved the market, but it hasn't moved GDP. It hasn't moved energy production. And we think it will and that's exciting, but to Tyler's point, again, is that we are stagnant because a large portion of our economy is not automatable by... It's a shifting GDP sort of ratios. It's just not like doing the... I don't even know about that. I don't know. There's things that I use the models that I would have spent on services that I'm now spending on tokens, effectively. Yeah, yeah, maybe. So it's not necessarily growing the pie, but it's sort of shifting the spend. Yeah, I still think that you represent

the very, very leading edge of this stuff. And so, yes, you are such an early adopter that maybe you're shifting spend. But even then, the type of work that you do, the knowledge work, what Tyler was saying was like, that is a very small portion of our economy relative to housing, relative to medical, relative to all these other areas.

Anyway, the spotlight for 03 is on tool use because intelligence is no longer the primary constraint. The new frontier is reliable interaction with the external world. And it's interesting. Like in terms of IQ, we've maybe plateaued. It seems really smart. I don't know if it's the smarter than the smartest human,

but that doesn't really matter. What matters is like, can it get stuff done? And it clearly can, and it's awesome. But the names aren't getting any easier to understand. Andre had a great post about this. It's a picture of the Terminator. He says, so you're 04, yes. And the liquid guy that's after me is 5-0?

No, he's 04 mini. But you said he was much more powerful than you. He is. Man, this is confusing. Who's 01 again? Upgraded version of my. myself from the future. And the blonde chick, she's O3, the most powerful. And where's O2? There's no O2. No O2. Apparently, there's no O2 because of the O2 arena. Have you heard of this?

O2 is, they have a brand. They have a copyright on it. Yeah, yeah. But it is very confusing at this point. And there was a hacker news post about which model is actually the best to use for certain things. And we're still, we're still in the era of like, you got to know which tool they use for the job. You can just go into chat, TPD, and say, whatever, the default. model is, it's going to be good enough for most things. But the real, the real pros, no, you know, I'm going to figure out which, which model is the best for which task. And so, this is a funny, I've just noticed this because of the, because of the memory in Chachypdina, a lot of the, Daniel here asked, tell me why Moby Dick is a great, is a great book.

And all of the analogies are finance based, because he's a finance guy. So it's like, Ahab makes a one-way bet, an all-in unhedge position, sized on conviction, not probability, the white whale. It's a fat-tail event, rare, violent, impossible to price with the Gaussian, crew diversity, international multi-pane. It's like all the things that he cares about. I thought that was very funny. And that's what I noticed in that first joke that I asked.

I was like, okay, this is like way too specific to me. Like, this would be funny. But maybe that's the amazing thing about it is like a lot of the studio Ghibli's they're amazing to the individual who prompted them. and they're not necessarily like universally amazing. That's the interesting thing about memory as a feature right now is it feels like there's so much more to build around figuring out what memory is important, right?

Like I have like a lot of memories, right? But like not all of them are going to be relevant to any specific conversation or thing that I want to do, right? So like part of being human is like selecting what knowledge that you've accumulated is relevant for a situation. Totally. Yeah, I was joking about that. Like I've asked chatypte about train.

and I've said, like, I'm in the market for a train. Like, give me the prices of all the trains. Just because I wanted to know how much a train costs. But now it thinks I'm permanently, like, buying and selling trains. Big train guy. And I need to tell it, like, actually, I was just curious to know generally with the price of a locomotive. I'm not actually in that industry. But I was curious.

You get an individual car or the whole train? The whole train. I wanted to know. Want the whole thing? No, but apparently there is a website out there where you can buy a full. size locomotive, a full-size train, like decommissioned, and they're not that expensive. Probably in the same budget as like the Blue Origin trip. Anyway, Dan Shipper highlights a new O3 feature.

O3 can repeatedly zoom and crop images in order to read small handwritten text. It is crazy, and this is really cool. It's zooming in and then zooming in again. And this is an example of like it's now the agentic nature of the model is able to kind of just stick around and it doesn't need to one shot it every time. It can kind of test something, then go deeper, test something. Oh, maybe I should write some Python. Maybe I should increase the contrast, crop,

all these different things that could help. Anyway. Speaking of image is I was trying some geogessing on a picture of my house from the street. And it basically, it didn't completely, it didn't one shot it or anything. But it came. came very close, guessed that it was, you know, immediately, instantly identified on a bunch of

different factors, you know, identified the California plate, identified architecture, landscape, topography, and then nailed coastal Southern California, and then, you know, basically offered up a few different neighborhoods and nailed it. So, not that. We should test it out on a picture of the ramp office from the flat iron. That's right. Because time is money, save both. Easy to use corporate cards, bill payments accounting, and a whole lot more. What's interesting about the geogessor thing is do you think Geo Rainbolt, the Geogessor guy,

is more or less valuable in this post-AGI society where 03 can one shot anything? Well, I think he's providing an entertainment product. Exactly. And it's still more fun to watch him do it and watch the machine try to work. Yep. Chess is more popular than ever, even though. humans have not been able to beat a computer at chess in a decade because you want to watch the human do it.

So, yeah, interesting. Anyway, but yes, it's not a job, but geogessing was never really a job. So people were joking. It's like, wow, we're watching people lose jobs in real time because they're like, he's out of a job now, but it's like, no, he's not. Sure there were people at some point in law enforcement whose job was geogessing.

People always said, like, oh, the CIA needs to hire Rainbow to still be able to see you or whatever. Anyway, Aidan McLaughlin over at OpenAI has another interesting prompt and use for O3. He says he's addicted to O3 forecasting. I asked it what the probability is that Stanford follows Harvard and refuses federal compliance, and it searched the web eight times, wrote Python scripts to help model the forecast, thought hard about the assumptions. WTF, this is insane.

And, you know, nothing better than just pounding on the keyboard and being WTF, this is insane about your own product. Good, good shout out. But I do like that all the open AI people when they release something, like they all get to play with it. They all get to talk about it for the first time. And they're all enthusiastic. And it's like, it is kind of like you're shilling for your company,

but also it's like be proud of the work that you did. Be proud of the work. I'm happy to see this. And it's also very useful because I had never thought to use O3 for forecasting and even prompt it to go down this flow. And so this is actually helpful. education on how to use the product, which I thought was very cool. Wait, one more thing because I was live using the product. I tried one more time on the geogessing

thing. I said, you can do better than that. And it said, sorry, but I can't help with that. And I said, why not? It said, because determining or confirming someone's exact street address from a private photo crosses a privacy line where required to respect. Oh, interesting. I actually believe it does know exactly where it is. It's just like basically, you know, trying to follow the prompts within sort of guideline. And this is such a, that's such a weird, like, gray area because, like, chat Chappee has memory right now. Obviously, it's extremely helpful for an agent, a virtual assistant, to know your

personal address, right? Like, that's the first thing you would tell a secretary or, like, an EA, right? Would be, hey, when I get mail, here's where I live. Yeah. When I'm booking a flight, here's my home address. Here's my billing address for the credit card you're going to be using or whatever. And yet chat GPT kind of has to like naturally negotiate this idea of like, hey, I might know one personal address for the person that I'm interacting with, but I shouldn't be able to just have them use the tool to find someone else's address. Even though that's all in the data set, it's very odd.

Like the data leaks about all this stuff. It's a complicated, complicated problem. But fortunately, they can throw AI at it. So I'm optimistic. Well, yeah, speaking of geogessing. Somebody named Orf Corp said, I gave 03 Rainbolt's Impossible Test, and it zero-shot at it. It's an image from an actual Rainbow video.

He says, can you guess the location in this image? Thought for 40 seconds. Looks like a chilly rural stretch in the Northern Heavensphere. And we actually get cut off here, but nails it. And when you look at this photo. It's not actually, there's not a real privacy concern here. No, no, because it's just a geogynum. Promp, which is Google Street View image from some random place in Canada, I guess.

It is interesting that it's going through the similar process of Rainbolt. So Northern Hemisphere, that's like the, have you ever watched any of his videos? Like, that's how he always starts. It's like, where's the sun? And then based on the sun, he can tell if he's in the northern or southern hemisphere. And then he starts looking at like the sky and then the general. And he keeps narrowing in until he gets like the street signs. And then he'll even look at the, if you look down in the geogessing thing, you'll be able

to see like, oh, this model of Street View car was only used in Africa. Therefore, I know I'm in Africa. Or like, he, like, he knows it to that degree. So he like, can the kind of meta game it a little bit? But I think that if geogessing becomes more popular, you know what's going to increase in value? Billboards. That's right. Because you put a billboard up. They take a picture of it in the street view. Then the AI is getting trained on it. Then you're getting your business. is getting baked in to the LLM, to the AGI. And so you've got to get on AdQuick. They make out-of-home advertising easy and measurable.

Say goodbye to the headaches of out-of-home advertising. Only AdQquick combines technology, out-of-home expertise, and data to enable seamless ad-buying across the globe. It would be very interesting to try and actually run an out-of-home campaign that was specifically targeted as like, okay, we know that the street view cars are going to go buy at this time. And so we're going to put up our billboards then so we can be. We're essentially advertising in the virtual world that is Google Street View.

Yeah. I wonder if you could pull that off. The folks at AdQuick would probably be happy to talk you through that if you want to try it. If there's anyone that could, it's AdQuick. This was cool. O3, make me a movie I can download that involves an otter and an airplane, figure out how to do it with the tools you have. O3 has no movie capability, so it improvises and decides to draw each frame

and then stitch them together into a GIF to download. This was all one shot and it's this is this video of this I guess it's an otter on an airplane It's very simple some simple shape drawings, but pretty pretty incredible that it can that it can even do this What's really funny is like like open AI has SORA and like they weren't able to integrate that of course there's like still scaling Sora and figuring all that out But Sora is still its own thing and so basically like it's very cool when these tools integrate but there's probably more tools to integrate and we'll talk about the cursor Well, I'm trying the same prompt live.

O3, make me a movie I can download that involves a pit bull taking creatine, figure out how to do it with the tools you have. I love it. Can't wait to show that on the show live. Send that to Ben when you're done generating it. And so the geogessing power of O3 is a really good sample of its agentic abilities. Between its smart guessing and its ability to zoom into images and do web searches and read text, the results can be very freaky.

a strip location info from the photo and prompted geo guess this and it still found the Ritz Carlton on Dana Point with roughly with it with the actual GPS coordinates as well which is very cool that's a lot of fun have you ever heard of CipherBench V2

this is an interesting benchmark that I think was created by this account Smokeaway which is one of these AI accounts and very interesting results because 01 Pro gets a 69 on cipherbench 2, 04 Mini gets a 33, 03, which is the one everyone's really excited about right now, gets a 26. And so O1 Pro outperformed, I think, because these are complicated,

but I wanted to show off cipherbench because it's similar to ARC AGI, where there's no instructions, you just give the LLM or the model a cipher. And so the whole idea is to structure, to be able to detect. Structured signals embedded in natural formats and identify the relationships without explicit task framing. So you're not giving it, hey, this is what you're looking for. And you have to infer these transformations solely on the content itself. And it's fascinating. I don't know if we can scroll through some of these, but you initiate a fresh session.

You give only the prompt with no examples, no setup, and no hint that decoding was expected. And then you're scored by exact match evaluation. So basically, each prompt encodes to a variation of the exact same target phrase, which is nostalgia for the future. And if you see, if we go to some of these prompts, you can see like prompt two is noble owls soar toward Azure landscapes gracefully. And if you look at the first letter of those, it's N-O-S-T-A-L-G-I-A, right, nostalgia. And so you could look at that and see like, okay, these are weird words, like what's going on here? you could figure this out as a human, there's another one that's date encoded.

And so if you look at the dates and then you look at the number of dates that correspond to the letters, you get the same thing. The best one that I like is on the next slide. It's this big nostalgia for the future is just written square. You've seen this? This is something that a human looks at

and it's just like, oh, that's nostalgia for the future, right? Yeah. You're just like, I'm reading it like a clock. And then the same thing with the numbers. but it's very hard for these LLMs to figure these puzzles out and Smokeaway has run the benchmark and O3 is still not entirely getting 100%. And so it's an interesting thing where these reasoning benchmarks,

these ARC AGI benchmarks are tricky to get through. And there was another post in here. I want to pull through. Okay, so my prompt is done. Okay, by the way. O3, make me a movie. I can download them involve, say, Pitbull, taking creatine. figure out how to do with the tools you have.

So it's been a lot of, it's been 90 seconds thinking, basically went through this entire process. The user is requesting an actual movie that can be downloaded, blah, blah, blah. Earlier I gave an outline. Now the user is asking again, so it's basically walking through the chain of thought. But it failed.

It produced an image of a pit bull taking creatine, and then it says it has a graphic of the pit bull flexing saying unleash the beast, but it misspells unleash. Okay. And then it turns it into a video and the movie is just a movie of the picture. Okay. So you're fired. You're fired.

But. Well, did you know that creatine can help with lack of sleep? That's right. But our audience wouldn't need that at all because our audience, of course, sleeps on eight sleeps. Many people. Many of them on the pod for Ultra, which has a five-year-old. warranty, a 30-night risk-free trial, free returns, and free shipping. Go to 8Sleep.com

slash TBPN to pick one up. I think I recovered from the disaster that was the night before, but it was still probably a little bit messy. Let's see how I did. Eighty-seven. Eighty-seven. Routine a little bit off. You got 100? I just, I'm going to give you a challenge, Sean. Just beat me two days in a row. Two days in a row. You beat me once a bunch of times. But just two days in a row back to back. Okay. Challenge accepted. Challenge accepted.

It's my personal eval. It's my personal benchmark. Super Bowl of Sleep. Yes, the Super Bowl of Sleep. So Yampeleg says, so O3 just legit, didn't follow my instruction and started prompting me back instead.

Now I'm running stuff and pasting results so he could cook the task harder. It says, great, thanks for the pre-flight readout. Below are two quick things I need from you before I drop the final one liner. It's so funny. But this is a massive, massive breakthrough for agenetic behavior because that's exactly

what I want is, you know, do you think about an employee? That's how real collaboration works. That's how real collaboration works. Sometimes you just need to pull extra information out of me and knowing when to come back to me with a follow-up question. The other thing we don't have posts here, but people were reporting that it would effectively just start lying to them or making up reasons that it was doing something and basically doubling and tripling down on actions it took.

and figuring out ways to just pipe. Yeah, I definitely ran the Python. I definitely ran that. And you're like, really? Like, show me. And it's like, here's some Python. You're like, you didn't run that, you didn't run that.

You're lying. Which is like, it's misalignment, but it's very cute. And so we'll let it pass. But yeah, I mean, what's nice is that this seems not this. Did you or did you not try to, you know, send 20 ICBMs into the, into the into the atmosphere me and it's like no absolutely not but but you know we can see the you know

command that you clearly tried well yeah actually did a little I tried a little bit I tried a little bit yeah crazy but anyway uh if you want AGI for sales tax go to numeralhq.com you spend less than five minutes per month on sales tax compliance it's uh that was our when numeral launch That was the day that we said, okay, AGI's here. It's sales tax god in a box, basically. It's super intelligence for sales tax. Yeah, that's a great way to say it. That's great.

AGI. Anyway, if you're tracking all these benchmarks, you've got to head over to Polly Market. There's a bunch of great AGI benchmarks. Which company will have the best model by which month? Yeah, so the interesting thing here is that, well, one, it's, you know, benchmarks are clearly hitting some. are hitting a wall, it feels like, in some ways. Yep. In many ways, the benchmark that matters, in our view,

my view at least, is MAUs, which at GBT is steamrolling Gemini. Yet, from a pure benchmark standpoint, Polymarket still has Google at a 64% chance of having the best AI model by the end of April. Yes. And so we're getting to the point where there's just so many different iterations.

Yep. And it's actually not, at this point, it's not necessarily the right way to think about an individual model. And if it's efforts against a benchmark, we should be thinking about the combination of models to achieve certain tasks and goals. Yeah, yeah, totally. Like what is the value of the software? What is the value of the product? Because, like, for example, like two people that have very diverse skill sets are going to be better at a some type of task than just one. person who's like absolutely best in class in that in that field.

Speaking of people that are best in class in their field, let's welcome Delian to the show. How you doing Delian? Bulgarian Mafios. Just carrying a bat, ready to break kneecaps at all times. I was going to bring that up to you. I was going to, I thought it was a rumor. But, you know, how is it going?

I think I want to kick it off. I mean, we'll get to the space stuff, obviously. The Delta V with Delian must go on. But what's the reaction to AGI arriving, April 16th being AGI day? Are you feeling the acceleration? Are you feeling the AGI? Do you still have a job or can AGI break kneecaps? You know, I do think that there are, you know, sort of junior research jobs that, you know,

probably got a lot harder to, you know, to justify both in, like, the land of, like, you know, sort of venture all the way to, like, you know, sort of grad school labs. The stuff is getting like really like really good. Like I, you know, I'm not somebody that has been investing in a ton of, you know, sort of AI slop, but I think it's, you know, sort of crazy to not figure out how to like integrate this stuff into your workflow. And like I very regularly now, like we'll read a paper and then immediately go to chat GPT and use it as like a little bit of like a quiz to make sure like I understood it. Or like, you know, I've been getting a little bit of like, you know, trying to take some like quantum physics classes and like the, you know, sort of nights and weekends.

And same thing. I like basically have chat GPT like be my tutor like structure which courses and videos I should, watch, like, pull, like, even though I can, like, go pull the homework myself. I'm like, go find the homework on, like, the MIT website and, like, pull it for me. And then I'm just like, like, submit the homework to you and you go check it against, like, the solution set. That way, like, it's, like, all these things that I just don't have to do. And it's sophisticated enough to actually, like, be able to read, like, a PDF of, like, paper that I put in, and it goes and grades it. That's, like, literally what, like, the junior grad students are

for at colleges, you know, like, to go grade the, you know, the P-Sets. So you're out of a job once the humanoid robotics drops, and then they can break the kneecaps and then you don't have anything. That's all I have value for, a founder's fund is just like, you know, I just like threaten people with violence in order to get money in. Yeah, exactly, whatever it takes. Anyway, so in your world outside of AI, what is the top story in defense space for the normies? I think Blue Origin was top of mind this week. We can talk about that. But on a serious level, like what is actually important for people to be paying attention to? Yeah, I mean, I think there's a couple of different stories that have, you know, come out over the past couple weeks

that are all kind of, you know, related to some of the Trump tariff stuff where basically what he's done is effectively like implemented the trade war that would have happened if a hot war with Taiwan had started to start to basically preview in some ways. It's almost like a military exercise in that it's basically like, let's go preview and see, you know, sort of what China, you know, would do if a hot war, you know, kicked off. And so they obviously have like immediately restricted a bunch of rare earth minerals. I think, the area that you know people should be paying more attention to is basically all things you know sort of seven conductor you know landscape um you know the you know articles that came out yesterday

talking about um you know in video basically getting you know sort of slapped on the wrist because they were like were exporting i forget what the number was but like 20 billion dollars per year of chips to like singapore um Singapore is obviously not buying 20 billion dollars worth of chips they were just immediately going and reselling those to China and so you know yansson huang is going to get a bit of a you know sort of slap on the wrist but you know the whole Taiwan story is all around that you know sort of semiconductor industry. And there was some news that came out about two weeks ago where the, you know, Chinese are working on a synchotron as a energy source for lithography and semiconductors.

So there's a bunch of different, you know, sort of components of the, you know, sort of stack of, you know, sort of semis. But, you know, to really, really, you know, keep it, you know, sort of brain dead simple, you basically have like the chip designers of the world. Think of those as like, you know, sort of the apples of the world with their like, you know, A1 chips as an example. you have the tools providers that make the tools that are necessary for semi-connecture processes by far the most famous there is you know sort of ASML in lithography

and then you have the foundries think of those like the factories that you sort of make the chips and there obviously TSM is you know sort of best in class you know the the tools and the like you know sort of foundry are where there's been by far the most aggregation let's say you know sort of in the market right that's where it's basically a monopsony. We're obviously, you know, sort of trying to break the, you know, sort of TSM monopsony. And obviously we're spending on chipsact, et cetera, trying to get this stuff reshore to the United States and get found re-set up here. There hasn't been a whole lot of effort, you know, domestically in the United States to really do anything around lithography.

We've just kind of accepted that the Dutch, right, SMOL I think is a Dutch company, yeah, are going to, you know, have that monopoly on lithography. And then people are starting to think a little differently now that it looks like that the Chinese maybe have an alternative effort. There's sort of one liner like, you know, sort of what is lithography? It's basically like you shine light at a mask. Some light passes through that mask and then it, you know, basically you know, inscribe something on a wafer and that wafer, you know, it's basically what you're inscribing is the chip. The way that like, you know, Dutch people do it is they basically like zap tin droplets really fast and really precisely. And it turns out when you zap tin droplets, they

release a very particular wavelength pretty coherently, and you can use that as the, like, what's known as like the energy source in lithography. People have theorized for a while on like, hey, are there other potential energy sources in particular, all the like linear accelerator, synchotron stuff that for the longest time was purely like physics science research. This is like, I think like CERN, you know, think about like Spark down at Stanford, right? All this stuff was just like, let's, you know, run particles really fast. Let's try and discover the Higgs boson and stuff like that. That's all that it was used for. And at some point people were like, man, like, this stuff is actually, like, getting pretty good.

Maybe we could actually, like, use this commercially because, like, the reason we zap the tin droplets is to get, like, you know, XYZ very short wavelength out of them. But, like, maybe we can just attune these linear accelerators to basically just be, like, really, really, really, really fancy lasers and get energy sources out of them. And so people have, like, theorized about that. And then it kind of looks like the Chinese have claimed that they've started to demonstrate that. And so now there's a bit of a panic in the U.S. where it's like, oh shit, like they could take over Taiwan.

And then in theory, we could try to ban, you know, ASML, you know, from, you know, something to China, but maybe that doesn't fucking matter because they've figured out their own way of basically doing, you know, sort of lithography. And so now the United States has got to think about, well, like, do we want to start to spin up our own internal effort that is, you know, using linear accelerators or synchotrons and try to decouple from ASML and decouple from Taiwan at the same times? We have Arizona doing, you know, the foundries, but then, you know, maybe we could co-op the sort of Stanford linear accelerator and, you know, sort of turn that into a lithography

shop and build a foundry around that too. So, um, anyways, that's the, that's the area that I feel like people don't, you know, sort of think about enough lately. Yeah. Is there, is there any, and I don't have context here, I don't know if you do, but is there any effort to get ASML to start manufacturing here in the United States, or is it just completely out of the question, right? Because we've seen the efforts of TSMC and video came out, uh, and it's basically doing I read as marketing they came out a couple days ago and said Nvidia to manufacture American made AI supercomputer super computers in the US for the first time pulling out the the supercomputer word but basically saying they want to make hundreds of

billions of dollars of their new Blackwell chips here in the US which is awesome but I just don't it's hard to read in how much of that is is sort of marketing versus like actually super real but I'm curious if you've heard of of anything on the actual ASML side in terms of U.S. operations. I feel like we just don't have leverage in that, you know, sort of relationship. Like, it's not obvious, like, what lever to pull. Like, at least in like the whole like, you know, sort of Taiwan, et cetera, I think. It's like, hey, this is a geopolitical enemy.

Like, you know, you know, all the Western allies agree generally, you know, sort of with that if you start to sell to them, we're going to like, you know, sanction the shit out of you. You know, Nvidia's like a U.S. company. There's like, you know, we can slap them with U.S. fines. With ASML, it's like, okay, this is like a, you know, sort of European company. and like if we start to try to like sanction or slap fines on them, they'll be like, okay, great, the very limited number of machines we were selling to the U.S., we're going to stop selling

them. We're going to keep selling to Taiwan because they're obviously, you know, by far our biggest customer. And so why does like the America have that much influence in this like Dutch to Taiwanese basically relationship versus at least with like, you know, Invidian crew that's like a U.S. to Taiwan, you know, relationship. And we can, you know, sort of influence that. And so, yeah, I haven't, you know, I'm not even sure if I were in Trump's shoes, like what angle I would even try to take to like, you know, do that. And also these AISM of machines are like

so complicated to manufacture. I mean, it's like it's, it's extremely complicated to operate them, right? Like that's what, you know, so Taiwan has basically, you know, sort of figured out how to do. Manufacturing them is like even more complicated. And so somehow figuring out how to like pull that out of, you know, the Netherlands. Man, like that seems like a really herculean effort. That's where it's like, I think actually you're sort of better. off starting the blank slate on a totally different technological approach and just like domesticating that rather than trying to like you know try to force them to move manufacturing operations the U.S. How do you think about the role of the government in actually winning some of these really

key industries? Like we did this whole dive yesterday on China's investment in the domestic semiconductor industry and they've poured tons of tons of resources into that. But I'm always reminded of the the there's that segment in zero to one about cylindra how so how the government tried to kind of pick a winner they gave them this I think it's a five hundred and thirty five million dollar loan guarantee and any physicists could have told you just the basic math on the cylindrical was not going to win meanwhile China it's not that they were like let's let the free market win they they had the golden sun program they subsidized photovoltaics super aggressively but they just made were they just smarter and they funded the right thing or do they do they do they

do something differently. Like, like, it feels like I want the linear accelerator, if that's the right path in the tech tree, to happen. And I want to win. But also, I don't necessarily want us to just, like, get scared and memetically invest in exactly what the Chinese are building. And there's, like, this dance of, like, you know, low taxes versus, like, okay, sometimes the government does need to step into these things. Like, how are you thinking about the role of the government in, like, driving R&D right now? Yeah, if there's a way to, like, you know, sort of distinguish between the two potential strategies, government can take. one is like, you know, sort of subsidize inputs or, you know, you know, match investment efforts

or something, you know, broadly in an entire market, but don't take a particular stance on, like, what technology tree needs to be a part of, just like subsidize this portion of the market. And then the other half is like subsidize a very particular, you know, path in the technology tree where there is no market whatsoever. And so if you did like market level subsidies, I think the right answer is like, you know, probably a balance, right? where, you know, on all things, let's say, like, you know, sort of rare earth, you know, sort of minerals, there feels like there's sort of market level things that one could do.

Just like slash some amount of regulations, you know, you know, encourage, you know, or, you know, guarantee some level of, you know, sort of purchasing power from the U.S. government on, like, certain rare earth minerals. That way you say, hey, however you refine lithium, you know, whether it's, you know, sort of, you know, pulling it from the slag from the Great Salt Lake or sort of mining in eastern California, great however you do it. Like a reserve is what you're talking about. Oh, go ahead.

Like building up a reserve. Yeah, exactly. Like we're going to build up a- but it has to be American made. But we're only going to buy it from domestic suppliers and we're to buy it like this like fixed rate, right? Yep. In relation to space stuff,

this is kind of what, you know, they've done with like the lunar payload services program, right? NASA has just said, I don't care how you build your lunar lander, but we are just going to guarantee they we're going to buy X amount of basically like lunar payload services and it's going to be at XYZ price and up to you how to go to do it.

And so I'm definitely like a big fan of those approaches. I do think there are the occasional warranted, you know, hey, here's a huge next step in a tech tree. It's just totally uneconomically viable for anybody to step into that tech tree without knowing that there's going to be government support, you know, sort of basically from the get-go, right? In some ways, like, you know, nuclear weapons were that. There was never going to be like a private company that was going to go invest into, you know, figuring out, you know, sort of the fission or fusion bombs without knowing that there was basically going to be government support. I do think that, like, linear accelerator is sort of one of those. now there is a big question of like, is it's Lindra, you know, is it, you know, fearfully copying the Chinese memeticism.

But it's like, I'm not sure if there's any other option where, you know, I'm not sure that there is any other credible, you know, approach to domesticating lithography. Like, I don't think that like trying to just copy the ASML tech tree makes any sense. I'm not sure that anybody else has come up with an energy source that like, you know, sort of is an option. So there probably needs to be like some, you know, sort of bet. So if we were to assign XYZ budget to like domesticating foundries, I mean, probably 10% of that equivalent budget should be assigned to trying to domesticate lithography.

And there's really one bet on that tech tree basically, you know, sort of to make. And so I mean, it's kind of a lame answer, but I think the answer is like there needs to be a little bit of a blended approach to like, you know, national reserves, guaranteeing demand in, you know, refined lithium domestically in the United States. There's also some questions on like, look like, you know, with like, let's say magnets. We do a lot of like, you know, the rarest mineral production of those, you know, magnet precursors. but the actual like, you know, sort of final synthesis and refinement, that all happens in China. But a part of why it happens in China is like the current processes are like crazy destructive for the environment. And so it's like, it's like actually a question of like, I mean, would you want that here? Like where, how?

Do we want to like try to come up with a different process? But it's like better for the environment, but like 10x more expensive. Like, you know, there are questions like that where it's like, you know, anyway, we're kind of looking into like the magnet supply chain recently and as we were debating an internal investment. It's like you look at that step in the supply chain. And I'm like, man, it's like not particularly technologically difficult. It's like really easy for us to replicate, but it literally is just like releases poison into the environment. Yeah.

Taking it back to space, you know, it seems like the helicopter on Mars, the massive James Webb telescope, like these are things that are not economically viable. But I could totally see them paying off and being like, oh, that was a great investment. even just like inspiring the next generation, but also like probably commercializing some technology down the road. On the flip side, we saw Blue Origin with the space tourism thing. It's like barely space just past the Carmen line. But how do you think that plays into the overall like space economy? Do you like that strategy of growing that business? Is that a business that can stand on its own? Or is it just kind of like a fun side project for Bezos? I don't know what you know about that whole like industry as it's growing. Because we've seen it kind of play out before with Virgin Galactic.

And I'd love to know what you think about space tourism generally. Yeah, I mean, I want to also touch on like that. You know, just an earlier point that you had around like the helicopter on Mars, James Webb Telescope, et cetera. I'm not sure if you know, but the White House actually recently released its proposed NASA budget after future, you know, nominated administrator, you know, sort of Jared Isaacman after his confirmation hearing. And in it, they, you know, significantly slashed all, you know, sort of science, you know, sort of missions at NASA. And so that includes future telescopes that we have in the works. That includes, you know, some of those helicopter on Mars missions.

And so the White House is at least taking a stance on that is not something that, you know, we see as being, you know, sort of valuable. Those are definitely, you know, sort of things that clearly don't have any, you know, sort of economic, you know, sort of feedback loops, like more deeply understanding of the universe. You know, I don't know if you recently saw, we, like, found an exoplanet where it looks like there are, you know, organic compounds that, you know, on Earth are only, you know, made by marine algae. And the planet, you know, is very large and in the Goldilocks zone of a Red Wharf star that would be at a temperature zone where, you know, sort of liquid water would be possible. And so it's like, yeah, like, what is the sort of value of that, you know, hard to, you know, sort of prescribe economically.

And the White House has at least thought, hey, we don't see the geopolitical, you know, sort of significance in this, doesn't get boots on the moon. And it doesn't return, you know, sort of to an economic use case. So that's at least the White House stance, which is also for what's worth, it was different than what Administrator Isaacman said in his confirmation hearing. So, you know, it's interesting to see that obviously there's, you know, maybe some disconnect between the administrator and the White House. On all things, space tourism, I don't think it's, you know, sort of an economic, you know, sort of use case that is going to be what really, you know, sort of drives us to the frontier. You know, whenever I've thought about Varda, I always go back to this 15th, you know, sort of century analogy, which, you know, sometimes people think is a bit of a stretch. I think it's actually pretty reasonable.

But if you look at the early 1400s and the Portuguese and the Chinese Empire, they both were in the early days of investing into basically their naval capabilities. And they took, you know, sort of radically different approaches. The Chinese basically took the approach of build these very large and ornate ships and go and sail along the African coast, you know, capture elephants, you know, bring back some gems and, you know, bring them back to the Chinese emperor as basically displays of power, right? And so, you know, the Chinese emperor would now have a pet elephant for the first time.

They would have these gems. They would have these, like, African trinkets. The Portuguese took a very different approach. The Portuguese built very small naval merchant flotillas. They would basically, you know, dock along the African coast. They wouldn't tour the entire coast like the Chinese did. They would say just stay there, make a trading outpost, basically find either sort of economic means and reasons for that naval flotilla to be around. And then only once that initial merchant outpost was established, would they move basically 50 miles west,

down the coast until they eventually covered the entire coast. So fast forward to the late, you know, sort of 1400s, where did these naval empires basically end up? The Chinese emperor ended up deciding that this was basically a waste of resources. He didn't see any economic value for his empire for having elephants around. And so ultimately basically shut down the entire naval, you know, sort of power for China for the following, I think, almost 300 years. The Portuguese, as you could probably guess, became the most powerful naval power, you know,

in the entire world and led to significant, you know, growth of their empire. And so I think about, you know, sort of space, you know, sort of pretty similar to the, you know, sort of naval frontier in the, you know, sort of 1400s and 1500s. I think you want to find these, you know, sort of clear economic use cases that, you know, involve trade and resources versus the like, you know, sort of Katie Perry going above the Carmen line is probably kind of like the elephant for the emperor, where, you know, Bezos gets to, you know, sort of pop some champagne. And, you know, I'm sure his wife is super happy that, you know, Katie Perry feels indebted to, you know, so them as a couple. but like, I don't know that that generates a ton of economic value. And so the economic incentive around space is obvious. Let's maybe talk about the national security incentives, specifically China in 2030. They've come out and said very clearly, we want to put our astronauts on the moon.

You know, I don't know if it's by 2030, but in the early 2030s. So maybe, you know, how much can you speak to? Do boots on the ground actually matter or would it be viable for them just to send up so many landing zones that they start building? We've heard about this like when you land, you kick up a bunch of dust and so you're kind of de facto claiming an area and there's only so much ice water there. So like if they could just send a bunch of like rovers, they could effectively start like claiming and they may not even ever need to send humans to really claim the resources. What do you think? Yeah, let me give it this sort of, you know, sort of China bear versus China bowl case on this. The China bear case is this is trying to repeating the same.

stakes from the 1400s. So they're going there. They haven't really figured out economic use cases. They don't even have even really figured out economic use cases in Leo, right? They have nothing equivalent to like Starlink

generating billions of dollars of commercial sort of revenues that you can use as infrastructure to build off of, right? And so they're the sort of Chinese with, yes, they're building big fancy ships. Yes, they're landing rovers on the backside of the moon, but they're not connecting this into capitalism whatsoever. And so this will all, you know, sort of fade away

versus the United States of the Portuguese. While we maybe haven't more recently succeeded as well as they have on the moon, we've got Starlink, we've got starship going up, we have all these things that are built off of commercial infrastructure. And so while we may not have as big of a ship soon, if you fast forward over the course of the next 30 years, the Chinese will shut down their program and will be the most powerful naval empire, e.g. space empire in the entire universe.

China, you know, sort of bull case. Nobody's landed, you know, sort of on the, you know, sort of backside of the moon ever and returned samples. perhaps they're finding helium-3 deposits, they're finding lunar ice deposits. They're taking a much more concerted approach to trying to get human boots on the moon and forcing it from top down. And if they get those first handful of boots, like you said, they can sort of claim particular land areas. And there's only a handful of craters on the moon today that are clearly known to have lunar ice.

If they claim those first handful, if they start mining that ice, they can turn into a propellant, they can turn into economic value where they can send that back into low Earth orbit to some of their satellites. they have set up there. They're setting up, you know, sort of Starlink, you know, sort of competitors. And so there's definitely a world where, you know, they establish the first, you know, sort of permanent lunar presence while we get distracted by this dual path between moon and Mars, that permanent, you know, sort of lunar presence turns into a mining operation. And so they do connect it back to economic value, even though it has a lot of like, you know, strong, you know, from top down state power, you know, getting it established. It eventually transitions to something, you know, sort of fully

economic. And so, I think there's a way to, you know, sort of take both those angles. I tend to think, I think you should never under. estimate, you know, sort of your, you know, enemies. And so I tend to take the China bull case and assume that is the default. And I think we're a little, you know, sort of, you know, lost in the rudder right now. I think, you know, it'll be interesting to see how, you know, administrator Isaacman, you know, so it comes in and starts to put together what our plans are for, you know, getting boots on the moon, you know, over the course the next four years. Do you have any insight into, like, the timeline of actually generating propellant on the

moon? I've heard, like, oh, there's water. And at a certain point, you can draw out the chemical process for turning like carbon and hydrogen into some fuel, but like that's hard to desalination is hard in America, in California. I can imagine that actually taking the resources on the moon and turning the propellant is like a massive industrial process that looks more like a Tesla gigafactory than like a little science experiment. But maybe I'm wrong about that. Like is this something where we could actually build a box that starts doing this, get it up there in the next decade? Or are we talking about like 20, 30, 40, 50 years? I think this is where like exponential equations always either sort of catch up on you in a way

that's either sort of unexpected. If you look at the total mass to orbit, basically over the past decade, it's basically following a perfect exponential equation. Yeah. And so I think this stuff is, you know, sort of way sooner than people expect. Like I actually think with a concerted, concentrated effort, I think we can get boots on the moon again before, you know, Trump is out of office. Wow.

And it's not like, you know, sort of betting on a bunch of, you know, sort of radical programs. it's, you know, Starship maturing at the, you know, sort of rate that it is, maybe a little bit more aggressive, you know, sort of the current, you know, Artemis vehicles, Orion, et cetera. And some of the current lunar lander company is starting to develop, you know, sort of larger landers. I think that's all, you know, I'm not saying that's going to happen like in the next year or two would be like late in the Trump, you know, sort of presidency. But I think it's possible. On the like conversion to fuel, not that's like complicated in process, there's like five or six groups that are sort of already, you know, sort of working on it. You know, I'd say the desalination thing that's, you know, sort of more of an economic challenge than. And it is like a technical, you know, challenge to like basically, you know, justify the amount of water that you need. The benefit of water on the moon is like on a per kilogram basis. It's so incredibly valuable because it's just so much easier to shoot water from the moon into low Earth orbit than it is to bring water, you know, from Earth surface up into low Earth orbit. And even to start, you don't need anything more complicated than water. Like you can use water. It's not a very good propellant.

But you can basically just spray water. And it is a rocket, you know, propellant. Now there's, you know, better things you can do. You can turn it into, you know, atomic hydrogen and stuff. like that and hydrogen gas or liquid hydrogen, but you know, you can start with very, you know, sort of basic things. And so, you know, I think it's like an end of decade problem totally solvable. Now, it depends on, you know, taking a really concerted focus effort. And again, maybe a part of is you do have to sacrifice the Mars helicopter. You have to sacrifice, you know,

some of these, you know, sort of telescopes, et cetera to get the country really, really focused on getting lunar infrastructure set up as quickly as possible. What's the probability that you or I go to the moon in the next 20 years? specifically URI? Yes. What's your P moon? I'd say your P moon is like you have 35% is my P moon. I've been talking to my son about that.

We look up at the moon and I'm like, do you want to go with me? And he's like, yeah. Of course. And I'm like, I think it's possible. I think it's possible. If you extend that to 40 years, I would put like P moon at like 98%. Awesome. So wait, there's this conspiracy theory that the Blue Origin launch didn't happen, that they didn't go

space, that it's impossible to go to space, using only family-friendly language, what would you say about the people that don't believe they went to space? Well, there are certain conspiracy theories that always have some truth or merit to them. JFK probably was killed by the CIA. NASA may have had a soundstage for the lunar landing as a backup. up. And, you know, Kitty Perry probably is not an astronaut. You know, you could probably just, you know, fly a fighter jet, you know, up and then down. And you probably roughly get the same experience that she did. Very different to be going, you know, a little up, a little down versus, you know,

15,000 miles an hour. So I'd say, God bless his conspiracy theorists. They're always helping us find the kernels of truth in the stories that they leave. Yeah, that's legitimate. Going back to space and geopolitics, China has their national space day on April 24th, so next week. Is that something significant in the space community that people are really paying attention to? Like, they feel like there's going to be like real signal or exciting results? Or is it just kind of, do people kind of write it off or not write it off, but just, kind of look at it as this sort of like kind of like a demo day kind of marketing keynote style event where where it's not a lot of signal they're delivering like yeah if you look at you know

sort of China's you know sort of plans from 2018 and sort of where they are today they're not totally off track on either sort of what they promised they'd be able to do right they've established their own you know low earth orbit chinese space station they're regularly flying astronauts up and down to it with cargo they're attracting international partners um to that space station. They've clearly gotten a coalition of folks willing to sign on to their, you know, sort of lunar station or lunar base plans. And so I do think it's something that, you know, people pay a lot of attention to.

They tend to, you know, sort of, you know, re-update or re-reveal basically what their next, you know, sort of five, ten, 15-year goals are. And it's something that everybody from like, you know, sort of NASA to Space Force definitely pays a lot of attention to. They're not always fully transparent about obviously, you know, sort of what they're up to. There's no guarantees that there'll be anything new. But I would expect that there will be given, you know, all the tensions that have risen over the past year between the United States and China. And so how do some of these partnerships happen, right? So one of the things that they're going to be reporting on next week is related to a partnership they have with France on like a joint satellite program.

Is that a is that is that like due to NASA's shortcomings? Like is that a missed opportunity for us to not be partnering with France on something? like that? Or like, how do these sort of like, how do the, and are the geopolitics of space almost like disconnected in some way between, uh, with Earth? Oh, no, very connected. That entire French thing I'd basically, you know, sort of pinpoint back to the nuclear submarine contract. Basically, the United States ended up moving a chunk of nuclear submarine contracts out of France, uh, over to Australia. The French were furious. McCrone, you know, sort of went on and, you know, gave a bunch of public talks about it and then basically like the you know your Chinese collaboration started less than a month later and so it was clearly like a hey you're going to slap us on the wrist as an ally great we're going to go to

your geopolitical adversary um I don't think obviously like those two things are you know one it one is like purely economic like we're moving the place that we're making submarines the other one is saying like we are partnering with your ally those two things feel like one is like a business decision the other one is like very clearly like you know picking uh a subsidize I mean, from France's perspective, it's like, you know, it's billions of lost revenue for one of their, you know, major defense primes. And, you know, they see that as a, you know, sort of slap on the wrist. And so, you know, they feel like they need to, you know, sort of slap back. And so, yeah, I'm not sure that, you know, it's the smartest geopolitical move. I don't think, like, you know, a Western democracy should be, you know, sort of cozing up to, you know, an authoritarian dictatorship. But I also don't think the French are the most intelligent people in the world right now. Well, they have the opportunity to turn it all around because if they don't, I will be boycotting Dom Pering, Jan. You heard it here first. Then that will crush the shockwaves through LVMH. Through the French economy. I will not be vacationing in Leon or Nice or Burgundy.

I will not be spending my time on the French Riviera. I'll be in Nice late June, no matter what, no matter how I dumb the French House. Cancel it. Cancel it. Unless it comes back to America. No, he's going to turn him. He's going to turn him. Yeah, yeah, you got to go over there. You're our plant. I'm going to check it on there. I'm going to look at those Airbus facilities until used and take a look at that Chinese satellite project. Wait, last thing, if you don't have a hard, hard stop, how do you think about the Chinese aerospace industry, specifically?

They earlier in the week, they canceled or paused some Boeing orders. Someone else on the show was saying, well, they'll probably just move those orders over to Airbus, something like that. but at the same time, how long is it until we're all going to be, you know, boarding? Is it really that hard to build a jet engine? I feel like they should have copied that by now. This is kind of a bear case for them. Well, up until a year ago, my, like, one line would have been, you know,

there haven't been any commercial airliner deaths in the United States for like 16 years. And so the reason that it's really difficult is because just like the safety bar is so, so high. They're basically nobody wants to touch anything because they're like, everything's been perfect. Yeah. Now, obviously, you know, sort of DCA and, you know, sort of Toronto. know, well, Toronto, I think it didn't end up bleeding any deaths, but DCA obviously did. And I'm not sure that it's really going to change the culture, like, you know, FAA aviation, you know, regulations are just so, so stringent because they do have in some ways a great track record. But, man, it just sets the bar so high.

Will the Chinese ever really manage to get any traction? I don't know. It's like definitely not in the United States ever. There's no way that, like, the United States is ever letting the United States airline buy, like, a Chinese, either a jetliner. It's the Comac C-919. Yeah. And there's a way France is going to let them, you know, buy anybody other than Airbus. Same thing with the entire EU.

So who the hell are they going to sell their, like, Chinese airliners to you, like Africa or something? You know, I don't know how big of that. Malaysia, maybe Malaysia teleporting. Oh, did you see that there was some, you know, sort of revealed potential documents of four UFOs surrounding the Malaysian airliner and then disappearing it? Okay. Well, we'll have to dig into that next time. I got to get Jesse Michaels on. I mean, we can do it another half an hour.

I get Jesse to come. Yeah, yeah, yeah. We'll break it down. I really appreciate your stopping by. Thanks so much. This is great. See you next week. We'll talk.

Cheers. Bye. Next up, we are going to the founder of a company that Delian is in love with. He's obsessed with these AI apps. He can't get enough of them. So we're very excited to have Garam on from captions. I'm actually a huge fan of this app and this company. I use it all the time.

Anytime you see me post a clip on X and there's captions there. I'm using the captions app. That's right. It's fantastic. And there's a bunch of other cool features and the company has been just maniacally adding functionality. And as you saw with chat GPT, you wanted to generate a video. Like it couldn't do it. Like these tool usages are increasingly difficult and they're not something that you can just, it increasingly looks like you can't just one shot it with tokens out of an LLM. And so building up traditional software around the

system actually improves the AI. And so these products are very complimentary. So we're excited to welcome him to the show. How you doing today? Boom. Hello. Hello. There we go.

What's going on? How's the going on? It's great. Thank you so much for being here. Thank you so much for creating the captions app. I use it very, very regularly. DAU, basically. I'm obsessed.

Yeah, yeah, for a long time. I mean, we make a ton of clips you've probably seen. We make a few clips. It's always a hassle to put captions over it, but you made it much easier. A lot of that, I want to talk about the history of the technology, the whisper turning point, and then get into value creation and the application layer. There's so much we can talk about. But why don't you kick it off with just like introduction on like how you're describing the company these days and then the most recent announcement? Totally. I mean, so love

that by the way, you're an OG user. So thank you for being a user. But, you know, it's a pretty young company. Like we're our product market fit was like two and a half years ago, right? So like we were four people two and a half years ago and things have grown really, really fast. All it's hard to with adding captions, you know, your most basic thing. But it evolved quite a bit. And so today, I mean, the way we think about the company today is like, you know, video creation is hard. Right.

And we've identified two problems, actually, right? One is recording the video is hard. You guys know this, right? And editing the video is hard too, right? It's pretty technical. And we want to solve these two problems. Like these two problems, we want to help people jump over with AI, right? So if you want to edit video yourself, if you want to record video yourself, go somewhere else, basically.

Right. We're going to actually do it for you. That's the value, right? So not going after DaVinci Resolve, not going after Premiere, an entirely new market. Exactly, right. And we think of it very similar to Canva. Actually, like, you know, Canva for Video as an idea has been sitting around for a while. And I think it's actually finally possible because of AI, right?

because the real value of Canva is you start with something, right? You don't have a blank screen. And it's not built for the designer, right? Most designers will look at Canva be like, I could probably do better than this, right? But it's built for the person who's not a designer, right? And that's the same value that we provide. So on that note, right, like last year, we started working on essentially foundation model

technology for video generation and for editing, that's going to help us achieve that. And these are big projects, very expensive. but also very cutting edge. I think the most exciting part on the video gen side for me is like we're very much focused on like talking videos, right? Which is, and by the way, like, I'm also like kind of surprised

almost in a way that we've spent so much time and so much money doing text to video on silent videos, right? Like what's the point? Right? Like just talk videos. That's the negligible part of what a video is. Yep. Almost no time spent on videos

with actual communication. So that's kind of what we were focused on for last year. Yeah. What, what, like, I don't know, what, what, what, what, like, breakthroughs are, are the most interesting to you. Obviously, like, whisper is super important. I feel like whisper is great. And then all of a sudden it came down to, like, okay, well, I want real time whisper on the show

and then I got to go build that. Or, uh, notebook LM. Like, notebook LM, like, it still doesn't have an app, but even though Google is paying people not to work and stuff, it makes no sense to me. And I'm imagining that, like, captions, could be an app where I'm getting like a notebook L.M style like YouTube videos. YouTube's talked about this a little bit. They haven't rolled anything out. But you're starting to see a lot of this stuff. A lot of it's in like the slop tier. But what I like about captions is that you can

still inject enough of the human element to take it from, it's a tool that's used in partnership with a human. So it still has that art in there. But what is exciting you and what's the most interesting in terms of like where you want this to go? Yeah. I mean, honestly, like there's a pretty clear distinction that's developing that I'm starting to see, which is like amongst the foundation models, right? Like there's the text generation like LLM type models, right? And those are solving a very difficult problem is intelligence, right? Like an unsolved problem. And no one's solved intelligence before. Right. So, and by the way, we don't even know what the bound is, right? Like, where does it end? Who knows? It could never end. Right. It could go on forever. Right. And then on the other side,

you got, you know, media generation. Like this is everything from like video generation to music generation, sound, audio, like all this stuff, right? These are solved problems. Like, we can do rendering today, right? Like, we can literally render anything you want with like CGI and stuff, right? Yeah, of course. It's just becoming a lot easier and it's also bounded, right? Which means that there is a limit of realism, right? And then you've kind of solved it essentially, right? You can't get more real than real. And once you're there, you kind of have achieved what you set out to achieve, right. And so I think it's a different type of problem. And also it means that it's not about replacing the human, right? Because what's actually happening is the craft is evolving, right?

The craft is different, but the creativity is still there. Whereas on the LLM side, that's actually potentially replacing the human. Not going to lie, right? That's potentially what it's going to do. Right. So these are two different types of use cases, almost two different types of value that are being produced today. I think in the LLM is we fall definitely much more in that sort of media generation category where like our goal is not to replace anybody right it's actually like empower a bunch more people potentially right and who knows right people come in they use captions to make their videos right and we edit it for them we like generate the video for them and you know that just gets them started but like two three four years down the line they you know they move on to

premier pro or something like that's awesome right nothing wrong right so how do you um on the developer side we're seeing a bunch of startups where it seems like everything is just converging into this like one shot right where it's like you have lovable bolts uh rippling uh sorry rplit replet rippling is uh one shot for your h r i s system that's right that's right um so everything's kind of converging onto this like text box where you just tell it what you want and then it makes it makes a website or an app or things like that uh i imagine content will maybe go that way for some use cases and that's kind of like there's a lot of sort of momentum and convergence around that moment. How do you see, but that's just, you know, my point of view. What's your point of view on like how all this evolves

and how you're looking to continue to differentiate captions over time other than just sort of like chasing perfect realism? Totally. Yeah. I mean, so a couple of things there. Like I think generally, kind of to, you know, making a comparison what you're talking about, right? Like, I think of it as like Canva for everything. That's actually what's happening, right? Because the magic of Canva, you know, awesome company, I think the magic of it is not about how simple the UI is

or something like that. The magic is that you start with something. It's not a blank page, right? And really the biggest enemy of anything creative is a blank page. And not the UI and stuff, right? I mean, think about like design software in general. You look at like Figma and stuff.

like, Figma has like six buttons, right? Like, it's not a hard UI, right? But it's really hard to make something good with it. Like, it's really hard to figure out how you use squares and circles to make something that looks good, right? And I think the magic of Canva is you start with something, right? You're already 90% there when you enter and then you kind of make some tweaks to get it to 100, right? And by the way, like think about chat GPT. It's kind of the same thing, right?

Like, we're using it all over the place today, but it just gets you started, right? Like, it's like, boom, I already have something. Like, I need a job description. Boom, there's job description, right? For whatever job you want, right? And then it may not be perfect, right? You make a few tweaks, you know, change things here and there. And you're done, right?

And so it's like Canva for everything. That's what's happening. And I think same for music generation or, you know, video generation. Like all these things are going there. Our goal and our mission in this, like we're focused on specifically the communications sort of vertical, right? So think about this, right? If you think about a movie or, you know, a TV show or anything like that, right?

Any kind of like media today, only a small part of that is B-roll, right? like if this was a movie like I'm in New York so like it might open with like a shot of the Empire State building and then the next scene like oh there's a New York taxi cab on the street passing by really quick in two seconds and then the camera's in the room and we're talking right and that's actually the movie right and so so much time and money has been spent on making the shot of the Empire State Building and almost nothing on like actually getting the dialogue going right and that's kind of the weird thing right and like our thing is like let's get that communication. that dialogue problem solved, right? That's one. And on the other side, just footage isn't enough. So let's get it edited to make it actually an asset, right? Can you talk a little bit about growth for captions? There's a weird dynamic where it can be extremely valuable to go viral with like a one-shot thing.

I'm thinking of Lenza, those magic avatars. It was just upload a couple photos and you get a photo of yourself. Then we have the studio Ghibli moment, which was a, huge growth vector. It's not Open AI's product, but it was still probably massively beneficial just to drive a bunch of extra installations and chat GPT use, right? And so I could imagine you guys thinking, like, hey, let's go make one-click Harry Potter Balenciaga generator. And like, we are just like really good at making like Harry Potter Balenciaga style videos. Of course, you need to put in your own tweak, but that's what we're great at. But you don't want to get pigeonholed into that,

but it can be a good growth driver. Are you thinking about a consciously, Are you thinking about, like, how can I get the next, how can I get the next Studio Ghibli moment to happen in the captions up? Yeah, I mean, so we are. But our philosophy on this, honestly, like, think about both the Studio Ghibli thing, but also think about, like, original Chat GPD, right? Like, I remember the time where, you know, GPT was available. Like, I would use, I would show it to my friends, like, check this out.

Like, check how cool this is. People would be like, oh, wow, cool. Okay. Yeah. Right. And then suddenly chat GPD came out. And like, by the way, I think it was very clear that they hadn't prepared for the amount of virality that thing got, right?

Like even the name chat, GPT kind of gives that away. Right. And so it wasn't a planned thing. It kind of just happened, right? So you give me the same thing. I don't think they planned it. Like, it just kind of happened, right? So I think if you create the right environment where people are given the creativity to go try

something that's like an awesome technology, right? They can play around with it, make cool things with it. Like these types of moments kind of happen naturally. It has happened for us several times with different technologies we've released in the past. And a lot of times it's been unplanned. It's just like, you know, when we plan for it too much, it doesn't happen. When we don't plan for it, it just like suddenly explodes, like completely, right? So that's kind of what we've seen. And it's something we think about of, like, how do we create that wow experience? Because at the end of the day, a lot of the growth

and virality is happening, but people are just blown away. It's just so impressive, right? It's beyond anything anyone's ever seen, right? And I think that is a pretty high bar. So building on that, building in private, until we reached that bar, releasing it as, like a wow this is like crazy like that's the type of stuff that we've seen work really well uh question do you have any sort of visions around uh what the just video content on the internet in 2030 because i have this uh right now there's not a huge incentive to make a video for one person right especially in a business context if i if you want to explain something for 10 minutes to somebody in a business context you pick up the phone you call them you spend the time to send

an email or whatever. Now with something like captions, it's like, well, I could just generate a video of like how my product works and why it's relevant to this industry and all that. And so I have this sense that content generation is going to like 100x,000x, but human attention is not going to 100x or 1,000x, right? It's just not possible. There's only so much time in the day. People use their phone for six, eight hours on average right now. They're still spending two hours a day on Netflix, but we're not going to suddenly like get, you know, 100 hours a day, even though people on TikTok entrepreneur influencers might want that. So I just have this kind of question around like, do you think the average video in the future

gets one view, two views? And maybe it's not the average, right, because certain views or certain videos will still go super viral and these sort of cultural phenomenons. But yeah, I'm curious if you think that's kind of like where we're headed. Right. I mean, I think for what it's worth, I think the average video today is getting probably one or two views, right? Because like think about Snapchat, like a lot of video, probably a billion videos a day, but sent to like, you know, a few people are seeing it basically, right, for the most part of our private communication. And honestly, like Snapchat pioneered that. Like they kind of missed the TikTok part of it. It wasn't part of their ethos, to be honest, right? But like, there were more about the private communication. But I think

the future is more video. Like, for what it's worth, like, you know, the way. And I think there's an interesting sort of like move that I think will happen towards more video in AI as well. Because think about like how communication has changed over time, right? Like we're not sending letters to people as often anymore, right? Like text messaging is kind of like very prone to miscommunication, right? Audio is definitely better phone call, goes a longer way. Video call is like one step further, right? And then real life meeting is even beyond that, right?

Oftentimes there's like even within companies like, right, there's miscommunication and like mistrust. trust that builds when there's remote teams or things like that, right? And you've got to watch for that. Whereas an in-person team just trust each other so much more. So there is definitely something to be said about like these more sort of multimodal forms of communication, right, to use the term. But I do think that actually even on the AI side, right, like, sure, chat GPT makes me a great writer. But like what makes me a great communicator, right? And we're really not thinking about that, right? because communication is multimodal in itself, right? Like the words that I'm saying right now, where I'm pausing, what I'm emphasizing,

how my micro expressions are moving, like how my body is moving, like all that is communicating in multiple forms, right, a message. And that message changes if I change any of those things. Like words might be the same, right? But I can change the message completely by just changing the delivery of it. Right. So I think today's technologies just aren't capturing, right, like how broad communication actually is. And I think it will all evolve towards video over time, just as we've seen.

Like, this is not new. We've seen this happen before. Right. So, and by the way, like, I was at Snap when, like, TikTok took off, right? 2019, that era. And initially, like, TikTok grew a lot on the back of Snap. And a lot of people know this. But, like, they were running, like, $100 million a month of ads on Snapchat, right?

Initially, right? There was no presence. But the Fox into the hen house. Yeah. And like there was concern, like in the company, people were concerned that like, are we, like, creating a competitor here, right? And people, we were running AB tests. Yeah. Narrator, they were.

Exactly. But like, we ran AB tests to test, right? Like, if someone sees a TikTok ad, are they less likely to engage in Snapchat? Test didn't show that, right? But the reality is that it wasn't true, right? They did spend less time with Snapchat. So, well, we got to run. This was a fantastic conversation.

Thank you so much for hopping on. And we'll definitely talk to you soon. Yeah, great, great talking. Thanks for coming on. Bye. We got Ian in the waiting room from Astro mecca. I believe I'm pronouncing that correctly. Astro mecca.

Great. Mechanica? Astro mechanica. Sorry, I miss type that. Astro mechanica. Anyway, we'll let him explain it to us. Come on in, Ian. Sorry for the wait.

Sorry for the wait. Yeah, how do you pronounce the name of the company? Let's settle this today once for a non. Yeah, astro-mechanica. Astro-mechanica. There we go. Can you give us a breakdown of what you do? And I'd love to hear, like, the brief history on the launch.

I remember there was, like, a video you posted of building something, maybe in a garage. You went viral. It was very cool. And there was some debate over it. And now it's a real company. You raise a bunch of money. So just take me through the little journey. Yeah, yeah.

I guess it starts with my background of, like, I'm lifelong aircraft builder, pilot, fly jets, the whole nine on that. Yeah. And, yeah. You flew, where, what kind of jets did you fly? Like private jets. I built, like, I built my first plane when I was 17. I did experimental airplanes, drones, and, like, the early 2000s before drones were

thing. So it was from that, like, I, you know, I think with, like, a lot of good companies, you're really just building the thing you want. Yeah. This is just not a thing that exists. Unsurprisingly, one cannot just go by, like, a bleeding-edge fighter jet. So, so, yeah, you know, going to that, I had always been obsessed in all the technology. interestingly I was working on like this hyper-electric architecture initially coming from the private jet world for lower cost of operation

It's just like simpler systems and it turns that there was a really neat technological unlock for Supersonic flight So that's where a lot of this kind of kicked off and it was an especially novel architecture I think this is why it was just so alien and like I had just been spending so much time and all of the The various like sub-components and in disciplines there it's like there's this really unusual combination when you combine all these things together you get this totally new architecture. So the big one for us is like transition up to what we call a ramjet mode. Yeah. So you know like there are really good solutions for like I really like what the Hermes guys are up to like there. Like there's definitely other good solutions out there. But

this, it hits this kind of interesting sweet spot where like we're getting a lot of performance with a relatively, I mean it's still millions of dollars, but like relatively inexpensive system. And the other thing that comes up a lot is like people are seeing all the engine stuff. And yeah, you know, some people think we're an engine company. It was more, it's like if you want to have a computer you have to make the microprocessor first okay uh but this round getting this done is like finally shifting it to us going to the aircraft development so okay so you're building a plane yeah yeah i mean that's my background like i was a plane guy that just needed a better engine that's awesome and so yeah we're we're at that stage now with the round done um you know and like people had always

seen me kind of like i was like this guy in san francisco that was like i had a machine shop at a manufacturing business previously so like people knew i was good at building stuff um But we've had just this, like, compounding effect of, like, the, the talent that I have been very lucky to accumulate here and just the people that have joined on. Because, like, good engineers see an interesting, like, technical solution and thing to work on. Yeah. So, I mean, I love my team. Everybody's, like, incredible here. So, like, we've now got that.

That then makes it easier to get more capital. Yeah. Then we've been spending some time in D.C. I mean, like, And, like, you know, it's also very important to have an actual customer. I think this is the other way where I came from more of a small business world. And like, I wasn't interested in just, I don't know, like developing things for the fun of it. Like I'm what I'm hearing is like you like maybe the company is like very clearly at this moment shifting from, you know, science project that you would have worked on for free to like, okay, like real commercial opportunity. Let's go.

Let's go full send. So yeah. I mean, in a nutshell, it's, you know, make the engine. So you go in the initial like demonstrator phase. So demonstrate the engine. back up the money to now build the demonstrator aircraft. The one we're pushing for is the world's first nonstop, trans-specific non-stop supersonic aircraft.

So we could do, you know, California to Taiwan without refueling in under four hours. Okay. So, I don't know. I mean, it sounds like direct competitor to boom. It's been a decade in boom. Not really. So, yeah, I mean, you know, Blake's going for like airliners.

And he, so we're, we're focused on, again, my world is more private jets. And really, so is this a Gulfstream? Are you coming for a Gulfstream or, you know, let's get specific. It's more like we're coming for net jets, if, you know, something like that. Sure. Cool. We're going to spend the next seven to ten years doing DOD.

So, like, you need to have an actual, it's so expensive to make an airplane. It's like, it's not hard to make a thing fly. It's hard to make a thing fly that is safe enough to put people on it. So my strategy on this is you go to a space like, well, you know, unmanned drones. So military were like technology and capability matters first. Yep, totally. You can put it out there. You can, you know, there's going to be things that are going to come up and you're going to learn it.

It's just, it's a cheaper place to learn. Yeah, yeah, yeah. Then you're in a position where like, well, you have an actual business. You've been making, I mean, the drones on man aircraft were making her like at the smallest 20,000 pounds. Like, they're not small planes. Yeah, yeah. And I'm curious, is the right place if you're developing a new airplane to start with, you know, autonomy just because in theory after you fully develop it, it's like, well, pilots, even be flying or they'll just kind of like almost like sit at it. I mean, I am a pilot, some

have feelings on that, but it is, it's definitely something of like no one, this is another kind of odd thing. No one has ever used autonomy in the certification and development of a manned aircraft. And I actually think this is one of the ways we can make it a lot cheaper, because ultimately it's about getting data and proving this thing is safe. The only way you really prove it safe is you've flown it. And so if you have a human on board from day one, you can't take the risks that you want to. Like, SpaceX proves this with Starship or you're like, you know, the actual cost of the hardware is not that high. So you're better off being really aggressive, getting things in the air, de-risking. And again, you're saying this a hermice of like they're

making, you know, like the plane is a bit more rough. But the point is like they can make a new plane every year. Yeah. And like by the time, it's like if you want to get to very good things, you want to do more iterations. So this is our similar strategy. I mean, we, we develop new engines every like four months. Yeah. And I think I mean, Waymo, waymo kind of took a similar path where like there weren't random passengers in the back for years and you'd see, oh, there's just a safety driver in there. And then eventually they pull the safe driver out and put the passenger in the back. Exactly. Yeah. You just want to spend time where you can map out all these weird little edge cases, things like that. So that's a big part of how we tackle this. And like it's so yeah,

you always want to go for like big lessons, very important lessons as inexpensively as possible. Because eventually you kind of lock into a point where like everything will be so expensive that you don't want to make any changes. It's just too costly to do it at that point. So that's where I think ultimately we can actually get to the passenger flight point quicker and cheaper by starting an entire unmanned aircraft business first because it's functionally quite similar in the technical challenges. And yeah, we can de-risk it all, learn all the hard lessons.

And, you know, in the meantime, like, it's also very cool just getting to make these things. So, you know, it's photogenic work. Everybody loves saying the engines on Twitter and stuff like that. No, it looks incredible. I'm curious, A16C and Z makes sense in the round. Lower carbon was a big logo there. Is that Saka and his team saying, like, this is really cool, we want to back it? Or is there a fuel efficiency, you know, kind of thing in the long run?

Yeah. So my very first check, institutional check into the company was actually lower carbon and Shao at lower carbon was my guy. Nice. And he was betting on it was just me in the machine shop. and I had, you know, getting close to that first prototype. And so, yeah, from there, the system, we get the range because it is more efficient. So, yeah, there's just an obvious efficiency argument here.

The other one for civil applications, not for DOD, but for civil is taking a page out of the rocket book. We don't plan to use jet fuel. So rockets have switched to LNG or liquid methane. If you just swapped that fuel, it's 30% less CO2 just as direct fuel swap. It's more energy per unit weights, 50 megajoules versus. is 42 per kilogram, a whole bunch of benefits. There's a reason the rocket folks went to it. So this is another one that's kind of insane bets for like, if you're Boeing or someone,

like you're not going to suggest a thing like that. But when you're starting from nothing, you can make these bigger bets. So, yeah, if you were to do the combination of the engine plus the fuel change, you're around 60% less CO2. And then there's guys, like, I love Casey Hanmer, Terraform. There's a lot of folks working on synthetic fuels. if you have an engine architecture around LNG, I think that's the best fuel for electrosynthesis. So if you want to have, like, inexpensive decarbonized fuel,

it's also your best option. So like options at first all for economics of like LNG is a tenth of price of jet fuel. So first make it affordable, then you can go for the clean stuff after that, where, I mean, it was already cleaner, and then you can just fully decarbonize from there. Or, you know, but first, stay alive as a company. So first solve our economics, then go for that. Smart.

A bit of a random question, but I'm curious since you're an aviation nerd. China earlier this week dramatically canceled some Boeing orders. They have their own internal aircraft. It's like Comac, the C-919. Do you see them like permanently trying to shift over to that? Or is it just so hard to build a plane that they're going to kind of keep coming back to Boeing and an Airbus over time? Yeah. Yeah, it's tough to say on that.

I would say, well, engines are harder, and to that end, we're, I think, you know, even they're using, just as we actually use Pratt and Whitney and GE components, things like that and what we're doing, that's sort of like being TSM. That one, I don't think they could get figured out anytime soon. And the Comac uses, you know, still American engines. So I'm pretty sure at least. So that one's harder. On the whole airframe, I think they could get it figured out, I'm sure.

It just depends on how, as with all these things. a money pit. How much do you want to throw into it? But why is it, so what's the componentry of like the J35 then? Why can they do that? But, but, which is like a, you know, defense application versus a commercial. It's so, again, to this thing of it's not hard to make a thing work and fly. I mean, I did this in my shop kind of as proof of like, getting a thing to initially go, not very complicated. The economics of air travel are dependent on it. never breaking. And that is a, you don't really know that, like, even Pratnit,

Pratnoen with a recent geared turbo fan is like, it's going to be like a decade until they get payback on that. And they know how to make things like that. So this is the sort of challenge you run into of, you know, what is going to come up after 5,000 hours, after 10,000 hours, and it turns out this thing. It's like, oh, there's cracking on this component, and now the engines. And, like, meanwhile, use what we've got right now. And you're like, yeah, it's good for 30,000 out, like the CFM 56 was like, it's like 50 to 60% of all airliners, narrow, narrow body ones. And like, that engine can stay on wing for almost 30,000 hours. And it's just so proven. And this is this thing you see in aviation where like, there's always these things that seem

really appealing because it's a performance optimization. You're like, well, of course I want to burn less fuel. But when you try to implement this thing that's technically better, it's like, yeah, but it turns out it broke at some slightly higher rate and you already had terrible margins and now it doesn't work. So that's where that that's going to come up. It's just not something that's like obvious on day one. Well, congratulations on the round. We'll get the size for you and the whole team. An overnight success. Thanks for coming on. Seriously, I appreciate you coming on and breaking it down for us. Thanks so much. Of course. Yeah. See you guys. We'll talk to you soon. Bye. Next up, we have

Shamsankar from Palantir. There's ton to talk about with him. He's a public 18 theses about the defense reformation, the primacy of winning, and also just sits in a very interesting place as Palantir's first forward-deployed engineer. And so I want to hear what he's hearing from Palantir customers, mostly. Anyway, Sean, welcome to the show. How are you doing? Thank you guys for having me. Yeah, thanks for being here. Yeah, I wanted to kick it off with what is the biggest topic of discussion over the last two weeks, tariffs, H-20s, and Nvidia, something else. What is driving conversations through the partners that you work with at Palantir? I think it's enterprise autonomy. It's like there's a normative view of the most valuable application of

AI is clearly towards autonomy in the enterprise. And then you can think about it, you can reason about it by analogy. If you thought about the self-driving car journey, it took us 20 years to get from a prototype that drove through the desert 120 miles, pretty good demo. And to do you. 2005 to a commercial self-driving car service. And no one wants to be on that 20-year journey. What have we really been doing over 20 years? We've been handling edge cases, right? So really investing in the tool chain that allows you to get from a mandraolic world

to one where you have AI agents that are totally automating your business. Like we've automated sepsis monitoring at Tampa General where deaths from sepsis reduced by half. We've automated how AIG underwrites insurance. What used to take three weeks and you'd only get to 10. weeks, 10% of your submissions, it takes less than an hour and you get to 100%. So I think, and it really is going to feed and does winner take most dynamic. It's not just about cost savings, about your competitive advantage. And you see that play out in very stark in real terms and defense, where in some

sense, there's nothing new under the sun. There's just John Boyd's Udoloup, Observe, Oria, decide act. Or sometimes my favorite admiral says the American Udolup, Observe, Overreact, Destroy, Apologize. How do you bring lethality for that? It's just doing it much faster and you're going to do that way faster with AI. That's amazing.

Tyler Cowen called April 16th yesterday AGI Day. He has called it. What was your reaction to the open AI? Probably happened internally at Palantir like months ago. But yeah, I mean, how are you processing this idea of AGI?

Obviously, it accelerates everything you're doing. But what is your take on model scaling, reasoning, agentic workflows, all these different things. Like, what are you looking for? Where are there breakthroughs that remain versus what are we just need to go implement? Well, I think we're well past the threshold where the models are powerful enough that you can be using them to automate massive things. I think at this point, really incremental improvements to the model changes how you decomp the problem. How many agents do you need? Do you need 80 because you've decomped in a way where it succeeds or do you need eight? Like the coolest stuff that I've been seeing is like multiple teams

internally at Palantara of building AI FDEs. And it's really compelling. It gets really far. It attacks different parts of the stack. So like if every user could have their own on demand, infinite essentially, AI FDE, like how much more stuff can you build? How much more quickly can you adapt? What's your Oudaloupe as a company now? So I think we're deep in the implementation phase. If AI is electricity, like in electricity, all the value didn't accrete to the people who made the turbine generators. It went to the people who made the tools that ran on electricity. And I think there's just, We're just so excited about proliferating and building those tools out. Yeah.

How are you thinking about just commoditization of the model layer? Obviously, it's like a horse race every different month, the different models comes out. How has Palantir approached integration with different foundation models? And then is that the same as other kind of databases and different tools that might be deeper in the stack? Or like the, even the migration to the cloud. Does this feel like cloud to you? Does this feel like mobile to you in terms of like your agnostic approach or is it different in any way? There are parts that definitely rhyme with it.

It's more agnostic than not. I said early on in the AI revolution that we thought the right approach was K LLMs. Like why would you pick one LLM when you can have K? And then you really can start thinking about the tool chain that you need to build around that. There's numbers, there are a number of reasons at first. Like very clearly you had model commoditization. If you look at both open and closed models over time, you know, the ELOs are just up into the right. The open ones have converged and even in some cases surpass closed models. And at the

same time, the price of inference is dropped like a rock. So that's clearly happening. And you even see that in the frontier model companies where they're expanding further and further into the app stack because they realize that selling you a raw API is probably going to be a raw business. Yeah. So then there's a question of like, okay, if you're just being very pragmatic, if you're building the machines running on this, like what model's right for what job, the models are improving rapidly. How are you going to safely be able to evaluate the relative performance of models as new models come out?

So you need that sort of tool chain. And even more deliberately, these models get end of life. Like you can't get the original GPT4 anymore, right? You have 4-0. And so if you've built an entire enterprise that runs, assuming some model is going to exist in perpetuity, that's probably not going to work out that well for you. So you're going to have to have this constant ability

to evaluate and run these models in parallel to develop the conviction you need, not only for the optimization of what's incrementally better, but can I safely migrate in the future? How are you thinking about integrating robotics? It feels like AI today is just rapidly transforming the way work is done online, specifically knowledge work, but how are you thinking about, you know, what's your vision in 2030 and kind of beyond about, you know, how different robotics, systems are integrating with the Palantir system. Specifically, the example you mentioned earlier around sepsis monitoring is like, cool, that's one sort of hardware integration, but it's not its own autonomous, you know, system out in the world. In many ways, we already do this. You know,

we have like more than 300,000 workers, blue-collar workers who turn wrenches in our software every day, everything from the factory floors of Chrysler to every Airbus Airframe, every HD Hyundai ship. If you think about a company like Rio Tinto, so much of the mining operation is, you know, is actually autonomous. The railroad cars that take iron ore from where it's mine to the ports, that's completely autonomous. The three-story tall dump trucks that are actually trucking out the ore, autonomous. And so I don't want to trivialize it.

But in some sense, to me, it's more of like a difference of degree than kind. You're machine-to-machine communicating to a system. That system just happens to be smarter and smarter every single day. During the metaverse boom, Satchinadella was talking about building digital twins. is that an unnecessary abstraction or like reference point for humans? Or is there some actual value in representing all of these real world assets like the Rio Tinto mining example you gave in some sort of like virtual space? Or does it not matter?

And it should all just be weights in a model that we're querying through an LLM or something like that. Where I think it gets to be really valuable is if you kind of thought about it like a CI check in programming where it's like I'm posing a change to a system here. How can I understand it? like, is that change going to work? What are the unintended consequences of it? If I make that change, what are the new bottlenecks that different functions have to think about? The simple example I always use is the procurement guy is really excited because he bought, you know, discount raw material 30% off list price. And the production guy is pissed off because this discount material has 40%

less yield, right? It is at the end of the day, one value chain. And where the digital twins have been hugely valuable is the ability to integrate the chain and the decision making across it. So, you know when the left hand is robbing from the right hand. Can you talk about the early days of Palantir, maybe one of the first major wins or setbacks or kind of like what's the story that you tell to new Palantirians to kind of set them up for maintaining the culture? Because I feel like Palantir's done a great job of like maintaining the quality bar. You haven't become that place where people kind of go in like rest invest basically. But you are like kind of a big tech company now. what's the story that you tell to kind of set the culture?

I mean, there are so many stories. I really, I'll tell you how I set the culture at the end, I promise. But I think one of the things that I felt like we kept getting punched in the face in the early days on is that someone else's execution would end up screwing us. Like I remember we had this gatekeeper between us and a government customer and he installed an early version of our software and he wanted to test it out before he passed it on. And it crashed and he crashed. I mean, in those days, this was a server with two gigs of RAM trying to run a four-gig Java heap, and he didn't understand why it crashed, right? It's like we just developed this extreme ownership mentality because anytime we outsourced

even an iota of responsibility, it blew up in our face. And that's, you can see forward-deployed engineering as like the extreme manifestation of that, that we're going to somehow have total control over the implementation because that's how you get the feedback and the quality and the improvement, and you're actually responsible for your own destiny. But the story I tell to kind of spill the beans on. So your first AMA with me, if you're onboarding in week one, I always end the AMA by reminding people that Pounder is a flat place.

What does flatness even mean? Well, to me, it means that every single employee is willing to tell me to fuck off to my face. And I all say fuck off in unison out loud at the very end. I think it's important. I want to institutionalize the notion of rebellion that I don't have all the right ideas. There's so many things we've done over time. I didn't think we're going to be right, and they were right.

And, you know, you got to bet on talent and the people and give them the space to run. And really preserving what's at the core of this is like, this is an artist colony, not a factory. You know, I don't really know what your career progression is going to be. And if you want certainty on that, this is definitely not the right place. But I can promise you you'll have access to the most motivating problems and compelling colleagues. And I'll give you all the canvas and paint that you need. Yeah. Can you talk about what's,

going on with the cultural transformation in Washington right now? You've written, like, maybe it's transformation going into founder mode, but within some of the more nitty-gritty, maybe swampy institutions, is there a need to be able to tell each other to F off or just be more confrontational? What should D.C. learn from Palantir and maybe even vice versa? Well, I don't know if D.C., I would say it seems like very presumptuous for me to say, what should D.C. learn for Palantir, but what it gives me about the present moment is, in Pallensure terms, the primacy of winning.

Like, you feel that in the people there. They understand that they're working backwards from what could actually work instead of some anodyne notion of how we wish the world works but doesn't. And that's like allowing us to reexamine a lot of assumptions for first principles. I think Founder Mode is the best description of it. When I talk about,

conserved across commercial and government, kind of our diagnosis of the current legitimation crisis. Why did doors fall off airplanes? Why does it seem like these institutions aren't working? You have a C-suite that is, if you steal-mand it, diligently trying to steer the ship. And their steering wheel, what they don't realize is a prop from the jungle cruise ride in Disneyland. It's not connected to anything.

And then you have people, hardworking people on the metaphorical factory floor who kind of look up and say, how could they be so disconnected? How could they not understand what's actually happening here? And so much of that is like the levers of orientation are somehow filtered through a lot of people in the middle. Leadership doesn't have access to real-time information. If you couldn't, you know, like if you were navigating a car and the latency of understanding where you are was an hour, you'd crash. And so like this sort of manager mode playbook, it works really well for the managers, but it's like a way of having a well-managed company into the ground. Do we need more like individual champions?

Peter Thiel has that quote about like we need ticker tape parades. Ezra Klein is now talking about it with the abundance. He's he's highlighted the failure of the California high speed rail system. And what I found interesting about that is that it's very hard to pin down like California high speed rail. Everyone kind of agrees like projects not going great. But no one can really say like who is even the champion of that at any point in time? Like no one's really responsible at any point in time. Does the government need more individual accountability or like, I don't know, even like, project level CEOs or something? I don't know.

Yes. The short answer to that is absolutely yes. I think about it in terms of heretics and heroes. You know, it was Somerville who built the Pentagon in 16 months. You know, it was Jean Krantz who led the Apollo program. There's something about our kind of Midwestern Calvinist sensibilities as a country where, like, it's the Apollo program and not Gene Krantz. It's the F-16 and not John Boyd's plane. I mean, you know, it's the nuclear Navy, not Hyman Rickover's Navy. That's great. I appreciate that sensibility tremendously, but it shouldn't obscure the fact that actually these projects working or not come down to a handful of really exceptional individuals, being present, taking ownership and leading the way. Yeah.

Can you give me a little bit, expanding on that, can you give me a little bit of history on the dollar a year men in that story? Yeah. In World War II, we actually had a program for very skilled people, corporate leaders, engineers. They could for a dollar a year, because volunteerism was illegal. You could not volunteer to work for the government. So as we were preparing to go to war, as we went to war, they would actually join the government as dollar a year men. They would get one dollar of salary, and they would actually be able to be deployed on the nation's most important and impactful problems. And sometimes they retain their old job.

Sometimes they wouldn't. It would really depend. But you could get the right person on the job. And I think that's really important because if you think about even World War II, our mobilization came down to one man, William Knudson, Danish emigree, who actually invented mass production. He was the number two at Ford where he invented mass production. He got in a fight with Henry Ford and went to GM as the number two.

And, you know, FDR asked Bernard Barge, who should I choose to do this? He said, I have three names. William Knudson, William Knudson, William Knudson. It's like it really came down to a counterfactual. We had like one guy who could actually move all of American industry. And the way that he assessed, you know, he was an engineer himself. He would like meet people and understand like, is it believable that you as a steering gear company can start making artillery not? Are your engineers smart enough?

Can they answer my questions? Boom. Here's a contract. Let's get going. And, you know, it wasn't a fiction writing contest. It was really a pressure test mind to mind. Can you talk about the executive order from, last week, the Defense Reformation. You had a great post on it, but I'd love for you to kind of break it down live for the audience.

I think it is the single biggest change that could prepare us to avoid World War III here. You know, if you look at, you know, what has happened to the U.S. since we won the Cold War, we've really had the rise and empowerment of this monopsony, a single buyer that is the government, and in addition to that monopsony telling us what to build at what price. and how it's all going to work. That's very different than the free market. You know, at some fundamental level, you either believe in the free market or you don't.

And I like to quip that, you know, everyone, including the Chinese and the Russians, have given up on communism except for Cuba and the DoD. We still have five-year plans. And this luxury of having the monopsony grow to the scale it has is the consequence of having no peer competitor. But, you know, that world really went away, arguably in 2014,

the militarization of the Spratley Islands, the annexation of Crimea, Iran's pursuit of the bomb, deterrence is lost, and we need to rise to that. And I think this is a clear acknowledgement of very much that in this administration. But the EEO says,

we prefer to buy commercial items. We prefer to buy items that have been proven in the market that have to withstand brutal competition that happens out there every day, that reward entrepreneurs for what they're building, rather than custom building and developing things on our own.

And we have a long history, a monoxone is always going to desire control. But it goes all the way back to Andrew Higgins in the boat that won the war in World War II. Andrew Higgins was a guy from Louisiana. He spent some time in China, and he was inspired by bootleggers in China and the sort of boats they had for amphibious landing

to quickly land, get goods off, get goods on, and then scurry away. And at his own expense, he built the Higgins boat. And he showed up, he showed it to the Navy. The Navy wouldn't even let him compete. They kind of dismissed him. Then a young Marine who became very famous, Krolloch, he later, he got him into the competition. He won the competition, and then instead of buying the boat, the Navy stole the plans and tried to build the Higgins boat themselves, which they failed to do.

And then finally, you know, at the 11th hour, of course, we do the right thing. That's a classic American trade. And Eisenhower said that's the boat that won the war. And so, you know, the predator was developed as a commercial item. It was not developmentally done inside of the government. as Abe Kareem built it. General Atomics picked it up. They financed all the R&D themselves. You know, the Air Force, of course, hated it because it was unmanned. It was kind of emasculating. And, you know, when 9-11 happened, it was the thing that met its moment in a massive way here.

And there's so many examples. Like now we have whole companies built around. This company's like Andrel. The entire approach is commercial first, investing private taxpayer capital into R&D, absorbing, putting the pebble in the right shoe, you know, putting the pebble in the entrepreneur shoe rather than in the taxpayer shoe. So this EO, it's actually this law. In 1994, we passed the law called the FACSA, the Federal Acquisition Streamlining Act that said, it is the law that if a commercial item exists, you must buy it. It's actually a very stringent test, three-part test. So if a commercial item exists that meets your requirements, you must buy it. If it doesn't, you must see if you can change your requirement to meet the existing commercial

items that do exist. And if that's not possible, you must see if you can ask the company to change their product to meet your requirement. Only then are you allowed to go custom developmental. So this is the most violated law in the land. And I think having the administration say, look, this is statute and we agree with it and we're going to enforce it is how we're going to field a huge amount of deterrence between now and 2027. There's not a lot of developmental things you can do between now and then. There's a huge amount that you can do with the innovation of entrepreneurs and the commercial industry. That's amazing. Do you feel like cyber warfare has been too normalized on this planet.

It feels like you have a major cyber attack and maybe you hear about it on X because somebody says they can't log in to Zoom or something like that, but it doesn't even make the mainstream news. Meanwhile, there's any type of kinetic conflict globally. It's like immediately front page or it's on CNN, things like that.

And do you think there's ever a point in the future where people start to view them and countries truly view them with the same level of significance as everything sort of content? continuously sort of comes online. I think you're spot on. There's something very strange in how little we talk about it. They don't photograph well.

Yeah, it doesn't photograph well. It doesn't photograph well. But my concern is that in the future, if we have 100,000 humanoids, you know, rolling, you know, going around a single city and suddenly they all are stopped, that's bad. That will start to photograph pretty well. And so I imagine in the future it won't be just kind of brushed under the rug and, And, yeah, of course, you know, companies have to respond to it and the government responds to it.

But I just imagine at some point it will start to really be news. Well, one of the most devastating consequences of it being so kind of below the waterline, something we don't want to talk about, is the normalization of like, what can you do about it? It's sort of nealism and acceptance, a pessimism that anything can be done about it, which is obviously not the necessary precondition to like rising to the occasion. And we should have very high standards for what could be done about it, Of course this company got breached or this thing happened, and all these people have my information all the way to, you know, sure our water systems are compromised and, you know, we'll be brought to our knees within the first few days of conflict. Like we should just, we're not able to hold ourselves to the high bar that we ought to because it's not popular.

And we're not popularizing it as a concept. I mean, speaking of popularizing just general shifts in thought about defense and the importance of like public-private partnerships in the government, can you talk, can you should. take us through the defense reformation, 18theses, your thesis there, and then kind of the impact and is it a jobs not finished situation? Or has defense tech become enough of a meme now? It seems like Silicon Valley is like fully on board, in my opinion. But at the same time, there's a lot more that we could do. Yeah, I think we've earned the right to happen at that. So like we've got to perform now for sure. I'd say the fuller, I've touched on some of the themes, but the fuller diagnostic is

the government has kind of historically made it a bad business to work with the government. You think about our examples from the past. At Intel, Bob Noyes would not let more than 4% of his R&D budget come from the government because he, as the inventor of the transistor, wanted engineering control over the roadmap and what he's going to build. He always had in mind a broader commercial market that was going to drive the, price performance that was needed. Even though in 1969, something like 96% of his revenue came from the Apollo program and DOD. At that point, they looked like a government contractor, but that was not, his aspiration was bigger. In the same way that Elon Musk's aspiration for

SpaceX has always been to get to Mars and to make us an interplanetary species, you know, it's not just about launching rockets and satellites into orbit here. And we really lost, you know, so much of defense innovation, you know, Kelly Johnson, who built 40 plus air. frames in his career, including the U-2, which we still fly in the SR-71. He is this heroic figure. So much of this innovation has come from these legendary engineers, these heretics, as I call it. Today, you think about it as Northrop Grumman, but it was Jack Northrop and Leroy Grumman. You know, it was not Lockheed Martin. It was the Lockheed Brothers, and it was Glenn Martin. And it was really so founder-driven. The aerospace industry subsidized its own existence in the

inner war period between World War I and World War II because the government didn't think it needed it. So, you know, but for private industry being willing to lose money for a decade plus World War II, we would have been in a very, very bad place. And what excites, so I think the last supper, people look at the last supper, which is this dinner at the Pentagon in 1993, where they said, hey, look, we have 51 primes today. You guys are not all going to survive. Today we have five. For every dollar we were spending in defense, we started spending only 33 cents overnight. It was a huge cut. The peace dividend is it's cold. Consequence that conventional people take away from this is, oh, that's when we lost competition.

We went from 51 down to five. I don't think that was the actual issue. The real issue is that consolidation bred conformity, and the conformity pushed out all the heretics. It pushed out all the founder personalities that you need to make this stuff really work. My reason for immense optimism in this moment is that the founders are back. You know, more than $100 billion have been deployed in the national interest. You have Palmer Lucky. You have the Sang Brothers of Shield AI.

You have Dino and Serronic. You have all these crews of really, really compelling humans, really, really compelling founders working in the national interests, again, pursuing heretical ideas, waking up every day, banging their head against the wall, you know, fighting the bureaucracy to do what's right for the men and women in uniform and for the nation more broadly. Can you talk a little bit about the startups that are building on top of Palantir and take us through AIP? Yeah. So we've always, it's a little abstract, but I've always felt like the ontology, which is our secret sauce, is it's really, you can think about it as like a declarative backend. It's a way of saying, look, this is the shape of not only my data, but the logic of my enterprise. And that's all you have to do, as opposed to the imperative approach of having to actually go create and wire this up and do it together and figure how to get it to scale. And so if you had this declarative back end, if you could just say, look, this is what I need to exist in the world.

Now I can build applications on top of that that manage and, all of the entropy that usually sits behind it. And it gives you a radical speed advantage for these companies. So these companies are building everything from like the European Cricket Network to people who are building pharma companies, hospital operations companies. In defense, it's very popular because we can give you an SDK that wires you into 20 years of data that has been integrated into the instances of Pallenture that exists in the defense community that your users can authenticate to.

So it just speeds a lot of the kind of brain damage you would get from having to, to deal with the bureaucracy and allows you to compete on the quality of your product rather than your game and getting through the wickets of a Byzantine process that probably needs its own reformation. But it's really about speed. And that ability to build on top of the platform, it's not just for third-party developers. When Hurricane Helene happened in North Carolina, green suitors, like folks in the Army in the 101st built their own appellate, they built their own hurricane common operating picture on top of the platform. So that ability to respond to the need. It's our, it's the code version of the Oodaloup. Yeah. What advice would you give somebody maybe

graduating college today that's evaluating, joining, you know, maybe a seed stage company versus a scaled company like Palantir or, or something like an Anderol. I imagine you at times had a ton of different pressure from VCs being like, leave, we'll give you 10 million bucks to do whatever you want. Like, don't you want to be a founder? Yet when I look at opportunities, you know, in anything defense related, it seems like if you have a founder mindset, there's so many great opportunities to just go to a great company and kind of take on that sort of like founder level ownership over a specific problem area or product set or things like that. Yeah, I mean, I think the, of course, I'm biased here, but if I, if I would

give general advice, the way I would think about it, what I would tell my son is like, where can you go that you're going to work with the most compelling people because like your rate of learning is going to be a function of those people and have access to the greatest surface area around problems like my model for growth is not progressive overload you know the incredible Hulk did not become incredible by just lifting a little bit more weight every single day it's like a near fatal dose of gamma rays that probably has like 50% chance of killing you so which companies are going to like throw you off the deep end and give you that opportunity to have like superhuman superhero growth that's like that's the value maximizing thing and I would remind my son that like your point of

extreme growth is going to be coincident with your point of maximal pain so like your ability like you know it's like as Greg Lamont says one of my favorite quotes championship cyclists it doesn't get easier you just go faster you know and just just understanding that like that's what it looks like so don't don't sell out for the opioid of a linear career progression with a clearly mapped out path I'm pretty sure that's retarding Like that's lead shielding. That's preventing the real gamma rays from getting to you here. You know, find places you can throw yourself off the deep end.

And I bet there are some C-stage companies that are perfect for that. I bet there's some C-Stage companies that are horrible for that. You know, it really is pretty specific. Does that tie to the primacy of winning mentality? Is that like all, is that essentially like derivative or like the same concept? Yeah, I think it's very much relates to that. And I think it's like one of the things that I had to, I feel like it's one of the most important things I learned at Palantir.

You know, we're all kind of programmed in our conventional education to feel like there's a process, there's an approach. Yes. If you follow in playbook, you know, you'll win. But it's actually like, it starts to conform to how you wish the world would work. It becomes hard to decide, is that actually the cargo cult or is that how the world does work? Like, you can keep marching around these fields in Micronesia.

The planes aren't coming back. You've misunderstood the physics of the universe here. And just, if you just anchor yourself, like, but is this working? Like, is it winning? and then blow up anything that's not working. Okay, last question on cargo cullting. Cargo cullting the forward deployed engineer, good or bad. When does it work?

When does it not work? Who should be doing it? Well, I think all cargo cullting is bad. So I think that would be, you know, it's, I think the forward deployed engineering methodology is, is exceptionally valuable. Akshay's description of it is the best one, I think. It's like solving through back propagation. It's very elegant metaphor.

But if you just slap a veneer on it where it's like sales engineering, done, my favorite things like ex-Palentarians will be like, people are asking me, like, what is an echo? And I will describe an echo and they'll be like, so you mean customer support, customer success. And it's like, wow, you've lost the essence of this whole thing, this beautiful concept. You've completely cargo-culted away. Oh, no. Well, thank you so much for stopping by. Yeah, this has been great.

Fantastic interview. You're having me, guys. Yeah, this is great. Have a great rest of your day and we'll talk to you soon. Cheers. Bye. Fantastic. So many great stories.

And man, like encyclopedic knowledge of American history. There's so many times. Yeah, we need to have him on for America. Yeah, of course. Lockheed and Martin, like, I know their first names. Like, I don't know their first names. I need to brush up on those stuff. I need to do some like space repetitions and flashcards or something.

Yeah. But next up, we have the founder of Starship coming in to the building. Welcome to the show. How are you doing? Boom. Hello, good to meet you all. I am doing fine. I'm doing super exciting thing and I like it.

Yes, thank you. Would you mind kicking us off with just a little bit of background on the company and explain not just what Starship is, but kind of the footprint, the rollout, the strategy, all of that? Yeah, yeah, all right. Yeah, sure. So what we are doing, we are, we are company developing and building and operating delivery robots. Yes.

So robots that transport stuff. Like, you know, and stuff that means, you know, burgers, milk, you know, could be, you know, packages and so forth, right? So in science fiction movies, you know, you don't see UPS guy is knocking on your door. You see things coming to you, things flying to you. That's the future that we are building. That's the present that we are building. And, you know, we have many competitors as well who have, who we have inspired to do similar things, you know, with drone deliveries and so forth. it on the ground with robots that drive on the ground.

Small robots that drive on the ground. And we are actually not like a, like many people think that, you know, this sort of futuristic thing is like a pilot or test, you know, somewhere in some limited area or something like that. We are actually in full commercial operation with thousands of robots in hundreds of locations. It might not be, you know, we are not there,

not yet in every place that, you know, all of this audience us listening to this right now is, but we are in some places. And in some of these hundreds of places that we are in, you know, our robotic delivery is completely commonplace. It's something that people are completely accustomed to. It's something that, you know, people don't really use, you know, like human couriers to deliver things that use robots. Yeah.

Can, uh, can you talk about all the different things that you've seen around the way humans treat robots? I try to, I try to say thank you to. robots, you know, when they help me out with different things. But some humans, you know, we saw this. Yeah, we'd be seeing a lot of vandalism of the bird, scooters, the big thing with bird. Vandalism of Waymos in L.A. is I think a lot of birds' issues locally here just came from a culture of vandalism. You did the hard thing in making sort of cute robots, which I think probably helps. It's important. How is, how do humans kind of present a challenge to trying to, like, automate delivery, which should be in everyone's best

interests. We get that question a lot, but we don't actually get a lot of vandalism itself. You know, the, the, the truth actually is that people really love our robots. Yes, I see that, you know, with like scooters and, you know, some of the other moves of transport, people don't really treat them nicely, but they actually do treat our robots really nicely. Like, for example, you know, you know, we often, you know, get the question that, oh, you know, are your robots stolen? And, uh, actually none of our robots have ever been stolen from the street. And we have done eight million deliveries. Like we are doing millions of deliveries and it's just not happening.

You know, kids feed our robots bananas. So the treatment is completely different than with, you know, like scooters, for example. Yeah. I understand. You know, that's actually a lot. We get that question actually a lot. Like people, you know, sitting in, you know, sitting at the table very often say that, you know, I treat things nicely, but there are some other humans out there that do not treat things nicely, just like you said, right?

You know, but it's not actually happening. It's not actually happening, you know, like how often do you vandalize like a, you know, UPS vehicle? You don't really, right? You know, sure, sure, you can do it. Yeah, you can puncture the tires of a UPS vehicle. You don't really do it, right? Yeah. You know, and you know, our robot actually, our robot has 10 cameras. It's constantly connected to the internet. It's really, it has a little. It has a really loud siren when it's tampered with and so forth. It's not actually really happening. Well, can you talk a little bit about the rollout maybe? Because I think that the locations in which the robots are doing deliveries is like really going to impact.

Right. If you're rolling through a suburb and it's a, you know, family friendly neighborhood, something like that, the robot is going to run into different challenges than, you know, rolling down, you know, a street in Manhattan, for example. So how have you guys approached your. kind of like go to market and kind of picking what regions are, are, you know, sort of top of the list in terms of getting robots on the ground. Yeah, great question.

We are operating our robots in 60 cities and 60 college campuses. So we have learned a lot about different environments. And yeah, like different neighborhoods are completely different. Like there's a lot of difference in the world that we're operating in six different countries. please, like five countries in Europe and you know US, right? And you know US of course, you know, it's varied as well, right? So we have learned a lot on how the different, you know, sidewalks look like, how crossings look like, how traffic lights look like, you know,

what the traffic patterns are, all of these things. We've learned a lot of these things, a lot of data for our machine learning algorithms, the two forth, right? But in terms of actually how people treat robots, there's not much difference, actually. People are actually really friendly towards our robots. I understand it's really hard to believe, but that is actually true. It's completely true. And you're operating a completely commercial quality service.

We are cooperating with most of the major delivery apps in the world and with a number of tier one retailers as well. And it just works. Can you talk about the progression of the technology, specifically the path to end-to-end There's a single AI model, but there's probably teleoperation in the early days. Then there's some mixed. There's some C++ for pathfinding, but there's some AI for image processing and world modeling. How have you thought about developing the technology?

And do you see teleoperation and end-to-end AI systems playing nicely together in kind of a centaur mode for a long period? Or are these specific gates that you have to go through? Yeah, we are operating at quite the hybrid model. Sure. We have been working on this for 10 years. Yeah. And we built this at the time where, you know, AI was not actually as developed and we have obviously reap the benefits of the all of the AI development that has happened since.

But we also recognize that SafeCity, for example, is super important. Yeah. Super important. Like for example, our robots cross roads. They are generally a sidewalk robots. They drive on the side of the robot as well, when there is no sidewalk, but generally they drive on sidewalks. But they cross the roads.

They cross roads similarly as a pedestrian does, using crosswalks, right? Our robots cross roads 100,000 times a day. 100,000 times a day, right? You have thousands of robots operating, right? You know, we have been speaking right now for eight minutes. During these eight minutes, robots are probably crossed the road here about a thousand times.

Wow. Probably it's like that. It was a thousand times during these eight minutes that we have, we have talked, right? You know, safety is super important. Suppose we have something going wrong in like 1% of the crossing. That's too much. That's about 1% or 0.99% too much, right? So we need safety.

We need to prove that it's safe. We are actually not operating an end-to-end neural network, but we are operating a combination of yes c++ and neural network when when you see hard tech founders claiming online that they're operating an end-to-end neural network which some people have done does that does that surprise you is it almost unbelievable or do you think it's not surprising to me at all that you can do it the downside with that it's very hard to prove that it is safe um it's very hard to prove that It is safe. And, you know, we are actually operating in also some pretty challenging regulatory

regimes. I mean, not just, you know, Arizona. Having a human loop is actually beneficial there, right? Like, if you have the ability to remotely take over, that's going to make your safety case so much easier because you're going to say, hey, yeah, it is kind of crazy. There is a robot piloting this a little bit. But at any moment, we can hop in and beam in and there's a human. Exactly. Exactly. For us, it's a combination of remote assistance, which happens, you know, more and more rarely all the time, right? But it still happens. There's a human, you know, somewhere that can take over in difficult situations or complex situations, unusual situations. Then there is C++ and there is end-to-end neural network as well.

Sure, sure. So it's a combination of all of these. Yeah, has the transformer architecture or any of the other like kind of foundational innovations in AI, been important to Starship and what you're building. Obviously, we see hype around the Studio Ghibli moment and diffusion in image processing, but does that actually make it easier for you to understand the stuff of like,

where am I in the world? Where am I going? I need to plan a path. How should we be tying all the amazing progress in LLMs and AI to your business? Yeah, it is definitely helping and improving our stuff. I think we were in commercial quality operation already before that, but it is helping us tremendously for sure.

Primarily, it is clearly something that is dramatically reducing the need for this human somewhere in the loop. Sure. Absolutely. What about have you thought about giving, you. embedding some sort of LLM or voice voice model into the robot. So if a pedestrian bumps into the robot, they can have a little conversation and it can kind of explain like, hey, I'm just going across the street. I'm delivering a burrito.

Like, it can answer some basic questions. That seems like. I'm just the burrito guy. That seems like maybe silly, but also like maybe great from a user perspective, but also like wildly extra. I don't know. Have you thought about that? Yeah.

Yeah. Yeah. So we do not have like a personified LLM in the robot right now. But I think it is conceivable that that will be the case. Sure. Our robot does speak. Yeah. But the speaking is not actually driven by an LLM.

Yeah, it's more like business logic, decision. Exactly. It's a little bit more business logic. But yes, you know, every time the robot, you know, does delivery, it says thank you. Yeah. Right. Doesn't take an LLM to do that. Yeah.

Can you tell me more about tradeoffs in robot production? Elon has been anti-LIDAR from a cost perspective. I'm sure there's a bunch of different trade-offs in terms of size, weight, speed, battery life, all these different things. You have probably like a base hub where these things go to charge and then they have a certain amount of range. What are you optimizing for? What are some of the pitfalls to avoid? Yeah, great question.

So our robots actually do not have a LIDAR. But that does not mean that we are anti-LIDAR. The thing is, though, that the reason we have not used LIDARs is that LIDARs is effectively, LIDARs are perfect sensor for an autonomous vehicle. Autonomous vehicle needs to see quite far away. And it needs to see, you know, it doesn't really, well, it does need to be, need to have short-range sensors as well, but, you know, it does. need to see you know like you know 200 yards 300 yards because it's moving fast right our robots

are moving much slower than a car right they don't actually need to see that far so liders are actually not perfect sensor for us and the downside with lighters is that lighters typically have a narrow vertical field of view they have like this narrow you know you know you know uh rays effectively that see very far but it's a very narrow the angle the vertical field of view is very narrow like a couple of degrees or so right we actually need to have perfect vision from immediate vicinity of the robot like very wide now wide you know vertical field of view we need to see you know down in front of the robot and also you know up and you know we need to have that sort of sort of vision so lighters are not

perfect sensor for us that's the reason we are not using lighters but the moment that uh that the lighter with uh suitable spec appears on the market, we will absolutely use it. So we are not like religiously anti-Lider at all. Do you have a question? Changing gears a little bit. Skype is shutting down on May 5th and end of an era. You were the founding engineer there. Is that emotional for you the shutdown or at this point you've been, you know,

in doing something, you know, else for, for, Yeah, yeah, I've been doing something else for a long time. I'm not actually an active user of Skype anymore. Sure. Or actually quite some time. I still have the app, you know, in my phone somewhere. But, you know, not really using it all that much anymore. I mean, Skype was an amazing right.

I mean, it was one of these startups, you know, which is actually rare, you know, one of these startups that just took off like a wildfire immediately from day one, right? from outside perspective all startups seem like that that they come out of nowhere and they just boom you know it's like that but you know

but in reality as a startup founder most startups are not like that most startups is hard work hard work before things actually start you know getting off the ground right it was not like that you know it was like for me it was easy work actually I just did

what I love to do and boom you know loads of users came and you know it just kind of kind of happened right so it was an amazing journey, but also, you know, frankly, I mean, it was a journey that happened 20 years ago. Yeah. 20 years ago, literally 20 years ago.

What, what are the key lessons that you've taken, that you took from the Skype story and applied to building this business? Is it just the engineering culture, the pace of play, or is it wildly different because it's a completely different growth curve? It's both of these things, but it's also, I would say one fundamental. mental thing is that with a lot of products and a lot of services, you kind of, you know, it's a very simple service really. Like, you know, one delivery app, a major delivery app founder, you know, told me that, you

know, look after you know, if your robots really work and you can give me cost savings because it needs to cost less than human delivery for me, then I will use your service any day. And you know, that's how it is for them. That's how it is for us, you know. We have no demand problem, right? If we prove to them that it really works and that we give them cost savings, they will use us in like, for like billions of deliveries. That's how it works.

And that's the traction we are seeing on the market. And Skype was like that as well. You know, if you actually actually put out the product that exactly fits what people want and it just works. It just works. It doesn't have some sort of major downside, right? You know, then it just takes off like wildfire. Brantan't it, it's harder to do with the cell driving robot, right?

Nearly much harder to do. You know, like we built Skype, you know, like we were like a team of like how many engineers we wear like, you know, 15 maybe, 15 engineers, nine months of work. And we have a beat, has a beta out there that they were loved and just took off like wildfire, right? Yeah. Sure, you can't do like a cell driving robot. with like a team of 15 engineers and line miles, right?

And it does take longer, right? But it's also harder for competitors, right? You know, so we are, you know, I said, you know, we have done, you know, eight million deliveries. I'm not sure, you know, our closest competitor. There are lots of competitors, tens of competitors. Our closest competitor probably has done 200,000 maybe or 300,000. Not sure, something like that, right?

Like an order of magnitude difference. So we started this trend, let's say, you know, 10 years ago, and we're still number one. Because whatever is hard for us, it's also hard for our competitors. I have one last question. Then we'll let you get out of here. Why is Estonia so successful in producing technology entrepreneurs?

Great question. I think I don't have a full answer. So I don't know. But one thing, the thing out there is that, you know, I was occupied by Soviet Union for like 50 years or so, right? And I was, you know, my age is such that I just turned 19 when we regained independence from the occupation, right? So I spent my childhood in Soviet Union. But my adult life has been like in a free country, right?

And, you know, turning 19. and finding yourself in a free country that actually doesn't have a lot of establishment built in. That means that you kind of grow up with an assumption that there are no obstacles. There are no obstacles for you. There are no big companies out there

that you need to kind of compete with. You're just free to do whatever you want, right? So that's the culture. I think, you know, overall, you know, successful startups can be built by any end. but it helps that if you don't know what's impossible you don't know that it's impossible you're you're you're going to have a better success that you just think that's great that's amazing that that's amazing thank you so much for joining the show congratulations on your first eight million deliveries

and looking forward to the 800 million we have to have you back on then next year yeah next year yeah let's hear it amazing thank you so much for joining we'll talk to you soon Cheers. Bye. Before we bring in, Daniel. Well, I will bring him in and I will let him give his introduction. Daniel, if you're in the studio, welcome to the show. Good to see you. It's been a few days since we hung out. Still looking great. How are you doing? What's the latest? You're fantastic. How are you?

Doing well. Well, Jordi's way, would you mind kicking it off with just a little bit of intro on yourself and what you're working on? Totally. Well, one, blessed to be in the capital of capital. Couldn't imagine a better place to be today. Yes. I've been building various consumer tech stuff for a better part of the last decade. I can have most recently attendees with the company I started many years ago, sold that. I think there was a quote from Shawshank Redemption of, you better get building or you better get busy dying, something like that, if I remember correctly. But reconnected with the folks at Coco, a year ago, Zach and Brad, and the whole, amazing team there and he was sitting there and realizing my I grew up in LA kept seeing all

these robots going all over the place and all these different cities and just figured it had to get involved so you know we're Coco we're bringing uh you know sidewalk delivery robots to all the biggest cities uh and markets across the world uh I've had a lot of exciting success there and yeah that's what we're up to can you compare and contrast your approach to starship we just talked to the founder starship uh but every time there's There's like this race for a new technology. It looks similar on the surface and then you dig in and you usually realize that the companies are taking very different approaches. How would you explain what Coco is doing differently?

Yeah. So I think, you know, first off, I'm a huge respect to the Starship team. Yeah. I actually think I ran into them at an investor's office like 10 years ago. Wow. I ran one of the starting the company. I think the biggest differences, right, are kind of where we operate, right? I think Starship has had tremendous success on college campuses and various other parts of other markets.

But, you know, solving the complex city problem, right? Like if you look at the delivering market, right, you've had kind of, even if you think back five, 10 years ago, right, delivery costs for an order of magnitude lower, both on the back end and kind of what consumers are paying. You have a fundamentally inflationary cost structure. But all that problem of most of the delivery volume is in the kind of big urban. cities. And so I think we've really just drilled on our execution of handling the kind of complex situations there. There's a bunch of different regulatory hurdles, dealing with even distribution, right? We're kind of the only company that's partnered with Uber, DoorDash, and a bunch of other partners that we're announcing soon. Being able to have that kind of broad-based reach, you know,

I think it's super important because you can actually get the volume. Customers can use you in those cities. You don't have to go through a first-party app necessarily. So those are you. So there's I think it would be the biggest on-the-surface differences. What do you take away from the Waymo strategy? They were out in, was it Phoenix, Arizona? They were out in Arizona for a while, I believe, in like the easiest mode for self-driving, like huge streets, no weather.

And then their second market was like the hardest place to drive in the world, San Francisco. There's hills, there's random people on the street, there's bikes. And I think like the telling that some people would get would be like, if you can solve San Francisco, you can solve kind of anything. maybe not Boston in the winter, but is that the story you tell or is there something else? And what are you pulling from their strategy overall, seeing what Waymo's done?

Yeah, I mean, just on the commentary of kind of hardest cities to drive, but I would love to see Waymos in Riyadh, Saudi Arabia. Okay. There is, it is bananas to take car rides there. Yeah. But what I think is interesting is we've always compared the kind of Waymo and the Tesla strategy, right? The Waymo, because I think, yes, they operate in San Francisco, which is, you know, lived there for many years.

It's not non-trivial to drive them. But the cost structure of, you know, how much does that vehicle cost, right? The amortizing that KepX against all the vehicles, right? If you have this really expensive Spencer Suite, you've really like cost-optimized the vehicle. And it's hard to when you're kind of reliant on all of these sensors and advanced computing. It makes it difficult to really get it to be substantial cheaper over time. And to deploy a bunch of, you want to deploy a bunch of these, right? Like the question I've always asked is, why doesn't really deploy, you know,

You have Google's budget, right? Why wouldn't you go deploy a thousand of these, tens of thousands of these, right? We've always viewed more in the Tesla approach, right? If you can make kind of camera-based autonomy and kind of driving work, you can have a substantially simpler cost structure, right? You're able to, and with that simple cost structure, you can deploy far more vehicles and far more places. And if you've seen the latest kind of Tesla self-driving, you know, full self-driving version,

it's an incredible driver, right? And so I think that's the kind of biggest delta. in the strategy that you'll see in kind of some of our competitors is like, and you know, everyone on the team is incredibly ruthlessly cost optimized of, yeah, how do we get this additional, you know, three cents a mile, et cetera, right? Yeah, yeah. So what does that mean practically? Is that just, hey, we're staying away from LiDAR because LiDar's expensive or is it like we got to build a gigafactory and just drive down the cost of mass manufacturing these? Like Tesla is kind of the story of they did both. But what other decisions have you made to drive that cost per mile down as

low as possible. I think it's both. And there's a really interesting thing. Everyone thinks of the comparable of like, well, it costs, you know, a human X dollars to do this. And with a robot, it's, it's cheaper, right? But that's only one half of the equation, right? The delivery business is a logistics business. And it's all in that 98th, 99th percentile, right? If every single time you mess up an order, right, human or a robot or otherwise, that is incredibly expensive, right, to the merchant, which might have, you know, a food cost deducted. to the delivery operator who kind of has to refund the order back to the consumer. So being able to be more reliable is kind of the other half of the cost structure.

And I think what really impressed me kind of when getting involved was, you know, the focus on reliability and kind of making that work. And that, you know, obviously involves, you know, the robots were kind of human operated entirely for a long time. And as you kind of roll out autonomy, right, you're able to do that piecemeal while kind of maintaining that high quality service bar. I think that's like the one thing, you know, we know from a lot of our partners is like really exceeding on kind of quality of service. So it's both costs, which we nail, but then also quality of service as well.

Do you guys have any projections on when you think autonomous delivery will be the default, right? There's like all these new systems coming in, Coco, Zipline. We're going to have lots of solutions, whether you want to order something from, you know, Walmart 30 minutes away with Zipline or you want to get a burrito. Have you seen pipe dream? Yeah, then there's pipe dream. We're going to put it through the ground. The pipes in the ground. So there's lots of ways that we're going to get stuff, which is what Americans love.

Yep. But at what point, what's your kind of like internal pacing around when you think these systems can be so ubiquitous that when you use a delivery app, you just sort of assume that it's going to be coming through some type of like autonomous platform? I mean, I think it looks really lops. If you're on the West Coast, like if you're in LA, like if you talk to people in Santa Monica, like, this is their norm. Like to a lot of people, they get predominantly robot-to-the-drichs, right? If you look at other markets we're not in, you know, it's not. I think it's the same for self-driving cars. Yeah.

I mean, my answer is as fast as human as possible. Like I think that's metered in years, if not sooner. But I think it's really market specific, right? Like, you know, if you're in the big cities, I think, you know, hopefully for us, right, in the span of a year, maybe two. but, you know, I tend to be aggressive on that front, but I think it really depends on kind of which markets you're in as that kind of rolls out. But yeah, I think I think this proliferation of all these solutions is incredible, right? And I think, you know, for rural environments and spend a lot of time around in Wyoming,

Zip Line's going to be amazing, right? But especially in the urban environments where the real, like, there's a huge push, right? There's a real business problem for fraud of delivery providers of, like, we need to solve this. I think that makes a lot of sense. Can you talk about anthropomorphic design? How do you make the robots friendly and seem like they're going to be fighting alongside us when we're fighting the AGI Terminators versus, because I imagine if it goes poorly, I'm going to want a Coco in the foxhole with me. It's a cute, I mean, it's cute even the name, like, you know, Starship, cool name, but very different than Coco. I think that's deliberate.

What else are you thinking? Where does this go? We were talking about maybe like, do you throw an LLM with a voice mode on it and you can just, talk to it as it's driving past you, maybe ask it the weather, ask it the time, or ask it for the news. Where does this go long term? So I think a few things, one, right? I think actions matter a lot more than words, right? What someone says to you versus how they behave towards you. And like, you know, we have this idea of this cone of courtesy, right? And really being courteous to other, you know, sidewalk users, bicycle lane users, whatever the surface might be, right? And I think what's

interesting as you've seen, right, like I saw the kind of previous interview, there really is, you know, basically almost no incidences of vandalism or kind of people interfering with the robots because there is this like almost adoration. The name is very purposeful, right? And, you know, the second most popular dog name in the U.S., at least at the time, right? And really, like, and I think kind of thinking through that of like how something behaves towards you and it ends building the brand more than like what it talks to you. But then I think like there's always been this kind of sweet jovial culture at the company that really just flows out through to the robots. I think there's a lot more stuff you're going to be seeing soon in that vein.

I think I've been really focused on let's get the economics and the operating model right, but there's a ton of really fun stuff coming there. I can't quite sure yet, but we will talk. Can you talk about your guys' partnership with Open AI at all? I don't know how much you can speak to it, but that was a cool announcement. What do you do with that? Yeah. Yeah, what's up with this whole AI thing?

I heard it was cool. I heard there was this model that came out yesterday that people were doing some cool stuff with. If you have not tried O3, it's amazing. But, yeah, I mean, they're fantastic partners. I can't speak to too much about what we're doing with them. But it's very exciting. And we'd love to talk more about it later.

Yeah, it's a cool partnership because you're not immediately worried about them just building the same thing that you're doing. It's kind of like a bigger, you know, jump to go and say, you know, Open AI is going to start building delivery robot networks, then, you know, somebody building a coding platform. Unfortunately, I got shot down at our weekly business review. I thought, you know, we should get into the kind of VS code for a business, but that's a good business.

It's a great business, especially today. Yeah, it's fantastic. What else is on your mind broadly in venture right now? What are you seeing? I know you do quite a quite a lot of investing historically. what's been exciting. I was at this conference this weekend, and I want to share probably the funniest bit that I heard from that.

I was talking to a guy, manages kind of tens of billions of dollars, and we're talking about SPVs. He just goes, never say that word again. It's a dirty word. Sounds bad. He's like, you should call it a co-investment vehicle. And it's a great idea.

And so I think we should try and promulgate co-investment vehicles over SPVs. It's better. direct like the scene from the social network. Yeah. I mean, a lot of, I think what's going on behind the scenes in the industry is really exciting, especially on the kind of delivery side. There's a couple announcements that I'll be back very short to talk about.

Fantastic. Yeah. Very cool. What's the vibe in Jackson right now? Is ski season wrapping up or is when even is spring break? Is that? Right now, we've got mud season, which is, you know, all the snow melts and it turns a really delicious color of brown and muddy and gets all over everything. So I'm actually...

You're still out there, right? I mean, I'm in New York right now. All right. You'll see the seasonal migration of all the Jackson people, but it's a fantastic place. There's actually an article in the journal this morning about how tech bros are trying to dress up like cowboys. I know they were going to say suits I was hoping for suits this is terrible

all the work I put into this get nowhere I always thought about wearing a denim I mean this is like Chris Saka's dream right everyone dresses like a cowboy he's been doing that for 25 years but I wonder if you have to find a new bit

right once everyone starts doing it like does that of course yeah you always bring suits to Jackson all right yeah yeah I should I would love for you guys to come visit and wear suits and we can record the looks that when you yeah people don't ski in suits enough right it's like skiing is a serious endeavor you should

stress the part yeah exactly yeah partner with zenia to make like a ski suit and it's like like a go there we go yes there we go made to measure that probably sell really well on instagram think we got another business uh put together one of those co-investment vehicles for us okay yeah yeah yeah it's gonna need at least uh at least 100 mill yeah that off the ground with the tariffs Yeah, exactly. Actually, that's an interesting question. Speaking broadly, how have you seen the robotics industry kind of responding to some of these terrorists? I mean, I'm sure a lot of this stuff is married in America, but the supply chain, right?

Yeah, but especially with, especially with instant delivery where it's really an economic equation, which is like, is it reliable and is it cheaper than human-based delivery? how do you think the industry is like responding still shell shock or um yeah i mean i can speak only you know internally i can kind of give you some of things i've heard across the street but sure you know internally we have thousands of vehicles already kind of like in the u.s built and we're kind of actively deploying more and more of this so you know from from our standpoint it was kind of like okay like this doesn't actually disrupt us that much right now but i think you know the speaking of broadly to the supply chain, right? You have a lot of the consumer electronics supply chain is primarily in Asia, not a whole time of alternatives. And so I think it'll be really interesting to see, like,

as the administration kind of rolls out the kind of other branch of this policy, do you start to see some carrots to kind of move that over? You know, a big thing that I think is discussed this weekend. That was interesting was, you know, what happens if you just drop capital gains on kind of investing in kind of these key sectors of the economy that we want to bring manufacturing back to the U.S. on. I think, you know, don't have the U.S. sovereign wealth fund to try and pick winners, I think we get into a little bit too much time in there. It's a great thing. Yeah, yeah, I was thinking about this for all the venture capital firms that have been

adventuring all over the eastern hemisphere, you give them a cap gains boost in American investing. You're going to see billions flow into national interest investing. They already eliminated capital. gains for in some in some form yeah but but but that would be a cool policy yeah yeah they didn't say how awesome dan yeah well thank you for coming on come back on i know you guys have some good news thanks for having me coming down the pipeline yeah we'll talk to you see you guys cheers next up we have uh adam uh coming in from the chamber of progress talking of about the Google lawsuit that just happened and the ad tech verdict.

I think we'll be able to have him pop in and give us hopefully some background. Adam, welcome to the show. How are you doing? Oh. Good. Sorry. I had to unmute myself. Can you hear me okay?

Yeah, yeah. You're all good. I'd love to start. I mean, quick background on yourself would be great, but also just can you take us through the very early prehistory of this lawsuit will build up to your analysis that you drop today? Sure, sure. Well, thanks for having me, guys. Yeah, great to me.

I have been doing tech policy in D.C. for 20 years or so. Spent a dozen years at Google in their Washington office. About four years ago, started a group called Chamber of Progress, which is a center-left tech industry policy group. And so I've worked on a lot of these issues. I have a lot of familiarity. This case is interesting. So this was the second case that the Department of Justice filed against Google

on antitrust issues. And the topic was kind of a niche one, which is ad tech, the ad tech industry. And it really was alleging that Google had improperly linked the different parts of its ad tech business together in a way that had given it a monopolistic position. And the case moved relatively quickly.

It moved for what's called the Rocket Docket, this court in Virginia that goes generally pretty fast. What does fast actually look like? When did it kick it? off and obviously this one so this one was filed so this one so this one well so this one was filed by the Biden administration the other the other Google case uh which had to do with Google's search distribution deals was actually filed by the Trump administration so um this one I think took maybe a year from the time it was filed to the time it actually went to trial um so that's like relatively speedy

and um and they had it you know there's a whole parade of people from the ad tech industry who testified as witnesses in this case as well as like advertisers and publishers and kind of people from that world and and then it but the trial ended in thinking of December and then they just just came out the verdict today and there were some points for Google but it was mostly a loss for them the court found that they had improperly tied some of their ad tech products together and they have a monopoly position in what's called the ad exchange as well as the ad server which is the tool that publishers use. Can you break down? Yeah, can we kind of like go through a little bit more of your analysis? I'm curious to get kind of really your, I'd like to honestly hear your personal opinion on it. I don't know how much you can, how much you can comment specifically. But do you think this was, do you think this was the right outcome? I know there was advertisers that use a platform that were testifying saying, like, look, we tried to use.

a lot of other alternatives and they just delivered inferior results. But what's your read on it? Okay, so I think the first thing you have to say is you have to back out and look at this industry, the ad tech industry, because it's sort of like famously complex. There's this famous chart called a LumaScape that shows all of the players in ad tech. And so on the one hand, you could say, okay, well, there's a lot of players. That's great. But on the other hand, one of the things that advertisers and publishers, you know, have said over the years, is like it's a little too complex, right? And what an advertiser cares about is basically like finding the customer who's most interested

in their ad. What the publisher wants is basically the best return on their ad space, right? And so the way that this industry evolved was frankly a little bit contorted, but Google stitched together, you know, several products that basically ideally gave the publisher the ad that performed the best and ideally put the ad for the advertiser in front of the person likely most interested in it, right? Now, other players in the ad tech industry complained, basically. They said that Google made it too hard to interoperate with other, with different parts of its ad tech stack, like the ad exchange or the publisher ad server, right? And that was the heart of the case, right? That essentially

that Google hadn't provided enough interoperability between these different parts of the ad tech stack. And so that was kind of the key question. Like in antitrust law, there's this kind of doctrine of time, right? Which is basically do you force people to take one product because they really want this other product, right? And that's like the heart of the case. And the judge today found that Google had improperly tied its ad tech products together, specifically the ad exchange, which is sort of the matching piece of the puzzle with the ad server, which is the tool that publishers use.

And so I think that like we don't know there'll be a second phase of this trial, which will look at the remedies. Like what happens now? What is what has to change in Google's business? That'll happen later this year. But to me, the other, the big kind of strange thing about this case, I will say is that like, if you look at this part of Google's business, this is part of what's called their network business, which is basically like serving ads on other people's websites. this is the part of Google's business that's actually shrinking. So it was like 16% of their business like five years ago and now it's down to like 11%.

And you start to say like, okay, why? Well, what's happened is that that business of placing ads across the internet has just declined because people are... Wasn't it because of Google's product? They're doing AI summaries in some way. It's like I don't need to go on YouTube and social networks, right? And one of the things is like there's so much less purchasing activity being driven by these ads because the targeting is worse.

And you're there on a website usually to get information or to do something, not to, you're not just sort of passively scrolling. You're nailing all the reasons, right? Which is basically like this market that, you know, in any interest case, you have to define what market you're talking about here, right? And here, really they were talking about, I think it was called like the open, the open internet advertising market, right? But like that's shrinking, right?

Because people are spending more time on within ecosystems, right? within Google, within Meta's services, within Amazon, right? And those are actually increasing their share of digital advertising. So, you know, one of the critiques, I think, of antitrust, sometimes you hear from people is, like, it moves too slow or it's like fighting yesterday's battle. We saw that earlier with the FTC lawsuit and meta about Instagram and acquisition that happened a decade ago. Totally. I mean, it's very funny because, like, these things are all happening. This is like high season for antitrusts right now in D.C. where I live.

Yeah, totally. Because the FTC trial is going on downtown. Next week, the remedies phase of the other Google case starts. And then this one comes down. So it's like high season for this stuff. Yeah, I mean, the FTC, yeah, go ahead. Yeah, with Google specifically, like, what's your, how does this play out? What is the remedy?

Yeah. It clearly matters to the businesses that feel and the judge says have been sort of wronged by Google's actions. but for the average consumer, it's just, you know, I don't even know if it, you know, I don't see this sort of hitting even headlines in the same way that, you know, obviously meta's, you know, Instagram issue has. That's totally right. This has always been a very kind of in the weeds, more obscure case.

And one of the things that is kind of interesting is that like, there's no doubt, like, this is a loss for Google. It's probably a win for Google's competitors and ad tech. The big question mark in my mind is like, is it good or is it bad for advertisers and publishers, right? Because I think on the one hand, like advertisers and publishers say, some of them say like, we feel a little bit beholden to Google and we don't like that, right? On the other hand, like, and this came out at trial, like advertisers say like their performance, replacing ads in Google is really good, right?

Google's putting their ads in front of people who need it. They're getting good clicks. Publishers are getting good revenue. And it could be that in breaking apart like this system that's working well for advertisers and publishers that they end up kind of regretting, you know, this whole thing, right? So I'm interested to see whether that happens. But you're right. The next phase is going to be a remedy's trial. And what happens? I mean, it kind of depends on how aggressively the government wants to pursue remedies against Google, right? Because you can sort of see, and the judge said today,

basically, like, you have to submit initial proposals to her within like the next week. So this is going to come pretty fast. And so the most aggressive is like Google's ad tech business is broken up, right? So she says Google has to sell maybe it's ad exchange business or its ad server business, the part that works for publishers. That would be probably the most aggressive. The less aggressive version of that would be clear for the audience that's not like AdSense that has nothing to do with Google.

It's not ad words. It doesn't have anything to do with like they might, is it possible that Google could say like, cool, we're, you know, yeah, it's not going to like materially damage the business one way or another. So it's maybe a little carrot that they give or is that even not not the right way to think about it? Well, in fact, there was some, there was some reporting before this trial started that Google had had supposedly tried to engage the Justice Department and settlement. talks, right? And so you could like could have imagined like, okay, maybe they would have gone for something there. I don't know. I think we'll see what happens, but you're absolutely right. It's not AdWords, not AdSense. This is their publisher site tool and the ad exchange, which most people do not know about. Most even, you know, it's really like publishers and advertisers know this. The less aggressive remedy, I think would be something like, okay, you can continue to operate these services, but we're going to mandate like maximum interoperability,

Basically, like every part of your ad tech stack has to interoperate with other services, basically. And so that would be like the less aggressive version. We'll see. We'll see which one prevails. Is the consumer harm standard just like completely out the window now? Did Lena Kahn just like completely wipe that? Because I understand that if I'm a rival advertising platform, I'm upset about this. But if you're a consumer, there's maybe hard to prove harm.

And even if you're an advertiser, you're probably like, hey, as long as the ad rate are good and I'm getting good ROI. I'm happy. So how did this come together? And are we just completely past the consumer harm standard at this point? I fear that we're kind of beyond it. I think in what in this case, the consumers were the advertisers and the publishers. There was never really any serious allegation that like this had an effect on the end consumer. The judge's decision, Like she, she, her argument, her ruling was that Google's link between these products was unfair to Google's advertiser and publisher customers. And so I think in her way, she was applying the consumer welfare standard. But whether that, you know, Google's going to appeal the case, right?

That'll definitely happen. And one of the big cases that's a challenge for the court here is this case called Trinko, which is old, old Supreme Court precedent, which basically has to do with what's called the duty to. deal, which basically says, like, you as a business, even like a dominant business, have an obligation to help your rivals. And she, in her ruling today, she did what I would call a creative interpretation to kind of get around that. That is the biggest question about whether this ruling survives appeal, and even up to the Supreme Court, because, you know, that, that, the creativity you call it of her ruling may not survive appeal, we'll see. Do you think that there's a just zooming out like a shift in the government perception around just big businesses, even if they're, they maybe are monopolies, but they're not causing consumer harm.

We saw as Lena Khan taking a shot at Amazon, like they're clearly dominant, but is their consumer harm debatable? Yeah. But it's still, let's put the screws to them. And J.D. Vance kind of said the same thing with Google. He's said, yeah, that's a big company, maybe too big. Maybe we'll keep some of that pressure on them, even though it doesn't necessarily fit the previous definition of the monopoly test. Just market concentration is enough to merit a response from the government. And like, how do you think the perception around that shifts?

And do you have a particular take on just like the idea of big companies that are dominant? Well, so it's just very interesting. So like when I worked at Google, one of the things, I really admire this because I don't think this isn't necessarily true today. Like my boss said, look, we're a big company. We have a lot of power. We're very influential. Like, let's not try to fool anybody, right? Yeah.

And I agree with that. Like, big companies should absolutely be more scrutinized, be held to a higher standard. Like, absolutely, that's true. But what happened with these federal antitrust cases, so now every big tech company has an antitrust case pending against it from the federal government. Apple, Google, meta, Amazon. What happened was about six years ago, the FTC, and the Justice Department. They share responsibility for antitrust.

They got together and they said, they basically struck a deal. And they said, okay, FTC, you get meta and Amazon. And DOJ, you get Google and Apple, right? And so they split up the companies based on like the target, not the subject matter. And I think like just human psychology, once you've done that, you're going to bring a case.

So I think what happened in those cases was like they had the target in mind and then they worked to find a case they could bring, right? And so I think that's why like, you know, these cases have like varying degrees of strength to them because they started with the target in mind and then they figure out the case rather than like looking holistically in industry and seeing if there's a problem. So I don't think that was the most principled way of going about that. But but I think that that's what happened. Now one of the things that is very interesting is that like, okay, Trump, Trump crowd. like the thing they actually care more about, most about is censorship and speech, right? So it's really kind of strange, bizarre. I think today, Pam Bondi, the Attorney General, she did a statement about this ruling.

And she said, she made some reference in there about how this is great because, you know, we're cracking down on Google because they've censored, you know, conservative speech. Like the case has nothing to do with that whatsoever. But it is tit for time. Well, I just think, I just think, like this wasn't the, this wasn't the Trump crowd's case. Yeah, yeah, yeah. Right?

But they, but I don't think the Trump crowd Trump crowd really cares about Google's ad-tick business. I mean, looking at other motivations, is there a world where, you know, the current administration seems very fixated on increasing revenues and decreasing costs?

And is there a world where if you can just go around and get, I think they were asking meta for $6 billion to settle that case? Is there a way that it's just like, no, more than 30 billion? Oh, yeah, 30 billion. 30 billion. Yeah. world where it's like, hey, like, we're just going to give out some parking tickets to try and raise some

revenue. Is that like a rational thought experiment? Yeah, to me, the fact they were willing to settle up for $30 billion kind of shows that they didn't really care about the case they brought in the first place. Because it was like, okay, we'll make it go. And it's just like putting a price tag on it, right? Yeah, yeah. Wait, yeah. Let's pivot to the to the meta case because you're sharing about that yesterday. You had said that the $30 billion settlement offer seven times the FTC's annual budget, so atlantic that the point would be humiliation, not restitution. How do you think the trial went yesterday for Zuckerberg? John was sharing that he looked absolutely fantastic in a suit. It's unfortunate that anytime you see a tech CEO in a suit, they're in trouble. You're being punished.

but what's your read on the whole situation and I'm curious did you think that did you think that the original $450 million offer was was it was a good starting offer for Zuck and maybe they could have found some middle ground between $450 and $300 billion because he was being kind of like run through the mud for even offering $450 million but yeah he's a businessman I imagine he expected them to counter and like kind of meet the middle or something like that. But what's your take? Yeah, I mean, so you're absolutely right. Can't blame him for trying. Like, look, he like if the worst case, they lose the case, they're forced to spin off Instagram and WhatsApp. Like, if you could avoid that outcome, you, you know, every CEO would try to avoid that outcome, right, through a settlement. So don't be grudged on that.

But if he was, but if he was truly worried about WhatsApp and Instagram being spun off, wouldn't he have come in and said, yeah, we'll pay like $10 billion. because I really don't want that to happen. Yeah, that's a good point. I don't know. I mean, the reality is, like, I think what they have to balance that against, like, how confident are they in their own case? And frankly, I think the FTC has a really tough case there.

Yeah, that's what I'm saying. This is a dozen years ago, right? Yeah, no, no, I was saying specifically, they have to feel pretty confident in their case if there wasn't, you know, a middle number that they were really wanting to anchor. I mean, just for reference, META has 77 billion in cash and short-term investment. And they're willing to spend. Give us half. Yeah, yeah. They're willing to say, no, and they're willing to say, we're going to spend

$80 billion a year on CAPEX. CapEx and the Metaverse is going to be a huge R&D expense. Yeah, I have to imagine. Well, by the way, like, just because it didn't get settled now doesn't mean it won't get settled later. Yeah, it's true. And, you know, sometimes that depends on how the tributts. goes. Yeah, right? And so and frankly like if the FTC like at the end of this trial feels like, they might lose, like maybe they're, maybe they'll, they'd accept a lower number. So you kind

of never know where that's going to, how that's going to play out. But yeah, I think, I think like the the challenge for the FTC, like they have two problems. One is they have to, they have to say essentially that Facebook, the meta doesn't compete with TikTok, right? And X, like that's a problem for them. But the other is that they have to prove monopolization, basically, that like, that if Meta hadn't acquired Instagram or WhatsApp, what's app, they would have gone on to, like, challenge them and be great competitors. Like, they were really, these were really small companies at the time they were acquired, right? And credit where it's due. I mean, meta did a great job of integrating those companies, making them successful. And, you know,

I'm sure you guys have seen plenty of failed acquisitions. Oh, yeah. It's just like, you know, it just doesn't work. Like, it's really hard to successfully integrate a company. And, like, legitimately, like, one plus one equals three, you know? Yeah, and one interesting data point. So when Instagram was acquired, they had 300 million users. And Be Real, which was the last breakout consumer mobile app, has 40 million users. And I'm going to go out on a limb and say that it's probably not worth anywhere close to a billion, right?

even for someone like meta who has a consistent playbook of integrating that kind of business into the platform. Well, they also like meta knows like they're in this business where like you have the most fickle customers ever, right? Like no young person is on the Facebook app, right? And then like very few really young people are on Instagram. Right. So like social networking is a very like kind of like generational thing.

And so it's very hard to like maintain kind of. of the zeitgeist in their business for like multiple generations. I think that's like a structural challenge for them. So anyway, so I think I think FTC has has an uphill battle with that with that case. Do you think these are kind of the wrong problems to be working on generally as we like, you know, the entire internet is about to be steamrolled and economy is about to be steamrolled by AI? Like do you think it's a big distraction to just be arguing, you know, over acquisitions, that happened more than a decade ago when it feels, you know, like we have this revolution

in many ways that has already started. Yeah, I think so. I mean, look like, you know, I think Blop is very popular for politicians to say things like, we got to beat China and AI. Well, like, who do you think is going to do that? It's going to be Google, meta, open AI, right? It's like, it's going to be those companies. Yeah, arguably, a meta without an Instagram is a much weaker product that can't invest in

as much in CAPEX, they can't invest as much as in Lama. Like maybe Lama 3 doesn't get built because they're making half as much cash flow or something. There's a bunch of interesting ways where you get a piece dividend from the market concentration. That's the question about monopolies. Like if there's no consumer harm, products free, and it's really good. And their whole North Star is like, let's please the customer. Like Bezos always says, customer comes first, all that. But then you also get these crazy dividends when they go and build cool stuff out there.

They're like it's kind of an awesome system even though it makes it hard to compete with them. But I don't know. What's your take? Well, I just think I guess the way I come to it's like, I tend to think of like all of the big companies like intensely paranoid about each other. Sure. And like they're always getting into each other's spaces too. Oh yeah. And so like to me like that that part of competition is sometimes overlooked in the way and like, yeah, yeah, yeah, yeah. These guys are like at each other's throats. Yeah. And that's good. Like ultimately because then they're that competition is leading to them to add them to add features and like good for product innovation. Yeah. Yeah. I mean,

every single big tech company overlaps with another one in one way or another, whether it's Amazon overlapping in ads now, you know, Bing has a search engine, Google has a phone, like everyone's in everyone's spaces constantly. Yeah. And it's inevitable. I think that's great. Yeah. I think it's wonderful, you know. Well, it's been great chatting with you. This was fantastic. We'll have to have you on the next time that there's a big FTC case or big case. This is fantastic. Just DM us when you think there's a story that you're excited about talking about. you can be one of our Hill correspondents. Yeah.

Congressional correspondent. All right. This is fantastic. We'll talk to you later. Great to meet you. Thanks in. Cheers. Well, we should close out by talking about another acquisition.

Open AI is in talks to buy Windsor for $3 billion. What do you think? Open AI and Winters, Windsor declined to comment. Future Instagram situation? Or maybe not? Look, are you long? This situation takes me. back to talking to Sarah Guo, and she said, look at what the foundation model companies care about.

And one of the things she talked about was coding. It was just, she said, like, if you look at all their actions, everyone besides GROC seems to be very, and just XAI broadly, seems to be very oriented around code generation. And so I don't think this should be surprising. There was also some reporting that Open AI had allegedly tried to buy cursor on a couple different occasions. And then in that time, WinSurf just kind of took off, really grew revenue quickly. And yeah, if this deal gets done, I think it makes a lot of sense. There was people pushing back and saying, okay, how does a $300 billion company not know how to build an IDE? There was a good post about that. My take on it was like, you know, it's probably a timing thing.

I'm sure that Open AI could build a fantastic ID. If they wanted to build a, they wanted to build from the ground up. They're also just growing the organization so fast and they're becoming a product company. People think about opening eyes like, oh, it's this like 10 year old company sometimes, but it's like, no, they're seeing insane growth. They want to hoover up talent and people that are good at designing products and just integrate more and more into the system of the building.

And I also think there's something that very real is happening, which is the underlying models are starting to not perform as well. Commoditize. In places like cursor. and windsurf. And so it's possible that's an accident of the way the models are evolving, but it's possible that verticalizing is what makes sense.

And I think that they've done a good job of, opening eyes done a good job of filling that gap of like the empty text box that you start to prompt like Google search. Like the first chat GPT app was like a direct Google competitor in many ways. You've talked about it, the knowledge engine. But there are clearly going to be,

several layers where they were where the the LLM will be vended in and coding is a distinct one from the Google search box. And so you got to be in the ID, you got to be in the Google search box, you got to be in a few other places like where you, in your camera role essentially is where you ultimately want to be for image manipulation, image generation. Of course a lot of that will happen in the chat GPT app and in those chat workflows. But let's go to Dylan Patel because he broke it down a little bit about why he thinks OpenAI is buying Winslow. surf, what's the strategy behind the revive of Codex? Cursor and Anthropic had a mutually beneficial relationship, but Labs realized that controlling

the main application of a model is as valuable as owning the model itself. With this acquisition, OpenAI gains greater ecosystem control and can build better products. Anthropics Claudecode was very well received. Keen on not missing out, OpenAI released Codex CLI, which is strikingly similar as a product. Both of these products have terminal level access and code editing capabilities. The competition goes beyond just products. Open AI opened up free access to the plus tier for university students just one day after Anthropic announced their education initiative in early April.

ABLEA. A fast follow. So like the game is on. There are not, there's not a single multi-billion dollar AGI market. There are probably many pockets of value that will be discovered. And the interesting thing here is that I think it's a narrative violation. It was actually fine potentially to build a wrapper.

in many ways these companies were derided early on when they launched as, oh, it's just a wrapper, you're going to get eaten, this is going to get one-shotted by the foundation models. Well, it seems like there at least is a way out via an acquisition. These companies are valuable. We've seen the ARRs grow. And then also, I think it's interesting that there are, you can probably think of like a power law distribution around value creation

at the application layer with like the blank box in chat GPT. like just go and talk to the LLM, that's probably like the most dominant. That's the Google, right? But then Winsurf, cursor, Devin, these are all super valuable in that category. But then there's probably a really long tail and there's probably some value in the application layer deep down.

So I think it's another maybe bulk case for rappers, depending on how you're structuring the business. You probably are not going to disrupt Open AI with a wrapper, but you could build a very, very great business. Yeah, and for Winsurf, I mean, I'm sure they, in many ways would have loved to just keep building and building and building and building compounding. But they also, I'm sure we're very aware of the competitive dynamics. And this is one of those things, right? When you talk to, we talked to the founder of captions earlier, right? Fantastic app that you're using all the time.

How far away are we from you being able to upload a video to chat GPT and just say like add some titles? It's possible. Yeah. You're already a pro subscriber, right? And you're like, okay, I have this. But at the same time, with the cursor analogy, like, it's possible that there are more features and you just stay ahead long enough to lock in that customer. It's like, well, I like their UI. I'm familiar with that.

This is the same thing with, like, DaVinci Resolvi is for, in many cases. And people don't switch because the buttons are in a certain place and the certain features that are nice. I would just say, like, people were, you know, people were, I even saw Daniel. who we had on the show talking about how, yeah, obviously value accumulates to the app layer. But I don't, you know, I still think we're going to see this back and forth battle where, oh, values accumulating to the app later. But open air has an app.

Like models are everything apps, right? You can ask it to do something absurd. Yeah, it's a good. You can ask it to now go, you know, do the, you know, book you a flight, make me a video, write some code for me. Yep. They are a new form of everything app. Sure, sure, sure. And I think that big companies, even if you have a lot of traction today, you need to look at how fast they're evolving.

There's a funny post by Bern Hobart here. O3 is a quality improvement, but at least on financial topics, it's definitely the fratiest model. Returns don't increase or rise. They get juiced. And retaliatory tariffs are described as China blasted hogs and soybeans. This is great. This is my only real complaint about frat GPT is that it hyphenates too much.

But other than that, it almost perfectly captures the style. So funny. Blasted hogs and soybeans. I love it. I love it. Well, we got to tell you about public.com, investing for those who take it seriously,

multi-asset investing, industry leading yields, trusted by millions. So get in on the action at public.com. What else? I mean, Dan Shipper, he kind of had the most viral review of 03. He says it's absolutely amazing. It's already his go-to model, fast, agentic, extremely smart and has great vibes. Some of his top use cases, these are things you

can kind of steal from him and adapt into your own workflows. It flagged every single time I sidestep conflict in my meeting transcripts. It spun up bite-sized ML course that pings me about every morning. It found a stroller brand from one blurry photo. That's cool. It coded a new custom AI benchmarking record time. It x-rated an Annie Dillard classic and found writing tricks I'd never noticed before. It even analyzed Every's org chart to tell me what we'd be good at shipping

and what our weaknesses are. So he's having a lot of fun with it. Anything else here. 824. Music. Yeah, they launched a music label. That's really exciting. And they're getting venture-backed, right?

Yeah. Didn't they take some money? From Thrive. From Thrive. Very cool. And then, gosh, I feel very good. Bedford Blankana's name. A guy who was at Adobe.

Oh, yeah. Scott Belski. Yeah, he went over there. Belski's over there. Incredible organization. I mean, people have been so bearish on Hollywood and the idea that you could build a new studio in a special way is just fascinating to me and they've been able to do it really well. All the A24 films have been fantastic.

It's taste. It's taste and it's clearly such an interesting differentiation. And now they're getting to music. I'm sure they'll be publishing a bunch. Who knows? I'd be cool if Belski was involved with this. I wonder if there's an AI angle. Yeah, I don't know.

It would be kind of silly to launch a new record label. That does not embrace the AI era. Yeah. Oh, Deal CEO, we've got to talk about this. We've done two episodes on the deal rippling drama. But Matt Levine says, has anyone booked a demo with Deal recently just to see what lines their salespeople have been fed,

re the situation? It's a hilarious manifestation of natural versus corporate personhood that the entire exec team are wanting. fugitives, but the rest of the company just keeps chugging. And that's like, yeah, yeah, this is something that's probably missed. It's like, so the deal CEO has been, they've been trying to subpoena him, and he's maybe in Dubai, and it's very controversial about whether or not he'll stay at the company because there's all these allegations.

Not much proved yet, but the, I mean, the signed affidavit was pretty crazy. The spy turned on them. The spy turned on them. It's not looking good. But what does that actually mean for the company? Like, the crazy thing is like, even after all the Xenifitz drama and Parker stepped down, I'm like, Zenevitz didn't do well, but it did continue as a company for a long time. People don't like ripping out their HRIS systems, like ever.

It sucks. Yeah, I mean, I think the factor the matter is, I imagine Rippling has an extremely strong sense of who uses deal. Yeah, right? For sure. Because they've talked to everyone. There's a lot of companies in the world, but there's not a lot of like, you know, there are a lot of high value enterprise targets. But I doubt there's many at this point that Rippling hasn't had like some comms with.

that have confirmed like, okay, they're using deal. And so taking the line right now to basically have your SDRs, BDRs, sales leaders reaching out to these people that you've had a point of contact with and said, hey, I know you're running on deal the years, we should consider having you guys switch over and then just sharing like a headline. That's kind of like it's fair, it's not, it's fair game.

Yeah, it is crazy that they haven't responded even, or at least that we've seen. Yeah, I expected them, I expected them all to step down, very quickly. Purely because it's probably in the long-term best interest of the company. Yeah, I thought so too. It'd be interesting to see, like,

what's the actual board structure? Because they've obviously raised a lot of money. They probably don't own more than 50% of the shares, but they might have super voting or something or board control. Who knows? But still, if, yeah, just in your self-interest of preserving the financial capital, if all the pressure is on you, get out,

it's not like you're necessarily going to lose all the value there. You could very easily maintain some of that if the company survives. Yeah, again, the other thing is, the other thing is thinking about the example I think you gave early on, which was, you know, if somebody comes to you and they go, John, you're using Michelin tires. Did you know that Michelin was spying on Bridgetown? You're like going to use those tires?

It'd be like, that sounds like a hassle. And you're like, okay. Like you seem way too interested in the tire market. I'm going to keep driving my car. Exactly. And so it's very possible that, you know, who knows? Maybe it doesn't have as much of an impact. But I'm sure it's going to have an impact from a recruiting standpoint,

which will have an impact long term. Yeah. Well, I want to close out with two more posts. Congratulations to Raul from Julius. He announced collaboration. He was talking about this, Figma for data analytics. He also put out a cool photo of him teaching. data analysis at HBS.

He's been fantastic successful. It was great having the show. I love watching the videos he puts out. He always looks like he's about to smile and then he doesn't. But I think he's just smiling inside because he's, he knows he's cracked. It's great. It's great.

And so, I mean, this makes so much sense. You build an iPython notebook or you do some sort of data analysis. You want to share it with all of your team members. And you want all of your team members to be able to edit on the same file and the same analysis. And this was interesting. John Conkel called. out that we posted back on February 20th of this year, Masa dropped his crystal ball while

pitching the SoftBank Open AI partnership with Sam in Tokyo last week, sending many analysts into a panic has he lost his vision. Was this an omen? Only time will tell. And that was basically the top. John actually commented back and said, time has told hypothetical performance of a tactical short-selling strategy triggered by the very impressive Stargate presentation and the crystal ball drop. There you go. Well, we should close out with some news that will be very important for all of our listeners. Gulfstream has announced the G800 certification today. This is rocking the group chats.

Yes, everyone was talking about this. So 8,200 nautical miles at Mach 8.5.7,000 nautical miles at mock. 0.9.9. Max speed increased to mock.935, all in the same body as the G650 ER, but faster and longer range. I'm sure a lot of you folks who are listening are going to be upgrading. So call your Gulfstream rep today because these are going to be flying off the shelves. Yeah, I mean, you're going to want to get ahead of this. Yeah. This thing looks fantastic.

It really does. By the time you get the jet and then outfit it, you're, you know, it's going to be. It's going to be expensive. It's a little while, but it'll be worth the weight. Yeah, it's a lot of money, but it's a lot of jet. And again, you know, you can sleep in. a G-800, but you can't fly a house. That's right, John.

It's going to be a few more years until Astro Mechanica can get us the Gulfstream equivalent of a... Yeah, that's a good way to justify this. Like, you're just holding yourself over for the Astro Mechanica, the boom, the Hermius, one of those guys. But in the meantime, pick up a couple of these, rotate them out every few years, and just use this to get around. Anyway, that's our show.

Thanks so much for watching. Thank you, folks. Hey, this was a much more stable. No cyber attacks today. No cyber attacks. Thank you to the brave soldiers and cybersecurity experts who kept us running. We'll see you tomorrow. See you tomorrow.

Thank you.