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

Episode Date: October 16, 2025

(01:17) - Google’s AI Breakthrough in Cancer (22:44) - 𝕏 Timeline Reactions (30:37) - Protein Powders Exposed (35:41) - 𝕏 Timeline Reactions (01:36:03) - Dante Vaisbort, a University of Chicago graduate and co-founder of Albacore Inc., discusses his company's development of Ghostfin, a 6-foot-long autonomous underwater vehicle capable of patrolling over 1,000 nautical miles undetected and delivering precision effects. He highlights the vehicle's potential to deter maritime invasions, its ability to carry a 250-pound explosive payload, and its distinction from existing platforms like Anduril's DIVE, which focuses on deep-sea survey missions. Vaisbort also emphasizes Albacore's strategic location in Philadelphia, leveraging proximity to Washington, D.C., and the city's rich shipbuilding history to foster the next generation of maritime capabilities. (01:58:30) - Eiso Kant is the co-founder and CTO of Poolside, a company dedicated to developing advanced AI models for software engineering. In the conversation, he discusses his journey from being inspired by Andrej Karpathy's 2015 article on neural networks to founding Poolside, emphasizing the importance of reinforcement learning in advancing AI capabilities. He also highlights Poolside's strategic focus on enterprise applications, particularly in high-consequence environments like defense and government, and the company's efforts to secure substantial computing resources to stay at the forefront of AI development. (02:12:59) - Marc Benioff, co-founder and CEO of Salesforce, is a prominent figure in the tech industry, known for his leadership in cloud computing and AI integration. In the conversation, he discusses the rapid adoption of AI technologies, emphasizing Salesforce's commitment to enhancing customer success through innovations like Agentforce. Benioff also highlights the importance of maintaining a beginner's mindset and the role of AI in transforming business operations. (02:37:03) - Alice Bentinck is the co-founder and CEO of Entrepreneur First, a global talent investor that supports individuals in building technology startups, and also co-founded Code First: Girls, a nonprofit aimed at increasing the number of women in tech. She discusses the recent Entrepreneur First demo day, highlighting the participation of 40 founders who pitched AI-related products to an audience of 200 attendees, including partners from leading venture capital firms. Bentinck emphasizes the trend of young founders tackling complex problems in traditional industries and underscores the importance of revenue stickiness and long-term contracts in attracting investor interest. (02:43:17) - Eric Seufert, an independent analyst and founder of Mobile Dev Memo, discusses his belief that OpenAI will inevitably monetize ChatGPT through advertising, drawing parallels to Netflix's prior stance on ads. He emphasizes the importance of maintaining user trust by ensuring ads do not compromise the objectivity of ChatGPT's responses. Seufert also highlights the potential for ads to drive economic growth, citing their role in the success of direct-to-consumer businesses and the broader internet economy. (03:06:57) - Pim de Witte, a tech entrepreneur who founded Medal, a major social gaming platform, and previously ran a successful Runescape private server as a teenager, discusses his company's recent $133.7 million fundraising to develop general agents capable of deep spatial-temporal reasoning for applications like search and rescue drones. He explains their approach of training foundation models on diverse gaming data to enable efficient transfer to novel environments, leveraging existing gaming interfaces in robotics. Additionally, Pim highlights the importance of combining simulation and real-world data to ensure safety and effectiveness in deploying these agents across various applications. TBPN.com is made possible by: Ramp - https://ramp.comFigma - https://figma.comVanta - https://vanta.comLinear - https://linear.appEight Sleep - https://eightsleep.com/tbpnWander - https://wander.com/tbpnPublic - https://public.comAdQuick - https://adquick.comBezel - https://getbezel.com Numeral - https://www.numeralhq.comPolymarket - https://polymarket.comAttio - https://attio.com/tbpnFin - https://fin.ai/tbpnGraphite - https://graphite.devRestream - https://restream.ioProfound - https://tryprofound.comJulius AI - https://julius.aiturbopuffer - https://turbopuffer.comfal - https://fal.aiPrivy - https://www.privy.ioCognition - https://cognition.aiGemini - https://gemini.google.comFollow TBPN: https://TBPN.comhttps://x.com/tbpnhttps://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235https://www.youtube.com/@TBPNLive

You're watching TBPN. Today is 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 turn of events in the meantime. Well, Open AI 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. Which Way Western Lab? 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 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 paints, accounting, and a whole lot more all in one place. Go to ram.com. Just kidding. He said, an exciting milestone for AI and science, our C2S-scale 27B Foundation,

model, built with Yale, and based on gamma, 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. 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've got to

hunt them down. You've got to 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 head shots, double kills, 360 telescopes. 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 mini map. How do you do that? You've 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 immune signal, and it basically turns these cold tumors hot. And now, this is not new. There's a lot of 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 setting, but it's still very exciting. Now, just to put it 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 U.S. alone, and only about 5% of drugs that even get submitted to the FDA actually get approved. And so the total number of 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 them have, you know, less severe side effects, and so they become more popular. Some of them are just good at advertising probably, or good at, you know, 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. Something that's 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. If you, 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, uh, uh, 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 useful here. Uh, my question is, does, does open. AI 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 Open AI's like even roadmap, right? So this is that, 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 Sunder had this in his back pocket and was waiting to drop it. I'm not implying that either, but I'm sure he knew exactly what he was, he and the team knew what they were doing when they hit. 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? I'll listen. I mean, I think opening it definitely has the resources to do this kind of thing. you look at the actual model gym it's they used the 27 billion parameter model which is like i mean that's tiny compared to like what frontier models are now right like jbt4 was over a trillion it was done pre-training in 2022 yeah um like they could throw a couple people at this they they just started that new like physics team that's just doing like physics models yeah

opening i for science kevin real yeah so i i think this is like totally reasonable and it's in their capabilities yes 100 percent um i think it just uh like clearly it's a race even if this is just like positive PR, like, hey, AI is good for the world. Like, every foundation lab should be competing in that because that's the thing that when you get on Capitol Hill and somebody 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 it. Like, it's just good calms. And so you want to be putting the points on the board first. Yeah, and this is why, again, Open AI 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 Roon'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, 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 like 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 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 a vindication. I don't like this, like, was that big of it 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 biological systems based on just a pure textual representation of the cell and the interaction.

So I think that's an interesting vindication for like text is a 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. Yeah, I mean, so like if you look at Alpha Fold, 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 like areas. You apply them to bio and it works. Like, yeah, because it's not like those architectures are like fairly general.

It's not like just for gaming or just for like text. Yeah, yeah, yeah. I want to talk about scale, but first I want to talk about scaling your streams with one live stream. Destinations, Restream, multi-stream and reach your audience web there. So, yeah, the second takeaway for me is that, like, scaling laws apply to useful AI applications in bio, because that's, 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 a smaller model. And so there's an interesting, like, they're scaling up, and it's, you know, qualitatively better system and so yeah I don't know yeah I mean when I see stuff like this I kind of am wondering what people who are like oh this is the 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 yep in five 10 years this will be yeah 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, you know, has a novel insight, runs away with it, you know, fully robotic, lights out, lab, and then it just spits out a drug at the other end. So I think like I think it's absolutely incredible,

but and an amazing proof point for the potential for the technology. But 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 everyone, 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, like, 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 two gigawatt data center being planned. Just yesterday. He's 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. Yeah, I think the bear case is for like a fast takeoff scenario, I suppose, is something like, is this the chat GPT 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 test it in the chat says got to give Martin Schrelli 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 on. Anyways, we can. Is he live? He's probably live, but I'm sure he's down to co-stream. Uh, going on. Uh, there's a post here from G. Fodor, uh, who says utterly, totally destroyed worldview good riddance. He's, uh, quoting, uh, uh, a post, uh, featuring a line from Jan Lacoon, 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 a big moment. And it's a win for the world. Yeah. My question is like, okay, can Google whatever the process that led to this discovery can Google

produce thousands of these, right? It does seem like they're accelerating, yeah. Yeah, like I just imagine at some point, like a year from now is Sundar still sharing these breakthroughs or are they just kind of going out? Taking them through FDA approvals? I don't know. What do you think, Tyler? I mean, I assume

everything that we've seen in AI says that, like, yeah, they can keep scaling. It's only 27 billion parameters, scale it up to a trillion. Yeah. 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,

then, like, what does it really matter? Yeah, it's just a super drug screener that allows you, like, a first pass of ideas. Like, 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 like that was not the gating factor

here. I do wonder if they're somewhat data limited on like actually indexing the 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

mercores, like those companies have been able scale extremely quickly to do knowledge retrieval tasks and put tons and tons of new you know, answers to questions in the training set. Like, what is the Mercor-scale AI of AI bio? Yeah, like, I don't know if there is one yet, which just tells us that, like, this is still pretty early innings. Yeah, it seems early. Yeah, it's interesting to see these people that were hyper-bearish transformer models, like this, this Brett Hall post, he says, once you understand

the transformer, as Lanier does, and dare say 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 LLMs. 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, it could be very cool. And I would give you the example of like this, we just got these rugby shirts 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 like a 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 like 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 then sync 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 force and creativity.

it. 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 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 bare case to like 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 like, you know, fundamental law of physics around how fast you can move sand around the world and turn into silicon or how fast you can spin up a new power plant or new nuclear power plant. Like these, some of these things just take time and regulation, like 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 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 Pyridic 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. Yeah, no. What is it?

No, I'll keep saying about, like, natural gas. Is it like half a percent? Isn't it like 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, no, no. 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, uh, sell 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, uh, Gemini 3 that's launching. Like, it feels like, do you see the cutoff date for Gemini 3 is like, is like, uh, October of 25? And so it feels like that was a much bigger run, probably maybe a billion dollars 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 can underwrite it very clearly on an ROI basis.

This is much harder to underwrite right now, but it's going to get less and less fuzzy 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, it 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-class. Is there new information? Yeah. Was that not knowledge that existed that was discovered? Yeah. And now they're organizing it. Yeah. Their mission is to organize the world's information and make it useful.

Oh, make it useful. This is extremely on brand for them then. It's being made useful. Well, let me tell you about Privy wallet infrastructure for every bank. Privy helps you, it makes it easy to build on crypto rail, 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. Comrat. Comrade? Yeah. Bodging the pronunciation. Comrade. He says, someone needs to turn each TBPN card into a physical collectible, slowly getting activation energy for this to start taking pre-orders, limited to 10 prints, first edition reserve 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. We were texting about this over the weekend, and I wasn't sure about, like, how, what the right way to handle, like, card printing. They're very, like, 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'd say it's fun to print them all, honestly. Yeah. But I'm sure Greg will figure out, like, where the demand is. Like, 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 sign cards?

Yeah. Do they want, like, a piece of merch, like, you know, on like some, some uh trade you know actual trading cards they'll put like a piece of the jersey fabric in them oh really i've never heard i've never seen that 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 that'd be kind of like what what are you doing depends on wedding anyway let me tell you about cognition they're the makers of devon 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 TBPN 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-sized photos of things. And they're, like, 30 cents, I think, for, like, a set of four. Then 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 solve the problem. But I'm excited to see where it goes. Well, Monitis says, interesting cultural indicator that Volvo quietly rolled out armored versions of their family cars direct to consumer. over the summer, probably nothing. Direct a 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 former president of Kazakhstan. No way. Actual, like, presidential, like, car that he'd be driven around that my buddy Blake had imported from Kazakhstan. And 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 gap. Yeah, exactly. Yeah, it must be so insane. You're just filling up like, oh, I'm driving from Malibu to L.A. Got to stop for gas three times. Full tank. Yeah. Rough. Um, uh, let me tell you about Figma. Think bigger, build faster. Figma helps design development teams build great products together. Uh, 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 polymarkets that were 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, 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 I get how did our family get so rich memes going viral around the Founders Fund names Dad read the Simerillion and

bought a lot of domain names Sebastian Calieri says move quickly there aren't that many good Lord of the Rings names left to use. Arthur McWhorter says he's camping the following. Mordor, Lothorian, Merckwood, Numanor. I don't think. Valanor. Vallinor's been used. Valinor is an incubator, I believe. Ammon? I didn't know Hamon. Amon? Is that? No, there's no way. Did Peter Thiel back Amon Geary?

Joe Lonsdale quoted the announcement yesterday that we shared around Airbor. 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. A new with today's innovation frontier can upgrade how these areas work for all. I'm just the founding board member and early investor in Arab War with 8VC, proud to support Palmer and his co-founders. Yeah. I always liked Shadowfax. If you were going to start a Founder's Fund backed HelloFax competitor or DocuSign competitor. using Shadow Fax?

Wait, have you not seen Lord of the Rings? I have seen Lord of the Rings. Do you know who Shadow Fax is? No. He's Gandalf's horse. Oh, okay. Yeah, that's coming back. And I think Shadow Fax is a good name for, like, a docu-sat competitor. 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 Aribor. 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 four prices across a range of industries to combat market manipulation by China, according to Scott Besson. It's very odd that they're pulling that out as a tool instead of just focusing on tariffs. I don't know. 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 like sometimes a DJ's up there and like really turn in all the different knobs sometimes you just let the music play you know that is a very funny metaphor for David Sullivan like the knobs are there right you can yeah some people just choose to use them more than that you should 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 uh 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. 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. It was like, What was a 5% or 10% stake that the U.S. government took? And a lot of people were like, this is like communism.

This is command and control. But if you look at the, I believe TSM 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, like, doing worse than what's going on in South Korea or... I think, yeah, my framework is that as much as Trump and G beef, I believe that Trump respects G.

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 turn the company around, and then sold their position and gotten out entirely. Like, government motors. The U.S. does not fully – the U.S. government never, like, fully nationalized the GM ecosystem, even though they did bail them out at one point. Joe Wisenthal 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 wind up being,

but we've got to have him on the show. What are you thinking? I'm just laughing at the chat. Sam Schaeffer is asking about real competitors to the USA at the country level. And the chat says China, Russia, Uzbekistan, Guadalamo, Martinique, Mauritania. Yeah. Well, let me tell you about Vantza. Automate compliance, manage, risk, approved trust,

continuously. Vantage's trust management platform, takes the manual work out of your security compliance process, and replaces with continuous automation, whether you're pursuing in which framework or managing it. Did you see this consumer reports report? Paris did an investigation, 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 food that would appear to be like quote natural right so like dark chocolate is a good example of this 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 yeah and I think that people were particularly scared about Hewle, which the Hewle 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 Heel Black over the last year. You're just deranged from all the lead poisoning.

Yeah. We'll stick to the optimum nutrition, the transparent labs. Good to see transparent labs being exactly what they're name. 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 last year, Nat Friedman with the plastic list, like really shifted the conversation to levels of microplastics. And I think people kind of like stop paying attention to lead. But it's all important. You want to be aware of it. But like 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 skits I will be like, actually, actually, good for you. Because it makes more aggressive, right? Yeah, yeah. Doesn't it make you a little crazy? Yeah. If you want to have a strong Q4, you know, up the lead and then detox and see 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 illegal levels which like none of these, none of these brands were above the legal limit.

Oh really? None of them. Even even the fuel was just was just not recommended. So if you're a food company in America, you can legally 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. Oh, you ever take? Yeah, I mean, I think with plastics and lead, like the thing is, we've kind of ran the AB 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. Letted 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 you like, it makes you way crazier. But with plastic,

we haven't really run the AB test because like there's kind of just always been like a build of a plastic. Sure. There's like no one who like doesn't have a lot of plastic. So who knows what happens when we take plastic out of the system? Some potentially schizo alpha is that lead paint could actually be positive because it reduces the like EMF in a room. If you have like 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 poisoning because you have so much plastic in there. You wrapped your entire body in plastic. This went incredibly viral. So 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. I have every single day, basically. Yeah, 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 that if they ran one lab test per product, if you tested another product that has, that was made in an enough,

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 Soreland. We had some article that was like, it was very funnily worded. It was like, so super heavy metal or something like that because they were tested it 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 to. like change their supplier or something like that, the lead company. Bobby wants to see a lead gong in the Ultradome. For agro maxing, yeah. Let the lead fly. 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 kind of like it was written by chat to BT.

Well, let's pull this video up, but first let me tell you about graphite.com, code review for the age of AI. Graphite helps teams are good. 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 is really cool. 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? Fighting a robot. There we go.

It's discovered in obsession. Born from truth so raw, it feels like mad. That was a weird shot. This is cool, though. A beautiful disregard for your own limits. A truth your body understands before your mind. This really sounds like Morton Freeman.

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

Yeah. 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, giblied it and just posted that and it got thousands of likes because people were like

oh this is like something I like and I know and I like Oppenheimer and I like Ghibli so like but if I posted that today it would be a total flop because like it's been done so many times right I just think it's funny because we should actually test it you should post like just a Ghibli image and be like Ghibli of that horse and be like oh wow you know

or whatever whatever memes going viral that day you could Ghibli it and I mean there were so many Ghibli that went viral where it was just like every iconic image for the past few years, it was like a picture of J.D. Vance, Ghibliad. Okay, viral. A picture of Donald Trump, viral. Like, picture of Michael Jordan, Dunking, viral. And, but now it's like, we have all seen it, and it's been commoditized, so it doesn't, it doesn't, it, like, it ran its course. And that's, like, the, the worry with this is that, like, the first AI launch video goes viral and then it, like, 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, you know, the agents kind of plan, generate, edit every scene. So you're kind of like, you can still come with an idea. Like the thing that stuck out to me in that, 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, like that was an interesting artistic style.

Yeah. 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 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 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 part of called Zavo.

Okay, let's play that. Let's pull it up. I'm very interested in that. It's in the chat. It's in the X chat. We can try and put it in the timeline. I think you're going to absolutely. 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 are falling. Okay, meet Greg, your numbers wizard.

He'll tell you the Seabass special is secretly costing you a fortune. But the Truffle Frye special... Is that motion graphic AI as well? That's pretty impressive, if so... He knew you'd run out of Pino Noir by 8 p.m. on Friday, and already reordered it yesterday. He knows when your night will go from zero to 60. So is this vertical sass for restaurants? 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 like post, that looks like after effects. Wow, I hate this. See, this looks like not AI generated. This looks like they layered in their actual product, but maybe that's part of the workflow in 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 restaurant tour I hope I'm getting this right that's it's a it's an ERP for for restaurants yeah so you use that to run your restaurant that that's what I got from that is that is that correct the first agentic point of sale for restaurants and retail payments okay to sale on AI agents in one platform to build the future of autonomous commerce over 400 businesses already used. 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 scene was particularly bad, but at the same time, it was probably a good hook.

Exactly. But I think the hokeyness of that will dissipate. I think you're correct on that. Like, it will be, like, eventually you'll just scroll and be, oh, yeah, I've already seen, like, yeah, I've already seen, like, yeah, I don't want to see that. But, like, right now, if you're the first, like, Toast doesn't running an ad like that. And so if you're the first comfort. to be running that ad that grabs people in that way,

like you will break through. There's like a window of alpha, I think. Yeah, the question is, 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 founder would be. That's for sure coming. I mean, V-O3 launched yesterday and 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 find two models with serverless GPUs and on-demand clusters. But I agree with, I agree with, like, part of it as an advertising enjoyer. I like to, like, without having knowledge,

try to analyze it when I see a video, when I see a campaign, I'm trying to clock, you know, based on the stage, of the company, how much did they spend on this? For example, like if Airbnb comes out with a launch video, I feel pretty, I like expect it to be incredible because it's Airbnb and they probably spent like half a million dollars like producing this ad.

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, sure, sure. I like being able to 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 had maybe maybe there were white humanoid looking things like robots they were completely missed that so yeah it was very like it was very visually intense um i yeah i agree with that that's interesting um i think that there is a there uh there is a like a 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 is a lot with like Justine Moore. Like a new model will drop and she'll be, 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 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 mid-journey, but you can start with a mid-jorney 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? Are they spending energy, resources, time? Oh, I think it's definitely not just like one shot at this point. Right, right. But that's fun. So today it's basically like a managed, productized service business.

And I've heard of another company, an AI ad company, who is effectively, I'm not going to name them, They're just doing, like, creative services work, and they're selling it. They're selling it like its AI outputs, but it ultimately is just them running an agency. And are they prompting a lot, or are they actually filming stuff, too? They're doing everything. They're doing everything. Basically, it's an agency masquerading as, like, as an AI company. Yeah.

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 you know how we do things in in different you know 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. Yeah. 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 get you get clients that are looking for cheap work and those clients tend to be really rough. Like great companies. understand the value of creative, we'll 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 purely corporate, where if a new law firm emerged and they said, we want to have your legal bill drop by 70% every month. And we're going to 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 vector in full-text search.

Built from first principles on object storage, fast times cheaper and extremity scale. 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. set this, 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 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 D to C e-commerce era? We're basically who had brands who said, our secret, how we're going up against Nestle, how we're going

up against Coca-Cola is we have Shopify and like SaaS. We have SaaS and they don't. It turns out Mark Zuckerberg says, actually I'm planning to be the middleman here. I just I don't want to. That's a great take. Yeah. I don't want to. The money that you were going to spend on rent, well, I'm actually going to need to spend twice as much with me. So do you think that, do you think that's how it plays out in the next generation of creative agencies? I start a creative agency and I 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. Stuff, right? I don't know. I think you always, 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. It might take them weeks. It might take them, they might one shot, 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 on, I wouldn't respond well to a pitch today

from a agency that says, hey, you should work with us. We're going to charge you 60% less than your current designer by using AI. Like that doesn't, 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 rapper around it is going to be a little bit tricky. Yeah, and the question, so going back to, absurd, this YC company, there was a time when, like, the D to C 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 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 stopped 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. Yeah. It's a very narrow. It's a very narrow arbitrage. But the timeline is in turmoil on threads because paper. and WISE are taking shots at each other. Did you see this? PayPal said, normalize sending money instead of memes and got

9,000 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 one like... It was Coke and Pepsi

saying like, wag me. Yeah, the Metaverse. 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 Nano Banana.

You can talk to Gemini live. And continue. Speaking of ads. Let's go. Mika Amar made potential. 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 shocked full of ads. I wonder what website these are. Display ads. Also, it looks like in another era. In German, maybe. maybe or something? I don't know. Did you, did you hear that yesterday, Paxos mistakenly minted 300 trillion of their stable coins? I am not, I did not follow that at all. Exactly what happened? So, stable coin issuers like Paxos and Circle and Tether, they mint new stable coins.

Okay. They had an internal error apparently that, and they, they, it sounds like it was effective. that finger, they accidentally minted $300 trillion in their P-Y-U-S-D, their stable coin. Did this, like, crash their market or something? I wonder what happened here. No, I think they were, they were, uh, Paxos immediately identified the error and burned the excess. Uh, it's resolved. Yeah, they said that there's, uh, no security breach. Customer funds are safe. They've addressed the root cause. But how fat of finger do you have to add, like, I imagine like four extra zeros? Maybe, maybe the dev like fell asleep. It does

feeling a cat on the keyboard moment? Yeah, it doesn't, it doesn't exactly. Yeah, Joey's a 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. Yep. What was it backed by?

Or is it actually, is there some sort of disconnect? They bought 300 trillion of U.S.D. Glad they figured it out. Maybe they bought 300 trillion of treasuries, you know. Well, before we move on, let me tell you about profound. Get your brand mentioned in chat, GPT. Reach millions of consumers who are using AI to discover new products and brands. Born in Texas, in a Texas town of less than 1,000,

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 life story. 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. 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 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 tech.

Let's get 100 GACs in Open AI or something. That's the move. Breaking news. Hit, get that gong ready. Julius AI is now GDPR compliant. 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 somewhere. I've got to go deeper. Dean Ball says, could I? make a tax deductible donation to the Open AI non-profit? You've been joking about this. I think you can. Alexander Berger says,

True story. Open AI once came up on the checkout screen of my Safeway as the local nonprofit to donate to. I think that's got to be a joke, right? Yeah, that's got to be a joke. But 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 Open AI nonprofit, because they'll probably spin out another for-profit. That's my hottest take is that we're getting a... Well, being a donor in the nonprofit certainly does not guarantee you equity in spin-outs. 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 nonprofit, which is good. What about if you're just a donor? Yeah, you got to really... You got to post a screenshot of an email. A 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 Sheel, including Steve Jobs for California. They're doing Dr. Norman Borlau. The Kray won supercomputer for Wisconsin. Do you know who these guys are?

Steve Jobs for California. And Minnesota, with mobile refrigeration. That's a huge brain. Let's give it up for mobile refrigeration. I know the Cray's Supercomputer, and obviously Steve Jobs. 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 going to make?

Because I feel like these things could instantly trade at the moon, like a hundred times. Yeah. Yeah, I'll look that up. And then also Norman Borla was an agronomist, agriculture guy. Okay. And he kind of did a lot of stuff that influenced the Green Revolution. He, he, or a farm farming. Yeah.

who's a literal aura farmer. I love it. Rune was clapping back at George Hatz. We read George Hatz's pathetic losers blog yesterday, which was very blackpilling. 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 Reed says, wow, his website is very negative. And Rune says, feels like George Hatz is a zealot who has produced far less value than his clearly exceptional skills as an engineer should suggest. But George Hots can do whatever he wants. That's the beauty of the free market.

He can go and be a thinker. He can advocate for different things. He can open source software. Tyler, how have you been? reacting to the George Hots news. Do you think that you're, 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-bilt tool for planning and building products. Meet the system for modern software development. Streamline issues, projects, and product roadmaps. People were having a lot of conversations surrounding Open AIs numbers from the financial times.

Of course, they have 800 million weekly active users. Five percent of those are paying 40 million. 13 billion in AARR, which implies a $325 annual ARPU or $2. $87 a month per paying user. And this post 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 chat GPT is like they don't like you can't actually advertise on it yeah like they don't have an ads product and so uh 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 the ads product for 20 years now yeah and yeah and i think what what malte was meaning to say is that he's it falls it up he says the tweet was meant to say you can run this into a profitable business sure i expect value per is going to be higher than Google. Oh, yeah, because Open AI 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, like similarly, you get to break even pretty quickly. Yeah, and Open AI is projecting to get to $100 billion of revenue. The projections are crazy. We should go through those.

Doomslide says, extremely polite thread by E.Poc. Open AI's projection implies it gobbling up about half of all software revenue by 2028. We are going to be joined by Mark Benioff in just a few minutes, who, of course, captures a lot of that software revenue already. But gobbles up, is gobbles up the right way to do it?

Or is it creating new software spend, right? If OpenAI can build consumer, or agents, if they can monetize product discovery, right, 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 that, you know, basically money that wasn't being spent on software that will now go to software. They're going to be creating a 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 $100 billion in a few years is ludicrous, but I don't know that it'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 Open AI 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 opening and I 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 like instantiated on the fly when you go into the ChachyPT app where I was talking to a barber who was saying that ChachyPT 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 chat GPT just, like, does that for you on the fly. And it's something that Google used to, like, route you to that one-off piece of software. You can do a million of these, like, you know, calculate interest rates, compounding interest. You know, you wind up on those, like, home mortgage interest calculator websites sometimes. Calculate your BMI if you're into bodybuilding, like diet plan. Like, there's so many different pieces of software out there that are just kind of, like, getting rolled up.

And we haven't really seen, like, the SaaSpocalypse come to those, like, small, long-tail pieces of software. But that could be... No, I think Google has also made a pretty big push into that. Yeah, obviously, the calculator. Like, you're not searching calculator on Google, and then going to a website that's showing you display ads. It's so sad if you were like, I had an amazing lifestyle business. Just... Online calculator.

Calculator.com. It was a 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. Yeah, it would just generate. But that wasn't generated on the fly. That was something that they hand-coded and then would, like, route you to if you searched certain terms. But opening I 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 ChachyPT to pretend you are a casino game of roulette. I put 100 on black, spin that wheel. Tell me if I won. And Chachapit 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. dollars, 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 its due. And so, I mean, obviously that's a joke because they're not getting into gambling. But, you know, like, Chad GPD 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, pretend to be a blackjack dealer, and let's actually play with real money. Real money, yeah. I don't think they're going to do that, but yeah. There's probably an interesting eval where it's 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 I could do that definitely. And also, even if you just continue the chat and stuff. 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, 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 do you step away enlightened or spin again for revenge what do you think I should do should I double down again put $200 I should go to chat to me I put $2 million down well whatever you're winning make sure to pay your sales tax with numeralhq.com sales tax on autopilot spend less than five minutes per month on sales tax compliance go to numeralhq.com Did you see that Kevin from the office was wearing a cubitist yesterday? I didn't clock that on the stream. We should have talked about that. What a dog, the denim, the denim strap cubitist. 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 a royal flush by the team behind this. So congratulations. Nick Carter says the government that I pay 40% of my income too has been shut for three weeks and nothing in my life has changed.

It's a 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? So they can continue to do that. 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 Chachapiti has stalled since May, DB data insights. 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 don't need reasoning models to tell you. St. Trope. Yeah, where to park your yacht. You can, you can, you can,

ask your butler. With the free tier. The free tier is more than enough when you're hanging out in Monaco. Yeah, that was my initial read on this, is that you can't read too much into it, but it certainly is not good. Yeah.

Right. You'll be really interesting. I mean, I would love to look at the ramp data on chat TPT 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 expensive. They're being expensed, basically. I think that for the general average person,

They are very much just using chat GPT 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 or they and I wonder if that adoption rate is kind of plateauing in the, you know, broader, you know, 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 SuiteX, 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 would be somewhat disruptive to people. But whatever they do, hopefully they use fin.a.I.

The number one AI agent for customer service, number one in performance benchmarks, number one in competitive bakeoffs, number one ranking on G2. What's that company that's competing with Starlink? ASTS. It has a retail 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 consolation up there. We did see Amazon yesterday talking about Predocupier, 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. Amazon is building a satellite. Yeah, Coupier. 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 this point, 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 inside information here. Yeah, I mean, I think going back to our conversation, I think it was earlier this week on posts are so easy. 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 some of my best videos have taken the most amount of effort. Like my top five videos took the most amount of effort, some creative spark, and then a ton of investment and everything from production to the editing process. So, again, I think YouTube creators certainly deserve to be paid well. I have never felt as a, you know, relatively small creator on X that I should be paid. There's also the factor of allocating payment accurately. That is very difficult.

It's so hard. Yeah. What is the value of, is a, is a Jeremy Gaffan 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 Jeremy Giffon YouTube video, we're. to go up and YouTube were to run a mid-roll video ad in that video, it's very clear 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 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 interesting to see where it goes.

Bucco Capital bloke says the AI CapEx trade can't keep going up 2% every day on partnerships, projections, and information articles. using 20-30 estimates, and, of course, the AI KAPX trade actually can't. Foxcon pops today, right, based on just the news that they had a conversation with Open AI? Yes, so the news, Foxcon jumped 8% based on this. Foxcon shares rise after chairman says he met with Open AI, plans Nvidia next. So it's not actually a full deal yet. Frenzied investing in AI has fueled talk of a bubble, but the chairman of Taiwan listed Foxcon, the world's largest contract electronics maker, says there's

much further to go. The application of AI is just the beginning. Foxcon'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 Foxcon had met with OpenAI chief executive Sam Altman at the Taiwanese company's headquarters to discuss potential future collaborations. And so the stock jumped eight percent. Let's give it up for discussing future collaborations. And I mean, AMD shares jumped 24% after the chip company announced a multi-billion dollar tie up with Open AI earlier this month, another multi-billion dollar. After they announced giving away. With Broadcom, similarly sent chip maker. So I don't know. 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 a 20-year period and Sam is just saying, I want multiple data center providers, multiple clouds, multiple chips, multiple, you know, fabs. Like, 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. Like, people are already going to, the average person is going to be rooting for it to fail. Yeah. And then actually delivering enough value to earn a place.

Are you talking about, like, consumer hardware? Yeah, the Johnny iPhone. Oh, because, yeah, that Foxcon would be a logical partner there. 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's 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 I message. 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 Apple Home pods.

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 I 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 Open AI, 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 get a decent speaker out of it, at least can. I mean, Open AI 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 TV. Sonos is a $2 billion market cap company now. Should Open AI

just buy it? Equus Global says Sonos plus Open AI. I've heard rumors. I don't know where you'd be hearing those rumors, but I don't know. I wouldn't be surprised. I wouldn't be surprised. It's like they have, they did almost 2 billion of revenue. Open AI is great at software. Sonos is great at hardware. 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, like, if you're, if you're entering these like hypercompetitive markets, do you want to go for the most competitive with the phone? Sam Altman certainly indicated that he might not pull. Hello. Hello in the chas is how much paper towels is Jordy? Yeah, Jordy's obsessed 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, you know, 1,000 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, Open AI 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 and future spending with a private valuation of half a trillion is mental. You would think, John, would think, John, would. 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, like, you know, look back at Uber and Lyft. The big critique of Uber was it's losing money. It's never going to make money.

Yep. So. Have you seen Uber's market cap recently? I think it's like $200 billion. TK. Yeah, 192 billion And a 15 PE ratio Like this business matured and is doing

Fantasticly It's pretty pretty remarkable Like they very much like fought the capital war and won It's like a testament to like They you know they they they they went out I mean Lyft is an 8 billion market cap There was a time when those were When the spread between Uber and Lyft was like you know

2x, 3x, 4x, and now it's, and now it's, what, 20x or something like that. Yeah. Yeah, so I think it's, I just, 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 Adio, custom relationship magic.

Adio 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-164.

Portland has a retro-gaming expo. That's extremely on-brand. That's extremely Portland. Got to give them credit for that. This seems, maybe we should dispatch, Tyler, to the Portland Retro-Gaming Expo to go try the first playable M-64. That would be cool.

I think we'll be able to get one of these pretty soon. I'm really excited. Tyler can just spend a whole show. Beat Golden Eye. Playing the Mod Retro. I wonder how fast you could speed run Golden Eye. Games back then were so short. They just didn't make 100-hour games.

So people would speed through them. This is, what else is here? How did 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... 7 hours and 40 minutes. Because it's a little one, so it's not... We got to hit the gong again for Rahul and Julius because you can now connect Julius to your data bricks. Let's go.

Hit that. This is a relevant post... Matt Slotnick posted on July 16th, 2025. bowl case for nothing ever happens is that open AI runs its entire business on sales force and slack and this was in a blog posted an open AI employee posted they said an unusual part of open AI 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 um we got to ask benny off uh does anything ever happen and which way which way does the deal go? Because Salesforce has a deal with Open AI. Everyone assumes it's to buy a lot of tokens from Open AI,

but maybe it's a trade deal. You get your Salesforce installation. The money, it's one of those circular deals. Salesforce for tokens. Zephyr is sharing Broadcom's fifth customer

is an 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 like them to look 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 uh 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 um some folks on the timeline were talking about how how Apple might make a big acquisition, but maybe Anthropics 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. They own 14%. That's a lot. It's pretty remarkable. How much does Amazon own? 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 have multi-asset investing, industry-leading yields, and they're trusted by millions. Amazon owns an estimated $5. 15 to 19% of Anthropic. And it's so funny to me based on the way on Anthropics' brand and approach and all these different things, the way that you have Open AIs Cap Table,

which is owned by a nonprofit, the employees own a ton, and then the VCs are like at the very bottom, right? Just carving out a little bit. Meanwhile, Anthropic is like Amazon, Google. Like these big, big, basically... Tyler, did you see... this Eliezer Utikowski post. He quote posted barely AI. The financial times projected. Open AI's

cap table following its for-profit transition. Microsoft gets 30%. Open AI employees get 30%. Open AI 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 percent, but they're sort of getting hosed.

And so LAs are standing up for the venture capitalists. Yeah, he must be speaking up for Josh. Yeah, I would imagine. Just bigger ownership stakes would have been more. I don't know if it's technically theft, but maybe he's using that kind of in a loose way. I think we need some kind of Robin Hood figure to take from the rich and give to thrive. Yeah, exactly. Yeah. Take 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 at some point. But they did it.

Flock launched a American-made public safety drone response system. We were talking to the CEO, 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 in their 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. Nakunj says, accidentally said sales instead of distribution, and they kicked me out of SF. You got to 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.

I got to, we got to ask, uh, Benny off about sales specifically. Yeah. Like, what, what reps are actually outperforming today? Is that the reps that have adopted the most AI and have just increased volume? Or is it, um, is it, is it, is it reps doing things the old-fashioned way, which is, like, actually getting, 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 Open AI, I'm launching Mason AGI, a fashion house creating cultural artifacts for the AI era. Our first collection,

Relic of Thought, is a collaboration between Ilius Sutskiver, featuring his original artworks alongside his signature hat modeled after his iconic head. They're actually doing it. 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 in tangible objects we can hold, wear, and pass down. Working alongside some of the world's brightest researchers building Claude and Chat Chb-T 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 human, 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 Wynn left opening eye to do this. That's a bold move.

You know, been on a tear, you know, and now launching it entirely new. Bold to be post-economic and pursue the arts. Yes, you're right. Become a patron of the arts. Who knows, who knows. It's kind of a classic playbook, John. 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 out-home advertising. Only ad-quick combines technology, out-of-home expertise, and data to enable efficient, seamless ad-buying across the globe. The chat has been on fire.

Mark Benioff is a little delayed. We're going to remix the schedule a little bit. He will be joining later the stream. Taylor Hodge has a great take. He says that he's going to 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? 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,

a hundred feet of the ground, it's fair game to take it out. Yeah. I think that, you know, the police can drive up and down your street. Do 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 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 mayor. Yeah, imagine if the, the UK right now, let's just be real. very, you know, that 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, at your door. I think that might be overblown. I don't know. I feel like, who have you talked? Have you talk to people in the UK that say it's bad? Or is it all just like American posters who are taking victory lap? Because I love an American victory lap. Don't get me wrong. I will jump on any opportunity to, to wave the American flag. I think there's plenty of posts from people that are based in the U.K. Yeah.

So I got arrested for posting the midwit meme about my enemy. And I got, I depicted, I depicted my enemy as the sad wojack and me as the Chad Wojack. And the trained falcon taking out the drone is just too good. Great. It's been the perfect form factor. Do we have this video of,

we have Dante. Should we start early? Oh yeah. Lay out. Yeah, let's bring in Dante. Speaking of drones, underwater drones, we got Dante from Albuquer.

coming in the amazing name yeah great name there he is dante welcome thanks a much for joining us fantastic fantastic name i thought we were out of incredible names but albacore is strong albacore's a great one it's amazing i like the lord of the ring stuff was taken so we had to get created going to the tuna section do you know the uh the the headline that uh is is in the the group chat is why C-backed Albuquer swims into $6.5 million seed round. I love it. Hit that gong. We've heard it all. We've heard it all. We love making marathon funds. Congratulations. It's great stuff. 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 grown-focused fund. They're sort of mix of U.S. and Ukraine investments, but they backed a lot of folks like Neuros. I've got that. They made two sequels to D1 Capital. We're on D3 now. We're never going to be the names there. You just go to D4 and then D-5 and eventually go to D-10, D-11. A-17 Z is going to be a ripper for sure. Who else? Sorry, we got, we interrupt. Oh, yeah, no, all good, all good. Yeah, so Brave Capital, which is a bunch of ex-Incetal 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. That's great. A lot of different investors. You got Box Group in there. Oh, I love Box Group. Yeah. They've done a lot of defense.

I think they also did. Mock Industries. Oh, cool. pretty recently, yeah. How did you get into this? Yeah, so prior to this, I actually started a battery tech company, and this was 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 with drones, and started thinking about the power problems there, You know, 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 do underwater drones have just like such a 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 than 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 Anderol? 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 some sort of niche. Yeah, I mean, so actually the founders of dive were also investors in us really early on, Sam Rootson, Bill Lebo. They're super knowledgeable about the space, obviously. I was actually just with them earlier this week in Boston. Yeah, so the dive platform is focused on going really deep and doing survey type work. We're working on loitering munitions.

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 succinct that submarine. Wow. And what are the kind of, I don't know how much you can share, but like what, what's the goal in terms of, 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 have, use it for, 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 for endurance and a thousand 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. That's 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 fast. It's faster than anyone in the office. You lose a finger, though. That's 10,000 aura points. For sure. That's like eye patch level.

Yeah. We docked his salary by 10%. Oh, stop. That's so rude. Stop it. It's very dark. What's your path to program of record? How are you attacking?

Yeah. I mean, so contracting 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 a defense autonomy working group that just got released. There's a lot of things that are moving through programs like that with Socom. You know, Indo-Paycom has various flush funds that are available and we're, you know, we're focused on really connecting with the operators and making sure that, you know,

from the start of founding this company, we were really talking with end users of people who are going to be at the tip of the sphere in the Indo-Pacific could 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've worked with a firm now.

But, yeah, you have to basically approach it from all angles. And, you know, there's also a lot of, you know, FMS interests 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 I mean 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 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. We're a lot of the customers that we'd be speaking with are based. So it's a very quick trip. We can get up here on 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 too many hard tech companies that are based here. There's one that got pretty big called Ghost Robotics.

They make robot dogs in military. Yeah. But we're based in the Pennivation Center right now, which is where Ghost Robotics started. 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 Naval Yard, it's being re-industrialized, but it's actually not owned by a U.S. 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, you know, really strong. Amazing.

Does your office have a pool? So we're getting a giant fish tank and also a turtle. A turtle. But we don't currently have a pool. We have limited space. What kind of fish tank do you need to make sure that like 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'd have to look into the specifics more. You got to call Keith Rboy. He's got a big fish tank. Oh, yeah, he does. He's famous. Famously fish tank built. Well, thank you so much for coming on the show.

Great to meet you. I'm sure you'll be back on soon. And congrats on the milestone. Yeah. Yeah. Thank you guys. We'll talk to you later. Have a good one.

Cheers. and if you're interested in some hardware for your wrist head over to getbezzle.com your bezel concierge is available now to source you any watch on the planet seriously any watch under Carpathie

launched a new repo nano chat 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 8000 lines of code to actually build

basically chat GPT from scratch Is that how I should understand it? Yeah, so he's done this whole, like, a course, lecture series on YouTube before, where he goes through, like, different parts of how to train a model. 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-trained. 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 datasets, basically, for everything. Yeah, yeah. I forget the exact name,

but there's a good, like, RLHF data set that you can use to, 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 Ander, Carpath, he says, good question. It's basically entirely handwritten with tab autocomplete. I tried to use Claudex. agents a few times, but they just didn't work well enough at all in net unhelpful.

Possibly the repo is too far off the data distribution. That's kind of interesting. I think part of it is that like the end product of this is like the repo itself. So you want very like nice code that is very readable. It's in his style. I mean, he can like explain everything super easily. Yeah. Which is like a little different from if you're building like actual product or just like a

vibe coding thing where the endpoint is just like the interface. It's where you're actually like using where this is. is like the code itself is the product. So it's like you need super nice, you know, formatting and all this like kind of custom stuff. Do you feel that like the, like the, whether or not the task is in the data distribution when you're firing off Claude Code or Codex agents? Like are you noticing that it's like, okay, well, clearly, you know, there's plenty of data around how to build like a mobile, a modal or, you know, some sort of like rest interface. And so it's really good at that.

But if you're trying to do something that's, you know, probably a unique problem that there might not be, like, a direct open source example of, like, you're going to try and code it from hand? Yeah, that's probably true. Like, I find it's really good at writing React code and, like, tailwind CSS components. And then when I try to do some, like, custom, you know, AWS flow, it's a little harder. And it just takes longer. But I still find it does it very well. yeah uh jordan did you see that uh tyler cowen put the timeline in turmoil taking shots at europe so tyler cowen we talked about this yesterday he said that uh the european union is run by uh 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 one percent growth and uh and is a pretty pretty hot take uh cassie pritchard said never understood why people liked this quote the e u 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 Cowan, though, taking shots at Tyler, friend of the show, and young macro says, difficult to overstate the extent to which you need to be under-exposed to macroeconomic data to think that Tyler Cowan diagnosing Europe's growth as anemic is some sort of contentious statement, or that his negative 1% hyperbole is

anything but mild and contextually entirely suitable young macro continues and says i will offend tyler cowan i will defend tyler cowen with my life uh difficult to overstate the extent to which one's large audience must skew uneducated for this take to pass through with broad base support nothing to do with pedantry 10 out of 10 pedantic domain experts would say tyler tyler's quote is quite fine tyler cowan there's no way a regular guy would lose to a hobbit in a fight a regular guy is what six feet, no chance. A 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. One of the greatest up-and-coming posters. Should we watch this clip from Roloff Bota from Uncapped? He is the steward of Sequoia, Rola Potta. 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 they've 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 yeah 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 dollars a year and numbers you know these all estimates let's just say it's 250 billion a year a lot of exits well let's just do some very simple arithmetic for a second 250 billion dollars going in every single year if you assume that the firms generate 12% IRAs net net net of fees

and carry which isn't that great by the way i mean over the last three or four years the nassad stakeholders compounded at 16, 17%, let's just say 12%. Not spectacular. You basically just average performance. You'd need a 3.7x roughly. On 10 years. On a seven-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. 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 is that coming from? Well, Figma's worth what? It's a lot, but it's 40-ish billion. Let's say it's worth you on 0.03 trillion. So, you know, if you start thinking in trillions and Figma gets you 0.03 trillion. You need 30, 40, 50 Figma every year.

Yeah. To make that arithmetic. Figma week. It just, I don't see that many companies of that scale every year. So the only thing breaks is the return assumption doesn't hold. Yeah. And so Venture is a. return-free risk.

Not a risk-free return. That's terrible. You are better than investing in the index or holding T-Bills, honestly. And so I don't think Ventures and Asset Classes scale if you add more money. You can build more real estate. There's a lot of equities.

There are, you know, trillions and trillions of bonds to be purchased. Venture capital doesn't scale with more money. I mean, the most important, I mean, takeaway here is Bobby's question in the chat. What's on the wrist? risk check we'll work we'll work on we will get to the bottom of that venture return free risk is a crazy line yeah um taking shots but uh there was some good

ball up from harry yeah rogavan he said obviously roll off'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 is 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 an order of magnitude, the exit expectations make a lot more sense for true VC that is, over, let's say, eight to 10 year horizons, four to five X, so 100 to 125 billion in investor ownership at 150 to 200 billion in exits. For the PE growth equity class theoretically of

1.7 to 2x, clear similar hurdle return because the hold periods are, say, three to five years from series D, E, etc. 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 a 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 centicorn, which I guess... Triple 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 centicorn, if a unicorn's a home run, a decacorn's a grand slam, a centicorn. should be like a World Series clean sweep, something like that. Like you won the whole season. You're saying, I'm not saying we'll have one Open III per year. That'd be crazy, but we could see per year. One, impropics, stripe, data bricks, et cetera, two to three, or three to five Figma scale outcomes,

20 times wealth front 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 10 years ago. We had zero trillion-dollar companies. He says, maybe I'm just too much of a techno-optimist, and a ball comes in with the reply and says, nailed it.

Correction on wealth-front. Wealth-runs independent. You're thinking of a different company. Well-Front's the robo-advisor. Oh, Gusto, guideline. Sorry, guideline. Subway different. yeah the interesting like nuance here is how weird open AI is it just breaks the venture model

because yes it is you know 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 open AI and then also it just completely breaks the whole model of like who's founding these companies same altman's like much more experienced much older than you know, some, uh, 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

dollars at like a $12 billion valuation. It was like, you know, building like a 1% ownership in the firm. And so, uh, it's, uh, it's, it's, it's such, it's such an outlier. But this is the, that's the game of venture, right? That's the game of venture. Open AI just hired prize-winning black hole physicist Alexis Lopes Lousaska, as the first member of its new Open AI for science team per Axios. Good timing here because they are leaning more into science. Lopes Saska is a Vanderbilt professor known for his work on black hole photon rings, and now he will help shape how GPD-5 tackles advanced math and physics problems and guide open.

AI's push into scientific discovery. VP of science, Kevin Wheel, said GPT-5 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, open AI should hire you. I'm going to hit him up after this and say, I'm sure he said something like that. That sounds like it would 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 Open AI. Yeah. Well, we should prep for our next interview with ESO 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. when we talked about this yesterday, but CoreWeave is partnering with them to build 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 into, to the TVPN Elfrey.

Welcome to the show. How are you doing? Hey, guys. Good to see you. How are you doing? So pumped to have you. Yes. 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 pool size going this hard? Yes. So please. And when we realized that it was you, we were pretty excited. Yeah. So please tell us the story of the company. Can you start kind of at the beginning, a little bit on your back. 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 Andre Carpathie, I'm writing code. I've actually never said this. My story into space started because Andre Carpathie 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. He made an acquisition offer 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 had the world's first

code completion models that were working back there. 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 2020, you see Chachapit come up. 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, our view, and it's why we started PooleSight two and a half years ago, was their reinforcement learning was going to become the most important scaling access for model capabilities. First 18 months of the 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 cloud code. There's agents. There's windsurf.

There's cursor. And there's so many different positions. Like, how did you see the market map develop from your world? So we bring it back to like what ultimately our mission is, right? We're here to reach human level intelligence and go beyond. And in that world, intelligence in our view is a commodity. It's actually a commodity that gets created by only a small number of companies because it is 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 what I often refer to internally is that the barrels of oil business, right? 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 wide for projects 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 wanted to be for the world's, frankly, knowledge workforce. We wanted to be for the enterprise. We wanted to be for the world's most high-consequence environments. So our models are now rapidly progressing in capabilities. It's definitely been non-linear, 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 was really no one else was out, which was defense and government. Oh, interesting.

And so kind of our first customers were not just building the model, but also putting all the crazy enterprise stack to be able to deploy it anywhere. Like literally, and workstations and Humveys. all the way to like the larger models and like air gaped environments or guff clouds or places where you needed ATOs. 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 two. for yourself, it's less fun to build tools for a role that you've never done, right? Yeah, intelligent is, we treat it as this one North Star like intelligence, but in reality, it's actually really multidimensional, right? Like how good you are at writing poetry is very different than how good you are at writing code versus how you are, you know, being a researcher and biology. And so the thing that was the unlock, I think, in our space is that, you know, first generation of models had only been trained on the output of humanity's work, right? Like the way.

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. 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 basis where our agents can do hundreds and hundreds of billions of tasks. And 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, That's two, where it's like 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 H-200s. Think of that as like annually $1,20, $50 million worth of compute. But it's orders of magnitude less than what others had. And so we built our efficiencies around orders of. of magnitude, more efficient experimentation. But your big model run is still your big model run.

And so the size of a model you can train has a duration at which you train it has a clear correlation with the final intelligence and capabilities. So we 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 skilled 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, right, and for kind of obvious reasons. And we got really excited because we found the path to partner with CoreWeave that brought online more than 40,000 GB300s 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, you know, like the compute market. But right now, that scale of compute is impossible to get. It is sold out for the entirety of 2026 and already, like, well into 2027. So we found that a strategic partnership that allowed us to do so. And 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 frankly, you're not in this race. You're just cosplay. And so we had to take a step back. This was already a while ago. This was already starting looking at this, you know, a year ago. So 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 it's 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, well, last year I could call someone for 50 megawatts and I could kind of, you know, get it within six to nine months. If I needed to call someone for 250 megawatts, 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 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 side as well right we've heard you know there's plenty there's labs out there that have like they 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-founded like a day zero memo. We often go back to it to see, you know, like what's changed along on that. And we kind of understood early on it. 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, right, if you treat it like a barrel of oil, a barrel of tokens, someone else is going to deliver it, right? Exactly. And look, and so 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, 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 it's probably here and 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 uh 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 two megawatt facility, that feels like leapfrogging a lot of the one megawatt clusters that we've seen

talked about, whether it's Colossus 2 or what meta's doing with Prometheus and what Open AI 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 six gigawatts of gas that sits there that we bring turbines online for it 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 skill to how long it takes to get online. And this is where the partnership with CoreWeave is really interesting.

Because with CoreWeave is 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 pool side, and how much of that compute we might have to put in the free market. And there are, of course, world class. Like, I mean, I cannot speak highly enough of their ability to operate, like large-scale compute. And so for us, it was, it was not about the big headlines, but it was about building a company that could incrementally deliver data centers. And the first 250 megawatts are actually to deliver it in a quite unusual mass. matter. I 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 1,000 GPUs, another 1,000, not 1,000. And so we took a different approach here where 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-side manufacturing.

So a data center for a GPU compute is effectively three layers. It's an electrical skid. It's a cooling skid and it's a compute skid. And you actually designed them that they fit on the back of a flatbed truck. Ah, 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. We will talk to you later. Yeah, let's run it back very soon.

Thank you so much for joining. How are 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 mark benio from sales force he is the CEO of sales force let's bring him in to the show thank you so much for joining um our team will sort this out in just a second we can see you we can't hear you we are working on this um let's make sure that uh he is in the restream waiting room joining and i wish uh i wish we got to get eso back on the show i see him i don't hear him yet We got the ESO.

Hey, there we go. How are you doing, Mark? We got audio. Thank you so much. You got audio. You have John Coogan and Jordy Hayes from TBPA. AGI is here, but we still have video calling difficulties. But here we are.

It's amazing. Yes. I mean, it's unbelievable. We're feeling it. We've come. I mean, all of the AI is, you know, is it ESOCon or ISOCon, I can never remember. I think it's ESO, ESOCon. It's ESOCon.

It's ESOCon. Is it miso soup or myso soup? I think it's miso soup. Okay, and it's Esau Kahn. Okay, so Esau Khan, yeah, I love Esau, by the way. Oh, really? And I love Poolside. Also, one thing you guys know each other.

One thing you need to know, one thing you need to know. I also love Poolside, the music. Have you heard of you? Yeah, Poolside FM. Yes. And they do that great Neil Young cover. Yeah. Oh, the band.

And Harvest Moon. Yeah, yeah. If you haven't heard the Poolside cover on Harvest Moon. Yeah. And Eso Khan and I were talking about that. And I said, why did you call it Poolside? He's like, oh, this. and, you know, he's building this huge data center now.

Yeah, yeah, yeah. 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, pool side. 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. Uh, anyway, please give us, give us, give us, give us, how you feel? How you feel? 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 going to have to let Lars Orich 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 Sass. Why? Why? I mean, this is our 23rd Dream Force. You haven't been to any of them. I want an explanation. I want an explanation. I want to know why. You guys are the number one podcast in the world. I want to know why you're not at Dream Force.

And I want to know your music choice. I do love Metallica. I grew up on Metallica. Yes. Okay. I don't believe it. What song do you? What's your number one Metallica song?

Enter Samman. Oh. No, no, no. It's a layout. No, no, no. No, it's okay to like the best. That's like everybody knows that one. Let me ask.

It's, it's, it's, it's, uh, what do you got, what do you got? You're unforgiven. I do like Master Puppets. That's a great one. Master Puppets, plastic. But I was always in a little bit more of the slip knot, a little bit more of the corn and the tool. But, you know, I mean, Metallic is great.

It's definitely in the heavy rotation. It's on my metal playlist. We not only have the greatest heavy metal band of the United States here, Metallic, really the Bay Area, Lars Ulrich, and Rob Trujillo, and, you know, Jimmy Headfield was here. and Kirk Hammett, probably the greatest guitarist of all time, if you like that. But we also have here Yoshiki, the head of X Japan, who's also one of the greatest drummers of all time, and X Japan also, he does classical piano, and he's about to go on tour again, back to Madison Square Gar and Royal Albert Hall.

Amazing. But anyway, hopefully you guys will come next year. And works for X, the Everything app, that's correct, right? That's great. X the everything app. This is the different X. This is X Japan, not related to Lato. Okay, got it, got it.

I saw your post about being on X and I wasn't sure if it was a different thing. This is, we are X. Okay. We are X. Yes. It's never been more of a confusing time. Your Japanese viewers are going to go up. Oh, no.

Say, we are X. We are X. Which is Yoshiki official. On Twitter. Yeah. Or whatever it is. This is so fun. We got to, we got to, 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, uh, which somebody was highlighting that open AI runs on both Salesforce and Slack. Thank God. I wanted to ask you, does not, does nothing, does nothing ever happen? Nothing else matters. Nothing else matters. There you go. Yeah. Um, you're still unforgiven by the way. A super intelligence would use Salesforce and Slack for sure. That would be the first thing that they pull off the shelf. If you are your Slack user, you need to see the Slack bot that we introduced here, which is built on Anthropic. It's also amazing. So we use OpenAI. We use OpenAI. I just did an interview with Sundar. You know,

you can see it on YouTube. We, 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. 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? Tensore, right?

Yeah. He's got, he does a lot of 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 if the future is, well, what is the future? Is it billions of fast food restaurants on the planet, you know, all 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 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 gone for the first prompt. Absolutely. Gone. Gone.

Thank you. Thank you. Thank you. And listen, 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-Cont. Was it ESO or ISO? ISO, right?

ISO-Cont. He's in 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 Kwan from Alibaba. You didn't mention that. 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 will. We've had 23 of them, by the way, but I know. But here's the thing. Jordan's barely 23 years old. One of our team members over there, most of us weren't even alive. you're doing podcasts all day long you're doing your podcast all day long you're doing this podcast go

yeah yeah but listen listen i get it i know you're too busy for me but now listen to this listen to this is this interview is so fun we'll fly we'll fly no no no i know you're busy i get it i got it you're a master you're master no no i understand you're 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 sizes. They're small businesses, companies like zero to 200 employees. There's 200 to 1,000 employees, 1,000 to 2,000 employees. Very large enterprise, the Pepsi's and the Dell's. Michael Dell was here. You're not here.

Michael Dell was here. Okay. Laura Albert, the CEO of the Williamson. No, I'm 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. And that is actually complicated because you have different governance, like I was saying, different compliance, all kinds of different rules, based on country, so 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. Great. Okay, we love that. But here's one more 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, and 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, but we're not. He's there, 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 conference. Hit that gong again.

Mark, one question for you. Salesforce has been hugely acquisitive. What is your pitch to founders in 2025 that you want to acquire? That's great. Like, why is it such an incredible 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 Wiz, 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 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, our own. arm that acquires but also invests. It's a huge part of our business. It's a $5 billion fund. It's, I think, doing a 33% IRA. I don't know exactly the number is probably one of the highest performing funds in the world. Let us partner with you. Yeah. You know, come here. You know, these podcasters on F time, you do, come here, be with us. Customers are here to meet you. You're good at sales. I've got to say. You kind of got to be in the role. What's the current thinking on C-based pricing versus consumption, both at sales, or what you advise founders that might be working with you.

Yeah, I think people have this obsession with seat-based pricing right now and this idea that AI 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 understand they buy, 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 churn or they'll find another. But how do you think about this conversation? around seat-based versus kind of value-based pricing.

What a great question. Look at it. I just subscribe to ChatGPT. It's a seat-based model. You know that you probably pay $20 or $200. I just subscribed also in the XAI. Did you use that, by the way, to make that picture with Sam? It kind of that picture.

I use Brock. I said, because we were wearing like conference necklaces. Okay. And I said, take it off. And Chat-GPT goes, oh, no. No, no, no. Copyright infringement. And then Grock 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 C-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, 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 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 customers on Slack. Hopefully you guys use Slack. We do? Of course. When you have that many customers and 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.

We only had one product, one price, 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, 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, a quality 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 going, you know, like, by the way, for example, you probably heard of the Beast. Jimmy Donaldson, his team is here. They're running on Salesforce.

They're a huge media company now. Everybody knows the Beast. Mr. Beast runs on Salesforce. Wow. You know the Beast. The Beast. The Beast. He uses Slack very aggressively.

Of course. Jimmy is amazing. Yeah. Okay. He runs a great business. It is growing like a weed, Beast Media. all of these things. We have to adapt to him.

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 great life. You probably feel blessed every day that this is your career.

We do. And I feel that myself. I feel blessed every day that this is my career that I'm able to do this for a living. Last night, James Headfield, 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 makes me happy. I love making customers successful. I know that's what you like. You love making your listeners and your viewers. You're the final boss of Enterprise SaaS.

I absolutely love it. What makes, why do you believe that sales is still a great career path for young? 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 agent for sales rep companies getting funded and 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. What's your kind of world view 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 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, 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. 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 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 want to just say this to you I also just hired somewhere between three and five thousand more salespeople. Wow. Okay, I'm growing my sales force. It'll be more than, almost,

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 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? 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, look at 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 more deeply,

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, A.A. is not born. It's, you know, AI is made. So, you know, we, there's a humanity that still needs to be nurtured and governed and developed 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 that's my job right now right that is my job yeah 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 one to ten the ten one of the best communicators most aggressive salespeople in the world where did you put him one or 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 he's 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 if 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 tight. So trashed. Last week, I've been on the road nonstop.

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 MISO. You're doing great. I just want to thank you. I wish we didn't have to fade to black. We'll talk to you soon.

You see what I did there? Oh, yay. I know you Googled it. You're still unforgiven. 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.

Thank you so much, Mark. We'll talk to you soon. Great to see you. Have a good day. Super fun. Bye. Fire me up. Yeah, the maugging counter is relentless.

We got to add that. The chat is having a lot of fun at our expense. I wouldn't have it any other way. Everyone had a great time. Thank you so much to Mark Benio for hopping on the show. I am the show is in chaos at this point the timeline is in turmoil a show schedule is in turmoil I believe we have more guests I hope so you 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

Mark Benioff really knows how to take control of an interview it was a fantastic conversation but but we're here to talk about you so please introduce yourself and give us the news yeah great hi I'm Alice Benton 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. May I to share. Yeah, totally.

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 Cozler Andreessen, Zeta, Sousa, True Ventures, Basis Set.

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. And being actually, like, in the room and talking to so many companies, you start to pick out really clear themes. Get us to the frontier. What's the October 2025 theme or different, like, 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-ide. We help them go through that process of actually developing the idea from scratch. And Connell, the CEO, he had actually spent a summer spending eight hours a day hand processing these sort of purchase. orders for 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 Cherto, and what they're doing is it's like Vantor for FMCG. So it's automating the compliance process, but for fragrance companies, for cosmetics companies, is this is something that's holding back $400 billion worth of new products every year because it takes months and months to go through. We were just talking about the lead. Well, and probably a year ago at this point,

I had somebody pitched 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 office? audience around what, like, what's the bar to come out of Demo Day with, you know, a seed round from your view? So 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 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. And 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 the 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 it's here today, gone tomorrow, you've got to rebuild it. And often our company is getting, you know, 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 stay ourselves in the same way. We want to be the CAA for founders.

We want to be in their corner as their talent agent. And so we go out into communities across Europe, across the U.S. 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, their pre-team, their pre-ide. Our job is to work out from their behaviors, which are 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 on an application. You only see that from spending an extended period of time with these people.

So our selection process 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 if they're found a potential by looking at a piece of paper. Yeah.

Well, congratulations. Hackathon is a very cool selection process. Hit that gong for the whole badge. It's great to meet you. And I hope you back on the show soon. We'll talk to you later. Amazing. Have a good idea.

Thank you. And we have Eric Sufurt from Mobile Dev Memo. in the Restream Waiting Room. We will bring him in to the TBPN Ultradome. We are getting back on track. Eric, how are you doing? What's up? Sorry to keep you waiting.

Sorry about that. Absolutely chaotic day. I'm great. I've enjoyed your guest appearances on podcast, 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, Freeman 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. Do you, uh, uh, it, it seems like one of the main, uh, like, I don't know, takes you have that you're just waiting to cash in on is this idea that open AI will be doing advertising soon. Um, can you walk me through like how you perceive the messaging from open AI the firm and then how like what led you to believe that it is inevitable that we will see advertising. Yeah, that's my, my modus operandi is strong, uh, strong opinions forcefully stated. Yes. Uh, I've seen. Uh, uh, I've seen. Uh, uh, said very forcefully, Open AI will monetize with ads. I wrote a piece in May. The title was obviously Open AI will monetize with advertising. And here's where I think the sort of like most

profound clue has been. So Sam Altman in Ben's podcast last week, Ben Thompson, he said he pulled off like a rhetorical sleight of hand. He said like, hey, look, you know, we're not really considering ads, but I really love what Instagram does. 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 products from those ads. Those aren't ads.

That's a separate thing. Yeah, 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. And that was a Spencer Neumann moment. That was the Spencer Neumman moment.

It's a Spencer Norman, Netflix CFO, March 22, Morgan, Morgan, Stanley TMT conference, he said, look, it's not like we have religion against advertising. It's not in our plans 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 Norman moment. That was March 2022. When did Netflix launch ads?

November 2020, 8 months later. Obviously, they were already working on it. Yeah. And by the way, I would dispute that statement that it's not like we have religion against ads. They very much exhibited religion against ads for years and years and years. That's at 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 opening I with chat EBT. I said yesterday they're ineluctible. I got made fun of for using that 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 a 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 it 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 straightoffs, 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 you you get the the sort of the imperative is to reduce consumer 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 yeah some people the only accessible price point is zero right now for some people now now you have to weigh that against perception right and there's perception risk like Netflix I I had sort of argued at one point that I thought Netflix to go fast a freeming ad freeming uh sorry free ad supported tier yeah uh and

and just drop the price point to zero and have ads. Now I've changed my mind on that. I don't think they should do that just because it would, you know, it would 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 Chat-GTP-T has that risk, right? So I think Chat-GPT can introduce – so first of all, When we talk about ads in ChatsyPT, we're not talking about ads in the paid tier. I just want to be clear about that. We're talking about ads in the free tier, right? And so what's down the downside is. It's like if I'm on the 200 a month 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 yesterday the Financial Times posted some data about Open AI. So 800 million users, we knew that. 5% are paying, right? So you've 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 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's 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 ARPPU. It's ARPPU. Average revenue per paying user, right? But if you do the calculation for just ARPU, it's $11.38 cents ARPU, right? So how does that compare? That's less than META's. 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 Open AI for that, and we saw, that in the numbers as well, is that inference isn't free. Inference has, 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 chat chief chat, it kind of feels like a feed. I think the real challenge with chat to be, and I've seen like a lot of conversations about, well, the difficulty is going to be like the actual. or it's going to be the sort of middleware that connects things. I don't think, I think this is mostly a solved problem. I don't think attribution is going to be any different.

Attribution meaning, how do you credit chat GPT 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 question 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 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 the yeah because right now you're getting the average product recommendation from Reddit right yeah and then if they start to put their kind of uh they start to kind of tilt the scale yeah what are the landmines in terms of like just describing the nature of the five

between the results when you ask chat chipt be 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 an objective that's actually influenced by whoever's just paying the most to show the product to you then you lose the consumer trust right and actually that's like that's a very specific problem for chat bots 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, Google's, and the 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, right? Because the sort of the end price that an advertiser pays is actually modified by the relevance. So with chat TPT, though, so if you just see a stream of links, if the top link is an ad or if it's not an ad, if it's the most relevant, either way, everyone wins, right? But those interests are not aligned in that way with the chat box, 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 was paid for, that was paid to be shown.

But the other thing is like I, 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, and they kind of, they all serve e-com, 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 AI, 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, right? That's what the advertisers 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 open AI would be leaving money on the table by just sort of saying, okay, well, let's do an auction. But, sorry, like, just with the flat fee and then we're just going to like decide what to show that. Right. And the other problem with that is from like a merchant perspective, you don't have any control there. Right. So the thing is you can't change the price to sort of absorb that market.

margin shop, right? 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 sort of recover the the ad spend, but also make money on the checkout. But you can't really do that with the instant checkout because you just sort of submit like a JSON of your catalog, right, with the various price points. So I think instant checkout might, and the point I made in my blog post is 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? Like, 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 things 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, they don't, they don't, there's no, 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, you know, 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. You just, it knows that you need paper towels.

We're going to show you an ad for paper towels. It doesn't matter if you went and searched for, you know, the history of the Roman Empire. Yeah, we haven't talked about. 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 issue with that. Well, no, but so that, I mean, I think that could be a, so that's what I, that's kind of what I meant when I talked about, like the format's going to be the real challenge here. Like, I mean, and they've got, you know, exceptional people to tackle these problems. And I think one thing about,

you know, open AI that people sort of like discount is, is the number of people that they've hire from meta, right? So, I mean, like when people say, well, they don't have the team yet, of course they have a team. They've been hiring people for meta for years and years who've 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 history. 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,

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 Saurus side, do you think they can transition from a creative tool to a actual consumption? platform. It's still number one in the app store, by the way. It's fascinating. It's very sticky. So I think 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. Yeah, 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 hours and hours and hours.

or Pinterest or TikTok. Yeah. So 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, you know, opening I was not short of top downloaded apps, right? They've got chat GPT, and that was a top downloaded app. The thing is, like, you look at the staying power of these generative apps, and it's, you know, they're pretty short-lived, right?

I mean, like, you know, Lensa. What's that? Lenza. You remember that? Yeah, Lenza was like the original. Yeah, that's right. Yeah.

But like, remember Deep Seek? Yeah. Like, you know, I mean, the thing is like, you know, and then whichever, whichever company releases like the next, then sort of the latest next gen model, they usually take to the number one spot, but they sort of fall off after, you know, kind of a short amount of time. But do you think deep seek? I think it's a really novel idea. Do you think Deepseek was was paying for downloads?

Like, it didn't seem organic at all. Um, yeah, that's, I, I looked into that at the time. I don't actually remember what the conclusion was. I think they probably were. I mean, they weren't giving away the first reasoning model for free, like a week earlier than Open AI. So there was just like, it's the hot new thing. Yeah, but the general public was not like, I need to try a reasoning. 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 two, 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 Reel is a very serious competitor. Now you have SORA. How are you thinking about TikTok in the, in the ecosystem right now? I mean, TikTok not getting banned was a foregone conclusion. I wrote, I write an annual predictions posts 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, you know, or metas. I think, so TikTok is facing some challenges with its social shopping product. Yeah. Right. And I think that was the challenges that they're losing, it's a massive amount of money. Like, it's not really, it's like a retail platform that is massively loss making. Yeah. And they scale 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. So I don't think there was any news with this deal being reached. I think when people get worried about the algorithm being retrained, keep in mind, that's the content

algorithm, what the ads algorithms. 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 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's bringing its video podcast onto the YouTube platform. 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, for, you know, 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 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. 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 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 excited do you think the average app developer, business owner in general should be about

open AI launching ads because I know a number of businesses over the years that were basically be born because a new ads platform, like I have a friend's 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 get to hyperscale on Facebook in the early days. And so I assume Open AI will want to make the ads product cheap and performant early on to just drive this huge influx of volume. But I'm curious what you think the opportunity will be early.

Everyone should be excited about open AI launching an ads 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 DDC category won't exist. E-commerce is much smaller than it was. There would be far fewer small businesses in this country. Ads is a growth engine, not just for, you know, e-com 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's 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. 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 know, if a camp, if you spend 20, you know, 20 grand on a campaign and, CAC is too just high.

It doesn't make sense. You just turn it off. Like, that's the beautiful thing. Exactly. So, I mean, I think, you know, there's a reason to be very optimistic broadly about the new opportunity that will be engendered by, just by LLM empowered content engagement, right? Or AI empowered content engagement. It doesn't have be just LLMs. But I think, like, where people get mixed up is they think, well, you know, these ads need to be different.

Or 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 dry performance and you get conversion of 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 things. It's getting applied to the back end, right? That's what annoyed me about Facebook's last earnings or medist's loss earnings because people are 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 when 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 is recycled. And also those products are not static.

They're not frozen in time. They will also continue to improve. So the thing is like 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 about? And the thing is like 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 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. Incredible rant. I feel like I just got it. 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. Mostly, I just got fired up. This was 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. Thank you. Great to see.

We'll talk to you soon. Cheers. Have a good one. That was late. One more guest before we get a gun on with London. Intuition. 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 TVP and Ultradem. Welcome to the show. Sorry for keeping you waiting. Give us the news. Yeah. We've raised $133 million, $1.7 because we're gamers. 0.7. $0.7, yeah, to build general agents.

for environments that require deep spatial and moral reasoning. 1-337 LEAP. This is a gamer reference. Tell me about your, didn't you make money as a gamer at like 18 or something like that? Yeah.

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 Rinscape when I was a teenager did about a million to have in revenue about a time and short 18 years old. Yeah, it was pretty funny.

that's incredible I mean people come by with you know a hundred million dollar fundraises all the time but that is that is extreme well we don't we don't see 133.7 million dollars seeds 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 we're building general agents for environments that require deep spatial and moral reasoning so what just means is like look at drones for instance For instance, robotic arms, they all ship with game controllers already. So there is this general interface for applications that's already in all the robotics where we don't need to reinvent a wheel. And so the bet that we're taking is that we can train 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 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 Doctor 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, you know, it can basically cover a lot more space, a lot more quickly if you have general agents that can navigate these types of 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, you know, 100 square miles in an hour?

Yeah, I suspect there will still be like VLMs on the other side initially that sort of analyzed 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. Like, it's not, 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 we have these inputs that already gamers use to control in all these very diverse environments inside video games,

including things like drones, right? That then 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 sold for safety, right? And so 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 the specific action space, such as navigation, right, 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. And so, yeah, so our take is, so we're already working with game developers to get these models deployed into their games, which means much more fun playing video games. And then it's also a great flywheel, right? Because if we can verify, we prove

that these models are performing great against game players, then, you know, that's a great benchmark to also start transferring out to other environments. Uh, walk through the usefulness of something like Unreal Engine versus, uh, I'm going to mispronounce a Gaussian splatting and then a genie three style like generative world model where you're generating the flame, the, the frames on the fly. Uh, 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 an individual one of those technologies?

I, I've thought about this question a lot. 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. Love it. So, so, um, the most computationally difficult things 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 just have, you hit a point where you just have to bet on video transfer.

right and and so and also there's then in the same time for instance there's loads of things that we currently do not have video footage of which we cannot really bet on those things right and so like like cells and you know smaller smaller things that like just aren't well represented those type of things right so so you always have to use a combination of the two you always can have to use a combination of two the other thing is that when you sit inside like 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, I think where we end up is, it's actually really annoying not to have that like determinism in these world models, right?

Because you want to be in some form of variable domain. So I hope that somebody manages to create some form of hybrid architecture. Like, okay, so for instance, if you go on with chat, GPT, like 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 way, You draw Python, right? Yeah. Have you 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 build the Chedongued. So my point is, you know, we might actually end up finding, like, really interesting. Like, my dream is to be able to simulate a CPU inside a world bottle where he sort of

have some form of determinism. And, again, this is not at all possible or near possible today. But my point being, like, maybe a world model can, yeah. Yeah. And so... That's awesome. Yeah. So 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, 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. I'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. This is great. We'll talk to you soon. Have a good rest of you, 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. Sorry to let to leave you hanging there for a second. Oh, you're good. No, I'm fine. I didn't even notice.

Book of Wonder with Inspiring Views, Hotel Great a Mani's Dreamy Beds, top tier cleaning 24-7 concier service. It's a vacation home, but better folks. And good luck to Ragov in the chat. he is going to a job interview and we wish him the best leave us five stars in apple podcast and spotify and we will see you tomorrow thank you for tuning in can't wait goodbye have a great evening