AI vs. Dog Cancer, Oscars Reactions, How to Lose the AI Arms Race | Kevin Espiritu, Paul Conyngham, Tony Zhao, Drew Oetting, Carina Hong, Cameron Fink, Debra Birnbaum

You're watching TVPN. Today is Monday, March 16th, 2026. We are live from the TVPN UltraDome, the temple of technology, the fortress of finance, the capital of capital. Let me tell you about Ramp.com. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more all in one place. Uh, what a massive week last week. Alex Karp going back to back with Travis Kalanick. Uh, the reactions to the the Travis Kalanick interview were phenomenal. I was reading them all weekend.

I was still emotional the next day.

Yeah. And there's something— I posted this on one of those clips that someone just shared. I was like, this is a great clip, and I was there. Because you're in the moment, and you don't realize it.

And rarely do I reflect too much on different interviews, because there's always the next day of interviews. But watching some of the clips back— Guillermo from Vercel put together that hour-long—

So good. —motivational video.

It was so good. I think that the Travis Kalanick mindset has been missing. Totally. When he kind of left, there's been a Travis-sized hole in the industry, in the culture. And to see him come back and in 45 minutes basically just give the advice that I think everyone that's building in some way can benefit from. Not everyone is going to be Travis, but— there isn't anybody out there that's done what Travis has done that is kind of like preaching that. And I don't like listening to like founder porn content personally. It's not, it's not appealing. But when it comes from Travis, yeah, it is just another level. Yeah.

Like the right message at the right time.

Yeah. Especially the thing, the thing that I was, I was kind of pulling on is like right now, like there's a lot of easy money everywhere, right? There's teams that have built nothing that can raise between $50 to $1 billion at times. And his feedback on that, his point of view was like, OK, is capital really a constraint in your business? How much does it matter? How much is it going to matter in terms of the competitive dynamics of your market? And if it matters,, and in a lot of these AI categories, it does. If it matters and it was easy, that means you didn't go hard enough.

Yeah, that was the best line.

And that was like the best line.

If money matters, as we all agree it does.

So you raised a billion? Why didn't you raise $2 billion? If money matters, why didn't you raise $3 billion? Yep. Like, oh, it was easy? Yeah. That means you didn't go hard enough.

Yeah, I mean, that's somewhat the subject of what Dylan Patel was talking to Dwarkesh about on the Dwarkesh Patel Podcast, fantastic show, by the way, fantastic episode, about this like, being risk-on, being aggressive. And Ben Thompson wrote about that today through a different lens, talking about, are we in a bubble? Maybe, but all the numbers are penciling out, so go, go, go. Now is the time to scale. And yeah, it was fascinating hearing it from a completely different perspective at the perfect time. But I really— yeah, that was a great interview.

That was personal highlight. For sure. Building TVPN. For sure.

Friday. Yeah, that was great. It really was the conversation that we set out to have because, I mean, he mentioned he's leaving California, but we're not going to get bogged down in his political views or whatever like that. It's so much more about the actual craft of scaling a business. And I think we just nailed that.

And so that was really fun. Yeah, and the good thing is we have plans to do a show like that. Every single day of the year. Yeah, every day. Unfortunately, it's not possible. It's not very often that someone like Travis— world historic founder— comes out of media retirement after almost a decade. Almost a decade.

Yeah. So very special. But thank you for everyone who tuned in. Thanks to everyone who enjoyed any of the clips, saw whatever you saw of it. It was a really fun time.

And If you care, if you want to work in physical AI. Yeah. And you don't see yourself in the Elonverse. Yeah. I think that is one of your best possible bets. Sort of like an indexed approach to physical AI.

You don't work insanely hard for Elon, work extremely hard for Travis.

Yeah.

It's not— you're going to have to work hard one way or another. Exactly. But that's the nature of what you're working on.

If you don't want to work hard, there's probably a company out there that's competing with Travis or Elon in physical AI. You could work there. I just wouldn't put much value on your, on your RSUs.

Yeah, it's rough. Anyway, let's pull up the Linear lineup. Linear, of course, is the system for modern software development. 70% of enterprise workspaces on Linear are using agents. We have Kevin Espiritu from Epic Gardening coming on to tell his story about scaling his YouTube channel. I think we have a lot to talk about. We always love creator economy stories. Paul Coeningham, the dog healer, is is coming to break down how he used AI to augment, delay his dog's cancer. We're gonna be digging into, we'll first go through what actually happened. I have some opinions about this and then we'll talk to him to get his side. Then we are pulling our delayed lightning round. We went far longer than we expected with Travis on Friday. So we are catching up on our lightning round with Tony from Sunday Robotics.

And we have Drew from 8Bit Labs. VC, he's a founding partner there. He and the 8VC team backed Quince at seed. It's now a $10 billion company. The Quince founder is a little under the radar. Totally. But I wanted to get this story from the 8VC team, hear how they're thinking about it, and then a bunch of other folks joining.

So looking forward to that. Well, let's read through Brandon Gorrell's deep dive on the AI versus dog cancer. What happened? So late Friday, there was a story about an Australian tech entrepreneur. Paul Coyngham, reducing the size of his dog Rosie's cancerous tumor by designing a custom mRNA vaccine with the help of ChatGPT. And it produced a substantial amount of discourse over the weekend. Separating facts from the hype cycle around the story, Coyngham is an AGI-pilled tech guy with 17 years of experience in machine learning and data analysis, at one time being a director at a nonprofit called the Data Science and AI Association of Australia.

Talk about an incredible association. We don't have enough data science and AI associations globally. It's true.

It's great to hear that. After his dog Rosie had been diagnosed with a deadly mast cell cancer in 2024, Cunningham used ChatGPT to brainstorm ways he could help. And he did an interview on this. And here's a quote from him. He said, I went to ChatGPT and came up with a plan on how to do this. The first step was to reach out to the university to get Rosie's DNA sequenced. Is there not a 23andMe for dogs yet or something like that? Who's doing— Who's doing full genome sequencing these days? I guess dog DNA is probably a separate assay, separate process.

Embark. Embark, you could do it.

Okay, well, anyway, he went to a university, probably for a good reason, probably got good data. He said the idea is you take the healthy DNA out of her blood, and then you take the DNA out of her tumor, and you sequence both of them to see exactly where the mutations have occurred. It's like having the original engine of your car, and then a version of the engine at 300,000 kilometers, down the road. You can compare them and see where there's damage. So once the University of New South Wales produced the DNA sequencing, Mr. Cunningham ran it through a whole bunch of different data pipelines. So there's a— this is something that we're going to go into, you know, throughout this story is the question of like, how much was this cure my dog cancer, one shot it, don't make mistakes? I don't think anyone's saying that, but very quickly there was like an incentive to amplify this into like the hype, like this crazy story. And then there was also, incentive to like de-hype this all the way, and the truth of course is in the middle, so that's where we're gonna get today. So once the DNA sequence was produced, he ran it through a whole bunch of custom different data pipelines to find those mutations, and then used other algorithms to find drugs to treat the cancer. With the help of the University of New South Wales, Coiningham identified a pharmaceutical company that produces an immunotherapy drug that looked like a good candidate for Rosie. So the drug already existed or was available, but the company The company refused to supply it to him because I don't think it was approved for this particular indication in this particular species. So he was out of luck there. So he then turned to, again, the University of New South Wales, their RNA Institute, which used Coiningham's data crunched down to a half-page formula to create a bespoke mRNA vaccine for Rosie. Again, from the story, Coiningham, ran an algorithm to inform the design of the mRNA and sent it to us, and we made a little nanoparticle, and it's democratizing the whole process, they said. This is Paul Thorndyson, the director of the RNA Institute at the University of New South Wales. So after several months of navigating red tape, Cunningham and his team administered the vaccine to Rosie, which was effective. One of her tumors shrank by half, though she is not completely cured. And that's just, kind of the nature of cancer, like cells are dividing all the time, everyone has some sort of low-level baseline of cancer, most dogs have like a little bit, the question is like, is it runaway, is it bad, is it terrible? And then it's hard to just like snap your fingers and cure it completely, but if you get the amount of cancer down really, really far, then you will of course survive. So the important thing is that Cunningham says the quality of life of the dog Rosie is much better now. So on X, the news of the story turned into a heated debate on health regulation.

Yes. What is that? That was for Rosie.

That's for the dog. Air horn for Rosie. Air horn for the dog. That's great. Turned into a heated debate on health regulation after biomedical engineer Patrick Heizer posted that quote, "It is trivially easy to make a single-arm mRNA vaccine. It's not hard." And Hank Green, prominent YouTuber, issued something of a rebuttal, which we can go through later. A separate thread in the discourse is focused on the promise of LLMs democratizing access to medical science, with OpenAI President Greg Brockman quote tweeting the story with the caption, "A small window into the opportunity of AGI." Well, Cunningham didn't literally cure Rosie's cancer with ChatGPT, as Stripe CEO Patrick Collison pointed out. It acted as a high-powered search tool that ultimately helped his team get to an amazing outcome. Sort of George Hawking.

So we got to move the goalposts?

I'm ready to move them. I think— Moving the goalposts. Where are we moving them to? It has to actually— you have to be able to type cure my cancer, and then from your phone, it just deposits a pill that you just take.

Yeah, exactly. Is that what it is? It has to locally end to end.

No, ideally, it would be not even a pill that you take. It can just create—

a video that you watch. The right pattern of light.

The right pattern of light coming from, and sound. So the phone has light and sound, and so the light flashes in your eyes at a certain rate, it rewires your brain, and your brain decides to go kill the cancer.

Yeah, and we've talked about this a bunch. I think it would be helpful for the industry to refocus some messaging on not AI is going to cure cancer, but humanity is gonna use AI to cure cancer and do a number of other things, right? And so the bar is not just like one-shotting it with a prompt, and it sends it to a lab and you get some type of treatment in the mail. Maybe we, I can imagine that in the future, right? Something to that effect, but it is an enabler, it's a tool, and this has allowed someone to become not an expert in something, but to help somebody understand a process enough to go out and find the right experts to help them solve their problem, and I think it's incredibly inspiring. So excited to have him on the show later.

So there was a chemist who works in AI and biotech by the name of Ash Jogalkar, and he had a really good summary along those lines with a riff on Freeman Dyson's 2007 New York Review of Books essay, "Our Biotech Future," which we should read at some point, in which, in this article, Freeman Dyson argues that biotechnology will become small and domesticated rather than big and centralized. The full post is worth reading in full, and we might go through it, but the conclusion is particularly good. If AI continues to reduce the cognitive overhead required to navigate biological knowledge and assemble complex pipelines, the boundary between professional research and motivated individuals may begin to blur. That shift will require careful thinking about safety, governance, and responsibility, but it also carries an exciting possibility. Dyson imagined a world in which biological design might eventually become something like a creative craft practiced not only by institutions, but also by curious individuals experimenting at smaller scales.

Yeah, I think there's, there's the reality of cancer treatment from my understanding is— and this is based on a late family member that had cancer and ultimately passed away— during the process, during his treatment process, which was around a year and a half, he was getting looks at different treatments that were promising, some of which he was able to do, some he didn't qualify for just based on his personal situation, even though there was there was a decent chance that it could have had a positive effect. Yeah. And that sort of the insane frustration that an individual feels or a family feels when they're like, hey, this, you know, if something's terminal or it's looking really bad, it's progressing in the wrong direction. And there's a treatment out there that isn't— that is somewhat trivial to actually make. Yeah. You just don't qualify for it. Yep. That level of frustration will eventually drive more individuals, I think, to do this, right? And so there's definitely some like safety, there's huge safety concerns, there's ethical concerns. These are things that we have to work through, but ultimately, I just think there's going to be so much, there's going to be enough like human energy and just overall desire to live that people will take risks that they wouldn't take for a bunch of other more sort of like trivial sort of issues.

There have been initiatives with the FDA, something around right to try in certain scenarios, patients' rights, sort of removing some of the regulation and allowing people to make decisions like that. It does feel like the FDA's stance might need to change in this case. Like they clearly have a role to play currently and in the future where biotech becomes more democratized. But yeah, hopefully there's like some good symbiotic relationship there with the broader biotech community as it gets bigger. I have a similar story with someone who developed a rare illness and was able to go and read academic research at a very deep level. Didn't have a background in biotech or anything like that, but was able— this was pre-AI. Was able to read like every published research paper that was at all related to this particular illness and found the world expert in this particular disease, contacted the professor, and the professor said, yes, you have the thing that I've been studying, and I've only found 5 people or 10 people in my entire career that have this thing. Come down. I will operate on you. The operation happened, it was successful, and it was fundamentally like a high-agency person doing a lot of research. And if AI just acts as a search tool that democratizes that, you're going to get better results. So even if you're— even if we're not in like one-shot in curing cancer, that just feels like making search easier, making research easier, huge benefit.

Yeah. Cunningham. The guy in Australia, could have done a lot of this 10 years ago. Totally. He just would have needed to spend, I'm sure, a bunch of time in libraries. Yeah, that's the thing. Everything you do with all these things— you can do manually. You can just get a guy for that.

Yeah, you can get a guy, or you— I mean, you don't even need a spreadsheet. You can calculate the math by hand, but these things speed things up. So it's been a good time. Let's read through Ash's post, but First, let me tell you about Public.com, investing for those that take it seriously. Stocks, options, bonds, crypto, treasuries, and more with great customer service. And let me also tell you about Finn, the number one AI agent for customer service. If you want AI to handle your customer support, go to Finn.ai. Thank you for clapping, Tyler. How was your weekend, Tyler? It was good. Yeah, it was good. Yeah. Did you go to any data centers or are you—

No data centers this weekend. No.

I was in SF.

Didn't you go to a pig roast? Yeah, that was on Friday. That was in El Segundo. How was SF?

Is something big happening there? Does it feel like being in—

Wuhan in February of 2020. Something, something big was happening. Yeah, there was— I went to a debate.

Oh, you went to the debate? Okay, cool.

How was that? It was good. Yeah, yeah, it was about the billionaire tax. Yeah, yeah, yeah.

And did you go to the hackathon at all?

No, I missed that.

Okay, I saw that Semianalysis had a hackathon. The winners were crowned, seemed like a lot of fun. It really does seem like the best time to go to hackathon, uh, just because What you can actually accomplish in 2 days is remarkable.

Yeah, yeah, no.

Right? Yeah, it's like people used to do hackathons and it'd be like after 2 days they'd be like, we have a landing page. And now it's like—

And a cool idea.

We created a hackathon simulator with minigames for everything, and it's also making money. We need to give an update on TVPN Simulator at some point, but it is coming along. The development has continued at breakneck pace. Yeah, we got to work on the rollout of this. Yes. It might be GTA 6 level by the time GTA 6 comes out. I think we can get there. We need a new graphics package. What do you think the actual path to AAA graphics is? Do you think we should rewrite it in Unreal Engine with ray tracing and insist that people only run it on gaming PCs? Or should we do some sort of style transfer on top of it?

Yeah, I mean, I think the Unreal Engine is probably easier, right? Because you're just like moving the code over. That shouldn't be like that difficult. You can probably do that in like a day or two.

I think these things used to take like years. Like, it took Elder Scrolls like a decade to get to like Nintendo Switch.

You're like, yeah, it takes a day or two. The real-time rendering thing is interesting, but I think that's— it's just like expensive. That's the problem.

We had someone on the show that was doing it on Zoom over in real time.

That was Descartes?

Descartes, yeah. That was a cool demo. So you imagine like That tech prompted with, like, take this from boxy— I would say we're above N64-level graphics, but we're probably more like Xbox 360 graphics. And take us into modern-day PS5.

Yeah, I mean, this is why I'm very excited about doing— everyone's so up in arms about, like, oh, the new PSX isn't going to come out because you can't get the memory. And people are like, oh, I don't want to play games in the cloud, right? But if you're in the cloud, that means you can actually access a ton of compute because when you're not playing, when you're not using the GPU to run the nice graphics, someone else can be. You can actually get access to much better hardware when you're playing video games.

And then also, yeah, more iteration on the graphics. It should just be like live service model, basically.

Yeah, and if you get the Genie 3 model where it's actually generating on the spot, that's something you can really only do in the cloud.

I'm excited. Jensen is doing his keynote at GDC. Should we pull up the live stream? We can, yeah. Let's check in with Jensen.

Let me tell you about Okta first. Okta helps you assign every AI agent a trusted identity so you get the power of AI without the risk. Secure every agent, secure any agent. And let me also tell you about Graphite, code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. Continue. Let's play it. What do we got? We got Jensen. Institutional investor. These 3 people are deep in technology, deep in what's going on, and of course they have just a really broad reach of technology ecosystem. And then of course all of the VIPs that I hand-selected to join us today, all-star team. I want to thank all of you for that. All-star team. The leather jacket really has just aged so well. I also want to thank all the companies that are here. NVIDIA, as you know, is a platform company. Mic drop. We have technology. We have our platforms. By the way, everyone uses us.

He's mocking our merch. He is.

And today there are probably 100% of the $100 trillion of industry here. 450 companies sponsored this event. I want to thank you. $100 trillion of industry. $1,000— I love it. Technical sessions, 2,000 speakers.

This is 2,000 speakers.

Wow. Every single layer in one. They're going to do more interviews than we've done all year. The infrastructure in one day, chips for two days, to the platforms, the models, and of course, the most important and ultimately what's going to get this industry taken off is all of the applications. This really is the symbol for semiconductors. It all began here. This is the 20th anniversary of CUDA. We've been working on CUDA for 20 years.

20 years.

For 20 years, we've been dedicated to this architecture, this revolutionary invention. SIMD, single instruction, multi-threading.

All right, very, very cool. Let's get back to the timeline.

Let's go to Gemini 3 Pro. Gemini 3.1 Pro is here with a more capable baseline. It's great for super complex tasks like visualizing difficult concepts, synthesizing data into a single view, or bringing creative projects to life. And let me also tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more while Railway automatically takes care of scaling, monitoring, and security.

Back on the timeline case test. Okay, we gotta go through. Wait, yeah. You can just save your dog with the beautiful picture here. You can just save your dog.

It's remarkable. This is a heartwarming story and it also, yeah, I really like how it reveals like current AI capabilities, where things are, the benefits and sort of the diffusion narrative. Like this is fundamentally a diffusion story, not a superintelligence story in my opinion. But let's go through Ash's post here. My take on the whole AI cures dog cancer in Australia is a very interesting story, but perhaps not for the reasons that are being noted. In 2007, Freeman Dyson published an essay in the New York Review of Books called The Biotech Future. It contains one of the most memorable predictions about the future of biology that I've ever read. I predict that the domestication of biotechnology will dominate our lives during the next 50 years at least as much as the domestication of computers has dominated our lives during the previous 50 years. Dyson believed biology would eventually follow the trajectory of computing. At first, powerful tools live inside large institutions, universities, government labs, major companies. Over time, these tools get cheaper, easier to use, and more widely distributed. Eventually individuals start doing things that once required entire organizations. You will be the manager of infinite minds. You will have, you know, a million agents and you will also have access to the equivalent of a university lab filled with biotechnology equipment. Biotech will become small and domesticated rather than big and centralized. This is very interesting in the age of AI because there's been this narrative of like AI is a centralizing technology. It is very power law driven. But this is sort of counter to that. I don't exactly know how to piece those two things together, but it is interesting that his prediction was actual decentralization in this particular category. He even imagined genome design becoming almost artistic. Designing genomes will be a personal thing, a new art form as creative as painting or sculpture. Dyson's words rang in my mind as I read the AI cures dog cancer story. Much of the coverage framed in—

I got to say, it's very easy to imagine you in 20 years. I'm like, John, like, you got to tell us your anabolic steroids. And you're like, it's kind of a personal thing. It's kind of a personal thing. It's kind of like an artisanal process that I go through, like a sculpture. I'm sort of sculpting myself. I can't really— I can't— I'm sorry, I can't— I can't really share my stack with you, but it's a personal thing. So go and kind of figure out your own stack.

Speaking of sculptures, I was walking around my neighborhood and I looked through this gap in the trees into this large lawn and I saw on this person's front lawn behind gates and whatnot, just a full-size statue of a man playing golf who I didn't recognize. It was not Tiger Woods.

I think it was the owner.

I think it was the owner. I think the owner was like, I'm into golf and Or, you know, like one of his boys got it for him, which is a hilarious gift.

Getting someone a life-size statue of themselves.

And just having it delivered, and then it's like, well, it's impolite for you to turn it down. You know, what are you gonna do? Anyway, the scientific pipeline involved here is actually well known. It closely mirrors the workflow used in personalized neoantigen vaccine research that has been under active development for years. The steps are fairly standard. Sequence the tumor, identify somatic mutations, predict which mutated peptides might be recognized by the immune system, encode those sequences into an mRNA construct and deliver them to stimulate an immune response. The biological targets themselves were almost certainly not new discoveries. I have been unable to find out what they are, but mutations in targets like KIT, which are common, might be involved. Partly therein lies the rub, since the hardest part of drug discovery, whether in humans or dogs, is target validation. The lack of which leads to a lack of efficacy. The number one reason for drug failure in neoantigen vaccines. The proteins involved are usually ordinary cellular proteins that happen to contain tumor-specific mutations. AlphaFold, which was used to map the mutations onto specific protein structures, is now a standard part of drug discovery pipelines. That's fascinating. The challenge is identifying which mutated peptides might plausibly trigger immunity. What is interesting though is how the pipeline was assembled. Normally this type of workflow spans multiple domains, genomics, bioinformatics, immunology, and translational medicine. And in institutional settings, those pieces are distributed across specialized teams, document sources, and legal and technical barriers. Navigating the literature, selecting computational tools, interpreting sequencing results, and designing a candidate mRNA construct is typically a collaborative process. In this case, AI appears to have helped compress that process, pulling together data and tools from different sources. Instead of requiring multiple experts, a motivated individual was able to assemble the workflow with AI acting as a kind of guide through the technical landscape. That is fascinating. Anyway, it's a longer post, but you should go read the thing in full. Patrick Collison also chimed in. He said, according to the story, the dog's cancer has not been cured. I think it's just, 50% smaller, which of course is a win, but not using the term cure is always tricky, but it does go viral. Absent all regulatory and manufacturing constraints, we could not just synthesize magic RNA, mRNA cancer cures. The technology is very promising, but it's not any kind of panacea yet. The emergent system of regulators and manufacturers is indeed far too conservative, and small-scale experimentation is much harder than it should be. More people should read the first part of The Rise and Fall of Modern Medicine. So it's interesting.

Lee says, ChatGPT, cure cancer, make no mistakes. Biomedical engineering industry. Yeah, don't do this. It's easy and effective, but we can't make enough money off of it. Ridiculous.

It's surprising. G. Fodor says it's surprising how people are so blatantly talking past each other on this. The point is that the system of clinical trials is predicated on an assumption that a given drug will work on a cohort. What if there are lots of drugs that will only work on one person. So definitely a big desire and push for rethinking the system of clinical trials if you're going to have personalized medicine. What does that mean? There's already a lot of people, biohackers, that do all sorts of stuff like this.

Can't believe he wasted 2 cups of water to do this. #BanAI. It is ridiculous. It's a great counterpoint.

Point to the doomers. What else is going on? Mark Andreessen chimes in.

I can't load the post right now. We got to go to probably the most important story of the day. Gabe says he had a dream that Apple released a 32-inch MacBook called the MacBook Pro Ultra Wide, and it looked like this. I bought one and unlocked extreme productivity, and then it wouldn't fit into my backpack, so I had to leave it behind. Oh no.

This is sort of like that other laptop that we saw.

They should honestly make this. They should. This should— I mean, walking around looking like— maybe you could put skateboard trucks on it. Yeah, that you could use it as transport.

Yeah, it's more of like a snowboard build that you like carry over your shoulder like this, or surfboard. You know, people throw it on the top of your car like that.

Three fingers. Why?

You don't put a surfboard on the top of your car?

Uh, yeah, I mean, real ones don't.

Oh, what do they do? They put it inside the car?

Truck bed or inside? Truck bed. Okay.

Yeah, well, I don't pretend to be a surfer.

In the LA area, you can clock if somebody is actually a surfer or not just by the way they go to the beach with their board. No, but I think, yeah, throwing it under your arm, having some trucks on it, skating, being able to get where you need to go.

I like the ultra wide. What if they're driving a Huracán Sterrato? Where would you recommend that they put their surfboard then, Jordy?

Sterrato, I could make exceptions.

Okay. I like this. Dylan Patel said, on door the TAM for GPT-5.4 is north of $100 billion, but there's adoption lag. That's considered AGI as far as the Microsoft OpenAI contract is concerned. That's very interesting. Sam Carter says the reported $1 billion of profit is no longer the sole trigger for confidential IP research access. It reportedly includes an independent expert review. You were saying Joe Rogan would be on that. Andrew Huberman. The experts would be on there. You gotta trust them at all times. And— The Yvonne, maybe? Yo, the funniest thing about that joke is that like, I actually would like to know that panel of experts, whether, where they deem AGI, because I feel like between all of them, they could chat with the chatbots and be like, "Ah, it's like not that good yet." You know, like be very realistic about it. Yeah, they're not necessarily just gonna be like, "Oh, I'm pumping it for whatever reason." And they're like, "I have this weird bias," or whatever. And so it'll be very interesting to see how the AGI definition plays out because it does feel like we're close. I mean, Dario on DoorCash was saying like we're near the end of the exponential, which is like sort of crazy. It feels like, you know, Sequoia declared AGI, they're an investor in OpenAI. And so there's a lot of stuff. What do you think about the AGI time? Do you wanna be on the expert panel?

I think I would say that we already reached AGI. Like it was maybe, earlier.

You called it like 30 seconds before Tyler Cowen did. I think I remember it was like 30 seconds before you came, you tapped me on the shoulder and you said, it's here. And then we went and refreshed X and Tyler Cowen had come out.

Yeah. I think like realistically, I'd probably say it was something like when the, you know, agentic harnesses came out. So stuff like Claude Code. Sure. Where you can actually just like tell it to build a project. And then it would, like, it'll, there'll be errors. It'll see those errors, it'll fix them. It'll like keep working on it.

Not reasoning models? I mean, it's so hard.

It's like on like math or something like this, right? Like those basically unlocked like, yeah, now they can just do anything.

Yeah, yeah. I mean, the agentic thing was talked about for a full year and then it finally happened like in December and it was pretty broken up until then. And then all of a sudden.

Yeah, but I think like you can still just make like a very good case that like, yeah, ChatGPT, like that was AGI. AI. Like, you can just ask a question, it'll answer it. Yeah. If you'd never talked to an AI model before and you talk to that, you're like, this—

okay, this is a person. Microsoft Excel, 1985, AGI.

Uh, Jose Macedo says, ultimate narrative violation from the Dylan Patel to Arkesha. 3 years later, H100s are actually trading above launch price in secondary markets, i.e., negative depreciation. Yeah, that's called appreciation. Appreciation. I appreciate— I just want to go out and say I appreciate H100s. Negative depreciation. Yeah, that's funny. Uh, this completely flips the Michael Burry 2-year e-waste bear thesis on its head. Yeah, I mean, somebody's got to check on—

somebody's got to check on Michael Burry. It's such a different— it's such a different dynamic because, I mean, like, the whole— there was a reasonable underpinning for GPU depreciation, which was just look at 20 years of computer equipment history. It's like it all depreciates over like maybe 5 years, maybe 10 years. Some stuff sticks around, but like they burn out.

Yeah, it's just interesting. Jose said Coree probably benefits most from this. They have 250,000 GPUs and a $66 billion backlog. Depending where you think market was pricing depreciation, margins improved by something around 40%, which means $1 billion a year in additional earnings. Who knows where this stuff actually re-rates or how sustainable, but great, great time to be a Neo Lab. One of the founding team members at Lambda was posting last week, basically, congratulations to everybody that booked out GPUs on an annual basis in 2025. You're looking absolutely brilliant right now. Obviously Sam is starting to look extremely vindicated on all of the deals that he did last year. So Tomasz.

Quickly, let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it. And let me also tell you about Turbopuffer. Serverless vector and full-text search built from first principles and object storage.

Fast, 10x cheaper, and extremely scalable. Tomasz says, we've been growing a lot and are out of GPUs. This is Sam Altman in March of 2025. Satya Katz over at Oracle says, we are still waving off customers or scheduling them out in the future. This is a situation that we have not seen in our history. Satya says, you may actually have a bunch of chips sitting in inventory that I can't plug in. I don't have warm shells to plug into. Sundar says, what keeps us up at night? The top question is definitely around capacity. All constraints, be it power, land, supply constraints, how do you ramp up to meet this extraordinary demand?

And so quickly, between power, land, supply chain constraints, chips, you were saying that the takeaway, your takeaway, your read on Dylan Patel on DoorDash was that chips were the main—

Yeah, I mean, not even like a read, like he explicitly said, he's like between power and chips, chips is what's going to be the big bottleneck because at some point, like, yeah, there's all these ways that you can actually like maybe get like 10% of the, you know, US energy production to just like go to, yeah, you know, yeah, where like at some point, like, okay, we don't have enough like EUV tools. Yeah. And like they're not building them right now, which means that they're not gonna have them for at least 3, 4 years.

Yeah, yeah, yeah. This was Ben Thompson's, like, TSMC needs to step up and spend more on CapEx. Their, their CapEx guide is like a CapEx guide for ants, like a mere $45 billion or something. It should be probably much, much higher, basically.

Yeah, but I mean, it even goes down to the toolmakers below them, like really deep in the supply chain. Yeah. At least what I got from Dilip Patel on that interview is that they still are not really that AGI-pilled. They're not expecting this kind of massive increase in demand to stick around. Yeah.

Trey says a sign of taste is dabbling in the vintage GPU market.

A100s. Yeah, Ampere. V, you gotta go Volta. Go back to Volta. The, yeah, I mean, I remember I was digging into that like chips versus energy, what's the big bottleneck? And I think we're using something like 50% of leading edge capacity, like of the fabs that can make AI-powered GPUs, like GPUs that can run transformer-based large language models, we're using like 50% of that capacity already. Some of the leading edge nodes go towards like, you know, Apple Silicon chips that are maybe designed system on a chip, something for a phone. And only like 1% of energy right now in America goes towards AI or less. It's like 0.1% or something. So you can reallocate and everyone just turn off your air conditioning.

One more.

Close the door if the air conditioning's on.

Lip Buton says there's no relief as far as I know. No relief until 2028. Somewhat ominous. Keep reading. What— Tamaz says, what happens when your AI doesn't answer? Everything is in short supply. It's no longer just GPUs. It's power, data centers, memory, CPUs. If there's no relief for 6 more quarters, perhaps it's time to plan for a world where inference isn't freely available on demand. Inference prices, which have been static, will rise. Subsidies will be harder to justify. Enterprises will need to rationalize workloads, deciding which teams receive state-of-the-art models and which don't. Not every CRM update requires a trillion-parameter frontier model. Inference rationing normalizes market. Marketing receives this much, sales receives that much. Software engineers probably receive a lot more. Constraint will be the mother of invention. Companies will optimize what they have, adopt open source where they can, and likely move to smaller models for many workloads.

This is a really cool take. I like this. It's also interesting to me is that, that not every CRM update requires a trillion-parameter frontier model. That's another bull case for Hopper price stability going forward and less depreciation because if you can leave your CRM update workload where you're just going through and spell checking names and cross-referencing data sources and pulling from email and dumping some notes, summarizing, and you can do that on a GPT-4 class model instead of using 5.4, you can probably distill that model, boil it down, run it on an H100 fleet really efficiently. And so that's still economically valuable. And so you're able to continue that.

Should we, should we go over Ben Thompson's post from this morning? Yeah, we should. Now would be a good time. Yeah. It is.

First, let me tell you about Labelbox RL environments, voice robotics evals. That's where human data— Labelbox is the data factory behind the world's leading AI teams. And let me tell you about Vibe.co, where D2C brands, B2B startups, and AI companies advertise on streaming TV. Pick channels, target audiences, and measure sales just like on Meta.

Agent Thompson published this this morning. To me, the second I saw that, I started reading it, it felt like taking a double scoop of C4. Is that a pre-workout? Yeah.

I know the can. I didn't know it was a—

You never dabbled?

What was the one that we— I'm more of the Gorilla Mind one. That's the one that I know. Many people have said you have the mind of a gorilla. Yes, yes. Yes.

For more plates, more dates. You're a gorilla in sheep's clothing.

I think that's literally the pre-workout that I have, although I don't use it that often.

Anyway, so you got pumped up. I got pumped up. Ben writes, there is a weird paradox in terms of AI prognostication. Prognostication. Prognostication. That was a good, good effort, Jordy.

On one hand, it's like, what are some of the requirements?

There's just so many words.

What are the requirements for having a podcast? Like knowing how to say words?

No. Taste. I mean, yeah, ultimately there's a lot of words that you, when you read them, you're just like, oh yeah, you can just do it and then you try to rip it. On one hand, you don't want to be the one to completely dismiss the most terrifying doomsday scenarios. Who wants to be found out to be foolishly optimistic? At the same time, there's also pressure to give credence to the possibility that we are in a bubble and all of this hype and spending is going to go belly up. While I have argued against the former, I've very, I've, I've very much been on board with the latter, making the case that bubbles can be good. Sitting here in March 2026, however, on the morning of NVIDIA's GTC, I've come to a different conclusion. I don't think we're in a bubble. Let's go. Which paradoxically may be the truest evidence we are.

Where's the bubble gun? Let's get the bubble gun going.

He writes, LLM paradigms over the last couple of weeks, first in the context of NVIDIA earnings and then last week in the context I've talked to you about 3 LLM inflection points. I'm not gonna go through all these because ChatGPT—

We talked about this a few times. Oh, why? Basically LLMs, reasoning models, and then agents, and each one of those increases the demand exponentially for compute.

Yeah, so ChatGPT, oh, one, and then Opus, as well as Claude Code and Codex, basically getting to the point where tasks are being accomplished over hours. Yep. And getting to great outcomes. And this is the interesting point. Okay. The decreased need for agency. The reason Ben has been writing about these three inflection points over the last couple of weeks has been to explain why it is that the industry is so compute constrained and why the massive investment in the CapEx by the hyperscalers is justified. The first paradigm required a lot of compute for training, but inference actually answering a question was relatively efficient. You simply sent the user whatever the model spit out. The second paradigm dramatically increased the amount of computing needed for inference for two reasons. First, generating an answer required a lot more tokens because all of the reasoning required tokens in addition to the answer itself. Second, the fact that reasoning made the model so much more useful meant that they were used more, which drove increased token usage in its own right. It's a third paradigm, however, that has truly tipped the scales in favor of CapEx expenditure not being speculative speculative investment, but rather badly needed investment in meeting demand that far exceeds supply. First, generating an answer will often entail multiple calls to a reasoning model. Second, the agent itself needs more compute, and that compute and the tools the agent uses is better done by CPUs and GPUs. Third, agents are another step function increase in usefulness, which means they're going to be used even more than even reasoning models in a chatbot. It's how this third point will be manifested that I think is underappreciated. After all, far More people use chatbots than agents, but I would make the case that most people are not using chatbots as much as they should. It's been a question of agency. To get the most from AI requires actually taking the initiative to use AI. And he goes into a little bit talking about local compute, talking about how Apple's opportunity to run LLMs locally.

There was a very, very interesting take in here where he's talking about the Apple MacBook Neo launch, which is $599, I think $499 for education, potentially very disruptive to other laptop makers.

You said you still get discounts, Tyler, or does it? I think I'm still—

Oh yeah, you're still on leave because you're on leave. Yeah, that's great. There you go. There are some legendary honorary leave of absences where people have been away for like 10 years and then they go and do so many— see, the goal is to defer for so long, but then also have such a meteoric rise that they have to give you the honorary degree before— while you're still eligible. That's a good one. Because they're like, oh, well, we gotta— I think Mark Zuckerberg got an honorary degree from Harvard, but he was on delay for like a year. And I think they gave him the honorary degree a couple years later, so. So, you know, that's the speed run to beat. But the point about the MacBook Neo is that at $599, a lot of PC makers should be sort of quaking in their boots because you're selling at that price point. And for a customer who's just like, I want a $600 laptop, normally it was like, am I going with like ASUS or another brand? I'm not in the Apple, category. Like it's not an option because that store over there, those laptops start over $1,000. That's not my budget. So I'm not even going in that store. Well, now you can and you can spend $600 and get a pretty good computer. And the CFO, Nick Wu of ASUS, was on their recent earnings call and he said, actually, don't worry about it. It's not a threat. We found out about the MacBook Neo shipments. In the second half of last year. We made some internal prep, but now that it's out, like, we don't think it's that big of a deal. Like, it has some limitations. Specifically, it only has 8 gigs of RAM. So like, you know, this is more focused on content consumption. It's not a mainstream notebook for notebook usage, for creation, for working. It's not a work device. It's a consumption device. It's more like an iPad. And Ben Thompson's point is that, well, like, that's what people use these laptops for now. It is a lot of consumption. There aren't as many people who are in that $600 price target that are wanting to run powerful applications at that price point. As soon as you're running powerful applications locally, you're probably more of a business buyer and you can spend more. And so— and then he goes into apply that to to AI, talking about enterprise and the value of— companies have a demonstrated willingness to pay for software that makes their employees more productive, and AI certainly fits that bill in this regard. What makes enterprise executives truly salivate, however, is not the prospect of AI eliminating jobs, but doing so precisely because it makes the company as a whole more productive. So increasing production. Yeah, he basically—

my interpretation, he's making the case that there are companies that could cut headcount and actually just grow faster. Yeah. If they're implementing AI properly, not just replacing like the routine workloads. Yeah. So he says agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value. Okay. Actually, I'm going to start one paragraph. Yeah, please. It's always been the case, even in large companies, that a relatively small number of people actually move the needle and drive the company forward in meaningful ways. That drive, however, has been filtered through a huge apparatus filled with humans who accelerate the effort in some vectors and retard it in others. That apparatus makes broad impact possible, but it carries massive coordination costs. Agents, however, will tilt much more heavily towards pure acceleration, making those drivers of value much more impactful. I'm sympathetic to the argument that the best companies will want to use AI to do more, not simply save money. The reality of large organizations, however, is that the net positive impact of AI will not be in eliminating jobs, but rather replacing hard-to-manage and motivate human cogs in the organizational machine with agents that not only do what they are told, but do so tirelessly and continuously until the job is done. This only makes the argument that we are not in a bubble much more compelling. First, there's—

unless there's a compute constraint and then the models get lazy and they're like, I don't know about tirelessly or continuously, I'll I'll get around to it when I feel like it. I'll give it a crack.

I'll get around to it. Yeah. This only makes the argument that we are not in a bubble that much more compelling. First, all of the weaknesses of LLMs are being addressed by exponential increases in compute. Second, the number of people who need to wield AI effectively for demand to skyrocket is decreasing, right? You have one Tyler, just, you know, he's going to— Tyler's going to set up to be able to do sign language with his agent. Tends to just be not even speaking, just sending.

Do you actually ever use any of the voice models? Remember Karpathy was talking about that?

Yeah, a lot of people do this because you can just, like, talk much faster, I guess. I haven't done this really. I've used it, sometimes use the voice mode, but I don't actually use it in, like, coding agents yet.

I was using the ChatGPT voice mode, like the true, like, back and forth voice mode. Yeah, like real-time voice. Real-time voice mode in the car this morning. And they improved that thing dramatically. It's good. Yeah, it's so much better. So first off, it doesn't do that. Like, that's a great question or anything like that. Or that whole pause that was in the Super Bowl ad like that just doesn't exist anymore. It just answers and it answers in these like really short, punchy things. I was asking it about like how many jobs are actually in America and it just says like 164 million. And it just like gets me the answer and I'm like, how many jobs are there in China? It's like 730 million. And I'm just able to go back and forth with it and ask more and more detailed back and forth without needing to like dictate a whole prompt and then let Pro cook on it for 10 minutes, come back, have it read it to me. It was like a much better experience. I was very pleasantly surprised by how the back and forth worked. And they also changed it so that that you see the floating bubble of like the little animation, but the text populates in real time with your question and then the answer and then your question and the answer. So you can just scroll and read as well.

It's very cool. Anyway. Third, the last argument that we are not in a bubble, that economic returns from using agents aren't just impactful on the bottom line, i.e., saving on cost, but the top line as well. Let's go. In this context, It is any wonder that every single hyperscaler says the demand for compute exceeds supply, and that every single hyperscaler is, in the face of stock market skepticism, announcing CapEx plans that blow away expectations. So I encourage you to go subscribe to Strategery, max out your plan, pay annual. But this was extremely notable.

It's such a funny ending where he, He has this point about like, you only need to be worried about a bubble when— like, you don't need to be worried about a bubble if everyone's saying a bubble, because then everyone's like risk-off because everyone agrees that we're, oh, we're in a bubble. Let's not do bubble behavior. And so capitulation is the sign of a bubble. And he's like, I understand that. And still, this is my take. It's a bold take, but I think it's a good one. Really quickly, let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. We talked to John Zito at the New York Stock Exchange a couple months ago. Now he's in the Wall Street Journal talking about arrogance in private markets.

Take us through it, Jordan. He was going hard. He was going hard. Yeah, we'll click into this. Top Apollo executive sounds off on arrogance in private markets. He says, I literally think all the marks are wrong. Apollo's John Zito said of private equity in previously unreported comments. Apollo says comment was about software companies. Let's go through it. Executives at the biggest private credit lenders have sought to play down an exodus of investor money from their funds, making carefully worded television appearances to calm jitters about the sector. Apollo's John Zito, former guest of the show, co-president of the firm's asset management arm, that is one of of private credit's largest players, spoke more bluntly in a previously unreported discussion that UBS arranged for some of its clients late last month. Zito called out arrogance in private markets, predicted that a private credit loan made to a generic small or midsize Joe software company might recover 20 to 40 cents on the dollar, and said Federal Reserve Chairman Jerome Powell is needling President Trump with his inflation commentary. According to audio recordings of the comments reviewed by The Wall Street Journal, It sounds like this wasn't meant to be public.

I don't know.

Fido also detailed— Calling, you know, people in private markets arrogant is crazy. I feel like I know a ton of people in private markets. I don't know anyone who's arrogant at all. It's like remarkable. So a bold, bold call by him. But we'll see what evidence he has to back up that extraordinary claim.

He blamed the media for creating a frenzy around private credit. Obviously, we're in the middle of a private credit party. Apparently, if you do credit well, it's honestly, I would say we don't understand private credit well enough to like really put everyone up into a frenzy. He says if you do credit well, it's supposed to be pretty boring. If you do stupid things and you do concentrated things and you do things that you're not supposed to do in your vehicle, you probably will have a bad ending. Yeah. So, you know, talked about the sell-off in shares of large software companies, which was largely sparked by fears about AI. He cautioned that smaller software companies bought by private equity Many with private credit loans could face even more challenging conditions. Those dismissing concerns by pointing to strong results from public companies are missing the point. I'm not as rosy and I'm not as confident in what will happen with the technology. Anyone who tells you that the earnings last quarter were really good, so all is good. Anyone who says that clearly doesn't understand. Most of the businesses that were bought from 2018 to 2022 are lower quality than those companies because they're not public, smaller than those companies, and we're trading at a much higher valuation than those companies. And so I am concerned about many of those take privates.

Yeah, I remember a lot of— like, Logan Bartlett was doing a ton of analysis in the end of the ZIRP era at how high the multiples were in the public markets. And that's what was driving the 100x ARR transactions. And you have to imagine that even if we were like, oh, yeah, that VC-backed company was sort of overhyped at 100x ARR, Well, that still has a trickle-down effect to the private equity buyout that's just like—

Yeah, and remember last year when Figma went out, and they priced it very reasonably, right? They were very intelligent how they priced it. But then obviously, there was so much excitement because it's such a great company. It ran up. Round trip. It ran up. In the first couple of days, there was some late-stage private SaaS companies that I remember were posting like, maybe we should go public. I remember those, yeah. I think it was the— Parker Rippling was like, oh, if I can get—

that's a crazy multiple.

Yeah, if I can get, you know, some insane revenue, revenue multiple, maybe that's appealing. Obviously, a lot of those names still could get out this year. Strong companies, they're not, not as, as eager to get out. Zito pointed to Thoma Bravo's 2021 $6.4 billion take private of the software firm Adalia. In particular, several lenders to Adalia, including Apollo, have already written down its debt.. He says there will be an issue with respect to that credit, which I think will be worse than people expect. Asked what kind of recovery rates he anticipates on private credit loan to generic smaller midsize Joe software companies, Zito said Joe Software Company, if he's in the wrong place, I think is going to recover somewhere between $0.20 and $0.40. So 60 to 80% markdown. A lot of the private credit firms have been they'll mark down a loan, but like mark it down to like 95. Sure. You know, sure. Like, you know, nothing very significant. Tito noted that he expects private credit loans originated in the next 12 to 18 months to be a much better vintage as it relates to quality of company, amount of leverage, documentation, and spread. He also weighed in on redemptions and whether private credit managers should enforce limits, typically 5% of a fund's shares each quarter, or allow more investors to cash out. When they are flooded with requests. It is a topic he and others on Wall Street have recently been asked about as funds take different approaches. You're going to see elevated redemptions for a handful of quarters. I don't know how long it lasts. Making a decision in one quarter may be the right, like, decision for fundraising in the near term, and then a quarter later you'll realize it was a really bad decision. So my overall bias is to stick to the 5% to protect all of my existing investors. Mm-hmm. On vulnerabilities in private equity, Zito sought to shift the focus to private equity, where Apollo has less exposure than most of its peers. He suggested investors' voracious demand for buying stakes in existing private equity investments, but wariness of the private debt underpinning those deals doesn't add up since the equity would be junior to the debt if there were major problems with these assets. There's unlimited demand for secondary private equity, but they're worried about private credit, which finances 80% of those portfolios. I can't compute, but I'm the dumb guy. I don't understand. I start saying this and I get these blank stares back at me like, okay, I don't know. He said, I literally think all the marks are wrong. Is that what you're asking me? I think private equity marks are wrong. Wow. And again, I read into this, he's talking so candidly. Yeah. At least in private equity land, he doesn't feel exposed enough to be freaking out. Yeah. Let's see a couple more quotes. He says this next cycle is going to be a big moment in time for the private markets because people are way smarter than I think private market participants, particularly people in the wealth channel. Like, I kind of sense an arrogance of the people who grew up in the private markets business. If you don't mark your book, I think you actually lose trust with the clients. We're going to be the market leader in actually marking our book. Let's give it up for being the market leader. On the economy and markets, he said, I think it's more likely than not that we go into a recession, a consumer confidence-led recession. Most of the companies we lend to are getting a lot of pressure to show clear execution. It's forcing people to do stuff before the actual technology works. That's going to be the first step of just a slowdown in the broader economy. Interesting. He said, I literally think Powell, he's so upset at how it's ending that he's just saying there's inflation every day to piss off the president. Like, I literally think that's what's going on. It's so hard for me to see inflation. I don't see it anywhere. I see it much more deflationary. I think the technology is attacking every profit pool.

What do you say? Asked why a popular high-yield corporate bond ETF, seen as a benchmark for such that is typically under pressure in an economic crunch was relatively flat for the year? He says, I don't have any idea. The amount of dispersion going on beneath the surface is kind of crazy. I literally at home, I told my wife last night, I feel like the market should be down at least 10% and it's flat or up. Can't make heads or tails of it. On Apollo's credit business, he says, on our balance sheet, we are 95% investment grade, private and public investment grade. I have a view that bigger companies are going to do better than smaller companies. And so I've tried to position my—

The terminology. Gets me every time. Yeah, I know. Because they're like, you asked me, I run a private credit fund. We mostly back, we mostly invest in non-investment-grade opportunities. Yeah. It's like, brother, don't you want to be investing? What were you doing? It's in the name.

Now, of course, very, very fun end of this journal piece. They have a form. We want to hear from you. Are you currently an investor in private credit funds or are you planning to become one? We'd like to hear from you. Share your thoughts or experiences in the form below. They're looking for snitches.

Yeah, I'm going to gig along. Back to data center land.

Really quickly, let me tell everyone about Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. And let me also tell you about CrowdStrike. Your business is AI, their business is securing it. CrowdStrike secures AI and stops breaches.

Uh, got to give another shout out to George Kurtz, who went 1 and 2 again at the Chinese GP over the weekend. Uh, Mercedes on an absolute tear. It's the Kurtz effect.

I saw a— I saw some sort of promotional post for a vintage Le Mans racing series. So 24 hours, but there's some date where like all the cars have to be from early before 1990 or something like that. I don't, I don't exactly know how old.

I didn't dig into it, but it looked Very, very cool. Anyway. Cerebras just landed AWS. Amazon announced inference chips deal with Cerebras, which is big. They are proving the doubters wrong. Elon is saying that the TerraFab project launches in 7 days. Bezos says, what? Very, very fast timeline. Obviously, when people heard about his plans on Dworkash. A lot of people kind of questioned it, but Elon's used to being questioned.

Cerebras is such a cool company. Like, I just the first time, I mean, we've seen it with like the ChatGPT AI and, you know, just going to Codex Desktop, which is of course like a coding harness. But you can just ask it questions and you can experience Codex 5. I think 5.3 is on. Yeah, 5.3 Spark. Spark. Yeah, Spark. And it just gives you the answer immediately and it's actually very, very magical. And I think that's going to be really good for retention. Basically, everyone's going to be in the smiling curve, will smile more as people come back.

Matthew Zeitlin says data center capacity growth is slowing. He's pulling data here that says newly added US data center capacity slows down considerably in Q4 2025 as market struggles to keep up with explosive demand. 25 gigawatts of data center capacity added to the funnel in Q4, 50% less than Q3.

We'll see if that ramps up again or that still seems like a lot like the number that I was hearing was for this year. The target for Anthropic is like 5 gigawatts, which is like an insane amount of compute. But at the same time, like like in the context of 25 gigawatts in one quarter, like it feels like there is like still significant growth. But of course, you know, risks to all of this.

One Dozen over on X says they were right to take cigarette ads off TV. I would've smoked a pack a day if I saw this when I was 14. What is this video?

Let's pull it up. Is this a real ad? This cannot be a real ad. I think it's some sort of viral. I don't know if we're allowed to play this anymore. I think cigarette ads are banned. Is that actually— I'm so confused by this ad. I think that's Charlie Sheen, right? Is that the Arc de Triomphe in Paris?

Good music though. This should be the new launch video, Meta.

Oh, it was an international ad. The, the, the message there is that they're taking New York to France, but then it was Japanese text on screen. I don't know, it seems like some sort of, uh, some sort of mashup.

I don't know. Let's go over to Tyler Cowen. Yes, with how to lose the AI arms race. Lock in. Oh no, I was going to go to his "Why You Should Work Much Harder Right Now" over on Marginal Revolution. Okay. Before we get into the next piece, he says, if strong AI will lower the value of your human capital, your current wage is relatively high compared to your future wage. That is an argument for working harder now, at least if your current and pending pay can rise with greater effort. Not true for all jobs. If strong AI can at least potentially boost the value of your human capital, You should be investing in learning AI skills right now. No need to fall behind on something so important. You also might have the chance to use that money and buy into the proper capital and land assets. So work harder. He should have put this into a course. I would follow this advice if it cost me $999 in 6 installments. But because he's given away it for free, it can't actually be be that valuable. Kidding, of course. From Ricardo in the comments: Suppose you're the best maker of horse carriages in Belgium around the time the automobile is invented. You might want to take on as many orders as possible for new carriages because you know your future is precarious. Or maybe you get your hands on one of these newfangled automobiles as soon as possible and learn how to fix them. Both options require you to work harder, but these seem to be the two best options. Options available. Paradoxical but true.

That's a good take. I like that.

A little bit of a white pill. Never, never a bad idea to work harder. Never a bad idea. Should we go through this?

Yes, we should. First, let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco. And let me also tell you about Cognition. They are the makers of the AI software engineer Devin. Crush your backlog with your personal AI engineering Where do you want to go next, Jordy?

How to lose the AI arms race.

Let's do it. So investor Leopold Aschetbrenner is now famous for Situational Awareness. His essay predicted that major AI companies would end up functionally as part of the government-led national security project, possibly even nationalized. Along related lines, economist Noah Smith recently asked a critical question: if AI is a weapon, why don't we regulate it like one? We already know— this is Tyler writing in the Free Press, thefp.com— we already know that the Pentagon has been using Anthropic's Claude to interpret collected intelligence data and help plan the attacks on Iran. Advanced AI can also be used for cyberattacks, enemy surveillance and identification, followed by missile or drone attacks. Under most extreme scenarios, which may or may not be realistic, advanced AIs might design bioweapons, disable the nuclear weapons of an adversary by disrupting chains of command, or perhaps design and build a scheme to knock missiles out of the sky. Washington, D.C. is starting to ask very basic questions about where we are, what we are doing here. Anthropic and the Department of Defense are at loggerheads over whether Anthropic's AI should be banned from government work. Senator Bernie Sanders recently raised a broader set of concerns, calling for a moratorium on AI data centers with the intention of slowing down progress in artificial intelligence. Intelligence models. But circa 2026, neither nationalization nor an AI slowdown are feasible strategies for the United States. We need to keep our lead both in military and civilian uses of the tech, and that requires a dynamic private sector building our artificial intelligence models. Our federal government, working through the Manhattan Project, developed and built the first atomic bomb, but the strongest AI models are creatures of the private sector, whether we like that fact or not. Even China China, which is far more statist than the United States, has seen its cutting-edge models built by companies, not the government. The top AI models are far too complex and require too much high-paid talent, including, including international talent, to be done well by governments. Governments sometimes can succeed in building out massive hardware projects, with the space program being another example. They are very— but there are very few cases of governments succeeding with advanced software. On a large scale. For that, you need private sector dominance. There is no easy way to switch from that mode of organization, which includes salaries of tens of millions of dollars for top researchers, to a more bureaucratic approach. An attempt to do so would destroy or take down those companies, thus thwarting our standing in this new arms race. The general reality is this: we all benefit from living in an advanced civilization rather than eking subsistence as our ancestors did. But there is part of this bargain we have tended to ignore or take for granted. Now that human beings have developed advanced technologies, we, the freer and better societies, must commit to keeping the technological lead. We have not stayed ahead in every area of tech, but we need to be able to protect ourselves and our allies. It's a good thing that America built an atomic bomb before either Hitler or Stalin did. To the extent you believe AI AI is important for weaponry and national security, that means we need to keep up the pace of progress. You might find that a slightly unpleasant thought because even under positive visions of an AI future, it will change our world a good deal. Nevertheless, it is a part of technology, of a technological bargain we have been living with for a long time, arguably since the widespread deployment of firearms or explosives. We seem to have been lulled into state of stupor by the longstanding technological dominance of the United States after World War II. In essence, we have to fight and win yet another arms race. You can't blame AI for that. You can blame AI for that reality if you want, but the reemergence of competitive arms races was inevitable with or without AI. You should redirect your ire toward modern history itself. AI might, may have accelerated the world's new arms race, but there are many other technologies technologies that could play and yet may yet play a comparable role. Space weapons, anyone? How about lasers or new types of hypersonic missiles? At least with AI, the US currently holds the lead. The creativity behind top AI models plays into our national strengths. And he closes by saying, so today we need an odd complex, an odd and complex mix of not entirely consistent ideologies for the current arms race to go well. How about some tech tech accelerationism mixed with capitalism, and then a prudent technocratic approach to military procurement to make sure those advances serve national security ends. On the precautionary side, we need a dash of 1960s and '70s new left and libertarian antiwar ideologies skeptical of Uncle Sam himself. We do not want to become the bad guys. Do you think we can pull that off? The new American challenge is underway. I like this. There was a lot of back and forth around the Anthropic Department of War debate, and Dwarkesh had a great piece on it, and lots of people have chimed in now that the dust has settled a little bit. And I think this is a good nuanced take. It doesn't boil itself down to a tweet just yet, but I think we are getting somewhere with with the different trade-offs that are at stake.

What do you think, Tyler? Yeah, this is good. I mean, I think the whole thing that I basically got to when I wrote the nationalization thing was that there's just this pretty big scale of what actually means to nationalize something. There's the Manhattan Project, which is like, okay, this is full-scale, top-down, everything is decided by one person, it goes down the pyramid. And then there's the very distributed, Oh, like Intel, is that nationalization? Yeah. I think I broadly agree, like I don't think really, I don't think the Manhattan Project is really the best way to do this, right? Yeah. Because if you take like, you know, Tyler Cowen's whole thing is like, you know, state capacity libertarianism, like is the government like fully capable of continuing, you know, this AI progress that we have right now? Like would the US stay in the lead if the whole thing is like, you know, set by the government? Yeah. It's unclear.

This is sort of what I was going back and forth with Karp on, was like people have framed this as like a battle between Dario Amedei and Pete Hegseth. And I feel like we are a democracy and so like I would like more authority to be assigned to the individual American voter for a lot of these things. You know, you have that joke about like we got to talk about We gotta just talk about this. Like, what are we gonna do about AI? It's like, well, like, we can actually vote on it. Like, you can have a plan and then people can vote for it. And there are a bunch of different ways to exercise political will. And it feels like there is a trade-off, but we got to a good place with the nuclear weapons one. And I do feel like I, as a voter, I have a very small stake, one of, 300,000, you know, I guess 300 million. I guess there's like 160 million people that vote in the national election. But part of the national election is, you know, do you trust this particular person to have the nuclear football, to have their finger? They're going to have their finger on the button. Like, well, let that sit with you before you cast your ballot. And it will be a continuation of that. Like, this is the person that will decide AI policy. So vote according to that. And I hope that there's more of an understanding that the American voter, the American citizen does have a huge stake in the AI future. And it's not just the, like, you know, the high-flying personalities that give, you know, speeches and podcast appearances. There is a lot more to the American project than that. Well, there is a lot of news around Taiwan and what might happen over there. We found an interesting call sheet market that sort of tracks just general unrest in Taiwan. So the question is, will the United States issue a level 4 travel advisory for Taiwan? That, of course, would be very, very bad news if that did happen. It's sitting at 46% before 2028. January 1st, 51% before 2029, and 57% above for 2030. And so this is sort of a way to understand geopolitical risk. Obviously, we hope this calms down and this market goes to zero because we have—

you sent me that headline about increased activity around Taiwan. Yeah, some of it was fake news. Yeah, some of it. Yeah, I think the reason that it that it triggered—

Really calling me out here, sending you fake news. You're like, you actually fell for a viral hoax recently.

No, I mean, I looked at it and it was factually true. It was just that the activity had dropped enough. Okay. The increased activity looked like a really sharp growth, but it was just kind of normalizing.

Oh, okay. Okay. Interesting. Well, we are certainly hoping for smooth sailing in the Taiwan Strait. Let me tell you about Figma, no matter where your idea starts, Figma Make, Cloud Code, Codex, or Sketch, the Figma canvas is where ideas connect and products take shape. Build in the right direction with Figma. Uh, ASML.

Burn Hobart, funny post here. Yeah. ASML can't figure out how to make money from EUV machines, so they sell them to TSMC, but TSMC can't figure out how to make money from chips, so they sell them to Apple. Apple can't figure out a profitable way to use iPhones, so they sell them And there you go, the profit. And anyways, Dr. Kareem Carr is— Someone's saying— Bear posting.

Yes, bear posting that they don't know how to make money from AI directly.

This is really— This is such a funny criticism. It's such a funny criticism because if they were— if they actually were doing that, the criticism would be insane. It would be like they created superintelligence and they're keeping it to themselves.

Yeah, exactly.

Exactly. The whole point is that every single person on Earth, whether you pay for a plan or not, can benefit from today's models. Indeed.

Well, let's head over to Meta. But first, let me tell you about ElevenLabs. Build intelligent real-time conversational agents. Reimagine human technology interaction with ElevenLabs. So Nebbius and Meta have agreed to a $27 billion AI infrastructure pact. Deal. The talks are advanced to pact stage. 5-year deal, $27 billion to supply AI infrastructure capacity to Meta. Nebbius has really been on a tear. Fascinating company, formerly part of Yandex, spun out, independent now, publicly traded, and just one of the neoclouds that's figured out that Microsoft deal and now seems to be doing good work with Meta. So Nebbius said it will provide $12 billion of dedicated capacity across multiple locations. Meta will also purchase up to $15 billion in additional capacity over the 5-year, over the 5-year period. These deals are sort of squishy, but it doesn't matter because the people who actually need to know can underwrite them accordingly. Nebbius added that it will use large-scale deployments of Nvidia's next-generation Vera Rubin AI infrastructure, which Jensen is surely talking about at GTC right now. That's expected to be available in the second half of the year. And Nebbius will begin delivery of that capacity beginning early next year, which feels like a decade in AI timelines. Why do you have the paper in front of your face?

The team earlier said I look like a third base coach, so I'm covering up.

I'm covering up. Oh yeah. Because you don't want to let everyone know what play you're calling.

There you go. Exactly. There was news Friday, late, a rumor or some reporting from Reuters. Meta is planning sweeping layoffs that could affect 20% or more of the company. 3 sources familiar with the matter told Reuters as Meta seeks to offset AI infrastructure bets and prepare for greater efficiency brought by AI-assisted workers. How many employees does Meta have? I think it's like 60,000, something like that. Let's figure 75,000. June 30th, 2025. It will be the same number. Yeah. 78,000 as of December 31st. Somewhere in the same range as Salesforce. And again, not super surprising. Stock's up around 2% today. I would expect this to pop even harder once these layoffs are actually announced?

Yeah, I mean, the advice is, is, you know, become the— become aligned with the AI effort at Meta. Like, they're, they're, you know, if these layoffs happen, they're clearly cutting part of the workforce, but then they're also like acquiring and hiring all over the place, just more around AI. I mean, we saw that today with the Manus announcement.

They're taking new naming, Meta. Just call your product a computer. Indeed. Manus Computer.

We got— No, no, no, it's called My Computer. Manus. My Computer by Manus. My Computer by Manus. It works on mobile, works on your computer. Manus Desktop. But wait, again, this was— this was My Computer is the core feature of the new Manus Desktop app. It's your AI agent. Okay, so it's still called Manus. Direct competitor to Codex, Claude Code, Code Cowork and Microsoft Cowork. At this point, everyone's doing cowork, so maybe you just rip that.

Yeah, so the reason I thought the Manus acquisition was interesting at the time is people were positioning it as more of a talent acquisition. Like, these are great product builders that figured out how to grow products super quickly. I think at the time they sold, they were somewhere in the range of $100 to $200 million run rate. I was interested in it specifically because it seemed like Zuck was trying to take what they had built and actually just scale it, not just roll them into, you know, working on ads or whatever other products. So, Tyler, please download My Computer by Manus and play around with it and come back with a review.

So the top recommended action that they showcase here is is organizing thousands of unsorted photos. I'm not super into like organization for the sake of organization, but that does seem pretty useful. I was taking photos on an actual camera this weekend and had to transfer them from the camera to an iPad, then sort of scroll through them, favorite them, then share them over AirDrop, and there is a cool like agentic workflow, which is basically actually download the RAWs. Some of them were a little bit overexposed, some of them need a little bit of color grading, and if I could have a workflow where Manus or, you know, some desktop agent opens every photo individually in Photoshop and tweaks it and does it intelligently and crops it ever so slightly and is like thoughtful about it, like that would definitely speed up my life.

So Dodd says, would trust OpenGlaw more than Manus after the Meta acquisition with private data?

Yeah, well, the Manus branding, the Meta branding on this is so limited. I would be surprised if people sort of, you know, if this, if this goes broad, people wouldn't necessarily know that much. I wonder if they'll do the Oculus thing and you'll have to like log in with Facebook at some point.

You can log in with Facebook, you can already, but you can also log in with normal like email.

Yeah. I mean, MANAS before was, wasn't it a Chinese company? It was based in Singapore, but it was, you know, like, like rumored to be aligned with China.

And so, I mean, not rumored. They, they were building it in China. Okay. Okay. To Singapore because the optics were not good.

So, you know, as far as, as far as private data security goes, I think this is an upgrade. It certainly feels like it. Anyway, let me tell you about Phantom Cash. Fund your wallet without exchanges or middlemen and spend with the Phantom card. So the Oscars happened last night. Jordy, just to get you up to speed, the Oscars are an award show that are put on by the Motion Picture Academy of Arts and Sciences.

Yeah, I saw someone at the RampCap table got an award.

Yeah, yeah, yeah. Michael B. Jordan won best Best actor, and he won an Oscar for that. Best investor. Yeah, they should have a category for that. But Timothée Chalamet is getting taken to task in the Financial Times over his views on opera and ballet, of all things. The Financial Times writes, "It's quite sweet, really. So desperate are some people to get their knickers in a twist on the internet that in the face of a lull in the culture wars—" We have real wars now. The only thing they have found to get outraged about recently relates to a man saying nobody cares about ballet and opera anymore. The man I refer to is Timothée Chalamet, a talented young actor who stars in the multi-Oscar-nominated "Martyrs Supreme," which had a very unfortunate showing at the Oscars. I think they were nominated for 9 awards and they didn't win anything.

And so— And is your belief that it had to do with his were his comments disrespecting? [Speaker:MICHAEL_SEIBEL] No.

[Speaker:MALE_SPEAKER] Or it was just the people, the critics actually just said, "Hey, like, you know, fun movie." [Speaker:MICHAEL_SEIBEL] I think in every category, he was, Marty Supreme was up against like a Goliath. Like, it was, every fight was sort of a David and Goliath. And there were just no upsets because he was going up against Sinners in one battle after another, which were heavy favorites, I think, from the very beginning, before these comments were made. Timothée Chalamet was talking with a fellow actor, Matthew McConaughey, at a town hall event organized by CNN and Variety in February, but the comments actually just got clipped and went viral recently. It was a 2-week delay. The slicers over there got to step it up. He said, I don't want to be working in ballet or opera or things where it's like, hey, keep this thing alive even though, like, no one cares about this anymore. All respect to the ballet and opera people out there. And then he said distinctly disrespectfully, I just lost $0.14 in viewership. Damn, I just took shots for no reason. There is evidence of Chalamet showing having made similar comments before, such as on the Graham Norton Show in 2019 when he called opera a, quote, outdated art form. And at an event the same year where he was where he was worried that cinema would become like opera or ballet or something, kind of a dying art form or something. He also, as many people—as many of those who claim to feel so offended have pointed out, has close family connections to the world of classical dance. His mother, grandmother, and sister all danced with the New York City Ballet. Wow. And he has spoken out about growing up dreaming big backstage at the Koch Theatre in New York where the ballet performs. As someone who tried to pursue a career in pop music while my older sister— this is the writer in the Financial Times— my older sister pursued one in classical piano, I would wager that he has been honing this particular attack or perhaps defense line since adolescence. So his apparent instant regret, his slip, felt a bit disingenuous. Are you an opera fan, ballet fan? I like the opera. Yeah, me too.

Although I actually have not been to the LA Opera yet, so it's hard for me to—

And I just think like there's a world where, where, you know, the film and movie industry like does become like opera and ballet, but that's still like a beautiful thing with an amazing culture.

Well, you've seen this in LA where I think a lot of movies are now releasing only at these like kind of fancy theaters. Yeah, Chinese Theater has one. Chinese Theater. Yeah, these kind of things where it's like much more like kind of upstage. It's like a real event that you go to.

Yeah. And of course, it is like a technological disruption with social media. And there's a lot of other gyrations in the transition there.

But I'll tell you why I think this whole kerfuffle happened. And as someone who doesn't really follow Hollywood, doesn't follow film, doesn't follow Timothée Chalamet, etc., etc. I think what is happening is he came out with like this new, like, it's okay to pursue greatness. Yeah. On the path to greatness. Sure, sure. I'm trying to be the GOAT. I'm trying to, you know, like coming out with this kind of like bravado. Yeah. And if you do that and it's like me, me, me, me, me, me. Sure. I'm trying to be the greatest. Yeah. And then you start just randomly taking shots at another art form where other people are pursuing greatness. Sure. You just invite a lot of criticism. Yeah. Because I think like everyone's okay, I think, with somebody like, you know, being on their own personal pursuit of greatness. But if you're doing that while trying to tear down other art forms, yeah, you're just going to invite massive criticism.

Yeah. It does feel like he's sort of collapsing market cap and TAM of like, yes, the opera TAM and the ballet TAM is smaller than film, but it would be odd to be— Let's play the actual— Yeah, let's play this. People that are here that are younger than me, where people desire, are desiring things that are more patient and that pull you in. I just saw another article article that says Gen Z is a bigger movie-going audience than the millennial audience. You know, I feel like a fucking grandpa saying that. No, but point being, I think even like Frankenstein, which is like a hugely popular movie this year, I didn't think that pacing was extraordinarily fast or anything, but it pulled people in, you know. But it does take you having to wave a flag of, hey, this is a serious movie or something. And some people want to be entertained and quickly. I'm really right in the middle, Matthew, 'cause I admire people, and I've done it myself, to go on a talk show and go, "Hey, we gotta keep movie theaters alive. You know, we gotta keep this genre alive." And another part of me feels like if people wanna see it, like Barbie, like Oppenheimer, they're gonna go see it and go outta their way to be loud and proud about it. And I don't wanna be working in ballet or opera or, you know, things where it's like, "Hey, keep this thing alive even though it's like no one cares about this anymore." All respect to the ballet and opera people out there. I just lost $0.14 in viewership. But, um, crazy shot.

That's not a shot. I hear what you're saying.

Yeah. Yeah.

So, but, um, yeah, yeah, I don't know.

It's, it's, it's interesting. I was thinking about like, if, if like the creator of like GTA 5 like stood on stage and was just like, We are 10 times the size of the— Baseball. Baseball, but also the movie industry. The video gaming industry has been basically 10 times the size of the movie industry for—

You mean the movie theater business?

No, like Hollywood. Gross revenue. Yeah, totally. I'm almost positive.

Not 10 times the size if you count streaming platforms.

Yeah, maybe streaming. Includes TV shows. And then do you include mobile games or not? That's a big question. But the video game industry is definitely bigger.

Raghav in the Twitch chat from Deep says Nvidia CEO just said he sees $1 trillion in revenue through 2027.

That's a gong. Bring down the gong.

Bring down the mallet. Bring it down.

Let's go. Congratulations. And we have our next guest in the Restream waiting room. First, let me tell you about Sentry. Sentry shows developers what's broken and helps them fix it fast. That's why 150,000 organizations use it to keep their apps working. And we are joined by Kevin from Epic Gardening in the Restream waiting Kevin, how are you doing?

What's going on? What's up, brothers?

How you doing? Good to see you, brother. Thanks so much for taking the time to join the show. First up, we got to talk about that tank.

Yeah, what's in the tank?

What's in the tank?

We've been talking about— I'm breeding rare Costa Rican tree frogs in this tank.

No way! Yeah, they're endangered.

Okay. Oh, they're endangered. Okay, what else is, what else is special about, uh, and you're planning to release them in, in all 50 states once you have enough.

This is the goal.

This is the goal. We're always scaling over here.

Yeah, of course. I love it. Is it challenging? Like, how much of your time is devoted to that particular tank?

Almost none. Almost none. I just need to make sure that they're fed. Yeah, that's cool. What do they eat? Yeah, they eat crickets, which I'm breeding in that little tank right over there. You can see that. Yeah, we're breeding the different trophic levels over here for sure.

Okay. And then, and then do these frogs have Is it use in your garden? Is it purely just for fun?

No, I'm just branching out to flora or to fauna now, I guess. Yeah.

Here at Epic, you know? Okay, well, first time on the show, so I want to kick off with your backstory. I want to know about the decision to start making content. I feel like that's always an interesting origin story. Like, when did you think, okay, I need to make content?

Dude, I mean, I'm an internet OG. So I was on GeoCities. I was on AngelFire back in the day. I was on AngelFire too. Anime tutorials, you know? So I don't know what it is. I think genetically I'm designed to make content. But for Epic, it was really a calling card for— remember when you used to design WordPress websites back in the day, like when people actually paid for that service? I used the blog as like a calling card or a digital business card for like designing websites for local businesses. And then just kind of kept plugging along with it and adding different platforms.

And here we are today. Yeah. Well, what about the first YouTube video? Like, what was the backstory behind choosing to go to YouTube, choosing to go to video? It's a big lift for people if they're on Substack or they're a writer and they don't know how they're going to do in front of camera.

First YouTube video was 2013, so it was a long time ago. And ironically, back then, I mean, SEO and blogs were kind of the thing. Yeah. And so So for me, the first YouTube video, maybe it's the second YouTube video, you can see me using a screen recording app, reading a blog article, just literally reading the article. And with the hopes that people would watch that video and click the blog link and I would make money off of the advertising on the blog. So it was a completely backwards logic to today, of course. Yeah. But then obviously discovered YouTube is a far better platform, especially these days.

So what was the, what was the flow of traffic over time? Were you able to reroute blog viewers to YouTube or did the algorithm eventually kick in? Because you're pre-algo feed, right?

Yeah, yeah, I think so. Right. I mean, I think it was back then if you subscribe to a channel on YouTube, that subscription would just show up. Yeah. Which was a beautiful time. But no, I think every platform, as you expand every platform, you think like, okay, well, I can get someone from this one to that one. It tends not to work. You tend to have to just just play each platform for what it is. And so like YouTube became its own thing. Insta, all the other social media platforms have become their own thing now.

How do you think about like serializing content, creating through lines, like the initial formats? Like what was the actual development of the playbook that you ran on YouTube?

On YouTube, I think in the early days, 'cause remember like I'm 13 years old as a YouTuber, which is like, 2 YouTuber lifespans. I think YouTubers last about 6 years or so. And so back in the early days, it was just pure SEO, especially for a gardening channel. It's like, hey, how do you— how do I grow basil? How do I grow tomatoes? How do I prune tomatoes? Sure. These days, those videos have all been made either by me or someone else. Sure. And so we've had to come up with formats that, that work repeatedly over time. So for us, it's great. I mean, it's a very seasonal business. So in March, what to plant in March. In April, what to plant in April. Or, um, you know, in June, how to take care of your garden in June, that kind of thing. And then also coming up with like formats that are a little bit more high effort but tend to do better, like garden makeovers or garden tours where you actually have to go somewhere. Sure. But it's easy to kind of like bulk those into a week and produce them.

What did the journey look like of transitioning from media into actually making products yourself? Because that is an idea that that is, at least in the venture world, people talk about. It's just a very obvious transition. You just go from content to commerce. And yet there's actually so few creators who have made that transition well, actually created products that go on to have equity value. I mean, people bring up all the time, oh, you guys have this audience in— in tech, you should create software or various products for the audience. And our answer has always been, look, if we do that, we're competing against someone in our audience who is spending 100% of their time on that business. And— They could be a sponsor. Yeah, and they could be a sponsor. But more so, I don't want to compete with someone in our audience that gets to spend 100% of their time on something when we can only spend like 10% of their time on it. Like they're gonna smoke us. But I think in what you're doing, like very, very niched, very, very niched down. And maybe the companies that you're competing with are not like, they can't go out and get $100 million of funding necessarily right away. But talk about that transition and how it's evolved.

Yeah, yeah. I mean, I think up until 2019, EPIC was just media business and that's it. And it would be Google Ads, it'd be YouTube ads, and maybe some brand stuff here and there. And I think in 2019 we did out of just that pool a quarter million in revenue. Wow. And then that was the year I decided to do product. Yeah. And so the whole logic being I can't really control any of those three streams of income. Like if traffic goes down for one reason or another, all of those go down commensurately. Yeah. And so I thought, okay, well what, what can I sell? And the beauty of having content is that you kind of get like a pre-validation engine for, for what you might want to put out there. And so there was this raised bed that I had, it's just like a gar— a metal garden bed that had been sent to me. And I was like, this is the thing I get asked the most about, so I'll figure figure out how to sell it. I didn't even know who gave it to me initially. So I tracked down the manufacturers, Australian company, and I just kept emailing them every quarter. I was like, can I sell this? Can I sell this? They said no, no, no, no, no. They eventually said yes. I think I had $70 grand in the business bank account. I spent $40 on a shipping container. I knew nothing about e-com. So what I thought I would do is, this is the most crazy stupid e-com logic of all time, but what I thought I would do is bring it into the port of San Diego, which does not take containers. So that, that was already a no-go. Um, it goes into the port of Long Beach. I thought I was going to go up and get it, like me at the port, drive it, driving the container down.

I have a container here.

I'm just picking it up. Yeah. Just like hauling it down. And then I was looking into Costco self-storage to like rent that, unload the container and like get like some sort of satellite internet to print the orders. And I talked to a couple of friends and they were like, yeah, have you heard of a third-party logistics company? Just ship it there. And you know, it was just so stupid, but that's how little I knew at the time. And so. What happened is made the order, got it on the water, made an Instagram story and said, hey, all these beds you guys keep asking about, they're here now. I have 550 of them. They sold out in 2 days. Use that cash to buy another container. Sold out that out in 2 days. So like by the end of the year, I think we did quarter mil in just that. So the business doubled. And then of course, setting that up before the global pandemic was insane. So we went from 500 to like 2.8 million to 7.1 million the next year. And then raise a Series A. But yeah, I mean, immediately I was like, oh, this is obviously the actual revenue driver behind this business, at least, which I agree, like a lot of media businesses don't have that easy plugin.

Yep, totally. Yeah. What was the team like before and after this transition? Did you have to hire business people? How did you feel your role was changing? I mean, we've had Doug DeMuro on the show a few times and He was very happy to hire a CEO to sort of run Cars and Bids and go back into content mode, do podcasts, which grew a ton. But every creator has sort of a different journey as they evolve the business.

Yeah, it's so weird because I run into Doug all the time at the coffee shop down the street. And we share the same investors. But yeah, so up until 2021, at tail end is when I raised the Series A, it was me or contractors. So it was me, editor, a writer, and an assistant, and that was it. And we had did— we did about $7.5 million that year, mostly product sales at that point.

So it was like way— wait, wait, wait, wait. So, so you have 4 contractors, but all those contractors are on the content side, but mostly it was—

yeah, so I was doing all the commerce stuff.

So you have like— but you're single, you're single product at this point. You just made the best single product. I'm just gonna sell. I made— and you didn't have to develop help the— I'm sure you made changes to the product. I didn't make the product.

I mean, I think that's the biggest thing here is I did not make—

you're living the dropshipping dream. Yeah, like literally it was like that everybody gets sold and then it doesn't actually work.

Yeah, it was crazy level dropshipping, I guess you could say, except for— I mean, I, I own the inventory. I brought it in, I had a 3PL, like, so it wasn't true dropshipping. It's just that I didn't invent the product. It was a distributor relationship. Eventually, of course, we've started inventing products, and, you know, we scaled, I think, from from December '21 to December '22, from 4 people to about 90, because we used some of the funding to buy a seed company that had 60 people. So yeah, that was a pretty crazy transition.

And talk about that, the buy versus build decision on the seed side, because I'm sure you had opportunities to do both.

Right. So with seed, it's almost always going to be a buy because the infrastructure to actually acquire seed, we sell almost 800 varieties of seed. Vegetables, flowers, herbs. It's nearly impossible to scale that really quickly. You have to have like buyers relationships. The buy orders are out a couple of years. You need like pretty specific infrastructure to actually like germinate and test those seeds to pack them appropriately. I think there's like 3 or 4 companies maybe that sell the packing machines and they're all in like Germany. So some German guy will fly over and like fix a machine for you. So yeah, I mean, and plus, let alone like we bought the brand of seed that actually started gardening with back in the day. So there's like a heritage sort of story angle there that worked out really well.

Yeah. What about the, like your role shifting as you bring in those 60 new people? I imagine that they had a leadership team at a company of that scale. How are you interfacing with them? What does your role look like then?

Yeah, I mean, the first year or two was like all-out madness. It was like whatever I could do at any point in time. So like still be the face of the content and architect that. But, you know, hiring, scaling, all sorts of ops types of decisions. Now we have a president similar to Doug's setup, which is extremely, extremely helpful. He's ex-chief growth officer at GameStop back in those crazy days. So he's got some pretty, pretty wild stories. Yeah. And with the Seed brand, the founders wanted to leave. And so we had like this little holdover position for them. And she kind of coached our leader in and they were just ready to go. And we can always call on them if we need them. But we don't, we don't really anymore, yeah.

That's great. Talk to me about seeds as a particularly good e-commerce business. I imagine, like, when I think about the worst e-commerce brand, it would be like, I sell a gallon of water. It costs $20 to ship, and then people buy it for $1.

And it's a low margin.

Low margin. Seeds, it feels like great e-commerce product that maybe people just needed to be educated about. But was that your experience and what was it like actually scaling up?

Yeah, I mean, I think like those original products, the raised beds, like I didn't have to invent them, right? Which is great. But every, by every other metric, they're not a good e-com product. Totally. The lightest one is 20 pounds. The heaviest one is 60 pounds. And then you're also, you're charged on dimensional weight of the shipping as well. And at the time my 3PL was out of like Thousand Oaks. So I'm shipping from SoCal to the whole country. A 60-pound box, which is just terrible. The beauty of that time is that I was charging shipping, which is kind of unheard of these days, and I had no customer acquisition costs. My customer acquisition cost was actually negative because I was getting paid to make my YouTube videos, you know what I mean? And that's what was selling it. And so I remember back in those times, pre-funding, let's say, kind of like laughing at all the D2C e-com bros because I was like, you're running paid ads, like, you're such a clown. And now I'm like, okay, I understand the model a little better. But yeah, I mean, once we got the Seed brand, that's a primarily wholesale business. And so when we looked at it, I would say about 15-20% of the revenue was direct to consumer and they had not focused on it. And so we've tripled D2C just by saying we own the business. Basically, we haven't done like a crazy amount of improvements as far as like D2C goes. We just, we just actually paid attention to it and plugged it into content. But you're right. Yeah. The gross margin on Seeds is quite good relative to everything else in the marketing space.

Yeah. How do you think about the transition from— I mean, it sounds like you're actually doing the backwards transition. Most of like the D2C bros start online, and then eventually they realize that, okay, well, I found the efficient frontier of cocktail TV on Meta and Google. Now it's time to go into retail. And then the whole company needs to pivot. They need to hire retail salespeople. Are you going in one direction or both directions? I've always wondered about the retail side of the business.

Yeah, yeah. I mean, I think the logic of the seed brand logic to me, I think there needs to be like a first order logic of buying something and that needs to be true. And then the second orders can be like very beneficial and may or may not play out. For me, the logic was like what we just talked about. The seed margins are very good and it's actually the only item in gardening you literally need every year. Every other thing you technically could get away with not buying again, like a raised bed or something like that. Sure. And so there's a repeatable addition to our business that we now have. That's great. But yeah, I mean, the, the sort of like second-order thoughts of, of buying the seed brand was, can I introduce the raised beds, these seed trays that we developed, to the wholesale network? Yeah. Because that is very, very hard to build out. We're in 75% of all independent nurseries in the country, which it would be different if like we had Home Depot or Target or something and we could just say, take this line. Yeah. Instead, we have reps that can go out to like 5,000 stores and say, do you with this line, which if we can get penetration on like some of those harder goods, then that's a huge benefit that could play out for us.

I have this thesis about creators that do, that launch products is that they typically underrate the number of B2B buyers in their audiences. And you might not, you might think, okay, I'm selling a protein shake, I'll sell it to the consumer. But you might have like literally someone whose job is to buy the next protein shake who works at Target or Walmart and they might be familiar with you. Have you had any of those experiences? Has that been advantageous or, you know, is this a unique, unique industry?

It's, it's, it's actually really weird because like the advantages you get, let's say, in like pre-validating a new product you might launch by teasing it in content and sort of seeing early demand, you actually get that to some degree with the wholesale relationships. Like we're in, we're in about 1,300 Petcos now. And I would say the sole reason is because because the major buyer at Petco has just been an Epic fan for a long time. So we were warmed up. I don't have to go chase that down and prove it out. We're talking to Walmart for some stuff. Hopefully that comes to be, but it's a similar sort of way that relationship started too. So I think the content angle, if you can convert it, you have some interesting doors automatically open.

Do you think you have the most AI-proof business in the world? As we talk about, one, you're obviously not trying to sell like vertical software, but two, even the Citrini piece pointed out that there's a lot of AI-proof businesses where if demand gets destroyed because your buyer is no longer making $250K a year to do some email job, your business might be fine, but maybe there's less demand. But I feel like even in these AI doomsday scenarios, like, I probably, you know, if I, you know, lose my job, I still probably want some seeds to put them in the ground.

Oh, I saw that Anthropic piece that came out saying like which industries are the most vulnerable. And I saw groundskeeping at a near zero. Near zero, yeah. Which is, you know, gardening is just a, you know, a recreational version of groundskeeping. So I think we're fine.

Yeah. Yeah. Oh no, it's true. How are you using AI? How are people using AI? In gardens. Like, I can imagine taking a picture of something happening in your garden and just being like, how do I fix this? Like, a lot of, a lot of ways that, that it could be very useful.

And we have that. Yeah, we have that. So what we did is we launched this membership program that comes with commercial benefits. You get like 10% off the store, free shipping, free returns, which is great if you want to buy like a couple seeds here and there. And then we paired that with an educational sort of side because we have more or less the biggest gardening audience on any of the platforms. So we trained a model just on our own internal content and then like licensed databases of, let's say, plant facts or weather or something like that. And so if you ask it a question or send it a picture, it'll give you the answer that the closest answer you could get to what we would actually say, not just like what GPT or Claude might say. And then it'll kind of funnel you to live support if you want it. So you can get actual humans too. So it's kind of like a two-tier thing. And then we're just using it like along with anyone else that that's how you'd use it inside of a company for operations and stuff like that.

What about on the content side? Are you finding it useful for scripting or thumbnail development or prototyping or sort of like layout, anything there?

I think it's good, John. It's a game changer.

It's game changing. Yeah, I mean, I think it's like the way we try to use it for content is like your really good first draft that you would normally have to spend however long to script out. The beauty of gardening, I think, is it's so bespoke to a particular individual's approach or a particular geography that AI's not really crushing that right now, nor do I really want it to be. But it's really good for first drafting a lot of different things in content.

Yeah. Yeah. How do you recommend I fall in love with gardening? I grew up, my parents basically forced me to do a lot of weeding, a lot of mowing, a lot of just random stuff around our yard. I had bad allergies at the time, so I would come out of that and be like destroyed. And so I've had zero desire to get into gardening as an adult, but I feel like I just gotta find the right wedge products. So is it raspberries, tomatoes?

Look, I mean, if you pick the crop that you are the most excited to eat and cook with, then you grow that. So if it's tomatoes, I mean, I'll send the Technology Brothers a bed and some seeds, no problem, if it gets you in the garden.

Fantastic. I love it. Yeah. No, happy to support you. Just tell us what to get.

Yeah. How are you thinking about the interaction between the creator economy, YouTube content, and Hollywood? We've seen like, you know, MrBeast is all over Amazon Prime now. I could imagine you doing content content with more legacy institutions. What's your philosophy around those distribution channels?

So the two things we've done that are kind of tasting that world is we have a Samsung FAST channel now that we've licensed 200 hours, hopefully more soon. And then we just launched last week an 8-episode series on Home Depot's YouTube channel. So kind of like a co-produced series, not a show, like not on streaming, but That's coming around too, I think. I don't know. I mean, I think that if you're Jimmy and you can get a massive check to do something on Prime, like, why would you not? Right. But a lot of us on the smaller scale or like maybe industry-wide big, but not like global big. Yeah. The fact the fast channel deals are looking really good right now, the, the sort of shows, if you can brand even if it's just a YouTube series as a show versus just like a video or a series of videos. It seems to be pretty palatable to advertisers these days. That's good. Um, which is kind of interesting because like fundamentally it's just a list of videos. There's nothing really different about it. But if you call it a show like Michelle Khare Challenge Accepted, I don't know if you know her. Oh yeah. Yeah. She's great. That, that is very much like a show on YouTube for sure. And it plays really well for, for those types of networks.

Yeah. Yeah. And eventually you, you have the seasonal element that you were saying, like eventually you can be like, here's 100 hours of just April-focused gardening content. And that's like, That's super powerful because the content is evergreen. The plants and the earth and all these things aren't changing really in any sort of meaningful way year over year.

That's such a funny mind shift because I don't know if you've had this experience, but it feels like playlists on YouTube never really got what they deserve. Did you feel that way too? Yeah, yeah.

I mean, playlists, maybe back in the early days, you would crank through a let's play video game series or something like that. Just let the playlist run. Exactly. These days, I told the team actually, I was like, look at every playlist we have, prune them down, and then like bucket them into more conceptual shows rather than like, this is my grow tomatoes playlist. That's so smart. It's more like, you know, so that's what we're trying to do right now.

Yeah. Like with Doug, you'll see like, you know, car reviews or like listicles and like it's more the structure. But yeah, that like you could imagine, there's also the question of like, How set up is Hollywood to work with someone like you? Because even if they're like, yes, we want you to do a full season on HGTV, but we're going to need to pull you away from everything else for— and I think the Corridor Crew guys went through this a little bit, where the numbers just never matched up. They would get bigger, and then Hollywood would get more interested. But then the opportunity cost of taking 6 months off to do a real Hollywood movie is never there.

Dude, this happened to me. Yeah, this happened to me in the pandemic. So 2020, it was June 2020. I did a deal with Chip and Joanna Gaines' then burgeoning network Magnolia. I think they were taking over DIY at the time. Yeah. And it was supposed to be this transformation show. You go back, you have this beautiful sort of thing, and obviously the pandemic kind of hampered that. I had just bought a house, so I pitched this idea of I'll just build out this house and we'll show you and we'll go through that. So it was like 45 days straight of hardcore filming, like 10, 10+ hours a day trying to get this done because there's a skeleton crew. And 2020, of course, was the year— I think we started that year at $180K on YouTube. I ended that year at over $1 million plus another channel almost at $100K. And so if I take those 45 days and just calculate, let's say I was making, call it even just 15 more videos, it would have been not only more money straight up, but more sort of brand value to the business to just make the YouTube videos. And I think that's what all the creators are running into. Yeah. Yeah.

Have you— Have you ever gotten tempted to do any of the, like, selling the actual end product? There's been a number of venture-backed companies that are like, you know, trying to make the perfect strawberry or any of these kind of vertical farming things. I always wanted somebody to do one of those, like, one of those, like, like Butcher's Box style thing, but give me a live video feed from the ranch so you can actually, you know, if you have this, like, real-time, 24/7 idyllic ranch, and then you're able to— I don't know.

I'm low-TAM there, but I mean, look, it's hard enough shipping seeds around the world and shipping hard goods that don't expire. I can't imagine how hard it would be doing perishables.

I would never want to do it, honestly. Yeah. How are you thinking about product expansion? If you go to the nursery, there's so much there. There's certain things that you're you're operationally set up for, and there's other stuff that's maybe better content, you know, could be marketed, but might be an operational challenge.

How do you assess like new— Rome Hearts Epic Gardening Wheelbarrow.

I would do it. I would do it honestly. Yeah, sure. Why not? That'd be fun. That's awesome.

I mean, look, like for us, there's a lot of room to run and seed. There's like tens of thousands of stores we're not in. Just on our botanical interest line. We launched an Epic Gardening line, which is like a guaranteed-to-work sort of beginner's line. Maybe that goes through to big box because two-thirds of gardeners spend their first dollar at a big box. We're not in any of them. And the mission of the company is to help people grow anywhere they are. So if that's where they walk in, we want to be there in some way. So I think you can, you can run this business quite, quite a bit further just on seed alone. Sure. And then we're architecting the rest of the product strategy around that. So our second best our best-selling line of products is seed starting trays and equipment and lighting. And then from there, it's raised beds. I think we probably do something in soil or fertilizers next. But again, do we own a fertilizer and soil mixing facility? No. And do we just want to white label?

Probably not, right? Yeah, yeah, yeah. On fertilizer, is the fertilizer, broader global fertilizer crisis because of the strait, is that gonna trickle down to everyday gardeners or are they not? It doesn't really matter for them if the price even were to go 2x, they still don't need enough product.

It probably is fine for us. I don't know. I mean, I think it's way more of a problem for like industrial agriculture. I think for us we're probably fine.

Yeah. Yeah. That's great. Very cool.

Well, what about tools?

Linus Tech Tips has a screwdriver. I know. Yeah, I was just hanging with their CEO and I don't know, you might even see an LTT do an LTT Epic collab at some point.

That'd be great. Yeah, yeah, he's the king of collabs. Well, thank you so much for taking the time.

Great to finally meet you. Absolutely love— congrats on all the progress, everything that you're doing. And, uh, say hello to Doug DeMaria. This year I'll give, I'll give gardening another shot. I want to see it. Now that we have Kevin AI.

Yeah, I'll hook you guys up. Don't worry about it.

I got you. Fantastic. You're the man. Great to have you. Good one. Take care.

Let me tell you about Restream. One live stream, 30+ destinations. If you want to Restream, go to restream.com. And I believe we have our next guest already in the Restream waiting room. Paul Cunningham is the dog healer, and now he's in the TVPAN Ultra Room.

Paul, how are you doing? How's it going, Jess?

What's happening? It's going fantastic. Great. I don't know if it's early or late, but thank you.

Or is it, is it shockingly early?

The sun is just rising. Okay, there we go. Rising. Well, we appreciate you getting up early to come chat I guess. Why don't you take us through some of your backstory, your history? I feel like you have a very interesting career that led up to this moment, and then we'll go into the actual story and process of what went viral over the weekend.

Sure. I've been doing machine learning since about 2009. Went full-time about 2015. I ran the Sydney Machine Learning Meetup group here for 6-ish years. I worked with Sholto and Tristan on a robotic arm project. Oh, yeah. And yeah, right now I'm doing consulting.

Oh, yeah. The Sydney Mafia, I forgot. Yeah, that makes sense that you did it over there. Not, not, not, you weren't in the States for that project. That's fun. Yeah, correct. Yeah, that's great. Very cool. And so, yeah, take us through the story of— I actually lost this in the story. Like, when did you find out that your dog was suffering from cancer? What was the initial process? At what point did you leave the traditional veterinarian system?

Sure. So what actually happened, the pre-story was Rose had some, like, skin rashes appear on his skin. I took him to the vet, uh, and he misdiagnosed it for 3 times for about 11 months. So over a period of 11 months, took it to the vet, was misdiagnosed, and on the 3rd time it started bleeding. So I decided to have the, the tumors removed, and that's when it came back as cancer, unfortunately. Tried really hard to have additional surgery just to remove as much cancer as possible and to essentially try to look at the stem. And because it had been misdiagnosed for so long, one of the tumors that got so large that it wrapped around her leg and we weren't— there's just not enough skin to close it. So that's when I kind of realized we needed to try different options. Sure. And then tried to put her on chemotherapy. But none of the traditional stuff was essentially like stopping it. It was continuing to grow.

Okay. Okay. And then so when do you actually first go to AI tooling? Do you start at a very high level just sort of asking about dog cancer broadly? Like at what level did you come into the conversation with AI just understanding the capabilities?

Abilities? Um, well, I, I knew about AlphaFold from, um, the AlphaGo days, so it was the progress, the progression technology. And, um, I just decided to ChatGPT one day in November 2024, like, come up with a, come up with a plan on how we can, like, potentially make a drug to block this cancer. I didn't really know anything about cancer at the stage. I was just going through the process of trying to figure it all out.

Yeah. Yeah. And so what happens next? Who do you actually call to? Because at some point, you know, it is just text in a box in an app or a website. At what point do you need to go back into the real world to advance the next step? I imagine that ChatGPT at one point tells you like, okay, well, we'll need the DNA sequence and we can't get that just from a text box.

So where do you go next? Yeah, so correct. So the first actual piece of data we needed was the DNA sequencing. Yeah. And yeah, ChatGPT recommended to reach out to Professor Martin at Stelvio. It provided 3 other people, but it was like it gave all the reason that this is the reason why you should reach out to Professor Martin. That's remarkable. And through a mutual friend here in Sydney, I was connected to Professor Martin, and he was very receptive to just taking it on. Extremely receptive. Yeah.

And so at some point you walk me through, you know, for those who are familiar with 23andMe, it's a saliva swab. What's the actual process for getting a dog's DNA sequenced? And then what's the file type that comes back?

Do you just get a text file? So this is considerably more advanced than 23andMe. It is like We have— I have Rose's entire genome on my— on an external hard drive I bought. Wow. So the process to submit the DNA sequencing was quite cumbersome. So it was filling out spreadsheets and stuff to submit. But what came back 2 weeks later was 300 gigabytes of data. Wow.

Yeah. And had to push through that. Yeah. And so at this point, you're not just dragging that file into a consumer chat Bot at this point. You're starting to build custom pipelines, correct?

Correct. So, again, I use ChatGPT, I use Gemini, and I use Groq. Constantly switching between the two. And, yeah, built out the pipeline to essentially go through the steps of computational pipeline to get to mutationally to sort of see what's causing the cancer, the root cause.

And did you actually use AlphaFold? Is there like an open source package that you can download and run?

Yeah, yeah, we use AlphaFold 2. Okay. So from the literature and from also additional LLM sessions, I find out that there's a gene called CKIT that is one of the primary drivers for Rosalie's cancer. And what we, what we essentially did was take her healthy DNA. So we sequenced her healthy DNA and we sequenced her cancer DNA, compared them side by side, got like a genetic diff between the two. Yeah. And, and then focused in on like the secret gene, pulled it out, modeled it in AlphaFold. Okay. And And I used two different techniques to essentially look for drugs to, to try to block the cancer. One was genetic algorithms. So I ran genetic algorithms and we actually came up with a unique chemical compound that could block it. But the reason I didn't pursue that is because I actually talked to a chemist about having it made. But the problem with that is you have to like go through the the steps of, uh, you know, um, first doing it in like, uh, in, in a test tube, then moving to mouse models and moving on further. So that's too complicated. Yeah. And, um, yeah, uh, the, the other technique was docking. We docked a whole bunch of these chemical compounds called ligands to the AlphaFold, uh, 3D protein, 3D structure of CKIT, and, um, these mutated CKIT and essentially discovered a drug that was very, very strong at blocking it. But unfortunately, the drug is owned by a major US international company. I reached out to them for compassionate use and they politely declined, which is fair enough. But that was kind of the— there's a second part of AlphaFold we used late in the pipeline, but That is kind of the start of the journey. And around this, around this time was about June 2025 when I went through all of that. And it really took the wind out of my sails because I tried everything. I tried like to see if I could synthesize it. I tried to see if I could like get hold of a preexisting chemical. And yeah, one day I was walking Rosie down the street and I realized maybe I'm actually close to making a vaccine myself and got back on ChatGPT and like typed away and it said, yeah, you know, you're halfway there. You've already done the DNA sequencing.

These are the next steps you need to do. That's amazing. So, so, so back to the lab. You did mention we at this point. So I imagine you've looped in friends, colleagues, like who is around you on on this project at this point?

It's myself, and I run a small AI consulting firm here in Sydney. Yeah. Yeah. So just like— I kind of worked on it in part-time for about 2 hours every day.

Wow. Yeah. It's remarkable. So back to the lab, and they wind up finally making the drug?

Yeah. Process in and of itself. I went through and did the design of the vaccine construct and pushed— I literally emailed it over to the mRNA Institute at UNSW. Yeah, it was like half a page of text. And, and the, the major blocker was actually getting an ethics approval because you can't just like go and make an mRNA vaccine in your garage. Like, they don't let you do that in Australia. No. So I'd been notified that I had to create an ethics approval. And again, I spent— I don't know, that was 3 months of my life creating that. And it got to a point where we were actually going to have to modify the university's license with the government because the vaccine was going to be administered offsite. So the ethics approval would have only been approved in June. This year. So, uh, oh wow, uh, through a connection in, in America, in Seattle, I was connected to Professor Mary Maeda. I don't know how to pronounce his surname. And, uh, she, she is like the preeminent canine cancer person on planet Earth. She connected me to someone in, in, um, a professor in Queensland, which is a state that's about 1,000 kilometers— I'm not sure what that is in miles— uh, north of here. I was talking, chatting to her, and then I was just saying like, I'm having trouble with ethics approvals. And she said, oh, I actually have a— I have an ethics approval with the government for that specific type of novel immunotherapies. And, um, you know, I'm happy to take you under my wing. I just played it completely cool. I was like, oh yeah, cool. Cool.

Um, but actually, like, you're jumping up and down. That's amazing. Yeah, inside my head. Oh, that's so cool.

So, um, yeah, yeah. So, uh, once we got the green light to do that, uh, uh, it all sort of like lined up in, uh, in parallel. I drove Rose up to Queensland. Um, we did the induction phase of the vaccine, and then, uh, just sort of waited to see the results.

And so, yeah, cancer is like a long fight, but it seems like there's at least some really positive signals, something like a 50% decrease in the size of the tumors. Is that roughly correct? Like, how are you measuring progress these days?

Okay, there's been a lot of talk about that. So obviously the visual, the best trait is the, the reduction in the cancer size. Yeah. Um, we also took, uh, blood work, which is going to be published in a paper later this year. Um, and, uh, just continuing to visually monitor her, her tumors, essentially.

Yeah, that's great. So where do you think this goes next? Uh, obviously there's a lot of attention. Uh, some people are saying, oh, maybe you'll launch a startup around this concept or try and democratize biotech further? Do you want to just continue the story in some way, or is it back to work as usual?

I think like the process itself was way too hard, and I think there's room to make it much, much, much easier for not just people like me, but everyone. Yeah. Yeah. So I think there's definitely room. I even know I could probably do the pipeline now in maybe 4 to 6 weeks. Okay. And that's important because the faster you can do the pipeline, cancer is constantly mutating. So if you can run the pipeline fast, if you can outrun the speed of the mutation, you can essentially clamp it down.

Yeah. So talk about the long pole in the tent. I imagine that just sequencing DNA takes time. Actually producing a chemical or synthesizing the actual vaccine, producing the product takes time. It sounded like ethics waivers and the approvals also took time, but some of those can be shortened. Some of those are going to be harder to shorten. Where's the biggest opportunity that you see to speed up that cycle time?

The computational pipeline itself can be sped up. The sequencing can be sped up. Sequencing is getting better and better every 6 months. It's like on a double exponential. Whoa.

I had no idea.

Then we could probably do the vaccine itself manufacture faster. And then ethics as well. I think there's different— that's probably the biggest room for improvement right there, to be honest.

Like, yeah. I think for sure. And I think that's going to be a story that we'll see in a bunch of other use cases and categories where like the technology's advancing faster than the society and our legal system can even adapt. But what I just love about this story is, as amazing as it is, the role that ChatGPT and these other LLMs played in this process, it really is a story of your just like insane determination and agency agency and high effort over such a long period of time to, uh, to save your dog. And, uh, it's just incredibly admirable. I love it. And, uh, I hope that, uh, many, many more people hear about this story. I'm sure you've been, uh, I'm sure some like documentary crews and things like that have reached out, but it's really, really special.

Well, thank you so much for taking the time.

Yeah, it's great to meet you. Keep, keep us posted. Send us your progress when you, when you about the paper later this year, come back on. And we're sending our prayers to Rosie.

Yes. And if you can see it over there. Oh, there we go.

Amazing. Little air horn.

Air horn for Rose. Well, thank you so much for your time. Incredible stuff.

To come chat with us.

Yeah, great to meet you, Paul. Have a great rest of your day. Cheers, guys. We'll talk to you soon. Cheers. Bye-bye. Goodbye. Let me tell you about Shopify. Shopify is the commerce platform that grows with your business. And lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents. The US Army awards Anduril Industries a contract with a total value of as much as $20 billion to buy the defense tech startups, software, hardware, and services.

Is that good? Let's put this into context. It seems really good.

It seems really good. I actually, I don't know where the the revenues are, but I feel like this is a significant jump up. And folks in the comments are asking when IPO. It is still private, but people are getting excited. And of course, the team over there, most of which, almost all of them have been on the show, are doing a fantastic job. So congratulations to them. There's never been a better time in history to be a shrimp. Are you familiar with this? This, Tyler? Of course, yeah. So Anthropic's founders and employees are about to get a lot of cash. Anthropic is at over $330 billion in valuation. I think I saw rumors that secondaries were trading at almost twice that. Obviously the company's on an absolute tear, revenues up and to the right, uh, they're doing very well. Many of the employees and founders have pledged to give away huge amounts of that cash, but where's it going to go? And people are wondering about about what, you know, if it goes to a nonprofit, well, what is the nonprofit gonna do? And of course the joke is shrimp.

Yeah, I believe that there was some system where if you say that you're gonna like send, like give a lot to charity, your comp is actually much increased. It's like 1.5x or something.

Yeah, you could get a multiplier, which is cool. I do wonder, you know, the shrimp thing obviously is a joke, but it will be interesting to see where the nonprofits go. There's been a lot of talk about mosquito nets for a long time. There's been talks about previous tech booms have created nonprofit funding booms, a lot focused on health and wellness and development and all sorts of different stuff. I would like to see a nonprofit that builds data centers. I think that would be sort of, if I had excess money, that's what I would put it towards. But— A Dyson Sphere. Dyson Sphere, yes. An Anthropic, maybe an Anthropic that's structured like a hedge fund. And it can like trade— because the public good that it would be delivering is like stronger price signals to the market, and it would be creating more efficient markets.

More efficient markets is— I mean, everyone benefits.

It's a public good. Everyone benefits. And so maybe on like a microsecond or millisecond basis, that could be the job of the nonprofit, would be just to create more aligned price incentives. That could be good. Also just like buying companies out like private equity style, loading them up with debt. That could be another option for a nonprofit. LBOs? I think LBOs. I mean, there's an option there. I actually have no idea what you can do.

Many of the software LBOs from that 2018 to 2022 are already effectively going to be nonprofit endeavors.

I mean, there is a question about like, you know, a lot of the, you know, the ruthless business people, they'll say like, yeah, I make a lot of money, but I do I do it for the love of the game. I'm like, we're about to find out. Because do you go work for the nonprofit then that does LBOs? Do you go work for the nonprofit high-frequency trading firm? Put your money where your mouth is. Actually, put the no money where your mouth is if you've been saying that you do it for the money. The lack of money.

Put the lack of money where your mouth is. Anthropic is hiring a national security policy lead. OK. And Alexander McCoy says, LOL, no kidding.

It is time. For that. Although, why are they headquartered in the San Francisco Bay Area? You got to set up the DC office, folks, uh, if you're, if you're going, uh— they're like, DC comes to us. Yeah, uh, or, or do the Jensen thing and fly around, go, go all over the world, uh, doing— talking to world leaders.

That is the way to— sure, Zepatide will do $45 billion in global sales this year. That is so much higher than I thought. Global iPhone sales will likely be $230 billion this year. For perspective. Yes, peptides are popular, especially the rigorously tested FDA-approved kind. Uh, yeah, huge, huge numbers. Feel like, uh, pharma needed a win. So some of these pharma— big pharma, then big pharma, low-key had been getting kind of hosed. It had been. And biotech broadly, I mean, yeah, we've had biotech investors come on here and be like, I don't know why you'd invest in biotech. This is absolute garbage, but I'm doing it. They were doing it just for the love of the game, basically.

I mean, with biotech, like, there's a very, very, very straight line to helping people live healthier lives. Semaglutide sales, which is Ozempic and Wegovy, which you might be more familiar with than tirzepatide, which is sort of the next-gen semaglutide. And then there is Retta, which is the popular one in San Francisco. That is the third-gen. Per peptide for a bunch of things, but weight loss is the one that people know. Semaglutide is projected to remain high but potentially decline in 2026, with estimates for revenue hovering between $36 and $39 billion. So that is a huge, huge market. Those are AI lab numbers for revenue, and the margins are great too. These things, I mean, huge R&D budget. Huge R&D cost, but once the manufacturing plant is up and running, and the demand is there, of course, you have to move through insurance, there's a bunch of other dynamics, but what a remarkable business.

Someone over on Reddit that said, funny story about Retta. Basically, he injected himself with Retta True Tide, but he didn't take the cap off, Yes. I don't understand this.

So these peptides, they come in a plastic shell device that has a needle inside of it. And then you press the pen against your skin, I believe. And when you click it, it shoots the needle out, very small, and does the injection and then retreats into the device so that it can be thrown away. And it's sterile and it's single use. And so you're not like doing the bodybuilder thing with the needles and the bottles, right? I think that's generally how it works. And so you have to prime the device properly. You have to remove the cap. I guess the guy messed up. He thought he gave himself peptides. He did not.

But didn't stop him from losing 10 pounds. He lost 10 pounds.

He says over the week, since he took the fake peptides. He lost 10 pounds, got amazing sleep, woke up happy, zero pain in his feet and Achilles in the mornings for the first time in 9 months. Food appetite felt suppressed, but I was still able to eat anything. Life was good. Last night, on the 6th night, I figured since I had zero side effects and life was great, I'd take an extra click or two. Nothing crazy, just a tad more. This morning when doing so, I realized that I never took the cap off the needle of my upon my— upon injection. I have literally placebo-affected my way into feeling absolutely amazing. Who knows how real this story is, but the placebo effect has been studied a ton and it is very real. So Tyler, do you have something?

I was gonna say like 10 pounds in a week seems like super fast.

That seems higher than— yeah, than what Reda should do to you. I mean, it's crazy.

I fasted for a week one time and I didn't lose— I think it was like just under 10 pounds.

You ate nothing for a full week?

Yeah, like water and salt.

Yeah, just water and salt. Should do it again.

Zero calories. Should do it again and we can do a time lapse of you talking across the 5 days.

Yeah, see how it goes. Let me tell you about Vanta. Automate compliance and security. Vanta is the leading AI trust management platform. We have our next guest in the Restream waiting room. Tony from Sunday Robotics is here. We had to delay. Oh, let's start the Lambda lightning round. Play that. Thank you. You're overdue for a lightning round. We went long on Friday, but we have a lightning round today and we will start with Tony from Sunday Robotics. Welcome back to the show, Tony. How are you doing?

There he is. Very good.

It's awesome to be back. Thank you so much for coming back.

Congratulations.

Extremely busy. Extremely busy. Incredible progress. Take us through the progress. How are you framing like this, the most recent era of Sunday Robotics?

Yeah, I think the biggest announcement or commitment that we made is that that, hey, we're ending the era of demos. We're focusing on deployments now. And I think really what's behind it is that there are so many robotics projects that start as a demo and end as a demo. And that was unfortunate.

No, no, they start as a demo and they end as a YouTube video that goes viral.

Yeah. And I think that just from the how much progress we've made through the beginning of this year and all the accumulation of infrastructure and systems that we felt like we can deploy it to real homes this year. And that's the premise of the whole beta program that we talked about. And yeah, and we're just like, hey, that's our sole focus and we're just going to do it really, really well.

Okay. What types of tasks do you have line of sight to everyday consumers benefiting from with, I'm assuming, some level of supervision, but enough autonomy that they can be, I'm assuming, valuable? Are you going to sell them initially or are you just going to place them into homes? Like, how are you thinking about going post-market?

I'm looking for a robot that will ride a horse and pull me in a chariot behind it.

Yes. That's what I'm into. Yes. But what are you working on?

Yeah, 100%. So on the demo, on the beta program, we are actually going to document it. We're going to be very transparent. We're going to be autonomous as well. And the reason is that we have so much data and the robot will be generalizable. But I think at the same time, when it comes to tasks that we'll address, I think the fun part is it will not be surprising. If you look at all the things that you spend the most amount of time on, like the thing that you hate the most, we're talking about laundry, we're talking about dishes, we're talking about like organization, cleaning, these type of things. And we're just going to pick a focus and do it very well. And be able to provide value. So that is how we think about it. There will not be like this super surprising pick of tasks.

Yeah. What do you think about the opportunity in offices versus the home? Everyone's focused on the home, but I feel like an office is potentially like a less chaotic environment. People are generally more like not leaving a trail of clothing around or whatever. Like maybe there's more straightforward tasks. Tasks. Where do you see the divide?

Yeah, I think home to us is such a good, like a long-term goal that's to drive the AGI moment for physical intelligence, right? To get there because it's so diverse, so many tasks, so much like objects, things are moving. But I think as we approach it, there will be lots of— as we build up the capability of the robot, it starts to unlock other use cases, maybe it's in offices, maybe it's in hotels, maybe it's somewhere else, that we're actually very open-minded to that. And that's something that we are going to think a lot about this year.

What's going on on the data side? You said you have a lot of data now. Where is that coming from?

Yeah, so I think for folks who haven't read our website yet, we have this new way of doing data collection, which is building gloves that are mirroring the robot's hand. So instead of needing to deploy like thousands and thousands of robots, we just need to make all these gloves and people can wear them and collect data in their own homes. So this gives us really high quality data, but also really high diversity and quantity of data. So I think this year we're going to scale to like a few thousands of these people to be collecting data for us every day. And we're going to build a high quality and diverse dataset that will be kind of powering the foundational model.

That we're going to train. Is there a value in having a less transferable, less precise dataset with higher volume, maybe recorded through like a face camera, like a Meta Ray bands? I've seen some examples. I think it was in the LA Times today about people doing chores with basically a GoPro on their head and they're just recording what they're doing while they're doing chores. And it feels like maybe that's not the perfect data. But if you can transfer that data over to the gloves, and you can transfer that to the robot, maybe you get extra data. But what's your thought on the continuum of data quality?

Yeah, I think you're talking about egocentric cameras, right? Yeah. People strap a camera here to record their movements. I think if you think about the quality side, we're definitely compromising. For example, we do not have precise movements of how people use their hands. Context. We do not have force information, tactile information, these type of things. So just that data will not bring us all the way through. But at the same time, egocentric data and all the data we already have in the public domain on the internet is going to help the robots because you can learn the more general physics. You can learn some intuition around how rooms are arranged, all those common sense. So I think the eventual recipe will be a combination of those those video public data with our proprietary datasets. And the way we think about it is that we're going to use our data to bridge this bulk of knowledge that we can extract from the internet to be a deployable product that is actually useful. Kind of bridge the gap from demos to something real that's providing value.

How do you incentivize people to wear the gloves? We pay them. I'm not, I know, but like, is it, is it, is, are you paying them? Like, good, good, good answer. Um, no, but I, but I'm curious, like, are you, you're giving some of the gloves and saying like, hey, I want you to do like at least an hour of activity a day? Like, like, is it per task?

Like, what's the structure?

Yeah, because if they just, if they just put them on and then watch Netflix, I can't imagine this. That's valuable.

100%. I think we both need to give requirements on the quantity of data and the quality of data and everything else. But I think it's actually a really good part-time job to have that you can collect the data anywhere, like in your phone, and you can do it anytime. You can do it right super early in the morning. You can do super late at night, in between your shifts, whatever it is, we're going to be really happy about that. And you don't need to even leave your homes.

Yeah, that's cool. How are you feeling about simulated data? We've talked about the sim-to-real gap before, and there's always this— there's not enough variation in some Unreal Engine environment that you build a kinematic model in. But it feels like with generative AI, you should be able to— to sort of stochastically generate different variations, create better synthetic data. It feels like the LLM companies are doing very well with synthetic data generation in certain cases, the various rollouts. Like, how are you feeling about it? Do you think it's gonna be in the playbook this year, maybe for a few years and then not anymore, or it's something that would be valuable farther out and maintain from there.

How are you thinking about synthetic data? Definitely. I think we talk about world models a lot these days, right? Yeah. And they're like, hey, can we generate synthetic data out of the world models? How good is that? And I think there are two sides of it. One is training a world model can allow us to leverage even more compute and even more data, like all the internet, right? That can bring us a lot of knowledge. Without collecting any additional data. But at the same time, I think it is neither going to bridge the deployment gap, which means getting from 95% to 99.99%, nor is it going to bridge the last millimeter for a certain manipulation task because you're just certain the fidelity of the data is slightly worse. But what I see that is being a layer that lifts everyone. Everyone will become better because they now pre-train on all these synthetic data.

Yeah, that makes a lot of sense.

You raised any money lately?

Yeah, we actually did it mostly last December, but we raised a $165 million Series B led by CalTwo.

What was the valuation?

Just out of curiosity. $1.15 billion. Unicorn already. I love it. Well, great set of investors. Congratulations. We love Coatue over here and they know what they're doing.

So good luck. Send us a pair of gloves too. Yeah. You don't even need to pay us. We'll just—

we'll continue. How else will you know how to adjust a podcast mic 75 times over the course of 3 hours? You can't solve that.

You can't simulate that. I'm not afraid to be automated by Sunday. I invite it.

Challenge accepted. Have a great rest of your day. We'll talk to you soon, Tony. Good to see you, Tony. Goodbye. Cheers. Let me tell you about Lambda. We are, of course, in the Lambda lightning round, and Lambda is the superintelligence cloud building AI supercomputers for training and inference that scale from 1 GPU to hundreds of thousands. And without further ado, we have Drew from 8BC. He's a founding partner there.

He's in the virtual waiting room now.

What's going on? TV Penultimate. How you doing, Drew?

Hey, doing well. How are you guys doing? Doing great. Good to finally have you on the show.

Yeah, thanks for having me.

Talk about Quince. Let's get right into it. I feel like somehow this company came out of nowhere for me. I haven't purchased anything there yet, but it didn't come out of nowhere for you guys, given that you backed it at seed. So I would love to Yeah, understand how you initially found the company, somewhat of a contrarian move to back a company like this, given it's kind of not necessarily right in the sweet spot for Eight, if you look at the rest of the portfolio. And then there's a kind of a graveyard of companies in Silicon Valley that have like made different attempts at this opportunity generally.

Yeah, I mean, we've been involved in one of them that was actually instrumental in this. But we, uh, I met Sid a long time ago, the founder, um, and kind of basically like it was just incredible. He was running a business called Lollipops, which I'd never thought about before. It was basically a, uh, a luxury candy company. And the way he talked about the, about the business and the stuff he wanted to do there, I was just super impressed because I, I kind of maybe didn't think there was all that much when it comes to candy, you know, luxury candy. Uh, and so he kind of told me he was going to think about doing his next thing. So we spent months kind of just ripping on ideas. And, um, it was— this was 2018, and Wish.com, which, uh, we'd invested in from a, uh, one of our— actually our first fund, uh, was, was kind of a high flyer at the time. And he was very intrigued by it, interested in, in the business model and, and their use of technology, I frankly, and an ever-expanding category of things to sell. And so, he was studying it. And I'll be totally honest with you, I was trying to convince him to work on some defense tech or biomanufacturing or some other business.

I just wanted to stay in the equity. Yes, something more in the typical line of sight for you guys.

Exactly, which shows how stupid it is to try to be thematic about things.

And so, well, you guys were smart. You didn't let— you didn't let the— you just bet—

you backed the jockey. Exactly. So we had already agreed on what we wanted to do. And he came and he pitched us what was called Last Brand at the time. And the rest is history.

And then— and then how quickly did you realize he wasn't too crazy?

I mean, I think pretty early on the business was working. I think what's, you know, it's kind of like what— it's always tough to see like when an exponential curve kind of early on. And it was, um, and it really started to become obvious maybe, uh, just after COVID about how fast he was growing and really how unbelievable, uh, the operations were. I mean, the, the amount of products they were bringing in with really high quality and the amount of retention and repeat purchasing, it's just totally nuts. I mean, I think that at this point, Quince is now a top 10 retailer in the United States in terms of repeat ordering. And the other top 10 all have grocery or pharma that drives people in every month or whatever, which has been massive. And the company grew 100% year over year. Last year, cash generative at like a multibillion-dollar scale. So it's just, it's just really a testament to the way that Sid and the team he's built have been running the business.

Zooming out, how do you think about D2C retail changing in the AI era? Like we've been tracking the agent-to-commerce stuff pretty closely. It feels like it could go exponential this year and go from like 0.1% to 1% and it wouldn't necessarily move the needle, but like, how are you thinking about it as like the next decade?

Yeah, I mean, it's, it's something that Sid has been investing in a huge amount. And really, I think the way he describes it is just like he wants to build the world's most efficient supply chain. And so AI is used in like every, like literally every single aspect of the business. Sure. Obviously, there's some places you can have, I think, a much sort of maybe more— there's a lot more like of a call option of something being totally transformative, maybe more on the front end side. But the reality is, like, one of the ways that he's been able to just be so unbelievably capital efficient and cash generative is because their operating expenses are just incredibly low. And that their sales and marketing is super, super dialed in and just basically every dollar they spend in it is cash generative. So I think that AI will probably have the biggest impact just in terms of taking a business which maybe you thought, okay, the max ceiling here and what this thing could generate would be 20 or 30%, you know, operating margins and just drive that even higher. And there's obviously interesting stuff on the consumer front end too. But to me, that's what's most impressive, at least right now.

Just thinking about the broad startup landscape, are you more bullish than in previous years around consumer broadly? There's a lot of like, oh, B2B SaaS vertical stuff, it's going to get steamrolled, blah, blah, blah. It feels like D2C commerce, it sort of went through a wave. Wave, and then there's winners and losers. You're clearly in the winner. But how are you thinking about just startup opportunities these days?

I mean, I think it's actually interesting being based here in Austin because Austin maybe is like the— maybe probably the most— sort of has the highest density of successful CPG founders. And I think the interesting thing about them is that the vast majority have run super, super capital efficient. Maybe they've raised a little bit of money, I think the sort of traditional software—

Yeah, they're sort of forced to, they're really forced to be capital efficient. I'm sure a lot of them would love to just be growing, paying whatever it took to grow. But you look at the rounds that even great brands with great economics put together, it's like they're pulling in $500K from over here, a few million over there. And that's with being an 8-figure revenue business when you look at any other counterparty, even in defense or AI, if they had that level of revenue, they'd be raising at 50 to 100 times revenue potentially.

Yes. So I think AI offers the potential that, and we're talking more on the actual physical consumer product companies side, I think it offers the potential for higher margin structures. So if you find the right entrepreneur, I think there are going to be more businesses built, even if there's not some super obvious, like Big Wine Now, just because people who deploy this the best are going to be— they're going to have a superior cost structure, which means they can spend more on marketing, grow faster, invest more in their product quality. And then I think obviously on the software side, I think anytime you see one of these big tectonic shifts, there's going to be tons of businesses built on the consumer side. I think some of them are going to be tiny but insanely profitable. I already— as we all know, there's people who have of 1 or 2 people that have thrown together something making like $1 million or $2 million or $3 million a year, and they're running it basically for cash. But I think we're always looking for super high-quality founders that match well to the business they want to build. And at the earliest stage, I think can't be overly thematic. So if another you know, Sid walks in and wants to build something other than, other than Quince, I think I would back him into it at the early stage. But I think there will be, we will see more founders that are able to run these businesses at pretty, pretty insane margins because of AI.

What categories in physical AI or industrials or even defense do you feel like is still underinvested? Invested today? It feels like obviously there's great company. Sometimes you wake up and it feels like there's great companies in every category. But from your vantage point, where do you want to see more new company formation?

I mean, I think that we're kind of just still the first innings of just the— I mean, basically every huge technology wave right now is bottlenecked by the physical world. And I think there's just going to be an insane amount of companies that are built. Some of them will be incumbents that figure out how to use AI to lower their cost structure, change their incentive structures they have. I mean, and maybe those will be, you know, those won't be done by startups. But I just think that pretty much anyone looking to build a big company that is enabling the amount of energy, cement, steel, copper wiring, you know, etc., right now is, is super, super well positioned. If they can figure out how to run those businesses, they're very different than running a software business. And so, you know, I think, you know, kind of to the point of what we saw with Quince, I think a lot of people who were great software engineers and maybe would have built great SaaS companies wouldn't have necessarily been the right founder. I think that, like, when it comes to the reindustrialization that's happening in the United States, capital allocation is just this unbelievable— like, it's probably the most important sort of lever. And so if you're building billions of dollars worth of facilities a year and you don't have a relatively sophisticated finance function or a CEO who has studied this and comes from that world, I think you can be in trouble, which is kind of an interesting thing to see. For the first time, I've started seeing co-founding CFOs of companies. That's crazy. Or maybe they have a different title, but that's based on their background is like they're a partner in investment bank or or whatever.

And it's so funny because like a decade ago, if somebody comes in to pitch you and they've got this CFO as like a co-founder, you're like, hey, I think you have much more important problems to deal with than the finance function. You should probably get some revenue and happy customers first. But now I can see that flipping.

I mean, totally. It used to be like, my advice would be like, I don't want to see a CFO person around until you at least raise like $30 million in revenue, I want just like hire the smartest investment banking analyst you can find who'll work like 120 hours a week and like make sure you don't run out of cash, right? But now it's like, okay, well, we're gonna go raise a $500 million like, you know, project financing deal for this facility. It's a little different.

Yeah, that makes a ton of sense.

Awesome. Well, thank you so much for taking the time.

Yeah, thanks for coming on.

Austin, how's the weather? Weather's great. It's, uh, Probably great at least for another few weeks and then we'll get really hot. But it's good. Yeah.

Come visit now. Then it's time to go water skiing.

Hill and Valley next week.

Hill and Valley.

We'll have a bunch of folks out there. Fantastic. Yeah.

Looking forward to it. Looking forward to it.

Well, have a great rest of your day. We'll talk to you soon.

Great to meet you guys. Some Quint stuff. Let's go do it.

That sounds amazing. Thank you. Yeah. Well, that's it. Yeah. That's it. Bye. Let me tell you about AppLovin. Profitable advertising made easy with Axon.ai. Get access to over 1 billion daily active users and grow your business today. And without further ado, we have Karina Hong from Axiom in the TV studio. What's going on? Karina, welcome to the show. How are you doing? Hi, great to meet you. Great to meet you too. Great to meet you. Since it's your first time on the show, please introduce yourself and the company.

Hi, I'm Karina Hong, founder and CEO of Axiom Math. We are building mathematical superintelligence that will be a critical path to verified superintelligence. Amazing.

What's your background? How did you start this company?

Yeah, so I did my undergrad at MIT, math and physics, kind of did math Olympiads since I was a kid.

And we are—

Overnight success. 7 months old company. So we're downtown Palo Alto.

Very cool. And I mean, it feels like so many of the math benchmarks have been unsaturated? I don't know. What is the goal? How do you know that you're making progress when the— just the frontier models seem really, really good at math?

Yeah, so we combined very interesting techniques in post-training reasoning with formal verification. We use this language Kotlin, which is a program for proofs. And at the PanAm competition last December, which is the hardest undergraduate math test, we competed in real time. And we got a perfect score, 120 out of 120, where the best scoring LLM is 103. So by DeepSeq. So there's a lot to be done if you combine informal and formal approach, and that will have a really strong superhuman mathematician.

Okay, talk about why you think math is the pathway to general superintelligence.

Yeah, so we think that math is the sandbox for reality. You will be very quickly seeing verifiable rewards because in math there's absolute right or wrong. And especially when you have Lean, you can check the proof, the solution step by step. You'll be able to apply reinforcement learning in a much more efficient way. And we have currently scaled from winning Putnam perfect score to solving a batch of research problems that professional mathematicians find really challenging. And we also see this transfer to code verification.

Okay. On the last IMO, I believe everyone struggled with question 6. I believe OpenAI and Google both were unable to answer question 6. It was this sort of like tessellation of triangles question. I don't understand it. I didn't get it right. But do you feel like you're making progress? You came close. I didn't even try. You came close. But do you feel like you're— The read on that particular IMO question was that it required a lot of outside-the-box thinking. And that's why both many of the students who took the IMO struggled with it. And that's why also the model struggled with it. Do you feel like the progress that you're making will transfer to that type of math question?

Yeah, so there is this going joke that no AI could solve it because they were not at the Australian airport because the actual solution is the tiling of the floor. No AI was able to look down to the floor. They can't solve Problem 6. And we have seen that consistently since 2024. There were two problem. No AI was able to solve that. Stays the same in 2025. You know, still haven't seen an IMO perfect score. But in this year's Putnam, we have seen some really difficult questions by the Math Olympiad hardness scale by Evan Chen. There is a question much harder than any of the 5 questions on the IMO that Axiom Prover solved perfectly. And we don't believe any other AI has done so.

How are you thinking about the impact of advanced AI that can solve math? Is there, is there a direct benefit for just advancing the basic research that is done at a high level in the mathematics profession? Is it just once you're good at math, you can also go and, you know, write software that generates economic value. What are you most excited about?

Yeah, so let's just take the time machine back to 2024. Sure. I think everyone was kind of aware that Anthropic was working on coding, and people didn't really think much of it and thought of that as just one vertical in the enterprise AI applications. Turns out coding is a much more horizontal bet. And we believe the same thing with math too. We believe informal math gives us a pathway to verified AI, to be able to revolutionize how verification has been historically done in, say, hardware and software. We think the TAM is all AI code. And we think that in a way, we are looking at the ROFR, right, the first refusal, to verify or not all the AI code that will ever be produced. So this is a bet that's really relevant even if you assume AGI. I think to a lot of people, verification is about correcting mistakes, like erasing hallucination. To us, we see the upside. We think of verification as a way to have AI agents work with each other, human AI work with each other to compound and scale the brilliant, the super intelligence in a way just like Ramanujan after he learned proofwriting from Hardy and Littlewood came out to be a much more powerful mathematician, turn his intuition into theorems, and theorems have proofs. We think that we are seeing a similar thing happening here, and math reasoning will transfer to other parts similar to code and logic.

Talk about the impact of verifiability on mechanistic interpretability. The idea of like, you know, a lot of people have fear around AI. They think it's a black box. They don't understand it. Is this going to make it more of a black box because it's doing math at such a high level that no mathematician can understand it? Or does it make it more interpretable because you can verify what's happening once the AI generates its output?

Yeah, I think instead of being sort of like, you know, going to the machine interpretability, which some other great companies are doing, and we also work with them. That's about like the black box, what's going on inside. Verified AI is about being able to trust the output once it is generated. And so mathematicians will be able to function at a much higher abstraction than they ever, ever have been. So, I mean, if you ask me what is the purpose of like, you know, proof checkers, like the formal language Lean in the mathematical community, if they already have a pretty rigorous peer review process. All the mathematicians will just peer review each other's work. I would say that it's because the lean and all the tactics within, such as Grind, is able to cover all the low-level deduction and for mathematicians to focus on the high-level navigation and then do math at a much more compressed timeline. I think we're quite excited about the future where mathematicians, because of this sort increased supply of reasoning can produce way more breakthroughs than before. And that can then flow to other applied scientific domains.

That should be quite interesting. You raised a lot of money, $200 million. Congratulations. Are you— Thank you. Is there a large compute budget within your organization because of that?

Yeah, we are going to spend the new capital in compute and hiring. We are also very excited to continue the amazing team building progress we have been. I think I feel fortunate every day to work with this world-class team. And we want to, you know, have people who are interested in program verification to also join us.

Amazing. Well, congratulations. Thank you so much. We got to hit the gong for the $200 million.

Awesome. For Axiom. Wow, that was a, that was a big hit. Thank you. That's, that's—

it's a big round and it's a big idea. And this has been a great interview. Thank you so much for taking the time to chat with us. I'm sure we'll be back. We'll talk to you soon. Have a good rest of your day. Goodbye. Let me tell you about Gusto, the unified platform for payroll benefits and HR built to evolve with modern small and medium-sized businesses. I nailed it. I didn't say smother. I didn't say smother. I said modern small and medium-sized businesses. Our next guest is in the Restream waiting room. We have Cam Fink, the co-founder and CEO of Aru.

Welcome to the show.

What's going on? How you doing?

How you doing? Hey, thanks. Thank you guys so much for having me on. You know, it's been a dream since the day I was born to be a TBP.

Incredible. I know you're younger. You're the youngest ever guest. You're 11 years old.

Yes, yes. No, coming soon. But since it is your first time on the show, introduce yourself and the company.

Yeah, I'm Cam. I'm the co-founder and CEO of Aru. We're a business that predicts human behavior for almost every type of business on the globe. So we tell people who's going to win elections, what products you're going to purchase. We help people predict the outcome of marketing campaigns no matter who you are or what type of business you run.

Okay. Predicting elections is interesting because we just went through a huge prediction market boom. That financial instrument was one way to harness the wisdom of the crowd. You're sort of doing that through AI and data that you collect, or is it all simulated? Walk me through, like, how do you actually get to a better prediction than what the state of the art is?

Yeah, I mean, it all starts with our idea of rather than training off of what humans say they do or who they say they are, right? Like, we all know polls, focus groups, surveys are fundamentally wrong. There's survey bias, sampling bias, incentive bias, let alone the fact that people lie, right? We train on ground truth behavioral data only. So we're looking at things like credit card purchase history. We're looking at real marketing campaign click-through rates. We're looking at health insurance information, we're looking at real election results, that is all way more indicative of the actual decisions that people make and thus far likelier to predict elections better than anything else.

And so is that like if somebody buys at REI, they are more likely to vote a certain way and that factors in? Like how many different ways are you trying to triangulate? And then help me out in understanding how the actual platform works. Is this like effectively you have your own set of data and then you're spinning up a bunch of agents and you're basically prompting it to say, pretend you're this person and then this event happens, what is your response? How does it work? Explain it to me like I'm a podcaster.

I mean, 100%. When you start by asking, right, How is it working in terms of what sort of different data are we including? Your REI example, it's like that, but at massive scale. We can understand how the differences in the price of eggs in someone's zip code is going to change their likelihood to vote for one candidate or another or to care about some different marketing campaign or another. And then as far as it comes to an individual simulation, one simulation has been composed of tens of thousands of agents for any audience on the globe. So remember, because we're not constricted by what you can reach in a traditional survey and then going trying to train a model on top of survey responses, we can generate any audience we want, right? So we can generate maybe an audience of social media influencers. We can even generate an audience of podcasters. We have a client in the podcasting business, probably simulated you, John and Jordy, somewhere in there. But because of that, and then each agent gets given to a model that we build in-house, right? So it's our core model, our foundation. And then that model is able to take on that profile for all 10, 20, 50, 100,000 members of an audience and simulate their behavior more accurately than anything else on the globe.

How are you working to build sort of confidence within your customer base? This feels like the kind of thing that a lot of businesses would be down to try. But how do you actually prove accuracy over time?

The person that they predicted gets elected. Great question. I mean, that's a lot of what happened with prediction markets. It's like they got it right, and then everyone was like, OK, I guess it works.

Well, they get it right most of the time, not all the time. Take the Virginia Attorney General's election as a good example. But as you look at us, we've been around for 718 days. It's almost 2 years of ARU. If we didn't work, then we wouldn't exist as a business anymore if we weren't able to predict behavior accurately. But let alone that, we do have tons of external validation. We did a really good study with EY on a really tough-to-reach audience, like 3,500 individuals who are ultra high net worth, right? Can't imagine anyone with a $30 million net worth or more taking a survey. And we were able to recreate their behavior even more accurately than the survey, which was pretty cool. So great example there.

How concentrated is the customer base? Because I feel like with prediction markets, it was just like, you know, people had the page bookmarked and they were refreshing it going into the election. Very like general consumer. It feels like there's like a lot of customers maybe. Of course there's like whales, but with you, I imagine that you can walk into like the CEO of Coca-Cola's office and say like, I can move the needle for you. And that's like a big ticket client. What's the shape of the customer base right now and where do you want it to evolve? Yeah, I mean, behavior is everything, right?

So when we talk about predicting behavior, we can sell to like every type of business on the globe. We have film studio clients, we have We have podcast company clients, we have CPG clients, we have utilities businesses, and we sell to governments, right? So it's super widespread. But I would say the 3 biggest areas today are consumer, you know, whether that be retail, technology, or CPG. And then we do a lot of work as well for financial services businesses. And then I would add on top of that kind of like the government policy use case. We do a lot of work like simulating the impact of new tax changes.

What about like a self-serve, like, you know, for small business that might want to put down like $100 a month for a service on a credit card? Is that an option, or do you think that will be an option, or do you want to stay in like enterprise-y, like let's actually—

They want to help big business get even bigger.

I mean, I'm not going to be upset about that. Sorry, little guys.

I'm not saying we reject small business forever.

We work with plenty of really cool businesses across the full size, just like we work We work with the massive CPGs.

We work with Spindrift as well. But, you know, in terms of what our core customer base is, look, I think every business on the globe is going to want to use ours someday, right? Like, in 5 years from now, there shouldn't be a decision you're making without using our software. And I would like that to be as accessible as possible. I think it's just a question of making sure that people are primed to use technology as powerful as this.

Is this company somewhat, and I'm going to give you ample time to push back, but somewhat short AGI, like I assume a sufficiently advanced model from a frontier lab in the future, I could talk to it and say, hey, I need to make a decision on this product launch. This is my customer base. I can feed it some data, predict the outcome for me based on all the data that you have access to. I imagine you guys, if you just work harder on collecting the right kind of data, could always have a differentiated data source. But how do you imagine kind of competing with other— you're an intelligence— I view you as like an intelligence provider, right? Like you guys are predicting, but more narrow than some of the more general.

I've been running every life decision through GPT-2, and people say that my behavior is really chaotic, but it's been working out so far.

Well, it's actually funny you mentioned that because what we've noticed is the foundation models over time, they actually get worse and worse at predicting behavior, right? Like this is something we've seen. We used to be a business where we just give like, you know, survey data to an LLM and then tell an LLM, try and predict the behavior based off of this past survey data. But what you notice is it just doesn't reach the edges, right? Like really consistently, it is not going to be able to predict things at the margin. And that's a big issue because it's the predictions at the margin that are the most valuable.

Table, right?

Those are the predictions of like, what are Fortune 500 CEOs going to do that we can nail that an LLM can't? And so sure, you know, there is a future where like Claude is going to be able to tell you, yeah, I suspect that American household purchase decision makers might buy this product. But for the biggest decisions on the globe, why would you risk it? And that's why people trust us for the toughest decisions they have. Makes sense.

Very cool.

Well, thank you so much. We have a— I think we have— there's a gong.

Is there an official round announcement?

Announcement? Uh, I'll say we're very well capitalized. Very well capitalized.

We need a soundboard cue for that. Well capitalized. That'd be fantastic.

Thank you guys so much for having me on. I'm sure you'll be back on, and, uh, congrats to the whole team. Yeah, the progress. I look forward to, uh, uh, we don't— as well, we don't really—

when you want to send for TDPN, we're here anytime.

We'll figure something out. Yep.

We'll talk to you soon. Good to see you.

Have a good one. Cheers. Let me tell you about Plaid. Plaid powers the apps you use to spend, save, borrow, and invest, securely connecting bank accounts to move money, fight fraud, and improve lending now with AI. We have our next guest already in the Restream waiting room, so we will bring in Deborah from—

What's going on? Great to meet you, Deborah.

Great to have you on the show. Thank you so much for taking the time to join. Us. First, since it is your first time on the show, I'd love for a brief introduction on yourself, and then I want to get into the Oscars. But tell us a little bit about yourself.

Well, thank you for asking. I'm a longtime journalist. I've been covering entertainment for a long time. Let's just say a lot of years. Great. You cut me open, you can count the rings.

Let's put it that way. Fantastic. And just give me your overall reaction to the Oscars this year. What took you by surprise? Impressed you? Uh, what was your personal highlight? Uh, walk me through what you thought the, the story of the night was.

Um, the story of the night has to be one battle after another. I mean, I think it was definitely the film to beat going into the night. I don't think it definitely, you know, I don't think it counts as a surprise. Yeah. But I think it was great to see that film definitely take away all the wins that it did. I think the winner of the night was definitely Warner Brothers, you know, given all the coverage that that studio has been getting. It's sort of ironic that it was is the studio that walked away with the biggest wins of the night. I think Conan did a great job as the host. It was exciting to see that. So I think there was a great story. But to me, I loved Michael B. Jordan's win. I think it was historic. It was wonderful. I think he was so emotional about it. And that's kind of what you want out of the Oscars.

Yeah. How are you thinking about— I mean, one of Conan's funniest bits, I thought, was his jokes about the Oscars moving to YouTube and that you'd be interrupted by some sort of sloppy ad. Very tongue-in-cheek, but what do you think might change about the Oscars as they move to a more internet-native model?

I know, I thought he did a great job. He was really so self-aware about it. I thought, you know, like, I think one of the most awkward moments of the night with the speeches getting cut off. It's always uncomfortable. It's really hard to watch. You feel so bad for people. It's the moment of their lives. Yeah. And suddenly they're just sort of jumping up and down down on the stage going, why don't I get to say thank you?

That's a really great point. I mean, one of the amazing things about the internet, we've been tracking Apple's work with F1. And with the new F1 broadcast, you can say, I want to watch just this car on just this feed, or I want to watch these 3 teams, and I want this announcer. And I could imagine in the future a feed where you say, you know what, I actually don't care about the intros. I want to hear the race. The acceptance speeches, give those to me in full. I don't care if I'm here for 5 hours. And you could just let the person continue and then cut away on the other feed. And maybe that comes. It feels like the first version might just be the standard Oscar program on YouTube, but certainly some silver lining there if that happens. What, what, what, what, how would you characterize the AI this year at the Oscars? It feels like there was a lot of demand for statements and people to share their opinions about where things are going. Yeah, well, there are some comments. Yeah. At the same time, I mean, artificial intelligence in the machine learning context has been recommending people what to watch on Netflix for 2 decades. And so how are— how is Hollywood grappling with the AI issue at such a big event like the Oscars?

The answer is they're not. Okay. I think it's a raw nerve. I mean, I think no one is willing to admit, to your point, just how much it factors into it. Yeah. Um, I don't know that I can talk about what this season of The Comeback is about, but I think everyone is definitely addressing it, let's put it that way. And I think, um, I think it's definitely something that's on the forefront in everyone's mind. We all use it, and to your point, we use it in ways that we're not even aware that we're using it. We're using it unconsciously. And I think everyone is very sensitive about it, and we're seeing the guilds— God forbid there's another strike. Please, God, I hope there is no strikes. Everyone's doubling But I think we have to get ahead of it and come to terms with the real ways that it's helping us, but also getting ahead of— no one wants to see AI writing a script. No one wants to see AI making movies or making creative decisions. But we can also recognize there are a lot of ways that it can help us and make our lives better. So how do we find that happy medium?

I remember when Avengers— it was probably Infinity War, maybe Endgame— won best visual effects. And in the the CGI that went into Thanos's chin, they used AI to transfer the data from the facial capture of Josh Brolin. They had a camera pointed at him with all the dots, but they needed to be higher resolution. They used AI to actually up-res that data to make a more compelling character, which was Thanos. And it was a beautiful synergy between the VFX shop that needed to do more and better graphics and Josh Brolin, who still delivered a great performance. So, hopefully, there can be more storytelling there, but it is such an ambiguous time.

For sure.

And it really, like, it came up last season with The Brutalist, you know, and it's a really good question of how much it really hurt The Brutalist campaign. But it's like suddenly you had Adrien Brody talking Hungarian, and there was a controversy about how much ADR came into impacting all of that and how much it ultimately hurt his campaign and all of that. But it's sort of like, let's all be aware of how much it really has to do and how much it actually helps helps the making of the films. And if it's going to help films get made, and it's not really impacting the acting and the performance, is it really that difficult?

And is it really that painful? Yeah. And harmful?

The other thing is, even if you mentioned on the script side, how many movies have been created that in hindsight you're like, oh, yeah, AI could have made that exact plot.

Could have punched that up or found that plot hole, maybe.

Look, I'm not defending AI writing. I'm not saying that. I'm just saying— I mean, if it can help, to your point, the VFX, if it can help, and if it helped the film get into theaters earlier in the technical aspects of it, maybe there's some sort of happy medium to be found there.

Stuff like formatting a script, I mean, it's such a hassle sometimes, and there's little things about—

What do you think the big, like, goals from, for the guilds will be around AI? Do you have any sense of, like, what their asks are, what they're pushing for?

I think it's about protecting the writers' rights for sure. I mean, I think it's really making sure that AI does not come in and write scripts and that the writers' rights, which is really hard to say, but you know what I mean, that the writers' rights are protected. It's probably smart of them to have done, as painful it is to talk about another negotiation so soon after we just had one of those, because all of this is changing so quickly and that this just happened that we're making sure that we're staying ahead of it. So I think, I think knowing how quickly all of this is evolving and how quickly these conversations are happening, making sure that this is a thing that they're ahead of and that they're not gonna, that suddenly some new technology isn't gonna emerge that they haven't thought about.

Yeah, yeah. Have you been tracking the debate over dialogue legibility or how hard it is to hear dialogue in Hollywood movies these days? I heard that there's a, there's an interesting loop from as TVs got cheaper, a lot of the speakers went out the back towards the wall. And so the sound quality got worse. So Hollywood sort of had to adjust. And I'm wondering about your thoughts on how Hollywood is changing as we move to a culture that consumes movies on their phones, at home, and less in the theater. Or in your case, on the Apple Vision Pro. I did watch 2 movies last weekend in VR fully. I'm probably the only person. I think it's amazing. I actually, we talked to James Cameron about this on the show and he was sort of, it was very clear that he had gotten access to the next VR headset but wasn't able to talk about it yet. And I think from his perspective, perspective, you know, his movies, the Avatar franchise, it's so visually rich that being able to deliver something that instead of a 55-inch TV that maybe is from Costco and is tuned wrong, he can have more control over the actual visual experience. It was something that he was cautiously optimistic about, I think. And at least in terms of the really odd silver lining for me in VR It sounds very like anti-movie theater, but it puts you in a virtual theater where you actually can't use your phone. And so that whole Netflix thing about they have to restate the plot 7 times, that's not an issue. And so I watched Citizen Kane from start to finish, no breaks. I didn't, you know, that's a movie that would challenge the most brain-rotted of the younger generation., and I enjoyed it. It was great, and it felt like the movie that I should have gone to the theater to see, and I was able to do that in the Apple Vision Pro. So I've been having a good experience, but I don't know, VR is probably not in the conversation at all in Hollywood right now, is it?

Not so much, but I do think things that enhance the theatrical experience, anything that can get butts in seats in theaters, is definitely going to move the needle and it's going to be top of mind because I think that really is what is very much on top of mind for people and is really concerning the studios. And writers and guilds and actors and all of that, because I think that's what's really the biggest concern right now, because there's nothing compared to the theatrical experience. There's nothing compared to, you know, seeing one battle after another on a big screen. And I know I keep going that, you know, that chase scene.

Yeah, that chase scene is incredible. There's nothing like that. Yeah. It's incredible. Yeah, I wonder how the— you know, obviously, the Netflix-Warner Brothers deal didn't pan out. But one of the interesting— case studies that I heard was about how K-pop Demon Hunters sort of got a second run in the theater once it had gone basically viral online. Then there was a sing-along version, and then it became this experience where even like probably 90% of the audience that saw K-pop Demon Hunters in theaters had seen it before. But the kids loved it and the parents had seen it, and everyone agreed, this is a great movie, let's go and see this experience. I wonder if that could be something that the theaters lean into in sort of bringing back the movies that have already been de-risked. There's already this audience, but it's the spectacle, and you know that the tickets will sell.

But who knows? Nobody thinks that communal experience— there's nothing like sitting in a theater with an audience and experiencing it together. And on the flip side, you know, I'm a ride-or-die Hamnet fan, and sitting in the audience and riding that emotional wave of that movie with people in the audience, and not just sitting on my couch crying alone, but crying with people next to me, and someone turning to me and going, Oh, okay. That was really, it was a visceral experience and there's nothing that could compare to it. So, you know, I think to your part about K-pop, Damien Hunter, putting concert films, Taylor Swift saw that, putting butts in seats where people can experience something like that together, that's what theaters are all about.

Yeah, I had a similar but much dumber experience with the sequel to the Planet of the Apes movie. I saw it in what's called 4DX, which is, is a 3D movie, but then your seat moves left and right, and there's water that sprays you when something happens on the screen that has water. There's smells that are piped in. And at the end of the movie, there's this crazy avalanche, and they all survive and stuff. And we were high-fiving with the people next to us, and it just created this wild, hilarious experience that I still remember to this day.

I'm there for it.

I'm all there for it. Bring it on. I think whatever fits the right experience experience. For certain things, it'll be just a group of friends. For others, it'll be the full-tilt 4DX experience, or maybe VR. Who knows?

And I can't wait to see what Oscar category they come up with for that.

Yeah.

I don't know if it would be a category. Yeah, any predictions on AI-specific categories in the next few years?

Do you think we could see that? Ooh, that's a third round.

Yeah, I know. Yeah, I think about it.

It took them 25 years to add the casting category. You know, we just saw that one last night and next year we're getting stunts. So the wheels of change move slowly.

They come, but they move slowly. Yeah, that's interesting. Stunts being added at the time where I feel like AI will be most, as it rolls out, potentially more disruptive to stunts than anything else because it just— why risk human life?

Yeah, insurance and stuff. So yeah, interesting. Well, I mean, it'll certainly— I think just like cinematography, costume design, set design, there's so many things where when you're at the level of the Oscars, you are going to see the people— like, you think Tom Cruise is going to stop doing his stunts? No way. He's not going to let AI jump across a building. No, he's not. He's going to be jumping off planes. He's going to do it. And it's about the lore that he brings to that performance that part of when you sit down to a Tom Cruise movie is you've been hearing about the process of making that film for months and seen behind-the-scenes stuff of him jumping the motorcycle off the cliff. And it's real, but you can see some of the camera equipment. And so you go into it, and it's so much easier to suspend— you're almost not suspending reality. You know it's real. And so it just makes the excitement so much more thrilling. Well, thank you so much for taking the time to come chat with us.

Great to meet you. Great to meet you. Come back on anytime you're writing something you think our world would be interested in.

Yeah, we'll talk to you soon.

I'd love to come back. Take care, guys.

Have a great rest of your day.

Back to the timeline.

What did we miss? Well, John Collison asked on the cheeky pint Podcast, if Sierra is a short AGI company. That was a very funny one. I think everyone was wondering about this. And, you know, we'll have to watch the full episode to get the full answer. But there is this interesting dynamic right now. VC investments usually take 5 to 8 years to exit. This is from Ethan Mollick. That means almost every AI VC investment right now is essentially a bet against the vision of Anthropic, OpenAI, and Gemini have laid out. And so it's got to be a great time to be a VC if you're in those names, a little bit stressful, and it just requires like an extra layer of attention because you're not going up against legacy incumbents that have been rolled up and sold and gone public and gone private. That was so much of the early software boom was, okay, this company has been doing things in paper and we're gonna do it on a website. You're going up against Sam Altman and a lot of people that are in founder mode and very well capitalized and have a very broad vision and the technology shows a lot of promise.

So we will end here. Marc Andreessen says, "It is 100% true that great men and women of the past were not sitting around moaning about their feelings. I regret nothing." So we are gonna leave it there for today. I hope you spend the rest of your day Not remoting feelings.

Living with regret. Yes. Never introspect, always outro.

Always respect yourself. Here's our outro-spection. That's right. Uh, we made the new outro last week, and unfortunately we used a song they didn't want us to— they didn't want us to play. Yeah.

And so they took down— we're gonna work on a new outro. We're— we already got a bunch of ideas cooking. But leave us 5 stars on Apple Podcasts and Spotify. Subscribe to the newsletter at TVPN. Goodbye!