TBPN - AI’s Power Problem, Apple Goes Meta on AI Glasses | Pat Gelsinger, Josh Isner, Sheel Mohnot, Santiago Nestares, Austin Federa

Episode Date: October 2, 2025

(01:18) - AI’s Power Problem (22:04) - Andrej Karpathy AI Models Law of Diminishing Returns (35:40) - 𝕏 Timeline Reactions (56:18) - Apple Goes Meta on AI Glasses (01:01:31) - Josh Isner, President of Axon, has been with the company since 2009, starting in the Leadership Development Program and advancing through various roles to his current position. In the conversation, he discusses Axon's evolution from a hardware-focused company to a leader in software solutions for public safety, highlighting innovations like the Draft One AI product that generates initial police reports from body camera transcripts, significantly reducing officers' administrative workload. He also touches on the strategic acquisition of Prepared AI to enhance emergency response times and emphasizes the importance of open APIs to ensure interoperability with other public safety technologies. (01:19:31) - 𝕏 Timeline Reactions (01:30:12) - Intel CEO Pat Gelsinger has been vocal about the critical role of artificial intelligence (AI) in the company's future and the broader tech industry. He emphasizes that as everything becomes more digital, the demand for semiconductors, which power AI technologies, will continue to grow. Gelsinger refers to this trend as the "siliconomy," highlighting the increasing reliance on silicon-based technologies in our daily lives. Under Gelsinger's leadership, Intel has launched an "AI everywhere" strategy, aiming to integrate AI capabilities across its product lines. (02:16:04) - 𝕏 Timeline Reactions (02:29:41) - Sheel Mohnot, co-founder of Better Tomorrow Ventures, announced the successful raise of a new $140 million fund focused on pre-seed and seed-stage fintech companies. He discussed the firm's broad investment approach, including vertical SaaS and B2B marketplaces, and noted the evolving fintech landscape, highlighting recent public offerings by companies like Chime, Klarna, and Etoro. Mohnot also emphasized the importance of resilience for founders during challenging periods and shared his perspective on the competitive dynamics within the stablecoin sector. (02:40:30) - Santiago Nestares, co-founder of Dual Entry, an AI-driven ERP company, discusses their recent $90 million Series A funding and the challenges of modernizing legacy ERP systems. He highlights their innovative migration engine that enables rapid, low-risk transitions, allowing clients to experience the platform with their own data before committing. Nestares emphasizes the importance of reducing repetitive tasks for finance teams, enabling them to focus on strategic decision-making. (02:47:35) - Austin Federa, co-founder of DoubleZero and former Head of Strategy at the Solana Foundation, discusses how DoubleZero is building a high-performance, permissionless fiber network to enhance blockchain infrastructure by increasing bandwidth and reducing latency. He explains that the current internet was not designed for high-performance systems like blockchains, leading to inefficiencies and vendor lock-in due to data transfer costs. By creating a decentralized network of dedicated fiber and subsea cables, DoubleZero aims to provide faster, more efficient connectivity for distributed systems, starting with Solana validators as their initial market. TBPN.com is made possible by: Ramp - https://ramp.comFigma - https://figma.comVanta - https://vanta.comLinear - https://linear.appEight Sleep - https://eightsleep.com/tbpnWander - https://wander.com/tbpnPublic - https://public.comAdQuick - https://adquick.comBezel - https://getbezel.com Numeral - https://www.numeralhq.comPolymarket - https://polymarket.comAttio - https://attio.com/tbpnFin - https://fin.ai/tbpnGraphite - https://graphite.devRestream - https://restream.ioProfound - https://tryprofound.comJulius AI - https://julius.aiturbopuffer - https://turbopuffer.comfal - https://fal.aiPrivy - https://www.privy.ioCognition - https://cognition.aiFollow TBPN: https://TBPN.comhttps://x.com/tbpnhttps://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235https://www.youtube.com/@TBPNLive

You're watching TBPN. Today is Thursday, October 2nd, 2025. We are live from the TBPN Ultradome, the Temple of Technology. The Fortress of Finance. The capital of capital. That's right. It is good to be back. That is a lovely fedora, Jordy, if I say so myself, a yellow fedora celebrating ramp.com.

Time is money. Looking good. Save both. Easy to use corporate cards, bill, payments, accounting, and a whole lot more. If you've been sleeping under a pile of receipts, you probably haven't heard the news. Ramp is doing a publicity stunt in Manhattan, in New York, the big apple, the city that never sleeps. With Kevin from the office. Yes, he's going to be locked in a box, I believe.

I don't know how much they've actually teased about this, but we obviously know a lot. But there are some billboards up and some out-of-home ads, and they look great. And the tagline that I did see on the timeline that is public says, I'm CFO now. of course a reference to the social network and Mark Zuckerberg. And it's a lot of fun and we will be tracking this stunt as it is right. I cannot get enough of this fedora. I might start wearing this every day. And I believe it might be streamed live.

It might be on Restream. One live stream 30 plus destinations. Multi-stream and reach your audience wherever they are. Right. The top story today is of course about the slop trough. How are we going to keep the lights on at the slop trough? That's right. The slop trough, it takes an incredible amount of electricity to keep the lights on for the

piggies. Everyone was worried about water. Yep. Turns out electricity may be a bigger deal. Data centers need water. They need water, but they're pretty efficient. They recycle the water. So the water isn't that big of a deal.

Maybe we don't know all the data. But the electricity does seem a good thing. Your electricity prices to keep skyrocketing. Yes. And what in the, you know, among, you know, many of the issues with that. is that it will just ultimately make huge amounts of everyday Americans genuinely hate AI. Yep. Because chat GPT is cool.

Yep. They might use some other tools. Yep. But if somebody's sitting there watching their electrical bill, just climb month over month, year over year. Yep. At some point, they're going to look around and say, I miss when I could just Google something and I didn't have to pay an extra $100 a month for power. Totally.

So this is already hitting the timeline. timeline, it's clearly in the zeitgeist. This question of electricity prices, is it worth it culturally as a trade-off? We get AI, but we pay more in electricity, whether or not using AI. This has been bubbling up for a long time, but the Sora thing seems like a big deal. I don't even know if this is a Sora video, but there's a post from September 5th. Oh, that must have been pre-Sora. So this wasn't even SORA, but someone generated an AI video, of a woman smashing. Play this video because there's going to be maybe one person on Earth that thinks the, you know,

massive increase in electrical. Is worth it? Okay. So let's play this video. Oh, God. Not with the sound. Okay. Horrific.

Worth it. Worth every penny. And clearly fake and weird and the glass flows through. Pretty interesting. Tyler's over there being like, we're going to train robots on that. We got it. It's an accurate physics, Sam. It's worth it.

We're getting humanized. So this is how we get to the Dyson sphere. Yes, this is a critical path to the Dyson sphere, obviously. If you're doubting that, you are so... So what's notable here is that the guy, Chipotle Rowan says, pretty cool that my electric bill has gone up 25%. So other people can make this crap.

And he gets 300,000 likes. On X. Like, this is the... That's basically the... All the user. That's basically every user like this post. Eight million views, 17,000 reposts. It's a lot.

And I wanted to dig into some numbers, so I did fact check that 25%. I think it's real. So in 2021, the average electric bill for an average American was $1,452. It's like a significant amount if you're an average American, you're making 60K or something like that

or 100K a year. And now that's up to $1,728. That's exactly 25% increase. So it does seem like electricity prices have gone up by 25%. And if you look at some of the charts, there really is a spike right after 2020. When AI investment started ramping, electricity prices went up. Now, some of this might be COVID,

but it does seem like from 2010 to 2020, electricity prices were pretty stable, about 12 cents per kilowatt hour in the U.S. city average on the St. Louis Fed. But then 2020 hits, AI comes, and things have been up into the right ever since and not in a good way. J.B. Morgan estimated that 70% of last year's increased electricity cost was the result of data center demand. And the, and the, and the, and the, here is the logic is not quite perfect, but it's not great either. So it's that AI slop will be irresistibly dominant. We've seen this with everyone who works at our company becoming entirely one-shotted by short form AI video. No one can get anything done. It's a

disaster. It's a crisis. Of course, John's being sarcastic. Somewhat sarcastic. There's a couple people that have been a little... They're on notice. They're on notice. But yes, so you have assume meta vibes, Open AI, SORA will be massive successes and energy prices will skyrocket as humans spend all their time at the slop trough. I don't think personally that there'll be a fast takeoff in AI slop consumption. I think it will kind of grow slowly. I think this TikTok is just one of many that people will watch. But there's no denying that energy prices are rising. Per capita consumption, interestingly in electricity has been flat to down, mostly thanks to efficiency standards, but also there's just not that many new things that require a lot of energy.

So a lot of people are migrating from incandescent light bulbs to LED light bulbs to more efficient cars, more efficient washing machines, dishwashers, everything's getting more efficient. So the use is actually pretty flat, but the price is going up because we haven't been building any new electrical infrastructure. But that is changing. So the good news is that American energy production is growing, with tons more capacity coming online soon. companies are working across all categories. And there's actually a bit of a fight developing between nuclear and solar maxis. Elon is a total solar maxi, but there are a bunch of folks who are going all in on nuclear.

Including Rick Perry. Rick Perry. He's company Fermi. IPOed today at a $15 billion valuation. Not bad for a nine-month-old company. Not bad. A land lease. Yeah.

And not a lot else. And the hyperscalers are doing deals to bring decommissioned nuclear power plants back online. I see that as a very positive development. We've talked to probably, what, five different, 10 different nuclear CEOs on this program, from Radiant to Valar, to the nuclear company, to general matter.

There's so many companies in the space that are doing cool stuff in nuclear. And hopefully all of that comes online, And although the nuclear projects are very slow, but there is more capacity coming online. And, you know, the tech industry just has no incentive to become maybe more of an enemy of average Americans. They're not polling super great, but this could potentially make it worse. Yeah, the tech industry is addressing that. The tech industry got through the social media is terrible and bad crisis.

Everybody's sort of, a lot of the people that were most negative. around it are still power users of social media? I mean, the social network too, like the leaked plot is about how they were like meddling and all the, I think it was like election stuff. I don't know if people are fully over it. Yeah, yeah, yeah, people are upset generally. And that gives the tech industry an incentive to build more electricity so this doesn't become a bigger issue.

Sam Altman's guided towards 250 gigawatts of capacity at OpenAI by 20303. That's an insane number for reference. 23, 10 years before his prediction, America as a whole produced average power of around 500 gigawatts, 485. And so he's saying, I need half of the power in America currently for chat GPT and AI slop, right? And if he doesn't build more and doesn't justify more builds, that would mean way higher prices, right? Because opening I would be buying all of that electricity and selling it.

Not to mention, he'd have to choose between curing cancer free education. That's true. That's true. Which he's effectively requested. No one wants to make that trade. Not make him choose. No one wants to make that trade off. So that 250 gigawatts will not just be pulled from the grid broadly.

It will be built. There will be new builds, and that's already happening. So AI companies, the problem right now is that AI companies are pulling forward energy way faster than new capacity can be built. Like it's very easy to buy some GPUs, rack them in a tent. and run the one gigawatt cluster. Like Colossus 2 is an example of that. Like build it and then pull the electricity from the grid. It's not as easy to build a new nuclear reactor.

It's not that easy to build entirely new energy capacity. But there are certainly plans to do that. Back in the day... There's also some new... I saw a good post earlier from an account called Pantex Cap talking about how power companies. Companies are overwhelmed and tired of a lot of the BS that's happening in some of these new data center buildouts. They're putting up huge barriers to weed out the BS.

If you want 500 megawatts from Georgia Power, you now have to put up $600 million up front. And it's held against a 10-year power agreement and withdrawn monthly. Alabama Power is making people to commit to a minimum of 90% of the power requested. So you can't ask for 500 when you want 200 and release it later. So a lot of people will be requesting a lot and just not even necessarily confident. that they're going to use it, but understanding that it's an asset to have access to it. Yeah, no, that makes sense. I mean, years ago, the hyperscalers, Amazon, Google, Meta, Microsoft,

they were in a very stable place in terms of energy consumption, pre-AI boom. And it was so stable that they could basically promise to go all in on clean energy going forward. They made net zero pledges. They satisfied a bunch of ESG requirements. But now the AI race is on. And so new data centers need to come online, and we're seeing that with these new data centers in tents and all these new construction projects.

Natural gas, solar, recommission, nuclear, anything's on the table. We've heard that quote before from Chase over at Crusoe. AI companies would burn whale oil if they could. The energy is so scarce. It really is the most important resource for all of this. And so data centers consumed about 4.5% of U.S. electricity, in 2023, and that number's growing quickly could be 6% in the next few years.

By late 2024, 6.5 gigawatts of new capacity was underway. That's actually not just for the data center world, but just in America, they're adding 6.5 gigawatts of capacity. Now, remember, we're almost at 500 gigawatts. So that's like a 1 to 2% increase. It's like not that much, but there are now plans for up to 100 gigawatts of new buildout. over the next few years, obviously a lot of that is up for, it's like still at the LOI phase, but that would mean a 20%, like 20% increase in energy production is sort of like being papered right now.

You can think about it that way. Like America and the business community and the energy world and the AI companies, they've all come together and said like, okay, let's sign and let's jump over the threshold and let's go build an extra 20% capacity, which would be incredible. because if AI flatlines and people don't show up to the slop trough and they're just like, I'm good, just give me a database query, please. And then, and they go back to 10 blue lengths, who knows, then basically you have 20% more capacity. Price should fall.

It's a basic supply and demand equation. On the flip side, if tech builds the infinite just ultimate slot machine, but nobody comes, you know, in that case, energy prices fall. on the flip side, if demand is certainly, demand is certainly there today. You can just look at Foundation Labs at OpenAI, Anthropic, Google Deep Mine. They're all seeing exponential growth

and token generation. If it's an exponential curve, get ready for expensive electricity. But if it's a sigmoid curve, and we're on our sigmoid grind sets, America probably overbuilds, and then prices fall. The internet overbuild that followed the dot-com boom was pretty positive externality

of birth a ton of new tech companies that were able to leverage cheap broadband, and backhaul and internet services once the application is actually caught up. It's even easier to imagine a positive scenario with excess energy lying around because you can do even more with it than just fiber lines. But all of this is tight walking on a power line. And so interesting to keep track on it. Sager and Jetty had a good take.

So Derek Thompson shared a series of charts in the first slide of our deck today. Derek Thompson, a friend of the show, says data centers, AI are gobbling up electricity, but the share differs significantly by state between 2010 and 2025. Data centers went from less than 5% to roughly 40% of Virginia's electricity consumption, sweet Jesus. And of course, the city, the states that are not on here are like California. California produces an immense amount of electricity, doesn't have a lot of data centers. When we use Netflix in California, it comes out of Oregon, usually. When you're on the East Coast, the energy is not made in New York State.

It's made in Virginia, and then you access the U.S. East and AWS. cluster. And we're seeing that with Texas and Mississippi and Alabama and all the new data center construction, not really happening in the biggest states. And so you can look at this and see, wow, Virginia is at 40 percent electricity consumption. Well, nationally, we're still at, you know, 4%, 5% of data center consumption. But it's rising and it's a big deal. And it's going to be crazy in those state legislation. So yeah.

Soger's point is that, you know, right now effectively single-family homeowners, businesses are subsidizing the, basically subsidizing the cost for these data centers that are coming online. Yep. And so, yeah, he says, why should we be paying sky house? high power bells for Sam Altman's AI slop video generator. And it's a question that tech will have to answer. It's crazy because of course, Sam Altman, leader of Open AI, you know, was a driving force and creating SORA, but he's also videos of him are the ones, are some of the most viral. Yeah. Videos created on the platform so far. So it really is, it really is him. But we are at this precipice where AI is

cool and tech people are into it but it's becoming a it's becoming culturally like the danger here is become it becomes like NFTs right I think you were saying this yes I think yeah potentially hated yes

even faster than social media yes especially with the video stuff the video stuff has the same potential for like oh this just bugs me in the same way that the expensive monkey picture bugged people People weren't that annoyed by Vitalik Buterin building Ethereum and nerding out about secure blockchains. It was when it became, okay, wait, what?

Celebrity is hawking a million dollar monkey picture on a late night TV show and I'm losing my mortgage off of, like, why is this happening? That's where people got really frustrated. And I think it'll be the same thing if people are like, okay. I don't like this and it's increasing my. Yeah, yeah. It's very different when it's like, like, I have a job. job and there was a little bit of a boring thing and now I use chat chupit to summarize those meeting notes and it's delightful and I'm okay with that versus like this stuff is not good and I don't like

it and and also I'm paying for it that sentiment that got 300,000 likes that there's also there's also a growing sentiment around even the the non-coding use cases what the value is we had we had dinner last night with a very successful entrepreneur who's not really, he's built a widely known SaaS company, but not really in the Silicon Valley bubble, not kind of caught up in the AI hype cycle himself. And he said something, a line that stood out, which is AI is really good at helping people do unnecessary work faster, right? Just like producing big reports, producing research, right? What was your example? It was like... And you gave the example of you know, if you're prompting an AI and you're like kind of bullet-bulleting out, for example,

an email or a document, you can kind of get, you could kind of just send, give somebody the bullet points. That's because I have chat GPT running in my brain. Like if you send me three bullet points and you just say, hey, John, imagine this is like five paragraphs. I can just do that in my head in two seconds. It's very easy for me to do that based on facts. Organic intelligence. Yeah. Also, you know what else I can do? I can imagine that it's being read to me by a monkey and a jet ski. I can just imagine that. If you tell me three bullet points, hey, today we're going to talk about AI,

we're going to talk about crypto, we're going to talk about Ramp. I can just imagine a monkey saying that to me. You don't actually have to instantiate it. You can just share the idea, and I can instantiate it in my head. But, yeah, the example you had was funny, was like if you're going to a five-minute meeting

and you're like, I used AI to generate a 50-slide deck in a 400-page memo. And it's like, did that help? That seems entirely unnecessary. Yeah. Yeah, a while back when GPT psychosis was becoming more discussed, I was predicting that AI was going to go through a really, really rough PR cycle on par with what social media started to go through.

I think that's generally happened. You correct in the macro, maybe wrong. Well, no, I actually think generally correct, right? There's an article today. Yeah, every legacy media company has been putting out articles on people that are using AI in weird ways and going crazy, or talking about how, you know, chat, you know, Chat GBT or another product is like mentioning, you know, the word suicide, a bunch to users. They, it seems like they release some pretty big updates recently that basically take people

from 4-0 back to GPT-5 reality, if they sense behavior that's unhealthy. But, again, every legacy media company has been pushing out content around people going crazy through AI. Now you add in, you know, electricity costs going up. And I think it's just going to continue. Well, the good news is that we've been here before and there are two positive possible outcomes here. One is just that the tech community comes together and builds a bunch of nuclear and solar and just power gets cheaper. Like that would be amazing. We need more energy production and clean energy production is kind of limitless. It's magical when it works. The problem is AI usage is exponential tokens.

But maybe it's a sigmoid function and we wind up building even more and then prices do fall. It's totally possible. The flip side is that we've seen there are solutions to energy intense algorithms. We've seen this in past with crypto. So there was the same during the NFT bubble, there was a huge push for, oh, Ethereum is using so much energy for proof of work. they were able to migrate to proof of stake. And, you know, energy prices fell. And it actually took a huge hit to NVIDIA's sales that year

because NFT companies and blockchain companies didn't need to buy as many GPUs because they were able to do the same algorithm effectively on a more efficient system. So you're saying Jensen is Satoshi? Maybe. Maybe. I mean, it was extremely bullish for him, for sure. But I do think that we,

We've successfully moved past the crypto is using all the energy narrative, and that was a technological solution to that. What do you laugh at that? The Fedora finance. Well, speaking of crypto, Privy, the wallet infrastructure for every bank, Privy makes it easy to build on crypto rails, securely spin up white label wallets, sign transactions, integrate on-chain infrastructure all through one simple API. The official wallet infrastructure provider. And you're not going to be needed a gigawatt of energy capacity to spin up a... an application on crypto rates. Great line paracletes, the Fedora finance.

Yes, it's great. Well, Andre Carpathie had some reactions to the Dwar Keshe Patel podcast with Richard Sutton. He said his background, the bitter lesson has become a bit of a biblical text in frontier LLM circles. Researchers routinely talk about and ask whether this or that approach or idea is sufficiently bitter lessen-pilled, meaning arranged so that it benefits from added computation for free, as a proxy for whether it's going to work or even worth pursuing. The underlying assumption being that LLMs are, of course, highly bitter-lessened-pilled. Indeed, just look at LLM scaling laws, that Chinchilla scaling laws, discovered by Google, where if you put, compute on the X-axis,

number go up and to the right. So it's amusing to see that Sutton, the author of the post, is not so sure that LLMs are bitter-lessened-pilled at all. They are, trained on a giant data set of fundamentally human data, which is both human generated and finite. What do you do when you run out? This is the data wall that was a popular point of contention for a while. How do you prevent a human bias? So there you have it. Bitter lesson pilled LLM researchers taken down by the author of the bitter lesson directly. Rough, he says. In some sense, Dwar Keshe who represents the LLM researchers viewpoint in the pod and Sutton are slightly speaking past each other because Sutton has a very different architecture in mind and LLM's break a lot of

principles. He calls himself a classicist and evokes the original concept of Alan Turing building a child machine, a system capable of learning through experience by dynamically interacting with the world. There's no giant pre-training stage of imitating internet web pages if you're a kid. Dylan Patel was on Investing the Best and he keeps talking about how he has a baby cousin, I think, who learns by like, and he keeps using this example of like the baby puts the hand in the mouth or puts the foot in the mouth and like learns immediately. And it's not like the baby looked at you and saw like,

this is what a hand tastes like. This is what a foot tastes like. Or like this is what a rock tastes like. The baby definitely learned that from first principles through exploring the actual physical environment and that's not what LLMs do. And so there is a good point there. So there's no supervised,

there's also no supervised fine tuning at which point, which he points out is absent in the animal kingdom. It's a subtle point, but Sutton is right in the strong sense. Animals may, of course, observe demonstrations, but their actions are not directly forced or teleoperated by other animals. Another important note he makes is that even if you treat, if you just treat pre-training as an initialization of a prior, before you fine-tune with reinforcement learning,

Sutton sees the approach as tainted with human bias and fundamentally off course, a bit like when Alpha Zero, which has never seen human games of Go, beats AlphaGo, which initializes with them. In Sutton's worldview, all there is is an interaction with the world via reinforcement learning where the reward functions are partially environment-specific, but also intrinsically motivated, e.g. fun, curiosity, and related to the quality of the prediction in your world model. And the agent is always learning at test-trial. by default. It's not trained once and then deployed thereafter. Overall, Sutton is a lot more interested in what we have in common with the animal kingdom instead of what differentiates us.

If we understood a squirrel, we'd be almost done, he says. Well, speaking of agents, go check out cognition. They're the makers of Devon, the AI software engineer. Crush your backlog with your personal AI engineering team. So, Andre Carpathie gives his take. If we understood a squirrel, we'd be almost done. Undercarpath gives his take on the Dorcasch Patel, Richard Sutton Pod. He says, first, I should say that I think Sutton was a great guest for the pod, and I like that the AI field maintains entropy of thought and that not everyone is exploiting the local iteration of LMs. I completely agree. It was great pot. AI has gone through too many

discrete transitions of the dominant approach to lose that. And I also think that his criticism, his criticism of LLMs is not bitter lesson-pilled is not inadequate. Frontier LLMs are now highly complex artifacts with a lot of humanness involved at all the stages, the foundation, the pre-training data is all human text, the fine-tuning data is human and curated, the reinforcement learning environment mixture is tuned by human engineers. We do not, in fact, have an actual single clean, actually bitter-lessen-pilled, turn the crank algorithm that you could unleash upon the world and see it learn automatically from

experience alone. Does such an algorithm even exists? Asked Andre Carpathie, co-founder of OpenAI, head of Tesla's autopilot program for a while, now working on something new. What is he actually working on? He is working on education technology, correct?

Yeah, it's called Eureka Labs, I think. They haven't released anything yet, though. He's in the trenches. Yeah, I think the question that I have is, you know, Ilya has said broadly they're focused on their own definition of super intelligence and that, and it seems like he might not just,

it certainly doesn't feel like he's going to launch a chatbot anytime soon. Ad network. Ad network or business process optimizer. Enterprise software. I wouldn't be mad at that. I would love that. But I would hope that he's working on something

Maybe just an entirely novel. Yeah, entirely novel. That he could unleash on the world. Yeah, I mean, it makes sense to give him the budget and time to go figure that out and explore. I love it. Two example, proofs are commonly argued, are commonly offered to argue that such a thing is possible. The first example is the success of Alpha Zero learning to play Go completely from scratch with no human supervision whatsoever. But the game of Go is clearly such a simple, closed environment that it's difficult.

to see the analogous formulation in the messiness of reality. I love Go, but algorithmically and categorically, it is essentially a harder version of TikToktoe owned. Sorry, Go players. Stick to TikToktoe, buddy. The second example is that of animals like squirrels. And here, personally, I am quite hesitant whether it's appropriate because animals arise

by a very different computational process and via different constraints than what we have practically available to us in the industry. animal brains are nowhere near the blank slate they appear to be at birth. First, a lot of what is commonly attributed to learning is, in Andre's opinion, a lot more maturation. And second, even that which is clearly learning and not maturation, is a lot more fine-tuning on top of something clearly powerful and pre-existing, the brain. For example, a baby zebra is born and within a few dozen minutes, it can run around the savannah and follow its mother. This is a highly complex sensory motor task, and there's no way, in my mind, that this is a very small.

This is achieved from scratch, Tabula Raza. The brains of animals and the billions of parameters within have a powerful initialization encoded into ATCGs of their DNA, trained via the outer loop optimization in the course of evolution. If the baby zebra spasmed its muscles around at random as a reinforcement learning policy, would you have you do at initialization, it would not get very far. Similarly, our AIs now have neural networks with billions. of parameters. Andre goes on.

He does give a summary. He says, anyway, in summary overall, in actionably, I think this pot is solid, real talk from Sutton to the frontier LLM researchers who might be gear shifted

a little too much into the exploit mode and not enough in exploring. Probably we are still not sufficiently bitter, less, and pilled, and there's a very good chance of more powerful ideas and paradigms

other than exhaustive bench building and bench maxing and bench pressing, and animals might be a good source of inspiration. Just look at how a gorilla can bench press. That's what we need to do. Is anybody actually working on that? Working on getting guerrillas. Come on. You got the robot Olympics. Let's see some Olympic lifting. Yeah. Intrinsic motivation, fun, curiosity, empowerment, multi-agent, self-play, culture, use your imagination. I keep coming back to this idea of like,

we're obsessed with this like agentic horizon, one hour, two hours. I'm like, my initialization, my initialization prompt, humanity's initialization prompt, is like, be fruitful, multiply, right? That's Genesis. And I think there's something where humans, the goal that we seek is like survive forever, reproduce forever. And so I don't really see my life as like a string of one hour tasks or five hour tasks. I see it as like a hundred year journey, potentially thousands of. of years when I think about my multiple generations. And so I, it just feels like we're pretty far from something that is satisfying, all the different criteria.

And of course, you know we need smell of vision. And we're not making any progress there. Tyler, what was your reaction to this? Yeah, I mean, it's very interesting. I think, like, Sutton is very RL-Pilled. I think, like... Wait, wait, RL-Pilled? Yeah, Sutton.

Are we in the RL paradigm? Yeah, but it's still, I mean, we're still on like very early days, right? I think Sutton envisions this thing. It's like, it's just RL, right? At its core, you start with nothing and you learn the world where the current paradigm is like, first you basically mimic the world. And you're basically like at first, when you're training a model, it's literally just like memorizing facts.

And then you slowly get it to generalize. Yep. So right now, the way I think about it is like for a modern chat bot that I'm using, there's maybe like 300 megawatts of pre-training compute and then 300 megawatts of RL thrown at it. And it's like a 50-50 balance or something. And then maybe he's saying it should be like 100% balance towards RL. And like there should not immediately. So to ground this in SORA, part of why this new app is strategic is they're generating

an asset and then they get to immediately see feedback. See feedback. Basically the reward is more engagement in the, you know, watching it. it longer. If you see a sore video and you're just on X, OpenAI is not really able to leverage that. Yep. Right?

You could indirectly be like, OK, this went viral. Maybe that's good there. But in the app, if you're in the trough, you're immediately the time that you've spent watching the video, did you like it? Did you send it? All of that is providing feedback for the model. RL.

Yeah. Tyler. Yeah, I mean, so I think if you look at just training, like RL is nowhere close to half of computer. It's like way less. But I think it is like it is like a very cool idea to basically have this very like pure model in some sense. It's just like experiencing the world and learning from that. But I don't know like if you go to the Dylan Patel kind of example of the baby like putting its finger in its mouth to like figure out where its fingers are.

Like the baby is not just like randomly twitching like its nerves are not just randomly firing. There's still some like prior that comes from evolution or whatever. but there is like, it's not just like purely experiencing things and then updating. Yeah. There's something there. I was thinking about the zebra example. Like that's true, but also like the baby is kind of randomly twitching in the womb. And like the baby, like you can see on the, on the, what's it called the oscilloscope or something?

I forget what it's called. There's some sort of like device that you can like look inside and see the baby. And the baby's moving like very basically, but probably like doing some reinforcement learning essentially like pre-birth. I think, I don't know. Yeah, I mean, on some level, like, there is some encoding that is going on into the baby of, like, this is how, like, you can walk around or something like that. Yeah. You're experiencing world, then you're updating on the actions you took.

You have some goal. You either complete the action or you don't, and then you update. I think imitation is more important. There's like the kind of like the Gerardian take or whatever. Is that like humans are apes, right? To apis to imitate. If you drop a baby in a prison cell from birth, it'll never learn to walk because it needs to see a human walk to learn that.

Yeah, something like that. I think they'd figure it out. Okay, maybe not walking, but like doing, you know, useful things. Like, it's not going to. I think, I think if, I think if my son was born in a padded closet and just given food and an Excel sheet, he'd be writing VBA macros in five years with no external. Well, maybe it's been encoded into your DNA. Pure, pure reinforcement learning, experimentation from first principles.

I think it's possible. I don't know. Anyway, Figma, think bigger, build faster. Figma helps design and development teams build great products together. Hey man, just wanted to say I think we talked about this yesterday, but I love it so much I put it back in. Hey man, I just wanted to reach out and say

I loved how much you drank at the networking event last night. There are a few of those. This is becoming a new mean format. There was another one. Find the other one for me in the meantime. Apparently there's a game that makes you a YC partner for a day. You watch old pitches and guess if they made the batch.

I wonder where this data came from. Usually YC is pretty locked down with these old pitches, but this seems like a fun game. I went to the top of Hacker News yesterday. Here's the other post. Hey, man, just wanted to follow up and say, I admired your ability after three drinks to persistently ask who had a bag at the super serious networking event last night.

Super serious networking event. Yeah, I could never. What do you think about? I was texting with Tyler about this. Like there's this post. Sam Altman is playing 4D chess. Sora 2 is about to take over social media. the virality is guaranteed once the scales.

Billions in ad revenue will flow straight into more compute, fueling the flywheel in a year, so or two will be so efficient and cheap that margins explode. You can't out-accelerate Sam Altman. And I wanted to reflect on this idea that Sam said, it's way less strange to watch a feed full of memes of yourself than I thought it would be not sure what to make of this. And Gabriel, he's, again, my read on this is,

I don't, I don't fully believe that he actually, I don't know. I don't know how to read into this other than he is very incentivized to normalize people being deep faked all the time. I mean, it's a crazy, crazy move to just be like, somebody had to go first. My likeness, I'm going first. My life, I'm merging.

I'm merging. It's like, it's being on the first SpaceX rocket. Hey, it's not so bad. Look at this. It's not so bad. I don't want to turn this off. I don't want to see that. I don't like the skibity of thing to begin with.

and then you mash it up and it gets even more horrifying. But anyways, he went first. He's very incentivized to get a bunch of other big names, personalities to agree to do the same thing. Is this an example? I would imagine it's very weird to watch videos of yourself committing crimes. You know, there's a video of, I guess a video of him in Blackface loading around. There's a bunch of videos that I think would be incredibly strange to watch

of yourself. But stranger than your expectation? Because I would expect it to be weird. I expected it to be weird when Tyler sent me a deep fake of myself bench pressing and then turning into a golden retriever. I was kind of just like, eh, yeah, this is below my expectation.

Like, it is less weird to me. I don't know. You haven't seen one of yourself yet, right? Has anyone made one of you? You haven't uploaded. I don't think I've done, I haven't. I haven't agreed to do the cameoia thing. What about you, Tyler?

Did you expect it to be weird to see yourself in a SORA video? and then was it weirder or less weird than you expect it? I don't think it's that weird. Yeah. But it's also like it's still not like that good where like it's kind of me, but it's not like all the way there. Totally, totally.

But I think, okay, Jordy, do you think that him basically releasing a social media where like half the content is just him? Is that an aura farming move? I think it increases his fame, which is a resource that he's able to leverage. But I don't know that it didn't, you know, a lot of the videos that I've seen don't,

I don't think are necessarily good for the brand, but it's also just putting out so much stuff into the world that it just all gets diluted and you don't read too much into any one video because it's not created by him. Do you believe in Cosgrove's law? That ore farming is all you need. Break it down for us. Yeah, always be ore farming.

Okay, I think I have a text somewhere, but yeah, okay, basically ore farming, right? Yeah. You can think of like most things are on some level like based off like interacting with other people. Like any kind of if you're, you know, raising around, if you're going to make some deal, right? It's based off like your interaction with someone else. Yeah, people raise, want to raise from tier ones because every dollar is the same, but you get to aura farm the, you know, the GP. Yeah, and then it's like in, like when you're raising around, you want to be ore farming the VC, right?

because they want to be like, oh, this guy's, like, pretty sick. Like, I want to give him money or something like this, right? So basically, the idea is always be aura farming, right? You always want to be doing something that increases your aura because everything you do, like any kind of interaction you have, will be influenced by, like, your personal aura. Yes, but stretch the analogy. Like wearing a fedora of plants.

But stretch the analogy to the absolute limits. Sometimes when you farm, your lands grow fallow. and you have a, you have a failed harvest, and you can also salt the ground so that nothing will ever grow in. No bounty. And so there is a world where you go out to ORA Farm, you set up your ORA farm,

but someone has salted the ground and you don't grow a single crop. Yeah, so that's like a, I mean, that's a failed oar farm. Failed Oro Farm. So the question is, is this, the sort to putting yourself in the cameo, it is an attempted aura farming. Was it a successful aura farm?

Yeah, I think most, like, I would say in most cases, ore farms, you can only tell the success of a farm in retrospect. If it's successful. Yeah. So it's like, in six months, we'll see. I think if people are still on SORA, if the app is still like being used as a social media platform, not just like a creative tool,

then I think it will be a successful farm because now Sam is just, you know, he's just everywhere. It does feel like there's always a winner and a loser in an ore farm. And like if you are ore farming and you're going down. Yeah, VCs will do this. They'll buy secondaries in a company to say they're an investor, even though they didn't participate in a primary round. That's true.

And then also for every time a founder, aura farms a tier one venture capitalist and then blows the company up in a major scandal, that reduces the aura of the fund that put the, the money in and they offended. They got farmed. Yeah, it makes, if you backed FTCS, you got farmed, you look bad. Yeah, like there's always two sides, right?

Because in Jordan's example, the VC is farming company, but when it gets exposed that they actually bought secondary that they weren't real investors, then it's a negative aura. The company is farming the VC. Yes. They're like, oh, we're so great that. But there must be

positive some aura farms where two people come together. They They grow a beautiful garden together. When an emerging fashion brand partners with a legacy brand, let's say a streetwear company partners with Solomon, they can both benefit because the streetwear brand is getting new legitimacy. And Solomon gets to be like culturally relevant and cool. Yes. And that is a positive sum or a farm. Sydney, Sweeney and American Eagle.

Positive sum or a farm. The stock's way up. everyone's doing well in American Eagle world. Sidney's back in the news and dominating. It seems like a win-win. They oarformed each other successfully. They grew a new bountiful harvest together. It was more positive for American Eagle.

It was more like net neutral, I think, for Sidney and that she got a bunch of blowback from it. She got a bunch of attention, but it wasn't all positive, whereas American Eagle just the stock ripped. Anyways, I have an interesting post here that I wanted to throw in. and it's from a guy named Robert. If we can pull it up in the timeline.

Let's pull that up and I'll tell you about Vanta. Automate Compliance Managed Risk and Prove Trust continuously. Vantta's trust management platform takes the manual work at a security and compliance process and replaces it with continuous automation. There we go. Let's look to go to this post. Went super viral. This is Robert says,

Good game everyone. And he's charting total job openings versus the S&P 500. And he shows that when chat GPT, was released there was this massive divergence now a lot of people took this at face value and they thought okay chat gbt is just killing jobs yep but i put through in another post here um anybody that's used chat gpte understands that it's valuable but it doesn't replace them at work yep like maybe i i i haven't met somebody that's been like hey jordy like i have to admit i've lost my job because chat chpt just does it for me right yeah it's there's certainly in key areas

is like Finn is like replacing like the need to add more CX reps. But but by and large chat GPT was not the catalyst. That's just like, you know, totally reducing the number of job openings. So there's another person here from Simon, Saris, who says the zero interest rate era is going to become lost history because people want to make up a narrative around AI. The white collar bloodbath didn't happen because of a chat app release. that happened because of the end of ZERP, which occurred rapidly in 2022. So literally when FTCT was released when FTCT was released when FTCL collapsed effectively during, you know, within like the same 30-day window.

Yep. And so it really was a lot of CEOs woke up and said, we have to get a lot more efficient. We have to get more lean. Let's not hire, you know, a ton of new people. Let's do more with what we have. The AI narrative from a job loss standpoint has also, it's also in some ways I feel like been a lot more narrative driven and that CEOs have a massive incentive to show they're getting more efficiency out of AI.

And so they're just saying to their teams, like, I don't want you to hire more people, just be more efficient. And even that by itself of just forcing people to say like, no, we're not expanding your team. Just do more with less. Yeah, there was also a huge nomadic contagion from the Twitter buyout where X was able to run with like one-fifth, the staff and the service doesn't really go down in their shipping features.

And so I think a lot of CEOs just kind of adopted a year of austerity, a year of efficiency, and wanted to test what they could do with less. And so they pulled back. And so, yeah, you have to look at the labs are scaling headcount rapidly. Anthropic is, you know, hiring teams to, you know, capitalize on all the demand they have. and they're some of the most, you know, AGI pilled of anyone, right? And so you have to read into what people are actually doing. And certainly there's been massive job creation in the kind of startup ecosystem broadly

just because of, you know, we were talking yesterday with a friend. It seems like every, it's become so normalized, but it feels like every day there's like just on TVPN, like at least 100 million of new funding announced, usually like, you know, half a billion, at least, at least lately. And so all, you know, a lot of that is going to go into hiring teams. And so anyways, this, this original chart thought it was worth calling out. Yeah, it's important to reality check the put one chart over the other and see if they line up perfectly. We had some updates from Tyler Cosgrove. Yeah, there was another chart. We were asking, somebody had made a chart tracking the NASDAQ.com bubble to today. And this person effectively did graphic design to make them look like they match up perfectly by the day. It looked so perfect. And it looked so perfect. And Tyler spent 24 hours this week trying to recreate it.

And no matter which index he was looking at, it could make it. Yeah, take us through the thought process. Yeah, I just sent it in the chat, but basically it's supposed to be NASDAQ from 1990, 2001, and then overlay on top is 2023 to now. Yeah. And then, yeah, I mean, they just like don't line up. Like, you can kind of see, like, there's a little spike, and it goes down a little bit and it rises. So the person just, like, made up the data or something?

No one would ever do that. I don't know if it's, like, literally made up, but you can tell that. This is the original post, right? Yeah. So the time scales are not lined up correctly. Okay. And then... Oh, so they compress time on one of the charts and not on the other.

Yeah. Yes. And even when I did that, they still wouldn't really line up. So I think certain parts of the graph are more, like, compressed than others. Okay. To make the kind of, you know... Yeah. So it's like a linear scale, then a log scale, and then a linear scale.

It's the ultimate chart crime. Yeah. This is, I think this is pretty bad. Do you have your actual overlay available? Can we look at that? Yes. Let me pull it up. I would love to see that.

While you pull that up, let me tell you about graphite. Code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. I love that when I hit that sound effect. I know you're going to amp up the WW. Fire me up. While he's pulling that up, John Galt shares websites should be this again, waiting for the first company to authentically do it.

Everything looks like a venture capitalist yoga retreat website at the moment. It's from the Sony website in 2002, Web Design History. Remember when we pulled up the old iPhone website on Apple in 2008? And we were like, oh, it's going to be like all iPhone focused because it was the iPhone moment. It was the new MacBook. It was like the MacBook air and like the iPhones off to the side and said, high technology is one of the features that you get with the iPhone. I like this. I wonder if this would convert, how bad this would be.

I wonder if someone's actually going to take this post and run with it and build a site that looks like this. We've seen some cool people do interesting things with web design. Post-Hog had a very cool website. There's still room for innovation in the website. Have fun. Do something different. Don't just copy Linear. Our sponsor, Linear is the tool.

It's a purpose-built tool for planning and building products. Meet the System for Modern Software Development, Streamline issues, projects, and product. your favorite products are built using linear. So do we have the actual truth zone edition of the stock chart? Okay. There's some overlap there. Yeah, I mean, they're both rising. Yeah, they're both rising.

They're both up into the right. But yeah, it's a wildly different time scale. Yeah, and that little like dip and then pick back up does not line up at all. Yeah, like, you can line up. Can line up certain parts of it. Yeah, that was the big dip. Yeah, that's that big dip. Yeah.

And you can line up certain parts of it, but then the rest of it doesn't work. So if you shift it around, like, if you shift the actual time scales of certain parts of the graph, then I think I could get it to line up. Yep. You need to copy the red line and make it a different color and just offset it by a little bit and be like, I'm going viral. That's the secret. That's the secret. Well, if you want to do your own analysis, head over to julius. dot AI, chat with your data, get expert level insights in seconds. It's the AI data analyst that works

for you. And so back on Sora 2 and the question of energy, Zephyr says pretty sure opening I made an algorithmic breakthrough. SORA 2 is a much smaller model. Tyler, I don't know what you think about this, but Andrew Main is saying another interesting metric is that V-O-3 costs $3.20 per eight-second clip. You get three a day. that's 10 bucks a day. I pay $250. So they're basically pay, like if I use all my V-O-3 credits on Google's top tier of Gemini, I'm break-even. But, you know, SORA currently gives you a hundred 10-second clips per day. That's $320 of video generation per person for free right now if that's what they're doing. I think they'd eat that cost all day long. What do you think? Do you think that they've actually

brought the cost down to 32 cents? per clip to three cents per clip. Where do you think it's going? Yeah, I mean, I would be very surprised if it was like the same kind of cost as a V03. I would, if I had to guess, I would say maybe like 10x. So maybe it's like 30 cents. But even then, that's like pretty steep. That's like pretty hard to squeeze that out.

I just think like they gave me a hundred. I've generated six over three days. Have you generated 300? No way. Yeah, I'm at like five. You've done five. and we're like kind of power users, early adopters, really testing this for work. And then they're also rolling out the invite slowly.

I wouldn't be surprised if this is, yeah, it's three bucks a prompt, but they're just going to eat that for a while until it gets to scale where, I mean, if it's a million new videos a day coming on the platform, that's only $3 million. That's only a billion a year. Like, they can eat that, right? It's only a billion a year. It's only a billion. That's really nothing for Open AI. They're a $500 billion

company that's raising tens of billions of dollars every second. And they, and, and, and what's at stake? Like, winning the next social platform, like, like, really eating the Troph Wars. The Troph Wars are going to be fought one 10-second clip at a time, no matter what it costs. Well, people are still doing things the old-fashioned way. Ramp is doing things the old-fashioned way. we talked about this earlier in the show, but Kareem says SOR2 is insane, unbelievable, and he got a community note on it because Ramp published a launch video

that it was shot with a real camera, and the community notes team put him in the truth zones that it was not created by SORA. It's a real commercial. I think that helps with the virality, and it helped. Karim got 400,000 views on this, 450,000 views.

I wonder what percentage actually just believed it was Sour 2. I saw someone in the chat saying, they think it's Sora, I don't know if they were joking. but I do think some people, you know, it's hard to tell. It's hard to tell these days. Well, we have an amazing announcement, which is that there are 19,000 private equity funds in the U.S.,

and there are 14,000 McDonald's in the U.S. We have more private equity funds than McDonald's, and I just think it's important to, you know, recognize this moment, this milestone. I wonder how that's calculated. Is that like, does that count like an LLC that holds a sports car in Montana? is technically a private equity fund? Or are we talking like there's a real office with a couple people that are buying companies

with that equity? Like, is it a true private equity fund? Like, what's the scale? There's probably some crazy power law, right? Where like KKR is like here and then there's like some company. Yeah. What's the accounting funds for ants? What's the median size of that private equity fund?

I would love to know that. Anyway, we can dig into it more. Speaking of budgets, the government is about to shut down. And it's already a polymarket. Isn't it already shut down? I believe it did shut down. Not good. The party's fight to cast shutdown blame.

You know, do people like the government? I don't know. It seems like a mess. Delian was on the show laughing about like how no other country just shuts down the government. Do you have the polymarket on when the shutdown might end? That seems like a useful, useful data point for us to track. We pull it up. Hours into the government shutdown, the White House issued.

New threats, congressional leaders were digging in, and some rank-and-file Democrats and Republicans were searching for a way out. Senate Majority Leader John Thune and Vice President J.D. Vance said they are open to talking to Democrats about extending, expiring health care subsidies, but won't negotiate as long as the government is closed. Democratic leaders said they won't agree to reopen the government until their health care demands, at least some of them are met. And when does Polly Market think it's going on? vast majority of folks, over 70% think that it'll be October 10th or beyond. 44% believe. October 10th, that's our one-year anniversary, I believe. Really? Something like that.

So, you know, hopefully the government reopens and we can celebrate the one-year anniversary of this show. In other news, Apple is shelving development for the cheaper, lighter Vision Pro, the Vision Air, planned for 2027, and they're prioritizing a more urgent effort developing AI smart goals. glasses to rival meta. This is interesting. There's some debate over whether or not this is actually going to happen. Apple's a big company. They could just be working on multiple things.

But the news from Mark German is that they are reorienting. So the company, Apple, had been preparing for a cheaper, lighter variant of its headset, code named N-100 for release in 2027. But Apple announced internally last week that it's moving staff from that project to accelerate work on glasses, according to people, knowledge of the matter. Smart glasses have emerged as a critical arena for tech companies, which are racing to develop AI-centric devices. That's crazy. That meta's making them dance. Making them dance. Ducks making him dance. Tim Cook was watching our Meta-Connect coverage and was like,

I want to hit that gong with those guys. Let me do it. I got to make some glasses. Dumb day. I'm day. I got to get my own Rocco Basilisk. I got to get, I got to get my own Rocco. Yeah, what could they do? Oliver Peoples. Is that owned by Luxottica as well. Also. Also owned by Luxottica? Where will Apple go? I think Apple will not try and partner

with a fashion brand. I would make that prediction that they will do their own and they will look like AirPods and they will be mostly all white and they will be a big status symbol to let people know that you have an Apple product on your face.

And there will be a couple of different colors. There will be different colors every year. But Apple will try and own that design language. Apple has done a few partnerships. Didn't they do an Hermes band for the Apple Watch? I believe they did that and a few other collaborations,

but Apple generally has been pretty limited on the, let's partner with Nike. They've done a little bit there on the running side, but in general, they have not done a lot of like, this iPhone is co-designed with another brand. It's just you can get a case from them. We'll give them the design. It's up to them to do whatever they want.

You can get the Louis Vuitton colors and whatnot on design. One more post before our first guest. There's a post from Modest Proposal who's sharing a screenshot from a transcript of Patrick's interview with Dylan Patel. Patrick says, yeah, he's doing demand guarantees. He's doing all this crazy stuff right now. Of course, they're talking about Jensen and Nvidia. Dylan says, yeah, right, right. He's using his balance sheet to win, try and win more, which is an interesting dynamic.

I don't know if there's ever been anything this in terms of the anti-competitive nature of this where you backstop clusters. Correve recently got a deal with in video where it was where they backstopped a cluster, right? So that's saying like if you buy this and and bring it online, we will guarantee the demand is there Dylan continues now Corweave would have never built this cluster because it's for short-term demand and renting GPUs on short-term is a terrible business model You want to do long-term contracts and you want to do long-term contracts with people with balance sheets. That's the golden goose but that doesn't exist so much, an important point. So you do long-term contracts with people who don't have a balance sheet, OpenAI.

And if you can't do that, then you'll do short contracts with people who do have a balance sheet. There's this whole matrix of who you rent GPUs to, but from Nvidia's interest, it's, you know what I really love is when venture capitalists fund a company and then 70% of their round is spent on compute.

They effing love that. They're talking about NVIDIA. Right? And that's what's happening with all these companies, whether it's physical intelligence, they're spending a lot on robot arms and SHIT too, but they're also spending a lot on compute, or it's any other startup that's raising cursor, et cetera.

And even if it's not directly, it's indirectly going to GPUs. They love when people spend their entire round on GPUs. It would be really good if this wasn't a two-year deal or a three-year deal for that compute. If it was, you can spend 70% of your round on one training run, leave a company with these ideas, gather the data, do the training run, and then you have a product and you show how good the model is and try it and raise again.

That would be really great for Nvidia, but no one wants to build a cluster who's predicated on that as a business model. That's crazy. So they have to backstop a cluster to do that. Yep. Before we bring in our next guest, there's a related article in the Financial Times.

One of Britain's best-known tech investors has warned of a disconcerting rise in artificial intelligence valuations, saying Nvidia's planned 100 billion investment in OpenAI brought uncomfortable echoes of the dot-com bubble. This is about James Anderson, who is at Bailey Gifford. He had a massive position in Invidia. He was his largest holding. He's now in Chinese battery maker C-A-T-L, Kettle. And he said, I have to say the words vendor financing do not carry nice reflections

to someone of my age. And so the, yes, the echoes of the dot-com are growing louder. Well, we are pleased to be joined by someone who is building in a completely different space. We have the chief operating officer of Axon. The president. The president of Axon. I was reading the chat. A little misinformation. Sorry to get you. What's going on, Josh. Welcome to the show. Oh, you. Good to see you. Thanks for hopping on. What is this setup here? That looks incredible. That looks incredible. You're in the command center. Yeah, yeah, we have some folks at AXon who just on their own come out, they just churn out new Zoom background. So every time I open Zoom, there's like five or six options to choose from that I've never seen before.

So this one looks pretty cool. It's looking great. Can you take us through kind of the shape of the business today, maybe a little bit about your journey, just to kind of set the table for those who might not be familiar with the business? Sure. At AXON, we make pretty much all the cool technology that police officers use day to day. Everything from their taser gun to their body camera to all the SaaS software in digital evidence management to all their reporting products, drones, virtual reality police training. We really believe that we can add a lot of value in a lot of different spots where really software and hardware intersect in policing.

And so I've been at the company 16 years, two months or sorry, two weeks after graduating college. Wow. We drove across the country to join Axon and here I am. That's amazing. What's been the most recent transformation in the business? I was talking to somebody about this category and very experienced investor. He was like, oh, no one's really made it work. And I was like, well, what about like AXon? They seem to be doing pretty well. And he was like, oh, it's not that big of a company. And then we looked up the market cap and it's a huge company. And you've been massively successful. But what has been the, like, the biggest transformative moment over the past five years for the business? Sure. I think really for the first 25 years of the company's history, it was really about hardware. And in the last five years, it's been far more about software. So we manage more than 40 times as much content as the Netflix library combined. So we're managing all of this police evidence in the cloud. And in figuring out how to make use of this massive data set for our customers has been

the name of the game. So most notably, a couple of years. ago we released a product called draft one, which is a Gen AI product, analyzes the body camera transcripts and creates the first draft of the police report for the police officer. So cops spend about 50% of their time writing police reports. We think we can make that like 10% of the time. Such a good use of Gen. So is that something where an officer in the middle of an action, let's say they're pulling somebody over for speeding, do they talk out loud and say, all right, you know, how are you, how, where is that data coming in from? Because I could imagine they're just kind of, John actively uses like transcription with chat chbt. So we'll be like talking about something and then he'll,

you know, be prompting it just with voice. I could imagine an officer does the same thing where they're saying like, hey, pulling this over. I'm at this, you know, intersection. I, you know, clocked them going 20 over. And, and then that just gets built in. Or what does that workflow like? Sure. So it's a. It's a. great point, cops are getting better at prompting the AI to, uh, to, you know, improve the quality of the police reports. But even without doing that, uh, just in the normal conversation, you can imagine that same incident when the cop walks up to the person's window and says, hey, you know, you're speeding, you're going 20 miles an hour over the yeah, I guess speed limit.

Daria you get to talk about. Yeah. Yeah. And even just like, uh, what's your name? And then the person says that. And then that's in the audio file. And then that gets transcribed. And that's just like, So much manual work. You're not putting anyone out of a job. You're just making their life better. It feels like a fantastic use of artificial intelligence. Very exciting. And you guys have been super acquisitive. I think you've done a number of acquisitions, but recently completed, it was an $800-ish million acquisition. We got a gong here. We'd love to, you know. Yeah, it was about $630 million. It was preparing AI. There we go. Love that. Congratulations. Congratulations. as well at headquarters. Nice. Big deals.

You hit the gong. I love that. Of course. We're excited about those guys joining the company. Their founder, Michael Chime, super, super talented, and very happy to have them as part of Axon. Is the strategy with that deal to kind of diffuse the technology over everything you do, keep it as a separate business line, cross-sell it, how are you thinking about that? Yeah, so our mission is to protect life. We want to keep people from getting killed in policing incidents to include.

include police officers. And the best way we can do that is shrink the time between the incident taking place and help being on the way. Right now, that's like a two-minute cycle. Someone calls 911. They talk to a call-taker. Call-taker talks to the dispatcher. Dispatcher figures out the right officer to send to the scene. And we think we can automate that entire process in partnership with prepared. So essentially, call 911, based on the metadata of the call, has already taken off to head to the incident. Over time, we think we can route the proper police officers to those scenes without a lot of human intervention. And so that's a lot of what we have to look forward to with prepared moving into the future here. You guys are at the stage where you're a true

platform, multi-product. I'm sure you started with, you know, I don't know the exact, the first product, But in this category, how hard is it to get point solutions through? I'm at, you know, if you're just building for other startups, like, you know, a typical YC company, for example, they can build a really narrow, simple point solution, sell that into somebody in their batch. It's pretty easy. You know, these aren't, you know, super sophisticated operations. Whereas selling into police departments, I imagine just the timeline alone of building that relationship is probably really hard for a young company.

I guess like talk about the different maybe stages of the companies and getting to the point where you're truly a platform. Sure. The biggest part is, you know, you have to recognize that U.S. public safety is less than a million total users. So a lot of startups wouldn't even be able to get funding based on that tam, you know. And so when you go into public safety or defense tech or, you know, this segment, you really have to be confident, you can build something for the long term where you can win the

majority of the users on your platform. That's really the only way you can have a viable business model in this space. And so, you know, we look at this like, hey, you know, we start with the most talented teams we can find and then the most product synergy. And so with a company like prepared or one like FUSS that we acquired last year, we really believe we can help accelerate their growth by plugging them into our ecosystem. And, you know, the products, nicely play off each other. And, you know, with police, they want to outsource as much of the technology as they can. They don't want to build things internally to, you know, to capture their use cases. And so when we can show up with a really nice end-to-end workflow from what we call

captured a courtroom, that really has a lot of value to our customers. And so that's part of our thinking and hypothesis as we look at M&A and which companies to partner with. In terms of that platform, Is there demand from local law enforcement agencies for you to play nice with other companies, build some sort of open API? Is there another kind of enterprise resource planning backbone that they want you to plug into? What's kind of the software supply chain of a modern police department look like these days? Yeah, it's a great question. And the short answer is, yes, you want to build something.

that's open that at the end of the day, this isn't about trying to help people make more money or make more effective ads on the internet. This is about fighting crime and saving people's lives. So when you do those kind of sketchy things where people don't have access to your data or you don't, you know, you don't work well with others, you're ultimately undermining the police department's mission of protecting their communities. And so for us, we try to look at this like, hey, we want to be the Switzerland in this space. And regardless of the hardware, inputs or the software outputs, we want to play nice and leverage open APIs to help our customers achieve what they're sending out to do. What robotic form factors are you guys most bullish on

over the long run? I'm sure you leverage a lot of drones. Is there any other, we had Tony from DoorDash on this week who built a really cute robot for delivery, L4 autonomy. What are you guys thinking about are drones enough or other things on the steves we want so the reason i'm smiling is i thought that like i pitched our founder rick on this idea like two years ago to do this big hugging robot like the michelin man just imagine this thing kind of walks down the street it's like big euro six criminal and takes them into custody so people aren't as scared of robots i don't think that one's in our future but the thing i'm most excited about is bear hug ability to put less lethal

on drones. So you can think of a school shooting taking place. And if you have drones in nests in the roof of a school, there's no reason those things can't be equipped with taser technology. And they can fly autonomously. They have targeting software. For us, an end user, you know, a police department would always decide when to use force. That would not be autonomous. But this idea of finding ways where we can keep people safe in these mass shooting incidents before first responders ever get there, that's very interesting to us. And over the next few years, we hope to be playing in that space more and more here with some tools that can really, you know, again, shrink the time frame where these incidents are happening from when they start to when they're concluded.

What about somebody in the chat is asking about technology for, you know, firefighters, specifically wildland firefighters were here in LA. As the fires unfolded at the beginning of this year, it felt like they were so, they had too little technology, at least that was the perception. We had firefighters relying on the nonprofit fire tracker that kept going down because they didn't even have capacity.

What are you guys doing on that front? It's a tough market because fires are so unpredictable, but certainly feels important. sure thing sure thing i think the most interesting thing i've seen in that space to date is these drones that carry thousands and thousands of pounds that can stay in the air for hours and you can imagine you can actually string those along uh with you know with very lengthy fire hoses and fly out into the fire and try to contain the fire from within like those types of uh improvements uh through the air i think will help i think in software

It's a little bit of what you said. It's predictive. It's based on the wind. I'm not sure that's like a major game changer. I think you've got to find a way to get water faster and more reliably into these fires. And it feels like drones over time, especially as they can carry more and more weight, will be a reasonable kind of creative way to help there. What are your current thinking around facial recognition, obviously image generation or generative

AI is getting so much better. But there's different parties. How are police feeling about it? How are different communities feel about it? Does it differ from township and city to city? How are lawmakers thinking about the tradeoffs there? It seems like one of those conversations that's like forever ongoing. And maybe that's the best process in a democracy, but how are you thinking about it? Yeah, I think you characterize that well. You know, the United States is one of the few markets right now where facial recognition is not used in policing. And we have, you haven't participated in that market yet, I do think we're more interested in it now. The thing we've been waiting for is to be able to demonstrably show that the algorithms are rid of racial bias.

I think that is certainly something that is worth making sure that you got that right. And ultimately, you know, up until now, we haven't felt comfortable with the technology. But it is improving and we're seeing those improvements. And so as we start to kick the tires on what this looks like ethical design principles and, you know, like we say internally, uh, helping make the right things easier and the wrong things harder, but that's going to be kind of at the center of our thinking as we, as we think about facial recognition on fixed cameras, on body cameras, and so forth. Yeah, where's the, uh, who has ball control on that narrative? Uh, someone in the chat was asking about this documentary, all light everywhere, very small box office. I hadn't heard of it

until this person mentioned it. Are there other communities or nonprofits? I mean, we talk about AI all day, and there's this guy, L.Azer Utikowski, who's kind of defined like the AI doom scenario of like super takeoff. And there's a couple thought leaders. The Dwar Keshe Patel show is talking about

tradeoffs and AI building. Who are the key thought leaders, influencers, groups, public groups, nonprofits that are actually defining the conversation there? Sure. I think a big piece of it on the policing side is major county, major county sheriffs and major city chiefs. Those are the large organizations that, you know, that kind of paint the path forward on what the priorities are for public

safety. But I think ultimately it's going to be very political. Like we need to get to a place where mayors and city counselors and all, you know, the downstream folks, they serve, the community is looking at this saying, hey, there's, there's far, far, far more good than harm here in this technology and it's progressing, albeit slowly, but I do think we're getting closer to the point where we can say, hey, look, you know, there's everybody should be interested in equipping public safety with ways that you can reliably fight crime and keep constituents safe. And between mayors and city councils and the police chiefs and sheriffs, I think that's the key combination of folks coming together on this one. What about on the hardware side, it had, it,

It feels like that piece of the business is pretty stable, but is there anything interesting happening that you can share on, you know, there's this big American dynamism movement, make stuff in America, there's a trade war, like, has anything shifted your thinking about manufacturing just hardware devices in America or for, you know, your business over the past few years? Yeah, for sure. And as a company that's sold to not only U.S. public safety, but the federal government and federal law enforcement for a long time, we've always manufactured in the United States. So Scottsdale, Arizona is where our headquarters is.

Hopefully it will be in the future. We've got a lot of drama around that right now. But we can get past that. We'll keep building tasers and body cams in Scottsdale. And we're proud of that. You know, we're supporters of the Buy American Act. We think it's in the best interest of the company to have all of our defense equipment made here. And so we're happy to participate in that process.

How crazy valuable is an individual person in the manufacturing process? There's been this debate going on in Silicon Valley about process knowledge, this idea that, I don't know if you've been tracking this, but Mark Zuckerberg has been paying hundreds of millions of dollars to people that know how to train just a very specific. set of skills and how you train a large language model. And once they know that information, once they know how to do it, they're worth so much because they understand how to do it and no one else does. And there's been this debate about, well, could we do the same thing with semiconductors and bring some people over from Taiwan and know how to make that, make it in Arizona where their TSM's already set up? But does the same thing exist in the world of like hardware

manufacturing? Is that even something that exists or is it more diffuse? It exists. It exists a some extent, I think, you know, for us, all of our cartridges, the things that go in the tasers, that have the darts in them, that fire out, those are all made by robots. They're all automated at this point. So, you know, we make millions of those every year. So the idea of people kind of hand-winding darts and so forth, that gets expensive. And so robots are expensive too, but they scale really well. And so ultimately, we've really made a big bet on automation for taser manufacturing. And we think it's the right bet. We've had automation for years and years, but we continue to invest more in it. And frankly, the people who know how to do that really well, like you said,

they might not be, you know, as valuable as some of the folks on the cutting edge of LLMs right now, but they're very valuable to us. And got to make sure, you know, manufacturing automation is one of the things we continue to be really, really good at. Awesome. Well, thank you for the updates and come back on when you guys do your next. Maybe the next acquisition will hit the $1 billion mark, but always welcome in the 600 million range. Sorry for keeping you over time. Thanks so much for hopping on.

This is great. Congrats on all your success. You too. You too. You too. Talk to you soon. Great to meet you, Josh. Cheers.

You got to look up the Axon Market Cap stock chart. It's beautiful. It's a thing of beauty. Up 128,000 percent of, for all time. IPOed, according to Google, in 2001. And just went out of absolute time. It's a $55 billion company now.

Yeah. It's really big. Well, back in the startup world, Deish Jane is announcing Mosaic, an agentic video editor. I definitely want to give this a try. He says, in a world trending towards AI slop, create something real. There's no wait list. There's a public beta. We gotta pull up the video.

Let's pull up this video and see, is it slop or not slop? How are we feeling? Comment, mosaic, and you can get some credits. Let's watch this launch video. Mosaic. The Mosaic. The real test, was this built-in Mosaic? Or was this after effects or premiere?

I can see this being pretty good. Yeah. I think, so node-based workflows are really popular like this, where you wire things together, because then you can repeat the process very quickly by just dropping in a new base clip and all the process. So if you say, I want to use AI to remove the background, add subtitles, we have a workflow for this that we've defined with clips in code, but oftentimes it helps to just be able to have a node-based workflow that you can just like wire up and understand where

each piece goes to the next, overlay the text, DaVinci Resolve is doing this and a lot of other, like the next generation of Blender and Houdini have all moved to node-based workflows. Cinema 4D recently moved to a node-based workflow. Well, I like that. I thought it was a good video focused on the actual product, what you could do with it. It shows you how the UI works. I was worried because he was, the founder said, in the era of AI slop, and then it looked like, like, it was just going to be a bunch of clips of. No, no, no, no. This is very clearly stolen footage from Interstellar. I doubt that they licensed it, but that's fine for one second. Yeah, I think that it, it's

isn't it legal to do it use it for like yes it would be transformative it would be fair use yeah uh probably you know i wouldn't you know you gotta talk to a lawyer if you want to uh dd das is sharing some news from earlier a lot of people have already seen it probably opening i wrapped a 6.6 billion dollar uh share sale at 500 billion this wasn't primary capital this was employees uh selling into it the sale Diversify the short of the $10.3 billion that was authorized. But does that mean it was oversubscribed? Does that mean that 10 billion of demand came in? They went to the employees and said,

you can sell up to $10 billion, the employees held. Is that what happened? I want to dig into it. Okay. Yeah. So it says with insiders seeing this as a sign of employee confidence and continued investor demand.

I was hearing that people were getting cut down. Yeah. That could be a spin. Sure. By being like, employees don't want to sell. Yeah. But I think a lot of employees at 500 billion are probably, you know, happy, happy to convert a meaningful amount into liquidity. Just given, just they're smart.

They're running the comps. It's life-changing money. They can see, like, you know, I'm sure you can, you can still be very bullish on the business and want to sell even half of your position, right? At 500 billion. You know, many people weren't underwriting that when they joined the company. So it's $8.5 million per employee on average, apparently. And so D.D. says, thanks, chat, GPD for absolutely destroying the SF real estate market. So even after taxes, these employees are all going to be sitting there being like it's time to hit market buy on Zillow.

Will they be buying in San Francisco? They might be renting in San Francisco and buying an inclined village getting those Nevada taxes going. I don't know. They're in office. They might be buying, they might be buying in Menlo Park, you know, Atherton. Yeah, Bay Area Broadly. They might be, yeah. Bay Area broadly. Yeah, the challenge is like, you know, hundreds of employees sold into this. And if you look at how many nice single family homes that are moving ready are available, I'm sure it's in the, you know, less than the number of employees, right? Yep. If everybody rushes to buy a home, it's going to really spike prices. It doesn't feel like it

happened yet necessarily, right? It feels like there was a time when SF real estate was really hot and then it sort of sold off, I believe, during the pandemic. You could buy a you could buy like a two-bedroom condo and SF, you know, not in the nicest area, but for like

600K. But that was down from like really elevated prices, right? Totally. And so we went, we crashed and now we're predicting a resurgence. And I mean, it'd It feels like it will come back. I mean, San Francisco has been, you know, on a tear in a bunch of different ways. It seems like it's more like a better place to actually settle down.

But still, there's always the pull of South Bay of what's up. Yeah, it's interesting because it still looks reasonable, maybe because we're in L.A. Obviously, at home prices are crazy here too. But you can get a four bedroom for $1.8 million. That seems reasonable. Take the money and run to the Ozarks. Did you see this story? First, let me tell you about Fall,

the generative media platform for developers, the world's best generative image, video, and audio models all in one place, develop and fine-tuned models with serverless GPUs and on-demand clusters. There's a very cool video that we are featured in that was generated on all that we will show later in the show.

But we only have five minutes until our next guest joins. But the article from Performative Mail says, This is the biggest toughest article I've done about so-called AI psychosis. It's the story of a man who committed a horrific crime in his youth, but served his time and against all odds found love and a new life, one that swiftly unrattled after he started talking to a chatbot.

He grew obsessed with an AI chatbot. Then he vanished into the Ozarks. What a wild story. John Gantz submitted a terrible crime. Very dark. Always tough to read into these anecdotes and see like how widespread is this. Would this have happened without AI? Would this person be deranged by something else?

A book. Did you get one-shotted by Walden? The book about going in living by a lake? Who knows? But you can go read it on Rolling Stone if you want to dig in to that piece. Petrini says as highlighting or quoting one of his own posts saying Korea's financial news is so ridiculously lit.

It's a bunch of sad ants crying and the Samsung chart just got down. Oh, that's hilarious. And Cetrini says this article was posted like 24 hours before the absolute bottom for Samsung. The ants are very happy now. Let's go.

The ants are so back. I have never been a Samsung shareholder. No, but Samsung is, I mean, they did that big deal with Elon for Tesla AI chips, I believe. So, you know, they're certainly getting it on the AI boom. Well, econ kids will see this and think I should give this guy money to help me get rich. And it's two of the most crazy-looking individuals.

I don't recognize either of these. This is the guy on the left is a rapper. What's his name? Oh, yes. He was the basis of spring breakers. This is riff-raff. Riff-Raff. This is Rif-Raff.

I don't know that he's actually done a get-rich quick scheme or like a hustle. Now, there are some guys out there that look like that. somewhat like this. It's a very evocative post by Rob Frund. We haven't seen a course bro just sell a course called How to Make Money, just reasonably priced and just really breaking down, you know, here are the different ways that- It's Econ 101, baby.

Tyler Town, conversations with Tyler. You can sell services yourself. Yep. You can sell services that you deliver through a team. You can buy something for one price and sell it for a higher price. Yep. and that's pretty much it. Yeah, Naval kind of one-shot at that entire category

with the How to Get Rich without getting Lucky Thread. He should have dropped a court. He didn't need to because he summed it all up in a series of like 20 pithy tweets and you can just scroll and be like, yep, that's the algorithm. Yep. Anyway, mounted cash flow says, always exciting when you see a new letter

get added to the end of EBITDA. What is he talking about? What's the latest one? I did a sale. Community adjusted was the classic. Community adjusted. What was that? C-A-E-B-D-D-A-E-D-A-C-A, I guess?

I bet-D-K-K. I don't know. I don't know if there's a new one that's going around. Because the AI companies aren't even talking about EB-D-D-A-E-D-A-E-D-A-E-D-A-G. It's all A-R-R-R-R-R-Is and just top-line. Invita-V-E-V-E-T-E-A-KKUKUKUKUKU. Yeah, so they haven't had to dip into some crazy EB-D-D-A calculation. I don't know who is, but stay alert.

You get to talk about cash flow. Yeah, and if you're diving into EBIT numbers, head over to public.com. Investing for those that take it seriously. Multi-asset investing, industry leading yields, and they're trusted by millions, folks. Adam, person of swag says you can just do things. Option on the left. Quietly announced a $2,200 million Series A in a tweet saying back to work. And the second option, get black hat drunk at cheese candy after closing 120K precede.

a little specific there, Adam. They think those are both great, great options. One more post from Bucco Capital Bloke. Most people think investing makes you rich, but that's wrong until you're making at least 500K a year. The best RY is growing your paycheck, get the cash flowing first. Totally agree. We had this conversation with the team,

no matter how much, how excited you are about investing when you're young, focus on developing the skill set that allows you to be more valuable. to your market. Without further ado, I think we should bring in our next guest, a man that should need no introduction. Pat Gelsinger. He's in the Restream waiting room.

We will bring Pat into the TVP and Ultramm. Welcome to the show, Pat. Great to meet you. So good to be able to join. Thank you. Thank you so much. We're so excited. I would love a little bit to catch everyone up on what you're doing today.

take us through the current project, the current business, and what you've been focusing on over the last few years? Yeah, thanks, John. Thanks, Jordi. So since Intel, I've been focusing, you know, I've been wearing two hats. One is the deep tech investing hat with playgrounds, so the hard stuff, you know, quantum computer, superconducting, next generation light sources, new materials, you know, those types of things. And then my other hat is being the head of technology for glue, a faith tech company. I've lived my life at the intersection of faith and technology. And glue is building the platform for the faith and flourishing ecosystem, you know, from running IT all the way up through the latest and greatest trained models for AI services, you know, for the church and the faith ecosystem.

Yeah, I'm fascinated by glue. We've talked to the founder of Hollow on the show, and it feels like an industry that is super ready to adopt technology, and yet there just aren't that many founders that are pursuing it. I would love to talk more about some of the findings that you put out when you investigated the foundation models. How did you process that information? What was the methodology for evaluating these models? And then I'm sure we can dig into more about the consequences of those findings. Yeah, you know, and maybe the first point is, you know, I've been driving benchmarks for 40 years. Yeah, that's right. You know, my name, you know, right, in fact, some of the benchmarks that are still used today have my code in them.

Yeah. I was sort of like, it's sort of embarrassing. Like, you get a move. It's amazing. Yeah, you should have moved on by now. But I, yeah, my code is still there. So I've been deeply associated with, okay, if you can't measure it, you can't manage it. So, you know, this idea of measuring AI models and, you know, people are working on this today, but, you know, we're also seeing that they're just measuring performance or cost per first, you know, time to first token, cost per token, right? You know, total token throughput. Those are great metrics, but it doesn't measure.

Is it good? Right? And to be, you know, the absence of bad does not demonstrate the presence of good. So what we did, right? The second point is, you know, there's this body of work that was fairly recently released by Harvard, Baylor, and Gallup measuring human flourishing, a five-year Lilly study that across 22 nations, across 50-plus cultural groups, what is human flourishing? So now we can answer fairly definitively across multiple dimensions, across, you know, relationships, character development, you know, finances, spirituality. and faith, is it good? So now we have, you know, a solid view for that. And then the final piece is, okay, let's create benchmarks for AI systems using that foundational research. And now we have, you know,

a corpus of qualitative and quantitative results to measure good. So if you say, which is the better model for human flourishing, open AI, you know, Gemini, Anthropic, you know, Claude, I can give you an answer based on a solid corpus of data. with the most rigorous of, you know, benchmarking experiences and numerical analysis sitting behind it. And we believe, and our objective is, as the result of that, is we will make models better because they can start measuring not just first token time, but did I give a good answer that's aligned with human values? How do you think about that benchmark of human flourishing? Like, when you go through your life, are you thinking about your...

finances, character, happiness, relationships, meaning, faith, health. Are you consciously checking in with yourself, or is this something that's more reflective on your life where it all kind of rolls up into happiness or well-being? And then when you investigate your life, you realize that those are the things that you've done. How intentional have you been, I guess? Yeah, you know, I'd say it's sort of in two different, you know, dimensions there. I mean, hey, I don't go benchmarking myself all the time. But, you know, that said, you know, every time a major new model comes out, I want to ask the question, is it better? Yeah. And on these dimensions, so, you know, we'll be producing these benchmarks on a periodic basis every couple of weeks. You know,

we'll be updating those benchmarks to say, okay, are we better? You know, which model is better? is Deepseek better than OpenAI than OSS? And so I want that rigor to be in place that everybody is, you know, okay, you know, are we improving in this area? Because you don't fix these things overnight. You know, it's a long trajectory of, you know, evolution and development of the technologies. You know, secondly. Oh, yeah.

You know, and then the second side is, okay, you know, you'll hear disastrous situations. where we have kids committing suicide, their best friend has become the chatbot. Wow, when those things come up, would our benchmarks be detecting those? And then do we have to add more and make them more rigorous that those kind of things can now be avoided in a constructed way that we really are happy

that we are making technology better or the language I like to use, technology as a force for good. technology itself is neutral. Are we going to bend that arc toward good? How receptive have the labs been to the benchmark? You know, I'll say it's early at this point. And, you know, so, you know, we're interacting with all the names that you would like.

And I'll say, you know, they're getting better. You know, their scores have improved somewhat. You know, so I do think we're having some influence with them, but it's still early. And, you know, as I would like to, you know, also say, I don't think the benchmarks are rigorous enough yet. You know, we got to, you know, we have sort of, you know, an early version of them out there. These are good, you know, 1,230 questions or so are part of them as well. But I expect that becomes a bigger corpus, right, of questions. You know, also, we got to go deeper, you know, for the most part, these are single shot, you know, questions.

We got to get into multi-turn conversations. We have to, you know, have more. of a mixture of experts, you know, that you're testing different experts. Also, how do we test the safeguards, right? You know, we're now at the sixth turn of a conversation. You know, this individual was clearly suicidal. You know, how do we go handle that? Yeah, I think a lot of the safety concerns that at least we have is when people are, you know, many, many, many, many prompts deep, right? In a, in a specific conversation, it could be thousands, right? Deps to which that personally I've never gone, right?

I don't know how crazy it gets that far, but certainly the labs need to be thinking about all those extreme edge cases. Yeah. Yeah. So I think we have more work to do, you know, I'll say multi-turn conversations and how we build more of that robustness into the benchmarks. But, you know, what we have today

and you can go to the GLEW website, FAI, right, you know, and see what we have today. You know, this is a good start. But, you know, I do think, the speed of evolution of the AI technologies, we have a lot more work to do. And I'm sort of excited now as we get some of that momentum underway, you know, really start crawling into, you know, just like you said, more of the multi-turn conversations. And we can be pretty aggressive in the questioning sets for multi-turn to drive some of those

extreme behaviors, you know, because, you know, the first few questions is mostly driven by prompts, right? You know, as you get deeper into, it's much more reflecting what's the underlying. model weights, you know, themselves. So all of that, we have more work to do, but I'm excited to have this first version out there. We're starting to get some impact and good results. Yeah, there also seems to be some sort of unresolved question around alignment by default. How much do the models just reflect you? If you show up with a lot of negative energy, they can kind of become negative, and it feels like it's the self-reinforcing spiral. If you come to it just asking interesting questions about the world and wanting to understand how math works.

Like, you're going to get a tutor if you come to it with something bad. So a lot to sort out there. I have a question. One of my favorite quotes from you is from this talk you gave where you sort of lightly accused Silicon Valley to be a group of miserly pagans. It's a great quote. I think it's a very apt analysis. And when I look at the performance scorecard on glue.com slash fluris

Hub research. I noticed that the models seem to be doing very well at finances and very poor at faith. And I'm wondering if that's a reflection of the humans that train the model or the data set that's going into the model or both or how intractable is that problem. Yeah. You know, and I think that's great because, you know, I mean, where's a lot of the training data coming from? Yeah. Right. You know, associated with that, you know, and I view it as read it. Hey, you put bad stuff in, you're going to get bad stuff out. You know, you put good stuff in. You're going to get good stuff out, you know, for it. And, you know, I mean, we're doing some work, you know, we glue, you know, we'll have our own values aligned models, right? And part of the reason we're going to do our own foundational training, you know, for those models is specifically, I'm not going to put the bad stuff in.

Yeah, that's good. I don't have to worry about it ever coming out because it was never in. so that we can never probabilistically producing those things. So, you know, but, you know, getting more specifically to your question, I think a lot of it does have to do with the implications, right, of those training sets in the first place. You know, whenever, you know, you've seen lots of conversations about the challenges of we don't have enough training data. You know, we don't have good training data.

you know, the quality of synthesized data, you know, versus raw data, you know, associated with it. So I do think there's a valid problem, you know, and, you know, I'm always a bit optimistic on the individuals that their motives are good for it, you know, but we've clearly seen in the social networks, you know, the algorithms were bad, right? You know, it leads to more degrange behavior, right? you know, it's only about dwell time on sites. You know, and I think those same motivations would, you know, naturally drive the AI models if there wasn't further accountability. Yeah.

And that's what leads us to saying we have to have better benchmarks that are rigorous and values based that get the answers we want. Well, yeah, even with with human-driven social media and, and how it can take people down these extreme rabbit holes, that they're not as worrisome as LLMs because LLMs can be totally private. They can go to these kind of dark places, and it's just one human interacting with the model. Whereas social media, at least if somebody's on some crazy forum,

there is a point at which they can say something that goes too far. People can say, hey, I think you should go out and touch grass, right? And I haven't seen a model tell somebody to touch grass yet. maybe that's built in. Maybe that'll come to the next version. Somewhere. Did you always predict, you know, it feels over the last few years, Silicon Valley has a, you know, massively renewed interest in faith and religion. How much did you anticipate that?

Clearly, you know, you never cared what the industry thought you had your own faith, but is that something that you predicted? Did it come faster or sooner? Yeah. Yeah, maybe three different dimensions, you know, there. You know, I mean, one is there is, I'll call it a post-COVID and a post-success renewal, right, where, you know, this is very endemic of Silicon Valley, but you see it in lots of other, you know, communities, New York, Austin as well. You know, people presume success. And now they're asking deeper questions like, is it significant? So, you know, there is this national, right, you know, shift in what's important. Some of that might

have been driven and some suggest it was driven by COVID isolation. You know, some of it is driven by the success phenomena, you know, that we're in, you know, but it isn't just Silicon Valley. You know, there's a number of statistics, you know, those that, you know, believe in a faith view of Jesus Christ, the church engagement or spiritual engagement, you know, we're We're seeing very much, you know, 20 to 35-year-olds, you know, is going up in very meaningful ways, you know, four data points in a row of rising, engagement or spiritual value engagement. So it's a national phenomenon at this point. Clearly in Silicon Valley, we've seen a greater willingness to talk about faith. You know, I started an organization about 12 years ago called Transforming the Bay with Christ.

one of my other hobbies. And that organization now is over 800 churches strong, right? When I talk to people, they didn't believe we had eight churches in the Bay Area. I have over 800, you know, participating in transforming the Bay. And I'll just say there is a renewal, a connectivity, you know, associated with that. You've seen other things recently, you know, some news and, commentary about Act 17

Dre and Michelle Stevens and Peter Thiel and Gary Tien friends of mine that all of a sudden we're hosting these events they're sold out events where hundreds of people are showing up asking life questions

some of us hey I want to come and hear Pat speak I want to come and hear Peter Thiel speak but some of it is they want to have life questions from someone they respect. So I'll just say, you know, there is this renewal going on. You know, and personally, I've been at it so long. People often say, well, do you feel more comfortable? Well, you know, I felt comfortably uncomfortable 30 years ago to speak about my faith in the Bay Area. You know, 30 years later, well, I've been doing so long, everybody expects me to do it.

So it's not as uncomfortable, you know, but in many regards, you know, it's like, it's who I am, right? And if you have a different faith or a different worldview, you know, everybody should be comfortable talking about that. And I'd say from my faith perspective, it's, you know, like I spoke to a Baha'i womanly recently. I don't know anything about Baha'i. Tell me about Baha'i, right? And the more I engage you where your worldview is, the more freedom I have to talk about my Christian worldview and why I think that's good, powerful and meaningful as well. What do you make of,

it feels like an easy prediction to make that there will be religions built around AI. It feels like they're already emerging. If you go into the Reddit forum on chat GPT, you have people that worship 4-0. It feels scary. Even in technology circles, people kind of joke, oh, we're building AI God

or God in a God. box. It's going to be so power. It's super intelligence. It'll be above humans in some way. And it's a little bit tongue in cheek with some people, but some people seem to believe it. And I think there's a long history of Silicon Valley and even parts of the Bay Area trying to recreate religion from first principles. I have, you know, my family has been in, you know, Palo Alto in some way or another since the early 1900s. And, I've, heard stories of like cults emerging around Stanford in the 60s where people were basically creating a club that looked like a church. They would meet like a church, but ultimately we're

just basically trying to create new religion. And I think these, yeah, these conversations are uncomfortable, but they're important because they're, I think it's inevitable that they're going to pop up and it feels like, you know, potentially can go to a very dark place. Yeah. And I'd say, you know, maybe three different comments there. One is, you know, technology is neutral. You can use it for good and bad, right? AI, most powerful technology, you know, that we've created yet. And no small part because it's built on the shoulders of all the other technologies, you know, connectivity, cloud, et cetera, right? So it can grow and expand more rapidly. But our job is to make it good, right? Can it get used for cults and other things?

Of course, right? Just like everything, you know, what was the early driving use case of the internet, right? I think the number one websites early on were pornography. Not what we intended, but you know, gambling and pornography is like, oh, man, you know, man, what have you done? Right. So I have no doubt, no hesitation that there will be, you know, these cultish experiences that emerge, right? But again, our job is make technology a force for good. And where and how do we build safeguards and other into it? Second general comment is, you know, people talk about super intelligence and those, you know, and some of it is, you know, it's like, man, you know, the calculator is a superintelligence. It does math and better than I do. And, you know, my, you know, my, Excel spreadsheet, man, I ain't giving that up. And I don't want my accountant to give it up. Right? You know, and so on. And, you know, we think about LLNs and the futures, man, they're going to become indispensable, right? You know, because it is like the real-time, contextually relevant encyclopedia of my life. Right? You know, it's got a better memory, better knowledge,

better connectivity. I mean, we're going to use AI to conquer language for the first time in humanity's history. You know, I'm going to be able to teach every kid in their native language contextually relevant using the power of AI. And because of that, you know, almost all the people that live in poverty live in the fringe languages that are not conquered. Wow, you know, how exciting, you know, I mean, the extraordinary opportunity. these, you know, this is going to give us to truly improve the life of the planet. But, you know, ultimately... Yeah, it's been interesting.

I've been debating with John how, you know, how bad would it be if I suddenly didn't have access to LLMs? And we've gone back and forth on this. I personally would much rather live without LLMs than a mobile device or the internet, right? I think some of these enabling technologies that we have that make LLMs valuable and useful and widely available are feel today more foundationally important, but that is coming from a viewpoint where the most of the Internet is English, most of the media that I want to watch is English. And so the point you made about, you know, being from somewhere in the world that English isn't your native language,

being able to educate yourself on, you know, and watch any content, you know, things like that are pretty exciting. Where do you think? Continuing and going back to you the core of your question, though, you know, will there be AI gods, right? God designed us as humans to be in relationship. And any technology that's driving us apart, you know, three teenagers sitting on the couch texting each other, right?

How terrible. Talk to each other, right? God designed us to be relational individuals and, you know, a foundation of religion, spirituality, you know, isn't just the self. It's the community, right, that you're part of. So, you know, I fundamentally, you know, things that take us apart, you know, I think, okay, we have to build safeguards to have them be tools bringing us together. You know, I do not want my AI doing suicide prevention cowardsline. I want a real human doing those things. You know, I want a real human interacting with kids and teaching and learning. You know, I want AI agents that are helping the teacher do better education, right? You know, because, you know, of the modalities, the training and so on. But ultimately, these are very, you know, powerful, right, human experiences that we need to keep as human experiences. How have you been processing the dialogue around the Antichrist?

You mentioned Acts 17, Act 17's in Peter Thiel. It's something that's really sparked a huge discourse in Silicon Valley. I think a lot of people weren't even thinking about Christianity is somewhat unique in the idea of the Antichrist, but how have you processed it through your life and then how are you thinking about that concept today? Yeah. You know, I do think there's, you know, there was, you know, when I was in my, you know, whatever, 20s or 30s, there was a series called Lessons. behind.

It was like, oh, you know, and if you go back and read that, you know, and the late great planet, you know, there's sort of been this episodic flow of end times discussions and discussions on the Antichrist and what they would be. So, you know, this topic has, you know, only persisted for about 2,000 years. You know, it just sort of goes way. It's Lendy. It's been around forever. Yeah, yeah. So, you know, hey, having another cycle of that, great, because it causes people to ask deep questions.

Right? You know, to me, that is a good side effect to it. Now, from a biblical scholarship perspective, you know, Antichrist has mentioned one time in the Bible and not contextually in revelation. Right? It's like, okay, this is a thin discussion. Right? If you look at it theologically, you know, that way. You know, when most of the revelation and, you know, apocryphical conversations in the Bible are actually much simpler, right? You know, people of those days were looking at they were seeing the Antichrist and his name was Caesar, right, you know, in that sense. And, you know, the Bible was written to be, you know, I'll say eternal, but current at the same time. So I think a lot of this, it's this episodic flow of excitement on the topic, but I think a theological study of it would say, calm down a little bit, right? You know, we are living in the end days, meaning that, you know, right, you know, there isn't another expression of Jesus Christ coming until the end of time, you know, it's seen through scripture. You know, and because of that,

you know, hey, if this causes people to be more interested in religion, yes. But at the end of the day, let's make sure we have good theology behind the conversation as well. What about millinarian thinking generally, apocalyptic thinking generally? It feels like maybe I'm out of touch with what the vibe was in the semiconductor industry when you were active in it, but it feels like the semiconductor companies have not historically said, if we don't build this, the world will end. But we've seen a rise of entrepreneurs who have been talking about apocalyptic consequences to either building or not building what they're working on.

And it feels like it's more in the zeitgeist. It's kind of the final conversation you can have about the impact of a technology. And so far, I mean, it's not just the AI folks. Even Elon Musk talks about the consequences of not becoming multi-planetary being apocalyptic potentially. And so we have to explore the stars, which I completely agree with.

I love space exploration. And so I'm wondering about that as a rhetorical tool in entrepreneurship. Is that something that you've noticed? is on the uptick, or has it always been this way? Well, again, I think there's an episodic nature to it. You know, and let's, you know, when, you know, go read some of the press when Sputnik flew through the sky.

Yeah. The Russians hear everything, right? Yeah, they can read my mind. I just saw it in the sky last night, right? You know, you know, so my father, you know, talked to me about when he saw Sputnik, in the night sky, right? You know, so, right, you know, and I, so I do think there's somewhat of episodic nature. And when you hit these huge inflection points, right, like AI, that is just exploding in capacity

and capability, I think, okay, we're in the next episodic discussion, right, as a result. You know, but I think if you look over history, we've seen that, you know, go through over time, you know, in the first days of the internet, you know, myself, phone is going to, you know, right, turn my brain to jello and, you know, all of these other kind of things. You know, I just don't get too ex. We might have a technical issue. Let's give just a second. Hopefully we can get back. Pat down, the deep state took down the, it was getting too real. That was a fantastic interview. We were enjoying that. Hopefully we can bring back Pat Gelsinger. I didn't get to ask him about AI CAPEX.

we'll work on getting him back in the meantime let me tell you about turbo puffer search every bite serverless vector and full text search built from first principles on object storage fast 10x cheaper and extremely scalable uh head over to turbo puffer and we have pat gelson here back sorry about that aGI is here but the internet still doesn't work sometimes yeah i always get a kick out of this when the cell network, here, let me expose this as well. Yeah, we are. The cell network or the internet, you know, drops the Silicon Valley. Oh, yeah.

What's going on? What's going on? This is a place we should get this right. I don't know exactly where we got cut off, but I wanted to ask you about there's been so many insane headlines this year around AI, AI CapEx, AI talent wars. you know, rewinding in your career in the 90s, early 2000s, like what parallels are you personally drawing? Yeah, you know, I do think, let me give two different answers to that first. Because, you know, I mean, you know, people are talking about these one, five, ten gigawatt

data centers, right? Every time somebody says gigawatt, the image I want in your mind is a nuclear reactor. Because one nuclear reactor is about to gigawatt of power capacity. How many nuclear reactors has the U.S. built in the last 25 years? I think it's one or three at Vodal, right? Yeah. Remember, China, I think China has 60 under construction today. Right?

And fundamentally, energy capacity, you know, we sort of lost a decade. right, you know, as we were so focused on, right, you know, climate that we weren't building our energy capacity and renewables. Hey, I'm the biggest believer, but for the most part, they're not economic, right? There's a lot of work going into making them economic. You know, some of the playground companies are working on that. Yeah, but for those part, they weren't economic and they didn't add that much to the grid, right? Yeah. So there's this craziness going on, right?

of that cap-backs expectation and so on, you don't have the energy to do it. Now, obviously, we're working on dramatic improvements to make inferencing 10 or 100 times more power effective, right? Because ultimately, you know, I think the demand for inferencing capacity is the ion infinite, right, to Jevin's Law or the gas law. You know, it really should be permeating every aspect of what we do. going forward. But some of the crazy predictions associated with this, you know, it's not going to be chip

limited. It's not going to be land limited. It's not going to be capital limited. It'll be energy limited going forward. And I do think it's inducing a lot of investment into the energy systems, which I view is very positive because our long-term economic growth as a nation will be energy limited. So even if we're, quote, overbuilding capacity, I view it as a super good thing. Yeah, right?

Because, hey, whether I need that for my electrification, whether I need it, you know, for my quantum computers, whether I need it, you know, for... Just make my power bill cheaper. If you have more energy to go around, the price is going to fall for average Americans. That's a win. Yeah. You know, I mean, the oil crisis. Yeah.

Great. You know, it can become the electrification crisis in the future, right? if I don't. So I just view that as such a good thing. Yeah. Because if so, mismanaged our nation's energy capacity going forward. Now, you know, some of these forecasts are just nutty. It's a little like that. It's 100 nuclear power plants next year.

Who knows? Yeah, I mean, we did have an IPO today, Bermi, that is planning to make a nuclear reactor by 2038 or something. Like they raise his money to try to bring it online. So I was poking a little bit of fun at that because the company's nine months old. They're valued at $17 billion. But maybe it will be net good if we can bring a new nuclear plan or two online. Yeah.

And, you know, by the way, you know, I just became the general partner lead on an investment that we're doing in the nuclear space at playground. So I'm learning a lot about this. It's great. And a lot of the small nuclear reactors, they're not economic, right? You can't connect them to the grid. You can't scale them, you know, the regulatory thing. So, you know, I do think it's a very fundamental problem, you know, for the nation. And one, hey, I'm now diving into that problem as well.

So, you know, we need to bring on many gigawatts of additional capacity for, you know, the future of our nation. Yeah. Now, you know, some of the other, you know, projections there, you know, that people, you know, point to, I think they're way over their skis, right? You know, not everybody is going to add 10 gigawatts of additional capacity. And I am excited that, you know, like one of our company's snow cap, you know, we're out to make AI a hundred times more power efficient than it is today. Not 10%, not 10x, 100x. That's great.

Right, you know, as we move to superconducting. So instead of that gigawatt, you know, I'm going to have a hundred. 100 megawatts of cryogenic capacity that produces 10 times more inferencing. Okay, that's industry reshaping. Those are the kind of things that are truly going to, you know, say do a better job of the energy capacity we have and, right, you know, creating more ultimate capacity for the future. And what about, you know, some of the vendor financing going on,

some of the demand guarantees that people have been making? does that immediately worrisome to you because of some of the negative events that came after, things like that in the dot-com era, or do you think it's natural part of the cycle? We've seen this before. You know, look at all the optical companies, you know, in the early days of the Internet, right? Well, you know, everything is up and to the right, you know, for, you know, exponential growth for unlimited periods of time. You know, okay. Yeah.

I mean, it was a, it was a stunning 10 years, but it didn't extrapolate to 30 years. Yeah. Right. And I see most of the AI, I think, you know, I see, you know, fundamentally no change for the next two, three, four years. But by the end of the decade, some of these transformational technologies will be at scale. So I do see a real shift, right, in them before the end of the decade, particularly on inferencing costs, imprinting performance, and power consumption for imprinting, which is sort of like the big one for a broad deployment of AI.

How do you think about, I mean, there's a lot of founders in our audience who are, you know, it's raining right now and they're happy and everything's good, but you've been through so many market cycles. Is there a story or advice that you give to entrepreneurs or founders who are going through a reset in going through a bear market. How do you think about getting through those hard times? Well, you know, so they're going to come, right? Right now, you know, as a CEO, right? Your job is the high is never as high and the low was never as low. Right. You know, your job is to sort of, you know, cut the tops off of the crazy highs and, you know, manage through the crazy lows, right? That's your job.

And if you're not ready to do that, you know, if you're the, you know, the super enthusiastic champion that everything's up and to the right, you know, for the next two, three decades, you shouldn't be in the CEO's job, right? That's called a cheerleader, not a CEO. And, you know, we have way too many companies that are led by cheerleaders as opposed to CEOs, right? And we can go back over history and identify a number of those. Also, you know, like I say, I think the next couple of years, I think are going to be just fine. But as we get to the end of the decade, okay, things are not going to be just fine. There's going to be fundamental shifts in the economics and the capacity. And a CEO's job is to get the balance sheet in place that they can navigate through those times. You get cash when you don't need it because when you need it, you can't get it. So you have to be managing thoughtfully about when those disruptions occur. So, you know, I do think, you know, particularly a couple of years out from now, okay, that's your job as the CEO, as the leader, as the entrepreneur. Do I have enough capacity to take a meaningfully negative time, right, and be able to go navigate through it?

What about team building leadership? How should you be communicating with potentially thousands of employees who are looking to the CEO for leadership, not just cheerleading? Well, you know, the first job of the leader is to represent reality, right? You know, where it is. You know, that is the first job. Here's where we are. Life is great, but it's not as good as that. You know, life is bad.

And here's the situation. And it's not as bad of that. First job of the leader being that point of truth, that point of reality. You know, as I say, you know, the leader's job is to. communicate the vision over and over and over again to the point that you are absolutely sick and tired of it, right? You know, you think there can't be a single person that hasn't heard you communicate the vision. And just about the time that you are absolutely dreadfully bored by it, it's the first time it's really getting through. Right? It was just that hard. And about the time

your organization knows, it usually one to two years. It takes another year or two for the customers, understand it. Yeah. Right. You know, and if you change, right, in less than three or four years, nobody understands it. Yeah. You know, so, right, that constancy of vision, purpose, mission is a super important thing, you know, for the leader, you know, preparing for the bad days, like we already touched on.

There will be bad days that, you know, you have to navigate through. You know, and then when you have success, you give it all to the organization. and when you have failure, you claim it all. In a transparent, you know, a facing way, you know, this, right? And that's how you build, you know, loyalty and commitment from your teams. You know, being a CEO, it's not cut out for everybody. Period. Stop.

You know, if you're not ready to take accountability, if you're not ready to make the hard calls, you know, if you're not ready to hire but also fire at the same time, if you're not ready to demand your organization to excellence, you shouldn't be a CEO. At the same time, if you're ready to do that, it is one of the most fulfilling, satisfying opportunities of a leader's career. It's fantastic. Well, very well said. Do you have time for one more question?

I know we're 10 minutes over. For you, absolutely. Thank you. Incredible. I'm really enjoying this. We were talking with Brett Taylor of Sierra about, kind of the SaaSpocalypse and how there's a lot of legacy enterprise software companies that are needing to react and respond and evolve as, you know, LLMs and generative AI sort of comes online.

And he said something I wanted to get your thoughts on and maybe some examples. But he believes it's easier to change your technology stack than your business model. And he was saying in the context of it's great. if you're a software company that's started and is charging based on value, but if you have seat-based pricing today and your competitors are now selling, you know, not trying to get you to scale headcount, but just trying to deliver more and more value, that can be tough. I wanted to ask if there was any notable kind of moments in your career where, you know, you witness a company, sort of be able to effectively evolve their business model,

and or on the technology side and how those two things play together in your view. Yeah, yeah. And, you know, first, you know, Brett, you know, just he's a world-class guy. Yeah. So if you have the chance to bring Brett on or Pat on, take Brett. Well, we've enjoyed both. Why not both? But Brett is great.

You know, the, you know, we took VMware through the transition from enterprise licensing to a SaaS business model. It is but ugly hard. Right? You know, a business model is like a deep rut, right? You know, one of my favorite posters on this was, you know, it showed a deeply runted Indian road and the caption read, right? You know, pick your rut carefully. You'll be in it for 100 kilometers.

Right. You know, and there's sort of this truth to a business model because your competitors get used to it. Your customers get used to it. your supply chain, you know, you just get funneled into it. And when you try to change your business model, man, it is dreadful, you know, the factual, the legal, the renegotiations, etc. The team. Yeah. Yeah. Everything is hard about it. You know, because, you build, you know, every aspect of the business around, you know, those, the characteristics of the business model you've chosen. So, you know, Brett is right on. Changing the technology is quite a bit easier than changing

the business model. Now, you know, the master stroke is if you can keep sort of that interface to the customer and the business model the same and do a major retooling, right, of the technology underneath it, okay, now you win, right? By the way, you know, that, you know, the first major success in the enterprise to SaaS conversion, you know, my, you know, friends at Adobe, they navigated through it. You know, I mean, that was, you know, their masterstroke and many others followed in their footsteps after it. But that was hard, you know, for the transition.

Now, as we go to AI, most are going to be an SaaS-like business model going forward. You know, that said, the user experience in an AI native application is dramatically different, right? You know, you know, and if we think about it, you know, when I, you know, very broad terms, the computer has, I have been adopting to the computer for the last 50 years. In an AI world, the computer adapts to me. Right. You know, if you think about the fundamental shift of that where now I am speaking to it on my terms, it's hearing me in my way. You know, it's understanding my face, my recognition, my thoughts, right, contextualizing them, you know,

in my language and understanding. So I do think the idea of just flipping to a SaaS AI business model, you know, versus an AI native experience that truly harnesses the full breadth of that power, I think many people as they talk about the AI conversion of their applications wholly underestimate what will be associated, right, with truly creating an AI native experience for the future. Everything changes.

Yeah, it's not just even business model going maybe from, seat-based and fixed contracts to value or selling the work, but it's even the entire user interface and the, like, what does the application look like? What does it feel like? Is it feel like hiring and managing a team? Does it feel like hitting buttons yourself? Yeah. So doing both of those things at the same time. Okay. Actually, last question. Is it true, or do you still have a VMware tattoo on your arm? So, no, I know. Okay. And it was so funny because when, you know, my... But you did at one point.

I did. Right. But it was like one of those 30-day tattoos. Okay. Okay. That's great. So, you know, and I came home from that trip and my wife, you know, somebody sends the video to my wife. Yeah.

Right? You have a tattoo artist and everything like that. And she says, you may not have that on your arm by the time I go on vacation with you. That's amazing. Yeah, people think tech people are hardcore today when they launch you like a video or something or they put up a billboard. Like the 30 day, living with your brand for 30 days is a new level. And we thank you for your service to the technology industry, Pat.

Thank you so much for coming on the show. This was fantastic. And thank you for the work that you're doing. It's very important. I'm very excited. I mean, we'd love to have you back, dive into all the different portfolio companies, also what's happening on the glue side in terms of actual productization. I know we spent a lot of time in the AI world and talking.

about the other labs. But there's so much more that we could go into. But we'll give you back the rest of your day. Have a great day. Thank you. This was a lot of fun. Thank you so much. Thanks so much. Talk to you soon. Cheers. Bye. Let's pull up this video of the meta rayband display. Okay. I will react to this. The meta rayband display has a new video that we are going to pull up in the meantime. Let me tell you about profound. You've heard that you can now buy products on chat GPT. You're going to need to get your brand mentioned in chat GPT. You can reach millions of consumers who are using AI to discover new products and brands. We did this just yesterday. We found some new microphone stands. I took a

picture of a microphone stand that I saw, dumped it in chat GPT. What is this? Wound up. I didn't check out actually in the chat GPT workflow, but we were pretty close to getting there and you're going to want your brand mentioned in chaty pt with profound. Let's watch the meta ray bands video right now. This is, so we've demoed this, both Jordi and I have looked at this. This is what the heads-up display looks like. It looks better when you're actually using your human eyes. instead of trying to film through it. But the navigation, the heads-up display, the Cod mini-map,

I think they should give it much more Call of Duty aesthetics if they wanted to really pop and go viral. Tyler, what's your reaction? Would you wear this? So is it like running through your phone at all? So I believe, we were debating this with Simon on the show when he was here, I believe that almost all of the compute is done on the phone

and it is serving up maybe like a small JSON file that it renders on the headset, or maybe it's even just streaming images. The bandwidth over Bluetooth has to be limited, and then the compute on the device is also limited. So I would imagine that it doesn't have GPS. It's not doing any on-device inference. It's really doing limited just,

this is the graphics that I need to show, render that graphics and pipe it in there, but then keep that battery life going as long as possible. Yeah, because, I mean, with these, These are the Oakley ones. And these, like, when you set up, you have to connect to your phone. That's, like, the very first thing you do. Yes, yes.

So these you definitely have to connect to your phone. They do not work without a phone. Over time, you would imagine that they try and bootstrap a true platform on top of this and say, hey, as soon as we have developer access to this, I'm going to add the sound effects. So using the live AI feature, you got, you walk outside, you're walking down the street. The glasses are, you know, monitoring the situation. And you're getting these kind of same. It would be electric, electric.

Bobby in the chat asked, why don't we use lav mics? Well, we were discussing it. It turns out lab mics are complicated. We're figuring it out. We're going to see some upgrades. Lab mics are coming. We want to be able to walk around. Enjoy the horse.

Yeah, enjoy the horse. Can we go to the horse cam? Do we have a horse cam today? I know I'm putting the production team on short notice with that one. Get the horse cam going. I mean, this thing. Let's go. We got the horse over there and the gong over here.

We're filling out the studio. We're having a lot of fun. Zoom in on the horse. It has been unboxed. Some folks in the chat were asking, what's the genus of the horse? There's a few things. We enjoy equestrianism.

We enjoy polo. This was from the early brand exploration that we did. We found that there was a bit of white space there. A lot of people in tech are equestrian. Too many people in tech media were afraid to show their appreciation. In some ways, the horse is a monument to technology journalism. Yes.

Obviously, many technology journalists grew up with a very equestrian lifestyle. Of course, of course. We thought this was considering the show wouldn't be possible without the brave reporting from many technology journalists. We thought, why not put a horse? Also, I have a buddy Truman Sachs. He's been on the show. He runs a company called Cubby. And he's obsessed with horse power, horse metaphors.

He's obsessed with the idea of channeling the energy of a horse, much like I channeled the energy of a golden retriever. The horse is a very, it's a noble stallion. It has a lot of noble qualities that you should be emulating. And so the horse made its way onto the soundboard. And now the horse lives physically in our world every day at the TBPN Ultradome. Anyway, let me tell you about numeral sales tax on autopilot. It's a horse for your sales tax. Spend less than five minutes. per month on sales tax compliance. That's faster than a horse can run around the Kentucky Derby. I think.

I don't actually know how long it takes. Some breaking news, Barry Weiss to be named Editor-in-Chief of CBS News. I think people were predicting. Promotion. Trade deal. This is such a brutal deal. There's another post here. Paramount will announce deal to acquire free press and bring Barry Weiss to CBS News on Monday.

I feel like this deal has been announced. 50 times so that when it actually gets announced. I wonder why. Are media people obsessed with media people? Turns out, yes. Yes. People love reporting on this. You could never get this much press for a $250 million acquisition of your database ERP enterprise AI company. You'd be begging for Bloomberg and tech crunch pieces. But when you're in the media, the media loves to tell a story about media. And so you get a lot of press. In other news, Arvind over at Perplexity is announcing the browser, which I think was teased earlier, but he said announcement in 3.5 hours. I'd love to

get an update from Tyler on what this announcement was and how it's shaping up, what the timeline saying. Do you know what's going on in perplexity world, or are you perplexed? I believe that the main announcement is that comment is just free. Okay. It was originally 20 or 200 a month? Yeah, well, there's two different plans. Okay. I think it was like usage based. And I think it was also, I mean, it was recently, it was in beta as a very recently. Okay. And I think it was released probably to just the paid plans. Yep.

And now it's fully free. Well, you should give it a try again. See if you turn. See if you stick around. Hi. Some people have just kind of stuck with Chrome. I'm kind of still living in Chrome world. Although I do Chrome on desktop, Safari on iPhone, which I feel like is the worst possible

combination. Like, I should have some bleed over. But I posted a screenshot of some app store rankings for SORA and, Microsoft Teams. So is this a mobile-based browser? Can you get Comet in the... Good question. Let's go to Tyler Cosgrove, our IRL, Cluley.

Anyway, let me tell you about fin.a.i. While everyone looks that up, it's the number one AI agent for customer service, number one in performance benchmarks, number one in competitive bakeoffs, number one ranking on G2. Yeah, I don't believe Comet is available as a mobile app. Yep. which, again, it's very hard to be a $20 billion company trying to grind to product market fit. Yeah.

And the browser company was working on something similar, and there's news that Open AI will be launching something as well. I mean, even Anthropic now has an extension, which it's very similar. Okay, yeah. So it bends in Claude 4.5 into your browser, basically. Yeah. And is it taking, do you know if it's taking, I wonder if it's taking screen? shots or looking at the HTML and JSON as it comes in, acting at a lower level. Yeah, I don't know exactly.

I don't think it's agentic. Like, comment, one of the main things is that it can actually make actions. I don't think the Claude one is actually agentic yet. Perplexity is a fascinating story. You know, we've had fun with the founder on the show, but it'll be interesting to see how it plays out. It feels like there's a whole bunch of big deals that need to happen under the hood, dance partners with, you know, they tried to buy TikTok, and you could see.

partnering up with another legacy tech company revitalizing them. We saw the browser company go over to Atlassian, and there's so many different ways this could break. But good luck to the team over there. In other news, open source, Pantex CAPs says, here's the thing on data centers. I laughed at all the dumb sell side reports on data centers being canceled earlier this year.

Remember those? People were saying, oh, yeah, we're not going to build data centers. Turns out that was pulling back. And I read a little bit from this. He went back on pulling back. there's news today in the journal that Sacha said that oh, where is it? It's somewhere in here

amid a bunch of other articles we won't be reading. Microsoft CEO gives up some responsibilities. Nadella said in a note to employees that Judson Altoff has been promoted to helm the company's commercial business in essentially overseeing a large swath of the technology giants operations.

Nadella is saying that he needs to focus on the company's biggest artificial intelligence bets. So he's going back in and really doubling down on AI personally, and he's promoting other folks to take stuff off his plate so he can refocus on AI. So kind of a retrench into, let's go big on AI. Yeah, it was, I was pretty, I got personally a little bit bearish when Satya was pulling back, right? He has more data than almost anyone on Earth. Right, very smart guy. He was basically actively getting fit. They're doing rounds of layoffs.

As he pulled back, a lot of other people, you know, have stepped up. Oracle has obviously seen a massive appreciation from leaning into, totally. And you could see Microsoft feel like filling that gap if Oracle hadn't stepped up. Totally. But what open source or Pantex cap is saying, but he says, you know, I laughed at all the dumb sell side reports on data centers being canceled earlier this year, made a ton of money going long against that nonsense and bearishness, but now we're on the other end. I don't think people understand the actual logistical hurdles in place for data centers now. First, power companies are overwhelmed and tired of all the fud.

They're putting up huge barriers to weed out all the BS. You mentioned this earlier. If you want 500 megawatts from Georgia Power, you now have to put up about 600 million up front. That's just such a massive number to be locked up. I'm sure people are also like lending, now lending helping people finance that

and he says second communities are pushing back more deals are getting canceled on it. The water, we saw the Amazon data center was in Tennessee, where was it, that just got canceled. He's saying water consumption is a real issue. Data centers actually

help electricity prices go down but overall rates are shooting up and they're getting blamed. So I want to You need to dig into this more. Unpack this more because that's not. This was always my anti-fast takeoff take, which was even in this world where the algorithms hold the scalings all you need, the bitter lesson holds, it's like at a certain point

you're just moving like billions of tons of sand and refining it into silicon to make chips and like the laws of physics come into play at a certain level if you're just moving an immense amount of equipment around the world and you need more boats to move the GPUs from one place to another, you need more energy. You have to build a new Hoover Dam every two minutes because you're accelerating so fast. And it feels like there will be some laws of physics that will kind of act as a drag. I don't know. Tyler, what do you think? I mean, yeah, but pure Arctic labs is kind of like the answer to that. I mean, obviously, like, at some level you always have to like, you know, conform to physics. Sure. But I think we have a long way to go where like you can actually have

the LMs improving the chips. How many days until age? GI on the on the on the on the on the day tracker I know it's it's 3,600 so that's what 12 years or something yeah I think so based on today based on this week based on SORA 2 has been updated since SORA I think I need to kind of give a good think about this okay figure out where I am but what about you Jordi if the singularity is is is is roughly 10 years away does the SOR 2 news does the those the energy consumption news move your estimate out or move it forward? I think I'm moving a little bit out.

I'd probably add another 100 days to the singularity counter personally, but feel free to fence. Yeah, it signals, it signals, yeah, this is Wilmanitis post from yesterday saying, is this the product you release if you believe you're three to four years

away from super intelligence? Yeah. Well, if you're looking for super intelligence for CRM, head over to Adio. Adio is the AI Native CRM that builds scales and grows your company the next level, gets started for free. Also, another extremely AGI-pilled news headline, Google ads are now live in Gemini. Let's go. There we go.

Do you love ads? Where was DeepMind on your AGI-pilled category? Ads were always on the critical path. I think this is critical path. I don't think this is bearish. Let's take 2,000 days off the singularity counter. It's coming tomorrow. Considering Gemini is the only provider which doesn't allow users to opt out of training,

their models on their data, which I don't think is true because we just got news about the meta team will be training ads and targeting ads based on your interactions with meta AI chatbots, which makes tons of sense. Targeting here will be off the charts as Wasteland Capital. Every data point about you, every question you've ever had sold to the highest bidder. And so if you're just chatting with Gemini or you're chatting with meta AI, you'll start seeing ads that reflect what you've been searching for in the chatbot or talking to the chatbot about makes a ton of sense. I'm sure there's going to be a bunch of weird things. It was already weird where people would say, like, I talked about this product with my friend and then I saw a targeted ad. How did that happen? It was already

feeling like creepily accurate and I'm sure this will make it way creepier. But it'll be good for the business probably, right? What do you think? Bullish? You know we're bullish ads. Bullish ads. Well, we are also Bullish Sheel, a good friend of the show. He has some big news and we're going to bring him into the TBPN Ultrodrome from the Restream Waiting Room. Shield, how you doing? Boom. Great. How are you guys doing? Fantastic. It is great to have you back. It's been too long. Awesome chat with Palik Gelsinger. Really went deep with him. It was a lot of fun. That's amazing. Yeah. I mean, I feel like the only, like, since I was last on, you guys have had like, Zuck, Gelsinger, like a bunch of, I feel like the level getting on TBPN now is a

really big deal. Well, we're happy to have you. Give us the news. What's new in your world? Yeah. So we just announced yesterday that we raised a new $140 million fund. Multiple times. Massive new fun. Congratulations. One gong hit for every fund. You're on fund. It's crazy. I thought you guys were on. I would have, if I had to guess what fund you're on, it would have been like at least five by now. Maybe you're like you got that wise unct status or something. But yeah, just getting started. I appreciate that.

Yeah, it feels like we're still brand new though. Like it I still feel I still refer to myself as like a new investor, but we've been running BTV for six years now. Any change in strategy with this fund? Or can you just restate the thesis for me? Yeah. So the thesis is investing in the best pre-tecuit. seed and seed stage founders building fintech companies. We take a pretty broad view on fintech.

So for us, there's a lot of stuff that you might not think about as fintech. Vertical SaaS is probably the main one. So if you look at toast, Shopify, those kind of company, mind-body, those kind of companies are actually largely fintech companies. So 80% plus of the revenue for those companies comes from financial services. So we invest in those kind of companies. We invest in some B2B marketplaces, but primarily we're just investing in the future financial services. What's the mood on the ground with LPs? It was easier than in past years to raise this fund. Is it contrarian to be focused on fintech when everything's AI? Or were you able to kind of roll that into the narrative because every fintech company is obviously leveraging AI now? What talking

points were you running through? Yeah, it was definitely, I mean, the mood on fintech is definitely not what it was when we last raised, which was 2021. Nothing could have been, I mean, in 2021, 200x revenue multiples for a team that had started two weeks ago and had, you know, an idea of a customer. Yeah, it was absolutely crazy. Exactly. We actually, we announced it yesterday and we haven't even started deploying from the fund, but we actually raised it a little while ago. I think now things are probably a little bit easier, but it was a harder fundraise for us than the last one in 2021 for sure. But we actually are larger than we had planned for. We had planned to shrink our fund size. And the idea was last fund took us three and a half years to deploy.

This one, we wanted to go back to a three-year cycle. So we were going to go from 150 to 125, but we ended up going to 140. We had some LPs that we really wanted to have involved. So overall, it wasn't too bad, and we're excited to deploy from it. But I think since we raised it, and now there's actually, it's actually like the market is ripping. You had Chime Clarnet, E. Toro, go public, circle, and you have so many more coming, Wealthfront, Navon, ethos in our portfolio.

And so I think the market is pretty hot on fintech. So actually, I was just before this talking to another fintech investor who's about to raise. And I said, now is a great time to raise. The market's awesome for fintech. What, I mean, you've been through a proper cycle. What advice did you give to founders at that time? What are the stories that you share about like buckling up and getting through rough periods, rough patches as a founder or entrepreneur? Yeah, it's just like do what you can to survive.

And if you live another day and you're a smart team, you'll figure it out. And we've had that happen. Like some of our companies basically had no growth for two years. And then all of a sudden, the last year had total inflection. And I think 23 and 24, 22 and 23 were really hard for fintech because a lot of these companies are selling into other fintech companies. If you're an infrastructure company, you're selling into other fintech companies. fintech companies and the funding completely stopped. There were all these, you know, NeoBank for X that got funded in 2021, like way too many of them. And so you had, it was just

so easy to grow in 2021. And then 22 and 23, you had to grow significantly just to treadwater, just to stay like even. And then now we're seeing companies grow a lot. So it was really just stay alive, hunker down. A lot of our companies got to a point where they didn't need to raise money. And obviously that's the best time to raise money. Now, now investors are throwing cash up. How are you thinking about stable coins? I think we've talked about them before. It's in such an it's an interesting time to me because people are so bullish on stable coins as a category. And yet it seems like the people, the company's best suited to capitalize on this opportunity or not startups. And I don't know if you feel the same way, but you see,

Stripe's launch of open issuance you see which was interesting to me because we had a company on like a week and a half prior that was building the exact same product and then Stripe launches it and they already have you know. There's also a public company circle that like is mature in that space and there's a tether. And then it feels like also interesting because even the the scaled stable coins that tether UST, USTC are now under attack from every institution

on earth and you know the the pie will grow, but at the same time, it's going to get sliced up. So where are you thinking about the opportunities to deploy against this kind of technology at the early stage? Yeah, I think you're right. I think one challenge we're having right now is we're just seeing so, especially we talked about last time I was on the bridge acquisition. Since then, we've just been seeing so many companies built in every sector within stable

coins. and I haven't seen that many that compelled me to say, okay, this is, you're going to beat Stripe on this. Like, Stripe is executing at an amazing clip. And I haven't found, the use cases I found are specific to a geo. So like, you're solving a problem in Latin America. And I think you might be the best person for that. You might be able to get a license in that region.

Yeah, I'm an investor in Menteo. which is doing something, you know, similar to bridge, but just for Latam, right? And so they've been building banking relationships for years now. And so they're, they're in a good position. But, but, yeah, it's a, it's a weird thing for venture capitalists and angel investors to be, at least from my point of view, super bullish on the category. And it's like, it's hard to deploy against it other than just buying some Stripe secondaries. It's totally true.

One thing that I, like a couple days ago, Stripe announced that they were doing. doing agentic payments. And obviously that was coming. But it's so hard. Like I met at least a dozen companies in the last year doing agentic payments. And I feel like as a VC, you're not supposed to say, you know, what if Stripe does this? But it was so obvious that Stripe was going to do it. Yeah, but saying what if Stripe does this is very different than what if Google does this, right?

Yeah. Or what is like Morgan Stanley does this or J.P. Morgan or like an older company that maybe isn't in founder mode? maybe not in Silicon Valley, maybe not Sam Altman on the cap table with the company that's doing the thing. It's like, you know, this is going to happen. Anyway, we have one last question. It's clear that the gentic payments are going to be awesome. Yeah.

And it's and like, so it's the same thing that you just said, Dordy, which is like, I believe in the category. I believe that I will be making purchases through like through my chat interface. But I don't think that there's necessarily a lot of opportunities for new startups. Yeah. Last question from the chat. They're calling you the People's Mayor of San Francisco, and everyone wants to know, what's the best place to get Indian food in San Francisco or New York? It's a great question.

Okay. All right. So New York, my favorite restaurant in the country, in the world, Indian restaurant is, my favorite high-end Indian restaurant is a place called Ambassador's Clubhouse. It's in London. And there's Investors Clubhouse in Jim Kana. and they're actually opening in New York next year. So that's going to be big.

Lines out the door. Lines out the door. And then there's a book of flight today. Stema in New York is very good. I got to say in San Francisco, I don't think we have excellent high-end Indian food. There's a place called Aton in Palo Alto. And then my best, my favorite spots are Udipi Palace on Valencia Street.

Colapasi also is. in the mission. And then there's a new spot called Jalebi Street on hate or in the hate. So those would be my three go-tos in San Francisco, all kind of hole-in-the-wall places. But that's the best Indian food. Well, thank you so much for breaking it down and coming on the show. And excited to get a...

I wouldn't be surprised if it's impossible to get a reservation in any of those places after this pod drops. Thank you so much. Great job. We'll talk to you soon. Let me tell you about... Adquick.com. Out of home advertising, easy and measurable. Say goodbye to the headaches

of out-of-home advertising. Only ad-quick combines technology, out-of-home expertise, and data to enable efficiency and this ad buying across the globe. We have our next guest in the Restream Waiting Room. We will bring in Santee from dual entry, sorry. What's happening? How you doing?

Sorry to keep you waiting. What's the latest in your world? Do we have video? Do we have sound? No. Well, we do have a... Another advertisement. We got... Don't tempt John.

Get bezel.com. Your bezel concierge is available to source you any watch on the planet. Seriously, any watch. What did you call it yesterday? AGI for your wrist or something. I don't know. It's probably on one.

Out of home ads for your wrist. Yeah, I think it's out of home ads for your wrist. All right, it works now. Hey, welcome to the TBP at Ultradome. What's how are you doing? Thanks for having me. Sorry for keeping you waiting. We were going along with Sheel.

but good to have you here. Please introduce yourself the company. And what's the news today? Thank you. I'm Santi. I'm a co-founder of dual entry. We're an AERP and we're announcing our $90 million series a day. Let's go.

I love ERPs. There's nothing like the feeling are coming out of stealth with an AI ERP with 90 million of funding. It just doesn't. I don't think it gets better than that. It's fantastic. How long you've been working? How long you've been working on it? What were you doing before?

We've been working for about a year. It does feel good. I will say it does feel good. But it feels even better to be able to run an ERP that doesn't suck like the 1990s. For sure. And we've been working for about a year now under stealth. We used to run a private company before. And we suffered one of those legacy European migrations for over nine months.

Oh, yeah. And as good entrepreneurs, whenever you have a problem, that means there's an opportunity to solve. When, I'm going to steal Jordy's question. when did you know you wanted to build an ERP? The first day we were out of the other prior company we promised ourselves. That was the day we were going to do it.

I'll tell you it was a very unpopular. We had a ton of people saying it would back you guys for any other idea other than that one because that sounds like a very stupid one. Yeah, so what is the traction? Yeah, so why? How are you, how are you, I mean, it's the hardest system of record to rip out. It's an enterprise. It's a company that is built on something else out there.

they're maybe unhappy, but they're really stuck. Are you better at migrations with AI? Are you just better at sales? Like, how do you actually onboard customers? Because I have to imagine that you have some serious traction to hit this series A so quickly. Yeah, that's exactly right. I mean, we understood from the very beginning that, like, ERP is like having chronic back pain, right?

It's sort of a pain that keeps growing with time. But if you have to think about vaccine surgery, it's very risky, it's expensive, it might go wrong. And we had to fix that problem. And the backpan surgery in ERPs is the implementation process. It takes nine months. The failure rates are over 50%. It's very expensive at times,

two, three times a software fee. So we fixed that with our next day, European migration engine. So we actually plug into whatever system you're coming from, whether it's legacy or IP or starter cloud. And we actually have you live by the time that you're doing the demo in most cases. Whether they're like it or not.

Wow. Pull the data from their cold dead hands. I love it. It's our customer's data. Yeah, of course, it is. Speaking of customers, what's the median size of a customer? Are you going mid-market to start? Are you going for the huge Fortune 100, Fortune 10?

How do you see the market now and then where will it develop? Yeah, so we're targeting the mid-market. We're starting with companies as small as $5 million, but we have public listed traded companies running on dual-lunch today. They're full set of books, not a subsidiary, not a department, just the entire company. And the most important thing for us is to give customers uncertainty that this is the last therapy

that they'll ever have to do. You know, you jump on, you upgrade from Starter Cloud, and this will take you all the way. Why do you think you're having more success getting people to trust you, switch over, is a lot of it helping, I imagine, agentically sort of do the migration or what's working now where, you know,

some of the investors you talked to maybe a year ago were saying, I'd fund you, but doing anything else. Well, I think AI is bridging that last mile in migration, especially when it comes to the mappings that need to be done. They're very tedious, and they're conventionally done by overseas teams. Then, you know, it takes weeks and it's error prone. But I think, you know, the best thing is we don't have to ask our customers for trust. We just say, go use it, demo it out. You don't have to come with us with questions and think about all the hypotheticals. Because you also don't know how your business is going to evolve.

And, you know, we're the only European in the world that we know of that can actually give you a free trial that lets you use your data and consume it. And by the way, this EOP migration engine is perpetual. So, like, you can plug it in. And if it takes you a month to make that decision, the data will continue to flow. You can show your CEO, your reporting. I know you can really play around with it, test all the edge cases before you make a decision. And that's how we've won the trust of our customers. Interesting.

are you, how are you actually, and feel free to pass on the question if it's too much detail, but are you basically saying, like, you have an existing legacy ARP, finish out your contract, but, you know, we'll start the migration over. You can start building trust with us now, and then when it comes up for your contract renewal on your existing ARP, you can make the call then. Do I want to stay with the legacy, or do I have enough trust with you guys to just, you know, keep running here? That's exactly right.

We send you a link. You click on it. We authorize the connection and the data is flowing within 24 hours. And then, you know, at times people like it so much more, and they realize how much time they're saving that the actual, you know, starting to pay for dual entry is not even the biggest issue or the biggest part of the budget is the fact that they're saving, you know, two or three people doing manual categorizations

and the same journal entries every single month. And accounting is a very repetitive task by nature, but every accountant doesn't want to be blocked down in the repetitive tasks. Every finance team member wants to be a forward-looking, and thinking, you know, sitting down with the CEO and saying, these are the areas where we should be investing more. This is the areas we should be cutting back. Not chasing invoices or categorizing transactions and drafting the same journals over and over. So oftentimes they see that and they make the jump even before the contract collapses. Very cool. What's the go-to-market going forward?

You guys have been obviously in stealth. You're out of stealth now. Are you going to get loud? Are you going to buy up all the billboards on the 101? What's the next move? Well, we like to get referrals. That's the best way. So far, we've gotten most of our customers through referrals from existing customers, and we hope to continue that. We've never gotten so many demo requests in an hour in the past, but we're excited to use those customers as a way to get the word out

and have more people come to us. There's no reason why you need to have a dinosaur in your office today. I love it. Nice. You should skip the dinosaur, get a horse. We have a full-size horse here. Statutes are in. Dual entry is really sounding like a stallion of the ERP market.

I love it. Dual interest stallion. I like that. Congratulations. Very inspirational office view right there. Wonderful having you on the show today. Thanks so much for stopping by. Thank you for having me.

We'll talk to you soon. Good to meet you. Let me tell you about Wander. Find your happy place. Find your happy place. Book of Wander with inspiring views, hotel great amenities, dreamy beds, top tier cleaning, and 24-7 concier service.

It's a vacation home but better. Our last guest of the show is Austin from Double Zero. He's in the Restream waiting room. Let's bring him into the TBP and Ultram. Austin, how are you doing? How's it going? Good. Thank you for having me on.

Thanks for hopping on. Kick us off with an introduction to you and the company and the news today. We'd love to get up to speed on Double Zero. Yeah, certainly. So, Double Zero is a blockchain protocol building an alternative internet for high-performance distributed systems and blockchain. And so at its core, this is a physical fiber infrastructure network run by multiple independent contributors to allow systems like blockchains to go faster than the public internet will allow them to go. What's the status quo in the industry?

I feel like if I sign up for AWUS, I'm going to get charged ingrass and egress bills, and they're kind of abstracting the underlying infrastructure, the physical infrastructure. At what level are you operating? How does all of this fit together? Yeah, so this is what's called an OSI layer one, two, and three networks. So this is pretty much the lowest level down in the networking stack. And so this is the physical cables, the protocols that transport over them. But fundamentally, this is exactly what you're talking about, the egress and ingress problems in cloud data centers. That's pretty much the only thing that you get billed for in the data center when you really think about it, especially the cloud data centers. And so this is a way that they generate lock-in. It is easier to stay in the Amazon ecosystem because data moving within Amazon does not get charged, whereas data leaving Amazon does. And this is why you have such a vendor lock-in in the cloud market today.

And so if you think about the ultimate mission of decentralizing the entire AWS stack or the GCP stack and basically rebuilding these on decentralized rails, not just for the point of decentralization, but so you have a greater choice from a consumer standpoint, The network and connectivity layer really is, in my view, the first piece of that that we need to tackle. And it's in some ways the most broken piece of it today. The Internet is an incredible resource. It connects us all. It serves data in far-flung regions of the world.

But it was never built for high-performance systems. And so Wall Street does not use the public Internet. Facebook does not really use the public Internet. Even all these big Internet companies that we think of, they run private networks. And they run private networks because the Internet is too slow. to show you Instagram reels. And so why are we building what is supposed to be

the alternative replacement to traditional finance on those rails? Who are the suppliers that are bringing internet connectivity to the network? Do you need to do deals with companies like Akamai?

I don't really know who the major players are in the space that you would be pitching to bring resources to the system. Yeah, so today we have 11 independent

basically fiber contributors to the network. And this ranges from the folks you expect, you know, large trading firms that own and operate their own private networks for high-frequency trading purposes like DRW Cumberland and, you know, jump to digital asset native firms like Galaxy that are, this is a new thing for them running private fiber, but it supports a lot of the other business lines that they already do, like market making and trading. And then you have data centers themselves, bare metal providers like TerraSwitch, Latitude, server, dot com, cherry servers, getting involved on this as well as a way to basically make their regions more attractive for this new burgeoning blockchain and AI infrastructure. And then you sort of have the last category, which are blockchain native groups themselves like GEDA.

How does the high-frequency trading, like an example work? I remember reading, what was it, Flash Boys, where they build the fiber lines between Chicago and the New York Stock Exchange so they can trade commodities fast. or something, do those just go dark outside of market hours and then that you can resell the slack capacity? Is that the idea? Or like, why would a high frequency trading firm, if they built this pipe and they got it and they're making money off of it, like in what world do they end up with extra capacity that they want to sell back to basically a grid that you're building? Yeah, so there's two things that are going on there. One is simply the technology changes very

quickly. And so if you are in the business of being a high frequency trading firm, you pay a pretty extreme premium to be on the absolute fastest path. In blockchain, the second fastest path or the fourth fastest path is actually totally fine. These are still orders of magnitude improvement over the public internet. And so sometimes they have what you'd call sort of secondary inventory that they can contribute to something like double zero. The other component of this too is there's not that many financial centers in the world when you think about it. It's a pretty tractable problem for a large trading firm with a large budget to go out and have a team that connects all these major financial centers. You look at just Solana, it's running over 300 data centers around the world.

And when you include other blockchains, that number goes up dramatically. It's too big of a problem for any one organization to solve an address. And so what do we turn to? We turn to a consortium model powered and aligned through tokens. And that is what the token is the best unit at. It can bring disparate efforts together, again, a common goal to benefit all. And it really is an innovation on the traditional corporate structure, which, of course, in itself, changed the world in its own right. Sure. Tokens just happen to be the best way to create this type of alignment.

Yeah. Your guys' token launched this morning, sometime in the early hours. It's now trading at somewhere around a $6 billion fully deleted valuation. What goes into, like, what's the process to get to get to this point? from a more from like a financial standpoint. Obviously, you need traction with the network and usage and a bunch of different partners, but we talk to a lot of, you know,

more traditional public company CEOs. We'll often talk to them on their IPO day and we'll talk about the different processes around the road shows, but on the digital asset side, launching a network like this, what are kind of the key steps leading up to this moment? And congratulations, by the way. Thank you, yes.

One of the things that is different about a blockchain protocol than, you know, any sort of traditional structure is we don't own and operate any of the fiber. And now you can say, sure, Uber does not own and operate any of the cars. But at the end of the day, the role of the core contributors to the double zero protocol is to build the software layer with other core contributors that then allow this contribution of fiber units to it. And so a ton of the work in the early days on this project, apart from the software side of how do we actually build these things in decentralized fashions that allow for permissionless contribution of fiber. But how do we go out and convince these people who have access to fiber that they should

take a quite valuable asset that has fairly high OPEX and fairly high CAPEX and contribute that to the double zero network? And a lot of that comes back to, you know, our contributors for the most part are fundamental believers in the power of this technology. We don't have today like an AT&T or a Verizon or a Lumen contributing capacity to the network because those are largely either publicly traded companies or companies that are owned by PE firms, and they're expecting a very traditional,

dependable type of return from that. Crypto is not that. This is very much the Wild West. The hope, of course, is that over time, we basically prove that it's worth contributing even a small slice of a global fiber network owned by one of these sub-C cable firms or these traditional telcos into the mix.

And we've had a number of hours outreaches just today from more traditional players asking how they can get involved. The real thing here is all of these companies, it's the same thing that the cable companies were going through in the United States maybe 10 or 15 years ago. They're priced and they're valued and they sell effectively dumb pipes. And I don't mean that in a derogatory way. They're quite sophisticated advanced technologies, but they don't have the same ability to take a cut of what flows over them the way you see in other types of spaces, especially in software. And so the promise of going from selling something for a fixed dollar amount to a token model

where validators operating on it and traders operating on it pay a percentage of their earnings into the network is a potentially disruptive business model opportunity for these traditional players, the same way that the cloud providers, yes, there was a technology revolution there, but it was as much a business revolution as anything else. How has the new admin been helpful to you guys, even just in terms of being generally pro-Crypto? I don't know how long you've been building the foundation in the network, but I imagine, you know, this wouldn't have been the kind of thing you wanted to launch two, three years ago for a number of reasons. Yeah, I mean, I got involved in crypto in late 2017, early 2018,

worked for a company called Republic, worked for a company called Bison Trails, which then got acquired by Coinbase, and spent the previous four years working for the Salana Foundation. So I've been through the changing regulatory landscape. You know, we received a no action letter from the SEC for key components

of the flow of tokens in the double zero network. Shout out Cooley. This is really important. Shout out Cooley. That was your law firm, right? Shout out Cooley, yes. Let's go. We love lawyers here.

They don't get enough right. go, I owe a drink. Yes. But I will say that the, you know, the whole thing that came together here was, you know, going to the SEC and asking for relief. So this is a process where you can basically go to them and say, we would like to do this. What do you think of this? Do you see any problems with this under current securities law?

And so we applied for relief on two specific areas. The first area was around the reward distributions to contributors. So basically the network programmatic. paying out tokens proportional to the utility that a contributor brings to the network. This is not like proof of stake. This is much more similar to proof of work, where if you generate 10% of the hash power for Bitcoin, you get 10% of the rewards. But this is really important to us because many of our potential contributors, especially as we sort of move lower down and closer to the direct fiber owners, they are not sophisticated crypto lawyers. They may not even be

sophisticated securities lawyers. And that sort of reassurance that we can get from the SEC of, hey, this programmatic payment based on this type of model, they don't see any issue with that from a security standpoint. Yeah, makes a ton of sense. Absolutely massive day. I hope you, it's unlike, you know, public company CEOs, you know, the market closes. And then you get a little breather. You're not so fortunate, but massive, massive milestone. And as you start knocking down some of these massive partnerships, come back on and tell us about it. Thank you for hopping on the show. We'll talk to you soon. Cheers. Appreciate it. He said that the internet is not a series of tubes. Are you familiar with the famous line that was said in Congress, I believe, the internet is a series of tubes?

There's a music video. This is a huge sign of respect. They turned it into a meme is from 2006. It's one of the great sign of respect in my culture to show you a YouTube video that you might not have seen. I want to show you this video from Internet Explorer 4.0, the Internet being a series of tubes. Let's play this. The Internet is not a big truck. The Internet is not something that you just dump something on. The Internet is not a big truck. The Internet is a series of tubes. This is a real clip from real soundby.

We can cut it. This is a real clip from a testimony that happened in Washington, D.C. with a lawmaker who's trying to explain the internet to the, like, to Congress, basically. It's not a truck. It's not a truck. It's not something you just dump something on. It's not fixed. It's a flow of tubes.

And he was lambasted and made fun out for this. But it's kind of true. It's like the flow rate matters more than the, fixed quantity. It's not a data center. It's a series of tubes. It's a series of tubes. Anyway, how'd you sleep last night? Brutely, it's just been such a, it's honestly been a brutal week of sleep. I got an 85. I got a 65. Oh, give me the sound. Give me the sound. Let's go. Yes. There you go. Oh. Yeah, they have this new feature that tracks your sleep debt. Oh, I saw this. The sleep debt is really, really nagging me these days like, I've accumulated two and a half. Okay, I'm at

Zero. I'm at perfect sleep balance. Not in debt. No surplus. No debt. I'm doing good. It's great because I have this handicap of having a much longer drive-in to work. And so it kind of, I really got to push myself. Equalizes for the twins. Earlier. Yeah, it equalizes for the twins. Yeah. It was great to see. We saw Will Robbins in the chat earlier. He would pump in up, pumping up dual entry. We had dual entry and then double zero back to back. That is a funny. A little bit confusing, but apparently Will did the seed of dual entries. He's feeling pretty good with this announcement. In other news, Inversion Space announced Ark, the world's first space-based delivery vehicle. This video looks remarkable.

Arc enables the on-demand delivery of cargo and effects to anywhere on Earth in under an hour and offers unparalleled hypersonic testing. Sounds very sci-fi. They have a rendered video here. We'll have to dig in and understand. and the actual timeline for this. It seems like a very ambitious project, but congrats to everyone over there at Inversion Space.

Very cool. And we got to get on with London. We do. Tomorrow, unfortunately, we will not have a stream. We encourage you to... Touch grass. Watch our recap of the week. Maybe we'll release the recap early.

Go back. No show tomorrow. Watch the first episode we ever dropped. There's a lot of lore in there. A lot of lore. Enjoy that. A lot of fun. We'll be off tomorrow.

We'll be back Monday with full force. I cannot wait. And we hope you all have a fantastic weekend. But don't start until tomorrow. Maybe around five, maybe around 10. Maybe burn the midnight oil. But have a great weekend. We love you.

We'll see you Monday. See you later. Bye.