NVIDIA Beats Earnings, Google Launches Nano Banana Pro, 𝕏 Timeline Reactions | David Chang, Loredana Crisan, Tarek Alaruri, Tony Zhao, Nikita Rudin

(00:15) - NVIDIA Earnings Reactions (18:23) - Google Launches Nano Banana Pro (22:56) - David Chang is an American chef, restaurateur, author, and TV personality, best known as the founder of the Momofuku restaurant group. He has received six James Beard Awards and was named one of Esquire's "most influential people of the 21st century." In his conversation, Chang discusses his culinary journey, the evolution of his restaurants, and his experiences in the food industry. (55:52) - David Chang is an American chef, restaurateur, and media personality, best known as the founder of the Momofuku restaurant group and host of "The Dave Chang Show" podcast. In the conversation, he discusses his diverse culinary ventures, including restaurants ranging from quick service to fine dining, and his expansion into consumer packaged goods like sauces and noodles. He also highlights the evolution of his podcast, noting its upcoming move to Netflix as part of a broader partnership to bring video podcasts to the streaming platform. (01:34:54) - Loredana Crisan is the Chief Design Officer at Figma, having joined the company in August 2025 after nearly a decade at Meta, where she led teams for Messenger, Instagram Direct Messaging, and consumer AI products. In her recent conversation, Crisan discusses the transformative impact of AI on product development, emphasizing Figma's role in creating a creative environment that bridges various disciplines to bring ideas to fruition. She highlights the importance of AI as a tool that empowers designers without constraining them, enabling precise control and fostering inspiration throughout the creative process. (01:49:25) - Nano Banana Pro Reactions (01:56:57) - 𝕏 Timeline Reactions (02:23:19) - Tarek Alaruri, CEO and Co-founder of Stuut, an AI-driven accounts receivable automation platform, discusses the company's recent $29.5 million Series A funding led by Andreessen Horowitz. He explains how Stuut's AI technology streamlines the collection of overdue invoices, enabling large companies to implement the system within three days and recover 40% of overdue payments in the first six months. Alaruri emphasizes the platform's user-friendly design, allowing finance teams to efficiently manage receivables and focus on other priorities. (02:31:42) - Tony Zhao, co-founder and CEO of Sunday Robotics, discusses his transition from academia to entrepreneurship, emphasizing the limitations of traditional research in advancing robotics and the need for real-world applications. He highlights the development of Memo, a household robot trained on data collected from over 500 homes using the company's patented Skill Capture Glove, enabling it to perform complex tasks like dishwashing and laundry. Zhao also addresses the importance of safety in design, opting for a wheeled base to ensure stability and prevent accidents in home environments. (02:44:04) - Nikita Rudin, CEO and co-founder of Flexion Robotics, discusses his company's mission to develop an intelligence layer for various robots, including humanoids and mobile manipulators. He highlights the use of large language models for common-sense understanding and reinforcement learning in simulations to train robots for tasks like object manipulation. Rudin also announces that Flexion has raised $50 million in funding and plans to establish a U. S. headquarters in the Bay Area. (02:53:03) - 𝕏 Timeline Reactions TBPN.com is made possible by: Ramp - https://ramp.com Figma - https://figma.com Vanta - https://vanta.com Linear - https://linear.app Eight Sleep - https://eightsleep.com/tbpn Wander - https://wander.com/tbpn Public - https://public.com AdQuick - https://adquick.com Bezel - https://getbezel.com Numeral - https://www.numeralhq.com Polymarket - https://polymarket.com Attio - https://attio.com/tbpn Fin - https://fin.ai/tbpn Graphite - https://graphite.dev Restream - https://restream.io Profound - https://tryprofound.com Julius AI - https://julius.ai turbopuffer - https://turbopuffer.com fal - https://fal.ai Privy - https://www.privy.io Cognition - https://cognition.ai Gemini - https://gemini.google.com Follow TBPN: https://TBPN.com https://x.com/tbpn https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231 https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235 https://www.youtube.com/@TBPNLive

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Jordi Hays?

  You're watching TVP.

• 0:01John Coogan?

  Today is Thursday, November 20th, 2025. We are live from the TVP and UltraDome, the temple of technology, the fortress of finance, the capital of capital. Ramp.com. Time is money. Save both. Easy use corporate cards, bill pay, accounting, a whole lot more, all in one place. Nvidia beat earnings and job, the job numbers came back very positive.

• 0:22Jordi Hays?

  We are back.

• 0:23John Coogan?

  119,000 new jobs and Nvidia beat earnings. The revenue came in at 57 billion for the quarter, up 62% from this quarter last year. Fantastic result for Nvidia. Of course, that's why the stock's selling off and the market's melting down and Bitcoin's down 10%.

• 0:46Jordi Hays?

  That was.

• 0:47John Coogan?

  Bitcoin's down 86%.

• 0:48Jordi Hays?

  That was my prediction after yesterday morning.

• 0:50John Coogan?

  It was one of your predictions.

• 0:51Jordi Hays?

  But I was, but I was, I was wrong on the timeline.

• 0:55John Coogan?

  Yes.

• 0:55Jordi Hays?

  It was interesting.

• 0:57John Coogan?

  Yeah, yeah, yeah. Pumped shortly and now, now everything's selling off. Very unclear where we go. But I did think it was just interesting that the, I didn't really piece this together until like some grand thesis or actually write a piece about it. But I did think it was funny that we are in a world where demand for robots is surging and also demand for human labor appears to be surging, like Nvidia, you know, the chips that they make sell artificial intelligence. That should be replacing human labor and yet the job demand is, is surging as well.

• 1:30Jordi Hays?

  And it's notable, they, they said they have visibility for a half a trillion dollars in revenue.

• 1:37John Coogan?

  Yeah.

• 1:38Jordi Hays?

  Through 2026, which.

• 1:39John Coogan?

  I mean, it seems crazy, but.

• 1:41Jordi Hays?

  It's not enough anymore.

• 1:42John Coogan?

  There, but I mean, they're making 57 billion a quarter. Just for the next quarter, guidance is at 65 billion. Analysts had predicted that revenue guidance would be 62 billion. So everything is trending up. Jensen said, we've entered the virtuous cycle of artificial intelligence. AI is going everywhere, doing everything all at once. What a great quote. Tyler's very happy about Jensen. And I'm also happy about Restream, one live stream, 30 plus destinations. If you want to multi-stream, go to Restream.com. So.

• 2:19John Coogan?

  We talked about it a little bit yesterday on the show. There's new product from Travis Kalanick, the founder of Uber, of course. It's called Picnic. We discussed it on the show yesterday, and we got a reply from none other than Travis Kalanick himself. Why don't I read his reply, and then you could kind of take us through the, what you wrote in the newsletter.

• 2:42Jordi Hays?

  Why don't I start with a little bit of context, and he can add to it?

• 2:45John Coogan?

  Please.

• 2:46Jordi Hays?

  I wrote in the newsletter this morning, of course, the subject of the newsletter was Daddy's Home. And that is, of course, Travis Kalanick. Travis is back on the timeline.

• 2:57John Coogan?

  It's so good to see him back on the timeline. Like, he's never been, not been doing business, but he's been so quiet.

• 3:03Jordi Hays?

  And just dropping, like, deep alpha on the market that he's operating in.

• 3:07John Coogan?

  Yes, yes.

• 3:08Jordi Hays?

  That is, like, non, like, in my view, like, I never, it makes sense what he shared, which we can get into, but I never looked at it exactly like that. But he's obviously back on the timeline with Picnic. Picnic is a new business under City Storage Systems.

• 3:23John Coogan?

  Okay.

• 3:24Jordi Hays?

  So you don't know the name City Storage Systems, but Cloud Kitchens is actually a subsidiary of City Storage.

• 3:29John Coogan?

  I thought Cloud Kitchens was the top.

• 3:31Jordi Hays?

  That's what I thought, too.

• 3:31John Coogan?

  No?

• 3:31Jordi Hays?

  But it's actually the opposite. City Storage Systems, great, great name if you want kind of an under-the-radar, holding company to, you know, verticalize food delivery.

• 3:41John Coogan?

  Sure.

• 3:42Jordi Hays?

  So Picnic is kind of a front-facing platform focused on meal delivery. The offer sounds too good to be true. Meals delivered from 50-plus restaurants with no tipping and no fees. They also bundle orders, so a company can order from 10 or so different restaurants, get it all ordered at the same time. He's got a bunch of customers already, Wells Fargo, Live Nation, EY, KPMG, PWC, and a bunch more.

• 4:09John Coogan?

  Sure.

• 4:09Jordi Hays?

  And so we were talking yesterday about how, like, broken the tipping experience is. When you're tipping directly, it's a way to encourage great service by, like, tipping if you're checking into a hotel. And you're tipping, you know, somebody on the way in. They're incentivized to make your stay great. Same thing, you know, valet tipping on the way in. They're gonna park your car right up at the front.

• 4:32John Coogan?

  I saw a viral, uh, uh, maybe Instagram reel or something about a guy who says that whenever he checks into a hotel, he says, you know, uh, we always tip the valet when we're here. We tip the bellman and the person that cleans, but you know, you folks at the front desk just don't get enough love. And so here's a nice crisp $100 bill. And he says.

• 4:52Jordi Hays?

  The front desk folks really.

• 4:54John Coogan?

  They never get tipped. And he said every time he does it, he gets upgraded to an insane suite. And so he was just like sharing this alpha, I think.

• 5:01Jordi Hays?

  No, the hotel, the hotel I worked at, Michael Jordan would stay and he just actually would carry around like 10 grand. And just any, he was just handing, handing it out like candy on the property. And he would have a very nice stay, as you might imagine.

• 5:16John Coogan?

  Yeah.

• 5:17Jordi Hays?

  Um, so anyways, we were talking about that. Travis responded and you can get into it and kind of give your reaction.

• 5:22John Coogan?

  Yeah, so Travis said, uh, delivery app tipping isn't about feedback, feedback mechanisms. It's a tool for maximizing the price paid by consumers. Eaters are economically irrational with tip. For every $1 in tip, they economically behave as if it were 80 cents.

• 5:41John Coogan?

  This is just a hypothetical figure, but it's directionally true. Because you feel emotionally good about tipping, mentally it, you, you give it less. It feels less painful to part with those dollars.

• 5:54Jordi Hays?

  Another purchase.

• 5:55John Coogan?

  Yeah, yeah, yeah.

• 5:56Jordi Hays?

  Buying gas.

• 5:56John Coogan?

  Exactly. So if you, the way you look at, if you, if there's $10 in taxes and $10 in tip, you'll be like, oh, I feel good about the $10 in, in tip. That feels like $8 and the taxes, that feels bad, right? Um, and, and, and it happens on the other side. This means that, uh, less price elasticity for the same price. So couriers are also economically irrational with tip. For every $1 in tip, they economically behave as if it were $1.20. Again, directional. And so you feel good when you're tipped and so you treat those dollars as more valuable. And so this is a hack on the human psyche, which apps must implement and maximize,

• 6:37John Coogan?

  or miss out on economic surplus that their competitor will use to defeat them. And so even if you have, your whole brand is built around, "Our app doesn't tip." Remember this happened with Uber. If your competitor is using tips, if they implement tips, they will just be making more money than you because of this economic inefficiency that arises from the nature of the human psyche. That is very, very interesting. The app that decides to pay the same net amount to the courier, but as a square deal via a drop fee plus tip, will lose market share every day to an equal marketplace player that implements and maximizes tip. Now, equal marketplace player, that's doing a lot of lifting because it's hard to just spin up.

• 7:21John Coogan?

  Like, you know, I can't just start an Uber competitor right now. It's hard.

• 7:26Jordi Hays?

  Yeah.

• 7:27John Coogan?

  But he makes a very good point here. And so what's interesting is that I read this as adding tipping is inevitable. Adding tipping is inevitable. We're not doing it right now, but eventually someone will come to the market, do it, we will have to in order to compete. Is that not the read here?

• 7:48Jordi Hays?

  So the difference here is that I think that one picnic is already counter positioned, right? So it's pricing thing, it's a flexibility standpoint. It's also counter positioning on like focusing on one key buyer. Obviously, you know, the DoorDash's, the Uber Eats have their kind of like corporate offerings. But I think like just creating a different and more transparent model makes a lot of sense. He's also like, I think you have to factor in, there's a lot like TK's been kind of like secretive about cloud kitchens, secretive about Otter, which is like the toast or square competitor that he has. And so.

• 8:32Jordi Hays?

  When I hear like no fees, no tips, like it just screams like there's been so many attempts at food, food delivery and just like new restaurant concepts that have been venture backed and a lot of them haven't worked out, right? Because it's just like becomes unsustainable. And so I think that, I think Travis is basically by focusing on a key customer type, trying to make it up with volume and then having this like vertical approach. I just like, I want to, I want to believe that, I believe that he's somewhat of a masochist and that like going and trying to win in food delivery is just like the hardest arena,

• 9:14Jordi Hays?

  just like food in general. We have David Chang coming on at noon, which I'm excited to talk with him about, but it's just like the most competitive space. It's low margin all the way down. But I think that he, I believe, just given the domain expertise, I believe that he's, he has a real play here and a real strategy. And I think that, I think that already we were talking with the person on our team that handles like food ordering. He got on the phone with Picnic yesterday and he was like, this offering is way better than what we're seeing with the, with the other delivery apps and wants to switch to it immediately. So again, if, if, if TK can make the model sustainable, I think it'll be quite competitive.

• 9:59John Coogan?

  Yeah. I mean, you would, you would imagine that vertical integration should allow true lower prices, like true cost competitiveness. That's like, you know, an age old business adage. If you vertically integrate, you can undercut your competitors and just offer lower prices. Almost like, you know, buying Kirkland brand at Costco is typically like, sort of like the canonical example of like heavy verticalization. And there's, there's a ton of other examples, but I wonder, I still wonder is, is this?

• 10:32John Coogan?

  Like, I remember in the early days of Uber. Like, it was amazing because you didn't need to think about the tip. And so that mental load wasn't there. And there was the star rating system and it felt like they're actually like the VCs might have been subsidizing it a little bit. But it felt affordable on the rider's side. And on the driver's side, it felt like people were getting paid pretty well and everyone was sort of happy. But maybe the VCs weren't, but they wound up getting, you know, a stake in a $200 billion company. So, you know, I think it all worked out for everyone involved. But it seems like Travis is reflecting on this idea that tipping was inevitable to come to the Uber ecosystem. Is tipping going to come to the Waymo ecosystem?

• 11:16John Coogan?

  Is tipping going to come to this picnic ecosystem, the picnic product eventually?

• 11:21Jordi Hays?

  I don't know.

• 11:22John Coogan?

  Do you think picnic will have tipping in 10 years?

• 11:25Jordi Hays?

  I just view this more as a corporate service in its current positioning.

• 11:30John Coogan?

  True.

• 11:31Jordi Hays?

  Than a consumer service. And when a consumer is buying food.

• 11:35John Coogan?

  Yes.

• 11:35Jordi Hays?

  It is, if you're ordering food delivery. It is, it is not a, it is like, it is a luxury, right? Like, food delivery has been extremely normalized. But if you, you know, rewind to 40 years ago and ask like, oh, how often do you get food delivery? Most people will be like, I never get food delivery. I just go pick it up myself, right? So, it is a luxury, but this is being positioned like as a corporate offering. And I think that if picnic can get just like deep relationships with a bunch of these different companies that have, you know, I listed off some of the logos before. If they can, if they can just become embedded in these companies and part of their workflows, I think they'll, it's possible to like make it up in volume.

• 12:19John Coogan?

  Yeah. This is so funny, the way Sheil puts it. "Truth bomb" from TK. "Tipping is a hack to maximize price. It's psychology. Consumers are willing to pay more in tips than they are willing to pay in fees or menu price. So a $16 burrito plus a $4 tip feels far cheaper to people than a $20 burrito that has a no-tip option."

• 12:40Jordi Hays?

  But again, from a, from a, from a, from a business standpoint? I don't, I don't know, I don't know if it's exactly the same thing. I feel like businesses like wanna have like more predictable, more predictable costs, not have that like variability and like, okay, sometimes the fees are like this, sometimes the fees are like that. I think this will be a better consumer experience. A lot of companies, you know, will give like credits to their employees, which is like you get $20 of credits.

• 13:07John Coogan?

  Yeah.

• 13:08Jordi Hays?

  Every day and then whatever you're kind of like spending on top of that, you have to eat.

• 13:12John Coogan?

  Yeah.

• 13:12Jordi Hays?

  And so I think consumers will could very likely like picnic more, so we'll see.

• 13:18John Coogan?

  Yeah, the, uh, uh, this was one of the original like D to C evolutions that happened with a lot of like Shopify merchants. I remember looking at, I think it was like Kylie Cosmetics. And, uh, there was a trend for a while that was like, uh, consumers want transparent pricing. Uh, don't do all the crazy psychological hacks. So you'd be like, yeah, I'm just gonna put, it's 30 bucks and that's what it is. And it has free shipping and that tax and shipping is included. And it's just like what we say upfront is feels really good, feels really good to say that. And then you go to like the high performing stores and all across the board. It would be like $9.99 and then you go in and there's like $6.42 added in taxes and then you add shipping and it each gets

• 14:02Jordi Hays?

  It's like a pop up that says you have one minute to add this to your cart.

• 14:03John Coogan?

  And it's laddering you up, exactly. And it's laddering you up and just like keeping you on the reel, reeling you in like a fish, adding, adding fees, adding fees until you're like, okay, well now I'm like entered all my information and I'm ready to click the button. And so, yeah, okay, you added two more bucks, whatever, I'll just deal with it. Uh, so these psychological hacks are just like somewhat inevitable. Um, but, uh, avoiding them, I think in the short term is a great go to market. I just wonder if there's something, um, if there's something like truly, uh, like counter positioned that will be durable, like Costco, Kirkland, Costco, like has not, has been like the low cost affordable option and that model has held for like

• 14:41Jordi Hays?

  Yeah.

• 14:42John Coogan?

  decades, right?

• 14:43Jordi Hays?

  The one thing, the one thing that we learned from, uh, having somebody on the team call picnic is that they are focused on higher volume orders. So like teams of like 25 and up.

• 14:53John Coogan?

  Yeah.

• 14:53Jordi Hays?

  And so I think that, I think that they're just betting like, hey, there's, we can, we can get a lot of volume. We we will be able to like handle having it's one person that goes around and picks up every order from all the different restaurants, right?

• 15:06John Coogan?

  Yeah.

• 15:06Jordi Hays?

  And so it's quite a bit less of one individual's time, much higher order volume than when a company is like, hey, we're giving credits to people. And then each employee is making individual orders. And then there's like, ends up being like 20 drivers on the road.

• 15:21John Coogan?

  Yeah.

• 15:21Jordi Hays?

  To deliver one lunch, which like makes no sense.

• 15:23John Coogan?

  Uh, I have one more take on this delivery question, but first I need to tell you about cognition, the makers of Devon. Devon is the AI software engineer, crush your backlog with your personal AI engineering team. My question is, what is TK's drone strategy? What's his autonomous delivery strategy? Because he's vertically integrated at the kitchen level. He has the point of sale system. He has the the sort of ordering front end.

• 15:52Jordi Hays?

  Yeah.

• 15:52John Coogan?

  You can interact directly with him. He's going, he's cutting out several of the middlemen. But is he going to be a logical partner for ZipLine? Is he going to be a logical partner for Coco and Starship and these robotics companies that are delivering food? Ryan Oskinhorn here says, people aren't ready for how much better food tastes when it arrives 5X faster. That's a hilarious take because like, in fact, I have tasted food right when it's made. Like, it's not, it's not like an entirely novel thing.

• 16:24Jordi Hays?

  Yeah.

• 16:25John Coogan?

  But what he's pointing at here is that ZipLine is is getting food delivered in four minutes as opposed to cars that take 20 minutes. And so hot food arrives hot, which is certainly a benefit, but it just does create more of like a, you know, benchmark to the to the actual restaurant restaurants.

• 16:45Jordi Hays?

  I tried, I'm trying to find if Travis is an investor in ZipLine. The Google AI overview says. Yes, Travis Kalanick is an investor at Uber. Gemini says, I could not find definitive evidence that Travis.

• 16:58John Coogan?

  Of supply?

• 16:59Jordi Hays?

  Yeah. So the AI overview says yes.

• 17:02John Coogan?

  We will do.

• 17:03Jordi Hays?

  Gemini says there's no evidence.

• 17:04John Coogan?

  Maybe we can ask him. I wonder if that's like a logical partner. I mean, on the self-driving side, his original vision at Uber, it felt very much like he needed to own that technology. He wanted to be not just a buyer of it from a different company. It seemed like while he was at Uber, he considered self-driving technology as critical path, as something that should be owned by Uber. And then, once he was out, the company spun down ATG, their advanced autonomy group.

• 17:34Jordi Hays?

  I mean, but think about it. So, I mean, right now, if you look at city storage systems, you have Cloud Kitchens, which is making the food. You have Otter, which is like the payments and ordering infrastructure. And then now, you have Picnic, which is like the front end. And any type of delivery method actually like fits into that system, right? So, I think he's being, I would imagine, he'll either add a strategy, but potentially more likely, he'll just integrate with a variety of drone delivery, and then autonomous vehicle delivery, and continue to use traditional labor. So, we'll see.

• 18:11John Coogan?

  Well, let me tell you about Gemini 3 Pro. You've probably heard about it, but we're telling you about it anyway. Google's most intelligent model yet, with state-of-the-art reasoning, next-level vibing coding, and deep multimodal understanding. I took it for a spin in AI Studio this morning, had it build a scrollable, like, you know, as you scroll, the bubbles move around and tries to visualize how DeepMind and Google Brain merged, and these sort of like generative UI around deep research reports, I think are going to be really, really fun. I need to continue iterating on this particular one.

• 18:47Jordi Hays?

  Gabe says, how can it be a logical partner if his products is for larger teams? Like Jordy just said, teams of 25 plus don't think a zip line can fit 25 different orders in their drones.

• 18:56John Coogan?

  That's a great point.

• 18:57Jordi Hays?

  It's a good point. Keller said they can fit two full grocery bags worth of food in their drones.

• 19:02John Coogan?

  That's still 25.

• 19:03Jordi Hays?

  Yeah, but... I like what it'll come down to is the actual cost.

• 19:08John Coogan?

  Your team does a group order and instantly gets swarmed by drones. Yeah, that would happen.

• 19:11Jordi Hays?

  I mean, yeah, I mean, right now you order on picnic, one delivery driver is like driving around to a bunch of different restaurants and getting all that food and bring it to the office. You can imagine like six different drones end up carrying out.

• 19:23John Coogan?

  No, I mean, I think, I think for, for this particular, uh, for, for this particular, uh, like use, it just feels like, uh, they will be much, much more a buyer of like a Waymo type autonomy solution. As opposed to a zip line autonomy solution. I would assume.

• 19:41Jordi Hays?

  Yeah. I just think, I think the most notable thing about picnic is Travis doesn't want to just sit at the infrastructure layer of food delivery, right? He wants to own the end customer experience and the brand. So.

• 19:52John Coogan?

  Yep. Uh, well, uh, we have a beautiful picture of Alex Karp's watch, the Patek Philippe Aquanaut with the orange band. We love to see it.

• 20:03Jordi Hays?

  We clock this.

• 20:04John Coogan?

  TJ, the wheel.

• 20:05Jordi Hays?

  Long time ago.

• 20:06John Coogan?

  Jensen, one of Jensen's leather jackets. Jensen, of course, has many leather jackets. Uh, I've only seen Karp in one single Aquanaut. There is a fascinating story of how Karp wound up with this particular watch. We'll have to get him to tell it on the show though, uh, at some point. We'll also have to tell you about cognition, the makers of Devon, which I already did. I already did cognition. I'm out of it today. Adio. Adios, the end of CRM.

• 20:37John Coogan?

  Build scales and grows your company to the next level.

• 20:41Jordi Hays?

  We, our routine, our routine is so dialed in that if we go to bed like two hours later than normal, it just throws everything off. We had a very chaotic morning.

• 20:51John Coogan?

  Yeah.

• 20:52Jordi Hays?

  But we're back.

• 20:53John Coogan?

  We're back. Uh, Nano Banana is remarkable. Look at this Golden Gate Bridge image. It generates the image and also all of the diagrams around it.

• 21:02Jordi Hays?

  This is how Tyler sees the world, by the way.

• 21:05John Coogan?

  Sundar says, "You went bananas for Nano Banana, now meet Nano Banana Pro." It's state of the art for image generation, editing with more advanced world knowledge, text rendering, precision, plus controls, built on Gemini 3. It's really good at complex infographics, which is awesome, much like how engineers see the world. That's very funny.

• 21:23Jordi Hays?

  I even see that. We were playing around with it this morning. It is absolutely wild.

• 21:28John Coogan?

  It's really, really good. The text is flawless. There's just truly, it doesn't make any mistakes with text anymore. You know, we were in the era of like, the text looked good, but you would still see a double S every once in a while. One thing would go wrong. And now we're in a much better spot. There's still one test that it fails. That's the Where's Waldo test. If you have it go generate a Where's Waldo, it will not be, it will, like, you will clearly be able to tell the difference between a real Where's Waldo-

• 22:01Jordi Hays?

  You showed me this morning. Did you generate that?

• 22:02John Coogan?

  I had Tyler generate that one. I generated another one.

• 22:05Jordi Hays?

  This is like the most funny Where's Waldo ever. John like says, like, "Where's Waldo?" And I'm like, "Are you messing with me? Is this a joke?" And it's like, this massive crowd of people, and then Waldo is standing on a stage going like this.

• 22:21John Coogan?

  Yeah. For some reason, it did not hide the Waldo at all. Like Waldo was just perfectly in the center, very obviously.

• 22:27Jordi Hays?

  Most novice Where's Waldo.

• 22:28John Coogan?

  It was very novice. And then also, there were actually two Waldos. And as you dig in, normally, when you're hunting around a Where's Waldo, there's different little sub-stories that are happening. And this was more just like a generic crowd. I mean, still remarkably impressive. But that is currently my go-to evaluation. And we got a lot closer, but it's not, it's not superhuman. It's not super Waldo yet. Anyway, we have Doug Olafman from Semi Analysis in the Reacher Waiting Room. Let's bring him into the TVPN Ultra Dome. Doug, how are you doing? Welcome to the show.

• 23:02Jordi Hays?

  How did you decide to take a vacation in the fall of 2025? You were off for two weeks?

• 23:07John Coogan?

  Yeah, no days off. We're in the midst of the biggest bubble of all time.

• 23:11Jordi Hays?

  It's gonna be mellow. There's not gonna be any news.

• 23:14John Coogan?

  No days off.

• 23:14Speaker 02

  So dude, every single time I take a vacation, stocks always drop. But dude, I did, I proposed to my girlfriend in Japan. That's the reason why I went.

• 23:22Jordi Hays?

  No way.

• 23:23Speaker 02

  Yeah, yeah, well, big deal. Yeah. Whoa, we're gonna hit the gong.

• 23:28John Coogan?

  Woo! That's massive news. Anybody can pull together a $200 billion LOI, but to find true love is beyond special. So congratulations.

• 23:38Speaker 02

  Dude, what's a 100 billion between friends today? I saw the 100 billion, the 100 billion Brookfield thing. I was like, dude, don't even, it's just another day, man. Another day, and why?

• 23:47Jordi Hays?

  Did you, did you have somebody, did like a semi-analysis intern come up during your vacation and go on your ear like, "Sir, Sarah Fryer has requested a federal backstop."

• 23:59Speaker 02

  So, okay, I mostly consumed it on a 12-hour lag, and the 12-hour lag, I was like, holy shit. And it's just like very funny to get it in like slow motion, where I'm like, okay, that was a bad, that was a bad interview, Sam. I'm gonna be honest with you. I was like, oh, Sarah Breyer? And then I'm just like, ooh, ooh. And then I'm like, but I do think, and to the Fed, the Fed is what I think people are freaking out on the stocks-wise. But it's just like this weird thing to witness in like slow motion, half on the other side of the world when everyone's asleep and shit. It was just really weird. Yeah.

• 24:29Jordi Hays?

  Totally. Well, welcome back.

• 24:30John Coogan?

  Well, yeah, what's going on with Nvidia? Take us through how you're processing the news. We've been batting around two takes. One was, we're extra analytical over here. The first take we had was Jensen was seen drinking a beer, and therefore he will beat our earnings.

• 24:49Jordi Hays?

  Not just drinking a beer.

• 24:50John Coogan?

  Chugging.

• 24:50Jordi Hays?

  He was linking arms in South Korea.

• 24:53John Coogan?

  This is, this is our rigor. Our rigor.

• 24:55Jordi Hays?

  And so I was confident that they were gonna do quite well.

• 24:59John Coogan?

  And then we just seem to be in the era where things beat on earnings and then immediately sell off for some reason, because expectations are so high. And maybe we're in that era now, but how are you processing it?

• 25:09Speaker 02

  I think it's like almost a perfect beat. It's very clean. You have like almost nothing to complain about. Margins, which was like a story last year, doesn't matter. Like they did a great job. They had a pretty solid, like a meaningfully above buy side consensus. It's, it's like a perfect quarter. You have no problems with it. Um, but the thing is you're the biggest, most profitable company, or not most profitable, but like you're one of the biggest companies of all time.

• 25:32John Coogan?

  Yeah.

• 25:33Speaker 02

  Uh, perfection is expected every single time you report. So I think it's totally fine, dude. I'm, I'm being serious. It's just totally fine. Like stocks do go down.

• 25:39John Coogan?

  Yeah.

• 25:40Speaker 02

  People forgot about that.

• 25:41John Coogan?

  People forgot.

• 25:42Speaker 02

  They do go.

• 25:42John Coogan?

  Stocks can go down.

• 25:43Speaker 02

  They do go down. They go down sometimes, man. It's crazy.

• 25:48John Coogan?

  What is the interpretation of, or what should the read be on Gemini three, the TPU? It, it feels like that's like, uh, um, you know, is Nvidia still a monopoly? If you can train the best model on all the benchmarks without a single Nvidia chip? That seems like maybe a crack in the narrative, but does it matter at all? Or is it, uh, just, uh, irrelevant?

• 26:12Speaker 02

  Um, I think it matters a little bit. I think Google being really aggressive is really nice. Cause like, do they have power and they're like waking up and TP V7 is gonna be awesome. And Anthropic and we're doing it. We're doing like actual deals. So like, that's good shit, I think. Uh, because you need like, like Gemini has been like, I don't know, asleep at the wheel despite inventing all this stuff. And so it's really nice for, to see them be back, but I don't think it's, um, I mean, I think it's a big deal. Clearly TPU is number two and it deserves number two. I think Nvidia being number one. What I really want to see is like, why isn't there a new pre-training run from OpenAI? Like I gotta, I gotta ask that question out loud again. We've seen so many, um, RL scaled up versions,

• 26:53Speaker 02

  but we know there's no new base pre-train and we know that there's Gemini three cooks because, uh, it's a new base pre-trained model. So like, where is that happening? Like, is it because the GB 200s aren't stable enough? Is it because like they're just totally not like dialed in? I think that that's like the question to be answered and, um, OpenAI is just, I don't know, they're not cooking. I want to see them cook. So right now, Gemini three.

• 27:17John Coogan?

  The. On the on the base pre-train, is it still is it still fair to like kind of set up that storyline with GPT now called 4.5 now sunset used to be GPT 5 potentially didn't really pan out. There's been a lot of debate over what went wrong with that pre-train. Is it fair to say that that was like an order of magnitude more compute spend like cost went into it? Is there anything real about like it was expensive to serve it? I've heard that bandied about as like why a lot of people said it actually was a better pre-train. It was a better model. It would have been amazing, but we just messed up something about the economics.

• 27:59John Coogan?

  And so, once we tried to deploy it, it wasn't very economical. And so it was slow and that's why we pulled back. Not that it wasn't a good pre-train, not that it wasn't a good model.

• 28:09Speaker 02

  I still think it's a failed run, dude.

• 28:10John Coogan?

  Okay.

• 28:11Speaker 02

  I still think it's a failed run. I don't think it got quite to where it should have been, given its size and something was wrong with that. 4.5 was decent and like a really good creative writer. You talk on the economic side, this is where I have to to to pump inference max.

• 28:26John Coogan?

  Yep.

• 28:27Speaker 02

  Right, which got shouted out three times. So yeah, look, man, I think the economics work now or could work with GB 200 because it's like, you know, 10x better performance. So you can, in theory, you probably could serve it, but for some reason they still don't want to. And it's probably something on the RL like it's just a bad base model that is not able to scale with higher chain of thought or like maybe because of how much compute it takes. It's, you know, the still distilled version wasn't doing as well. All this stuff matters and for whatever reason, 4.5 isn't it. We know there's been, we know there's been failed training runs. And so it's like, dude, OpenAI, I want to see it. And I think we'll, I think we'll get it, right? Like, nothing makes them excited, but like competition.

• 29:10John Coogan?

  Of course.

• 29:11Jordi Hays?

  Well, will there? So, so OpenAI has a consumer business. They have a front end for AI. It's the brand that people think of when they think of AI. At some point, you could imagine them not doing another scaled pre-training run because they're they're just like, you know, it's not really worth it to take it from, you know, this IQ to this IQ. It's like, our average user is just not really going to care. Meanwhile, if you have a company like Anthropic, which is like an API business that like relies on kind of like raw horsepower capability intelligence and maybe is like easier for end platforms to switch in and out of, like, I don't know that they can afford to not keep doing the bigger and bigger training run. But do you expect OpenAI

• 29:56Jordi Hays?

  to at some point just say like, yeah, we're kind of good on the core product. Maybe we don't even need to do the next run?

• 30:04Speaker 02

  Dude, I think I think at the same time you say all that, if they're being a consumer business, but you know, OpenAI has massive like, they have coding FOMO, man. They're really, really, really concerned about the coding models, right? The I'm sure you saw like the, I can't remember what it's called, but we'll say like Project 2027 or no, no, not Project, but it's like the 2027 scenario or something like that.

• 30:25John Coogan?

  Project 2025 was the right way agenda.

• 30:31John Coogan?

  You're talking about AI 2027, the fast takeoff scenario, this scenario where OpenAI buys Ford Motor Company to make to make humanoid robots.

• 30:42Speaker 02

  Yeah, to make more widgets, bro.

• 30:43John Coogan?

  Yeah.

• 30:44Speaker 02

  No, no. So I think I think that while like, hey, look, it doesn't seem like it's on track anymore.

• 30:49John Coogan?

  Sure.

• 30:49Speaker 02

  I do think the thing about the fast takeoff that people feel very strongly about is better coding models means better AI agent, you know, AI agents and those AI research agents will make better models. And that's there is a recursive loop there. I think that that's where, dude, that's where like, you know, they were they had so much Codex Fobo. And and despite all this, man, like Gemini still doesn't have the Anthropic lead today. So, yeah, yeah. I mean, I think everyone wants that sweet bench. And yeah, I don't know. I just think. It's it's just like this weird. I I think it's just like a perfect vibes time on the like the finance side, dude. People are freaking out about the market, uh, Fed cut. Uh, it's not gonna

• 31:31Speaker 02

  happen. And so it's like, uh, we, I didn't, I learned this yesterday, but this is like the second longest run above the 50 DMA, which is like, you know, stock chart, men, males astrology vibe. Um, the, the, the second longest run since like 1997. And so it's just like we've, we've been, we, stocks have been going up for quite some time and sometimes they can go down or even sideways. And so I think people are freaking the fuck out and it's kind of a long, long, long, powerful run. And, and, and also I think that people are freaking out because like stocks go down, people's vibes get bad. And then they're like, bro, maybe it's actually over. Maybe it's actually over. Like, nothing changes sentiment like price, man.

• 32:11John Coogan?

  Totally.

• 32:12Jordi Hays?

  Uh, what did you think about the Financial Times published an article that, uh, was pretty, I felt pretty misleading. They said Oracle is already underwater on its astonishing 300 billion OpenAI deal. And they said that because the stock.

• 32:27John Coogan?

  This was AlphaVille. It's their blog. They're having fun, but they're rage baiting. They're rage baiting.

• 32:32Jordi Hays?

  They rage baited.

• 32:33John Coogan?

  What do you, what do you think?

• 32:34Speaker 02

  They, they rage rate pretty hard, bro. Like, let's be, let's be clear. Like, I don't think, uh, I don't think, you know, making $400 billion of revenue is, uh, being underwater. But I, if you, if you're, if you're betting on just a stock, then sure, yes, they are underwater. I think they're gonna be okay.

• 32:49John Coogan?

  Their headline today from AlphaVille, it was like, it was like.

• 32:52Jordi Hays?

  Who is OpenAI's auditor?

• 32:54John Coogan?

  They're really taking shots at everybody. Um, it's funny. Um, I do, I do, well, yeah, sorry.

• 33:00Speaker 02

  I, I was gonna say, man, but AlphaVille, like, I don't know, AlphaVille doesn't cook, man. Their alpha is so, is, it's kind of, it's kind of mid. I don't know what to tell you, man.

• 33:08Jordi Hays?

  Petition to rename AlphaVille Midville.

• 33:10Speaker 02

  To Beta, to Beta.

• 33:12John Coogan?

  BetaVille.

• 33:13Speaker 02

  Beta buff, Beta boys.

• 33:15John Coogan?

  That's rude. I, I've enjoyed AlphaVille from time to time. I, I want to know about, uh, this Gemini three pre-training run. Is there any way for us to understand the rough order of magnitude of compute or dollars that went into it? I, I, from what I understand, Google has more of a distributed. Training system, they train across data centers, that might be right. And so before with like the GPT three training run, the GPT four training run, it was like they raise a bunch of money, they go build a data center, they acquire a whole bunch of GPUs. And then you kind of see like there was this much energy that went into it, this many GPUs were marshaled for it. But it feels like with these Google training runs, they're harder to understand the actual

• 33:58John Coogan?

  scale of the investment. But do you have a, do you have a more solid understanding of how, how big the Gemini three project was like from a capex perspective?

• 34:09Speaker 02

  I have no, I mean, I can't tell you because I don't know how big the model is like on a parameters basis, but I do, I have a pretty good vibe that it is multi data center. They were first to do that. Pathways has always been first in terms of like the OCS, the distributed scale out network, like they've always are sorry, scale across, like they've always been first in that. Yeah, I don't know. I don't have an actual number, but I don't think the actual pre-training of of like the final run probably wasn't that much money. Right. But the thing is all the experiments to get up to there, all the other things that that goes into training a really big model costs a lot in R and D. And so I think the thing, the final shot or whatever in terms of compute is probably paltry compared to like the actual total spent, right? You probably have

• 34:54Speaker 02

  a multiplier of like 10x on top of it of what the final number is. But I don't know. It's probably, it's probably a billion bucks if I had to guess. There, yeah, I'm just gonna throw out a number. You heard it here first. Billion.

• 35:08John Coogan?

  Because I mean, we've heard about training runs that were, you know, like a couple of years ago, they were in the hundred million range and that the billion dollar training run was kind of rumored. I wonder, I mean, I wonder if they, if we get another 10x next year or the year after, and we're seeing $10 billion flow into a single training run, like from an SEC perspective, does that need to be disclosed at some point? Does this wind up going into the filings into earnings kind of just as an individual?

• 35:35Speaker 02

  Opex number?

• 35:36John Coogan?

  Yeah.

• 35:37Speaker 02

  That'd be pretty sick, honestly.

• 35:38John Coogan?

  Yeah, I would just imagine that at a certain point, investors would want to know. I mean, it's like a mega acquisition. It's a, it's a significant slug of.

• 35:49Speaker 02

  I guess it goes into cost of goods sold like AI accounting is like completely made up today. So no, who the hell knows, but it's probably a cost.

• 35:56John Coogan?

  I feel like it should be CapEx. I liked, and I don't know what your take is, but I liked Dario's framing of each model is individually a profitable company. When you you spend a billion dollars and then you make a hundred million a month for, you know, a whole bunch of time, hopefully.

• 36:12Speaker 02

  But okay, in order for it to be CapEx, like to be an accounting brain is it has to have a multi-year lifetime. And so if you if you train a model every year, it's R and D. So that's that's the issue, right? You can't capitalize it. So I don't know, man.

• 36:24John Coogan?

  Well, I mean, I have a big question about this. This is something that I've been I've been going back and forth with with is like, we have seen that there is demand for 4-0 chatbots from like a group of editors, potentially forever, because those those people are like, 4-0 is my friend. I don't care about Gemini 3. I don't care about GPT-5. I don't care about 035 thinking Max deep reasoning. I want 4-0 and I'm willing to pay for 4-0, maybe forever. We don't know. Maybe the churn rate will be very low for a long time. And so you wind up with this weird thing where you can actually amortize 4-0 over years with that cohort. Now, we don't know how big that cohort is and what the churn will be, but it could be 10 million people for 50 years.

• 37:07John Coogan?

  It's just like their buddy. And and I'm wondering if the same thing will happen in businesses where you have some company that's like, we have a model that is 4-0 level intelligence or Gemini 2.5, and we have no reason to update to Gemini 3 because this this model just sits there and it looks at papers, scans them, summarizes them, and it does that a million times a day, and we're happy with that. And we don't need it. We don't need it to be more intelligent ever. So we're just going to keep that workload going in perpetuity. And we'll leave it on A100s if we need to. Like, we don't need to go to the the latest and greatest. Do you think that's going to happen in the enterprise?

• 37:43Speaker 02

  I mean, I don't know if it'll happen. I mean, enterprise, just like, let's get like more enterprise bullshit-y. People have to have price raises and you have to be like, well, why did you raise my price? And they, the, the single best way to do this is say, uh, we spent more compute, we have a better model, we do something like that. But I also want to say like in the consumer side, something that's like a good example is like, dude, RuneScape Classic is probably like a perfect case study of this. People want to play RuneScape Classic. They don't give a shit about like the, and like, obviously RuneScape Classic is like, kind of become a fork universe and there's like a lot of other stuff, but it's run by like 10 people, bro. And there's like millions of, probably like, you know, tens of thousands, hundreds of thousands of people who play it. And like, yeah, I, I wouldn't be surprised that we see like these long-lived little projects that are really stable.

• 38:27Speaker 02

  And they're like, dude, no notes, do not change it. I don't care. I want to play this one. I want to use this model forever. That's probably like a really good example of like, uh, but I feel like that's like a niche. And it's very hard for you to underwrite everything, becoming these like weird cohorts. That's like a massive fragmentation, the internet and everything, like everyone just has their one little like, you know, you know, freeze all my, my memory at this exact point, no new information, you're my favorite version of Gemini 3.5 or whatever. I don't know. I, I think that that's, it's kind of hard for us to be like, like, and also dude, that, that's not AGI. Like I, if you were talking about like vibes, that's extremely depressing. Like if we're talking like last year to now, that's like so depressing. Like that's why I think this is like, I think this is right, why markets are sad. People are sad. They're like,

• 39:12Speaker 02

  dude, you tell me 40 is all I want, then what do we buy? Why are we spending 100 gigawatts? So I think, I think we're just in a weird time, dude. It's in, it's, you know, the market didn't go down at all. And now the market's going down and everyone's getting sad.

• 39:26Jordi Hays?

  But, but the, the, the leverage is still coming in. The, the circular deals haven't actually hit the books yet. They've just been announced. Like that's.

• 39:38John Coogan?

  Didn't, didn't Jordan from Semi Analysis say that, uh, uh, there was potentially going to be like an H100 index that retail investors could like buy into or something like that? I'm just interested in like, how many different pools of money haven't like come online to the AI trade yet? Obviously, private credit is coming online now. There's obviously corporate debt. There's just sucking down all the big tech earnings. There's also potentially like retail traders getting in on the action one way or another. If some of these foundation model companies go out and go public.

• 40:11Speaker 02

  So the last note, I had my team who is sitting in the office right next to me and behind me.

• 40:16Jordi Hays?

  You're in an office right now. It looks like you're in a forest.

• 40:19John Coogan?

  Yeah, I love the wallpaper.

• 40:20Speaker 02

  We are in a forest. Thank you. To be clear, dude, we are squatting. Thank you to our squatting overlords who let us work here. We're sick. We're super happy about that. But okay, if you just do the math, man, because here's the thing. I think all the hyperscalers could raise like $2 trillion. Like, I really think the number is so large. In fact, I'm trying, I'm trying to-

• 40:43John Coogan?

  I was just looking at the free cash flow and then you multiply it by 10. If they were paying 10% interest, like 10% interest, and it's trillions of dollars because they produce so much, hundreds of billions of dollars, no?

• 40:55Speaker 02

  Okay, so I'm gonna, I'm gonna, I'm gonna give you the maxed out version of how I think about what we could do. So, uh-

• 41:03John Coogan?

  Leverage Max, the new, the new report from Southern Analysis, Leverage Max.

• 41:07Speaker 02

  Also, also, I've been told by, I've been told by my corporate overlords, you have to star the inference max. That's super important. I, you have to star inference max on GitHub. Sorry, before. Wait, so how?

• 41:17John Coogan?

  Everyone, go star inference max on GitHub, please.

• 41:20Speaker 02

  Thank you.

• 41:20John Coogan?

  We'll make a call to action. Cool.

• 41:22Speaker 02

  That'll help. So I think, I think they could probably raise something like $6 trillion by 2029.

• 41:30John Coogan?

  And, and is that like a 5% interest rate you're assuming on like corporate debt, basically? And that's what they're paying?

• 41:36Speaker 02

  Um, yeah, so we just essentially, we're, we're doing the current corporate interest. We're just saying like, hey, the current market rate. There are actual problems with how this is done, but like, let's use Meta. Meta is the most, like, the most aggressive version of this. You completely do all your data center CapEx off the balance sheet. You have Blueout come in, pay for all of that. You do a sale leaseback and then you spend the rest of the money just buying GPUs. And you could probably do like, you could, and then they can issue debt in the market that's like 50 bips above the government. And also the rating agencies are like, okay, as long as you don't have more than one turn of debt by 2029,

• 42:13Speaker 02

  you're good to go. So, so we did that number for all, all of them. For all the hyperscalers, ex Oracle, Oracle's pretty tapped out, $6 trillion. That's like the, that's the, the big number.

• 42:23John Coogan?

  That's great.

• 42:25Jordi Hays?

  So, yeah, we're somewhere between-

• 42:25Jordi Hays?

  But here's the thing. So if Oracle's tapped out already and they're about to spend four years where free cash flow is going to be negative, how does that actually work?

• 42:36Speaker 02

  So here's, here's the thing about this, though, is free cash flow doesn't like, free cash flow goes negative if you assume there's no revenue growth. But um, this is like, kind of like a shale well, okay? You get a lot of your money on a GPU cluster upfront. Uh, let's say five year economic life, people are gonna fight me about this, but whatever, you, you will have your payback for a brand new cluster in something like eighteen months. And so after that, on the, like, let's say on the eighteen to twenty-four to thirty-six months, which is like the two to three year, you're just gonna start, now you're gonna start to gather in cash. And that cash you can go turn around and borrow more against or re-spend again. And so that's where this, like, you know, the shale, one of the reasons why shale went so insane in terms of supply is like,

• 43:20Speaker 02

  twelve month payback period, which is way more insane than what this is, but like, um, twelve month pay, so you get all your money back, you can just do it again. Do it again, do it again. So I think next year Oracle will make a lot more money and they're gonna be able to raise against a lot more money.

• 43:33Jordi Hays?

  There we go.

• 43:33Speaker 02

  Yeah, but they're, they're, they're tapped out this year.

• 43:36Jordi Hays?

  Yeah, yeah, yeah, that makes sense. Uh, okay, I've kind of a lightning round, uh, cause I know you have a hard out in a few minutes. What is going on with CoreWeave and Core Scientific? Like, uh, CoreWeave is, is reliant on Core Scientific. They, they've had, it seems like some kind of frustrations around getting capability delivered from Core Scientific. They tried to buy Core Scientific, Core Scientific rejected it, Core Scientific has now traded down twenty percent or so since the acquisition uh, was attempted, uh, but can you explain that dynamic and why the Core Scientific shareholders are kind of still holding out at this point?

• 44:18Speaker 02

  Okay, so Cores uh, got offered to be bought in all equity and this is before all this, all the other Bitcoin miner, like, energy names ripped. And then I think a firm called Two Cs wrote this thing, be like, hey, look at everyone else's results and how much they've ripped, and you're telling me you're selling out at this price? And so rightfully, if you do the math, you're like, maybe we should just not get sold, or we should uh, deal break and we should ask for a higher price. And so most of the investors went for a deal break and asking for a higher price. But uh, you know, people who are, like, the stock does go down and you have a shareholder turnover when, when you. Rejective deal because like a lot of people are in it for the deal and then they have to sell. They're like, no more deal, I'm selling. And so that's like, that's pressure.

• 45:00Speaker 02

  But at the same time, Core Scientific is not delivering their Denton facility on time. And that delay is like, kind of a big deal for Core Wave. Yeah. So I mean, that's, that's like the, the Sparknose version of it. I think the problem is like, you look at iron or something like that and you're like, wait, wait. Why isn't Core is getting the iron multiple YOLO? This is a 2X and that's how the deal broke.

• 45:20Jordi Hays?

  Got it. That makes sense.

• 45:23Jordi Hays?

  Product idea for you guys, maybe it's something you're thinking about. Maybe it's something that doesn't make sense, but I was pitching John last week on an idea for something called a semi-analysis product called Diffusion Max, which would be like, how, what I want to understand is like, how is AI actually diffusing across a bunch of different key industries? So legal, accounting, you can just go on and on and marketing, on and on and on. Actually understanding like, I would, I would want you guys to have phone calls with like, thousands of business owners and employees in each of these different like categories. And then give us a read on, okay, are they actually laying people off because of AI? Are they hiring more people because of AI? What tools are they actually using? Are they getting a lot of leverage?

• 46:08Jordi Hays?

  Are they increasing earnings?

• 46:10John Coogan?

  It's affecting margins.

• 46:11Jordi Hays?

  Yeah. Is it affecting margins? And I don't feel like that, I, I don't know that there's like a definitive data source that I trust on that. And that's something that like-

• 46:19John Coogan?

  I only trust semi-analysis for everything.

• 46:21Speaker 02

  Thank you. Thank you. It's, we're the only source of truth, bro. Um, thank you. I thank you. I want to, I appreciate it.

• 46:27John Coogan?

  I think I pick quite maxy about semi-analysis.

• 46:31Speaker 02

  Same dude. So, so that sounds like I need an AI agent to call, you know, like a hundred thousand people. But dude, honestly, that kind of survey work is stuff that we're really interested in. But I don't think we're, we're, dude, we, dude, there isn't something we haven't thought about. Yeah, but we have a lot on our plate. It's like a throughput problem.

• 46:48John Coogan?

  Yeah.

• 46:48Speaker 02

  It's like another trading run.

• 46:50John Coogan?

  Super important too, yeah.

• 46:51Speaker 02

  Yeah. Yeah. Yeah. We're doing, we're really, really interested in the energy side. Like, we're going to do a grid by grid breakdown. Like, we're very focused on all the, I mean, we, we, like, we're putting a lot of effort in, and energy, energy into it. And I'm really excited about that, but even that is still probably a little bit far out. So each of these like bets take a little bit of time and you have to reinvest in them and give them time to work out. And so, yeah, man, I would love to do like, I don't know. Diffusion, but the problem is diffusion max is a bad name. You know, it sounds like diffusion, right? But like, I don't know, AI penetration max, who the hell knows?

• 47:24Jordi Hays?

  Don't, don't name it, don't, don't, don't name it that. Yeah. Another question, another question. XAI and Nvidia announced like a new data center project in Saudi yesterday. I don't know if you caught that. Do you see? Do you see XAI just getting into the cloud? Into the AI cloud business and, and helping power? And, and helping basically deliver infrastructure for other, other companies?

• 47:54Speaker 02

  I, you know, I know, but that I'm not until this conversation, but they're the best. They're the best at being quick. So if you don't, if you have like infinite capital and like the zoning laws can be whatever the hell they want them to be. I would sign up XAI to put up a cluster as fast as possible. They're the quickest with Colossus, like I think they literally have the speedrun record. And Saudis want it, dude. They want it so badly. And I think with this new, we're like allowing to export it. And I mean, yeah, if they can buy it, bro, and XAI is gonna be like, give me money in my pocket. I'm gonna make a Saudi Grok and then I'll make a Grok five. So I think that X, dude, X is down for business. And I think they like Tesla's always been supported by a lot of different funders.

• 48:37Speaker 02

  And I bet you some of the people who took X private probably were Saudis. So yeah, might as well.

• 48:44John Coogan?

  That's true, they, yeah, they actually were. I remember.

• 48:47Jordi Hays?

  Quick take on the, on the Brookfield deal. You mentioned it early in, in, after you joined.

• 48:53Speaker 02

  Mm hmm. Hundred, a hundred billion, but you should look at the actual numbers, five billion committed. So just, just.

• 49:01Jordi Hays?

  Not beating the press release economy allegations.

• 49:04Speaker 02

  Yeah, so, so, so, yeah, dude, this is the press release economy, man. I mean, I mean, you couldn't do to, you saw the 10Q, right?

• 49:11John Coogan?

  We gotta do a deal, we gotta do a press release. We'll pay you a hundred billion dollars if you pay us a hundred billion dollars.

• 49:18Speaker 02

  Dude, I think we can make that work with our accountants.

• 49:21John Coogan?

  I think we can do it.

• 49:21Speaker 02

  And then, yeah, we'll have a circular deal, bro.

• 49:24John Coogan?

  We'll write checks to each other, we'll hand them off right at the same moment, dramatically.

• 49:29Speaker 02

  Exactly, well, yes.

• 49:30John Coogan?

  Right there, right there, yeah.

• 49:32Speaker 02

  And then the economy will just circulate, bro.

• 49:34John Coogan?

  Exactly.

• 49:35Speaker 02

  And then we can use that money to raise capital.

• 49:37John Coogan?

  Yeah.

• 49:38Speaker 02

  It would be beautiful, holy shit.

• 49:40John Coogan?

  No, no, no, no, no, we don't want to, we don't want to take advantage of it and raise capital. We just want to, we just want to make fun of all the other. We just want the aura. Yeah, we want to aura farm everyone who's doing it unironically. That's what he does.

• 49:52Speaker 02

  I mean, yeah, well, okay, my most galaxy brain take from Nvidia, actually, that I feel pretty strongly about is, and this is, you know, in our call with them, they're like, you know, this $10 billion check is like kind of pennies compared to what we're trying to do. The real game we're trying to play, and if you think about it, being closer to their customers and understanding what they're doing is probably the number one thing that they have to do as Nvidia. To understand where the technology is going, right? And so I think if you, if you think about it, it's actually just R and D OpEx, bro. It's just a check that you pay in order to make sure that you're closer to OpenAI and Anthropic, and you know exactly what's going on in their data centers. So that's the, that's the most like bullish take I can think, but realistically, man, the Fed said something and everyone's freaking the fuck out.

• 50:38John Coogan?

  Yeah, yeah, that makes sense.

• 50:39Speaker 02

  Yeah.

• 50:40John Coogan?

  Yeah, I mean, on that, are you referring to that, that line that's getting shared around from the Nvidia earnings around the quality of the deal with, with OpenAI versus the strength of the deal with Anthropic? Did you read into that? Did you read the same thing that everyone else read into it?

• 50:55Speaker 02

  Yes, we, we did read into the same thing that everyone else said. It's actually kind of funny, though, because in the same language, they're like, the opportunity to invest. And it's like, yeah, it's just like, they glazed Sama, like, they glazed them a little bit, but then they also were like, yeah, this also couldn't happen at all. So I think, I mean, dude, a lot of the press release and the press release economy, as you know, you just say the biggest number and you're like, dude, until 2030, you have, let's say, I'm going to invest $600 billion in the United States. I'm going to be meta, right? I'll invest 200 billion from here to 2028 and then 400 billion from 2028 to 2030, right? You just push it into the back half.

• 51:33John Coogan?

  Yep.

• 51:34Speaker 02

  And then like, if it comes, it comes, right? That's, that's how you do it.

• 51:36John Coogan?

  I feel like people could go further here, too. I mean, Ray Kurzweil famously said Singularity 2045. You should be doing RPO all the way out to 2045. Why not?

• 51:48Speaker 02

  Yeah, I'll just have my children, actually. We'll do a deal with our children.

• 51:53Jordi Hays?

  My favorite big number of the week was MBS. Was hanging with Trump and he was like, I said $600 billion yesterday.

• 52:02John Coogan?

  Let's do it.

• 52:02Jordi Hays?

  But actually, let's actually, and Trump literally like hit, went like this and like hit him on the knee. Like he was so, he was so happy to hear that one trillion.

• 52:14Speaker 02

  He's like, finally, finally, someone said the trillion, bro.

• 52:17Jordi Hays?

  Everybody, everybody gets around, everybody gets around Trump and they just like detach from reality a little bit. And they just start saying like, Zuck, Zuck had this at the, at the AI dinner, remember? Like he just threw out, he threw out a number and had to, had to correct it the next day.

• 52:31Speaker 02

  He's like, uh, what, what number are we going with? Oh, six, six, six hundred, six hundred trillion, six hundred billion. Yeah. Dude, I mean, it's like, uh, it's like his, uh, his, his warp field is that, uh, everyone just says the biggest number around him. I love him. Like it's kinda, it's like, kinda ridiculous, dude.

• 52:45Jordi Hays?

  It's powerful.

• 52:47Jordi Hays?

  It's a, it's, it's, uh, it's real stimulus.

• 52:51Speaker 02

  It's, it's stimulating, that's for sure.

• 52:52Jordi Hays?

  Stimulating. I don't know if it's real, but stimulating.

• 52:55John Coogan?

  You need a semi-analysis plan, like, like a subscription where it's like $5 a month for the first 20 years. And then $20 billion in 2045. I will sign up for that and then you can book it and diffuse it back.

• 53:11Speaker 02

  For the revenue, it'd be beautiful, bro.

• 53:12John Coogan?

  Yeah.

• 53:13Jordi Hays?

  And yes, there's an out, you can cancel anytime.

• 53:16John Coogan?

  We need to bring massive RPOs and press release economy to the, to the newsletter analysis. We need, we need a Substack feature baked in for this.

• 53:24Speaker 02

  We can do, we can do, we can do a deal, bro. It'd be great. We'll do, yeah, we'll, we'll have a preliminary $1 billion advertising deal. We'll do some circular economy.

• 53:34John Coogan?

  Well, thank you so much for taking the time to hop on.

• 53:37Jordi Hays?

  Great to see you and congratulations again. So, so happy for you and your fiance.

• 53:42John Coogan?

  Inference max, inference max, inference max.

• 53:44Speaker 02

  Yeah, inference max, inference max.

• 53:46John Coogan?

  Inference max, inference max, inference max, inference max. Go start on GitHub.

• 53:52Speaker 02

  Okay, did you know, did you, sorry, did you, are you familiar with, we actually made a video of Dylan at our offsite just screaming analysts? And I was like, that's literally, it sounded exactly like that. I was like, what the fuck, how do you know about that?

• 54:06John Coogan?

  I have spies.

• 54:07Speaker 03

  Oh, yeah, we have, we have eyes and ears everywhere, Doug.

• 54:10Speaker 02

  Your information is incredible, bro. Holy crap.

• 54:14John Coogan?

  We were just having fun. Thank you so much for stopping by the show on a busy day. We'll talk to you soon. Have a good one.

• 54:20Speaker 02

  Yeah, nice to see you. Take care.

• 54:21John Coogan?

  Let me tell you about Linear. Meet the system for modern software development. Linear is a purpose-built tool for planning and building products. Our next guest is David Chang. He is an American chef, restaurateur, author, and TV personality. I believe he is in the Restream waiting room.

• 54:37Speaker 03

  Is he?

• 54:37John Coogan?

  He's the founder of Mama Fuku Restaurant Group.

• 54:40Speaker 03

  Not sure we have him quite yet.

• 54:42John Coogan?

  Well, we can also go all over the world. We have lots of content today. We will work on getting him in the studio. In the meantime, I will also tell you about Fall, to build and deploy AI video in image models trusted by millions to power generative media at scale. Speaking of generative media, Gemini 3 Pro Image only has an 8% error rate when generating text. OpenAI's model is at a 38% rate, and so there's been a very significant quantification of the improvement in Nano Banana Pro, I believe, or Gemini 3 Pro Image. I don't know exactly the difference in the naming conventions. Do you know?

• 55:22Speaker 03

  Yeah, I mean, I think, so originally Nano Banana was like the insider, like, that was the code.

• 55:28John Coogan?

  It was the code word.

• 55:29Speaker 03

  Code word, and then, then they're like, oh, let's just bring.

• 55:31John Coogan?

  Interesting. I was wondering how that happened.

• 55:33Speaker 03

  Yeah.

• 55:33John Coogan?

  Because it's very.

• 55:35Speaker 03

  They do the same thing.

• 55:35John Coogan?

  It's cool and it's quirky and it's actually very on-brand for Google, in my opinion, to run with a keyword like this. But, but at the same time, it's added a lot of confusion because they've done so much work just to establish the Gemini brand. And now they're also have a Nano Banana brand and it's a little bit confusing. But we can talk about that more after our next guest joins. We have David Chang in the studio. Welcome to the show.

• 55:59Speaker 03

  What's happening?

• 55:59John Coogan?

  How are you doing, David? Good to meet you. Thanks so much for taking the time to talk to us.

• 56:03Speaker 03

  Great, great fit, too.

• 56:05John Coogan?

  Looking great.

• 56:05Speaker 03

  You're ready, ready for anything.

• 56:07John Coogan?

  Yes.

• 56:08Speaker 04

  Well, I'm prepping out for a bunch of things right now and then again on plane and we'll be in Vegas by dinner.

• 56:16John Coogan?

  For F1, right?

• 56:17Speaker 04

  Yes, sir.

• 56:18John Coogan?

  Fantastic.

• 56:19Jordi Hays?

  We'll be there Saturday.

• 56:20John Coogan?

  Can you help everyone in the audience understand just the shape of your business between?

• 56:26Jordi Hays?

  The shape of your empire.

• 56:27John Coogan?

  The shape of your empire. Empire is the correct term. Sorry. Not just business.

• 56:32Speaker 04

  Well, it's restaurants. We have some quick service, fast casual.

• 56:39John Coogan?

  Yeah.

• 56:39Speaker 04

  We have casual, we have fine dining and we have a couple of spots in Vegas. We have a couple of places here in Los Angeles. I think it allowed us to pandemic allowed us to sort of refocus exactly our growth strategy.

• 56:53John Coogan?

  Yeah.

• 56:53Speaker 04

  Instead of trying to open up all around the world, which we had been doing until 2020, I think.

• 57:00Speaker 04

  We had a lot of sort of plans in place for doing CPG. So that, like many other things in the world at that time, sort of expedited the plan. And we went headfirst into, you know, we had dabbled in making some sauces here or there, but we had always wanted to go into making noodles. And that's sort of a good part of our business too. So it's equal parts restaurants, even though they're now split out into two completely separate entities. They take up a lot of my time and then me, it's mostly media stuff these days. So I'm dressed up right now because we're doing practice runs for our Netflix show on, which we air at 4pm Pacific Standard Time, dinner timeline.

• 57:46Jordi Hays?

  Have you, have you ever tried live streaming? We've, we've wanted somebody to do the version of our show that's just like a live, like a daily live cooking show. And I feel like you've got, you've certainly got the personality for it. It's, it's basically a full-time job, but I could see that being a hit.

• 58:03Speaker 04

  We tried to do that. You'd be surprised how adverse I think a lot of people that are running networks still. Because I think if anything, it might have to go on, you know, one of the three streaming platform services. But that's not a question. It's certainly on the long-term projects list for major domain media is to do just all day. You know, total transparency, what you see is what you get. And you do see some people doing it on Twitch.

• 58:29Jordi Hays?

  Yeah, that's what I'm saying. You don't need the networks. Just create a Twitch account, create an X account.

• 58:34Speaker 04

  But I'm currently pretty preoccupied with Netflix and Amazon and Spotify these days, especially since our podcast is moving over to Netflix.

• 58:43John Coogan?

  Oh, really?

• 58:43Jordi Hays?

  Oh, no way. You're in that, you're in that package.

• 58:45John Coogan?

  Yeah, talk, talk about, I mean, obviously, there's a bunch of particulars that are probably under the, under the hood, but what excited you about taking a podcast to Netflix? It's an interesting strategy. It wasn't on my list of predictions for what Netflix would do. What have they shared with you about how they'll surface that, what the audience might be like, what the Netflix viewer is looking for in podcast content? I find that whole strategy fascinating.

• 59:11Speaker 04

  I would love to answer all of those questions, but I don't think I'm the...

• 59:14John Coogan?

  Okay, anyway.

• 59:15Speaker 04

  I, I, I think if I was the person to answer that, it would be funny to both Spotify and Netflix because there are other people that are certainly designed to answer that.

• 59:24John Coogan?

  We'll have someone from Netflix on.

• 59:26Speaker 04

  I will tell you that it's, it's, it's been something, you know, we have, what, 600 plus episodes of our podcast and it's certainly changed over the years. When I first did the podcast, it was much more of an insider's take on the restaurant industry and, and sort of a sneak peek in terms of the thought process of opening restaurants or, you know, pre-opening of a friend's restaurant. Like my buddy, you know, opened up England, San Francisco. We, we sort of gave everybody a sneak peek of the, you know, the, the, the philosophy behind it. And then the pandemic happened and you couldn't travel. So it just sort of shifted and we've been waiting for this moment, quite frankly, where instead of talking heads, because food is the one thing which is sort of dumb, right? It's the one kind of podcast that we can't really react to culturally.

• 1:00:10Speaker 04

  I'm very close with Bill Simmons. I'm part of the Ringer Podcast Network. And, you know, we can't watch some movie and react to it. And we certainly can't watch a Monday night football game and then react to that either, right? So food is so ephemeral and in the moment, and also more importantly, not necessarily scalable. So in now with video, right, and more people, you know, with all the data and like more people are watching podcasts and actually listening to it. You know, there's certainly a lot of people still listening to it. Don't get me wrong, but it's, it's certainly in the near future going to outpace it if it hasn't already. And now it gives us the opportunity to evolve again and to offer a podcast that is somewhere between a TV show and a podcast.

• 1:00:52John Coogan?

  Yep.

• 1:00:54Speaker 04

  And I can do that because cooking is something I can do that, you know, other people that are maybe doing interviews or such as yourselves like, you know, would you be cooking and doing this interview right now? I don't know if a lot of people would sign up for that.

• 1:01:06Jordi Hays?

  It'd be very hard. It's hard to eat and, and do a podcast. Eating on mic is impossible, but even just cooking. We were, we were talking earlier on the show about kind of the delivery app experience, like the, the dynamics of like tipping and, and delivery apps. Travis Kalanick was commenting on, on, on a, on a post of ours yesterday. He's got a new product called Picnic, which is like a front end delivery platform, uh, just focused on corporate meals. And like the key value prop is uh, no tipping and no fees. So they're focused on like higher, higher volume orders. So like, uh, teams of like 25 people plus.

• 1:01:48Jordi Hays?

  I'm curious like how like getting, I wanted an updated take on from you on like what it's like working with the delivery platforms today, where you think there's opportunity, uh, and all that stuff.

• 1:02:02Speaker 04

  Well, um, I don't know if many people know or remember in 2016, we created the very first, to my knowledge, um, there might've been something called a ghost kitchen, but no one called it a ghost kitchen. We, we teamed up with, um, uh, Thrive and Will Gabrick and Josh and, um, uh, Caleb, et cetera. Great team. And we opened up Maple and we were doing like 10,000 meals a day out in New York City as a full stack app.

• 1:02:28Jordi Hays?

  Yeah.

• 1:02:28Speaker 04

  We were delivering.

• 1:02:29Jordi Hays?

  I had no idea. I, I didn't, I didn't realize you were behind that. I remember, I remember Maple, but, uh, that's cool.

• 1:02:35Speaker 04

  And, um, yeah, like it's something that I've been wanting to do and had done for a long time because I saw that's where food was going, particularly because of the, the, it's, it's been a 30% cut for, for some time on delivery fees. And that's just not a sustainable model for the, the delivery companies and the restaurants. So, uh, it was a real, uh, opportunity to sort of bridge that and to, to do everything oneself.

• 1:03:03Speaker 04

  And uh, yeah, and and so much so, I believe in that. We started another one uh called Ondo, which was more of a fast food. So like, maybe you might get a nice kale salad and then butternut squash soup. And Ondo, we started with um Garrett Camp's um uh fun expo. And we did Ondo, which was like a cheesesteak and fried chicken. So I was all in. I was all in. We were probably just six years too early.

• 1:03:28Jordi Hays?

  Yeah.

• 1:03:29Speaker 04

  It worked, but not enough where it is today because.

• 1:03:33Jordi Hays?

  But is it so is that is that model? Is the model like the the hard thing is like Cloud Kitchen's dominant like like company in that space, but they're just incredibly like secretive, right? And so it's hard for me to get a I haven't done, you know, much digging, but it's hard for me to get a read on like, is the model durable? Is is is there going to be a lot of value creation there? Or does it ultimately? Does it ultimately kind of fragment like a lot of the the restaurant industry has as well?

• 1:04:03Speaker 04

  Listen, I I think it's like anything else in tech, right? Remember in the dot com bubble? You had like tube socks dot com, right? Like,

• 1:04:10Jordi Hays?

  Yeah.

• 1:04:11Speaker 04

  There's only like three or four companies that really came out of that.

• 1:04:15Jordi Hays?

  Yeah.

• 1:04:15Speaker 04

  Uh, same stuff, what I imagine with all this AI shit, right? Um,

• 1:04:21Speaker 04

  you know, like, I can't even tell you in the food space, how many companies are a food logistics company, but now it's AI, right?

• 1:04:30Speaker 04

  It reminds me again in ninety nine, when at least in New York City, every company, I wouldn't say every company, but a lot of lot of places were now changing their name to sort of pizza two thousand, dry cleaning two thousand, right? It just is sign of the times. And and I I just, I think that in food delivery, you're over, you're gonna have about three to four winners, if that. And certainly, I would never bet against Travis and and the team there at Cloud Kitchen. Uh, definitely not. I I think what Tony and the team at DoorDash is doing is just unbelievable. And clearly, you have Uber and the Postmates guys. So um, that yeah, to me, that's pretty much going to be the space. And I think when

• 1:05:13Speaker 04

  I'm not sure when, but when they are able to be more open and transparent about everything, people are like, wow, that's a, that's a pretty goddamn huge business. Yeah.

• 1:05:21Jordi Hays?

  Makes sense.

• 1:05:23John Coogan?

  Drinking culture and like.

• 1:05:24Jordi Hays?

  Any restaurants been caught using AI generated imagery for their, for their menus or, or any of the food delivery apps yet?

• 1:05:32Speaker 04

  No, but you know, like I, I, I, I think it's been, it's the same shit that's happening, right? And that, and they're like, AI to me is like the, getting the lowest common denominator things and sort of like crowdsourcing and just getting something that is not necessarily perfect, but just good enough. And I, I just think that in restaurants, that's basically been like consultants, right? Um, they've just been, that's just like, I can go, I, I feel like I've been going to AI generated restaurants for some time now, you know? It just hasn't called AI.

• 1:06:03Jordi Hays?

  That's funny. It's a good take. Uh, are you excited about drone, drone delivery?

• 1:06:09Speaker 04

  You know, that's one thing where I thought it was going to be a total zero. Uh, I'm dead wrong about that one. I think it's definitely going to be a thing and, and, um.

• 1:06:17Jordi Hays?

  Well, isn't it, is it exciting as a chef to, to know like, if I make this, it will arrive hot?

• 1:06:23Speaker 04

  No, it's not going to arrive. Well, that's a whole other thing with this whole, the one thing I will tell you on the food delivery space, and I've talked to just about everyone under the sun over the past 10 years, um, that's tried to start up a food delivery company because they're like, oh, this guy's done it a few times. Let me just sort of steal all the ideas. And I'll tell them every time, I'm like, unless you created some kind of new technology to cook the food, it's going to be hard to really, um, make the food hot ultimately, right? Like, um, well, how should I say this? There's no new technology to make the food better. None. So, um, the delivery drone, unless it's cooking the food as it flies, it's always going to be limited.

• 1:07:05Jordi Hays?

  An oven, oven in the air, basically?

• 1:07:08Speaker 04

  Yeah. I mean, like, that's just the truth, right? Like, and also a lot of these places just have a bottleneck because everybody wants to eat at six thirty, seven o'clock. So there's just, there's not much you can do to make the food go out faster.

• 1:07:22Jordi Hays?

  Yep.

• 1:07:22Speaker 04

  Or hotter. And more importantly than not, everything can be delivered well. Like french fries will never be able to be delivered well, right? The next step is going to be whoever makes the food literally right outside the house or apartment.

• 1:07:34Jordi Hays?

  Have you heard any pitches on that?

• 1:07:36Speaker 04

  I have.

• 1:07:37Jordi Hays?

  Any, any?

• 1:07:39Jordi Hays?

  No, you don't, you don't have to, you don't have to give names, but like I'm, I like it, it gets to the point where it's like a street, we just have like, like a massive proliferation of like, street carts. And it's, and maybe, maybe like.

• 1:07:51Speaker 04

  Well, I mean, I'll, I'll tell you this, a lot of these pitches, I haven't been to one in a couple of years because I just don't want to do it anymore, but um. A lot of times there'll be a very success, like a, a chef that's worked 20 years at a three mission star restaurant. Making the food, you know, and it comes out and I'm always like, is this person going to be making the food? You're going to get this quality talent making the food at every single sort of satellite location. And the answer is they haven't even thought that far. And the other answer is that's not a reality, you know? You might as well be pitching me a unicorn, literally a unicorn with a horse and a horn on it because it's not going to ever work because. That's the hard part about this business, right?

• 1:08:35Speaker 04

  Uh, cooking is still a physical endeavor. And for all the, the, the VC money and tech money, it can't sort of.

• 1:08:46Speaker 04

  Solve that riddle of how do you make physical labor go away? Or done, or done better.

• 1:08:52Jordi Hays?

  So you don't, you don't think we're gonna, you're, you're very bearish on the humanoids chefing up.

• 1:08:58Speaker 04

  No, no, no, I'm not bearish on that either. Um, I spoke to somebody, pre pandemic, we did a show and we did some research and, and uh. Uh, a robotics expert and we talked to people at Caltech and a few other experts. And then somebody was like, oh, maybe 40, 50 years away. And I talked to, I talked to someone recently and they're like, yeah, we're probably 15 years away. Um, from getting somebody that has a robot that has the dexterity of a high end, best in class sort of chef.

• 1:09:31Speaker 04

  So no, I, I, it's going to happen. If anything, I think you're going to see, um, the next five, 10 years, you're going to see robots. You already see it in, um. I mean, I, again, like I, I don't think it's been a sudden, oh my God, there's robots, there's robots in the kitchens all the time. Like if you go to a good restaurant, there's dishwashers, a, a pretty much a transformer robot. It's amazing and that does the work of like twenty people.

• 1:09:56Jordi Hays?

  Under, pretty underrated.

• 1:09:57Speaker 04

  Underrated and, and if anything, you're gonna see machines that take the binary movements out. So a fryer that goes up and down, a bathroom cleaner, things like that. And already dishwashers are pretty advanced and that can handle very expensive stemware. So finding somebody that can polish stemware. You know, you might get a wine glass that could be two hundred fifty dollars per glass. And if you have a hundred of those, that's quite an expensive inventory for a restaurant. You need someone specifically trained to do so. And that's a hard position to find. So yeah, that, that kind of position will be a robot, no questions about it.

• 1:10:35Jordi Hays?

  That makes sense. Uh, what's the most overrated trend in food right now?

• 1:10:41Speaker 04

  Oh, man. I'm trying to stay positive these days, guys.

• 1:10:46Jordi Hays?

  I, I, I, uh, uh, I, I think, I think there's a way to answer the question by just saying like, there's things that can be popular now that are not dur- like durable trends, so.

• 1:10:59Speaker 04

  Well, I, I would say the, the most annoying trend is that everything has to be the best. This hyperbo- hyperbole, right? I have to have the best X. This restaurant has to be, you know, world class, number one. And, um, I hate to tell it to you guys, but I think most people don't, wouldn't even know what best is if they ate it.

• 1:11:22Jordi Hays?

  Yeah.

• 1:11:22Speaker 04

  Um, and I think for the most part, I'm just now on this mantra personally of, does it bring you joy, does it bring you happiness? And that's really all that should matter. Um, it's such a relative subjective thing, but more importantly, I'm just trying to tell people like, good is fucking hard to do. Like just good is hard. And I think we need more people to sort of appreciate just good or like even boring good. Then the world's best, oh my God, this is the greatest thing I've ever had. That to me is the worst trend in the world is, you know, and, and the media and chefs, we're all part of the problem too, right? You know, all these lists and it's all stupid ultimately.

• 1:12:00Jordi Hays?

  Yeah.

• 1:12:00Speaker 04

  But.

• 1:12:02Jordi Hays?

  Do you think, do you think we'll ever get to a point where like in tech, in the technology industry, many, many products get better with scale for like a variety of reasons. Food has felt almost always the opposite of that where you take an amazing concept in a, in a tiny, tiny little restaurant, right? Like, you know, a thousand square feet. And the second you add the second restaurant and the third, it just kind of tends to get, it tends to get worse and worse and worse and worse over time. And that's just like, there's one, there's one, you know, maybe it's one chef who had, had an amazing idea and it, and it just is very difficult to scale quality. Are you optimistic that there could be any new technology introduced that would kind of change that dynamic?

• 1:12:46Jordi Hays?

  Or is that just kind of an iron law?

• 1:12:49Speaker 04

  No, I know. I think it's not an iron law. It's not like a law of thermodynamics. I'm sure somebody could figure it out, but you know, the, the, you just mentioned something, the fact that something that's great is not scalable.

• 1:13:01Speaker 04

  And for years, you know, I've certainly tried to do it and scale these things. I, I just sort of spoke about this at Reid Hoffman's Masters of Scale Conference, right? Like I sort of, everybody's at that conference because they want to scale an idea. And I said, you know, the, the easy ideas are to scale an idea in food that is, um, I don't say cheap, but affordable and mass-produced, right? The other end is, um, high-end experiential dining, which weirdly has become very scalable because of its inaccessibility. It was, it's the equivalent of like getting front row tickets at the Knicks or, you know, Chase Center or something like that, because you're now eating at, say, the French Laundry. No one else who can get there. You may not even appreciate the food,

• 1:13:46Speaker 04

  but it's now a social flex. It's cultural currency that you can sort of have and it's ephemeral. And because no one can have it, weirdly, now that experience is weirdly.

• 1:13:58Speaker 04

  I mean, it's scalable because that actually is crazy marketing for the French laundry. Right? And the demand for that kind of restaurant is through the roof. And what I mean by that is restaurants, it doesn't have to be super high end in in Napa Valley. It has to be anything that can't be copied immediately. Right? You can't watch a YouTube video and decide I want to open up a restaurant like this. I can't just, you make it easy facsimile. So it could be barbecue, it could be sushi, anything that is best in class and people have a hard time copying. That's like the barbell. So you have really affordable, make mass produce stuff on food on the one end. On the other end, you have things that very few people are going to be able to experience or eat.

• 1:14:40Speaker 04

  And, you know, that's been sort of elevated because of technology, right? In different ways. But I chose sort of challenge the audience that, you know, because every, I mean, I mean, you guys know, I, I, I talked to a lot of people in tech and probably a lot of your peers and they're always wanting to know the next big thing in food. And I'm going to tell them like the hardest thing, the answer that needs to be solved is how do you scale the middle?

• 1:15:09Speaker 04

  Right? Does that make sense? Like the mom and pop restaurants, the diner, the, the restaurants that are just like good. Again, uh, how do you make it so they can survive? Because they're like cultural banks. They're great, but they're not, they don't have the sizzle. They don't have the, maybe the bottom line that makes it sort of cool for investment.

• 1:15:30Speaker 04

  And again, it's not about creating a company that's like the pickaxes and shovels for that middle market restaurant. But there's got to be something else that can like be something that's game changing. I don't know exactly what it is, but, uh, I never talked to the people that are trying to make food concepts and invest in food concepts that are actually concerned about the middle. And I'm not talking about credit card processing and shit like that. I'm just saying like in general, um, there's a lot there. That's the media's part of the food industry right now. But it's just too damn hard and nobody really wants to touch it. Interesting.

• 1:16:03Jordi Hays?

  Have you seen any interesting experiments on like the capital side of, of, uh, like new restaurant creation? Has anybody tried to make like a Y combinator for, for restaurants where there's, you know, a talented, you know, uh, owner operator chefs can get some seed capital and kind of support, uh, to go from zero to one?

• 1:16:27Speaker 04

  Um.

• 1:16:27Jordi Hays?

  Cause I feel like you would have tried that by now.

• 1:16:30Speaker 04

  We have definitely tried it. I I won't say there there. We've tried it. I know a lot of restaurants have tried it. I know there are spuns out there um that try to do this, but um I would say um. You know, Ron Parker created something called Hospitality NX, and that's a website that is a little bit like a job board, LegalZoom, but also a place for people that want to raise funds. So that that's something. Um, but I think for the most part, um, it's not as organized as other. You know, at the end of the day, it's because um it's hard to create an idea that uh um has a high barrier of entry in food.

• 1:17:16Jordi Hays?

  Yeah.

• 1:17:16Speaker 04

  Right? And there's no moat to really create, right?

• 1:17:19John Coogan?

  Yeah.

• 1:17:19Jordi Hays?

  Yeah.

• 1:17:20John Coogan?

  On the on the topic of like, you know, starting up and go to market strategies, are there uh? Are there risks to like going too viral early? Uh, we've experienced a bunch of uh like rage bait in tech recently where people have sort of designed products that are that are in designed to the the whole product is just designed to enrage and go viral and then get some attention?

• 1:17:45Speaker 04

  What do you mean by enrage? I I I'm not familiar.

• 1:17:47Jordi Hays?

  So so so a company, a company made a product that is like a developer tooling. So it's like software to help you make software. And they use AI in it. So there's minute, there's time periods where you have a little two minute break. And so they added the ability to gamble with steak while you're while you're making software. And that made a lot of people mad for for obvious reasons. So.

• 1:18:10John Coogan?

  Yeah, or just deliberately picking messaging.

• 1:18:12Jordi Hays?

  Like rage rage bait in food would be like a product that has like a single meal that has a thousand grams of protein. Like I could see a restaurant doing that just for the just just to try to get people to make TikToks about it.

• 1:18:26Speaker 04

  I mean, yeah, I mean, I don't know a rage bait, but like, again, like this has been happening on for for forever anyway, you know? Doing something that is probably gonna, you know, shit, I've opened up restaurants that I guess have been like that too, you know, it's just.

• 1:18:43Speaker 04

  You're not, you know, I I think if anything, it's just taken to another level because.

• 1:18:48Speaker 04

  I would say that a lot of chefs now, when they're talking about dish, is it, is it something that the younger generation will find appealing to, to, to, to, to record?

• 1:19:01Speaker 04

  And that's what I mean. It's like, it's this, it's vaguely experiential, but it's very ephemeral at the same time. So, um, I don't know. I want to be optimistic again. I'm usually Mr. Eor over here about this, but I do think that with all of this aside, with all of this access, with all this democratization of knowledge, because culinary knowledge with the younger generation is higher than it's ever been. I mean, it's never been better to eat in America. It may not have the sort of the titans of the industry as it used to because things have sort of leveled out, but eating today, like, I talk about this with people a lot in the industry that travel. You can find a great restaurant in every city in America for the

• 1:19:45Speaker 04

  most part now. It's pretty remarkable. If you just look at that, right? So maybe New York or San Francisco or other metropolitan cities are not as great. They're still great, but it's really broadened out and flattened out across the country. So Oklahoma City and, you know, places that are tertiary cities to most people are actually, might have some of the best restaurants in the, in the country. And I think that sort of pattern is what you're going to see throughout food. And it is a long-winded way of sort of answering this sort of rage bait. And I think because of that need to sort of find something that is going to create kind of some kind of spark in food, that is the catalyst that's going

• 1:20:26Speaker 04

  to cause people in food to get better at their craft. Because at some point, all of that bullshit is just going to wash away and you're going to be left naked with something. And if you want to be able to have the real goods to show for it. And I think that I really feel strongly that food is about to go into this very specific point of.

• 1:20:52Speaker 04

  Like a little bit like Japan where you can open up one specific kind of bakery that that makes one specific type of thing and you do it better than anybody else. And you're gonna see that here in America. I I feel very strongly about that.

• 1:21:05Jordi Hays?

  Yeah, I I love that approach. What uh what advice do you give to uh kind of emerging chefs on media strategy? I think in in tech, there's like like we tend to see kind of a high-low strategy where you want to be like super online, getting a lot of attention, or you want to be kind of the the mysterious dark horse that's kind of going over the radar and there's like a messy middle that's probably a disaster.

• 1:21:34Speaker 04

  Yeah, I I don't think that that pattern is any different than than what you see in food. Um but at the same time, I think uh, you know, I don't know if apathy is the right word, but I don't care about it as much anymore either. Um because it just, I know I'm not the only chef that feels this way. It's just some people are doing it more than ever and getting better at it, but others I think are just sort of getting exhausted by the whole thing because um I just don't know what that best long-term strategy is. Um and now you have an older generation of, you know, I'm I'm forty-eight years old. I know chefs that I won't say who, that are like clearly gotten a social media strategist or somebody because their content is really fucking good right now.

• 1:22:21Speaker 04

  And I it it we we I I still don't know which one works, right? Um because once you feed that beast, you have to do it all the time.

• 1:22:30Jordi Hays?

  Oh, yeah.

• 1:22:31Speaker 04

  And that's a lot of time. So I I I don't know if the better thing is to just be word of mouth because ultimately all of this is is word of mouth.

• 1:22:39Jordi Hays?

  Yeah.

• 1:22:40Speaker 04

  Right?

• 1:22:40Jordi Hays?

  And and do you build a relationship in that repeatability and and like there's my favorite restaurants in LA, like they don't have to do marketing to me, you know? I don't need to get an email. I don't need to see them on Instagram. I'm just gonna go there. Like when I have the time, right? Um, and so I, I think.

• 1:22:58Speaker 04

  I mean, yeah, I think, I think that's the zag, right?

• 1:23:00Jordi Hays?

  Um.

• 1:23:01Speaker 04

  But you can't do that unless you actually have a point of view that resonates with somebody.

• 1:23:05Jordi Hays?

  Yeah.

• 1:23:06Speaker 04

  And if you are constantly sort of pandering and figuring out like how to execute other people's dreams, wishes and visions and what the hell are you actually making?

• 1:23:15Jordi Hays?

  Yeah, and you don't want to get to a place where your content is better than, better than the product. And I'm sure that's like, uh, you know, a lot of people, the more, the more you time you spend on content, like the, the more greater there is a likelihood that, that it could get to that point, I think.

• 1:23:29Speaker 04

  Yeah, I mean, but like, do you guys care about what you see on social media still? Like, I, I actually think there's a bifurcation that's happening with what people see versus what actually people are going to eat.

• 1:23:39John Coogan?

  I do think that there's, I, I don't know, maybe the, the, the steel man argument for the, the viral, over the top, uh, you know, Tik Tok that gets me to go to a restaurant is that it can in some ways create like a shelling point and like a, a, a coming together, like a, if there's something that's trendy and, and I, and it's an excuse for me to pull my extended family, my friends, different people, and it just gets us a, a, an opportunity to kind of come together there and experience that like even silly trendy, over the top thing. Uh, I think that there's something that can be good about that, but, um, but it's certainly not like the primary uh, reason why I go to a particular restaurant.

• 1:24:24Speaker 04

  No, I mean, that's the thing is like, I actually, I, I, um, we're working on a show and I can't say which or where, but um, you know, it's sort of the thesis is we're going to take these lists that people find or things that are viral and actually go out of our way to avoid it.

• 1:24:41John Coogan?

  Okay.

• 1:24:44Speaker 04

  You know?

• 1:24:44John Coogan?

  Yeah, yeah.

• 1:24:45Jordi Hays?

  Fade, fade everything.

• 1:24:46Speaker 04

  Go next door to the restaurant that you're supposed to go eat at.

• 1:24:49John Coogan?

  Okay. Oh, that's cool. That's very cool.

• 1:24:52Speaker 04

  Sort of that, that in principle, right?

• 1:24:53John Coogan?

  Yeah, no, I, I like that as a philosophy.

• 1:24:55Speaker 04

  It's just like, the other thing is.

• 1:24:58Speaker 04

  I I I I sort of mentioned that earlier in this, the conversation about if somebody was tasting something that was truly good and remarkable, would they actually know what's good and remarkable? And I I think currently, we we again have a a knowledge uh uh that is greater than it's ever been in terms of food, but. And maybe this is same way in fashion and architecture and film and other arts, but.

• 1:25:20Speaker 04

  Does your audience actually know what good is anymore? Because I don't I don't know, right? And and I'm not, it's not trying to be snooty or an artist. I'm just saying like, let's just talk about wine right now. If if I'm giving somebody like a like. Nineteen ninety-eight Ravino, you know, from Bir- like White Burgundy. To somebody that has never tasted it before, I know that it might taste good to them, but will they appreciate it? Because this person might be more into natural wines than, you know, oxidization, et cetera, et cetera. So it's like, I'm not saying that they're not right, but. I always joke, like, you can't, you know, my friend used to say, um, you can't, um, you can say that.

• 1:26:05Speaker 04

  You can never say that Salieri was better than Mozart. Right?

• 1:26:10Jordi Hays?

  Yeah.

• 1:26:10Speaker 04

  You he he was good, but he was not better. And that's just sort of unequivocal.

• 1:26:14Jordi Hays?

  Yeah.

• 1:26:14Speaker 04

  And you can appreciate Salieri, but you can never say that he's better than Mozart. My concern is people don't even know who fucking Mozart is right now.

• 1:26:22Speaker 04

  And and I and that's sort of my concern when it comes to um sort of social media and food is. Who's deciding what is actually good? Just because something looks good, doesn't mean it actually is good. And I know this is getting into a meta sort of philosophical conversation, but this is the shit I think about.

• 1:26:44Jordi Hays?

  Oh, I love it. Last question on my side, I'm curious how uh restaurant operators are planning around America just drinking less than ever.

• 1:26:57Speaker 04

  Yeah, well.

• 1:26:59Speaker 04

  That is the, you know, I, I feel like the boy who cried wolf. I've been sort of screaming this flag for a long time. Um, this has been, this is the real existential threat. Um, like for example, LA, the biggest thing that happened in LA over the past 10 years in food was really ride sharing. Cause people were getting drunk. And you saw that in revenues, restaurants who are going through the roof. And if anything, restaurants was a bubble, right? Um, too many restaurants and I think we're still sort of in this bubble. Um, that's a whole nother conversation, but um, I, I think that you can see now, at least in LA. People are drinking much less. I think you see a younger generation, maybe taking some edibles.

• 1:27:43Speaker 04

  They're just not, you know, the crazy thing is I, I, kids just don't drink anymore. Like kids start, when they start a tab, which is crazy to me. They close it out every time.

• 1:27:53Jordi Hays?

  Yeah.

• 1:27:56Speaker 04

  Is going on. Like they're never going to know what it's like to wake up at three in the afternoon and be like, shit. I left my credit card at that bar. I gotta go back and get it.

• 1:28:05Jordi Hays?

  They're too responsible.

• 1:28:05Speaker 04

  You know, they're closing out every time. There, there's a responsibility.

• 1:28:08Jordi Hays?

  It's hurting small businesses. It is.

• 1:28:09Speaker 04

  It is hurting small businesses and, and um, but I think that there is, if you look at the sort of the, only look at the blended numbers for most. Restaurants or bev sales, I think that it might look flat or down, but it's actually, I think way worse because once you split out. The 1% of the 1% that are drinking like these huge bottles of expensive wine, right? And that is through the roof right now. Again, talking about the barbell experiential thing, like people that are drinking things that no one else can really afford. That's gone like three X, four X of the past five years. It really has. And you know, younger people are not drinking cocktails and they don't want mocktails. Because mocktails are actually way more difficult to make than a regular cocktail with alcohol in it, but nobody wants to drink it for the same or more.

• 1:28:52Jordi Hays?

  Why is it, why is it more, why is it more difficult just to actually deliver something that's like?

• 1:28:57Speaker 04

  Imagine if we were making the alcohol too.

• 1:29:00Jordi Hays?

  Yeah.

• 1:29:00Speaker 04

  From scratch. That's hard to do. That's, and that, you know, a normal restaurant, restaurant ratio was 70 to 30% for the most part. You want 70% food. I mean, this is not the, I like.

• 1:29:11Jordi Hays?

  Roughly.

• 1:29:13Speaker 04

  Roughly 70% food to 30% bev sales. And I think that is completely shifted. And for a good restaurant that might have like. Yeah. I mean, like if you want 10% of your, you know, profit, for example, right? Like. Something's gonna give when you're down like eighteen percent on bev sales.

• 1:29:31John Coogan?

  Yeah.

• 1:29:31Speaker 04

  You know, I think that's the average right now or something like that, fifteen, eighteen percent. Um, so I don't have an answer. Uh, food needs to get more expensive. I've been saying that for a long time. But that comes across as terrible when people read that as a pull quote.

• 1:29:45John Coogan?

  Yeah.

• 1:29:47Speaker 04

  Uh, because it's already expensive. So I don't know what the answers are. I will tell you that like, you know, it's one of the reasons why I invested in athletic brewing. Uh, in, in, in, uh, two thousand and nineteen. Uh, because I saw the data within our own restaurants. It was slowly going down year after year, just a little bit, like half a percent, one percent. But, you know, I, I, and I'm, I think that's what we can do is sort of figure out what the alternatives are. I don't have the answer.

• 1:30:14Jordi Hays?

  But isn't one of, isn't one of the challenges is like these non-alk products, like somebody's not like, there's not the incentive to have the second or third. Like I, I feel like a lot of the stuff people just have one, they, they get a little bit of the taste, but they're not getting like a real, they're not like getting, they're not. They're, they're just like not getting drunk, right? So they're not going for that third or fourth.

• 1:30:38Speaker 04

  Are you guys drinking as much as you used to?

• 1:30:39John Coogan?

  Absolutely not.

• 1:30:41Speaker 04

  No, you know, I feel like the way I used to was like Don Draper in Mad Men, the amount I used to drink.

• 1:30:47John Coogan?

  Yes, yes.

• 1:30:47Speaker 04

  You know, and I, you know, part of that is just a generational shift, but I, I can assure you, if you talk to people under a certain age group, the younger Gen Z. They think of drinking like it's smoking cigarettes.

• 1:30:58John Coogan?

  Oh, yeah.

• 1:30:59Speaker 04

  It's just not something they want. I, I've seen this in kitchens, like you finished your twelve, fourteen hour day. All you wanted was that cold beer at the end of the shift. And now they don't want that.

• 1:31:09John Coogan?

  Yeah.

• 1:31:10Speaker 04

  And I just don't, I'm just like, what is happening? You know, and I'm not saying they're wrong. It's just so, so that we're sort of dinosaurs.

• 1:31:16John Coogan?

  It's just different interpretation. Like the data didn't change, but it was contextualized through podcasts and there's a lot of health data out there. Uh, I, I, you could maybe call a little bit of the Huberman effect, but there's a whole bunch of, there's a long lineage of folks who have been like, actually ringing the alarm bells on the health consequences of, of drinking alcohol, even in small amounts. And so that's, I feel like that's what's really cascaded.

• 1:31:39Speaker 04

  Yeah, I mean, maybe what we should do, restaurants should start a lobbyist and just muzzle Huberman and Peter Tia and we'll be okay.

• 1:31:47John Coogan?

  Yeah, live life, uh.

• 1:31:49Jordi Hays?

  Yeah, I mean, I like the, the dual pressure right now from, from, uh. Like just labor labor costs on one side and then and then uh just like declining alcohol sales like it's just creating I mean I've seen some uh the the place we go for breakfast adds like four percent on top of every bill for for benefits I'm I'm sure that that's helpful but like it's a very real cost right it it's now twenty five percent between effectively for or twenty four percent for service.

• 1:32:18Speaker 04

  At the end of the day, food needs to be more expensive and I and I'm not it it just sort of has to and it can't be sort of be passed down I think I I I've been talking about this for many, many years. I don't know why, but people have a real allergic reaction when it talks to raising prices. For example, I think, you know, it's good, I I I'm pro when a restaurant jacks up their prices to like I'm hoping we see a restaurant where the the the the the ability to eat there is basically going to a Taylor Swift conference, a secondary, a mixed secondary market, you know, like that's sort of what

• 1:33:00Speaker 04

  has to happen. I I do believe there's gonna be innovation and again, the problem with the restaurant industry as a whole to sort of mitigating this decline in beverage sales is that we are too slow and prodding to to try new things out, to embrace new technologies. And as my sort of spiel and joke about this, as a whole, we're so goddamn allergic and slow to changing things, we still are using the metric uh imperial system instead of the metric system. I mean, that's so dumb. The metric system is scientifically proven to be more accurate and more effective. Why are we still using ounces, pounds? It's so dumb.

• 1:33:41John Coogan?

  America, baby, is because we're Americans. We do things the dumb way sometimes.

• 1:33:45Speaker 04

  Americans can still do it, but as an industry, as restaurant leaders, we can just use the metric.

• 1:33:51John Coogan?

  You can just use metric, but.

• 1:33:53Speaker 04

  And again, you know that it's bad when drug dealers use the metric system.

• 1:33:57John Coogan?

  Drug dealers use the metrics, that's right.

• 1:33:58Speaker 04

  So what the hell are we doing here? So if we can't adopt the metric system as an industry, what what are we doing here?

• 1:34:06John Coogan?

  Yeah, yeah, yeah. What a mess, what a mess.

• 1:34:08Jordi Hays?

  Last question, we've got a bunch of people in the chat have asked, who do you think's gonna win the AI race? Hot take.

• 1:34:14John Coogan?

  What? Really?

• 1:34:18Jordi Hays?

  We had to ask, this is a tech, we use a tech show, just give me your gut answer. First reaction, one word, one word.

• 1:34:26John Coogan?

  Google, Anthropic, OpenAI, who you got? Commodore Computers. There we go, Commodore 64. Coming at you.

• 1:34:34Jordi Hays?

  Dark horse in the race.

• 1:34:35John Coogan?

  I love it.

• 1:34:36Jordi Hays?

  Love it.

• 1:34:37John Coogan?

  Thank you so much.

• 1:34:37Jordi Hays?

  Great, great hanging.

• 1:34:39John Coogan?

  Well, we'll see you guys at F1. Yeah. With the Blagio? Yeah, we'll be very excited to see you there.

• 1:34:43Jordi Hays?

  Can't wait.

• 1:34:44John Coogan?

  Have a great one. We'll talk to you soon. Be good.

• 1:34:46Jordi Hays?

  You're the man.

• 1:34:47John Coogan?

  Have a great rest of your day. Uh, let me tell you about graphite.dev, code review for the age of AI. Graphite helps teams on GitHub ship higher quality software faster. We have been keeping our next guest waiting for far too long. Uh, Lordan from Figma. Thank you so much for, uh, holding tight.

• 1:35:04Jordi Hays?

  Look at this, look at this background.

• 1:35:06Speaker 05

  I'm hungry now.

• 1:35:07Jordi Hays?

  I couldn't even tell if it was, I just noticed that it was a TV, but it took me a second.

• 1:35:10John Coogan?

  A lot of the, a lot of the professional TV hits on CNN and CNBC, people will be sitting right in front of a TV and they'll put, uh, some sort of fake background.

• 1:35:19Jordi Hays?

  Very cool. So great to have you on the show.

• 1:35:23Speaker 05

  Thanks for having me.

• 1:35:24Jordi Hays?

  We, we've been reacting to, uh, Nana Banana Pro this morning. Very, very impressed on a bunch of different dimensions, but before we get into that, would love, uh, an introduction on yourself for the audience and, and your background.

• 1:35:38Speaker 05

  Yeah, of course. So I joined Figma two months ago as their chief design officer. And before then I spent close to a decade at Meta, primarily working on messaging. So I led the Messenger and Instagram DM teams and more recently leading consumer AI on the product side. And, um, you know, before you ask me, I'll tell you why I joined Figma. I did so because in my seat, watching all of the, um, AI improvements that, that we're seeing with these frontier models, it became very, very clear that product development as a process is going to change drastically. And, um, I truly believe, and I saw that Figma has the opportunity and I think the responsibility from my, um, point of view.

• 1:36:22Speaker 05

  To really build the creative environment that helps people like me, uh, people that really love to live at the intersection. Um, I used to be a musician, I became a designer, the, you know, product leader. I really believe in the thing we're making more than like how different disciplines kind of like line up to, to get the product done. And so I believe that a creative environment that helps to get that idea from your head into a finished product is what we need right now, and I'm excited to help to Figma, um, build this.

• 1:36:52Jordi Hays?

  Amazing. So, so many different ways that you can integrate AI into Figma. You guys have been doing Figma Make, there's also like the core product. What, what's been your priorities kind of in the first couple of months?

• 1:37:06Speaker 05

  Yeah, so for everyone who doesn't know, Figma Make is the place in Figma where you could take your ideas or designs and prompt them into working software. And that's really important because it takes your design and like helps you understand what it looks like, what it feels like in motion. Um, and what we've been, um, looking into with this, um, is an aspect of AI that, um, I think gets overlooked sometimes. As a creative tool, it's important for AI not to box you in. Um, so you want to be able to take your design from Figma, Figma design and, um, generate it, but then you want to take those generations back to Canvas and be able to like manipulate them as well. So,

• 1:37:47Speaker 05

  and we're moving pretty fast. In the last two months, I think we've shipped over 20 major features. And a lot of them have to do with like putting the designer, yes. Putting the designer in the driver's seat and enabling the designer to take these AI tools, but really wield them as tools that are precise and that go in their direction versus just kind of like.

• 1:38:09Jordi Hays?

  The number one most frustrating thing with generative AI right now is you generate an asset that's like 98% amazing. And then there's like one tiny element and you try to like re-prompt it and you try to say like, uh, can you remove that? Could you like try again on that? You know, you.

• 1:38:23Speaker 05

  And then it all changes.

• 1:38:24Jordi Hays?

  Yeah, yeah, yeah. And then it's all changing. And so just like. Yeah, making it making it easier to like go back, go back and forth, I think is like probably some of the most important work on the on the creative side.

• 1:38:34John Coogan?

  Do you have a personal evaluation that you run when a new generative image model drops? I have this, the Where's Waldo test, I try and get it to generate a full Where's Waldo? Because that's, there's a lot of detail in there, it's this whole laddered up image. Do you have a favorite image that you go to as like your ground truth? Just to kind of get the flavor?

• 1:38:56Speaker 05

  I don't necessarily have a shit ton of styles that I put it through the wringer with and a number of like creative tasks that I want to see if it does. What's really important with these images and hasn't happened necessarily predictably so far is that they take that certain first scene that they generate or the photo that you give them and then they're dependable in recreating the style and telling the second part of the story. Otherwise, they're not helpful. An image that doesn't tell a story is not helpful. This is why I like actually Nano Banana Pro, because it's dependable. The way Weeby, one of our companies, said today is like, it's a model that behaves. You should actually watch the video that they've put out.

• 1:39:41Speaker 05

  It's hilarious and it's in Weeby with Gemini 3.

• 1:39:46John Coogan?

  Yeah, yeah, apparently, I mean, Prins here on X is saying Nano Banana Pro is a reasoning image model and here's a quote, "This enables enhanced image quality, better rendering of long text passages in many languages, improved factuality," which is something like, we didn't, like, I was never thinking about the factualness of an image generator, but that's actually extremely important, like, you don't want errors.

• 1:40:07Speaker 05

  Of course, is it realistic? Is it something that that really connects? Like, our our eyes, right? Like, will pick up on details even before you understand what's going on and you'll understand that this is an AI-generated image. And the truth is that people prefer to look at things that feel human, that a human has put out there in the world. And what's really cool with products like Nano Banana Pro is that you're able to manipulate that and because it's your creative tool, you could layer all of the different elements. Like, as an example, you know Dylan loves to post. Videos with his like Figma quilt behind him.

• 1:40:43John Coogan?

  Yeah.

• 1:40:43Jordi Hays?

  Yeah.

• 1:40:44Speaker 05

  And so I took that, I then generated the quilt directly, then I turned it into a sweater. And then I put it on, you know, one of Dylan's photos and all in all of these steps. It kept each square of the quilt exact. It did not distort Dylan's face. I could do what I what was in my head versus in in other types of tools like this. This has not been possible yet.

• 1:41:10Jordi Hays?

  What? How how much do you care about like leveraging some of these models to help people generate new ideas? Because in my in in my creative process is just like creativity oftentimes is just like taking two different kind of like random disconnected ideas and bringing them together. And sometimes it just hits. And I feel like.

• 1:41:32Speaker 05

  Yeah, creativity is messy.

• 1:41:34Jordi Hays?

  Yeah.

• 1:41:34Speaker 05

  It's like bringing a lot of like disparate things into into the canvas in one way or another and letting them inspire you and like take the taking the next step with those. And that's probably the biggest role of AI in the creative process right now is like, how much can you explore because these tools exist? Because in the end, what you're trying to create is still the thing in your head.

• 1:41:57Jordi Hays?

  Yeah. Yeah. That makes sense.

• 1:42:00John Coogan?

  Have there been any internal memes that have been floating around in within Figma? Like, I'm thinking of the Studio Ghibli moment that was really big on the internet broadly. But have there been any like, like just fun prompts? I I'm seeing people use Nano Banana Pro to make RPG style maps. People are are using, there's always like a new like fun prompt that kind of goes viral on the internet. I'm wondering if you have any glimmers of what might be the fun prompt from Nano Banana Pro based on what you've seen in in the internal team chat.

• 1:42:35Speaker 05

  I haven't seen any in the team chat outside of like just broad variations. So like none of them really came up to the repeating like patterns so far. But insane variations, like you know, taking things that are just sketches and like filling them in with like complete 3D. And like as designers, really what we love to do is explore. So we push this thing pretty hard.

• 1:42:56John Coogan?

  Yeah, yeah, I'm excited to get deeper into it. It's such a fun tool.

• 1:43:00Jordi Hays?

  What's your, what's your updated read on just like general designer sentiment around AI? Cause I feel like it fluctuates from, from fear to excitement. And you have pockets where people are super excited and you have pockets where people are kind of not excited about it or calling it slop. But what is like the most up to date read from your view specifically with like?

• 1:43:23Speaker 05

  Yeah, I could relate to all of those points of view in some way, right? Because if AI is just about speed and mass production of software and design, like that is very anti what I'm here to put in the world. But at the same time, if design becomes a tool that you could actually control and it starts to inspire you, as you were saying, that's a very different thing. It really widens the canvas. And this is why we're so interested in all of the new models and we put them through the wringer. Because we want to see how in the hands of designers these become clay that they could mold. And so I think the, the different opinions are just really at which point, which part of it do you look? Do you look at the potential and what's, you know, kind of what's, what's coming up and how it could work? Or do you look at exactly what it produced yesterday?

• 1:44:08Speaker 05

  In which case, a lot of times it is not great.

• 1:44:11Jordi Hays?

  Yep, makes a lot of sense.

• 1:44:13Speaker 05

  David Chang was saying something about like, you know, good enough and you know, producing just something good. Like we have this thing, like at least the thing where we believe good enough is not good enough. If all, you know, we're able to do in the future is create the same software a million times. That is just humanity losing.

• 1:44:31John Coogan?

  Yeah, yeah, it's interesting. I process those two things very, very differently. But yeah, I understand where you're coming from on that.

• 1:44:41Speaker 05

  Well, say more.

• 1:44:42John Coogan?

  I just, I process just this idea of like, I guess my question is like, what is the mom and like, what he was getting at was like, what is the mom and pop restaurant that's not going to make an awards list, that doesn't have the most viral turducken where it's like-

• 1:45:00Speaker 05

  Oh, reliable, yeah.

• 1:45:01John Coogan?

  Yeah, yeah, it's not, it's not superlative. It's not the world's heaviest doughnut, the world's like most, you know, gold flakes on a steak possible. Like, it's not viral, it's not the best. And even just in terms of fine dining, it's not, oh, it has the 10 Michelin stars, it's the best, the best, the best. There's this demand for the superlative in the restaurant industry. And then there's also the demand for just the cheapest fast casual, just get in, get out, it's a complete commodity. And I understand what both of those are in the design world a little bit. I mean, I feel like we've, we've seen design trends, you know, like from Apple and, you know, where we've all been like, wow, like that is truly like the best UI possible, right?

• 1:45:45Speaker 05

  Emotional connection, yeah.

• 1:45:46John Coogan?

  And then we've also seen just like, okay, like, that's just like the bootstrap design library that everyone uses for everything. And that's like the, the fast food of design. And what's interesting is to think about that messy middle of design. Like, what is the mom and pop shop for, of design that's been there for decades, that's reliable, that's not, you know, it's not going viral and winning awards, but it's good and you love it. I don't know. It's a hard, it's a hard, I don't know about enough design to like draw an analogy, but maybe you can, I don't know.

• 1:46:17Speaker 05

  Yeah, I think it's, it's really use case dependent, right, in some way. Like, you want a, you know, Tuesday night restaurant that is not all the bells and whistles and you want it to just deliver in some case, and maybe that's your to-do app where you keep your tasks for development, et cetera. Like, there is no reason for design to kind of get in the way like in, in those use cases. But then there's moments even in those flows where you want to feel something. You want to feel like that developer that thought about like the app had you in mind. And those are really the, the surprise moments, the delight moments that, that make people be loyal to an app.

• 1:46:53Jordi Hays?

  Yeah, something, something I've been thinking about is like, what will be the product design equivalent of the M dash? Or like when you're when you read, let's say, somebody like publishes an essay and then you start reading it and you get to the second paragraph and you just like immediately close it because you realize like they just fully generated all the text. I feel like we're gonna start to get that with software more and more where you'll go to a website or an app and from afar, or at least when you first land on it, it looks like, cool, this looks like a nice product. And then you start using it and you realize like, okay, like they generated a bunch of like nice animations and it like looks okay, but then the second that you actually start using it, you realize there was no real human thought

• 1:47:35Jordi Hays?

  put into the product. Like it is now, you can now make a product that looks like linear in one prompt. You cannot make a product that's gonna feel like using linear.

• 1:47:46Speaker 05

  Exactly.

• 1:47:48Jordi Hays?

  And so that's where that's where the human element is just gonna continue to be super, super powerful and that and that and taking user feedback and like having that empathy with the user and being super thoughtful and using the using the products to yourself and not just because, yeah, it's never been easier to create any type of application. It still feels like just as hard in many ways to create like a product that's truly magical to daily drive or rely on.

• 1:48:14John Coogan?

  Yeah, makes sense.

• 1:48:15Speaker 05

  Yeah, and I think AI will play a role into that. But actually, to go back, I am so pissed about M dashes.

• 1:48:22Jordi Hays?

  We lost a good one.

• 1:48:26Speaker 05

  And every time I write now, I use them and I'm like, whatever.

• 1:48:30John Coogan?

  Yeah, people are gonna accuse me of using AI. Yeah, I think you just have to use the minus sign.

• 1:48:37Speaker 05

  But yeah, I do think it's gonna be recognizable when, you know, a website is just like kind of vibe prompted, vibe coded and put out there in the world and you're going to wanna feel that the developers spend more time considering that.

• 1:48:54Jordi Hays?

  Yeah. Amazing. Well, thank you so much for for joining.

• 1:48:59John Coogan?

  Yeah.

• 1:49:00Jordi Hays?

  Congratulations on the new role as a Figma DAU for going on a decade now. I'm very, very happy that you're on board.

• 1:49:09Speaker 05

  Try out all the all the new toys.

• 1:49:11John Coogan?

  Yeah, thanks for stopping by. We'll talk to you soon. Have a great rest of your day.

• 1:49:18John Coogan?

  If you want AI to handle your customer support, go to fin.ai, the number one AI agent for customer service. Let's react to some of these Nano Banana prompts. They look fantastic. Here's one where someone took a map, a Google Maps screenshot, and just turned it into an RPG-style map, a San Francisco monster map. And it's, it really is that reasoning model. Like, you can see, the Golden Gate Bridge is there, and what would be logical to have attacking the Golden Gate Bridge? A giant octopus. And then Alcatraz Island is there, and there's this sea monster next to it. And everything kind of like fits. Like, you didn't get, the dragon is up at Twin Peaks. You don't get the dragon in the water,

• 1:50:01John Coogan?

  you get the sea monster in the water. And so all of these things are like, pretty logical. There's, of course, some things that are a little bit repetitive and stuff.

• 1:50:09Jordi Hays?

  Like ogres in Golden Gate?

• 1:50:11John Coogan?

  Yeah, it's just very, very cool. I think this is gonna be a lot of fun. Then, there's someone else with a benchmark here. Angel, it says Nano Banana nailed the burger test. It's the first model to truly do this perfectly. And so, the prompt is, remove the ingredients, leave just the top bun and the bottom bun in the exact same place, and render the rest of the image, just with a prompt. And previously, this would sort of confuse models a little bit here and there, because it would be, it would sort of shift the colors, or shift the sections, and kind of not, not, not...

• 1:50:52Jordi Hays?

  This next one's wild. The Make It Lego?

• 1:50:55John Coogan?

  The Lego prompt is crazy. This could be the next, like, Ghibli moment, for sure. If you can just take a whole bunch of photos, and pipe them through. We should take some of the R iconic photo-

• 1:51:04Jordi Hays?

  Ben, make a, take a picture of us and put it through Nano Banana Pro and Make It Lego.

• 1:51:08Speaker 03

  I did, I tried doing this earlier. It works pretty well, sometimes with the people. It doesn't use like, the mini-figure, like-

• 1:51:15John Coogan?

  Oh, it doesn't. Okay. So this is particularly good, because it's already, because, so this dog one is remarkable, but it's a cartoon character. And so, yeah, is it gonna make us a minifig? I mean, maybe that could be worked into the prompt. Do you want to take some of the iconic

• 1:51:31John Coogan?

  photos that we've used through the press images, through the Wall Street Journal photo? There were the New York Times photos. Let's take some of those photos that we have. And let's put those through Nano Banana and ask them to render us as mini figures in this Lego world. I want to see the Ultra Dome in 4K Lego. While we're doing that, Pietro Shirano shares that Nano Banana is wild, Nano Banana Pro, that is. Here's my favorite use case so far. Take papers or really long articles and turn them into a detailed whiteboard photo. It's basically the greatest compression algorithm in human history. This is a very cool video where attention is all you need gets turned into this image.

• 1:52:16John Coogan?

  I can already imagine, you know, when we got ChatGPT, it was like, oh, wow, you can take you can take bullet points, you can expand it into into an essay. And then you can take an essay and expand it down to bullet points. And I imagine that people are going to be sending these. And then they're not going to be reading them. And then they're going to be like, actually like

• 1:52:33Jordi Hays?

  Okay, turn this diagram into an essay and then summarize it.

• 1:52:36John Coogan?

  And then summarize it. Turn it into two words for me. Just one, just turn it into just one word. But it is very cool. And I'm excited where people will play with this. It is, it does look really good. I think we're going to play with this tomorrow. We have a diagram, a market map of our own coming tomorrow. We're going to break down the state of AI from the TBN perspective. Dee Dee shares that he literally fed Nano Banana Pro raw graph viz of AI compute commits, generated by Gemini 3, and it one shotted rendering it with logos perfectly. What in God's name is in this model? That is very, very cool. I've seen, so that's not quite at the level of

• 1:53:19John Coogan?

  the elegance that I've been seeing from the Wall Street Journal's visualization of all the circularity in the we've seen that circular graphic from the Wall Street Journal.

• 1:53:28Jordi Hays?

  But it's like 70% of the way there.

• 1:53:31John Coogan?

  Yeah, yeah, yeah, I would say it's 70%. I'm, I'm just excited that it actually puts, it gets the logos correctly because with the right direction, um, it can definitely do some. Also, I imagine that sketching a little bit of the ground truth of like how you want this laid out, um, would probably give it a lot of like scaffolding to build off of that would be very cool.

• 1:53:51Jordi Hays?

  Look at this.

• 1:53:52John Coogan?

  Okay, let's see this. Uh, let's zoom in on this.

• 1:53:54Jordi Hays?

  That's Lego us.

• 1:53:55John Coogan?

  We're not mini figures, we're just Lego'd Lego people. Okay. Color temperature's a little off, I'm not into it. Let's move on to the next one. The gong looks cool in the background, I like the Lego gong. Uh, have we, have we done any others?

• 1:54:09Jordi Hays?

  This is the only one I made so far. It, it's pretty slow.

• 1:54:11John Coogan?

  It's pretty slow. Well, um.

• 1:54:13Jordi Hays?

  Wow, this is funny, so on this, on this next one.

• 1:54:15John Coogan?

  Yeah, Gemini 3 Pro Image versus GPT.

• 1:54:18Jordi Hays?

  Okay, so I didn't read the caption, I just looked straight at the image, I just assumed that the image on the left was an actual image.

• 1:54:24John Coogan?

  Yes.

• 1:54:25Jordi Hays?

  And then this was the output on the right.

• 1:54:26John Coogan?

  That is crazy.

• 1:54:27Jordi Hays?

  Cause it looks terrible.

• 1:54:28John Coogan?

  Yes, yes, yes.

• 1:54:28Jordi Hays?

  But it's actually the, no, this is a real image.

• 1:54:31John Coogan?

  Same prompt, two different, uh, two different results, that's pretty, pretty remarkable.

• 1:54:37Jordi Hays?

  Crazy.

• 1:54:37John Coogan?

  Yeah, VO4 is gonna be a big, big moment, I'm very excited. Cause VO3, I mean, such a huge leap over, uh, the original Sora, was it, uh, it was pre-Sora. What was, what was ChatGPT's video model, not Dolly, but pre-Sora app? Was it called Sora?

• 1:54:56Jordi Hays?

  Uh, yeah, it was always called.

• 1:54:57John Coogan?

  It was always called Sora, okay, yeah, that, cause Sora 1 or whatever, the precursor, was really kind of hallucinatory and, and, and, and crazy. VO3 got a ton of the physics down, but it still has this sort of like plasticky look that you can just clock. Um, but, uh, whatever they did with Gemini 3 Pro Image, uh, is really pushing the photorealism much farther. Very exciting. What else is going on here? Uh, Nano Banana Pro, edit this image and face swap it with Sam Altman. Slow, show thinking, Nano Banana Pro. Does that, is that a good face swap for Sam?

• 1:55:33Jordi Hays?

  I think it's just okay.

• 1:55:34John Coogan?

  It's okay. That one's a, that one's, that one's a five out of 10, I think. Uh, let's see. Uh, someone is dropping ChatGPT and saying, I don't wanna play with you anymore.

• 1:55:45John Coogan?

  Wait, but this is bad.

• 1:55:46Speaker 03

  No, no, they're dropping Gemini.

• 1:55:47John Coogan?

  They're dropping Gemini and they're going back to. The model wars are really, really heating up constantly. Rota is saying that they're all in on GPT 5.1 Pro because it's rolling out to all pro users. It delivers clearer, more capable answers for complex work with strong gains in writing, help, data science and business.

• 1:56:07Speaker 03

  What is this?

• 1:56:07John Coogan?

  That's exciting.

• 1:56:08Speaker 03

  I made this one, but it only turned you into a Lego, Jordy.

• 1:56:12John Coogan?

  What did it do to me?

• 1:56:13Speaker 03

  Look at John.

• 1:56:14John Coogan?

  What did it do to me?

• 1:56:15Speaker 03

  Look at John.

• 1:56:16John Coogan?

  What did it do to me? I'm just a human. They miss me entirely. What's going on here?

• 1:56:21Speaker 03

  You got Lego Jordy.

• 1:56:22John Coogan?

  Lego Jordy. And what is it? It doesn't know what to do with the turbo puffer because the turbo puffer is like already Lego.

• 1:56:26Speaker 03

  Wait, I know what you are. You're already a Lego.

• 1:56:29John Coogan?

  That's funny. That's funny. Well, speaking of turbo puffer, sign up today. Serverless vector and full tech search built on first principles and object storage. Fast, 10x cheaper and extremely scalable.

• 1:56:42John Coogan?

  So GPT 5.1, have you had a chance to take its first spin, Tyler? What's the latest with GPT 5.1?

• 1:56:48Speaker 03

  Well, so the main new thing is the new model is, it's not 5.1 because that came out like, what, two weeks ago? It's 5.1 Codex.

• 1:56:55John Coogan?

  5.1 Pro. Well, so there's Codex, but then there's also 5.1 Pro for like research tasks, I believe.

• 1:57:01Speaker 03

  Okay.

• 1:57:01John Coogan?

  But anyway, Codex, how are we doing on the benchmarks? Oh, oh, and you have, you have a take. You have a take. Give me your take.

• 1:57:07Speaker 03

  Yeah, I mean, basically, so I think the main graph or the main kind of benchmark that everyone is now kind of watching.

• 1:57:14John Coogan?

  Yes.

• 1:57:14Speaker 03

  Is this one from Meter.

• 1:57:16John Coogan?

  Yeah, show us where the goalposts have been moved to most recently. Where, where, where do we move the goalposts most recently?

• 1:57:22Speaker 03

  We still need to get goalposts over there.

• 1:57:24John Coogan?

  We do need goalposts. Okay, so we moved the goalposts from, from like, you know, just surprise me with something that's remarkably human and that's.

• 1:57:34Speaker 03

  Totally qualitative. You can't measure that at all. And I think the reason that people are using this benchmark is because like you can't saturate it. It's not like it's, you just can keep measuring. It's not like MMLU, like there's math questions and then at some point you just answer them all correctly.

• 1:57:48John Coogan?

  Yes, yes.

• 1:57:48Speaker 03

  So it's not like interesting.

• 1:57:49John Coogan?

  Okay.

• 1:57:49Speaker 03

  Where, where this is like, it, this is a benchmark that you could keep doing in 30 years, right? Because it, the time just goes up and up.

• 1:57:57John Coogan?

  Yes, yes.

• 1:57:58Speaker 03

  So if we're going to pull up this graph, it's the time horizon one. And there's basically what you've seen for the past like five years is every eight months, the. The time that a model can do, and this is just on coding tasks, but it's kind of generally applicable.

• 1:58:14John Coogan?

  Yep.

• 1:58:14Speaker 03

  It doubles.

• 1:58:15John Coogan?

  Yep.

• 1:58:16Speaker 03

  And so I think this is kind of the main thing that we should be looking at, like, are models stagnating? Are they decelerating? And what you see is it's basically a straight line. If you put it on a log, it's exponential, but if you put it on a log scale, it's a straight line.

• 1:58:29John Coogan?

  Yep.

• 1:58:30Speaker 03

  And the new model is like perfectly, basically on that line.

• 1:58:33John Coogan?

  Yeah.

• 1:58:34Speaker 03

  So I think it's like, this is just a great sign, like.

• 1:58:37John Coogan?

  Okay, so how long, like, what time, what task duration, measured in time, would you, would you say qualifies as AGI?

• 1:58:51Speaker 03

  Yeah, I mean, I don't know if it's exactly, I don't know if that, if that's my definition of AGI, because I think there are a lot of tasks that take a long time, but don't really require general intelligence.

• 1:59:04Speaker 03

  But I do think if you're getting into like weeks or months, that's like a big kind of project that would take a person. It's like a big part of their life.

• 1:59:11John Coogan?

  Yes.

• 1:59:11Speaker 03

  I think if we get up to there, and I guess, I mean, you can just chart it out to see if you follow this path.

• 1:59:17John Coogan?

  Yeah.

• 1:59:17Speaker 03

  How long would that take?

• 1:59:20Speaker 03

  But I mean, you said it's like basically a human lifetime, isn't that your?

• 1:59:24John Coogan?

  That's my, I think that's my correct benchmark.

• 1:59:27Speaker 03

  But I think that's wrong, because if you think of like, build a, build a company, that's not your entire lifetime. That's like, for some people, that's only like.

• 1:59:35John Coogan?

  Thirty years.

• 1:59:36Speaker 03

  Thirty years, okay.

• 1:59:37John Coogan?

  Four years, maybe, yeah, I don't know.

• 1:59:41John Coogan?

  The initiation prompt is the genesis prompt, it's be fruitful and multiply. Like, that is the AGI initialization prompt. Just, just replicate, you know? That's your goal, AGI. Just, just go create value, just go exist. And then it goes and does whatever it needs to.

• 2:00:01John Coogan?

  That's when it's like truly like, you know, embodied, I suppose. I don't know.

• 2:00:07John Coogan?

  All I do know is that you can go to profound, try profound.com, get your brand mentioned in ChatGPT, reach millions of consumers who are using AI to discover new products and brands. I guess the question is, this, this task duration thing is so odd, because like, does time move slower or faster in AI world? You would assume.

• 2:00:29Speaker 03

  Alex Karp manipulated time.

• 2:00:31John Coogan?

  You should be able to manipulate time if you're in the computer, right? So. You know what I mean, right?

• 2:00:37Speaker 03

  Uh.

• 2:00:38John Coogan?

  Okay, so, so what I'm saying is that like, is it like, uh, if, if GPT-5 can do two hours of, i- if it can work for two hours without losing consistency and still complete long tasks, um, if you get a new chip that speeds that up, you do the same amount of work in half the time. Like, if you just actually speed up the inference, you're, you're bringing this curve down. And so you have this weird countervailing force where, um, like, I would expect a computer to be able to do problems faster than humans, right?

• 2:01:15Speaker 03

  Uh, yeah, I mean, so, so-

• 2:01:16John Coogan?

  Oh, at least, at least over time.

• 2:01:17Speaker 03

  Compared to, uh, like, how long it takes a human to do it, right?

• 2:01:20John Coogan?

  Oh, is that what this is?

• 2:01:21Speaker 03

  It's like thirty, thirty seconds-

• 2:01:22John Coogan?

  It's a two-hour project?

• 2:01:24Speaker 03

  Is answer a question, six, like, five minutes is count words in a passage.

• 2:01:27John Coogan?

  Huh.

• 2:01:28Speaker 03

  Find fact on web. Uh, yeah, it's like, compared to how long it takes a human.

• 2:01:33John Coogan?

  Interesting. Okay, so they have to-

• 2:01:35Speaker 03

  Because obviously you could just, like, if you-

• 2:01:37John Coogan?

  How are they going to, how are they gonna benchmark? I, I, I always thought this was like, cu- come up with a prompt. Like, do they even have a prompt that can, that theoretically could take months to do? And, and, and it wouldn't just be sleeping?

• 2:01:51Speaker 03

  Build a, build a co- massive company that takes months, years.

• 2:01:54John Coogan?

  Yeah, years. Is that where we're gonna be with this meter chart in, in like, what? Six more doublings or something like that?

• 2:02:01Speaker 03

  Yes.

• 2:02:02John Coogan?

  That's what they're, yeah?

• 2:02:03Speaker 03

  I mean-

• 2:02:04John Coogan?

  You think so?

• 2:02:04Speaker 03

  I don't rea-

• 2:02:05John Coogan?

  That seems-

• 2:02:05Speaker 03

  That would, would it be comparable to, the, the time scales?

• 2:02:08John Coogan?

  Yeah. I just, I just, I wonder how they're, how they're mapping that.

• 2:02:13Jordi Hays?

  Just ten more doublings, sir.

• 2:02:16John Coogan?

  Yeah. I mean, it certainly does seem like, like, good progress. And, I mean, everyone, I, I, I feel it very much in the sense of like, um, just the, the amount of work that a single prompt can kick off just feels like it's doubling for sure.

• 2:02:29Speaker 03

  Yeah, I, I think this is just a good benchmark. A lot of people, i- it's getting harder and harder to find good prompts-

• 2:02:34John Coogan?

  Yeah.

• 2:02:34Speaker 03

  That show a model is like, actually better.

• 2:02:37John Coogan?

  Yeah.

• 2:02:38Speaker 03

  Um, and this is like a very kind of like, objective thing that there's a, there's a, you know, what we expect it should be, where it actually is, and it actually is where we expect it to be. So this is like a good model.

• 2:02:48John Coogan?

  Yeah.

• 2:02:48Speaker 03

  Like, this is, we're on track.

• 2:02:50John Coogan?

  Well, if you're looking for sales tax AGI, head over to numeralhq.com. Let numeral worry about sales tax and VAT compliance for you.

• 2:02:59Jordi Hays?

  Well said, John.

• 2:03:00John Coogan?

  Meter says, what is my purpose? All you put new A bio models on the graph. Meter. Oh, my God. Guys, please, I need to see Sonnet 4.5 on this. So Sonnet's not on there. And this is

• 2:03:12Jordi Hays?

  Well, no.

• 2:03:13John Coogan?

  Update the graph. There seems to be, there seems to be a mistake. I planned on assessing risks from automated AI R and D. That's funny. They're having fun. What else is going on in, in, in AI world? These things are looking smoothly exponential for AI over the past several years, and I continue to think this is the best default assumption until the AI R and D automation feedback loop eventually speeds everything up. I, we gotta have the meter folks back on the show and, and, and understand this a little bit further. I, I really, I really wonder how they're actually developing the prompts.

• 2:03:48Jordi Hays?

  I really want to get their take on protein, the amount of protein in, in fast casual concepts.

• 2:03:55John Coogan?

  That would be great, too.

• 2:03:56Jordi Hays?

  And potentially get them to chart that out as well.

• 2:03:59John Coogan?

  Okay. And then someone, someone took this chart and put it next to the AI 2027 graph. Is this correct? So there's meters data, GPT-5, Codex Max. It looks exponential, but not super exponential. Is that what this read is?

• 2:04:14Speaker 03

  Uh, yes. Yeah. So this is still on the log graph. You see the blue line is the meter.

• 2:04:18John Coogan?

  Okay.

• 2:04:18Speaker 03

  And then the green is the AI.

• 2:04:21John Coogan?

  So AI 2027 was, was, was expecting like even more of an exponential.

• 2:04:26Speaker 03

  Yeah. And I think that's mostly because they thought agents that would help develop the next AI would come a little bit sooner.

• 2:04:32John Coogan?

  Okay.

• 2:04:33Speaker 03

  Um, but I, I think, I think Gemini 3 seems to do very well in the kind of computer use stuff, which you should imagine should like greatly help out kind of agents.

• 2:04:44John Coogan?

  Yeah.

• 2:04:45Speaker 03

  So. Maybe there's a month or two ahead.

• 2:04:49John Coogan?

  Oh, so you think we're going back to the green dots there? You're optimistic?

• 2:04:53Speaker 03

  I mean, it's reasonable.

• 2:04:55John Coogan?

  You think we might jump from one line to the other? From the linear to the super linear? Or super exponential? From the exponential to the super exponential? Daniel says, yep, things are going somewhat slower than AI 2027 scenario. Our timelines were longer than 2027 when we published, and now they are still a bit longer still. Around 2030, lots of uncertainty, though, is what I say these days. Meter, of course, is evaluating GPT-5.1 Codex-Max triggered drastic AI acceleration or autonomously replicate. They concluded this was unlikely. Survey said unlikely. But obviously, big growth in the capabilities. Sweetbench, I wanted your reaction to this from vowels.ai.

• 2:05:41John Coogan?

  Different evaluation, different eval. But with this company, they say Gemini 3 is number one on the independent Sweetbench leaderboard.

• 2:05:50Speaker 03

  Yes, this is their own Sweetbench. It's also, they did not test the, actually, the newest OpenAI model. It's not Codex-Max. X-Hi or whatever, like the maxed out. I don't know, yeah, they named it kind of poorly, but.

• 2:06:06John Coogan?

  For the 10th time.

• 2:06:08Speaker 03

  But yeah, I mean, I'm curious where that'll end up. And also, people are saying, there's like rumors of Gemini 3 Flash, the small model. There's also rumors of Anthropic releasing a model soon, and I assume it would be Opus 4.5. Right? Because that's like, they have like three tiers. They have the Haiku, Sonnet, and Opus. So I'm very curious to see where all those end up.

• 2:06:34Jordi Hays?

  Very curious to see where Doug landed with his.

• 2:06:39John Coogan?

  Oh, we didn't ask him this, did we?

• 2:06:40Jordi Hays?

  He got.

• 2:06:41John Coogan?

  He tried it.

• 2:06:42Jordi Hays?

  The 100 gram max protein bowls from Sweetgreen yesterday.

• 2:06:46John Coogan?

  He got three of them. For the core team.

• 2:06:48Jordi Hays?

  The core research.

• 2:06:49John Coogan?

  The core team. No, no, no, he didn't do three. But wow, look at this. It's really on there. Chicken, chicken, chicken, chicken, chicken, chicken, chicken. It's so insane.

• 2:06:57Jordi Hays?

  So funny.

• 2:06:57John Coogan?

  So if things go badly.

• 2:06:58Jordi Hays?

  It's listed out.

• 2:06:59John Coogan?

  If things go, if things go well, we can big bulk on Nvidia. Little update, I ate over, a little over half and my tummy hurts. What is this? It's not good. Protein Max is tummy hurting. Protein Max is the, is the semi-analysis of.

• 2:07:14Jordi Hays?

  Phil Ehrenstein said yesterday, you're telling me the CEO of Sweetgreen is on TBPN?

• 2:07:19John Coogan?

  Oh, yeah.

• 2:07:20Jordi Hays?

  And that he's a Chad with slicked back hair, a golden tan and a sick leather jacket. And his handle is Johnny Nemo and he's a vibes guy disregarding surveys and he added seed oil free 106 gram protein bowl. Sweet green, Sweetgreen won me. So. Uh, I love it. Uh, Doug spoke a little too soon yesterday. He said, I survived the great bear market of October 29th to November 19.

• 2:07:45John Coogan?

  Uh, let me tell you about public.com, investing for those who take it seriously. Multi-asset investing trusted by millions.

• 2:07:52John Coogan?

  Nvidia emerges successful. Um, and yet the market is still selling off. Uh, Nvidia saw its shadow, six more months of bull markets as high yield hairy, although who knows, the market is tanking still. Uh, the NASDAQ is down 2.1% now and Bitcoin is down at $86,000.

• 2:08:10Jordi Hays?

  They're 5% today.

• 2:08:12John Coogan?

  Significant sell off.

• 2:08:13Jordi Hays?

  Let's check in on the sailor himself.

• 2:08:17Jordi Hays?

  Also down 5%, but at least he's tracking.

• 2:08:19John Coogan?

  Meltem.

• 2:08:20Jordi Hays?

  The underlying asset.

• 2:08:21John Coogan?

  Uh, Meltem says, uh, Nvidia earnings call first 60 seconds. We have line of sight to half a trillion in revenue in 2026. The bubble hasn't even started yet. Let's go.

• 2:08:32Jordi Hays?

  Uh, Michael Burry's still going incredibly hard.

• 2:08:35John Coogan?

  This is the circularity chart that I was calling out is like.

• 2:08:38Jordi Hays?

  And I think that Nano Banana could pretty much one-shot this.

• 2:08:43John Coogan?

  I don't know if it would be as, as overlapped and nuanced and like the, it's not.

• 2:08:49Jordi Hays?

  You can't just say make it more overlapped.

• 2:08:52John Coogan?

  Uh, no, I, I don't think you can yet. I mean, this is, we're, I mean, we're really, really.

• 2:08:56Jordi Hays?

  We gotta talk to somebody that has been making graphics for media companies like this.

• 2:09:02John Coogan?

  Yeah, it seems, it seems so easy, but if you, especially.

• 2:09:05Jordi Hays?

  It takes a lot of, it takes a lot of.

• 2:09:07John Coogan?

  Yeah. And especially if it's like.

• 2:09:10Jordi Hays?

  It takes a lot of deep thinking and reasoning.

• 2:09:12John Coogan?

  No one has, yeah.

• 2:09:13Jordi Hays?

  It takes so much deep thinking and reasoning and model could never do this.

• 2:09:17John Coogan?

  No, they will be able to. They will be able to.

• 2:09:20Jordi Hays?

  But Michael Burry says, "Every company listed below has suspicious revenue recognition. The actual chart with all the give and take deals would be unreadable. The future will regard this a picture of fraud, not a flywheel. True end damage is ridicu-

• 2:09:33John Coogan?

  No, no, no.

• 2:09:34Jordi Hays?

  Demand. True end demand is ridiculously small. Almost all customers are funded by their dealers. If you can name OpenAI's auditor in one hour, you win some pride."

• 2:09:45John Coogan?

  What do you, what does he mean, "true end demand is ridiculously small"? That's just not true. Like, there are tons of companies that are paying for subscriptions for all sorts of AI products. And, I don't know, I-

• 2:09:57Jordi Hays?

  He's a D cell with a crazy P doom.

• 2:09:59John Coogan?

  He is. He's a D cell with a zero P doom, I guess. I don't know. I mean-

• 2:10:05Jordi Hays?

  No, I do think he, he has, uh...

• 2:10:10Jordi Hays?

  Yes, if you're looking at the amount of investment happening now-

• 2:10:13John Coogan?

  Yeah.

• 2:10:13Jordi Hays?

  -in comparison to the demand-

• 2:10:15John Coogan?

  Yeah.

• 2:10:16Jordi Hays?

  And you don't believe that the products will get better at all-

• 2:10:18John Coogan?

  Yeah.

• 2:10:19Jordi Hays?

  If you don't believe that-

• 2:10:20John Coogan?

  It just flipped so much. Like, there was a moment where it was like, wow. Like, demand for this new thing went from zero to $10 billion in just a few years. This is remarkable.

• 2:10:30Jordi Hays?

  Yeah.

• 2:10:30John Coogan?

  And then people were like, "Let's invest a trillion dollars in that." And it's like, okay, well, at that price, it's like, it's actually kind of crazy. I don't know. It's a lot to deal with.

• 2:10:38Jordi Hays?

  Yeah, but if you think about any industry on Earth-

• 2:10:41John Coogan?

  Yeah, yeah.

• 2:10:42Jordi Hays?

  Do we think every industry on Earth will be using 50 to 100 times more tokens within five years, 10 years? Shrap- you don't even have to be that much of a-

• 2:10:54John Coogan?

  No.

• 2:10:55Jordi Hays?

  Of a, of a-

• 2:10:56John Coogan?

  Of a permabull.

• 2:10:57Jordi Hays?

  Of a permabull to believe that.

• 2:10:59John Coogan?

  Yeah.

• 2:11:00Jordi Hays?

  In fact, it's like hard, hard to argue.

• 2:11:02John Coogan?

  Tyler's permabulling. He's never lost sight at any moment. He's always been long. I love it. Let's read this from A Capital. "But first, let me tell you about Vanta, automated compliance and security with the leading AI trust management platform." So A Capital says, of course, that's your contention. Of course, this is... Do you know what movie this is from, Jordy? Pop quiz, hot shot?

• 2:11:24Jordi Hays?

  Uh...

• 2:11:25John Coogan?

  Do you know what movie that's from?

• 2:11:27Jordi Hays?

  I do not.

• 2:11:27John Coogan?

  You've got two quizzes.

• 2:11:28Jordi Hays?

  This one is...

• 2:11:29John Coogan?

  Do you know what this is from?

• 2:11:30Jordi Hays?

  Uh, no.

• 2:11:31John Coogan?

  No? Good Will Hunting?

• 2:11:33Jordi Hays?

  That's it.

• 2:11:33John Coogan?

  That's the Good Will Hunting image. You don't know that. Do you know what pop quiz hot shot is from?

• 2:11:39Jordi Hays?

  No.

• 2:11:39John Coogan?

  That's from Speed. It's issued to Keanu Reeves. He gets on the phone with a pop quiz hot shot.

• 2:11:45Jordi Hays?

  Speed worth seeing?

• 2:11:46John Coogan?

  Speed is definitely worth seeing. It's a crazy... It's a great movie. It's just a thriller. They crash a bus. It's a great, great, great movie.

• 2:11:52Jordi Hays?

  I'm in. I'm in.

• 2:11:54John Coogan?

  Back to the Good Will Hunting image meme that... I love this format. It's very fun. It's a very fun way to illustrate and like kind of tell a whole story. And so A Capital says, of course, that is your contention. You're a first-year AI skeptic. You just finished reading Andrew Ross Sorkin's 1929, and now you think you're reliving the roaring twenties with GPUs. You will cling to that until next month when you hear Jim Chang talk about unsustainable CapEx, and then you will start parroting that the entire AI ecosystem is about to collapse under the weight of its own spending. That will last until someone posts a CoreWeave CDS chart, and you'll repeat that too without realizing that it was just dealers hedging credit portfolios,

• 2:12:34John Coogan?

  not some cosmic warning sign. Then you'll probably start lecturing people about global crossing because you heard someone say 1999 fiber bubble, and it made you feel informed. Meanwhile, Nvidia just printed one of the biggest sequential growth quarters the sector has ever seen and guided higher again. The workloads are real. The demand is real, and the CapEx is already contractually locked. None of that came from a cash crash narrative paperback or a Chang soundbite. But sure, keep borrowing other people's takes and pretending they're your own. One day, you might look at the actual numbers and realize this is not a bubble. It is the early, the early stage of the largest infrastructure build-out in decades.

• 2:13:17John Coogan?

  Love it. Very fun. Let me tell you about Figma. Think bigger, build faster. Figma helps design and develop teams, build great products together. Oh, Sunday Robotics is coming on the show.

• 2:13:29Jordi Hays?

  Very shortly.

• 2:13:30John Coogan?

  We will get... Maybe we should...

• 2:13:32Jordi Hays?

  We should play the video now.

• 2:13:33John Coogan?

  Let's play the video now.

• 2:13:34Jordi Hays?

  Little teaser. Pull it up.

• 2:13:36John Coogan?

  I understand.

• 2:13:38John Coogan?

  Very cool.

• 2:13:39Jordi Hays?

  So, this is kind of a combination of the R2-D2 form factor with a humanoid.

• 2:13:43John Coogan?

  Yes. Look at that. Picking up two wine glasses is insane.

• 2:13:48Jordi Hays?

  I love the way it just bounces around.

• 2:13:50John Coogan?

  So, this is sped up, presumably?

• 2:13:51Jordi Hays?

  Yes. I think it's at like a 10x speed.

• 2:13:54John Coogan?

  Something about the lighting here looks CGI to me.

• 2:13:58John Coogan?

  I know it's not, but it looks CGI-ish.

• 2:14:04John Coogan?

  I'm fascinated. So many questions. Says it's in autonomous mode.

• 2:14:08Jordi Hays?

  Sunday has motion.

• 2:14:11John Coogan?

  Sunday has motion. I think that the I think that the design here is fantastic. I will have to debate it and you have to tell me what you think. But definitely beating the creepy uncanny valley, in my opinion, doesn't feel like, oh, that thing is about to pick up a knife. At least to me. I'm pretty into this design. And I think the internet was as well, since it got over a million views and over 3,000 likes.

• 2:14:42John Coogan?

  And Sholto Douglas over at Anthropic says, "This is insanely impressive." And I agree, it is.

• 2:14:48Jordi Hays?

  I'm excited to ask Tony how much they had to spend to get to this point.

• 2:14:53John Coogan?

  That would be interesting.

• 2:14:54Jordi Hays?

  Because I'm assuming it will be quite a bit less than many of the other players that are kind of competing here.

• 2:15:01John Coogan?

  The little telescoping pole is very cool.

• 2:15:04Jordi Hays?

  Taylor says, "It's the hat, not threatening lid." I agree. Just throw a cool hat.

• 2:15:10John Coogan?

  The hat looks... Yeah. So, Scott, I think the hat does look kind of dumb, but that's like, kind of okay. I'd rather it look dumb than scary or menacing or weird, you know?

• 2:15:24Jordi Hays?

  Think about how scary like some of these humanoid robots would be to a one-year-old.

• 2:15:28John Coogan?

  Like Wally looks kind of dumb. R2-D2 looks kind of dumb, but it's still like a friendly, you know? You don't want it to be overpowering.

• 2:15:34Jordi Hays?

  Like the Optimus or figure would be traumatizing to a one-year-old.

• 2:15:39John Coogan?

  Yeah, yeah, for sure. Well, before we move on to our next post, let me tell you about Julius AI, the AI data analyst that works for you. Join millions who use Julius to connect their data and ask questions and get insights in seconds. Andrew Reid says, "Every deal is a special situation if you're enthusiastic enough." Very funny. Let's move on. SAM3 video tracking is so good. Yesterday, collect data, train custom object detector, use tracker to estimate object motion, days, now track anything with a text prompt in seconds. Who put out SAM3? Is that Google as well?

• 2:16:16Jordi Hays?

  That's Meta.

• 2:16:17John Coogan?

  That's Meta?

• 2:16:18Jordi Hays?

  Whoa.

• 2:16:18John Coogan?

  Oh, this is Segment Anything?

• 2:16:20Jordi Hays?

  Yeah.

• 2:16:20John Coogan?

  Oh, okay, okay, okay. Got it. Okay. Wait, why is it on Google Research then? That's funny. Oh, it's how to segment videos with Segment Anything SAM3, and they just happen to be hosting this in a Google Colab notebook. That makes sense, because Meta does not have a Google Colab competitor that I'm aware of. Interesting. Well, that's very exciting.

• 2:16:41Jordi Hays?

  Very cool.

• 2:16:42John Coogan?

  You can track the, all of our gong hits potentially for velocity and understand velocity relative to the audio volume.

• 2:16:52Jordi Hays?

  If we had a live-

• 2:16:53John Coogan?

  And you can understand how the production team is doing their job to lower the levels for you.

• 2:16:58Jordi Hays?

  To blow your ears out. But if we had a live, like, speed tracker?

• 2:17:02John Coogan?

  Yes.

• 2:17:03Jordi Hays?

  Like, as you're swinging it?

• 2:17:05John Coogan?

  Predicted speed, yeah.

• 2:17:06Jordi Hays?

  Very cool.

• 2:17:06John Coogan?

  We could do it. Diet Coke tracker as well. We could automate all of this.

• 2:17:11Jordi Hays?

  Sheil has some great coverage. Grok says Elon is more fit than LeBron and would win a fight against Mike Tyson.

• 2:17:18John Coogan?

  Fact check, true. You're absolutely right.

• 2:17:21Jordi Hays?

  I'm gonna ask Grok if this is true.

• 2:17:23John Coogan?

  Is this true?

• 2:17:23Jordi Hays?

  Grok, is this true?

• 2:17:24John Coogan?

  Did somebody do that? I bet somebody did that in the, in the replies here. Is this true? Yes, very funny. I wonder how much of this is like in the pre-prompt or just in the X data set, you know? Elon's obviously, like, there's just, there's just an incredible amount of Elon fans in the X ecosystem still, since a lot of people that weren't Elon fans left. But even the Tesla bulls don't glaze to this level usually. Like, I, so I don't know where this would come from. This must have been in the pre-prompt or something. But it's a very silly- It's a very silly.

• 2:18:02Jordi Hays?

  Many people are doing this. I mean, you you went in Sora and you said depict me as a bodybuilder.

• 2:18:07John Coogan?

  Yes, that's true.

• 2:18:08Jordi Hays?

  Somebody tried to hack it to to give you small legs.

• 2:18:11John Coogan?

  They did successfully prompt engineer me.

• 2:18:14Jordi Hays?

  They got you good.

• 2:18:16John Coogan?

  They did. They did get me. But yes, I mean, I feel like I feel like at this point, like, we're we're past we're past this level of like novelty being relevant in a purchasing decision for an LLM. In fact, it might work against you, especially in light of Gemini three, very benchmark driven. They put out the model card. They there were a bunch of demos that went out. There were some clear examples of next next value coming from the model.

• 2:18:44Jordi Hays?

  Yep.

• 2:18:45John Coogan?

  Sort of a by the book launch and

• 2:18:50John Coogan?

  unclear how much this helps the Grok brand to have something like this.

• 2:18:54Jordi Hays?

  Yeah.

• 2:18:55John Coogan?

  Leak out, but certainly funny.

• 2:18:57Jordi Hays?

  Kevin Wheel, friend of the show, says, today we say hello world from OpenAI for Science. We're releasing a paper across 13 examples of GPT-5 accelerating scientific research across math, physics, biology and material science. In four of these examples, GPT-5 helped find proofs of previously unsolved problems. A lot of a lot of this type of posting has been heavily contentious in the past. But they are continuing to share their work.

• 2:19:28John Coogan?

  Yeah, and I think that this stuff will eventually be, you know, fully, you know, peer reviewed and and also there's just this interesting dynamic where like the other labs, they they won't really let you get away with anything like they'll fact check you so fast.

• 2:19:48Jordi Hays?

  Yeah.

• 2:19:49John Coogan?

  If this is if this is seriously impressive, like you'll probably see some congrats from.

• 2:19:54Jordi Hays?

  There's also a dynamic where if you were using ChatGPT to accelerate your own research.

• 2:20:03Jordi Hays?

  Are you gonna, are, is everybody gonna stand up and yell, hey, I use ChatGPT for this? Or are they gonna be like, my research?

• 2:20:10John Coogan?

  Yeah.

• 2:20:11Jordi Hays?

  You know, who, you know, I'm not sure that a lot of people that are leveraging the tool are gonna be quick to give OpenAI credit or an AI credit for, for something that they're working through.

• 2:20:23John Coogan?

  Yeah, but if they're doing it something in some sort of controlled environment, go after some specific problem. Before we move on, let me tell you about Privy. Privy makes it easy to build on crypto rails, securely spend of white label wallets, sign transactions, and integrate on chain infrastructure all through one simple API. Bern Hobart says, prescient New Yorker cartoon that saw prediction markets coming more than half a century in advance. Wow.

• 2:20:44Jordi Hays?

  June 27, 1970.

• 2:20:46John Coogan?

  If you can't see this, it's the arrivals at an airport and there's flights that are arriving from Chicago, Detroit, Philadelphia, Pittsburgh. They depart at 8:00 AM, they arrive at 10:20 AM, and then there's odds listed there because of course you'll want to bet on when the plane lands. And now you can maybe, you're, you're close to being able to with the, with the prediction markets on their relentless march to take over the world. Also, before we bring in our next guest, we have to talk about group chats in ChatGPT. We mentioned this earlier. It's official. They're rolling out globally. There was a successful pilot with early testers. Group chats

• 2:21:29John Coogan?

  will now be available for all logged in users on ChatGPT free, go, plus, and pro plans. I didn't know there's a ChatGPT go plan. We gotta figure out what that is.

• 2:21:37Jordi Hays?

  That was the India plan?

• 2:21:38John Coogan?

  Okay, that's interesting. And then also, I mean, it just says, of course, it's rolling out to, to the everyone because this one doesn't set the GPUs on fire. This is good old fashioned, stuff the text in the database and reduce churn in your product. So makes a ton of sense. Very, you know, we'll have to test this out and see if it's actually that useful.

• 2:22:07Jordi Hays?

  Yeah, I think this, I mean, this is the kind of thing that can help. OpenAI build more of a moat outside of a of a brand and just general distribution moat.

• 2:22:17John Coogan?

  ChatGPT is turning into a social app. Sam pulled it off. Before Zuck could make the Meta AI app good enough to compete compete with ChatGPT, says Yuchin Jin. Ax and Grok could have a real chance to do it too, but it's rough watching DMs and chat keep breaking. And this is from all the way back in February. CNBC said Meta plans to release a standalone Meta AI app in an effort to compete with OpenAI's ChatGPT. And Sam Allman said, OK, fine, maybe we'll do a social app. And he did. He did Sora, and now he's adding social features to ChatGPT Core. I am interested to see like how we I I send.

• 2:22:58John Coogan?

  If I do a deep research report, I'll send it around to people in the organization here at TVPN every once in a while. I'm wondering how much it makes sense to keep the chat running in ChatGPT. So they certainly do get value out of like putting together the queries, sharing that, sharing the whole theory. I I don't know how much I'll be a DAU of this in a month. I'll I'll need to test it out. But we have our next guest in the Rooster Room waiting room. Tarek from Stute here in the studio.

• 2:23:29Jordi Hays?

  It's time to Stute.

• 2:23:31John Coogan?

  How are you doing?

• 2:23:32Jordi Hays?

  We're saying it correctly, right?

• 2:23:33John Coogan?

  We, yes, I mean, first off, we have to say we love the brand because we love Day Job.

• 2:23:38Jordi Hays?

  Legends.

• 2:23:39John Coogan?

  Thank you for supporting them and.

• 2:23:41Jordi Hays?

  They did our brand.

• 2:23:42John Coogan?

  Excellent, excellent taste in branding agencies, of course.

• 2:23:45Jordi Hays?

  The best.

• 2:23:46John Coogan?

  But please introduce yourself and introduce the business as well.

• 2:23:49Speaker 06

  Hi, I'm Tarek Alaruri, and you got it right. It is Stute. It's actually played rugby after college.

• 2:23:55John Coogan?

  Oh, cool.

• 2:23:56Speaker 06

  And it it means prop in South African.

• 2:23:58Jordi Hays?

  OK, after college, that means you were on the on the pro track or?

• 2:24:03Speaker 06

  No, I was I was like this guy who wanted to drink some beers every now and then.

• 2:24:06Jordi Hays?

  OK, let's go. Let's give it up for those guys.

• 2:24:09Speaker 06

  But anyway.

• 2:24:10Jordi Hays?

  Underrated guys.

• 2:24:11Speaker 06

  Thank you. I'm I'm here to announce our Series A led by Andreessen Horowitz for $29.5 million.

• 2:24:18John Coogan?

  Boom.

• 2:24:22Speaker 06

  Also participating is Activin and Coastal Ventures. There we go. There we go.

• 2:24:27John Coogan?

  So we actually saw a preview of this, of this brand design when we were hanging out with the Day Job folks. And what I thought was interesting was the positioning of how AI comes through in the messaging to the customer. So maybe let's start with like the problem, the solution, what you're actually building. And then how you message it to, you know, an audience of investors or what you're building. But then also how you message it to the actual end consumer who might not care that much about the particular technologies that you're using.

• 2:25:02Speaker 06

  Yeah, and I think there's a lot of slop in AI or thrown around with brands. And that's why we use Day Job similar to you. You know, our customers, just for context, what we do is we help with accounts receivable. So if you don't know what that is, we help collect what you sold. And most of our customers are the kind of like PerkinElmer, Bishop Lifting, organizations you might not know, but they're flyover states, kind of where I'm from, which is Indiana. And what we do is we use AI as a platform and we help customers collect 40% of their overdue invoices in the first six months of using our tool. And it's not like a traditional software where, you know, they're promising you more seats, more people. We're live in three days for large Fortune 100 companies, which is crazy.

• 2:25:47Speaker 06

  Most people don't believe us, but then they start seeing the results. And, you know, most of these people we work with, you know, they have a nine to five. You know, they're not a startup hustler, they're not grinding, they're not, you know, working in New York on Wall Street. They want to go see their kid's game. And so we plug in at five to nine, so you can punch out and go to that game. And that's really the, you know, what we're about here at Stu.

• 2:26:10John Coogan?

  And then on the messaging side, do you feel like your customers want to know details about the technologies that you're implementing? Do they care about that?

• 2:26:22Speaker 06

  Well, you have everybody claiming AI. Like, I'm not Matthew McConaughey. You know, I got a brush on Matthew McConaughey.

• 2:26:28John Coogan?

  Yeah.

• 2:26:28Speaker 06

  And I have to compete with an AI technology versus Matthew McConaughey. It's almost impossible. Uh, and so, you know, our branding reflects our customers.

• 2:26:36John Coogan?

  Sure.

• 2:26:37Speaker 06

  Uh, you know, think Clippy, which Day Job did a great job with.

• 2:26:40John Coogan?

  Yeah, totally.

• 2:26:41Speaker 06

  And really helping, yeah, helping them be nostalgic, but it's like what software first promised you. It was going to automate things.

• 2:26:47John Coogan?

  Yeah.

• 2:26:47Speaker 06

  But instead, forty years later, you know, you need professional services, consultants. It just doesn't do the job.

• 2:26:52John Coogan?

  What's the best, uh, business model for this type of business these days? Consumption-based, seat-based?

• 2:26:59Jordi Hays?

  Success-based.

• 2:27:00John Coogan?

  Success-based, percentage-based?

• 2:27:02Speaker 06

  It's almost like you talk to the team at Day Job to team me up with these questions.

• 2:27:05Jordi Hays?

  Yeah, we did. We actually did.

• 2:27:07John Coogan?

  Uh, yeah, I actually didn't talk about the business model with them.

• 2:27:09Speaker 06

  You all, I mean, you all buy software. It's so confusing. I'm not a smart man, and I go on and they got multiple spreadsheets. You got a various version. You got like a pricing guide, my head spinning. We just charge a monthly fee like you would a coworker. The average accounts receivable person in the United States has paid sixty thousand dollars without benefits, and it takes three to four months to hire them. We can plug in the next day for a fraction of the cost.

• 2:27:34John Coogan?

  Sure.

• 2:27:35Jordi Hays?

  Uh, what, what, how, how does it, uh, like, how, how is the actual, like, product design work? Is this like an agent that gets integrated into communication channels? Like, what, what does it actually look like?

• 2:27:50Speaker 06

  Yeah, I'm glad you asked. Uh, so we do audio, so we'll actually place AI phone calls. Uh, well, we'll do emails. We'll, you know, even do SMS and WhatsApp in different areas of the world. So, you know, the way I always tell customers is we have two forms of communication, which is like outbound, hey, you need to pay me, or inbound. If you're a bigger customer and somebody calls you and you work in the finance team and you're like, hey, I got a question about invoice one, two, three, four. It's pretty hard. You have to pull up multiple systems. You have to answer questions. AI is great to live behind that IV archery and just answer it immediately. On the flip side, if we reach out to a customer and they might have like their generic invoice template that goes out,

• 2:28:33Speaker 06

  they'll have a question, hey, where do I send the check to? AI can instantly reply without a human being and even sit in the flow of funds where we'll send them a payment.

• 2:28:41John Coogan?

  Yeah, I want to, I want to dive deeper into your, what I think is a hot take about basically sticking with a seat based pricing model. Alex Karp was on the show and he was saying like, in the future, all companies will be paid on the value they deliver. And I'm just wondering what the difference is if you wind up going to a company that has a thousand times as many invoices, you collect a thousand times as many payments, you deliver a thousand times as much value, should you not get paid at least a little bit more?

• 2:29:17Speaker 06

  Well, I mean, Alex is an amazing entrepreneur and they're an established brand. Hopefully someday, if we keep winning each day and executing, we'll be where Palantir is. You know, right now, our customers, when we talk to them, you have companies that have been around since 97 saying they're AI now. And so, you know, we have to differentiate ourselves. And one of the ways we differentiate ourselves is with something very simple, very easy. It's like if you went to Chipotle for the first time, you line up, you get a burrito, you're like, wow, this is amazing. Back in the day, not anymore. So we, we want to make things as simple as our customers.

• 2:29:53Jordi Hays?

  Most brutal fall off of all time.

• 2:29:56Speaker 06

  I mean, look at me, it's brutal, right? I love Chipotle. But the, the tough part is a lot of the stuff our customers are looking at isn't simple. And they're looking at evaluating multiple days of presentations. They're getting grilled by salespeople. You know, we want to get in, demonstrate value and see a really quick ROI with these customers. And that's what we're helping them achieve. So great example is one of our customers, Bishop Lifting, reduced their invoices by 35% past due and have been able to free up that cash flow for other things. It could be like the holidays around, bonus time. You know, and they have these people across America in locations and AR is not or receivables

• 2:30:39Speaker 06

  isn't their first job. And so being able to offload that and get them a little more money in their pocket is something we try to achieve for our customers.

• 2:30:46John Coogan?

  Yeah, that makes sense.

• 2:30:47Jordi Hays?

  That's amazing. Well, I gotta say the chat absolutely loves you.

• 2:30:50John Coogan?

  They love you.

• 2:30:50Jordi Hays?

  This guy is great. This guy is a man. Midwest sensibilities in Manhattan is extraordinarily powerful. He's even drinking yerba mate. What a freaking legend. I'm liking the sound of this dude.

• 2:31:03John Coogan?

  Stute. We, we, let's give it up.

• 2:31:06Jordi Hays?

  No, I love it. I mean, I, I love an idea that, uh, that when you hear it, it's just totally obvious. It's like applying, you know, the same, there's like the capital war happening and like, uh, customer experience right now. They're using a lot of the same technology. You're applying it in a very, uh, clear way, uh, in a different part of the org. And, uh, I'm bullish. So thank you so much for joining.

• 2:31:28Speaker 06

  Really appreciate you having us on and keep crushing it on the show.

• 2:31:31Jordi Hays?

  Have fun out there. We'll see you back for the B.

• 2:31:33John Coogan?

  Yeah, we'll talk to you soon.

• 2:31:34Jordi Hays?

  Cheers.

• 2:31:35John Coogan?

  Have a good day. Let me tell you about adquick.com. Out of home, Matt, out of home advertising made easy and measurable. If you're launching a new company, growing, get on adquick.com, get some billboards. Uh, our next guest is in the restream waiting room. Let's bring them in to the TVP and I'll be done.

• 2:31:52Jordi Hays?

  There he is.

• 2:31:53John Coogan?

  We have Nikita from Flexion. Is also a day job? Is this a?

• 2:31:58Jordi Hays?

  No, no, no. We got, this is Tony.

• 2:31:59John Coogan?

  Oh, Tony. Hey, sorry.

• 2:32:01Jordi Hays?

  We got, we got mixed around. Uh, Tony, so great to have you on the show. Uh, I'm sure your 24 hours, last 24 hours have been absolutely crazy. We, uh, played your demo on the show earlier today and, uh, we're absolutely blown away. It's, it's really, uh, tremendous progress and we're, we're excited to meet you. So before we talk Sunday, uh, would love a intro on yourself, background and all that good stuff.

• 2:32:26Tony Zhao?

  Yeah, yeah, absolutely. So excited to be here. So before that, I was actually a PhD student at Stanford, working on robotics. So some of the works are like Aloha, you saw like the two robot arms clamped to a table. And it's not just about the hardware, but how it learns. Uh, how can we learn from human demonstrations? How can we learn to reinforce learning and all these things? And I think last, early 2024 is when I have the realization that like, you know, pumping out more papers and doing more research may not be the most direct way to push robotics forward, but starting a company and working on real product is.

• 2:33:07Tony Zhao?

  So this is why I co-founded this company, Sunday, with Cheng, uh, who is also a PhD student at Stanford. And uh, which leads to Memo, uh, Act One, and all these like, uh, new advances.

• 2:33:19Jordi Hays?

  Uh, incredible. What, uh, what has it taken to get to this demo that you released yesterday? Because our, our, uh, I have no idea how much money you've raised, uh, up until this point, but it feels like you guys have accomplished a ton, uh, in a pretty, uh, resource constrained way, at least compared to companies that you're competing with in the sort of like, helpful humanoid in the home category.

• 2:33:45Tony Zhao?

  Yeah, absolutely. So we're, we're functioning in a very efficient way. And I think as an early stage company, we think about it as a blessing that forces us to innovate. And finding out like these solutions are orders of magnitude more efficient than like 20% efficient and 30% efficient. Um, and I think a big part of it is also about like the culture and the team and all the people we have that are like really experts and really believes in what we're doing. Um, yeah.

• 2:34:20Jordi Hays?

  John?

• 2:34:21John Coogan?

  Yeah, I'd love to know some of the trade-offs.

• 2:34:23Jordi Hays?

  Also, the chat is mentioning you forgot to mention you worked at DeepMind, Tesla and Google. So, sort of a non-traditional background into robotics startup.

• 2:34:33John Coogan?

  Yeah, what are the key trade-offs? I mean, there's a lot of focus right now on tele-operation. Is it something, just a step in the path towards full autonomy? There are obviously some folks that are jumping straight to full autonomy and they say, well, we never use tele-operation at all. Other folks who say tele-operation is a really useful tool to pull forward some of the capability. Where do you stand on the issue?

• 2:35:02Tony Zhao?

  Yeah, so I think tele-operation is a really powerful research tool, but it's not necessarily the best tool to get to a product. Because if you think about robotics and, you know, put that right next to autonomous driving, right? Tesla has millions of cars collecting data for them every single day. And it still took almost a decade to kind of see light at the end of the tunnel that things are starting to work very well. In robotics, if the only thing we can rely on is tele-operation to gather the amount of training data, it will take like decades for sure. Because robotics is a harder problem than self-driving.

• 2:35:38John Coogan?

  Yeah.

• 2:35:38Tony Zhao?

  So, the way we think about it is that, how can we use human data to train the model? We have like 8 billion humans in the world. If you can use like 1% of that, that's already huge.

• 2:35:51John Coogan?

  Yeah.

• 2:35:51Tony Zhao?

  So, what we designed instead is, I actually have it here, is called Skill Transfer Glove, Skill Capture Glove.

• 2:35:58John Coogan?

  Yeah.

• 2:35:58Tony Zhao?

  That is one-to-one to Memo's head.

• 2:36:00John Coogan?

  Oh, interesting.

• 2:36:01Tony Zhao?

  And, yes, the idea here is that if you can wear the glove and do a task, Memo can also do it.

• 2:36:07John Coogan?

  Okay.

• 2:36:08Tony Zhao?

  And that essentially decoupled this whole, like, you need a robot to be deployed in the wild before you can gather the data to train the AI. We can train AI just by having people wear our glove and collect data.

• 2:36:20John Coogan?

  Yes, but, I mean, just to go back to the question of like capital intensivity, 1% of 8 billion people, that's 80 million gloves. If the glove costs even 10 bucks, we're back in, you know, you need a billion dollars to get your data set or something like that.

• 2:36:37Jordi Hays?

  You don't think Tony can raise the bill?

• 2:36:39John Coogan?

  I'm not saying you can't do it, I'm just saying like, is there a smoother path here? How many gloves have you shipped? Is there a scale thing? And then also I'd be interested to know about like transfer learning. Are you having luck with simulation? Are you having luck with, there's a lot of video, just content out there of people doing tasks? Is there any signal that you can pull from just a YouTube video of someone doing the dishes? Or do you need to simulate something in Unreal Engine or use a world model? Like what are the other tools in the tool chest?

• 2:37:12Tony Zhao?

  Yeah, I think robotics has got a point that there are so many of these ideas that we haven't converged to this like one single thing, which is like pre-training and post-training for LMS. And the way we think about it is that out of all these methods, some will be better than others. And as a startup, we should focus on that one thing that we believe in and build the best system and stack around it. And what we chose was using human data, like using gloves to gather data. And actually for all the models that we saw, we of course pre-trained on like internet scale data. But all the specific behaviors are learned only from the gloves that we make. We don't do teleoperation, we don't do simulation, and we don't have world models.

• 2:37:57John Coogan?

  Whoa, okay.

• 2:37:59Jordi Hays?

  Very opinionated.

• 2:38:00John Coogan?

  Then how do you see the data capture from the glove scaling? Like, do you think that there will be 80 million people in five years using this to create more training data? Or do you think it's a little bit more tractable of a problem where at a certain point, okay, yeah, it's been a big operation, but it's more like 10,000 people that you're employing or something like that?

• 2:38:25Tony Zhao?

  Yes.

• 2:38:28Tony Zhao?

  I think this question is more about like for us, how can this data generate value? So that we can keep this loop going now, right? And it's kind of similar to the whole large language model space that we need to spend a lot of money into compute.

• 2:38:41Jordi Hays?

  Yeah.

• 2:38:41Tony Zhao?

  But the model itself is generating like tremendous amount of value. And for us, we don't need to solve robotics to ship a product. That's the lucky part.

• 2:38:50Jordi Hays?

  Sure.

• 2:38:50Tony Zhao?

  And in the homes, there are lots of like, it's one of the few places you can do relatively simple tasks.

• 2:38:56Jordi Hays?

  Yeah.

• 2:38:56Tony Zhao?

  But give people a huge amount of value, both emotionally and functionally.

• 2:39:00Jordi Hays?

  Yeah. And it's much more low stakes tasks than self-driving, right? So self-driving, you said it's a harder, you said that uh home robotics is a harder problem earlier? If I heard that right, but at least, yeah, but but but at least it's lower stakes. And that if you have an air, if you drop a dish, it's annoying and you want to avoid that. But there's not like, nobody's gonna like die. Yeah.

• 2:39:23Tony Zhao?

  Yes. It's like the newer start of the prob, like of the company is to solve robotics.

• 2:39:28Jordi Hays?

  Yeah.

• 2:39:29Tony Zhao?

  But we don't need to solve robotics before we ship a product.

• 2:39:31Jordi Hays?

  Yeah. So yeah, talk about timelines.

• 2:39:35Tony Zhao?

  Yeah. So we've been around for a year and a half. And our next milestone is the beta program that will run late 2026. That is when we'll put memo and like tons of them into people's home. And actually see how people interact with the robot. And what do people want from the robot? Um, and the general availability of memo will be either 2027 or 2028. Depending on the progress we made uh through the whole beta program.

• 2:40:06Jordi Hays?

  Yeah. Uh, talk about form factor. Why not, uh, why not give it legs? I'm sure you have a, a, a reason for that and I'm curious. Uh, because I think, I think people's immediate question is, okay, I can see how a wheeled system, it makes a lot more sense in a lot of ways, but what happens if I have stairs?

• 2:40:26Tony Zhao?

  Yeah, absolutely. So the way we designed this robot is to put safety at a really high priority. And the way we define safety is we call it passively safe. That if the robot arm and torso is fully stretched out. And at that point, you cut the power of the robot. Can it stay stable or not?

• 2:40:47Jordi Hays?

  Interesting.

• 2:40:48Tony Zhao?

  A weird robot is actually like one of the few ways.

• 2:40:51John Coogan?

  So it can't fall over and crush your dog or even your foot, basically.

• 2:40:55Jordi Hays?

  Yeah, right. Or just like damage the floor.

• 2:40:58John Coogan?

  Or all sorts of stuff. That makes a ton of sense. And then also I imagine that there's you can just have more battery power or maybe dock easily. And there just aren't that many tasks that required. I feel like every demo is, I mean, the wine glass demo is remarkable. Holding two wine glasses is hard as a human, let alone as a robot with kind of odd fingers. But just the tidying up use case is potentially underrated. Because that feels like that feels right around the corner. Even if the like dealing with all the racks and different spoons and knives and wine glasses. Doing the full dishwasher feels a little bit harder.

• 2:41:38John Coogan?

  But there's a willingness to pay, at least for me, just to go around the house and pick up the ball that needs to be in the toy basket and pick up the shirt that's on the floor. Like, that's valuable. That is actually value if you can get the price right.

• 2:41:52Jordi Hays?

  What was the, what was the like key design inspirations? What matters to you with design? Somebody in the chat was asking if you were influenced by Homestar Runner or what.

• 2:42:01John Coogan?

  Oh, yeah, it is sort of Homestar Runner. That's hilarious.

• 2:42:05Tony Zhao?

  Yeah, so the way we think about design is we kind of think backwards of what do we want the world to be like? If the robots are ubiquitous, if you need to see it like every single day, what should it look like? And we lean quite heavily towards building a robot that is friendly, but also functional. And these two things, there's actually a small overlap in between them. So when we design the robots, one.

• 2:42:31Tony Zhao?

  I think detail that we decide to do is we do not put camera into the robot's eyes. The camera is actually right underneath its hat.

• 2:42:40Jordi Hays?

  Yeah, I saw that.

• 2:42:42Tony Zhao?

  Yes. So the reason is that like, you're going to make eye contact with the robot. You're going to like, look at his face. But if you look at someone's face and his eyes, you see like a camera watching you. It's a little bit creepy.

• 2:42:54Jordi Hays?

  Oh, interesting.

• 2:42:54Tony Zhao?

  So we kind of intentionally avoided that. Um, and yeah.

• 2:43:00John Coogan?

  That's very interesting. Yeah. The, yeah, the design, I, we were talking about it earlier. It feels good. Uh, it really, uh, just it avoided like the uncanny valley, the creepiness. Like, there's a lot of risk factors when you're designing humanoid robotics right now. We've seen all sorts of them. Uh, they can look cool, sci-fi, but maybe weirder in certain contexts. I think this one came across very well.

• 2:43:20Jordi Hays?

  Well, super, super excited for you. Thanks for coming on and breaking it down.

• 2:43:23John Coogan?

  This is really fun. Thank you so much.

• 2:43:24Jordi Hays?

  If, we'd love to be in the, in the demo program.

• 2:43:27John Coogan?

  Yeah, we got flat floors here.

• 2:43:29Jordi Hays?

  We got flat floors. We have, uh, huge messes.

• 2:43:32Tony Zhao?

  Bye.

• 2:43:33John Coogan?

  And we have a team of people that we will make wear these gloves all day long.

• 2:43:36Jordi Hays?

  And we will take care of, we will take care of Memo.

• 2:43:39John Coogan?

  We will.

• 2:43:40Jordi Hays?

  Because he's, because he's cute.

• 2:43:41John Coogan?

  Yes.

• 2:43:42Jordi Hays?

  And we want to see him win, so.

• 2:43:44John Coogan?

  Thank you so much.

• 2:43:44Jordi Hays?

  Congrats on all the progress.

• 2:43:45John Coogan?

  Congratulations.

• 2:43:46Tony Zhao?

  Thank you, guys.

• 2:43:46John Coogan?

  We'll talk to you soon. Goodbye. Uh, you know what we gotta do? We gotta get Memo a watch on getbezel.com.

• 2:43:55Jordi Hays?

  Yes, let's get Memo a hitter.

• 2:43:56John Coogan?

  Shop over 26,000, ice it out. Luxury watch.

• 2:44:00Jordi Hays?

  Iced out.

• 2:44:00John Coogan?

  Fully authenticated.

• 2:44:01Jordi Hays?

  RM.

• 2:44:02John Coogan?

  By Bezel's team of experts.

• 2:44:04Jordi Hays?

  Uh, he needs an RM ASAP. And some Chrome Hearts.

• 2:44:07John Coogan?

  Definitely needs a Richard Mill. Why not? Why not?

• 2:44:10Jordi Hays?

  Uh, next up, we got Nikita from Flexion. Excited for this one.

• 2:44:15John Coogan?

  Thank you so much for taking.

• 2:44:16Jordi Hays?

  What's going on?

• 2:44:16John Coogan?

  Time to talk to us today. Thanks for waiting. Good to meet you. How you doing?

• 2:44:21Speaker 08

  Hi, really excited to be here.

• 2:44:23John Coogan?

  Thanks so much.

• 2:44:24Speaker 08

  I'm Hidda, I'm the CEO and co-founder of Flexion.

• 2:44:27John Coogan?

  Cool.

• 2:44:28Speaker 08

  Where we're building the intelligence layer, or the brain that powers all kinds of robots, from humanoid to mobile manipulators.

• 2:44:35John Coogan?

  Yeah, I mean fantastic, we were just talking about humanoid robotics. How do you see the market playing out?

• 2:44:41Jordi Hays?

  Okay, so you, hopefully you caught at least the end of our conversation with Sunday Robotics. But something that I was thinking about, like a real challenge, is when Sunday gets good enough at picking up, manipulating objects. What happens if Sunday walks up to our table here after the show? We typically have lunch, and Sunday needs to figure out what's trash, and what's, like, what should be taken and thrown away, and what's actually should just stay there, right? Because that's actually like somewhat of a, like, it requires some memory. It's like, okay, this is an item that, that is, it needs to be able to identify objects, figure out what is like, what is something that is worthy of just throwing away? What is something that, like, I don't want thrown away, and if it gets thrown away,

• 2:45:26Jordi Hays?

  I'll be frustrated? So, I feel like there's like a lot deeper, more levels of complexity to a lot of these robotic tasks, than, than just like object manipulation, and kind of understanding, understanding the general environment, and really having like intelligence around the environments that it operates. And I feel like that might be something that you're solving, but tell me if I'm wrong or correct.

• 2:45:50Speaker 08

  Absolutely. Let me just say that Sunday's amazing. I think their videos are really, really impressive, probably the most impressive demo I've seen so far.

• 2:46:00Speaker 08

  Just start, let's just start with that. I don't know if you're doing it on purpose, but it's a great reference to the video we released this morning, where we have a robot walking around and picking up trash, and bringing it to a garbage can.

• 2:46:12John Coogan?

  Yeah.

• 2:46:13Speaker 08

  And, and the way we're doing that is actually splitting the problem into two parts. The first one has nothing to do with robotics, it's about common sense and understanding. And for that part, we don't really need to train a specific model ourselves, because the knowledge is already contained in large language models. Think of it as GPT-5, or all of these models. If you take a picture of that table in front of you, and you ask GPT, what is garbage, what is not, those models are already really good at understanding that.

• 2:46:45Speaker 08

  And once you have that, then the next part is actually the object manipulation, which we're also solving in a slightly different way compared to Sunday. We bet that the vast majority of data needed to train those models will come from simulation. Ah, great. You know, can have a look at that video.

• 2:47:03Jordi Hays?

  Uh, yeah. So, so say more or maybe even just narrate the video.

• 2:47:08Speaker 08

  Sure. So let me just quickly come back. We're bad on simulations. We train robots using reinforcement learning, not to humiliate humans, but to solve specific tasks and just through trial and error. So we have robots trying millions and millions of times and it's tens or hundreds of years of simulated data. And then they come up with very specific ways on how to walk across complex trains, but also use their whole body to manipulate objects.

• 2:47:34John Coogan?

  But isn't there a little bit of a problem? Yeah, yeah. Isn't there a little bit of a problem there where to perfectly simulate that forest path requires incredible, you know, just like CGI, just to, I mean, you need like Unreal Engine crank to max on every physics calculation because yes, you can model it like a video game, like it's all just one smooth surface, but it's not actually that. In reality, there's tons of different blades of grass. There might be slightly more friction over here on this blade of grass versus that one. You have to simulate all of that to actually recreate the real world. Is there not a gap?

• 2:48:15Speaker 08

  Yeah, absolutely. That's a great point. Usually we call it the simturiel gap.

• 2:48:19John Coogan?

  Yes.

• 2:48:19Speaker 08

  And once you train in simulation, the whole challenge is to cross that simturiel gap.

• 2:48:25Speaker 08

  For example, here in this video, everything is trained in simulation. And we were actually not even thinking about forests or mountains where we were training the robot. So you don't need to simulate every single piece of grass or every single rock. As long as you train on general enough scenarios that somewhat encompass what is happening here, then you can deploy the robot. And the other thing is that we're not training our policies directly from RGB camera inputs. Otherwise, you would actually need to simulate exactly how a forest looks. So we're doing some processing on top. Once again, using some other models, but we're actually trained on internet scale data.

• 2:49:03John Coogan?

  Okay.

• 2:49:03Speaker 08

  A good example, I think, is if you want to train a robot to open a door.

• 2:49:06John Coogan?

  Yeah.

• 2:49:07Speaker 08

  Um, either you have to simulate every single possible door that exists in the world with all the textures, the lighting, et cetera.

• 2:49:13John Coogan?

  Sure.

• 2:49:13Speaker 08

  Or you can use a model like segment anything.

• 2:49:16John Coogan?

  Sure.

• 2:49:17Speaker 08

  And then you paint the door in red and the handle in, let's say, green. Then all doors kind of start to look the same.

• 2:49:24John Coogan?

  Look the same. Interesting. And then you're basically training the motion against the segment anything version of the door of the world.

• 2:49:33Speaker 08

  Yeah, something like that.

• 2:49:34John Coogan?

  Okay, yeah. Uh, what technologies are you most excited about across these generative world models, uh, these Gaussian splats, uh, just Unreal Engine getting better, like traditional 3D workflows, uh, Houdini and Cinema 4D? Uh, are, of those tools, which ones will be most useful to you, uh, in the future? Or is everything kind of bespoke in its own world for you?

• 2:49:59Speaker 08

  All of this is super important. It's all about the timeframe.

• 2:50:01John Coogan?

  Hmm.

• 2:50:02Speaker 08

  So today, we're using physics-based simulators, um, just like Unreal Engine.

• 2:50:06John Coogan?

  Hmm.

• 2:50:08Speaker 08

  And that's, actually, my take is this is enough for way more than what most people think. We can go a long way with Jordi simulators.

• 2:50:16John Coogan?

  And, and, and, and explain, explain that. Is it, is it that, uh, if you have a physics simulation that's running fine, and let's say Unreal Engine, you might use something else, but Unreal Engine, is, do you think we're on a scaling curve where if you had a million GPUs, running a million instances of, of Unreal Engine generating simulated data, that that would actually result in better progress on the, on the robotics side, on the actual decision-making and planning side?

• 2:50:45Speaker 08

  Yeah, exactly. That mixed with one more thing, which is generative models that can create assets for simulation.

• 2:50:51John Coogan?

  Okay.

• 2:50:52Speaker 08

  You don't need to have humans coming up with a million different versions of all the things that a robot needs to interact with.

• 2:50:59John Coogan?

  So previously, that was programmatically, like to try and get to something with a varied world like that, where there's, you know, a little hill over here and a rock out of place that the robot might trip over, you would have to do all that programmatically, maybe through some node-based workflow in Houdini, or just kind of, or just inject just randomness, just random number generators, and then rotate this rock over here, change the geometry, et cetera, et cetera. But you're saying that generative AI can create even more variation, is that the idea?

• 2:51:33Speaker 08

  Yeah, exactly. You actually have two ways to add more variation easily. One is something like Gaussian splats, where you go outside, you collect real-world data, and suddenly you have a lot of assets. And the second version is you ask GPT-1 to do it for you.

• 2:51:49John Coogan?

  Yeah, yeah, that makes sense. That's cool.

• 2:51:52Jordi Hays?

  Any news?

• 2:51:53John Coogan?

  Yeah, what you got?

• 2:51:54Speaker 08

  Yeah, we announced this morning that for the first time, they will raise 50 million in just a few months.

• 2:52:01John Coogan?

  Congratulations.

• 2:52:03Jordi Hays?

  Who participated?

• 2:52:05Speaker 08

  A bunch of investors, DST Global, Nvidia, Unventures, Participated, ProSos, First, Windfire.

• 2:52:15Jordi Hays?

  Awesome. And then where are you building the company? You're in Europe, or have you moved over to the West Coast?

• 2:52:24Speaker 08

  So right now, we're all in Zurich, in Switzerland. This is why you see the robot walking in our nice Alps.

• 2:52:31John Coogan?

  Oh, yeah.

• 2:52:32Speaker 08

  But actually, right now, I'm in San Francisco right now for a few days, and I'm here to find the right team to start a second office here.

• 2:52:40John Coogan?

  Oh, nice. That's great. Well, good luck.

• 2:52:43Jordi Hays?

  If I were you, I wouldn't, it'd be tough to leave.

• 2:52:44John Coogan?

  Zurich's pretty nice.

• 2:52:46Jordi Hays?

  Switzerland, completely. Second favorite country in the world for me, after America. So hopefully, next time, next time I'm in Switzerland, I'll definitely, I would love to stop by the office and meet.

• 2:52:59Speaker 08

  Yes, please.

• 2:53:00Jordi Hays?

  But congratulations on the milestone, super exciting. And if you ever have hot takes on robotics, feel free to let us know.

• 2:53:07John Coogan?

  I'll buy it.

• 2:53:08Speaker 08

  Thanks, thanks so much.

• 2:53:10Jordi Hays?

  Awesome. Cheers.

• 2:53:14John Coogan?

  If you're planning to go to the Alps, book a wander with inspiring views, hotel great amenities, dreamy beds, top tier cleaning and 24/7 concierge service.

• 2:53:23Jordi Hays?

  Transition.

• 2:53:23John Coogan?

  To vacation home, but better. Did you see this, that we don't understand how ice is, why ice is slippery? I don't know if this is fully confirmed at this point, but Massimo Rainmaker 1973 shares, "New research," so I misspelled that, "new research shows ice is slippery because of electrical charges, not pressure and friction. For almost 200 years, the prevailing explanation for ice's slipperiness was that friction or pressure from a skate boot or tire melted a microscopic film of water on the surface, creating a lubricating layer. A new study from Sarland University has overturned that longstanding idea." Boris Power here says, who's the head of applied research at OpenAI, says, "Wow,

• 2:54:08John Coogan?

  this is one of the bigger firm beliefs I held that got overturned." I really, I like, it's like, this is the one thing I knew was true.

• 2:54:19Jordi Hays?

  This was one of my strongest beliefs.

• 2:54:20John Coogan?

  I knew that the world is round, the sun rises in the east and it sets in the west, and I know that the reason ice is slippery is because when you step on it, there's a layer of water. But it is a good point. If it's actually about electrical charges, then it begs the question, which he's asking, I wonder how long until we get non-slip shoes for ice? So you could have a shoe that has a battery in there, creates some sort of electrical field that cancels out the electrical field or something? Maybe that does something? I don't know.

• 2:54:46Jordi Hays?

  Ice is brilliantly humbling, you know? You think you're walking, you're confident, you know? You're like, "I'm handling this ice," and then you just, and then, and then suddenly it feels like you got a banana under your foot.

• 2:54:59John Coogan?

  One of my friends was worried about getting canceled because the first tweet he ever posted, like decades ago, and when he first got on Twitter was, "I just slipped on some ice. #Fice." Like, F-U-C-K-I-I-C-E. And he was like, "Am I being rude or in goose? Should I delete that post?"

• 2:55:21Jordi Hays?

  Uh, Jackson Doll pulled out 12 lessons from our interview on uh dialectic. His podcast, uh, I don't think we'd ever written down a lot of these ideas. I think we've certainly talked.

• 2:55:35John Coogan?

  We were talking about the need for principles and the need for uh some sort of uh, you know.

• 2:55:41Jordi Hays?

  But that was more like operating principles within the company. Some of these are relevant.

• 2:55:46John Coogan?

  Relevant. But yeah, this is more about the, the style of content.

• 2:55:50Jordi Hays?

  Good summary. Can't copy compounding.

• 2:55:52John Coogan?

  If you want to know more about us and how we think about the show behind the scenes, you can go listen to Jackson Doll's latest episode with.

• 2:55:59Jordi Hays?

  That was really fun.

• 2:55:59John Coogan?

  Other than yours truly.

• 2:56:01Jordi Hays?

  On our own very set.

• 2:56:02John Coogan?

  On our own very set. Yeah, we filmed it here. Uh, the dialectic pod.

• 2:56:06Jordi Hays?

  Uh, Aiden says, just so we're clear, anti-gravity is a windsurf rapper. Windsurf is a VS code rapper. VS code is an electron wrapper. Electron is a chromium wrapper. Chromium is a C++ wrapper. C++ is a C wrapper. C is an assembly wrapper. Assembly is a machine code wrapper. Machine code is a binary wrapper. Binary is a physics wrapper. Physics is a math wrapper. Math is a logic wrapper. Logic is a philosophy wrapper. Philosophy is a humans wrapper. Humans are a carbon wrapper. Carbon is a star forged matter wrapper. Stars are a gravity wrapper. Gravity is definitely not an anti-gravity wrapper.

• 2:56:42John Coogan?

  19K. Like, people enjoy it. Uh, this is very funny. Everything's a rapper.

• 2:56:48Jordi Hays?

  This is funny. Uh, Robin Hood, uh, had a post, trade the forecast, weather market predictions. And Augustus says, ha ha. This is how, this is how uh, uh, Augustus can, can, uh, one way that he can monetize is just, you know, getting a hedge fund betting on the weather outcomes.

• 2:57:05John Coogan?

  It's not gonna rain. I'd like to say it's not gonna rain with size.

• 2:57:10John Coogan?

  Everyone's like, is it, where, where is Augustus? Is he in town? Is he betting on it? Uh, he would be, I don't know. Would that be in, uh, would that be um investigative? Would that be insider trading? We'll have to figure it out. Uh, Sachin Adella had a banger. Um, uh, barely AI says, never forget Sachin Adella in 1993 as a Microsoft Technical Marketing Manager, showing how Excel works. We can play this clip. This is funny.

• 2:57:40Speaker 09

  As you can see, the most important architectural requirement for this piece is to be able to integrate data which exists on a host or a mainframe right now into Excel. Excel being our front-end tool and the AS400 in our case being the data repository. So what I'm gonna do now is exit out of this environment and show you how we can better integrate this data into Excel. And I'll go ahead and-

• 2:58:07John Coogan?

  Call in questions now. He's doing a live stream, basically. I mean, it's on TV, but-

• 2:58:12Speaker 09

  At this point, what it did was it talked to the MS Query, went ahead and talked to the DRDA driver and went and connected to the mainframe, brought down the relevant data and populated my sheet here with the relevant data. Going to, using Windows NTFS and a server, connecting to the database-

• 2:58:30John Coogan?

  It sounds like a Gentle AI. Sounds like a workflow that's getting automated-

• 2:58:35Jordi Hays?

  Superintelligence for numbers.

• 2:58:36John Coogan?

  This guy's been automating workflows since day one. Now he says he has less hair, but the same love for Excel. And he's posted a photo making Sheet happen since 1985. He's looking great. He's on top of the world.

• 2:58:53John Coogan?

  Suno, Raise More Money. We had the founder on the show, like, just a week or two before the round, so we didn't have him back on. But, congrats to everyone over there. And there's, the regulatory deals are getting worked out.

• 2:59:09Jordi Hays?

  Yeah. So, a company called Clay is the first music AI service to reach a deal with all three major record labels, Universal, Sony, and Warner Music. Clay plans to announce its agreements in the coming days. I guess they're kind of front-running him there. Clay is building a product that will offer the features of a streaming service like Spotify amplified by AI technology that will let users remake songs in different styles. I knew a founder that was working on this exact service and ultimately thought that it would be impossible to get all these deals done. So I'm glad that somebody persisted and built this product, because I think it's gonna be pretty fun to play around with.

• 2:59:51Jordi Hays?

  Clay apparently has licensed the rights to thousands of hit songs so that it can train its LLM. The company has positioned itself as a friend of the industry. Kind of letting a fox into the hen house, maybe. Offering assurances that the artists and labels will have some control over how their work is used. Clay is led by music producer Ari Adi and also employs former executives from Sony Music and Google's DeepMind. And anyways, so this, I'm excited to play around with the product when it comes out.

• 3:00:24John Coogan?

  We should close out with 8sleep.com. Exceptional sleep without exception. Fall asleep faster, sleep deeper and wake up energized. And I want you to tell me, which Ferrari do you like? Because we finally have the Ferrari bench results in a Ferrari in Minecraft. This one's from GPT 5.1 Pro with the same prompt as if we scroll down, we can see what Gemini 3 Pro did. Which one do you think is better? Which do you think is more Ferrari?

• 3:00:55Jordi Hays?

  I mean, the Minecraft Ferrari Gemini 3 actually looks something like a Ferrari.

• 3:01:02John Coogan?

  I think I like the Gemini 3 one, too.

• 3:01:04Jordi Hays?

  GPT 5.1 Pro doesn't look anything like much of anything.

• 3:01:09John Coogan?

  It got red. It's missing just like, with the Gemini 3 Pro, you can see, what I like about is, you see that little yellow dot on the hood? It's clearly like, that's where the Ferrari logo goes on an actual Ferrari. And it knew to put that in there. It's just a little bit more, the wing is a little more articulated and opinionated, it feels like. It feels like it's more disconnected from the overall structure. But still, an interesting challenge, and I'm very excited to see where this benchmark goes because it is, it's just so visual. It's so tangible. Like, okay, I understand what this should look like, and it really illustrates all the hallucinations. Anything else you want to close out with?

• 3:01:52Jordi Hays?

  I'll close out by saying it's pouring rain so hard that I'm hearing it through, through our earbuds.

• 3:01:59John Coogan?

  Okay.

• 3:01:59Jordi Hays?

  So if you are in LA.

• 3:02:01John Coogan?

  Yeah, yeah. Be safe.

• 3:02:02Jordi Hays?

  Be safe out there. Wherever you are in the world, we love you. Thank you for tuning in with us today. We will be back tomorrow for a Friday show. We got Saagar coming on.

• 3:02:13John Coogan?

  It's gonna be a hard one.

• 3:02:14Jordi Hays?

  I'm sure he, I'm sure he'll have fully 180 on AI.

• 3:02:18John Coogan?

  Yeah, our biggest AI bull. We got semi-analysis and then breaking points.

• 3:02:23Jordi Hays?

  We're trying to bring you diverse perspectives.

• 3:02:25John Coogan?

  We really are.

• 3:02:27Jordi Hays?

  We really do care about that. We don't want this to be an echo chamber. We obviously have strong views ourselves on many topics, but we're here to bring, to have real conversations. So.

• 3:02:39John Coogan?

  Thanks, everyone, for tuning in.

• 3:02:40Jordi Hays?

  Thank you for tuning in.

• 3:02:40John Coogan?

  Thanks for dealing with the chaos in the chat.

• 3:02:43Jordi Hays?

  Yeah, very chaotic day in the chat. That was our first time being raided.

• 3:02:48John Coogan?

  Yeah, we got raided a little bit.

• 3:02:50Jordi Hays?

  It was really funny because they made like seemingly like 20 fake accounts.

• 3:02:55John Coogan?

  We had 20 accounts.

• 3:02:55Jordi Hays?

  And then they were really angry at Ben.

• 3:02:59John Coogan?

  Yeah.

• 3:02:59Jordi Hays?

  For some reason.

• 3:03:00John Coogan?

  And then took the whole brunt.

• 3:03:01Jordi Hays?

  Which was funny. And then they also were really angry at Merkor. They kept dunking on Merkor.

• 3:03:06John Coogan?

  Oh, yeah. Why are they mad at Merkor? Okay. It was very distracting. I had to wind up turning off chat. But thank you to everyone who stayed the course, stuck with us for the show and made it through while the chat was getting wild. But we appreciate you all and we will see you tomorrow.

• 3:03:20Jordi Hays?

  Love you.

• 3:03:21John Coogan?

  Goodbye.

• 3:03:21Jordi Hays?

  Cheers.