Google’s AI Breakthrough in Cancer, Protein Powders Exposed | Marc Benioff, Eiso Kant, Dante Vaisbort, Alice Bentinck, Eric Seufert, Pim de Witte

You're watching TBPN. Today is October 16th. It's Thursday, 2025. We are live from the TBPN ultradome, the temple of technology, the fortress of finance, the capital of capital. Big news. We covered it a little bit yesterday, but big news out of Google.

Sometimes when something just hits the timeline and we're live and we're doing the show, it's hard to process how significant something is for sure.

And I mean, what a funny. What a funny turn of events. In the meantime, while OpenAI is fighting for their life in the timeline against allegations of moving into adult content, Google is saying, hey, we cured cancer. We've done it. We cured cancer.

Yeah. Which way? Western lab.

Yeah. Yes, exactly. They didn't actually cure cancer, but they made some progress, and it's definitely updating some people on what AI can do in bio, what AI can do in cancer research generally. It's very complex. I'm not an expert in bio or. Or any of this stuff, really, so I needed to use a call of duty metaphor. And so I read the piece. It comes from Sundar Pichai. He says time is money, save both easy use corporate cards, bill payments, accounting, and a whole lot more all in one place. Go to ramp.com. just kidding, he said. An exciting milestone for AI and science. Our C2S scale 27B foundation model, built with Yale and based on Gemma, generated a novel hypothesis about cancer cellular behavior, which scientists experimentally validated in living cells. Now, that is not in people. It's not in mice, it's not in rats. It's just in cells. And this is just one potential link between a drug and cancer cells. But it's very exciting, very promising. And so I was trying to figure out the appropriate analogy, and what I landed on was, of course, Call of Duty. And so in Call of Duty, there are a bunch of players on the map. Some of them are on your team, some of them are on the other. The opposing team. Exactly. The opposing team. You can think of them as cancer cells. When cells develop cancer, when tumors exist, they are on the enemy team. You gotta hunt them down. You gotta find them. The people that are hunting them down, those are the killer T cells. That's your immune system. Ideally, you want all the tumors, all the cancer cells to be really obvious to your immune system so your immune system can go around and get a bunch of headshots. Double kills, 360.

No scopes, of course.

Of course. But it's hard because a lot of these cancer cells, a lot of these tumors, they Exist in what Google puts them as. Like, it says they're cold tumors. Basically, they're invisible to the body's immune system. You want to turn them hot, you want them to light up on the minimap. How do you do that? You got to pop the UAV. And that's exactly how this works. So this particular C2S, which is cell to sentence model, ran a bunch of different correlations between different drugs and their effect on cells and found a drug that does just that. It's a conditional amplifier, meaning that it boosts the the immune signal and it basically turns these cold tumors hot. And now this is not new. There's a lot of drugs that basically act as UAVs for the immune system that allow the immune system to target cancerous cells or tumors more effectively. But why this is special, why everyone's obsessed with this, why everyone's white pelling so hard, is because they used AI to do this. And of course this was in conjunction with scientists and then it was verified in a lab setting in the real world sett. But it's still very exciting. Now, just to put it in concept, in perspective, Google found one drug that was correlated with this effect, which is great. There's something like 600 maybe FDA approved drug cancer indication pairs in the US alone and only about 5% of drugs that even get submitted to the FDA actually get approved. And so the total number of like discoveries like this, if you just think about it as like a drug that helps fight cancer or helps identify cancer, if you think about the pool of those humans have discovered probably tens of thousands, maybe hundreds of thousands of those, narrowed those down, sent those to the FDA, the FDA has approved maybe 5% of those and we get 600 that are on the market. And then of course, some of those are more effective than others, some of those have less severe side effects and so they become more popular. Some of them are just good at advertising, probably, or good at the cost benefit ratio. So they're very effective to a lot of people and it's really cheap. That probably is a blockbuster cancer drug. So that's really, really, really effective, but only in a very odd population. It's really expensive. Probably less of a blockbuster cancer drug. But what matters here is the slope more than the Y intercept. And so right now, if you just look at the scoreboard in your call of duty world, AI got one point on the board and humanity has like tens of thousands of these discoveries. Basically, that's a very rough estimate, but it's like it's A lot. We've discovered a lot of cancer.

And this is meaningful for one very real reason, which is that people have been desperately hoping that AI systems would be able to discover novel cures for cancer. Right. This has been something that if you look at anybody at any of the big labs over the years, they probably have at least one sound bite where they're talking about, maybe it'll be potential. My question is, does. Does OpenAI really have time to even compete on this front? Right. It's very easy for Google with hundreds of billions of dollars of revenue to dedicate resources to projects like this. Whereas this doesn't feel core to OpenAI's like even roadmap. Right. So this is that. But. But what ends up being funny is that the timing, right, within a. Basically a 24 hour period, you get the erotica announcement fast followed by this. And well played by Google.

Yeah. I don't think that Sundar had this in his back pocket and was waiting to drop it.

I'm not implying that either, but it is. I'm sure he knew exactly what he was. What he and the team knew what they were doing.

When they're probably having fun. They're like, this is the perfect thing to launch right now. The timeline is so primed to love this, but there are. Yeah. Tyler, what's your take on all this?

I mean, I think OpenAI definitely has the resources to do this kind of thing. Like, if you look at the actual model gem, they use the 27 billion parameter model, which is like, I mean, that's tiny compared to what frontier models are now. Like, GPT4 was over a trillion. It was done pre training in 2022. Like, they could throw a couple people at this. They just started that new physics team that's just doing like, physics models.

Yeah. Opening up for science, Kevin.

Yeah. So I think this is like, totally reasonable. And it's in their capabilities.

Yes, 100%. I think it just like, clearly it's a race. Even if this is just like positive pr, like, hey, AI is good for the world. Every foundation lab should be competing in that. Because that's the thing that when you get on Capitol Hill and somebody's saying, like, why are you using all the electricity? You say, well, look at all my press releases about AI and science and cancer. And it's good to point to. Like, it's just good comms. And so you want to be putting the points on the board first.

Yeah. And this is why, again, OpenAI has gotten so much pushback on both SORA and The erotica announcement. Because only a month ago Sam was saying, if I don't get enough compute, we'll have to choose between curing cancer and free education. And I don't want to have to make that call.

Yeah, Tyler. I think that this development updates me in two important ways. One is that it's a vindication of Rune's concept of like. Text is the universal interface. Like, if you look at the actual model, it's an LLM. And basically, as I understand it, the data that went in is basically for every cell. They have just a bag of words that are just like every gene that's expressed, ranked from most expressed to least expressed. And so it's basically just like a text representation of a cell. Which is interesting because I think when a lot of people thought about creating a model of the human body.

You.

Just go to like a 3D model or something. I don't know. I don't certainly go just to text.

Yeah, I mean, I don't know if it's. Is it literally text or it's. They just. It's like tokens, right? And then it's like, okay, DNA is already in tokens. It's just a string of numbers.

Yeah, yeah.

It's not that different.

Yeah, I guess.

Yeah. I mean, I don't think this, like, this is definitely vindication. I don't think this, like, was that big of, like, an update though, right? This is. Anyone who's, like, sufficiently AGI pilled is like, yeah, obviously, scale things up. We're going to get, you know, big.

Well, that's number two. So number one is that you don't need some sort of, like, new data primitive. You can actually just use tokens. You can use text to understand a cell and then find a connection between B systems based on just a pure textual representation of the cell and the interaction. So I think that's. That's an interesting vindication for like. Text is universal interface, not just in coding or text or knowledge retrieval or any of these other things. It's also applying to bio. You can take bio, transform it into text, and then do interesting things.

What do you think?

Yeah, I mean, so, like, if you look at Alphafold, Alphafold was like, not that crazy of a new architecture.

Right.

It was something we've already been doing on gaming or stuff like this. Then you apply to bio. So I think I'm not that surprised to see, like, architectures that we've already have been working well in other areas. You apply them to bio and it works because it's not like those architectures are fairly general. It's not like just for gaming or just for like text.

Yeah, right, yeah, yeah. I want to talk about scale but first I want to talk about scaling your streams with one livestream, 30 plus destinations, restream, multistream and reach your audience right there. So yeah, the second takeaway for me is that like scaling laws apply to useful AI applications in bio because if you actually go to the GitHub, you can see that they ran this model on something like I think like 4 billion parameters or something and then they scale this up to 27 billion. It might have even been smaller model. And so there's an interesting, like they're scaling up and it's qualitatively better system. And so. Yeah, I don't know.

Yeah, I mean when I see stuff like this I kind of. And wondering what people who are like, oh, this is a top, like we're going to crash any moment. Like how do they respond to this? Right, because this is exactly what everyone is like, oh, like oh, it's just, you know, it predicts the next token. Like yeah, it's slightly better at coding, it'll make someone 5% more effective but like this is not going to be substantial change and event like inevitably it's going to collapse. But then you see stuff like this where like obviously this can keep growing. In five, ten years this will be.

Yeah, I think, I think it's incredible. But it's more tools for humans. Right. Like this is human machine collaboration, obviously like the key insight, but still we're at a point where this isn't like an AI 2027 scenario where you see the machine running, has a novel insight, runs away with it, fully robotic, lights out lab and then it just spits out a drug at the other end. So I think it's absolutely incredible but. And an amazing proof point for the potential for the technology. But based on. This is exactly what you want to see based on all the promises we've been getting for the last decade. So I think Google deserves a huge pat on the back. But the industry should be breathing a sigh of relief being like, okay, we're actually doing what we've been saying we were going to do.

Yeah, I agree. I'm just saying if you listen to what super AGI people have been saying, like we're just right on track with everything they say.

Right.

Like if you look at, even if you look at like compute budgets, like we're now seeing, there was the 2 gig watt data center being planned just yesterday.

Yeah, Coming on the show today.

That's. I think that was ahead of, like, what a lot of people were saying. Even like Leopold or AGI 2027. Like, that's ahead of what they were saying.

Yeah.

So, like, I don't see. I mean, like, everything. It's not like we're just spending and then like, we're not getting anything in return. Like, this is.

Yeah.

I think the bear case is for, like, a fast takeoff scenario, I suppose, is something like, is this the chatgpt moment for AI and bio, or is this the 2005 DARPA Grand Challenge that took 20 years until Waymos were on the street? Like, there could be a rate limiting cycle time to actually testing these drugs in the wet lab, going through the FDA approval process. So even if the AI systems get better and better, how do you reinforcement learn on will the FDA approve this drug or not? Like, you have to actually, Gabe says.

Got to give Martin Shkreli a ring. We actually should. We should hit him up, see if he wants to jump on right now. Yeah, Tyler, figure it out.

Let me.

I'll DM him anyways. We can see live. He's probably live, but I'm sure he's down to. To co stream going on. There's a post here from G Odor who says utterly, totally destroyed worldview. Good riddance. He's quoting a post featuring a line from Yann Lecun who says it's neither optimistic nor pessimistic. It's just being real and stating what is the case. It's a system with gigantic memory and retrieval ability, not a system that can invent solutions to new problems. And then he shares Sundar's post. So anyways, humans are somewhat of a system with a gigantic memory and retrieval ability, and yet we find ways to invent solutions to new problems. So big moment. And so it's a win, win for the world.

Yeah.

My question is like, okay, if can. Can Google whatever the process that led to this discovery, can Google produce thousands of these? Right. Can they go.

It does seem like they're accelerating.

Yeah, yeah, yeah. Like, I just imagine. I just imagine at some point, like a year from now, is Sundar still sharing these breakthroughs or are they just.

Kind of going through FDA approvals?

I don't know.

What do you think, Tyler?

I mean, I assume everything that we have seen in AI says that, yeah, they can keep scaling. It's only 27 billion parameters. Scale it up to a trillion. You would think everything that we've seen says that. Yeah, this will give us More even if it's not like a novel insight, if it's speeding things up way faster than what does it really matter?

Yeah, just a super drug screener that allows you a first pass of ideas that's super valuable. I do wonder if there will be a data wall. There's obviously no compute wall because you can just train a 27 billion model. That was not the gating factor here. I do wonder if they're somewhat data limited on actually indexing every cell, every cancer type, all of that and what that cycle time looks like. Because we've seen with all the data brokers like the scale AI, the mercures, those companies have been able to scale extremely quickly to do knowledge retrieval tasks and, and put tons and tons of new answers to questions in the training set. What is the Mercur scale AI of AI Bio?

Yeah, I don't know if there is one yet. Which just tells us that this is still pretty early innings.

Yeah, it seems.

Yeah. It's interesting to see these people that were hyper bearish transformer models like this Brett hall post. He says once you understand the transformer as Lanier does and daresay I do and show on my website there you can literally see through what it does. Recombination and prediction. It's not creativity so it cannot possibly generate new explanations. And we see that with all alums. It's so funny because I like personally in my life when it comes to marketing and building brands and things like that, I think of creativity as recombination and prediction which is like if you put these two ideas together, that's great, it could be very cool. And I would give you the example of like this, this we, we just got this, these rugby shirts and that we, that we made that I'm very excited about. And it's like there's not a single new concept here. Just kind of taking rugby shirt which looks cool, making it kind of look like it's part of some formula one team and then putting our partners logos on it and yet it feels fresh.

This is the gwern take where it's like even if you can't just prompt give me a new idea for a joke or something, you can just brute force it by just giving it two random ideas and saying connect them and then you can ask if it's interesting, then you just compound that and then eventually you'll get. You're like brute forcing creativity.

Yeah, yeah, yeah. You're just kind of like indexing everything, ranking them against each other. I mean we know that like this works. This has worked in the TikTok algorithm. And yeah, just finding these correlations seems, seems really valuable. The question is just like, yeah, how long will it be until the first FDA approved AI generated cancer drug hits the market? Like, are there. Because the bear case to the fast takeoff of the compute build out is that at some point you tap all the debt, you tap all of the capital markets, you hit some sort of fundamental law of physics around how fast you can move sand around the world and turn it into silicon, or how fast you can spin up a new power plant or new nuclear power plant. Some of these things just take time and regulation. The, the, like, there is the chance that this stuff gets regulated to the point where there's no fast takeoff. Like, we should have seen a fast takeoff in nuclear energy production. Once we figured out how great nuclear energy was and we just kind of regulated it out of existence, the NRC stopped approving stuff and we just didn't see nuclear energy production. Like, you can see it's, it's exponential and then it's sigmoidal because it just turns into an S curve. Because we just said, yeah, we're actually good.

Yeah. I mean, that's kind of what Peer Deck Labs is working on.

Right.

It's like the lab in the loop of the LM or like RL step basically. But we're just like still so far from like actually tapping all the energy. Like, we're not close to that at all.

There's like, what is it?

I'll keep saying about like natural gas isn't like half a tiny amount. We're not. You know, we could like literally today, if we wanted to, we could like double natural gas like very easily. It's just regulation.

Yep.

And it's like, there's no incentive to do that right now because we're still not that close. It's like, oh, we're spending so much money. But like, yeah, I mean we could be spending way more. Yeah, I want to see way more spend.

I know. I agree, I agree. I really wonder like, how quickly will Google pull the trigger on. Okay, spend 10 times as much compute or 10 times as much energy on this? Because this cell to sentence scale 27b, this feels like they could have just put it on their ramp card. Like in terms of like the actual like, compute budget, how much weight went into that training run versus like the Gemini 3 that's launching. Like, it feels like, do you see the cutoff date for Gemini 3 is like, is like October of 25. And so it feels like that was a much bigger run. Probably maybe $1 billion of compute went into it or billion dollars of spend went into it. Like it's a very expensive project but they have, you know, a bunch of applications with it. They're selling it as an API. Like they, they can underwrite it very clearly on ROI basis. This is much harder to underwrite right now, but it's going to get less and less fuzzy as they, as they actually see. Okay, the things that we're generating, the ideas that we're generating. Yeah, they're actually making it through FDA approval. So exciting. Yeah, very cool stuff and obviously just amazing for the Google brand. Like doesn't quite fit with the concept of organizing the world's information. I guess you're organizing the cellular information. It seems like a little bit of a side.

Is there new information?

Yeah, they organize it.

Is that not knowledge that existed that was discovered and now they're organizing it? Yeah, their mission is to organize the world's information and make it useful.

Oh, make it useful.

Extremely on brand for them then being made useful.

Well, let me tell you about Privy wallet infrastructure for every bank. Privy helps you, makes it easy to build on crypto rails, securely spin up white label wallets, sign transactions and integrate on chain infrastructure all through one simple API. So the timeline is yeah, still in turmoil about this a little bit but mostly everyone's just like this is awesome.

Greg. Greg, Comrade comrad. Yeah, botching the pronunciation. Comrade Greg says someone needs to turn each TV TBPN card into a physical collectible. Slowly getting activation energy for this to start Taking pre orders limited to 10 prints, first edition reserved for the subject. All profits donated to TBPN cause of choice. He says he's pulling the trigger on it. This is cool. I think we were going to do this so maybe we were texting about.

This over the weekend and I wasn't sure about like what the right way to handle card printing. They're very news driven but I love that this just kind of organically happened.

I think we should do rookie cards for people that raise a seed round.

Yeah, first time you get on the show. I don't know. I would say it's fun to print them all honestly. But I'm sure Greg will figure out where the demand is. Do the big people want their cards or is it more like the seed stage founder? Is it more like the wedding announcements?

Do they want signed cards? Yeah, do they want like a piece of merch like you know on like some, some trade, you know, actual trading cards? They'll put like a piece of the jersey fabric in them.

Oh, really? I've never seen that. I, I feel like you'd want your own card if you did like a wedding announcement. But who would really want to connect, collect a wedding announcement that's not for their own wedding? Like that seems kind of like a niche thing. I'd be kind of like, what are you doing?

Depends on the wedding.

Anyway, let me tell you about Cognition. They're the makers of Devin, the AI software engineer. Crush your backlog with your personal AI engineering team. And he said he actually loaded the cards up into a system. He says UI polish is up next. Going to mimic TVPN website style. Thank you for your service. Greg, you've been crushing this. This is very, very cool. I would love to see how you go about actually getting these printed. You can go really crazy and actually print like real trading cards. Like there are systems and companies that do that. I was thinking that it's possible to just go to Walmart and print like wallet size photos of things and they're like 30 cents I think for like a set of four. And you just need to cut them out and people put them in the wallet photos. And it's a photo quality printer and you could just take that and then find a sleeve that fits it and for like, you know, a couple cents be like making these pretty easily. So there's a lot of different ways to, to solve the problem, but I'm excited to see where it goes.

Will Manitis says interesting cultural indicator that Volvo quietly rolled out armored versions of their family cars direct to consumer over the summer. Probably nothing direct to consumer.

You can just buy an armored Volvo now. It's pretty remarkable.

I almost bought an armored G wagon back in the day that my friend Blake had. It was the, it was the former president of Kazakhstan, actual like presidential like car that he driven around that, that my buddy Blake had imported from Kazakhstan. I really just wanted to own, you know, as a Borat enthusiast, I just really wanted to own that car. I didn't pull the trigger because it's so. It's got so much metal in it that it's like three times as heavy.

Yeah. The doors are massive.

And so it's just like not actually.

It must get like two miles to the gallon.

Yeah, exactly.

Yeah. It was so insane. You're just filling up like, oh, I'm driving from Malibu to L. A. Got to stop for gas.

Full times, full tank. Yeah. Rough.

Let me tell you about Figma. Think bigger, build faster. Figma helps Design development teams build great products together. President Trump says that India's Prime Minister Modi has agreed to stop buying Russian oil.

Wow.

Trade deal.

Trade deal. Yeah. The geopolitical trade deals are all over the place right now. There was also news that we're Polymarket said we're officially in a trade war with China. I'm not exactly sure how that's defined.

There was a market on that, I guess. Yeah, well, Trump came out and said yesterday. He said we are in a trade war with China.

Yeah. So that's the end of that.

That solves it.

But who knows? These trade wars can be very short and you know, I guess India's playing ball. Hanging out with Dario from Anthropic and hanging out with Trump and no longer hanging out with Putin and buying oil. The how did our family get so rich Meme is going viral around the Founders Fund names. Dad read the silmarillion and bought a lot of domain names. Sebastian Calliri says move quickly. There aren't that many good Lord of the Rings names left to use. Arthur McWhater says he's camping the following Mordor, Lothorian, Mirkwood, Numenor.

I don't think Valinor.

Valinor has been used. Valinor is an incub, I believe. Amman.

I didn't know Amand Aman is that. No.

There's no way did Peter Thiel back Amon Giri.

Joe Lonsdale quoted the announcement yesterday that we shared around Erebor. Of course getting approved. He said finance done right enables massive job creation and wealth for a civilization and complements builders like universities or media building anew. With today's innovation, Frontier can upgrade how these areas work for all. I'm just the founding board member and early investor in Erebor with 8 VC proud to support Palmer and his co founders.

Yeah, I always liked Shadow Facts. If you were going to start a founders fund backed hello Facts competitor Docusign competitor using Shadow Facts. Wait, have you not seen Lord of the Rings?

I have seen Lord of the Rings.

Do you know who Shadowfax is?

No.

He's Gandalf's horse.

Okay. Yeah, it's coming back and I think.

Shadowfax is a good name for like a Docus competitor. Yeah, Yeah, a horse. A stallion.

A real stallion of a company.

But also something that could be related to the fax machine, which would be funny. Anyway, congrats to the team over at Erebor. Palmer went on Rogan as well. So there's a three hour Palmer. Lucky Joe Rogan episode waiting in your podcast player of choice, if you want.

To go listen to that p. God says. Welcome back. Command economies and the classic template. Get ready to learn communism, buddy, which is that apparently the Trump admin is planning to set floor prices across a range of industries to combat market manipulation by China, according to Scott Besant.

It's very odd that they're pulling that out as a tool instead of just focusing on tariffs. I don't know, it seems like there's so many other ways to deal with.

A trade war than just they like to. They're like a dj. You know, sometimes a DJ is up there and really turn in all the different knobs. Sometimes you just let the music play.

You know, that is a very funny metaphor for David Solomon.

Like, the knobs are there, right? You can. Some people just choose to use them.

More than that, they should get David Solomon in there. He's already a dj. He's already turning the knobs. Why not have him turn the knobs of the global economy?

Yeah, I just, I do. Of course, we never discuss politics on the show, but I do worry about the opposite end of the political spectrum being heavily inspired by these actions. And then it just becomes the norm that you just have, you know, obscene levels of government intervention.

Yes, yes. All the time. I always go back to, like, trying to benchmark the level of government intervention in America versus other competitive countries. So with the intel thing, this happened the same thing as, like, what was a 5 or 10% stake that the US government took? And a lot of people were like, this is like communism. This is command and control. But if you look at the.

The.

I believe TSMC was basically like 50, 50 with the Taiwanese government. And so even now, there's a lot of other players in the semiconductor supply chain internationally that have a higher proportion of government influence in their cap table effectively. And so I'm not in total like, oh, wow, we're doing worse than what's going on in South Korea, or.

Yeah. My framework is that as much as Trump and XI beef, I believe that Trump respects xi.

Yeah. And I mean, America's bailed out companies successfully before America has come in, been the lender of last resort for struggling companies like intel and turned the company around and then sold their position and gotten out entirely like government motors. Like, the US does not, you know, fully. The US Government never like, fully nationalized the. The GM ecosystem, even though they did bail them out at one point. Joe Weisenthal says all businesses are banks except for banks. Banks are media companies. Wow. He Said this back in 2018. That's crazy. From Earth. This is because Mr. Beast has filed a trademark to launch his own bank. The organization will be called Mr. Beast Financial. We talked a little bit about this yesterday. I'm not sure what he will. What he will wind up being, but we got to have him on the show. What are you thinking?

I'm just laughing at the chat. Sam Sheffer is asking about real competitors to the USA at the country level. And the chat says China, Russia, Uzbekistan, Martinique, Mauritania. Yeah.

Well, let me tell you about VANT Automate Compliance. Manage Risk, Prove Trust Continuously. Vantage's trust management platform takes the manual work out of your security compliance process and replaces it with continuous automation, whether.

You'Re pursuing or managing it. Did you see this Consumer Reports report? Paris did an investigation.

Yes.

Did 60 plus lab tests on leading protein supplements and found that quite a lot of them had high levels of lead. This wasn't surprising to me. And the reason for that is that a lot of different food has high levels of lead. Even. Even food that would appear to be like, quote, natural. Right. So, like, dark chocolate is a good example. A lot of dark chocolate brands have high levels of lead. Something that. That you should be looking out for. Like, not every. Not every brand is created equally. And I think that people were particularly scared about Huell, which the Huell Black.

Had kind of an ominous name.

Yeah, kind of an ominous name. And of course, the timeline was having a lot of fun with it yesterday. People being like, yeah, I could tell you've been using a lot of Hewell Black over the last year.

You're just deranged from all the lead poisoning.

Yeah, yeah.

We'll stick to the optimum nutrition. The transparent labs. Good. Good to see transparent labs being exactly what their namesake implies. Transparent. It seems like they've done very well with the lead exposure. Is lead more of a cause for concern than microplastics? I feel like, what was it last year? Nat Friedman with the plastic list really shifted the conversation to levels of microplastics. And I think people kind of stop paying attention to lead.

It's all important. You want to be aware of it. But the way that you need to think about it is there's basically healthy levels, there's certified levels, and then there's like legal levels. Right. And like healthy levels. Like, ideally, it's zero.

Right.

Like lead. I'm actually. I'm sure somebody, some schizo will be.

Like, actually, I like a little.

A little bit of lead is actually.

Good for you because it makes you more aggressive, right?

Yeah. Yeah.

Doesn't it make you a little crazy?

Yeah, if you want to have a strong Q4.

Exactly.

You know, up the lead and then detox. Exactly one.

Yeah.

New year, new you.

Yeah.

But, yeah, I just think about it as, like, healthy, like certified, as in, like, you can. You can go out and get various certifications like, that generally show that your product is, like, healthy or clean. And then there's like, the legal levels, which, like, none of these. None of these brands were above the legal limit.

Oh, really? None of them. Even, even the Huel was just. Was just not recommended.

So if you're a food company in America, you can legally have. Have quite a significant amount of lead. But anyways, I think you could do these for, like, almost every category of food and get some pretty shocking results.

Yeah, I mean, I think with plastics and lead, like, the thing is, we've kind of ran the A B test with lead where it's like in the kind of early 70s, I think that was kind of the peak of, like, leaded gas. And then what you saw is, like 15, 20 years later, you basically saw the peak of, like, crime.

Yeah.

Because it's like you grow up as a kid, you're inhaling all this lead, and then it makes you way crazier. But with plastic, we haven't really run the A B test because there's kind of just always been, like, a buildup of plastic. There's like, no one who doesn't have a lot of plastic.

So who knows what happens when we take plastic out of the system.

Potentially Schizo Alpha is that lead paint could actually be positive because it reduces the amount of EMF in a room. If you have a bunch of lead paint on the walls, it's actually blocking some of that.

Yeah. All the plastic in your system is just bouncing the lead particles off, too. So you're just immune to lead poison because you have so much plastic in there. You wrapped your entire body in plastic. This went incredibly viral. 17,000 likes. I'm surprised.

Yeah. Because I think most people that look through this list are thinking, okay, I've had at least one of these in the last three months.

Probably every single day, basically. Yeah. Pretty. Pretty wild.

And it's just.

I mean, yeah, it's. It's. A lot of the brands are going to push back because not every. You know, you're not going to see the. If they ran one lab test per product, if you tested another product that was made in another batch, you could see wildly different levels yeah, the batches.

And also up the ingredient supply chain. Like a lot of times the lead comes from two steps or three steps deeper in the supply chain. This happened in Soylent. We had some article that was like, it was very funnily worded. It was like, like Soylent's like so super heavy metal or something like that because they were tested for heavy metals and they found something similar to like an early batch and it was like two steps up the chain. Like one of the, one of our suppliers, one of their suppliers had like changed their supplier or something like that. The lead company.

Bobby wants to see a lead gong in the ultra dome for agromaxing.

Yeah, let the lead flow.

I would hit a lead gong for absurd. A new company out of Y Combinator that makes absurd AI launch videos turning raw. This sounds, this sounds kind of like it was written by ChatGPT.

Well, let's pull this video up. But first let me tell you about graphite.dev code review for the age of AI. Graphite helps teams on GitHub higher quality software faster. You can get started for free. And now let's watch this launch.

Everyone wants to be different. It starts with a familiar feeling to feel stuck.

The circular motif here was really cool. And the, and the piano like changing out what's around it. I think that's a very interesting use of AI. Can we go Fuller screen Abity isn't designed in a studio fighting a robot. There we go. Discovered in obsession. Born from truth. So raw it feels like mad. That was, that was a weird show. Everyone else, this is cool.

A beautiful disregard for your own limits. A truth your body understands before your mind.

This really sounds like Morgan Freeman.

Beautiful.

Is this not stealing his likeness? I guess it's just like an homage. Yeah. Is Sam Jeffer in here? We need Sam Sheffer's comments on this for sure. He was in here before.

Yeah, it's just interesting when you. When it's difficult to make something, it's valuable and when anyone can have it, instantly it's absolutely worthless. So again, before I would judge this, I'd want to see. I want to see 10 actual launch videos for 10 actual companies that you've worked with. My guess is that. My guess is that these are might work as in they might get views, but I don't think they will build your brand at all.

Yeah, I mean I would assume it follows the same curve as like the first studio Ghibli that I posted got like thousands of likes and it was like super basic. I just took A screenshot from Oppenheimer of Einstein talking to Oppenheimer from the movie Ghibli'd it and just posted that. And it got thousands of likes because people were like, oh, this is something I like. And I know and I like Oppenheimer and I like Ghibli. So, like, if I posted that today, it would be a total flop because it's been done so many times, right? Why are you laughing?

I just think it's funny because we should actually test it. You should post just a Ghibli image.

And be like, ghibli of that horse. And be like, oh, wow. Or whatever meme's going viral that day. You could Ghibli it. I mean, there were so many Ghiblis that went viral where it was just like every iconic image for the past few years. It was like a picture of JD Vance Ghibli, okay, viral. A picture of Donald Trump viral, like picture of Michael Jordan dunking viral. But now it's like, we have all seen it and it's been commoditized, so it doesn't. It ran its course. And that's like the worry with this is that like the first AI launch video goes viral and then it gets less and less. But it does seem like they're building more of a SaaS like product on top of AI image gen. And so the agents kind of plan, generate, edit every scene so you're kind of like, you can still come with an idea. The thing that stuck out to me in that video, there were two things. One was the circle motif. Hitting the circle again and again and again was cool. And at the end, the very blurry images of the boxers, that was an interesting artistic style. And so right now, if I wanted to generate a video, if I come up with an idea of like, okay, let's, you know, let's do a horse themed launch video. It would be sort of a hassle for me to take an image of a horse, like, and come up with the concept of like the horse running. And then each frame of the horse running where it's in a different, you know, different segment of its walk cycle or its run cycle or its stride, each frame is transformed into a different art style or different world. It would be really hard to puppeteer that. I'd be copy and pasting the prompt again and again and again and having it an agent or having a system that kind of helps me pipeline that stuff where it's like, okay, I want to make. You've seen those videos on instagram that are like. It's a car and it keeps. And it just flips through like the headlights one after another and shows you like 50 cars. But it's always. The image is always focused on the headlights.

So let's. Jack Cohen over at General Catalyst just shared a video they did for a Zavo.

So let's play that up. I'm very interested in that. It's in the chat, it's in the X chat. We can try and put in the.

Timeline I think you're going to absolutely love.

While we pull that up, let me tell you about Julius the AI data analyst for everyone. Chat with your data and get expert level insights in seconds.

Do we have this video?

Let's play it. This isn't a restaurant. It's a battlefield. Every ticket is a ticking clock. Your best soldiers medium are falling.

Okay.

Meet Greg, your numbers wizard. He'll tell you the sea bass special is secretly costing you a fortune. But the truffle fries special, is that motion graphic AI as well. That's pretty impressive. So he knew you'd run out of pinot Noir by 8pm on Friday and already reordered it yesterday. He knows when your night will go from 0 to 60. So is this vertical sass for restaurants? And then there's Maria. That's pretty good. If so, she turns visitors into regulars by suggesting the perfect top shelf tequila for their favorite spicy margarita. She'll find hidden patterns in your audience and understand exactly what they want. That looks. That looks like post.

That looks like after chat. Wow, I hate this.

See this? This looks like not AI generated. This looks like they layered in their actual product. But maybe that's part of the workflow and the pipeline. They need some drop shadow on that logo. It's a little white on white. Little low Contrast the boys now. But I don't know. I think that is right now. That's attention getting like right now. If that pops into. If you're a restaurateur. I hope I'm getting this right. It's an ERP for restaurants. So you use that to run your restaurant. That's what I got from that. Is that correct?

The first agentic point of sale for restaurants and retail payments. Point of sale on AI agents in one platform to build the future of autonomous commerce. Over 400 businesses already use ZAVA to accept payments and manage operations.

I mean it did deliver that message to me and it grabbed my attention a little bit. So I don't know. Worth.

Yeah, the opening was particularly bad, but at the same time it was Probably a good hook.

Exactly. But I think the hookiness of that will dissipate. I think you're correct on that. It will be like eventually you'll just scroll and be like, oh yeah, I've already seen like the AI slot. I don't want to see that. But like right now, if you're the first, like Toast does isn't running an ad like that. And so if you're the first company to be running that ad that grabs people in that way, like, you will break through. There's like, there's like a window of alpha, I think.

Yeah. The question is if they had figured out a way to do this with the founder as like the key person and like some blend of AI plus the actual. I mean that's coming would be.

That's for sure coming. I mean, VO3 launched yesterday 3.1 and has like pretty phenomenal character consistency. I was looking at some demos. If you want to generate a bunch of video, go to fall Generative media platform for developers. The world's best generative image, video and audio models all in one place. Develop and fine tune models with serverless GPUs and on demand clusters. But I agree with part of it.

As an advertising enjoyer, I like to, without having knowledge, try to analyze when I see a video, when I see a campaign I'm trying to clock based on the stage of the company. How much did they spend on this? For example, if Airbnb comes out with a launch video, I expect it to be incredible because it's Airbnb and they probably spent like half a million dollars like producing Apple ads.

When they get like, oh, Spike Jones is directing it and you're like, he does movies.

And then with seed stage companies. One thing I like is like being able to clock like, okay, how resourceful is this company?

Sure, sure.

Right.

For sure.

I like being, I like being able. Personally, I enjoy being able to clock. Okay, they clearly had to be scrappy here. Like they spent like 10, 15, 20k maybe on this video. But it's amazing because it's like real. It's genuinely a really great idea. And so with absurd. I think like the, the actual output generally looks great. It doesn't. But. But I would say the, the idea itself is not. I don't know if it's like strong enough. Basically it's like white humanoid looking things in a restaurant, but maybe, maybe white.

Humanoid looking things like robots. I completely missed that. It was very like, it was very visually intense. Yeah, I agree with that. That's interesting. I think that there is an underrated side of using AI in a creative way that does show resourcefulness and it can show being on the cutting edge. It's a way to tell your audience that you know how to puppeteer. The models in unique and innovative ways. That just dropped today. We see this a lot with Justine Moore. Like a new model will drop and she'll have like converted her podcast into a cartoon with the correct lip syncing. And that's not just a feature where you just click a button and just like you have to actually wire up a few different systems and she usually will show underneath. It's like I use this for the cartoon effect and then this for the video and then I use flow or something else. And so there is a way to signal that you're sort of like tapped into all the different frontier capabilities because Sora is different, has a different look than midjourney. But you can start with a mid journey image and bring that into Sora and then you can use a different video or audio model or mix the audio especially like, I think that they added motion graphics on top that the AI model didn't just one shot those graphs, I think that that was after effects.

Yeah. So the question is like with absurd. Are they charging, let's say they're charging $10,000 for a launch video.

Yep.

Are they spending energy, resources, time?

Oh, I think it's definitely not just.

Like one shot at this point. Right.

But, but that's fine.

So today it's basically like a managed productized service business. And I've heard of another company, an AI ad company. I was effectively, I'm not going to name them, but they're just doing like creative services work and they're selling it, they're selling it like it's AI outputs.

Yeah.

But it ultimately is just them running an agency and, and are they prompting.

A lot or are they actually filming stuff too?

Or maybe they're doing everything.

You could do both.

They're doing everything.

Right.

It's basically, it's, it's, it's an agency masquerading as like a, as an AI company.

Yeah. There's still, there's still something there where, like, is it, I guess, like zooming out? Like, is it a better time to start a law firm from scratch where you can be AI native and you can just say from day one, we expect everyone at this firm to use AI as aggressively as possible or a creative agency and say, hey, we don't have. We're not a thousand person organization that has workflows for how we do things in different digital products and so we're not bought in or we're not like stuck in our ways. And so everyone we hire, we expect you to use every possible tool for the job. Like is it a better time than ever to start a creative agency because you can be AI native on day one as opposed to being like a legacy creative agency that then needs to go pull those tools? I don't know.

The challenge on the creative agency side is if you go and compete just on price, it's a really rough business because you, you get clients that are looking for cheap work and those clients tend to be really rough. Great companies understand the value of creative will invest in it. They're not just looking for the cheapest output, they're looking for something great. And on the legal side where you just there's certain amount of work, let's say like non transactional work that is, you know, purely corporate where if a new law firm emerged and they say said we want to have your legal bill drop by 70% every month and we're gonna do that by properly leveraging these tools and if that's their initial edge, I could see that working well because for a lot of just like day to day legal work that a company needs to do with external counsel, you don't really care. All you care about is that it's done well. It's not necessarily how it was done.

Let's stay on the legal example. But first let me tell you about turbopuffer search. Every byte serverless vectoring, full text search built from first principles on object storage, fast tanks cheaper and streaming scalable. So is there a good analogy between I'm graduating from law school, instead of going into the big law firm world, I'm going to start a law firm that uses something like Harvey on day one and really leans into the frontier of let's do as much as possible and set this company up from day one to be AI native such that maybe our business model is different, maybe we're not so focused on billable hours or we have some sort of different, different model that is enabled by AI. We're still a bunch of lawyers, but we're using AI very effectively. Is that at all? Can we draw any analogies between that and the D2C E commerce era where basically you had brands who said our secret, how we're going up against Nestle, how we're going up against Coca Cola, Cutting out the middleman is we have Shopify and like SaaS, we have SaaS.

And they don't, it turns out Mark, Mark Zuckerberg says, actually I'm planning to be the middleman here. I just.

That's a great take.

Yeah, I don't want to. It's 100% the money that you were going to spend on rent. Well, I'm actually going to need you to spend twice as much with me.

So do you think that's how it plays out in the next generation of creative agencies? I start a creative agency and I said, say I'm cutting out the middleman of the photographer, the videographer. I'm using AI tools from day one and I get a bunch of clients and I'm making a decent amount of money. But over time, the platforms that actually act as the AI generators say we're the middleman actually and we're going to take the lion's share of the profit.

Yeah, it's a good question.

It's tough, right?

I don't know. I think you always I. In the creative world, I think you want to be charging based on value, not how much time goes into the work. Right. Because the best logo designers in the world, they might be. If a startup comes to them and says, hey, we want a new logo, they might be able to do that in. It might take them weeks. They might one shot it. Right. They might sit down and just have this amazing creative spark. And so if you're the best in the world, you want to be paid for being the best in the world. And so it doesn't necessarily matter how long it takes. And so, yeah, certainly I wouldn't respond well to a pitch today from an agency that says, hey, you should work with us. We're going to charge you 60% less than your current designer by using AI. That doesn't get me excited on the creative front.

Yeah.

But maybe more so on legal.

Yeah. I don't know. It seems like it's like there's increasing value for true creativity. Like when I think about the AI tools that are rolling out, I'm like, I completely trust those in the hands of Gabe Whaley at Mischief. Like, I think that like when someone who comes up with completely novel ideas just has an extra tool in the tool chest, there's going to be interesting stuff. But just acting as a wrapper around it is going to be a little bit tricky.

Yeah. And the question, so going back to absurd, the YC company, there was a time when like the D2C style of doing a website really hit. Consumers would see it and you know, these brands were like simple they had a distinct style.

Right.

Product on this sort of like colored background. And consumers responded really well to it for. For a decent amount of time. Right. Kind of the same period that we've been in like the cinematic launch video era. And then eventually people just stop responding that well to it because it no longer sends the signal like this is a thoughtful person that's trying to stand out. It just becomes, you know, everything becomes that way. And then it became easier to get a response by going totally the other way and coming across like you were more almost vintage. Totally. Yeah.

It's a very narrow arbitrage, but the timeline is in turmoil on threads because paypal and Wise are taking shots at each other. Did you see this? PayPal said normalize sending money instead of memes and got 9000 likes on threads. A real ripper over there. I didn't realize how that Connor Hayes.

Connor Hayes is cooking.

He's cooking. Brands are really going wild over there. And Wise says normalize sending money with transparent fees. Oh, don't session. It's very funny. This is so like 2015. Twitter coded like. I feel like brands have not been.

I put this in. It reminded me of when like it was Coke and Pepsi saying like wag me.

Yeah, the metaverse. But hey friend, they're having fun for sure.

Wow, this is great.

But first let me tell you about Google AI Studio. The fastest way from prompt to production with Gemini, chat with models, vibe code monitor usage. You can try nanobanana. You can talk to Gemini live and continue.

Speaking of ads, Na Amar made potentially the product of the year. We will be doing awards later this year. She made a chrome extension that hides all website content except ads.

This is amazing. I love this so much.

I need a link here. I want to start running it. Yeah, it's remarkable.

The inverse ad blocker and some of these websites.

Wow.

Some of these websites are so chock full of ads. I wonder what website these are.

Display ads also it looks like from another era.

In German maybe or something, I don't know.

Did you hear that yesterday Paxos mistakenly minted 300 trillion of their stablecoins?

I did not follow that at all. Exactly what happened?

So stablecoins issuers like Paxos and Circle and Tether, they mint new stablecoins. They had an internal error apparently. It sounds like it was a fat finger. They accidentally minted 300 trillion in their PyUSD. Their stablecoin.

Did this crash their market or something? I wonder what happened here.

No, I think Paxos immediately identified the Air and burned the excess.

Yeah, they said that there's no security breach. Customer funds are safe. They've addressed the root cause. But how fat of a finger do you have to add? Like, I imagine, like 4 extra security.

Maybe the dev, like, fell.

It does feel like a cat on the keyboard. Yeah, it doesn't. It doesn't exactly.

Yeah. Joey says slight rounding error. Don't ask about the dollar backing. Yeah. So people are like. People are basically pushing back and be like, okay, like, you were able to just create $300 trillion on chain. What was it backed by? Or is there some sort of disconnect?

They bought 300 trillion.

Glad they figured it out.

Maybe they bought 300 trillion of treasuries. Well, before we move on, let me tell you about profound. Get your brand mentioned in ChatGPT. Reach millions of consumers who are using AI to discover new products and brands.

Born in Texas, in a Texas town of less than a thousand. Writes the most iconic rock album of the past 50 years. Quits music before its release. Becomes electrical engineer for amd. Makes it to Sony's vice president of technical standards. Helps create Blu Ray. Doesn't elaborate.

What a wild. What a wild life story. A true. Like, you can just do things. It's never too late to, like, just completely switch gears. This is James Williamson of the Stooges.

I never knew this story.

Story.

Me either. I'm not even really familiar with who the Stooges are. Look at this guy. American guitarist. What's the biggest song from the Stooges?

We need to convince the new, the younger generation that this is actually the path that you want to go on in life. Like, go on a short but generational run as a musician and then go work in. Go work in tech.

Let's get 100 gacs in OpenAI or something. That's the move.

Move. Breaking news.

We got some massive news hit.

Get that gong ready. Julius AI is now GDPR compliant. Perfect timing with perfect timing with Europeans getting back from summer holidays.

Oh, that's huge.

They can now get access to the AI data analyst that is Julius.

There's more Julius news in here. Somehow, somewhere, I got to go deeper.

But Dean Ball says, could I, like, make a tax deductible donation to the OpenAI nonprofit you've been joking about?

I think you can. Alexander Berger says, true story. OpenAI once came up on the checkout screen of my Safeway as the local nonprofit to donate to. I think that's. That's got to be a joke, right?

Yeah, that's got to Be a joke.

But maybe, maybe it's just like a random. They pick a random nonprofit or something. I don't know. But yes, you should definitely donate to OpenAI nonprofit because they'll probably spin out another for profit. That's my hottest take is that we're getting.

Well, being a donor in the nonprofit certainly does not guarantee you equity in spinouts. No, I've seen that with PT and Elon and others.

But it's probably a strong badge of honor. You're on the Wikipedia page if you're a co founder of the non profit, which is good.

What about if you're just a donor?

Yeah, you gotta. Really, you gotta. You gotta post a screenshot of an email, screenshot of a receipt. I donated before they invented AGI or something.

This is cool. The Mint is making some new coins showcasing innovation from each state. There are four ones next year according to Scheele, including Steve Jobs for California. They're doing Dr. Norman Borlaug. Borlo Cray. One supercomputer for Wisconsin.

Do you know who these guys are?

Steve Jobs from California and Minnesota with mobile refrigeration.

That's a huge breakthrough.

Let's give it up for mobile refrigeration.

I know the Cray supercomputer and obviously Steve Jobs. What a. What a great coin.

How much do you. I wonder. I wonder. Tyler, can you find out how many of the Steve Jobs coins they're gonna make? Because I feel like these things could instantly trade at moon like a hundred times.

Yeah, I'll look that up. And then also Norman Borlau was an agronomist, agriculture guy. He kind of did a lot of stuff that influenced the green revolution.

He ore farmed. Farming. Yeah, he was a literal ora farmer. I love it. Rune was clapping back at George Hotz. We read George Hotz's Pathetic Losers blog yesterday, which was very black pilling. Rune says contra to all this, it's completely fine to go work in technology or anywhere for that matter and not really give a damn. It is not from the benevolence of the SaaS founder, the currency trader or GPT rapper that we expect our dinner, but from regard to their own self interest. And Reid says, wow, his website is very negative. And Rune says feels like George Hotz is a zealot who has produced far less value than his clearly exceptional skills as an engineer should suggest. But George Hotz can do whatever he wants. That's the beauty of the free market. He can go and have his own opinion. He can advocate for different Things he can open source software. Tyler, how have you been reacting to the George Hotz news? Do you think that you expect that AGI will come from a corporation? Therefore it's worth it to be a corporate bagman?

I mean I'm definitely pro corporation.

But.

Also I'm kind of a George Hots truther. You know, I think he has a lot of good takes.

He does. He does have a lot of good takes.

He's pretty goaded in general.

Yeah, I'm excited to see what he does next. He's always working on a fun project. Hopefully there's something new dropping soon. But first, let me tell you about linear. LINEAR is a purpose built tool for planning and building products. Meet the system for modern software development, streamline issues, projects and product roadmap apps.

People were having a lot of conversations surrounding OpenAI's numbers from the Financial Times. Of course. They have 800 million weekly active users. 5% of those are paying 40 million 13 billion in AU ARR, which implies a $325 annual ARPU or $27 a month per paying user. And this post went, went pretty viral. Google would turn 800 million users into 32 billion of revenue.

Is that just based on their current revenue and their current user base? Because you have to think like Google's like so much more heavily monetized that the gap between 13 billion and 32 billion. It's not as much as I feel like it should be based on how young ChatGPT is. They don't. You can't actually advertise on it yet. They don't have an ads product. And so it's pretty remarkable that they're monetizing at the rate that they are already. That was my takeaway from this. Because Google's been grinding the ads product for 20 years now.

Yeah, yeah. And I think what Malte was meaning to say is that he follows it up. He says the tweet was meant to say, you can run this into a profitable business. I expect value per user is going to be higher than Google.

Oh yeah, because OpenAI is making 13 billion in revenue but burning like 20. Right. And so you add those together, you get about 33 billion in revenue to fully offset the burn, which is kind of like Exactly. Google's monetization rate. And so if you monetize the time and the products similarly, you get to break even pretty quickly.

Yeah, OpenAI is projecting to get to 100 billion of revenue.

The projections are crazy. We should go through those Doom slide says extremely polite thread by Epoch OpenAI's projection implies it gobbling up about half of all software revenue by 2028. We are going to be joined by Marc Benioff in just a few minutes who of course captures a lot of that software revenue already.

But gobbles up. Gobble. Is gobbles up the right way to do it or is it creating new software spend? Right. If OpenAI can build consumer agents, if they can monetize product discovery, which they're clearly planning to do. They hired the person that built Facebook ads. Right. It's like they're creating new software revenue. Basically money that wasn't being spent on software that will now go to software. They're going to be creating a mass massive ads business. They're going to be capturing a ton. They have partnership with Walmart that they announced, I think it was Monday. I'm sure they'll announce a big partnership with Amazon. They have a partnership with Shopify. They will flip the switch and start to take a percentage of all the transactions that they're already driving. So I think that of course the hundred billion in a few years is ludicrous, but I don't know that it's, that's impossible.

Yeah. Epoch AI says one way bubbles pop. A technology doesn't deliver value as quickly as investors bet it will. In light of that, it is notable that OpenAI is projecting historically unprecedented revenue growth from 10 billion to 100 billion over the next three years. It took Nvidia something like seven years to go from 10 billion to 100 billion Meta Tesla, Amazon, Apple, Walmart, Google, they were all in the six to 10 year camp to actually ramp from 10 to 100. And OpenAI is projecting doing it in three which is very, very aggressive, very very aggressive indeed. But yes, I agree there's a huge amount of software that feels like it's instantiated on the fly. When you go into the ChatGPT app where I was talking to a barber who was saying that that ChatGPT is phenomenal at creating packing lists for travel. And that's something that it could have been its own SaaS product. There might be a tool out there like packing list for travel.com might exist. And now ChatGPT just like does that for you on the fly. And it's something that Google used to route you to that one off piece of software. You can do a million of these, like calculate interest rates, compounding interest, you wind up on those home mortgage interest calculator websites sometimes calculate your bmi. If you're into bodybuilding diet plan, there's so many different pieces of software out there that are just kind of getting rolled up and we haven't really seen the SaaS pocalypse come to those small long tail pieces of software. But that could be.

No, I think Google has also made a pretty big push into that. Obviously the calculator. Like you're not searching calculator on Google and then going to a website that's showing you display ads.

It'd be so sad if you were like I had amazing lifestyle business just.

Calculator online calculator, calculator.com.

It was beautiful thing.

For 20 years they've also done that. They have, or at least they did at one point. Like a mortgage, mortgage interest calculator.

It would just generate. But that wasn't generated on the fly, you know, that was something that they hand coded and then would like route you to if you searched certain terms. But OpenAI was able to do that in a lot of different ways. We were talking about them expanding into the vice category with adult content and I was wondering if they could potentially expand into gambling content. So I asked ChatGPT to pretend you are a casino game of roulette. I put 100 on black, spin that wheel, tell me if I want. And ChatGPT says the ball spins and clatters around the wheel, a tiny metal comet tracing the red black cosmos. It bounces once, twice, then settles into its fate. Black 17. You win. Your $100 bet doubles to $200. The house size. The table cheers and the wheel spins on, indifferent as ever. To human triumph or ruin. Care to let it ride or cash out and walk away with your small but honorable victory before probability demands, demands its due. And so I mean, obviously that's a joke because they're not getting into gambling, but you know, they like ChatGPT can just act as a casino for you. Okay, that's an example of like software that it can just instantiate. You can just say like pretend you're generating a scratch ticket and it can just do it.

Yeah. In the same way you could ask a friend like percent pretend to be a blackjack dealer.

Dealer.

And let's actually play with real money. Real money.

Yeah. I don't think they're going to do.

That, but there's probably an interesting eval where it's like, like how close are the actual odds? Like odds to it to what it should be.

Yeah. Like you'd think that it would be able to like store probabilities in memory and then use Python to generate like true random Numbers, like, I would imagine that it could get pretty close to reality if it, like, wrote some code in the background.

Yeah, I think it could do that, definitely. And even if you just continue the chat and say, like, okay, do it again.

Oh, I did.

But if you do a new one every time, I wonder how close it gets.

I did continue the chat. I said. I said, let's double down. And it said, bold move. The dealer sweeps your chips into the felt circle. $200 on black. The wheel hums again. A perfect machine of chance, physics, and misplaced confidence. The ball launches, ricochets, clatters. Red 14. The house reclaims its offering. Your streak ends where it began. Roulette, in its quiet cruelty, is a philosopher's game. It teaches the fragility of momentum. Momentum. Do you step away enlightened or spin again for revenge? What do you think I should do? Should I. Should I double down again? Put $200. I should go to ChatGPT. I put $2 million down. Well, whatever you're winning, make sure to pay your sales tax with numerohq.com sales tax on autopilot. Spend less than five minutes per month on sales tax compliance.

Go to numerohq.com did you see that Kevin from the office was wearing a cubitus yesterday?

I didn't clock that on the stream. We should have talked about that.

What a dog. The denim strap. Cubitus.

Oh, it's a denim strap.

Very cool.

Yeah. It's an $88,000.

I honestly can't believe this is a real picture.

It's such a great pick for that character. It's so funny in so many ways. It's a very expensive watch, but at the same time, it's a watch that's not really loved by the watch lovers. And so it's sort of like cringy in certain ways, but that's exactly the character that he was trying to play. It really is an incredible amount of detail that went into this performance. Really, really a royal flush by the team behind this.

So congratulates Nick Carter. Says the government that I pay 40% of my income to has been shut for three weeks, and nothing in my life has changed.

That's a rapid rabbit staring in the mirror. Yeah, it is odd, but, I mean, the reason is that the government is not. It is shut down. But most of the government agencies have, like, six to eight weeks of cash, remember?

I'm glad they have Runway.

They have Runway. They have Runway. Maybe they'll have to call Softbank and say, hey, we need A bridge. You can take 20% of the United States government. Well, there's other news from Deutsche Bank. European spending on ChatGPT has stalled since May. DB Data Insight sites. I think we all know what's going on there.

Summer, Summer, baby.

Each season you don't need to get anything.

You don't need chat.

You could use the. You don't need reasoning models to tell you.

Saint Tropez.

Yeah. Where to park your yacht. You can, you can, you can ask your butt with the free tier. The free tier is more than enough when you're hanging out in Monaco.

Yeah, I, that, that was my initial read on this, is that you can't read too much into it, but it certainly is not good. Right.

It'd be really interesting. I mean, I would love to look at the ramp data on ChatGPT spend across ramp customers. Whether that's still going up because I have this thesis that a lot of the paid subscriptions are essentially like prosumer work. They're being expensed basically. I think that for the general average person they are very much just using ChatGPT on the free tier because $200 a month is a lot. Even $20 a month is still like well above Netflix. And people see it as a productivity tool. So they either figure out a way to expense it or they. And I wonder if that, if that adoption rate is kind of plateauing in the broader ramp customer base. That'd be interesting to look at. What else is going on? Starlink is live on United. This is great news. I wonder when this will actually be fully live. I imagine that it's going to be a slow rollout and they will work through one plane at a time. But it's now live on board the first mainline aircraft. And if you've ever used Starlink on a plane, it's remarkable. Jet Suite X. A lot of private jets have them and you can FaceTime on them and stuff. Although I doubt you could FaceTime on a United flight. It'd be somewhat disruptive to people. But whatever they do, hopefully they use Fin AI, the number one AI agent for customer service. Number one in performance benchmarks, number one in competitive bake offs. Number one ranking on G2.

What's that company that's competing with Starlink?

ASTS. It has retail ARP army behind it.

Down 5% today.

Yeah, the stock's been all over the place. They don't have revenue yet and so it's hard to exactly value. But they have a lot of contracts with cellular providers and the cellular providers don't want to be completely locked. They would prefer a duopoly. And so I think they're willing to do contracts ahead of schedule, invest, do whatever it takes to get a second constellation up there. We did see Amazon yesterday talking about Project Cooper, which is another LEO satellite constellation that could potentially be a rival to Starlink. But I mean man, the vertical integration at SpaceX is hard to compete with when you think about them just putting up so many of them.

And Amazon has their play too, right?

Amazon or Google?

Google have been doing this for a while.

Yeah, Cooper, that's what we talked about yesterday. Well, our next guest is joining us in just a few minutes. The timeline remains in turmoil over Nikita Beer. Nikita said at the I think creator payouts does more harm than good and we need to off ramp to a different system. And Elon said no, the issue is that we're underpaying and not allocating payment accurately enough. YouTube does a much better job and so Polymarket put up a market for Nikita Beer out as head of product at X this year and Nikita said this is how I win. So he is in the trenches.

Now he can hedge.

Now he can hedge. Oh yeah, I guess he could. He has the most insight in of front information here.

Yeah, I mean I think going back to our conversation, I think it was earlier this week on Posts are so easy. Every everybody that posts on X knows that oftentimes their best posts took the least amount of effort. Whereas on the YouTube creator side, for creators that are actually building a business, the average YouTube creator probably is inversely correlated in that. But some of my best videos have taken the most amount of effort. Totally. Like the top five videos took the most amount of effort, some creative spark, and then a ton of investment in everything from production to the editing process. So again, I think YouTube creators certainly deserve to be paid. Well, I've never felt as a relatively small creator on X that I should be paid.

There's also the factor of of allocating payment accurately that is very difficult.

It's so hard. Yeah. What is the value is a Jeremy Giffon post that gets 30,000 views. That's a lot more valuable to me than a meme that gets a million impressions. Right.

And so if a gifon YouTube video were to go up and YouTube were to run a mid roll video ad in that video, it's very clear that that however much the advertisers paid for that specific ad in that specific video, it's very easy to allocate payment based on that. But if you're scrolling a timeline and you see an ad here and then you see a post and then you see an ad here. It's very hard to say this ad was targeted at the reader of this particular post. And so you run into these challenges and that's why TikToks embrace the Creator fund which is just which is just distributed based on impressions. But impressions don't quite weight the quality of the audience, which is obviously incredibly important because advertisers want to get their message in front of customers at particular moments when they're primed to hear the message. So be interested to see where it goes.

Buco Capital bloke says the AI capex trade can't keep going up 2% every day on partnerships projections and information articles using 2030 estimates. And of course the AI capex trade actually can Foxconn pops today right? Based on just the news that they had a conversation with OpenAI.

Yes. So the news Foxconn jumped 8% based on this Foxconn shares rise after chairman says he met with OpenAI plans Nvidia next so it's not actually a full deal yet Frenzy Indeed investing in AI has fueled talk of a bubble. But the chairman of Taiwan listed Foxconn, the world's largest contract electronics maker says there's much further to go. The application of AI is just the beginning. Foxconn's Young Liu told local reporters on the sidelines of an event on Wednesday expressing confidence in the market's potential to keep going from strength to strength. He also said that Foxconn had met with OpenAI chief executive Sam Altman at the Taiwanese company's headquarters to discuss discuss potential future collaborations. And so the stock jumped 8%.

Let's give it up for discussing future collaborations.

Yeah. And I mean AMD shares jumped 24% after the chip company announced a multi billion dollar tie up with OpenAI earlier this month. Another multi billion dollar with Broadcom similarly so I don't know, I mean Sam is vertically integrating like crazy. We haven't seen this with a software company so fast. I mean Google eventually did it but over over a 20 year period and Sam is just saying I want multiple data center providers, multiple clouds, multiple chips, multiple fabs. He's really going deep into the supply chain and creating kind of a duopoly at every single phase. And that's really what I'm think about.

How much pressure there's going to be on the hardware launch. People are already going to the average person is going to be rooting for it to fail and than actually Delivering enough value to earn a place.

Are you talking about like consumer hardware?

Yeah, the Johnny I. Oh.

Because yeah, Foxconn would be a logical partner.

That's what I assume the conversation is about.

Yeah, I was thinking about like what a head fake it would be to be, you know, Samson saying, like, it's not a phone. And he's almost been saying, like, it's not a wearable. Like, it's not directly competitive with anything that's out there. Like, what if it is just like an Amazon Echo? Like, what if it's a direct competitor to like the Alexa? Like, that's actually kind of a great landing zone. I think the more I was talking about Alexa and Echo, I was thinking that, like, yeah, being able to talk to an LLM and that can, like, it's a much lower stakes environment to add on. Like, I think a lot of people, when they think about their phone, they're like, I'm tied to iMessage. And so I also have the Mac because I'm tied to the phone. And then I also have the Apple TV because I'm tied to the Mac. But I don't feel fully tied to Apple such that I have to have Apple HomePods. And the HomePod was not a massive success because people are like, well, I'll suffer through Sonos or I'll get a Google Nest thermostat. People have been a lot more open to having an Amazon Alexa and then also. Or Google Home and then also an iPhone.

Yeah. Part of the thing with Alexa is bought an Alexa in college because it was a decent speaker that was cheap.

Yeah.

And I maybe tried to use the feature order me some paper towels once. And so my. Yeah, I guess with OpenAI, it's like if it's just a puck that sits in your kitchen, around your home and is listening all the time, is that really going to be valuable? More valuable than your phone? Part of the draw with Alexa was, you know, you get a decent speaker out of it that at least can.

I mean, OpenAI could figure out how to produce a decent speaker at Foxconn, I'm sure.

Yeah. But then it's like, is it even really worth doing to go and just compete in the like low margin speaker market?

Yeah, I mean, Meta was trying with the Meta portal. Do you remember that device? It was a camera that would sit on top of your TV.

Sonos is a $2 billion market cap company. Now should OpenAI just buy it?

Equus Globo Global says Sonos plus OpenAI. I've heard rumors. I don't know where you'd be hearing those rumors, but.

I don't know. Sonos, I wouldn't be, I mean I wouldn't be surprised. It's like they have, they did almost 2 billion of revenue.

OpenAI is great at software. Sonos is great at hardware. The, the, I mean the speakers look great at hardware. Yeah, the speakers look great. And I don't know, there just is a question about like if you're entering these like hyper competitive markets, do you want to go for the most competitive. The phone. Sam Altman certainly indicated that he might not pull.

Hello in the chat says, how much paper towels does Jordy need?

Yeah, Jordy's obsessed with paper towels. Specifically paper towels.

I'll be honest, I have not bought paper towels in at least five years.

Yeah.

But when I think about buying things in the home setting, that's an item.

You specifically want the lindiest paper towels possible. Like paper towels from a company started in the year 1000 AD or something. That's like your dream. You're always on the hunt for an old school paper towel manufacturer.

Yeah, I want to know what tree it came from.

Yes.

Continuing. Vic says Derek Thompson said yesterday OpenAI is an amazing company and these are impressive numbers. Also, a company losing 20 billion a year with 13 billion of revenue. Making business deals that project hundreds of billions in future spending with a private valuation of half a trillion is mental. You would think, John, wouldn't you think that at a certain point people would stop if a company. It's one thing to be super bearish on a company that has massive losses and very minimal revenue. It's much harder given the history of companies that are in capital wars. Look back at Uber and Lyft. The big critique of Uber was it's losing money. It's never going to make money.

Have you seen Uber's market cap recently? Recently, I think it's like 200 billion.

TK.

Yeah, 192 billion and a 15 PE ratio. Like this business matured and is doing fantastically. It's pretty, pretty remarkable. Like they very much like fought the capital war and won. I mean it's like a testament to like they, they went out. I Mean Lyft is an 8 billion market cap. There was a time when those were, the spread between Uber and Lyft was like, you know, 2x, 3x 4x and now it's what, 20x or something like that?

Yeah, yeah. So I think it's, I just think that the company's losing money, hence it's crazy that it's valued at a lot. It's just not a good argument. Vic says stop looking at spreadsheets, go try out Codex. If anything they're undervalued.

Stop looking at these spreadsheets. Go try out Adeo. Customer Relationship Magic Adeo is the AI native CRM that builds scales and grows your company to the next level. You can get started for free. Also, it's the one year anniversary of the first playable Chromatic going on display at the Portland Retro Gaming Expo. This year they're returning with the first.

Playable M. Portland has a retro gaming expo that's extremely on brand. That's extremely Portland. Got to give him credit for that.

This seems maybe we should dispatch Tyler to the Portland Retro Gaming Expo to go try the first playable M64.

That would be cool. I think we'll be able to get one of these pretty soon. I'm really excited we can have Tyler can just spend a whole show beat goldeneye playing the mod retro.

I wonder how fast you could speedrun goldeneye Games back then were so short they just didn't make 100 hour games so people would speed through that them. This is what else is here. How'd you sleep last night? I got a ton of sleep but I slept poorly because I'm not feeling too well. I'm under the weather. I got a 61% on quality 79 overall. Go to 8sleep.com get a pod 5.

My tracker got messed up because of.

Hours and 40 minutes because it's a.

Little one so it's not we gotta.

Hit the gong again. Win for Rahul and Julius because you can now connect Julius to your databricks. Let's go get that.

This is a relevant post. Matt slotnick posted on Jul 16, 2025 Bull case for nothing ever happens is that OpenAI runs its entire business on Salesforce and Slack. And this was in a blog post at an OpenAI employee posted. They said an unusual part of OpenAI is that everything, and I mean everything runs on Slack. There is no email. I maybe received 10 emails in my entire time there. If you aren't organized you will find this incredibly distracting. If you curate your channels and notifications you can make it pretty workable. We gotta ask Benioff does anything ever.

Happen and which way does the deal go? Because Salesforce has a deal with OpenAI. Everyone assumes it's to buy a lot of people tokens from OpenAI but maybe it's a trade deal. You get. You get your Salesforce installation the money it's one of those circular deals Salesforce.

For tokens Zephyr is sharing. Broadcom's fifth customer isn't Apple or xai. It's Anthropic. They won't design a new chip. They will be buying TPUs from Broadcom.

That's very interesting that they're not buying them directly.

Expect Anthropic to announce a funding round from Google soon.

Yeah, the Anthropic thing has to heat up soon because Google like it feels like Google might be doing something. That's what Zephyr at least thinks. But then you also have Amazon partnering with Anthropic for a long time and probably ramping that relationship up. It seems like it would make a lot of sense for them to ramp that relationship up again. And then some folks on the timeline were talking about how how Apple might make a big acquisition but maybe Anthropic's too big to make. But maybe even just a deal there would make a lot of sense.

So Google already invested in Anthropic? Yes, they own a bunch 14%.

That's a lot. It's pretty remarkable.

How much does Amazon.

I don't know.

Look it up. But while you do, let me tell you about public.com investing for those that take it seriously. They got multi asset investing industry leading yields and they're trusted by by millions.

Amazon owns an estimated 15 to 19% of anthropic. And it's so funny to me based on the way on Anthropic's brand approach and all these different things the way that you have OpenAI's cap table which is owned by a nonprofit and the employees own a ton and then VCs are at the very bottom right just carving out a little bit. Meanwhile Anthropic is like Amazon Google basically.

Tyler, did you see this Eliezer Yuda Kowski post? He quote posted barely AI the Financial Times projected OpenAI's cap table following its for profit transition. Microsoft gets 30%, OpenAI employees get 30%, OpenAI nonprofit gets between 20 and 30%, SoftBank gets 10 and then the VCs get less than 10%. And Eliezer said, I wonder how the financial magnitude of this theft compares to the total amount of theft in the 21st century so far. Wouldn't surprise me if it's a majority. Is he talking about that he wants the VCs to have a larger ownership stake, I assume? Yeah, yeah, he must be just really because typically a company at this scale VCs could own 50, 60% but they're sort of getting hosed. And so LEAs are standing up for the venture capitalists.

Yeah, he must be speaking up for Josh.

Yeah, I would imagine just. Just bigger ownership stakes would have been more. I don't know if it's technically theft, but I mean, maybe he's using that kind of in a loose. In a loose way.

I think we need some kind of Robinhood figure to take from the rich and give to thrive.

Exactly.

Yeah.

Yeah, take. Take the. The nonprofit shares and just distribute them to the venture capitalist. That's probably what he wants. That's what he's advocating for. Did you see Flock Alpha launch today? We should pull this video up. Up at some point. But they did it. Flock launched a American made public safety drone response system. We were talking to the CEO of Flock Garrett about the role of drones, how you could potentially call a drone to escort you home, how you could dispatch a drone to an emergency situation, start monitoring the situation. And they did it. And they're launching Flock Alpha today. Raul says, five years ago, I stood on the rooftop of my police station and imagined this moment. There's no better feeling than seeing a vision come to life. I do not have kids yet, which is a good point. And so these can go on the top of all the police stations. The police stations, as we talked to Garrett about, are already geographically distributed across a city in a very equidistant way. And so as long as the drones can get from one station to the other or halfway from one to the other, you can basically have full coverage of everything and so you can read license plates.

Has there been a black mirror on drone? You know, drones for police that go rogue.

I'm sure there's been some stuff, but this is. Yeah, I mean, surveillance generally.

Nikunsh says accidentally said sales instead of distribution, and they kicked me out of sf.

Gotta use the lingo.

Sales is evergreen.

Yeah, I thought it was growth, though. People say distribution now. I thought people would wait. I thought sales was fully rebranded to growth. People would be like, yeah, I'm on the growth team. And it's like, you're on the sales team.

I don't think it ever.

No. Now it's on distribution.

Well.

Oh, well, we gotta ask Benioff about sales specifically. Like what reps are. Are actually outperforming today. Is it the reps that have adopted the most AI and have just increased volume, or is it reps doing things the old fashioned way, which is like actually getting to know existing and potential clients on a deep level and developing trust and making sure they're getting great service, et cetera, et cetera. Karina Nygan says, after a formative time at OpenAI, I'm launching Maison AGI, a fashion house creating cultural artifacts for the AI era era. Our first collection, Relic of Thought, is a collaboration between Ilya Sutzkever, featuring his original artworks alongside his signature hat modeled after his iconic.

They're actually doing this. Wow.

Do they have a picture yet?

Yeah. We have to pull this video up at some point.

It is a study in conviction and the clarity of vision that gives thought its form. We believe we're living through something extraordinary that deserves to be remembered not in research papers or technical model cards, but intangible objects we can hold, wear and passed down. Working alongside some of the world's brightest research researchers, building Claude and ChatGPT have come to realize they're among the most creative minds alive. Research, at its best, is an act of imagination, the ability to glimpse what doesn't yet exist and then build toward it. Many of them remain unseen even as their work quietly reshapes our future. Though AGI progress can feel incomprehensible from the outside, its story is deeply hidden, full of curiosity, conviction and creation. We're bridging that world with the creatives who give form to ideas. This may be humanity's last time to create a handcrafted project before what we build surpasses us. Each collection is also a message to superintelligence itself. That we cared and that we tried to make beauty out of understanding.

Can we play this video? I want to see how they actually launched this. Karina Nguyen left OpenAI to do this. That's a bold move, you know. Been on a tear, you know. And now launching an entirely new bold.

To be post economic in pursuit.

I guess you're right.

Become a patron of the arts.

Who knows?

It's kind of a classic playbook.

It is, it is, it is, it is. We'll try and pull that up. In the meantime, let me tell you about AdQuick.com out of home advertising made easy and measurable. Say goodbye to the headaches of at home advertising. Only Adquick combines technology out of home expertise and data to enable efficient, seamless ad buying across the globe. The chat has been on fire. Marc Benioff is a little delayed. We're gonna remix the schedule a little bit. He will be joining later the stream. Taylor Hodge has a great take. He says that he's gonna use a trained falcon to take down the flock safety narc drone. I wouldn't be surprised if we See a little bit of that. Hopefully not. I think that's wildly illegal. You can't just sabotage police equipment. You have to take it to the voting booth.

Taylor, what if it's flying over your private property?

I mean, I'm sure that there are. What is it, though?

You train the falcon to fly, like, low enough that, you know, there must be some law that, like, if it's within, you know, 100ft of the ground, it's fair game to take it out.

Yeah. I think that, you know, the police can drive up and down the street.

Yeah.

Drones have to have search warrants to go fly around your property, probably inside.

Your house, but not like on the street. A police can. A police officer can just drive down your street, no problem. Shine the light. Say, oh, is anybody breaking in? Like, you know, that's. That's helpful in many ways. You want that. You want, like a secure environment, ideally. This is why we live in a democracy. We elect the.

Yeah. Imagine if the UK right now, let's just be real, be very. You know, I don't believe there's a lot of free speech left in the. In the uk. Imagine if they had this setup and that you could. If you send the wrong post, a drone is actually at your door.

I think that might be overblown. I don't know. I feel like who. Have you talked. Have you talked to people in the UK that say it's bad? Or is it all just like American posters who are taking victory laps? Because I love an American victory lap. Don't get me wrong, I will jump on any opportunity to wave the American flag.

I think there's plenty of posts from people that are based in the uk.

Yeah, I got arrested for posting the midwit meme about my enemy. And I depicted my enemy as the sad Wojack and me as the Chad wojack, the trained falcon.

Taking out the drone is just too good.

Be great.

The perfect form factor.

Do we have this video?

Dante?

Should we start early?

Oh, yeah, yeah. Let's bring in Dante. Speaking of drones, underwater drones, we got Dante from Albacore coming in.

Amazing name.

Yeah, great name.

There he is.

Dante, welcome. Thanks again for joining us.

Fantastic. Fantastic name. I thought we were out of incredible names, but Albacore is strong.

Albacore is a great one.

It's amazing.

Lord of the Rings stuff was taken, so we had to get creative.

Going to the tuna section.

Do you know the. The. The headline that is. Is in the. The group chat is yc backed albacore swims into $6.5 million seed round I love it.

Hit that gong. We've heard it all.

We've heard it all about making marathon funds.

Congratulations.

It's great stuff.

Yeah, take us.

Who's in the round?

Yeah.

So lots of folks. It was led by Outlander. We also have our first institutional. Thank you. Page and AJ yeah, our first institutional backer was D3, which is a very drone focused fund. They're sort of mix of US and Ukraine investments, but they've backed a lot of folks like Nero. So I've got the.

They made two sequels to D1 Capital. We're on D3 now.

We're never going to run out of names there. You just go to D4 and then D5 and then eventually you go to D10. D11.

A17Z is going to be a ripper for sure.

Who else? Sorry, we got. We interrupted.

Oh, yeah, no, all good. All right. All good. Yeah. So Brave Capital, which is a bunch of ex inkitel people.

Cool.

R Squared, which is a national security focused fund. Managing partner there is a Navy SEAL commander, Rhodes Scholars on multiple national Security councils. A lot of different. A lot of different investors. You got Box Group in there. They're New York. They've done a lot of defense. I think they also did mock industries.

Oh, cool.

Pretty recently, yeah.

How'd you get into this?

Yeah, so prior to this, I actually started a battery tech company. And this was like while I was in college and really quickly identified. We were working on power systems that were high performance but would make batteries for really sort of complicated use. Cases last longer. And so that led us to the aerospace and defense market. Did a lot of work, work with drones and started thinking about the power problems there.

Right.

You know, the, the issue with drones is you're really limited in terms of how far things can go. And that problem just gets harder and harder when you start thinking about the oceans. Right. Because oceans are really big. You know, the people talk about in the Indo Pacific, the tyranny of distance. And so I was at a Shabbat dinner, ran into this guy John, who I'd known for a while through mutual friends. Brilliant engineer, can't share too much about what he was doing before this, but he's worked on some major unmanned systems production type projects. Very knowledgeable about the sector. And I was like, john, why, why do underwater drones have just like such niche use cases? Right. They're very focused on short range, you know, very human in the loop type of operations. There's nothing like an undersea loitering munition that we have Right now. And we really need like longer range capabilities for anything that's happening in the Pacific. So that's what we're working on.

So is that like distinctly positioned against like the dive LD program from Anduril? Like, how do you see. Like that seems like a bold company to go up with against, you know, it's a great company.

It's funny that you mentioned.

Right.

Yeah.

I mean, so actually the founders of DIVE were also investors in us really early on.

Sam.

Sam Russo and Bill Lebo. They're super knowledgeable about the space, obviously. I was actually just with them earlier this week in Boston and. Yeah, so. So the dive platform is focused on going really deep and doing survey type work. We're working on loitering munitions.

We're.

We're working on stuff that blows up. So you can fit a 250 pound explosive payload on our vehicle, and that's enough to do really severe damage to a large surface vessel. And if you're lucky enough to find a submarine, to sink that submarine.

Wow.

What are the kind of. I don't know how much you can share, but like, what, what's the goal in terms of, for like the timeline of a loitering munition? You want to go out effectively, park it somewhere strategic, and then it's just kind of playing a waiting game before you would want to activate it and use it for a certain use case. Like how long do you want one of your drones to be deployed without any sort of like human or machine intervention?

Yeah. So we're targeting 30 days, days for endurance and 1,000 nautical miles. And to kind of put that in perspective, the dive vehicle, which is, you know, a couple thousand pounds, that's less than a week and a few hundred nautical miles. And as you get to the size of vehicles that they were building, you know, our vehicle is like two man portable. And our lead software engineer claims that he can bench one of our vehicles.

Let's go. Yeah, well, he's really strong. The best news I've heard all day. That's amazing.

Well, he actually, he just got one of his fingers cut off and so not related to his work here, but once his fingers are healed up, he says he's ready to do it.

Actually Cut off?

Yeah, yeah. He works on a tall ship and so, yeah, log fell on his hand. It's a long story.

Brutal.

He can still type faster than anyone in his. In the office.

So you lose a finger though, that's, that's 10,000 orb points for sure.

That's like eye patch Level.

Yeah, you're. We.

We docked his salary by 10%.

Oh, stop. That's so rude. Stop it. It's very dark.

What, what, what's your path to program of record? How are you attacking?

Yeah, I mean, so contracting is, is changing really, really fast. Right. The current administration is doing things very differently and you can get pretty far without a program of record. But I'm not sure if you're familiar with the Defense Autonomy Working Group that, that just got release released. There's a lot of things that are moving through programs like that with socom. You know, indopacom has various flush funds that are available and we're focused on really connecting with the operators and making sure that from the start of founding this company, we were really talking with end users, the people who are going to be at the tip of the spear in the Indo Pacific who'd be using this. But you also have to go. Once you have an understanding of those requirements, you have to go to Congress too. And so we were actually lobbying Congress ourselves within the first couple of months of starting the company we've since moved on to. We work with a firm now, but yeah, you have to basically approach it from all angles. And there's also a lot of FMS interest as well, so can't get too much into specifics there. But there's a lot of companies have been really successful with a route to commercialization that consists of a number of different buyers. And you know, largely speaking, DoD can be really appreciative of FMS as well, because that means you can reach larger scale and, you know, potentially even battlefield test your systems.

Are you guys based in the gundo?

We are not.

We are based in Philadelphia actually.

Isn't there. Is there a great robotics program out there? Is that why.

So there's a lot of great stuff about Philadelphia. We're super close to Washington D.C. where a lot of the, you know, customers that we'd be speaking with are, are based. So it's a very quick trip we can get up here on, you know, very short notice, which is important. There's also a lot of great engineering and industrial talent in Philadelphia. And so I know there's not like too many hard tech companies that are based here. There's one that got pretty big called Ghost Robotics. They make like robot dogs.

Military.

Yeah, but we're, we're based in the Pennovation center right now, which is where Ghost Robotics started.

Sure.

And we have lots of access to brackish water just right next to our offices and Yeah, I mean Philadelphia has a really rich shipbuilding history and you know, recently the. The naval yard, it's being re. Re industrialized, but it's actually not owned by a US company anymore. And so we want to build the sort of next generation of maritime capabilities in Philadelphia and, you know, help sort of foster more of that industrial labor force here. That's still really strong.

Amazing.

Does your office have a pool?

So we're getting a giant fish tank and also a turtle, but we don't currently have a pool. We have limited space.

Okay, what kind of fish tank do you need to make sure that a presumably metal drone doesn't smash through the glass?

It's going to be a pretty big fish tank.

I don't know.

I have to look into the specifics more.

Yeah, you got to call Keith Raboy. He's got a big fish tank.

Oh, yeah, he's.

He does famously fish tank build. Well, thank you so much for coming on.

Great to meet you. I'm sure you'll be back on soon and congrats on the milestone.

Yeah, yeah.

Thank you guys.

Been a pleasure.

I'll talk to you later. Have a good one.

Cheers.

Bye. And if you're interested in some hardware for your wrist, head over to getbezel.com your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. Under carpenter, Matthew launched a new repo nanochat.

Repo alert.

New repo alert. Did you see this, Tyler?

Yeah. Yeah. It's super cool.

This is among the most unhinged he's ever written. 8,000 lines of code to actually build basically chatgpt from scratch. Is that how I should understand it?

Yeah. So he's done this whole course lecture series on YouTube before where he goes through different parts of how to train a model.

Yeah.

And then this is basically kind of all of them together plus some extra stuff. So it's training the full model, it's doing the pre train and it's also doing the RLHF to get it into like from predict Internet text to be like a chat model.

And it's just using open source data sets basically.

Yeah, yeah. I forget the exact name, but there's a good like RLHF data set that you can use to get it to become like a chat.

It's pretty remarkable. Is this. So this is mostly for like educational resources?

Yeah, I think mostly. I mean like presumably you'd have to run this on a much bigger scale to get like a really kind of capable model.

Yeah.

But yeah.

What was Your interpretation of the hot take that? Somebody asked, curious, how much of the code was written by hand? And Andre Karpathy says, good question. It's basically entirely handwritten with tab autocomplete. I tried to use Claude Codex agents a few times, but they just didn't work well enough at all in net unhelpful, helpful. Possibly the repo is too far off the data distribution.

That's kind of interesting. I think part of it is that the end product of this is the repo itself. So you want very nice code that is very readable. It's in his style and he can explain everything super easily, which is a little different from if you're building actual product or just a vibe coding thing where the endpoint is just the interface, it's what you're actually using. Where this is the code itself is the product product. So it's like you need super nice formatting and all this kind of custom stuff.

Do you feel that whether or not the task is in the data distribution, when you're firing off Claude code or Codex agents, are you noticing that it's like, okay, well, clearly there's plenty of data around how to build a modal or some sort of REST interface, and so it's really good at that. But if you're trying to do something that's probably a unique problem that there might not be a direct open source example of like you're going to try and code it from hand.

Yeah, that's probably true. I find it's really good at writing react code and tailwind CSS components. And then when I try to do some custom AWS flow, it's a little harder and it just takes longer. But I still find it does it very well.

Yeah.

Jordy, did you see that Tyler Cowen put the timeline in turmoil taking shots at Europe? So Tyler Cowen, we talked about this yesterday. He said that the European Union is run by people who have an incredible depth of knowledge in a bunch of different areas, and if they ran the world, the world would be growing at negative 1% growth and is a pretty hot take. Cassie Pritchard said. Said never understood why people liked this quote, the EU does exist. It is run by people like this and it does not have a negative 1% growth rate. I'd say that's about the typical sharpness of analysis for Tyler Cowen, though. Taking shots at Tyler, friend of the show and young Macro says, difficult to overstate the extent to which you need to be underexposed to macroeconomic data. To think that Tyler Cowan diagnosing Europe's growth as anemic is some sort of contentious statement statement or that his negative 1% hyperbole is anything but mild and contextually entirely suitable. Young macro continues and says I will defend Tyler Cowen. I will defend Tyler Cowan with my life. Difficult to overstate the extent to which one's large audience must skew uneducated for this take to pass through with broad based support. Nothing to do with pedantry. 10 out of 10 pedantic domain experts would say Tyler's quote is quite fine. Tyler Cowen there's no way a regular guy would lose to a hobbit in a fight. A regular guy is what, 6ft? No chance person who has come across height as a concept as insufficient number of times an insufficient number of times to be up to speed with baseline magnitudinal intuitions Ecstatic at the gotcha after googling the Average male height is 5:9. Actually having a lot of fun on the timeline. Wow. And Cassie Pritchard's really going back and forth a young macro. What about of the greatest upand cominging posters? Should we watch this clip from Roloff Bota from Uncapped Unhinged the Unhinged he is the steward of Sequoia Roloff Bota here. Let's hear it.

One of the greatest stewards of capital in history.

It doesn't support the numbers. So there was a lot of analysis back in the 1970s and 80s with the capital asset pricing model and people figured out that there's this asset class that supposedly has uncorrelated returns and a bunch of asset managers deem that they need to invest a certain percentage of their endowment or foundation or pension fund into this thing called venture capital. If you look at the Data, they're basically 20 companies per year on average over the last 20, 30 years then have ended up being worth in realized exits a billion dollars or more. Just 20 companies. Despite a lot more money plowing into venture capital, we haven't seen a material change in the number of companies that are outcomes that are that large. And I think part of that is that there's a lot more talent than really interesting ideas or interesting companies to be built. And I think we're spreading a lot of that talent thin right now, similar to what happened in 1999, by the way. So when you look at the data, the amount of money going into venture capital right now in America is in the order of $250 billion a year. And numbers, you know these all estimates let's just say it's $250 billion a year.

Need a lot of exits.

Well, let's just do some very simple arithmetic for a second. $250 billion going in every single year. Year. If you assume that the firms generate 12% IRRs net net of fees and carry, which isn't that great by the way. I mean over the last three or four years the NASDAQ has compounded at 16, 17%, let's just say 12%. Not spectacular. You basically just average performance. You need a 3.7x roughly on 10 years on a 7 year exit horizon. So I'm being a little bit aggressive. I mean maybe it won't even be that good. So 3.7x on 250 billion, that approximates to a trillion dollars a year coming out. Coming out by the way. That means. And that's what the investors own. So let's say that the investors own two thirds of the company. To make the arithmetic simple, that's 1.5 trillion annually in company exit value. Yes, just think about that for a second.

Where's that coming from?

Well, figma's worth what? It's a lot, but it's 40ish billion.

I mean that's, let's say it's worth 0.03 trillion.

So if you start thinking in trillions and figma gets you 0.03 trillion, you need 30, 40, 50 figmas every year to make that arithmetic.

Figma a week.

I don't see that many companies of that scale every year. So the only thing breaks is the return assumption doesn't hold. And so venture is a return free risk, not a risk free return. It's terrible. You are investing in the index or holding T bills, honestly. And so I don't think venture is an asset class, asset classes scale. If you add more money, you can build more real estate. There's a lot of equities. There are trillions and trillions of bonds to be purchased. Venture capital doesn't scale with more money.

I mean, the most important takeaway here is Bobby's question in the chat. What's on the wrist risk check. We'll work on that.

We will get to the bottom of.

That Venture return free risk is a crazy line.

Yeah, taking shots.

But there was some good follow up from Hari Raghavan. He said obviously Roloff's a legend and one of the smartest people in the valley. But I'd love to dissect this, add a bit more nuance to this. I completely agree with his directional point that Most of vc, everyone but top decile sucks as an asset class. But I think the same is true of hedge funds and the top quartile as less risky asset classes like real estate, et cetera. But the nuance is that the true venture capital segment formation through series A or modest series b is probably 20 billion a year, maybe 25 billion because the remainder of the 250 billion is really private equity being employed by a VC firm. When you take it down in order of magnitude, the exit expectations make a lot more sense for true VC that is over, let's say 8 to 10 year horizons, 4 to 5x. So 100 to 125 billion in investor ownership at 150 to 200 billion in exits for the PE growth equity class theoretically at 1.7 to 2x. Clear similar hurdle return because the hold periods are say three to five years from series D E, et cetera works out to about 350 billion. So the total now comes out to 400 to 600 billion. Nothing to sneeze at, but not one and a half trillion. And then you don't need 50 figmas a year, but 15. And more importantly, if a unicorn is a home run, a Decacorn is a grand slam. We're seeing more of the centic corn.

Which I guess Purple Crown. Let's go for the horse race analogy, switching out of baseball into horse racing.

And he says because you could have 15 figmas or literally one open AI a centic corn.

If a, if a unicorn's a home, a run, a Decacorn's a grand slam. A centic corn should be like a world swear, a World Series clean sweep, something like that. Like you won the whole se. The whole season.

He's saying, I'm not saying we'll have one open II per year, that'd be crazy. But we could See per year 1 anthropic stripe data bricks, etc, 2 to 3 or sorry, 3 to 5 figma scale outcomes, 20 times wealthfront scale outcomes. Wealthfront obviously was recently acquired by Gusto. All we have to do is adjust that there are 20 unicorns a year for inflation and that's very plausible. And that gets us to 400 billion in exit value. Markets are so much bigger than they used to be. Ten years ago we had zero trillion dollar companies. He says, maybe I'm just too much of a techno optimist. And Vol comes in with the reply and says nailed it.

Correction on wealth run, wealth runs independent. You're thinking of a different company. Wealth runt's the robo advisor.

Oh, gusto.

Guideline, Sorry, guideline somewhat different.

Yeah.

The interesting nuance here is how weird OpenAI is. It just breaks the venture model because yes, it is maybe going to put up a trillion dollars of value, but as we saw the venture ownership, Roloff was saying it'll be 50% venture owned and that's not the case for OpenAI. And then also it just completely breaks the whole model of like, who's founding these companies. Sam Altman's like, much more experienced, much older than, you know, some Gen Z needle in the haystack founder that you can pick out and get, you know, 20% of the company for, for $10 million. That just wasn't the nature. Like the one of the first venture deals we talked about with Thrive was what, $150 million at like a $12 billion valuation. It was like, you know, building like 1% ownership in the firm. And so it's like even it's just such an outlier. But that's the game of venture, right? That's the game of venture.

OpenAI just hired prize winning black hole physicist Alexis Loopska as the first member of its new opening Eye for Science team. Team Per Axios. Good timing here because they are leaning more into science Loop. Saska is a Vanderbilt professor known for his work on black hole photon rings. And now he will help shape how GPT5 tackles advanced math and physics problems and guide OpenAI's push into scientific discovery. VP of Science Kevin Wheel said GPT5 can already perform limited, novel scientific research, calling it the worst AI model you'll ever use again. I got to meet Alex earlier this year And I said OpenAI should hire you. I'm going to hit him up after this and say I'm sure he said something like that. Sounds like it'd be fun. I'm sure they were probably already talking at that point. So not the most original idea, but awesome guy and super excited for him to be over at OpenAI.

Yeah. Well, we should prep for our next interview with Isso Kant from poolside. The news that we'll be digging into at some point during the show is a giant new AI data center is coming online to the epicenter of America's fracking boom. We talked about this yesterday, but coreweave is partnering with them to build a a massive data center on more than 500 acres of land that sits on a sprawling ranch in West Texas. The site is owned by the Mitchell family, which has run on oil and gas companies. Which has run oil and gas companies for decades in the state and is located in the heart of the fracking boom.

All right, we've got our next guest coming in.

Okay, let's bring him in to the tv. Van Altrium, welcome to the shop. How are you?

How are you doing?

Hey guys, good to see you. How are you doing?

We're great. Pumped to have you. Yes. If, if. For people that were watching yesterday, we were reading through the article about your guys new project in west Texas. I was like poolside. Like poolside's going this hard.

Yes.

And when we realized that it was you, we, we were pretty excited.

Yeah. So please tell us the, the, the story of the company. Could you. Can you start kind of at the beginning a little bit on your background than the first act of the company? Because it does feel like you're at a very unique moment in the company history.

Yeah. So my story in this space actually started. I was listening a couple of minutes before to your interview. You were talking about Andrej Karpathy and writing code. I've actually never said this. My story in this space started because Andrej Karpathy wrote an article in 2015 called the Unreasonable Effectiveness of recurrent Neural Nets. It's one of the early kind of blog posts about language models. And it captured me so much that I ended up pivoting my entire company towards it called Sourced. And we spent the next four or five years building language models that were capable of writing code. Sounds great today. Back then 50 people in the world cared. But 2015 was an interesting moment in time because it was followed the year after by AlphaGo coming up. And at that point I built probably an unreasonable conviction that the combination of language models and reinforcement learning should be able to generalize to frankly anything and everything that you can approximate, including human intelligence. And that was kind of my origin story. I met my co founder in 2017. He was a CTO, GitHub.

I made an acquisition for that company.

Wait, you made an acquisition offer for what company?

So GitHub made an acquisition offer for my company in 2017 because we have the world's first code completion models that were working back then.

Makes sense.

Turned down the offer, but nonetheless became really good friends. And this company ultimately ended up not succeeding. We were too early. And so you can kind of figure out what happens when on November 2022, you see ChatGPT come out.

Yeah, right.

It's everything you've been saying for years, kind of into the void, somebody else just did and did it in an incredible manner. And so it kind of gave this really deep realization in that time that everything was about to change. It was kind of like a pre post electricity moment, like we're going to truly fully reach human level intelligence and go beyond. But the narrative at the beginning of 23 was all we have to do is skill language modeling, let's just size them up, predict more tokens on the web and just fundamentally disagreed. So you know, the our view and it's why we started Pool site two and a half years ago was that reinforcement learning was going to become the most important scaling access from all capabilities. First 18 months of life of our company that felt like one of the most contrarian opinions you could hold. I think today it's clearly not anymore, but that's kind of our origin story and why we decided to build a foundation model.

And what was the go to market that you had in mind? I mean we've seen the market play out where there are kind of like synchronous IDE based code completion tab completion models. There's a lot of custom models in that world, almost like I call them like prosumer use cases where I hit GPT5 and it winds up writing code and I didn't even ask it to. And then there's the Codex, there's Claude code, there's agents, there's windsurf, there's cursor and there's so many different positions. How did you see the market map develop from your world?

So we bring it back to what ultimately our mission is, right? We're here to reach human level intelligence and go beyond. And in that world, world intelligence in our view is a commodity. It's actually a commodity that gets created by only a small number of companies because of the sheer amount of resources and kind of compounding efforts that go into it. But I think at the limit we're not going to find very large differences between one foundation model and the other. And so I think as a foundation model company, you're in two businesses. You're on one hand in the what I often refer to internally as the barrels of oil business, you're just selling your tokens behind an API and that's your commodity business. And in a commodity business you care about cost and scale. And they'll come back to a little bit why the infra project is so important. The second though is what do you choose to do with that commodity Once you have intelligence available to you, who do you want to be for? And from day zero, we want it to be for the world's frankly knowledge workforce. We want it to be for the enterprise. We want it to be for the world's most high consequence environments. So our models are now rapidly progressing in capabilities. It's definitely been nonlinear and now we see a path to be at the frontier. But when we weren't at the frontier we kind of decided to cut our teeth in a go to market area, which is really no one else was at which was defense and government.

Oh, interesting.

And so kind of our first customers were not just building the model, but also building all the crazy enterprise stack to be able to deploy it anywhere. Like literally and workstations and humvees all the way to the larger models and like air gapped environments or gulf clouds or places where you needed atm. And now we're kind of on track to start expanding out of that defense sector and going into the wider enterprise. Not just with coding agents by the way. That's been really our starting point. It was our view that's where the market was first going to go adopt for frankly not a very intelligent insight, just the fact that developers, we've always been the first adopters of technology. So it's kind of clear that that was going to happen.

And it's really fun to build tools for yourself. It's less fun to build tools for a role that you've never done.

Right.

Look, intelligent is, we treat it as this one north star like intelligence, but in reality it's actually really multidimensional. How good you are at writing poetry is very different than how good you are at writing code versus how you are being a researcher in biology. And so the thing that was the unlock I think in our space is that first generation of models had only been trained on the out output of humanities work. Right. Like the web. But it wasn't trained on the thought process and actions that led to the creation of that work. And it was going to be clear that coding was going to be the first domain where we could do that through reinforcement learning because we could simulate it. Right. So we spent the last two and a half years building what I believe is the largest RL environment in the world. It's a million real world code bases where our agents can do hundreds and hundreds of billions of tasks. And so, so that's kind of. It was partially close to our heart, but it was also frankly very close to what we saw the path towards more capabilities and models was going to.

Run through, get into verticalization.

Yeah, talk about the core weave deal.

So the core weave deal we announced has two components to it. We've been able to do in the last two and a half years what we did with 10,000 H2 hundreds. Think of that as like annually 100, you know, $21, 50 million worth of compute. Yeah, but it's orders of magnitude less than what others had. And so we built our efficiencies around orders of magnitude more efficient experimentation. But your big model run is still your big model run. And so that like the size of a model you can train has, and the duration at which you train it has clear correlation with the final intelligence and capabilities. So we to need needed a lot of GPUs very fast because we saw now that, hey, we had gotten to a point where our models had gotten so good now that we knew if we'd scale them up, we'd be on track to be at where the Frontier was going to be. And so that really started with conversations with Nvidia and CoreWeave for kind of obvious reasons. And we got really excited because we found the path to partner with Corew Weave that brought online more than 40,000 GB3 hundreds really quickly. And I don't know how much you guys have discussed in the past.

How.

Much you guys in the past have discussed the compute market, but right now that scale of compute is impossible to get. It has sold out for the entirety of 2026 and already well into 2027. So we founded a strategic partnership that allowed us to do so. That gets compute online in December, but it gets you to the Frontier. But what then? You can build the world's most capable model, but if you're not able to serve it, if you're not able to scale it up further, if you're not able to train the next generations, you're not in this race, you're just cosplaying. We had to take a step back. This was already a while ago. This was already started looking at this a year ago. The true bottleneck in our industry is not chips and it's not energy. There's a lot of like 400 kilovolt electricity that comes off the grid. There's a lot of sources of energy in the United States. And while at the limit, it is the bottleneck, it's not the immediate bottleneck. The actual bottleneck is bringing it all together and actually having powered shells like data centers online. Because while last year I could call someone for 50 megawatts and I could kind of get it within six to nine months if I needed to call someone for 250 megawatts watts, guys, there's no one you can call at Least you can make like a multi billion dollar payment or commitment, you know, today on a 15 year lease and then you can get it in 18 to 24 months. Yeah, but I'm not meta. Right. And so we, we understood that we had to own that vertical stack entirely if we were going to be able to secure our future.

When how early did you guys make the call to focus? You know, clearly it sounds like you're focusing intensely on the go to market, market side as well. Right. We've heard, you know, there's plenty, there's labs out there that have like, they're taking the route of like, you know, if we build it, they will come type of thing. And it very clearly, even before you guys are at the frontier, you're saying like, no, we're going to get customers, we're going to get actual enterprise use cases, we're going to be focused on delivering value and then hopefully those products just get better for the customers as the underlying intelligence improves. But do you feel like that? Yeah. Like was that just an easy, obvious decision to make or did you guys like debate that a lot internally?

So it was kind of the DNA from day zero who we wanted to be. And so we wrote ourselves and my co founder like a day zero memo and we often go back to it to see what's changed along on that. And we kind of understood early on that there were three layers that were really going to matter for the next decade. It was going to be energy, compute and intelligence. And in our view a lot of things were going to become rounding errors compared to those three. But you also knew that if you see intelligence as a commodity, if you treat it like a barrel of oil, a barrel of tokens, someone else is.

Going to deliver it.

Right, Exactly. You care about your cost and your skill, hence infrastructure and vertically integrating. But you don't want to be in a commodity business. You want to be in a business that either increases someone's revenue, revenue. Right. Or improves their cost basis. Like it helps them grow their business. And so from day zero we said, well we're not a consumer company, probably hear it in how I talk. It's not in our DNA. I'm not a chat app in your pocket. This is not, I wouldn't know how to build that. But we deeply cared about businesses and enterprises. And so it was a day zero decision to do this. And so from day zero we built both parts of the org, our applied research Org and what we call our.

Production engineering like org, this 2 megawatt facility that feels like leapfrogging a lot of the 1 megawatt clusters that we've seen talked about, whether it's Colossus 2 or what Meta's doing with Prometheus and what OpenAI and Anthropic are doing. Is that the intention or do you think that you're more just like catching up to the frontier and they will be coming online with similar capacity around the same time and you're differentiated on the type of model that you'll train.

So there's big headlines of gigawatts of power and we had one of those yesterday and we have an incredible amount of power on that land. We actually have 6 gigawatts of gas that sits there that we can that we bring turbines online for to turn into electricity. But what I think really matters as a foundation model company is your lead time from when you need compute to scale to how long it takes to get online. And this is where the partnership with coreweave is really interesting because with coreweave as the tenant in the data center, we have the ability to determine ahead of that data center coming online how much of that compute goes to scaling poolside and how much of that compute we might have to put in the free market. And they're of course world class. I cannot speak highly enough of their ability to operate large scale compute. And so for us it was not about the big headlines, but it's about building a company that could increment mentally delivered data centers. And the first 250 megawatts are actually delivered in a quite unusual manner. There wasn't much about this in the press because it's kind of a geeky topic. But I think here we like geeky topics, of course. So if you look traditionally at data centers there are these big stick built buildings. Everything comes on site, is manufactured and put together on site. And that's been the vast majority of data center industry. But mobilizing large workforces and dealing with that complexity means that your ability to scale is not really incremental. I can't add an extra thousand GPUs. Another thousand, another thousand. And so we took a different approach here we've got the big stick build building that we're building. It's a long corridor and we're bringing around essentially data halls of two megawatts at a time. Current generation GPUs be a thousand, GPUs, next generation that's less because they're more dense in power. But what we're doing is we're doing off site manufacturing. So A data center for GPU compute is effectively three layers. It's an electrical skid, it's a cooling skid, and it's a compute skid.

Yeah.

And you actually design them that they fit on the back of a flatbed truck.

Hey, Mark.

Good to see you. We have Mark Benioff. We can go way deeper with you, but thank you so much for hopping on the show.

Show.

We will talk to you later. And we love it.

Yeah, let's run it back very soon.

Thank you so much for joining. How you doing, Mark? Good to see you. Congratulations on all the fantastic news. Thank you so much for joining the show. How are you today?

We don't have audio.

Let's make sure that we have audio for Marc Benioff from Salesforce. He is the CEO of Salesforce. Let's bring him in to the show. Thank you so much for joining us. Our team will sort this out in just a second. We can see you. We can't hear you. We are working on this. Let's make sure that he is in the restream waiting room joining.

I wish we got to get Eso back on the show.

I see him. I don't hear him yet.

I would love to Isso.

Hey, there we go.

How you doing, Mark? We got audio. Thank you so much. You have John Coogan and Jordy Hayes from tdpa.

AGI is here, but we still have technical difficult calling from tackle time, but here we are.

AGI. It's amazing.

Yes.

I mean, it's unbelievable.

We're feeling it.

We've come.

Yes.

All of the AI is, you know, is it ISO kant or ISO con? I can never remember.

I think it's ISO. ISO Kant.

It's ISO con. Is it miso soup or miso soup?

I think it's miso soup.

Okay.

It's ISO Kant. Okay. So ISO Kant. Yeah. I love ISO, by the way.

Oh, really?

I love poolside.

I'm so glad you guys know each other.

One thing you need to know. One thing you need to know, I also love poolside. The music, the. Have you heard?

Yeah. Poolside fm.

Yes.

And they do that great Neil Young cover.

Yeah, yeah, yeah.

If you haven't heard the poolside cover on Harvest Moon. Yeah. And Isso Khan and I were talking about that and I said, why? Why did you call it poolside? He's like, oh, this and that. You know, he's building this huge data center now. And I said, you should split the company. So you should have a software company building your model, which is amazing. Have you seen his model is amazing.

Oh, yeah.

And then he's also building a huge data center. And I'm like, you have one company, which is the model. Poolside one day. One company is the data center. Curbside, of course. So you could have poolside curbside.

That'd be fantastic. That'd be fantastic.

Anyway, great.

Please give us Congratulations. Break it down for us.

Dreamforce feels like something that would never, never get old.

Super bowl for sass.

Well, you guys aren't here, which I guess you're not Metallica fans, which is sad. And I'm gonna have to learn Z. Ulrich know that you missed it. And then also you don't like Benson Boone either. So you're not pop or heavy metal. What are you?

Count us in for the next one. We'll be there.

We'll be there. We'll be there. Next time. Super bowl of sad.

Why? Why? I mean, this is our 23rd Dreamforce.

You have 23 of them.

I want an explanation.

I want to know why.

No, you.

You guys are the number one podcast in the world. I want to know why you're not at Dreamforce. And I want to know your music ch. I want to know your music choice.

I do love Metallica. I grew up on Metallica.

Yes.

Okay. Believe it. What song do you. What's your number one?

Enter Sandman.

Oh, it's a layout.

No, no, no, no.

It's okay to like the.

Everybody knows that one.

Let me ask.

It's. It's. It's. What do you. What do you got, John?

You're Unforgiven.

I'm Forgiven.

I. I do like Master of Puppets. That's a great one.

Master Puppets.

But I. I was always in a little bit more of the Slip knot, A little bit more of the. Of the corn and the tool. But, you know, But I mean, metallic is great. It's definitely in the heavy rotation. It's on my. It's on my metal playlist a bit.

We not only have. Have the greatest heavy metal band of the United States here. Metallica. Really? The Bay Area, Lars Ulrich and Rob Trilo. And, you know, Jimmy Hetfield was here and Kirk Hammett, probably the greatest guitarist of all time, if you like that.

But.

But we also have here Yoshiki.

Oh, yes.

Head of X Japan, who's also one of the greatest drummers of all time. And X Japan also, he does class classical piano and he's about to go on tour again. Mad back to Madison Square Gar. And Royal.

Fantastic Albert Hall.

Amazing. But anyway, hopefully you Guys will come next year.

And works. And works for X, the Everything app. That's correct. Right? That's great.

Well, X. The Everything app.

This is the different X. Yeah.

This is X Japan. Not related to.

Okay, got it. Got it. I. I saw. I saw your post about being on X, and I wasn't sure if it was a different thing.

This is. We are.

Okay.

We are X.

Yes.

It's never been more Japanese. Your Japanese viewers are going to go up.

Oh, no.

Say, we are X.

We are X.

Which is Yoshiki official. Yeah. Or whatever it is.

Yes.

We're.

This.

This is so fun. I. We got it. We got to do this more often. We were joking on the show earlier. We were referencing a post, you know, the whole, like, nothing ever happens investment meme. Right. Which is like. Which somebody was highlighting that OpenAI runs on both Salesforce and Slack. Wanted to ask you. Does not. Does nothing. Does nothing ever happen?

Nothing else matters.

Nothing else matters.

There you go.

Yeah.

You're still unforgiven, by the way.

A super intelligence would use Salesforce and Slack for sure that. That would be the first thing that they pull off the shelf.

If you are a Slack user, you need to see the Slack bot that we introduced here, which was built on anthropic. It's also amazing. So we use OpenAI. We use anthropic. We use Gemini. I just did an interview with Sundar. You can see it on YouTube. Xai love Elon. We love all our children very equally here.

Yeah. How do you think about vertical integration? I mean, it's a fascinating company. You've had the opportunity to build your own cloud. I don't believe you've ever built your own operating system. But many companies now are saying, we want to build our own asic. And you've probably. I don't know. Have you ever thought about putting Salesforce on an ASIC or Slack on an asic? Like, how do you think about when to verticalize versus when to partner?

It's such a great question. And you probably will enjoy my interview with Sundar, because I asked him that because, you know, you think about it, he's really got it all together. He's got the data center, he's got the chip. Right. Tensor. Right.

Yeah.

He's got. He does a lot of the. The operating system type work, including the model, you know, the Gemini model, the applications he does as well. And he even gets all the way to the robotic layer. Right. So here he's got these waymos running around the street. What other company Is going from robot to chip, you know, to data center really only, I mean.

And it's a cancer cure.

Well, what is the future? Is it billions of fast food restaurants on the planet, you know, all having. Having their hamburgers made, you know, by robots. And that's the world we're about to enter into. I mean, I have no idea. But the one thing I know is if you look behind me, there's no robots walking around. Well, there might be one, but not making the burgers here.

Not yet. Not yet. How do you think about the different foundation model brands? Do your customers want to know that they can pick. Claude, is model switching important or is there a world where you to want wind up wrapping that at an abstraction layer and you're just serving them your product and it's powered by the different foundation models, but the consumer doesn't really care.

This is a great question because number one, you may know that Salesforce also has a huge research team. We invented prompt engineering. The first prompt right here at Salesforce Research, commercialized by others, of course.

Hit the gong for the first prompt. Hit the gong.

Thank you for your service.

Thank you, thank you. And listen. Yeah, look, some customers want to use our models.

Yep.

Some customers want to build their own models.

Yep.

Some customers like a certain brand of model, by the way. Some countries like a certain model. You just saw. ISO Kant, was it ISO or ISO? ISO.

Right, ISO.

Isocant. He's important. Portugal. He is going to make a play. He's going to have a data center in Texas, but he's also going to have a data center in Europe. There's Mistral in France. There's Middle east models, there's Asian models. You know Kwon from Alibaba. You didn't mention that.

Yeah, right.

There's models all over the world and different customers and different geographies are going to all want different models based on where they are, the kind of customer they are. They may want small models, they may want large models, they want micro models, they want foundation models. There's a lot of different models. So you have to give people choice. But what we're going to provide is this incredible platform that we call the agentic enterprise. And we have Agent Force. And Agent force is powered by all of those. And you can plug into whatever model you want, wherever you are, with the data residency, the governance, the compliance, everything you need, boom, right into Agent Force. Because look it here at this conference, you're not here, unfortunately, and I hope you do come. We've had 23 of them by the way.

But.

I know, but here's the thing.

Listen, Jordy's barely 23 years old.

One of our.

One of our team members over there.

Doing podcasts all day long, all day long. All day long. Doing this podcast going. But listen, listen, I get it. I know you're too busy for me. But now listen to this. Listen to this. This is.

This interview is so fun. We're.

We'll fly.

We'll fly to Salesforce. Salesforce Tower.

No, I know you're busy. I get it. I got it.

You're too busy. You're master.

No, no, I understand. You are not Metallica fans. You're too busy for us. Fine. But here's the thing. I want to just say this to you. Look, there's customers here from hundreds of countries who have flown in from all over the world. And they all have different needs, different industries. They're different size. They're small businesses, companies like zero to 200 employees. There's 200 to a thousand employees. A thousand to 2,000 employees. Very large enterprise. The Pepsis and the Dells. Michael Dell was here. You're not here. Michael Dell was here. Okay, okay. Laura Albert, the CEO of Williams Sonoma, is here. You're not here. That's fine. But here's the thing. Hold on. But here's the thing. Also, it's about governments. It's about software companies. You know, it's about. All of these people are here because it's a highly diversified market. Software. And these models don't just have to reach consumers. They have to go through these businesses and go B to B to C, you know, from business to business to consumer.

Consumer.

And that is actually complicated because you have different governance, like I was saying, different compliance, all kinds of different rules based on country based on. So it's. It's not like, you know, we're. Look, we're here. We're Americans. We're here in America. You know that we love our country and we have certain models that are built here. Yeah, great. Okay, we love that. But here's one more thing.

Thing.

There's many people who are not Americans who are here, and they have to comply to their laws in their countries, in their languages, in their governments, and we have to adapt to those organizations and those countries as well. So you have to think about AI as a global phenomenon. You have to think about AI as a highly diversified phenomenon, and you have to think of AI as the future. And we're just in. We're in the current moment. We've been doing AI at Salesforce for 10 years. We did Einstein 10 years ago. You know, obviously, AI has been around, I think since the 50s, the 40s, so. And you know, like the guy who wrote Minority Report and War Games and even Deep Impact, he's our futurist. He's 78 years old. He worked at JPL on the Apollo mission. He's one of the writers on all of those things.

And he's there, but we're not. He's there, yes, he's there, but we're not here.

But you're not. Everyone is here, but you're not. But here's the thing. Well, this is the largest tech conference in the world. You know that 50,000 people are here. The largest vendor led tech.

Hit that gong again. Mark, one question for you. You Salesforce has been hugely acquisitive. What is your pitch to founders in 2025 that you want to acquire?

That's great.

Like why, why is it such an acquisition opportunity?

We have bought more than 100 companies. You're 100% right. Over 26 years, we've bought more than 100 companies. We've invested in hundreds of companies. We own 1% of anthropic. You know, we just sold a company to Google, which is Whiz, the security company.

Oh, yeah.

We, you know, we grew Snowflake and took it public. We're investor in many companies. So we want to talk to you. If you're an entrepreneur, we want to invest in you, want to help help you grow. We want to bring you here, introduce you to all of our customers. We have 50,000 customers here to introduce you to who want to buy your product now. So come here to San Francisco or come to one of our world tours.

All over the world.

You know, from here we go on the road and we're going to be in every country in the world, every major city, and we will come and meet you. We want to meet you. We want to know you. Salesforce Ventures is our army that acquires but also invests. It's a huge part of our business. It's a five billion dollar fund. It's, I think, doing a 33% IRR. I don't know exactly the number. It's probably one of the highest performing funds in the world.

That's great.

Let us partner with you. Yeah, you know, come here. You know these podcasters, NF time. You do come here, be with us. Customers are here to meet you.

You're good at sales. Kind of got to be in the role.

What's the current thinking on seat based pricing versus consumption both at Salesforce or what you advise founders that might be?

Yeah, I think, I think people have this obsession with, with seat based pricing right now and, and this idea that, that AI may, may allow people to run. We're not seeing it a lot yet, but companies more efficiently. But in my view, I think companies are smart enough to, to understand they buy, sell, even if they're paying on a seat base, they're paying for value at the end of the day and if they're not getting value, they'll. They'll churn or they'll find another. So. But how do you think about this conversation around seat base versus kind of value based pricing?

What a great question. Look at. I just subscribed to ChatGPT. It's a seat based model, you know that you probably pay 20 or $200. I just subscribed.

Is that, what did you use that, did you use that by the way to, to make that picture with, with Sam? It kind of that, that picture you.

Guys grock because we were wearing like conference necklaces.

Okay.

And I said take it off. And chat GPT goes oh no, no, no, no, no, no. Copyright infringement. And then, and then Grok said great, I'll do that. And do you want me to put in the photo of the podcasters with you? I said I'd like to do that also. But you might have seat based pricing like those are. That's a good example. Right. You might have all you can eat pricing. So we have the agentic enterprise license agreement. All you can eat. Don't even think about pricing. We're going to give it all to you. Don't even worry. Three, it might be per transaction or per action pricing because you're more conservative. You want to know exactly what you're paying per action per transaction or it might be some other model that, some kind of flexible agreement. Right now, what we're having to do, and this is a great question, so thank you for asking me the question. We have to write pretty much a custom pricing agreement for every single customer. And as you know, this year we're going to do about 43.1 billion in revenue. And, and we just gave guidance yesterday that we expect to be able to do $60 billion by fiscal year 30. So that's, you know, it's still very fast growing, very exciting.

Fantastic.

But when you have. We have 150,000 maybe customers on our core. There's a million and customers on Slack. Hopefully you guys use slack.

We do.

Of course, when you have that many customers in that many countries and that many size companies, like I mentioned, it's across the board, it's every possible need. And so we have to adapt more than ever before. When we started our company 26 years ago, we had one product, one price, $50 a user per month for our sales product. But we only had one product, one price, one. One type of customer. That is not who Salesforce is today. With Slack, with Tableau, with Mulesoft, with our sales cloud, our service cloud, our marketing cloud, all, you know, our Agent Force 360 platform, all of the things that we offer, dozens of products, we ultimately have to adapt to the customer. And what I can say to you is Salesforce's core values, which are trust, customer success, which is right, what we're talking about right now, innovation, equality of every human being, including pay equality, and also sustainability, which is our Trillion Tree initiative and all the work we're doing for the oceans. And if you were here, you could see what we're doing. But listen, the number one thing, number one, you can say those are our core values. At the end of the day, it's all about customer success. Are you happy using Slack? Are you happy using our sales cloud? Are you happy using our service cloud? Is Agent Force fully satisfying you? Are you getting. You know, you guys have a huge business now. You guys are growing your business. Are you. You know, like, by the way, for example, you probably heard of the Beast, Jimmy. Jimmy Donaldson, his team is here. They're built running on Salesforce. They're a huge media company now.

Everybody, Mr. Beast runs on Salesforce.

You know, the Beast.

The Beast.

The Beast.

He uses Slack very aggressively, of course. Jimmy is amazing, okay? He runs a great business. It is growing like a weed. Beast, Media, all of these things. We have to adapt to him.

Yeah.

It's an example I'm giving you because I know you can relate to that in your own business, that you guys are trying to grow and make money and have fun. And look at this, look at this great life you guys are. You probably feel blessed every day that this is your career.

We do.

We do.

And I feel that myself. I feel blessed everybody day that this is my career, that I'm able to do this for a living. Last night, James Hetfield, probably one of the great singers of our time, I'm going to introduce you to his music. Listen, number one thing, he said, he's on stage, he said, Metallica for over 40 years gets up on stage and plays that song that you mentioned, Enter Sandman and You know what he said? I feel blessed that I can do this every time I get on stage. We feel blessed that we're here at Dreamforce. We're blessed every time we get to meet with a customer, every time we have an interaction, every time we are able to write a piece of code, sign a deal, see customer success. It's what gets us up in the morning. It's why we do what we do and it's why we're here at dreamforce. And after 26 years, I still have the passion, the energy and the excitement and the vision for Salesforce that I had 26 years ago because that's what, what makes me happy. I love making customers successful. I know that's what you like. You love making your final. Your viewers.

You're the final Boss of Enterprise SaaS. I absolutely love it. What makes. Why, why do you believe that sales is still a great career path for young people? I personally believe it is. I believe it is the most valuable skill in the world. But I think a lot of people see all these AI agents for sales rep companies getting funded and they, they might think, I don't want to go down that path because I don't want to go down the sales path because I think that's going to get automated away. Why do you. What, what's your kind of worldview there?

Well, it's a great question. I mean, you know, at one level there's between 20 and 100 million people we actually counted them that we have not been able to call back since we started Salesforce 26 years years ago. Between 20 and 100 million people. We didn't call them back not because we didn't love them, not because we didn't like them. We didn't have a people. And, you know, those are. That idea to be able to call everybody back, you know, how do you do it? Like, you can't call every listener back who's already contacting you. You know that.

Yeah.

So one thing is, yeah, you're going to have a SDR or sales development representative. The ability to call everybody back. Okay. The ability to qualify, to have a conversation, but listen face to face sales or face to face communication like we're doing right here. Like, by the way, I'm. I don't. I'm not an AI. As far as I know, I'm here. I'm not a biological computer running an LLM. At least I hope I'm not.

I feel like an AI would grab a water bottle and hit it like that.

Very suspicious.

What I want to say. I wanted to say this to you. I also just hired somewhere between three and 5,000 more salespeople.

Wow.

Okay. I'm growing my Salesforce. It'll be more than almost.

It's.

I think I'm going to try to get to 20,000 account executives this year. That doesn't include systems engineers, managers, the infrastructure sales engineers.

Yeah.

We have 80,000 employees at Salesforce. 80,000. And about a quarter of them. Okay. Are just people who are trained in our product who are designed to help you. We want to make sure that they can. Can help you be your success. That is why, at the end of the day, I think sales, listening, asking good questions, empathizing with the customer, connecting deeply with the customer, having fun, which is super important for us and for you. I know. I watch everything that you guys do. You love enjoying and having a good time. We want to do that, too. Isn't that what sales is really all about?

About?

And tonight, you know, look at the conference is happening, but across the street is the St. Regis Hotel. And last night I was there, and I'm coming down. It's a long day. It's a long two days. I haven't slept in two weeks. Coming down the escalator, looking, the bar is filled and the bar is filled. And what is happening in the bar, do you think? And it's not our people who are in the bar. I looked. It was all the customers talking to each other and connecting, going, going more deeply, you know, having that human touch, you know, because, look, AI, we love AI, okay. But AI, it's not the same. It's not. AI doesn't have a soul. It's not that human connectivity. It's not, you know, at our depth, AI is not born. It's, you know, AI is made. So, you know, we. There's a human humanity that still needs to be nurtured.

Yeah. People want to buy from people and.

Grown people want to buy from people. Thank you. I mean, that is what it's all about. And that's why I think sales actually is. And you can see right here, my job right now is to sell you to come to Dreamforce and also to look at our products to core. Sell you our core values.

I love it.

That's my job right now. Right? That is my job. Yeah, Amy. And how quickly.

How quickly can you clock if somebody's going to be great at sales? Does it take you, like 20 seconds, 30 seconds? A minute?

We've all seen. We just had a great conversation about Mr.

Beast. Yeah.

Would you say on a scale of 1 to 10. The 10. One of the best communicators, most aggressive salespeople in the world. Where would you put them? One or ten?

Ten.

Ten.

Ten.

Yeah. And boom. That is important. And I think that that is. And you see, like all of a sudden you're watching this thing. One of the games is happening. He's buried somebody underground for two years or something. And either digging the person up. Are you still alive? Whatever. And then just as they break, they go. And one more thing. I've got a chocolate bar to sell you. And this is the best chocolate bar you have ever tasted. Let me have the guys coming out of the ground now. Hey, will you try this chocolate? What do you think? Look, I've been underground for two years. This is the best chocolate I have ever had. I mean, he has to be one of the great communicators, but also one of the greatest salesmen I have ever seen. Is that valuable in the age of AI?

100%, absolutely.

100%, absolutely.

All right, where do you get.

Last question.

Where do you get your energy? It's off the charts. It's incredible.

I'm getting it right now from. You guys are obviously totally off charts. And I'm just vibing. I am so tired. So trashed last week. I've been on the road non stop. I haven't slept for an hour. I don't know where I am. I don't know what conference this is. I don't know if that was eso or myself.

You're doing great.

You're doing great. Thank you.

I wish we didn't. I wish we didn't have to fade to black.

We'll talk to you soon.

You see what I did there?

You're still unforgiven.

I think I have.

I think I have a clip of me playing a Metallica song.

I don't remember which song.

We'll send it to you.

I'll send it to you.

We'll work.

Thank you so much, Mark. We'll talk to you soon.

See you.

Have a good day.

Super fun.

Bye.

Fire me up.

Yeah, the mogging counter.

We got relentless. We gotta. We gotta add that.

That's the chat.

What a.

What a legend.

Having a lot of fun at our expense. Wouldn't have it any other way. Everyone had a great time. Thank you so much to Mark Benioff for hopping on the show. I. The show is in chaos at this point. The timeline. Timeline is in turmoil. Our show schedule is in turmoil. I believe we have more guests I hope so.

We have Alice coming in.

Bring in someone to talk to. Sorry. Hello, how are you doing?

Welcome to the show.

Sorry for all the chaos. Marc Benioff really knows how to take control of an interview. It was a fantastic conversation. But we are here to talk about you. So please introduce yourself and give us the news.

Yeah, great. Hi, I'm Alice Bentinck and I'm the CEO and co founder of Entrepreneurs First. And I'm coming to you live from the EF office where behind me I have 40 of the most incredible founders that were pitching at our demo day yesterday. So we had Jack Clark, who's one of the founders of Anthropic, who kicked off demo day, did an amazing opening talk that included the phrase I'm deeply afraid, which I think is always useful for grabbing the attention to share. Yeah, we had 200 attendees, an amazing selection of our previous co investors and the kind of leading lights of the Bay Area fundraising ecosystem. Partners from Khosla, Andreessen, Zeta, Suse, True Ventures, bases that and we had 20 companies, 20 companies pitching the next generation of largely AI related products. And it was wild, standing room only, Amazing vibes. So yeah, it was a great day.

What are the most common trends? I mean we've been to a couple YC demo days now. We and being actually like in the room and talking to so many companies, you start to pick out really clear themes. What's get us to the frontier? What's the October 2025 theme or different or even like anti theme.

I think the key theme for this one was really young founders, early career founders who are obsessively solving some of the most tricky problems in like old, old industries. So companies like Faction, which are these two insanely ambitious young guys who are automating the ordering process for industrial distributors and it's like where do these guys get these ideas from? And that's actually part of the EF process is because we work with individuals pre team, pre idea, we help them go through that process of actually developing the idea from scratch and canal the CEO. He'd actually spent a summer spending eight hours a day hand processing these sort of purchase orders, these big, you know, old industrial distributors. And he's like, right, I'm going to become a founder and I'm going to make sure that no other intern spends their summer doing this. And it's the most incredibly lucrative industry, an enormous industry. But I would say that's the theme. You know, another example would be got a great company called Chatter and what they're Doing is it's like Vanta for fmcg, so it's automating the compliance process. But for fragrance companies, for cosmetics companies, this is something that's holding back $400 billion worth of new products every year because it takes months and months to go through.

You were just talking. I had lead.

Well and probably a year ago at this point I had somebody pitch me this idea of something they wanted to do. But I believe, I don't think they actually went and did it. But I was super bullish on the idea at the time. What kind of guidance were you giving the batch on kind of metrics around what would be compelling for the investors that were going to be in the audience around? What's the bar to come out of Demo day with a seed round from your view?

It depends on the company, it depends on what they're trying to build. Half of this batch had more than $100,000 worth of, of traction. And these are really, really young companies. So because we build all the companies from scratch, we build co founding teams from scratch, we have this really unusual and unique process that creates companies. The companies that were pitching at demo day were incorporated often less than three months ago. So you're getting to really, really impressive revenue traction within a short amount of time. I think one of the things we do know that investors are looking for right now is the stickiness of that traction. And particularly because a lot of our companies are selling to older industries, the revenue is super sticky. So it's less the sort of the J curve revenue where you here today, gone tomorrow, you've got to rebuild it often our company's getting year long contracts, multi year long contracts with these big old companies. So yeah, I think one of the things we're really focusing on right now is the stickiness of the revenue that they are creating.

Last question from my side. How do you apply and do you have any particular application questions that you think make your application stand out so.

You can apply, but honestly, we like to find you.

Okay, don't call us, we'll call you. I like that.

One of the ways to think about EF is a little bit like, you know, caa, the talent agency. We see ourselves in the same way. We want to be the CAA for founders. We want to be in there in their corner as their talent agent. And so we go out into communities across Europe, across the US and across India and we build the relationships and the connections to actually work out who are the individuals that we should be going after. And because we're Going after them at the point where they're pre company, they're pre team, they're pre idea. Our job is to work out from their behaviors which the ones that stand out. So I suppose some of the behaviors that we focus on are pace productivity, their ability to build really fast. But you don't see that in an application. You only see that from spending an extended period of time with these people. So our selection processing includes a hackathon. It means that we're spending two days, you know, 48 hours pretty much straight working alongside them, seeing what they build, seeing how they interact with others. And honestly, the interacting with others bit, they don't always play nicely with others. We're not looking necessarily for team players, but it's really about the behaviors. You can't really assess an individual fair founder potential by looking at a piece of paper.

Yeah, well, congratulations.

Hackathon is very cool. Selection process hit that gong for the whole batch. It's great to meet you and congratulations.

Hope to have you back on the show soon. We'll talk to you later.

Thank you.

And we have Eric Sufert from Mobile Dev Memo in the Restream waiting room. We will bring him in to the TBPN Ultra Dome. We are getting back on track. Eric, how are you doing?

What's up?

Sorry to keep you waiting.

Sorry about that.

How are you? I'm great. I've enjoyed your guest appearances on podcasts, your writing, everything, especially some of your more recent takes. But maybe for those who aren't super familiar, would you mind giving us a high level, just introduction on your day to day how you describe yourself to folks these days?

Yeah, I call myself an independent analyst. I run the website Mobile Dev Memo. It's got a blog, podcast, newsletter, Slack Community. I wrote the book Freemium Economics and I spent kind of my operating career in the mobile space, mostly in performance marketing roles and strategy roles. And now I just kind of blog podcast full time and invest out of a fund called Heracles Capital.

Oh cool. It seems like one of the main, I don't know, takes you have that you're just waiting to cash in on is this idea that OpenAI will be doing doing advertising soon. Can you walk me through like how you perceive the messaging from OpenAI the firm and then what led you to believe that it is inevitable that we will see advertising?

Yeah, that's my modus operandi is strong opinions forcefully stated. And I've said very forcefully OpenAI will monetize with ads. I wrote a piece in May, the title was Obviously, OpenAI will monetize with advertising. And here's where I think, think the, the sort of like most profound clue has been, right? So Sam Altman, in Ben's podcast last week, Ben, Ben Thompson, he said he pulled off like a rhetorical sleight of hand. Right? He said like, hey, look, you know, we're not really considering ads, but I really love what Instagram does. Yeah, those aren't ads. We wouldn't do ads. Ads are attacks. But I love what Instagram does. Instagram has surfaced relevant products to me and I bought, bought products from those ads. Those aren't ads. Yep, that's a separate.

Yeah, those are, those are improving my life.

That's discovery.

That improves my life. Ads by definition, you know, deteriorate engagement. But, but these I extract value from.

Yeah, right.

And that was a Spencer Neumann moment. That was the Spencer Neumann moment. So Spencer Neumann, Netflix CFO, March 2022, Morgan Stanley TMT Conference. He said, look, it's not like we have religion against advertising. It's not in plans our right now. But never say never. But it's not in our plan. It's not like we have religion against advertising. That was the Spencer Neumann moment. That was March 2022. When did Netflix launch ads? November 2022. Obviously they were already working on it.

Yeah, right.

And by the way, I would, I would, yeah, I would dispute that statement that it's not like we, we have religion against ads. They very much exhibited religion against ads for years and years and years up to that point. But that was a tipping point. That was a tipping point when they indicated the market that ads are coming. They're inevitable. I've said ads are inevitable with OpenAI, with ChatGPT. I said yesterday, they're ineluctable. I got made fun of word. I've used it a lot, by the way. But they're inevitable they're coming and I think they've been working on that for some time.

Yeah.

Here's what do you think about ads where you pay for something and still get ads? Because for me, I signed up for some HBO plan a while ago and I thought I was signing up to not have ads, and now I've been a little bit lazy. I usually watch TV before I'm falling asleep and when I get like a series of like when I get like four minutes of ads before HBO and I'm paying for the product, that's to me a really bad user experience.

Well, I mean, it's a choice. I mean, there's trade offs, right. I Think, you know, so with any sort of, with any, with any digital product that has the capability of becoming like having a humanity spanning tam, the sort of, the imperative is to reduce consumer surplus surplus to as close to zero as you can, right? Because that gives everyone an accessible. And to give everyone an accessible price point, right? Some people, the only accessible price point is zero, right now for some people. Now you have to weigh that against perception, right? And there's perception risk. Like Netflix. I had sort of argued at one point that I thought Netflix could go fast. So freeming ad. Freeming, sorry, free ad supported tier and just drop the price point to 0 and a half ads. I now I've changed my mind on that. I don't think they should do that just because it would, it would, you know, deteriorate the perception of the brand.

Right.

They have a perception of being like a high quality, you know, content studio.

Right.

And so I think that would be a mistake for them to do so. It's just a trade off, right? Do you want to maximize TAM or do you want to sort of preserve some other asset? In Netflix's case, it's the perception of being like a high quality service. I don't think ChatGPT has that risk. Right. So I think ChatGPT can introduce. So first of all, when we talk about ads and ChatGPT, we're not talking about ads in the paid tier, just want to be clear about that. We're talking about ads in the free tier, right? And so yeah, that's, that's kind of.

What I was getting at. It's like if I, if I'm on the 200amonth plan and I'm getting slammed with ads like that's going to be, people are going to be really mad about that. But for people that are on the free tier, seems very fair.

Well, I mean that would create a lot of animosity on the $200 tier. But I think like. So yes, yesterday the Financial Times posted some data about OpenAI. So 800 million users, we knew that 5% are paying, right? So you got 40 million paying users, 13 billion ARR, but only 70% of that is from consumers. So when I saw a lot of the analysis, analysis about these numbers, a lot of them sort of anchored ARPU numbers to the overall 13 billion. But you've got to reduce that to the 70% that are this consumer. The rest of it is enterprise API access. So really that backs up to $228 annual, not ARPU, right? It's not ARPU. It's ARPU. It's our ARPU average revenue per paying user.

Right.

But if you, so if you, if you actually, if you do the calculation for just ARPU, it's $11.38 cents ARPU.

Right.

So how does that compare? That's, that's less than Meta. Metas is 1365. This is global.

Right.

It's much higher than snap. Snaps is 287, pins is 164.

Right.

But I mean the thing is like they're pretty close to meta's, right. On an ARPU basis. Now those 70%, sorry, the 95% of non paying users are essentially freeloading right now. The issue with OpenAI for that, and we saw that in the numbers as well, is that inference does, isn't free inference, you know, marginal cost of, you know, of providing that good. And so they have to find some way to close that gap. And I think introducing ads to the free tier will not meaningfully impact engagement. I think it could be additive to engagement. I think if you look at a ChatGPT chat, it kind of feels like a feed. I think the real challenge with, with ChatGPT and I've seen like a lot of conversations about well, the difficulty is going to be be like the attribution or it's going to be the sort of middleware that connects things. I don't think, I think this must be a solved problem. I don't think attribution is going to be any different. Attribution meaning how do you credit ChatGPT for driving a sale or something? I don't think that's going to be any different than a traditional social media ad for E commerce or like an app. I think the challenge is really going to be coming up with a format that is, that feels native for the experience without calling into questions the unbiasedness of the response. And that's going to be the real issue with ads. And I think it's also kind of going to be an issue with instant commerce.

Yeah, walk me through. When do we get to the line of as we're rolling out ad products at OpenAI hypothetically, when do we get to the. Okay, there's backlash. That's too far. It feels like if I see just an ad in the Sora feed that feels very native to Instagram. If I see, see an ad in my Pulse News feed that feels very native, that seems very reasonable. But if I ask it for what are the best paper towels and it shows me an ad and it feels like the ads team is actually Influencing the editorial, if you could call it that. It feels like right now you're getting.

The average product recommendation from Reddit, right?

Yeah.

And then if they start to put their kind of, they start to kind of tilt the scale.

Yeah. What are the landmines in terms of like just describing the nature of the firewall between the results. When you ask ChatGPT a direct question and ads, it's a bright red line.

You can't cross it because once you lose consumer trust, then it's gone forever. Right. If a customer thinks that the answer is not objective, that's actually influenced by whoever's just paying the most to show the product to you and you lose the consumer trust. Right. And actually that's like, that's a very specific problem for chatbots. It doesn't exist for Google Search, Right. Because Google wants you to click with any query, Right. You query for something, they want you to click. Now, when everyone's interests are all aligned, the advertisers, Googles and users, if that first link is the most relevant, well, you can kind of measure relevance through the price, the sort of modified price that the person pays for the ad, because the sort of the end price that an advertiser pays is actually modified by the relevance. So with what. So if you just see a stream of links, at the top link is an ad, or if it's not an ad, if it's most relevant, either way everyone wins. Right. But those interests are not aligning that way with the chatbot. Right. So I think once you start inserting ads into answers, you're going to call into question the unbiasedness of that answer and you're going to make users question whether they're actually getting the most relevant information. They're just getting the information that that was, that was paid for, those paid to be shown. But the other thing is like, so, so a lot of people think that instant checkout is like kind of precursor for ads, right?

Because.

Well, there are some partners now and they kind of, they all serve E Comm, but like they might serve different categories, right. So there's, there's probably not too much overlap in the type of products that could be served in any given instance. But there will be at some point and once there is a lot of overlap, well then how do you mediate that overlap? You run it through an auction. Well then it's ads, right. I actually don't think that's the case. I don't think that's what they're, what.

They'Re trying to do.

Here, you know, so the issue with that is what we know about instant checkout, and it's kind of light on details at this point, but it sounds like the way they monetize that is there's a flat percentage fee that gets applied to the product price, right? And so that's actually economically suboptimal for Open Air, right? Why? Because, well, if the lowest price item is the sort of most relevant and that's what gets shown, then they're going to make less money on that. Right. And so the issue with that is, like what you'd rather do is you'd rather combine the relevance with the bid, right? And that takes into account the price of the product. That's what the advertiser is willing to pay. And so if you modify that into like an expected value and you rank the potential ads on that basis, then that's economically optimal, right? So OpenAI would be leaving money on the table by just sort of saying, okay, well, let's do an auction and. But sorry, just with the flat feature fee, and then we're just going to decide what to show there. And the other problem with that is from a merchant perspective, you don't have any control there. So the thing is, you can't change the price to absorb that margin shock with ads, you price that, you price that specifically, you price the bid that you're willing to pay. That makes sense for you. That allows you to recover the ad spend but also make money on the checkout. But you can't really do that with the instant checkout because you just submit like a JSON of your catalog, right, with the various price points. So I think instant checkout. The point I made in my blog post about this when it was announced is I think it's probably just a way to bootstrap conversion data for users to ultimately target ads against. But I don't think that's going to be the ad surface area. I think the ad surface area. And so first of all, I think it's a mistake to think that ads need to be anchored to the content of the query. Right? That's sort of like a contextual targeting way of thinking. But that's not what Facebook does. That's not what Instagram does.

Oh, that's right.

You see an ad on Instagram, it's not because the previous piece of content you saw was related to the thing that's being advertised to you. It's all based on what there is a history of conversions for anchored to your account. Right? So these things don't need to be related. They don't need to be. There's no, there doesn't need to be like a direct relevance connection to the actual content of the chat.

Right.

It could be totally unrelated, except for there's some logic that came up with the idea that this ad should be targeted to you because based on previous purchase data, you seem likely to buy that thing.

Yeah, yeah. It knows that you need paper towels. We're going to show you an ad for paper towels. Doesn't matter if you went and searched for, you know, the history of the Roman Empire.

Yeah, we have, we haven't talked about the Pulse yet either. Which is another effectively feed that you can just slot in whatever you want in between the content. And I don't think users would have any issues with that.

Well, no, but so that, I mean, I think that could be it. So that's what I, that's kind of what I meant when I talked about, like the format's gonna be the real challenge. Like, I mean, and they've got, you know, exceptional people to tackle these problems. And I think one thing about, you know, OpenAI that people sort of like discount is, is the number of people that they've hired from Meta. Right. So I mean like when people say, well, they don't have a team yet. Of course they have a team. They've been hiring people from Meta for years and years who works worked on ads ranking because, you know, they worked on machine learning, therefore they were working on ads ranking. Yeah, right. So they've, they've tackled these problems exactly in their, in their career. Where I think ads will be very well suited for or what I think ads will be very well suited for is like the App Store that they are, that they've announced that they're building.

Right.

So what they said was, okay, we've got these app integrations, you know, Zillow, that's great. It's actually pretty functional. And a number of others canva, Figma. But at some point we're going to have more app integrations than you can remember. And so therefore we'll have a directory. Well, when you have a directory, what sort of like natively follows ads?

Baby, I love it. All the roads end in ads.

What about on the SORA side? Do you think they can transition from a creative tool to a actual consumption?

It's still number one in the plat in the App Store, by the way. Yeah, it's fascinating. Yeah, it's very sticky.

No, so I think I worked in, I've worked in mobile for Too long to be like very optimistic about the staying power of an app that's been number one for like a week or even.

Yeah, we had, we had this conversation earlier where I was like, I still put it at like a, I would say like a 10% chance that it becomes a content consumption platform that people are spending like on the order of Snapchat or Pinterest.

Pinterest or TikTok.

Yeah.

First of all, I don't know if it was a great idea to spin that out into its own app. It may have been. But the thing is, OpenAI was not short of top downloaded apps. They've got ChatGPT and that was a top downloaded app. The thing is you look at the staying power of these generative apps and it's, it's, you know, they're pretty short lived.

Right?

I mean like, you know, Lensa, what's that?

Lensa, you remember that?

Yeah, the original light bulb.

That's right.

Yeah.

But like, remember Deep Seek, right? Like, you know, I mean the thing is like, you know, and then whichever, whichever company releases like the sort of the latest next gen model, they usually takes the number one spot but they sort of fall off after, you know, kind of a short amount of time.

But do you think really not idea. Do you think Deep Sequence was paying for downloads? Like it didn't seem organic at all.

Yeah, that's, I looked into that at the time. I don't actually remember what the conclusion was. I think they probably were.

I mean they were giving away the first reasoning model for free like a week earlier than open AI. So there was just like, it's the hot.

The general public was not like, I need to try a reasoning model. I didn't see anybody outside of our bubble talking about Deep Seek and yet they were charting number one and it just felt totally manufactured to me.

Yeah. One, I don't know if you have a hard stop at 2, but we will let you go. But I'd love to know your thoughts on TikTok. There's a deal obviously in place. The valuation always felt low, but at the same time like competition in that category is pretty aggressive. YouTube has a very serious competitor. Instagram Reels is a very serious competitor. Now you have Sora. How are you thinking about TikTok in the, the, in the ecosystem right now?

I mean, TikTok not getting banned was a foregone conclusion. I write an annual predictions post in my post for 2025. One of the predictions was that TikTok's not going anywhere. That was priced in there was no movement basically in Snap, stock price or metas, I think. So TikTok is facing some challenges with its social media shopping product.

Well, the challenges that they're losing, it's a massive amount of money. It's not really. It's like a retail platform that is massively loss making.

Yeah.

And they scaled back the team there. I mean, just social commerce in general, I think is, I don't want to say a dead end, but it faces challenges in the West. But I don't think there was any news with it with this deal being reached. I think when people get worried about the algorithm being retrained, keep in mind that's the content algorithm at the ads algorithm. I really don't think we're going to see any meaningful change in the ecosystem as a result of this. I think though, because you mentioned, you know, YouTube being a competitor. I think the big news this week was the deal that got struck between Spotify and Netflix. That seems like a big deal. Right. Because they're sort of frenemies given the risk that YouTube poses to both their businesses.

Yeah.

And so I think it's really interesting that they're teaming up. Spotify is bringing its video podcast onto the YouTube platform. Sorry, sorry, the Netflix platform.

Yeah.

Not on YouTube.

Yeah. I think Ben Thompson called it the anti YouTube coalition. Right, right.

Well, but a similar coalition used to exist between, you know, amongst and at various times with different affiliations between Apple, Google and Facebook. Right. So I think these battle lines are being drawn there. But you know, YouTube is a CTV behemoth. Right. They, you know, they announced that TV is the largest consumption platform for the service. It's a behemoth. And you're seeing Netflix trying to sort of poach YouTube talent onto its platform and brought over Ms. Rachel.

Right.

So I think there is kind of like a war brewing, if not sort of raging at the moment across, you know, YouTube and Netflix. And it's interesting to see this alliance form between Netflix and Spotify.

Yeah.

Did you have a reaction to. Ben Thompson was kind of noodling on this. Take that. Essentially it's very hard for the second generation of YouTube creators who have built their entire business just on YouTube, YouTube to then take a Netflix or Spotify deal and get off platform because they're so tied into the ecosystem versus someone who maybe has subscribers on their own website. And then, yes, they use YouTube as a distribution platform, but that's not. If they turn off YouTube, it's not that big of a deal. And so the loose pitch as I was interpreting it was that Netflix and Spotify would kind of run out of talent to poach from YouTube because the next generation was so locked in.

It could be, but it depends on the structure of the deal. Right? So when Ms. Rachel went to Netflix, she wasn't forced to shut down her YouTube channel. She could continue to monetize that content. And actually she didn't even make any new content for Netflix. That's why it was such a win for Netflix. I mean, we don't know what, we don't know what they paid her, but she just took this content that was extant and brought it over to Netflix. They packaged it up into a short season and essentially just got paid for double dipping into an existing content.

Yeah, that's pretty remarkable.

Last question and then we'd love to have you back on. This was an awesome conversation. How. How excited do you think the average app developer, business owner in general should be about OpenAI launching ads? Because I know a number of businesses over the years that like we're basically be born because a new ads platform like I have friends company with hundreds of millions of dollars of revenue. They've told me like we would not be a big company if we didn't, if we didn't get to hyperscale on, on, on Facebook in the early days. And So I assume OpenAI will want to make the ads product cheap and performant early on to just, you know, drive this, you know, huge influx of volume. But I'm curious what you think the opportunity will be early.

Everyone should be excited about Open Air launching that platform. This is beneficial broadly for the economy. Ads are the driver of the Internet economy. Look, but for Facebook ads, but for conversion optimized Facebook ads, the DTC category won't exist. E commerce much smaller than it was. There'll be far fewer small businesses in this country. Ads is a growth engine not just for you know, e comm or not just for like tech bros. For everything. It is the beating heart of the economy. Everyone should be excited about this. Look, I remember there was Chamath was complaining at one point that like I don't remember the exact number, but he was like, you know, I invest in these companies and 70% of the money goes to Facebook ads. And I wrote a piece saying you should be thanking Facebook because if Facebook wasn't there to absorb this ad spend, those companies wouldn't exist for you to invest in. Right. This is great. This is.

Yeah.

Also the companies wouldn't spend the money if it wasn't driving results. It's not like you just, you don't, you don't. If a camp, if you, you spend 20 grand on a campaign and CAC is too just high, it doesn't make sense. You just turn it off. That's the beautiful thing.

Exactly. I think there's a reason to be very optimistic broadly about the new opportunity that will be engendered by just, by LLM Empowered Content Engagement or AI Empowered Content Content Engagement. It doesn't be just lms but, but I think like where people get mixed up is they think, well you know, these ads need to be different. There's some different way to sort of integrate ads. Well, no, the formats need to be unique and they need to feel native. But ads are ads, right? If you drive performance and you get conversion feedback and you get the feedback loop going and you deliver more value than you were paid, right. There's compounding there and it grows the economy. So I'm super excited. I call this commerce at the limit because I think what all. And a lot of this technology is not being applied to the consumer facing side of thing, it's getting applied to the back end. Right. That's what annoyed me about Facebook's last earnings or Meta's last earnings because people like, oh, look at all these big projects. Gem Andromeda Lattice drove 3% improvement to click through rate or 2%. You know how massive that is at that scale, how meaningful. But the other thing people ignore is that compounds every time I spend a dollar and I get more back, what do I invest the next time? Not a dollar. Right. I invest what I got back so it's more and it compounds over time. I'm going to make more money and it's going to grow over time. So that 2 or 3% this quarter will also exist the next time that money's recycled. Right. And also those products are not static, they're not, they're not frozen, you know, in time they will also continue to improve. So the thing is like these, these products and this is all in the back end. This is not seen by consumers but this is where a lot of the investment is going. And so people talk about the capex like the capex is delivering real returns. What are you complaining complaining about? And the thing is like that is that's driving real returns, but it's driving more spend, it's driving compounding spend and that's where a lot of the investments going. So I think there's, there's reasons to be excited across a number of service areas here. And it's not just the consumer facing side.

This is amazing. Thank you so much. I feel like I just got.

Let's get you back on the schedule next week.

We have to have you back. This is incredible. Thank you so much for stopping by, taking the time out of your day. I learned a lot, enjoyed it mostly I just got fired up. This is incredible. Thank you, Eric.

Thank you.

Take care, guys. And congrats on all the success. It's well deserved. The New York Times profile, Mike Isaac, that's the big leagues.

We survived.

We survived.

We survived. Thank you.

Great to see you.

Talk to you soon.

Cheers.

Have a good one.

That was one more last guest before.

We get a general with London intuition. We got to hop on with. We got to hop on with London in just a few minutes. But Pim is in the Restream waiting room and now he's in the DBP and ultra dump. Welcome to the show. Sorry for keeping you waiting. Give us the news.

Yeah.

We've raised $133 million. Seven.

Because we're gamers.

Seven.

Congrats.

Yeah. To build general agents for environments that.

Require deep space reasoning. 13:37 Lead, this is Gamer Reference. Tell me about your. Didn't you make money as a gamer at like 18 or something like that?

So I, so I grew up with Tourette's so I didn't really do much other than playing video games as a kid and built the largest private server on Runescape when, when I was a teenager. Did about a million and a half in revenue by the time I turned 18 years old. Yeah, it was pretty funny.

That's incredible. I mean people come by with $100 million fundraisers all the time, but that is extremely.

We don't see $133.7 million that is special.

So thank you so much for taking the time. Let's actually dive into the company. Tell me more about what you're building.

Yeah, so we're building general agents for environments that require deep spatial and temporal reasoning. So what this means is look at drones, for instance, robotic arms. They all ship with game controllers already. So there's this general interface for applications that's already in all the robotics where we don't need to reinvent the wheel, you know. And so the bet that we're taking is that we can train these, these foundation models on so much diversity represented in all the gaming data, which increasingly looks more realistic.

Right.

As physics engines get better and that we can transfer to novel environments with very, very minimal new data. And we're seeing, we know we're seeing the scaling for this happening now, it's very, very clear we can do this one application that we're going after. So after doing RuneScape, I worked at Dr. South Borders. I spent three years there. And so one of the applications that we're excited about is search and rescue drones, which is basically decoupling the need for humans to even be observing. And it can basically cover a lot more space a lot more quickly if you have general agents that can navigate these types of movements environments.

So the future of search and rescue is in your vision. Like drone swarms that get, let's say somebody's on a hike, they get lost, you'd be able to launch a swarm and cover 100 square miles in an hour.

Yeah.

I suspect there will still be like VLMs on the other side initially that sort of analyze footage and sort of process information. But yeah, I think the bottleneck is just make these things not stupidly run into things, to be honest. So that's what we're focused on right now is just general navigation for novel environments in devices that have gaming inputs represented specifically. And that's the key. Right, because you have these inputs that already gamers use to control in all these very diverse environments inside video games, including things like drones. Right. The agent only has to adapt to an environment, not a new action space. And so the bet is that transfers.

What are some other specific categories that are exciting to you or types of companies?

Yeah. So look at it this way. The model sits in a general action space, meaning that it has an understanding of how different actions relate to the world that it's acting in. You have to make sure that when you transfer it over to the physical world that you've solved for safety. What we generally do is we deploy agents, for instance, in video games that have a larger action space that can do loads of things because you don't need to solve for safety as much there. And then when you can verifiably prove that you deeply understand a specific action space, such as navigation, then you start transferring those actions over to the physical world. And we have sort of an internal joke that we think maybe simulation might actually be larger than the physical world because there's only one physical reality and there's many simulated ones.

I like that.

We're already working with game developers to get these models of played into their games, which means much more fun playing video games. And then it's also a great flywheel. Right. Because if we can verifiably prove that these models are performing great against game players, Then that's a great benchmark to also start transferring out to other environments.

Walk through the usefulness of something like Unreal Engine versus I'm going to mispronounce it, Gaussian splatting. And then a Genie 3 style generative world model where you're generating the full flame the, the frames on the fly. We've heard about a lot of companies using all three of those, mixing them together, using, focusing on one or the other. Do you have, are you particularly bullish or bearish on, on individual. One of those technologies?

I, I thought about this question a lot. I, I went through the depths of trying to actually start building my own physics engines, like actually understand this from first principles. So I'm going to try, I might take a little bit.

Let's give it up for first principles.

Yes. Love it.

So.

The most computationally difficult things to simulate are high degree of freedom agents. And that grows exponentially as you have more agents in an environment. Meaning that at some point in simulation you hit a point where you just have to bet on video transfer. Right. And also then at the same time, for instance, there's loads of things that we currently do not have video footage of, which means you cannot really bet on those things. And so like cells and smaller things that aren't well represented, those type of things. Right. So you always have to use a combination of the two. You're always going to have to use a combination of two. The other thing is that when you sit inside generative role models like genie, for instance, you're not fully in verifiable domain. Right. And so the problem with, to be clear, this is an incredible breakthrough.

Right?

But there's very much use for engines like Unreal. And I think, I don't think actually, I think where we end up is it's actually really annoying not to have that determinism in these world models.

Right.

Because you want to be in some form of verifiable domain. So I hope that somebody manages to create some form of hybrid architecture. Like okay, so for instance.

You can hallucinate all the text, but then you can RL it and then you have, oh, write some Python. So if you need to do a complex math equation, you draw Python, right?

Yeah. Have you guys ever watched Redstone CPU videos on Minecraft? So there are videos on Minecraft where people actually build Redstone CPUs and there's also now they built the ChatGPT. So my point is we might actually end up finding really interesting. My dream is to be able to simulate a CPU inside a world model where you have some form of determinism. And again, this is not at all possible or near possible today. But my point is being like maybe a world model can. Yeah.

That'S awesome.

Yeah, I think you need both right now. You want to bet maximally on world models and video transfer for things that are hard to simulate and then maximally on simulation for the things that you want to stay in verifiable domain on, which is most things in my opinion.

That makes a ton of sense. Thank you so much for coming on the show. Congratulations. Just very fun conversation.

Super fun conversation. We'd love to have you back on. Anything happens in the news that you're excited about, just ping us. We'll have you on.

Yeah, I'm sure.

Sounds great.

We'll talk to you soon. Have a good rest of your day.

Cheers.

We'll talk to you later. Before we hop off, let me tell you about wander.

Find your happy place. Find your happy place. Book of wander to leave you hanging there for a second.

Oh, you're good. No, I'm fine. I didn't even notice. Book a wander with inspiring views. Hotel graded many's dreamy beds, top tier cleaning, 247 concierge service. It's a vacation home but better folks. And good luck to Raghav in the chat. He is going to a job interview and we wish him the best. Leave us 5 stars on Apple podcasts and Spotify and we will see you tomorrow. Thank you for tuning in.

Can't wait.

Goodbye.

Have a great evening.