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

Episode Date: November 20, 2025

(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.comFigma - https://figma.comVanta - https://vanta.comLinear - https://linear.appEight Sleep - https://eightsleep.com/tbpnWander - https://wander.com/tbpnPublic - https://public.comAdQuick - https://adquick.comBezel - https://getbezel.com Numeral - https://www.numeralhq.comPolymarket - https://polymarket.comAttio - https://attio.com/tbpnFin - https://fin.ai/tbpnGraphite - https://graphite.devRestream - https://restream.ioProfound - https://tryprofound.comJulius AI - https://julius.aiturbopuffer - https://turbopuffer.comfal - https://fal.aiPrivy - https://www.privy.ioCognition - https://cognition.aiGemini - https://gemini.google.comFollow TBPN: https://TBPN.comhttps://x.com/tbpnhttps://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231https://podcasts.apple.com/us/podcast/technology-brothers/id1772360235https://www.youtube.com/@TBPNLive

You're watching TBPN. Today is Thursday, November 20th, 2025. We are live from the TVPN Ultronome, 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 the job numbers came back very positive. We are back. We are back. 19,000 new jobs. And Nvidia beat earnings. They, they, 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%. That was my prediction. That was my prediction after yesterday morning. It was one of your predictions. But I was wrong on the timeline. It was interesting. 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 into 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, uh, invidia, you know, the chips that they make sell artificial intelligence. That should be, uh, replacing human labor. And yet, uh, the job demand is, is surging. as well. And it's notable they said they have visibility for a half a trillion dollars in revenue through 2026, which... I mean, it seems crazy, but... It's not enough anymore. 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 multistream, go to Restream.com. So we talked about it a little bit yesterday on the show. There's new product from Travis Kalinick, 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 Kalenick himself. Why don't I read his reply, and then you could kind of take us through what you wrote in the newsletter, kind of sort of thing. Why don't I start with a little bit of context and he can add to it? So 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. It's so good to see him back on the timeline. Like he's never, he's never been not been doing business, but he's been so quiet. And he's just dropping like deep alpha on the market that he's operating in. Yes, yes. That is like non, like in my view, like I never, it makes sense what he shared, which I 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. Okay. So you don't know the name city storage systems, but Cloud Kitchens is actually a subsidiary. area of city storage. I thought Cloud Kitchens was the top. That's what I thought, but it's actually the opposite. City storage systems. Great, great if you want under the radar, holding company to, you know, verticalize food delivery. Sure. 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. Sure. And so we were talking yesterday about how, like, broken the tipping experiences. 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 going to park your car right up at the front i saw viral uh maybe instagram reel or something about a guy who says that whenever he checks a new 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 chris one hundred dollar bell and he says he always folks really 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 it's... I might give it a try. The hotel I worked out, Michael Jordan would stay. And he just actually would carry around like 10 grand. And just any... He was just handed it out like candy on the property. And he would have a very nice stay, as you might imagine. So anyways, we were talking about that.

Travis responded and you can get into it and kind of give your reaction. Yeah. So Travis said, delivery app tipping isn't about... 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. This is just a hypothetical figure, but it's directionally true. Because you feel emotionally good about tipping, mentally, you give it less, it feels less painful to part with those dollars. purchase.

Yeah. Then buying gas. 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 tip. That feels like $8. And the taxes, that feels bad, right? And it happens on the other side.

This means that less price elasticity for the same price. So couriers are also economically irrational with tip. For every $1 in tip, they economically be, they economically behave as if it or a dollar 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

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.

I thought 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 this tip. Now, equal marketplace player, that's doing a lot of lifting because it's hard to just spin up. like, you know, I can't just start an Uber competitor right now. It's hard. Yeah. 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? So the difference here is that I think that one picnic is like is already counter positioned, right? So it's a pricing thing. It's a flexibility standpoint.

It's also counter positioning on like focusing on one key buyer. Obviously, you know, the door dashes that 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, you know, when I hear like no fees, no tips, like it just screams, like, there's been so many attempts at 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. it just 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 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, 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 has a real play here and a real strategy. And 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 other delivery apps and wants to switch to it immediately. So, again, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, if, take the model sustainable, I think it'll be quite competitive. Yeah.

I mean, you would, you would, you would imagine that vertical integration should allow, low, 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 a ton of other examples. But I wonder, I still wonder is, is this like the, 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 there actually like the VCs might have been subsidized. it a little bit, but it felt affordable on the, on the, on the, on the, on the, on the, on the, on the, on the, on the, on the, on the, on the driver 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, so, you know, I think it all worked out for everyone involved. But, um, it seems like Travis is reflecting on this idea that, uh, it was, it was, it, tipping was inevitable to come to the Uber ecosystem. Is tipping going to come to the Waymo ecosystem? Is tipping going to come to this picnic ecosystem, the picnic product eventually?

I don't know. Do you think Picnic will have tipping in 10 years? I just view this more as a corporate service in its current positioning than a consumer service. And when a consumer is buying food, if you're ordering food delivery, it is not a, it is like, it is a lug. right like food delivery has been extremely normalized but if you you know

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, they'll, uh, it's possible to like make it up in volume. Yeah. This is so funny. The way

Schill puts it, uh, 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 for or menu price. So a $16 burrito plus a $4 tip feels far cheaper to people than a $20 burrito that has no, a no tip option. Uh, but, but, but again, from a, from a, 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, want to 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. Yeah.

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 every day. And then whatever you're kind of, like, spending on top of that, you have to. to eat. And so I think consumers could very likely like picnic more, so we'll see. Yeah. 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 there was a trend for a while that was like, consumers want transparent pricing. Don't do all the crazy psychological hacks. So you'd be like,

yeah, I'm just going to put, it's 30 bucks. And that's what it is. And it's free shipping. And that tax and shipping is included. And it's like what we say up front 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's like a pop up. And it's like a ladder. One minute to add this to your cart. 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. So these psychological hacks are just like somewhat inevitable. But avoiding them, I think in the short term is a great go-to-market. I just wonder if there's something, if there's something like truly like counterposition 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 decades, right? The one thing that we learned from having somebody on the team call picnic is that they are focused on higher volume orders. So like teams of like 25 and up.

And so I think that I think that they're just betting like, hey, we can get a lot of volume. 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? Yeah. 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 to deliver one lunch, which like makes no sense.

I have one more take on this delivery question. But first, I need to tell you about cognition. The makers of Devin. Devin is the AI software engineer. Crush your backlog with your personal AI engineering team. My question is, what is TK? a 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 sort of ordering front end. You can interact directly with him. He's cutting out several of the middlemen. But is he going to be a logical partner for a zipline? Is he going to be a logical partner for Coco and Starship and these robotics companies that are delivering food? Ryan Oskinehorn 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. But what he's pointing at here is that Zipline 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. 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. Really? Gemini says, or Uber. I could not find definitive evidence. Of Zip line? Okay. So, so the AI overview says yes. We will do. Gemini says there's no evidence. Maybe we can ask him. I wonder, 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, like 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. 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. 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 vibe 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, the bubbles move around and tries to visualize how deep mind 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. Gabe says, how could 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.

That's a great point. It's a good point. Keller said they can fit two full grocery bags worth of food in their drone. Yeah, but I like, what? What? it'll come down to is the actual cost. Your team does a group order and instantly get swarmed by drones. Yeah, that would happen. 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 bringing it to the office.

You can imagine, like, six different drones end up carrying out. No, I mean, I think for this particular, for this particular, like, use, it just feels like they will be much, much more a buyer of like a Waymo type autonomy solution as opposed to a zipline autonomy solution. I would assume. 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 in the grant.

Yep. Yep. Well, we have a beautiful picture of Alex Carp's watch the Patech Philippe Aquanaut with the orange band. We'd love to see it. We clock this. TJ the wheel A long time ago. Jensen, one of Jensen's

leather jackets. Jensen, of course, has many leather jackets. I've only seen carp in one single aquanaut. There is a fascinating story of how carp wound up with this particular watch.

We'll have to get him to tell it on the show, though, at some point. We'll also have to tell you about cognition, the makers of Devin, which I already did. I already did cognition.

I'm out of it today. Adio. Adio is the air. We had a late now. Build scales and grows your company to the next level. 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 very chaotic morning. But we're back. We're back.

Nano banana is remarkable. Look at this Golden Gate Bridge image. It generates the image and also all of the diagrams around it. This is how Tyler sees the world, by the way. Sundar says, you went bananas for nano-banana. Now meet Nanobanana 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. I can't even see that. We were playing around with it this morning. It is absolutely wild. It's really, really good. The text is flawless.

there's there's just truly it doesn't make any mistakes with text anymore. You know, we were, 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 a difference. You showed me this morning.

Did you generate that? I had Tyler generate that one. I generated another one. This is like the most funny. Where's Waldo ever? John like says like, where's Waldo? And I'm like, are you, are you messing this 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. Yeah. There's some reason it did not hide the Waldo at all. Like Waldo was just perfectly in the center, very obviously. Most novice. Where's Waldo? It was very enough.

And then also there were actually. actually two Waldo's. And as you dig in, normally when you're hunting around Aware'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 superhuman.

It's not super Waldo yet. Anyway, we have Doug O'Loflin from semi-analysis in the restroom waiting room. Let's bring him into the TBPN Ultradome. Doug, how are you doing? Welcome to the show. How did you decide to take a vacation in the fall of 2025? Yeah, no days off. We're in the midst of the biggest bubble of all time.

It's going to be mellow. No, there's not to be any news. No days off. 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. Yeah. Yeah.

A little big deal. Well, we're going to hit the gong. That's massive news. That's massive news. Anybody can pull together a $200 billion L-O-I, but to find true love is beyond special. So congratulations. That's incredible. What's 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? 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. 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 interview, Sam.

I'm going to 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 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. sleep and shit. It was just really weird. Yeah. Totally. Well, welcome back.

Well, uh, thank you. What's going on with NVIDIA? Uh, take us through how you're processing the news. Uh, we, we've been batting around two takes. One was, uh, we're, we're extra analytical over here. The first take we had was, uh, Jensen was, was seen drinking a beer and therefore he will beat our rings. Chugging. It was linked, linking arms in South Korea. This is absolutely, this is our rigor.

And so I, I, I was. confident that they were going to do quite well. 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? 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 by side consensus. It's like a perfect quarter. You have no problems with it. But the thing is you're the biggest, most profitable company, or not most profitable, but you're one of the biggest companies of all time.

Perfection is expected every single time you report. So I think it's totally fine, dude. I'm being seriously, just totally fine. Like, stocks do go down. Yeah. People forgot about that. People forgot. They do go down.

They go down sometimes, man. It's crazy. What is the interpretation of, or what should the read be on, Gemini 3, the TPU, do. It feels like that's like, 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 just irrelevant? I think it matters a little bit. I think Google being really aggressive is really nice because do they have power and they're waking up and TPV7 is going to be awesome and entropic and

we're doing it. We're doing like actual deals. So, like, that's good shit, I think, because you need, like, Gemini has been, like, I don't know, asleep at the wheel despite inventing all this stuff. And so it's really nice to see them be back. But I don't think it's, 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 Open AI? Like, I got to ask that question out loud again. We've seen so many RL scaled up versions, but we know there's.

there's no new base pretrain, and we know that there's Gemini 3 cooks because it's a new base pre-trained model. So, like, where is that happening? Like, is it because the GB200s 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 Open AI is just, I don't know. They're not cooking.

I want to see them cook. So right now, Gemini 3. On the, on the base pre-train, is it still fair to, like, kind of set up that storyline with GPD now called 4.5, now sunset used to be GPT5, potentially, didn't really pan out. There's been a lot of debate over what went wrong with that pre-trained. 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 have said it actually was a better pre-training.

It was a better model. It would have been amazing. But we just messed up something about the economics. 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-trained, not that it wasn't a good model. I still think it's a failed run, dude.

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 about the economic side. This is where I have to pump inference max, right? Which got shouted out three times. So, yeah, look, man, I think the economics work now or could work with GV200 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 thoughter. Like maybe because of how much compute it takes, it's, you know, the 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 failed training runs, and so it's like, dude,

open AI, I want to see it. And I think we'll get it, right? Nothing makes them excited, but like competitions. Of course. 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 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.

to not keep doing the bigger and bigger training run. But do you expect Open AI 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. Dude, I think at the same time you say all that if they're being a consumer business, but you know Open AI has massive, like, they have coding FOMO, man. They're really, really, really concerned about the coding models, right? I'm sure you saw like the, I can't remember what it's called,

but I'll say it's like Project 2027 or, or no, no, no. project was like the 2027 scenario or something like that. Project 2035 was the right way agenda. You're talking about AI 2027, the fast takeoff scenario, this scenario where Open AI buys Ford Motor Company to make to make humanoid robots. Yeah, to make more widgets, bro. Yeah. No, no. So I think I think that while like, hey, look, it doesn't seem like it's on track anymore. Sure. Do you think the thing about the fast takeoff that people feel very strongly, 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 what like, you know, they were, they had so much codex FOBO. 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 just like, this weird, I think it's just like a perfect vibes time. On the like the finance side, dude, people are freaking out about the market, Fed cut. It's not going to happen. And so it's like, we, 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. 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 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.

Totally. What did you think about the Financial Times published an article that was pretty, I felt pretty misleading? They said Oracle is already underwater on its astonishing 300 billion open AI deal. And they said that because the stock is just around. It's their blog. They're having fun. But they're rage baiting.

They're rage baiting. They rage baited me. What do you think? They rage bait pretty hard, bro. Like, let's be clear. Like, I don't think, you know, making $400 billion of revenue is being underwater. But if you're, if you're betting on just a stock, then sure. Yes, they are underwater.

I think you're going to be okay. headline today from Alphaville. It was like, who is opening eyes on the auditor? They're really taking shots at everybody. It's funny. I do,

well, yeah, sorry. I was going to say, man, but Alphaville, like, I don't know,

Alphaville doesn't cook, man. Their alpha is so, it's kind of, it's kind of mid. I don't know how to tell you, man. Petition to rename Alphaville midville.

The beta, to betaville. BetaVil. BetaBeta boys. That's rude. I've enjoyed Alphaville from time to time. I want to know about, this Gemini 3 pre-training run,

is there any way for us to understand the rough order of magnitude of compute or dollars that went into it? 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 GPT3 training run,

the GPT4 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 scale of the investment.

But do you have a more solid understanding of how big the Gemini 3 project was, like from a KAPX perspective? 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 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, or 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 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 goes into training a really big model costs a lot in R&D. And so I think the final shot or whatever in terms of compute is probably paltry compared to like the actual total spent, right? You probably have a multiplier of like 10x on top of it of what the final number is. But I don't know.

It's probably a billion bucks if I had to guess. Yeah. I don't know. I'm just going to throw out a number. You heard it here for a billion dollars. One billion dollars. Because I mean we've heard about training runs that were, you know, like a couple years ago. They were in the 100 million range.

and the billion-dollar training run was kind of rumored. I wonder, I mean, I wonder 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 lines? Yeah, something like... Yeah, something like...

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 significant slug of... I guess it goes into cost of goods sold. Like, AI accounting is, like, completely made up today, so know who the hell knows. But it's probably a cost of goods sold. 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 spend a billion dollars and then you make $100 million a month for, you know, a whole bunch of time, hopefully. 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 train a model every year, it's R&D. So that's the issue, right? You can't capitalize it. So I don't know, man. Well, I mean, I think, well, I mean, I have a big question about this. This is something that I've been going back and forth with, which is like, we have seen that there is demand for 4-0 chatbots from like a group of Redditors, potentially forever because those people are like, 4-0 is my friend.

I don't care about Gemini 3. I don't care about GPD 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. It's just like their buddy. 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 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. We don't need to go to the latest and

greatest. Do you think that's going to happen in the enterprise? I mean, I don't know if it will happen. I mean, enterprise, just like, let's get, like, more enterprise bullshity. People have to have price raises, and you have to be like, well, why did you raise my price? And the single best way to do this is say, 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. It's like, dude, RuneScape Classic is probably, like, a perfect case today of this. People want to play RuneScape Classic.

They don't give a shit about, like, and, like, obviously, Rooscape Classic has. 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 that, we're probably like, you know, tens of thousands, hundreds of thousand people who play it. And like, yeah, I wouldn't be surprised that we see like these long lived little projects that are really stable. 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, 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 frag.

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 think that that's, it's kind of hard for us to be like, like, and also, dude, that's not AGI.

Like, 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, dude, you're telling me 4-0 is all I want? Then what are we buying? Why are we spending 100 gigawatts? So 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. But the leverage is still coming in. The circular deals haven't actually hit the books yet. They've just been announced. Like that's some. We didn't Jordan from semi-analysis say, that there was potentially going to be like an H-100 index that retail investors could

like buy into or something like that. I'm just interested in 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.

So the last note, I had my team who is sitting in the office right next to me and behind me. You're in an office right now? It looks like you're in a forest. Yeah. I love the walk. And 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 a thing, I think all the hyperscalers could raise, like, $2 trillion. Like, I really think the number is so large. I'm trying. 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? Okay, so I'm going to, I'm going to give you the maxed out version of how I think about what we could do. So, um, leverage max. The new, leverage max. The new report from selling analysis, leverage banks. Also, also, I've been told by, I've been told by corporate overloads.

So you have to star the inference. that's super important. You have to star inference max on GitHub. Sorry, before. Okay, so how... Everyone, go star inference max on GitHub, please. Thank you. We'll make a call to action.

Cool. That'll help. Okay, so I think they could probably raise something like $6 trillion. $6 trillion. And is that like a 5% interest rate you're assuming on like corporate debt basically? And that's what we're paying? Yeah. So we just essentially, 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 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 blue out, come in, pay for all that. You do a sale lease back. 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 bibs above the government. Yeah. And also the rating agencies are like, okay, as long as you don't have more than one turn of debt, by 2029, you're good to go. So we did that number for all of them. For all the hyperscalers, X Oracle,

Oracle's pretty tapped out, $6 trillion. That's like the big number. That's great. So 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?

So here's the thing about this, though, is free cash flow doesn't like, pre-cash flow goes negative if you assume there's no revenue growth. But this is kind of like a shale well. You get a lot of your money on a GPU cluster up front. Let's say five-year economic life, people are going to fight me about this, but whatever, you will have your payback for a brand-new cluster in something like 18 months. And so after that on the like, let's say on the 18 to 24 to 36 months, which is like the two to three year, you're just going to start now you're going to 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 supplies, like, 12-month payback period, which is way more insane than what this is. But, like, 12-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 going to be able to raise against a lot more money. There we go. Yeah. But they're tapped out this year. Yeah, yeah, yeah. That makes sense.

Okay, I have kind of a lightning round, because I know you have a heart out in a few minutes. What is going on with core weave and core scientific? Like, core weave is reliant on core scientific. 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 is now traded down 20% or so since the acquisition was attempted. but can you explain that dynamic and why the core scientific shareholders are kind of still

holding out at this point? Okay, so Cores got offered to be bought in all equity, and this was 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 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, you know, people who are, like,

the stock does go down and you have a shareholder turnover when you reject a deal. Because, like, a lot of people who 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. But at the same time, core scientific is not delivering their dentin 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 spark nose 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. Got it. That makes sense. 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 on, 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? Are they

increasing earnings? Margins? Yeah, is it affecting margins? And I don't feel like that. I don't know that there's like a definitive data source that I trust on that. And that's something that like, I only trust some analysis for everything. Thank you. Thank you. We're the only source of truth, bro. I'm thank you. I appreciate it. I think of a Bitcoin maxi, but for some analysis. Same dude. So, so that sounds like I need an AI agent to call, you know, like 100,000 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, there isn't something we haven't thought about. Yeah, but we have a lot on our plate. It's like a throughput problem. Yeah.

Like, it's like another trading run. Super important too. 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're, like, we're putting a lot of effort and 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? Don't, don't name it. Don't name it that. That sounds weird.

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 XAI just getting into the cloud, into the AI cloud business and helping power and helping basically deliver infrastructure for other companies? I, you know, no.

not until this conversation, but they're the best. They're the best of being quick. So 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. I think they literally have the speed run 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 going to be like, give me money in my pocket. I'm going to make a Saudi GROC and then I'll make a GROC 5. 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.

And I bet you some of the people who took X private probably were Saudis. So, yeah, might as well. That's true. Yeah, they actually were. Quick take on the Brookfield deal. You mentioned it early after you joined. A hundred billion, but you should look at that. the actual number is $5 billion committed.

So just... No way. Not beating the press release economy allegations. Yeah. So, so, yeah, dude, this is the press release economy, man. I mean, I mean, you can do to, you saw the 10-Q, right? We got to do a deal. We got to do a press release.

We'll pay you $100 billion if you pay us $100 billion. Dude, I think we can make that work with our accountants. I think we can do it. And then, yeah, we'll have a circular deal. We'll write checks to each other. We'll hand them off right at the same moment, dramatically. Exactly. Yes.

Right there. Right there. Right there. Yeah. And then the economy will just circulate, bro. Exactly. And then we can use that money to raise capital. Yeah.

It would be beautiful. Holy sure. No, no. No, no. 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. Yeah, we want to aura farm everyone who's doing it unironically. That's what we done.

I mean, yeah. Well, okay, my most galaxy brain take from. video actually that I feel pretty strongly about is um and this is you know in our call with them they're like you know this 10 billion dollar 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 uh being closer to their customers and understanding what they're doing is probably the number one thing that they have to do as invidia to understand where the technology is going yeah right and so i think if you if you think about it it's actually just r&D opex bro it's just a check that you pay in order to make sure that you're

to Open AI and Anthropic, and you know exactly what's going on in their data centers. So that's the most, like, bullish take, I can think. But realistically, man, the Fed said something and everyone's freaking to fuck out. Yeah, that makes sense. Yeah. I mean, on that, are you referring to that line that's getting shared around from the invi earnings around the quality of the deal with Open AI versus the strength of the deal with Anthropic? Did you read into that?

Did you read the same thing that everyone else went into it? Yes, 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 Samma. 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 met up, 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. And then like if it comes, it comes, right? That's how you do it. 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 about to 2045. Yeah. Yeah, I'll just have my children actually. We'll do a deal with our children. My favorite big number of the week was MBS was hanging with Trump. And he was like, I said $600 billion yesterday. Let's do it. But actually let's round up.

Let's actually. And Trump literally like hit went like this and like hit him on the knee. Like he was so happy to hear that one trillion. He's like finally, finally someone said the trillion, bro. Everybody gets around. Everybody gets around Trump and they just like detach from reality. a little bit and they just start saying, like, Zuck had this at the AI dinner.

Remember? Like, he just threw out a number and had to correct it the next day. He's like, what number are we going with? Oh, 600, 600 billion. Yeah. Dude, I mean, it's like his warp field is that everyone just says the biggest number around him.

I love him. Like, it's kind of ridiculous. It's powerful. It's real stimulus. It's stimulating, that's for sure. I don't know if it's real. You need a semi-analysis plan, like a subscription where it's like $5 a month for the first 20 years and then $20 billion in 20-45. I will sign up for that and then you can book it and defuse it back.

Defer the revenue. It would be beautiful, bro. And yes, there's an out. We need to cancel any time. We need to bring massive RPAs and press release economy to the newsletter analysis. We need to, we need a sub-examination. stack feature baked in for this. We can do, we can do a deal.

Bro, it would be great. We'll do, yeah, we'll have a preliminary $1 billion advertising deal. Yes. We'll do some circular economy. Well, thank you so much for taking the time to hop on. Great to see you. Congratulations again.

So, so happy for you. Thank you. I appreciate that. Inference Max. Inference Max. Inference Max. Inference Max. Inference Max.

Inference Max. Inference Max. Inference Max. Inference Max. Go start it on GitHub. Okay. Did you know, Did you, sorry, did you, are you familiar?

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? I like it sounded. I despise the organization. We have eyes and ears everywhere. Your information is incredible, bro.

Holy crap. We're 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. Yeah. Nice to see you. Take care.

Let me tell you about. 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. Is he? He's the founder of Momofuku Restaurant Group.

Not sure we haven't quite yet. 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. Open AI's model is at a 38% rate. And so there's been a very significant quantification of the improvement in Nanobanana Pro, I believe, or Gemini 3 Pro image. I don't know exactly the difference in the name conventions. Do you know? Yeah, I mean, I think, so originally

Nano Banana was like the insider, like, that was the code. It was the code word. And then they're like, oh, let's just bring. Interesting. I was wondering how that happened. Yeah. Because it's very, same thing. 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 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 also have a Nanobanana 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. How are you doing, guys? Good to meet you. Thanks so much

for taking the time to talk to us. Great, great fit too. Looking great fit. You're ready. Ready for anything. Yes. 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. For F1, right? Yes, sir. Fantastic. We'll be there. Saturday. Can you help everyone in the audience understand just the shape of your business between media? The shape of your empire. The shape of your empire. Empire is the correct term. Sorry. Not just business. Well, it's restaurants. We have some quick service fast casual. We have fine dining. We have a couple of spots in Vegas. We have a couple places here in Los Angeles. I think it allowed us to, a pandemic allowed us to sort of refocus exactly our gross strategy. Yeah.

Instead of trying to open up all around the world, which we had been doing until 2020, I think 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. 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 of 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, which we air at 4 p.m. Pacific Standard Time, Dinner time line. Have you ever tried live streaming? 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 basically a full-time job, but I could see that being a hit. We tried to do that.

You'd be surprised how adverse I think a lot of people that are running network 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 media is to do just all day. You know, totally transparency. What you see is what you get. And you do see some people doing it on Twitch. Yeah, that's what I'm saying. You don't need the networks. I was wondering. Create a Twitch account. Create an X account. But I'm currently pretty preoccupied with Netflix and Amazon and Spotify these days. Especially when our podcast is moving over to Netflix. Oh, really? Oh, no way. You're in that.

in that package. Yeah. 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. I would love to answer all of those questions, but I don't think I'm the... Okay. Anyway, 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. We'll have somewhere from Netflix on. I will tell you that, it's, it's been something we, you know, we have, what, 600 plus episodes

of our podcast, and it certainly changed over the years. When I first did the podcast, it was much more of the insiders take on the restaurant industry 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 Angler in San Francisco. We sort of gave everybody a sneak peek of the, you know, the philosophy behind it. And then the pandemic happened and then 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. 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, also more importantly, not necessarily scalable. So now with video, right, and more people, you know, with all the data and, like, more people are watching podcasts than 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 certainly in the near future going to outpace it if it has a lot of 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.

Yeah. 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. It would be very hard for us. It's hard to eat and do a podcast. Oh, eating on mic is possible.

but even just cooking. We were talking earlier on the show about kind of the delivery app experience, like the dynamics of like tipping in delivery apps. Travis Kalanick was commenting on a post of ours yesterday. He's got a new product called Picnic, which is like a front-end delivery platform just focused on corporate meals. And like the key value prop is no tipping and no fees. so they're focused on like higher, higher volume orders.

So like teams of like 25 people plus. I'm curious 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 and all that stuff. Well, I don't know if many people know or remember in 2016. We created the very first, to my knowledge, there might have been something called a ghost kitchen, but no one called it a ghost kitchen. We teamed up with Thrive and Will Gaborick and Josh and Caleb, et cetera, a great team.

And we opened up Maple. And we were doing like 10,000 meals a day out of New York City as a full stack app. Yeah. We were delivering. I had no idea. I didn't realize you were behind that. I remember Maple. But that's cool.

And 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 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 restaurant so it was a real opportunity to sort of bridge that and to to do everything oneself and yeah and so much so i believe in that we started another one called ondo which was more of fast food so like maybe you might get a nice kale salad and a butternut squash soup and ondo we started with uh garik camp's uh fun expa and we did ando which was like a cheese steak and fried chicken

so i was all in i was all in we were probably just six years too early yeah it worked but not enough where it is today because but is it so is that is that model is a model like the hard thing is like Cloud Kitchen's the dominant, 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 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 restaurant industry has as well? Listen, I think it's like anything else in tech, right? Remember in the dot-com bubble, You had like tubesocks.com, right?

Like, there's only like three or four companies that really came out of that. Same stuff, what I imagine with all this AI shit, right? 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? It reminds me, again, in 99 when at least in New York City, every company, I wouldn't say every company, but a lot of places where now, changing their name to sort of pizza 2000, dry cleaning 2000, right? It just is sign of the times. And I just, I think that in food delivery, you're going to have about three to four winners, if that. And certainly, I would never bet against Travis and the team there, Cloud Kitchen.

Definitely not. I think what Tony and the team at DoorDash is doing is just unbelievable. And clearly you have Uber and the Postmates guys. So to me, that's pretty much going to be the space. And I think when, I'm not sure when, but when they are able to be more open and transparent about everything, people are going to be like, wow, that's a pretty goddamn huge business. Yeah. Interesting. Drinking culture and lots.

Any restaurants been caught using AI generated imagery for their menus or any of the food delivery apps yet? That's a good question. No, but, you know, I think it's been, it's the same shit that's happening, right? And they're like, AI to me is like 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 just think that in restaurants, that's basically been like consultants, right? They've just been, that's just like, I can go, I feel like I've been going to AI generated restaurants for some time now. You know, it just hasn't called AI. That's funny.

It's a good take. Are you excited about drone delivery? You know, that's one thing where I thought it was going to be a total zero. I'm dead wrong about that one. I think it's definitely going to be a thing. Well, isn't it exciting as a chef to know, like, if I make this, it will arrive hot? No, it's not going to arrive. Well, that's the whole other thing.

The one thing I will tell you under the food delivery space, and I've talked to just about everyone under the sun over the past 10 years. that's try to start up a food delivery company because they're like, oh, if 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've created some kind of new technology to cook the food, it's going to be hard to really make the food hot, ultimately, right? Like, how should I say this? There's no new technology to make the food better. None.

So the delivery drone, unless it's cooked. cooking the food as it flies, it's always going to be limited. An oven, oven in the air, basically. Yeah, I mean, like, that's just the truth, right? And also, a lot of these places just have a bottleneck because everybody wants to eat at 6, 37 o'clock. So there's just, there's not much you can do to make the food go out faster. Yeah. 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 food. the house or apartment. Have you heard any pitches on that? I have. Any? No, you don't have to give names, but like, it gets to the point where it's like a street.

We just have like a massive proliferation of like street carts and it's and maybe like. Well, I mean, I'll tell you this. A lot of these pitch. I haven't been two in a couple years because I just don't want to do it anymore. But a lot of times there'll be a very success. like 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, it 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? Cooking is still a physical endeavor. And for all the VC money and tech money, it can't sort of solve that riddle of how do you make physical labor go away? Or done better.

So you don't think we're going to, you're very bearish on the humanoid's chefing up? No, no, no. I'm not bearish on that either. I spoke to somebody. Pre-pendemic, we did a show, and we did some research and a robotics expert, and we talked to some people at Caltech and a few other experts. And if somebody was like, oh, maybe 40, 50 years away. And I talked to someone recently, you know, like, yeah, we're probably 15 years away.

From getting somebody that has a robot that has the dexterity of a high-end best-in class sort of chef. So no, I, it's going to happen. If anything, I think you're going to see the next five, ten years, you're going to see robots. You already see it in, I mean, again, like, I don't think it's been a sudden, oh, my God, there's robots. Those robots in the kitchens all the time. Like, if you go to a good restaurant, there's dishwashers, a pretty much a transformer robot. It's amazing. And that does the work of like 20 people.

Pretty underrated. Underrated. And if anything, you're going to see machines that take the binary movements out. So a friar that goes up and down, a bathroom cleaner, things like that. And already dishwashers are pretty advanced that can handle very expensive stemware. So finding somebody that can polish stemware, you know, you might get a wine glass that could be $250 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 kind of position will be a robot. No questions about it. That makes sense. What's the most overrated trend in food right now? Oh, man.

I'm trying to stay positive these days, guys. 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, like, durable. trends. So, well, I would say the most annoying trend is that everything has to be the best. This hyperbole, right? I have to have the best X. This restaurant has to be, you know, world class, number one. And they hate to tell it to you guys, but I think most people wouldn't even know what best is if they ate it. Yeah. 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.

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 than 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 the media and chess were all part of the problem too, right? you know, all these lists and it's all stupid, ultimately. Yeah. But do you think, do you think we'll ever get to a point where, like, in, 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,

tiny, tiny little restaurant, right, like, you know, a thousand square feet. And the second you had the second restaurant and the third, it just kind of, 10, 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 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? Or is that just kind of an iron law? 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, you just mentioned,

something, the fact that something that's great is not scalable. And for years, you know, I've certainly tried to do it and scale these things. I just sort of spoke about this at Reed 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 easy ideas are to scale an idea than food that is, you know, I don't say cheap, but affordable and mass produced, right? The other end is high-end, experiential dining, which weirdly has become very scalable because of its inaccessibility. It's the equivalent of getting front row tickets at the Nix or, you know, Chase Center or something like that because you're now eating at, say, the French laundry, no one else you can

get there. You may not even appreciate the food, but it's now a social flex. It's cultural currency that you can sort of have, and it's a femoral. And because no one can have it, weirdly, now that experience. is weirdly, 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 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 make an easy facsimile. So it could be barbecue, could be sushi, Anything that is best in class that people have a hard time copying. That's like the barbell. So you have really affordable, make mass-produced stuff on food on the one end. Other end, you have things that very few people

are going to be able to experience or eat. 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 talk 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 that the answer that needs to be solved is how do you scale the middle right as that makes sense like the mom and pop the restaurants the diner the restaurants that are just like good again 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. 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 I never talk to 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, there's a lot there. That's the mediest part of the food industry right now, but it's just too damn hard and nobody really wants to touch it. Interesting.

Have you seen any interesting experiments on, like, the capital side of, like, new restaurant creation? Has anybody tried to make, like, a Y-combinator for restaurants where there's, you know, talented, you know, owner-operator, chefs can get some seed capital and kind of support to go from zero to one? Because I feel like you would have tried that by now. We have definitely tried it. I won't say we've tried it. I know a lot of restaurants have tried it.

I know there's funds out there that try to do this. But I would say, you know, Ron Parker created something called. hospitality annex and that's a website that is a little bit like a job board legal zoom 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 has a high barrier of entry in food yeah what other right and And there's no moat to really create, right? Yeah.

Yeah. On the topic of like, you know, starting up and go to market strategies, are there, uh, are there risks to like going to viral early? Uh, we've experienced a bunch of, uh, like rage bait in tech recently where people have set sort of designed products that are, that are designed to the, the whole product is just designed to enrage and go viral and then get some attention. What do you mean by enrage? I, I'm not familiar.

So 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 time periods where you have a little two-minute break. And so they added the ability to gamble with steak while you're making software. And that made a lot of people mad for obvious reasons. Yeah. Or just deliberately picking messaging.

Like rage bait and 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. I mean yeah, I mean, I don't know what rage made, but like, again, like this has been happening on for forever anyway, you know, doing something that is probably going to, you know, shit, I've opened up restaurants that I guess have been like that too, you know, it's just you're not, you know, I think if anything, it's just taken to another level because I would say that a lot of chefs now when they're talking about dish,

is it something that the younger generation will find appealing to record? 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 I don't know.

I want to be optimistic again. I'm usually Mr. Eeyore 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 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 country. And I think that sort of pattern is what you're going to see throughout food. And it's a long wind of way of sort of answering this sort of rage bit.

And I think because of that need to sort of find something that is going to create some kind of spark in food, that is the catalyst that's going 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, like, a little bit like Japan, where you can open up one specific kind of bakery that makes one specific type of thing and you do it better than anybody else. And you're going to see that here in America. I feel very strongly about that. Yeah, I love that approach.

What advice do you give to kind of emerging chefs on media strategy? I think in tech there's 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 mysterious dark horse that's kind of going over the radar and there's like a messy middle that's probably a disaster. Yeah, I don't think that that pattern is any different than what you see in food. But at the same time, I think, you know, I don't know if apathy is the right word, but I don't know. apathy is the right word, but I don't care about it as much anymore either. 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 I just don't know what that best long-term strategy is. And now you have an older generation of, you know, I'm 48 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. And I still don't know which one works, right? Because once you feed that beast, you have to do it all the time. And that's a lot of time. So 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. Yeah. Right? And do you build a relationship in that repeatability? And like there's my favorite restaurants in L.A., 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 going to go there like when I have the time, right?

And so I think. I mean, yeah, I think that's the zag, right? Yeah. But you can't do that unless you actually have a point of view that resonates with somebody. Yeah. And if you are constantly sort of pandering and figuring out like how to, to execute other people's dreams, wishes, and visions and what the hell are you actually making? Yeah, and you don't want to get to a place where your content is better than the product.

And I'm sure that's, like, you know, a lot of people, the more, the more you time you spend on content, like, the more greater there is a likelihood that it could get to that point, I think. Yeah, I mean, but like, do you guys care about what you see on social media still? Like, I actually think there's a bifurcation that's happening with what people see versus what actually people are going to eat. I do think that there's, I don't know, maybe the steelman argument for the viral over the top, you know, TikTok that gets me to go to a restaurant is that it can in some ways create like a shelling point and like a coming together. Like if there's something that's trendy and it's an excuse for me to pull my extended family, my friends, different people and it just gets us an opportunity to kind of come together, they. and experience that like even silly, trendy over the top thing.

I think that there's something that can be good about that, but it's certainly not like the primary reason why I go to a particular restaurant. No, I mean, that's the thing. It's like I actually, we're working on a show and I can't say which or where, but, you know, 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. Okay. You know?

Yeah, yeah. Fade everything. Go next door to the restaurant that you're supposed to go eat at. Okay. Oh, that's cool. That's very cool. And sort of that in principle, right? Yeah, no, I like that as a philosophy.

It's just like the other thing is, I sort of mentioned earlier in this the conversation about somebody was tasting something that was truly good and remarkable. Would they actually know what's good and remarkable? And I think currently we again have a knowledge that is greater than it's ever been in terms of food. But, and maybe this is the same way in fashion and architecture and film and other arts. But does your audience actually know what good is anymore? Because I don't know, right? 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 I'm giving somebody like a like 1998 Ravino, you know, from 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. say you can't, you can never say that Salieri was better than Mozart. Right? Yeah. He was good, but he was not better.

And that's just sort of unequivocal. Yeah. 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. Yeah. And that's sort of my concern when it comes to 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. Oh, I love it. Last question on my side, I'm curious how restaurant operators are planning around America just drinking less than ever. Yeah, well, that is the, you know, I feel like the boy who cried wolf, I've been sort of screaming this flag for a long time. This has been, this is the real existential threat.

Like, for example, L.A., the biggest thing that happened in L.A. Over the past 10 years in food was really ride sharing because people were getting drunk. And you saw that in revenues. Restaurants are going through the roof. And if anything, restaurants was a bubble, right? Too many restaurants. And I think we're still sort of in this bubble. that's a whole other conversation

but I think that you can see now at least in L.A., people are drinking much less. I think you see a younger generation maybe taking some edibles. The crazy thing is kids just don't drink anymore. Kids start, when they start a tab, which is crazy to me, they close it out every time. Yeah.

That 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 got to go back and get it. They're too responsible. You know, they're closing out every time. There's a responsibility. It's hurting small businesses.

It is. It is hurting small businesses. And, but I think that there is, if you look at the sort of the only look at the blended numbers for most restaurants or beverage 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% are 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 got like 3x, 4x 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. Why is it more difficult just to actually deliver something on that's like imagine if we were making the alcohol too yeah from scratch that's hard to do that's and that you know a normal restaurant ratio restaurant ratio was 70 to 30% for the most part you want 70% food i mean this is not that i like roughly roughly 70% food to 30% dev sales and i think that is completely shifted and for a good restaurant it's like 10% yeah i mean like if you want 10% of your you know profit for example right like

something's going to give when you're down like 18% on dev sales. You know, I think that's the average right now or something like that, 15, 18%. So I don't have an answer. 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. Yeah. 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. in 2019. 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 think that's what we can do is sort of figure out what the alternatives are. I don't have the answer. But isn't one of the challenges is like these non-alc products, like somebody's not like, there's not the incentive to have the second or third. Like I feel like a lot of,

this stuff. People just have one. 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 just like not getting drunk, right? So they're not going to. Listen, are you guys drinking as much as you used to? Absolutely not. No. You know, I feel like the way I used to was like Dom Draper and Madman, the amount I used to drink. Yes, yes. You know, and I, you know, part of that is just a generational shift, but 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. Oh, yeah.

It's just not something they want. I've seen this in kitchens. Like, you finished your 12, 14-hour day. All you wanted was that cold beer at the end of their shift. And now they don't want that. And I'm just like, what is happening? You know, and I'm not saying they're wrong. It's just so that we're sort of dinosaurs.

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. 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 drinking alcohol, even in small amounts. And so I feel like that's what's really cascaded. Maybe what we should do, restaurants should start a lobbyist and just muzzle Hoverman and see it to you and we'll be okay. Yeah, live life.

Yeah, I mean, like the dual pressure right now from, from, from, like just 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 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's now 25 percent between effectively for or 24 percent for service uh at the end of the day food needs to be more expensive. And I'm not, it just sort of has to and it can't be sort of be passed down. I think 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'm pro when a restaurant jacks up their prices to like, I'm hoping we see a restaurant where the, the, the, the, the, the, The ability to eat there is basically going to a Taylor Swift conference, a secondary market. Yeah. You know? Like, that's sort of what has to happen. I do believe that it's going to 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 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 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. America, baby, it's because we're Americans. We do things the dumb way sometimes. Americans can still do it. But as an industry, as restaurant leaders, we're not.

As restaurant leaders, we can just use the metric system. And again, you know that it's bad when drug dealers use the metric system. The drug dealers use the metric system. That's right. So what the hell are we doing here? So if we can't adopt the metric system as an industry, what are we doing here? Yeah, yeah, yeah. What a mess. What a mess. Last question, we've got a bunch of people in the chat have asked. Who do you think is going to win the AI race? Hot take.

What? Really? Okay. We had to ask. This is a tech show. Just give me your gut answer. First reaction, one word, one word. Google Anthropic, Open AI, who you got? Commodore computers.

There we go. There we go. There we go. Dark horse in the race. I love it. Thank you. Great hanging. But we'll see you guys at F1.

Yeah. We'll be very excited to see you there. Can't wait. Have a great one. We'll talk. Be good. Go by. Have a great rest of your day.

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. Laura Dan from Figma. Thank you so much for holding tight. Look at this. Look at this background. I couldn't even tell if it was.

It was a TV. I just noticed that it was a TV, but it took me a second. 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 some sort of fake background. It works very well. So great to have you on this show. Thanks for having me. We've been reacting to Nana Banana Pro this morning,

very, very impressed on a bunch of different dimensions. But before we get into that, I would love an introduction on yourself for the audience and your background. Yeah, of course. So I joined Figma two months ago as their chief design officer. And before then, I spent close to a decade of 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 before you ask me, I'll tell you why I joined Figma. I did so because in my seat watching all of the AI improvements that we're seeing with these frontier models, it became very, very clear that product development is a process is going to change drastically.

And I truly believe, and I saw that Figma has the opportunity, and I think the responsibility from my point of view, to really build the creative environment that helps people like me, people that really love to live at the intersection. I used to be a musician. I became a designer, the product leader. I really believe in the thing we're making more than how different disciplines kind of line up 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 Figma build this. 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? 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 helps you understand what it

looks like, what it feels like in motion. And what we've been looking into with this is an aspect of AI that I think gets overlooked sometimes as a creative tool is important for AI not to box you in. So you want to be able to take your design from Figma design and generate it. But then you want to take those generations back to Canvas and be able to like manipulate them as well. So yeah. 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... 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 reprompt it and you try to say like can you remove that could you like try again on that you know you 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 do you have a personal evaluation that you run when a new generative image uh model drops. I have this, the Where's Waldo test. I try and get it to generate a full Where's Waldo because 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?

I don't necessarily. I have a shit ton of styles that I put it through the ringer 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 Weevy, one of our companies said today, it's a model that behaves. You should actually watch the video that they've put out. It's hilarious and it's made and levy with Gemini 3. Yeah.

Yeah. Apparently, I mean, Prins here on Axis saying, Nanobanna Pro is a reasoning image model and shares 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. Of course. Is it realistic? Is it something that's, that really connects, like our eyes, right, we'll 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 Nanobanata Pros 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 video with his like Figma quilt behind him.

Yeah. Yeah. 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 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 other types of tools like this.

This has not been possible yet. What how how much do you care about like leveraging some of these models to help people generate new ideas? Because in my 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 that's- Yeah, creativity is messy. Yeah. 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 taking the next step with those. 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. Yeah. Yeah. That makes sense. 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 just fun prompts? I'm seeing people use Nano Banana Pro to make RPG style maps. People 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 the internal team chat. 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. Yeah. Yeah. I'm excited to get deeper into it. It's such a fun tool. What's your what's your updated read on just like general designer sentiment around AI? Because I feel like it, it fluctuates 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...

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 ringer because we want to see how in the hands of designers these become clay that they could mold. And so I think 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 coming up and how it could work?

or do you look at exactly what it produced yesterday, in which case a lot of times it is not great. Yeah. Makes a lot of sense. 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 that Figma 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. Yeah. Yeah, it's interesting. I process those two things very, very differently. But yeah, I understand where you're coming from on that.

Well, say more. 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. Oh, reliable. Yeah, yeah. Yeah, it's not superlative. It's not the world's heavy. doughnut, the world's like most, you know, gold flakes on a steak possible. Like, it's not viral.

It's not the best. 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 getting, 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 you are possible. Emotional connection. Yeah.

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 fast food of design. And what's interesting is to think about that messy middle of design. Like what is the mom and pop show? 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.

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 wanted to just deliver in some case, maybe that's your to-do app or 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, 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 surprise moments, the delight moments that make people be loyal to an app. 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 immediately close it because you realize like they just fully generated all the text. I feel like we're

going to start to get that with software more and more where you'll go to a website or an 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 looks okay, but then the second that you actually start using it, you realize there was no real human thought 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 going to feel like using linear. Exactly. And so that's where, that's where the human element is just going to continue to be super, super powerful in that, and taking user feedback and like having that empathy with the user and being

super thoughtful and 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. Yeah. That makes sense. Yeah. And I think AI will play a role into that. But actually to go back, I am so pissed about M-Dash's. Such a good tool. And every time I write now, I, you know, use them and I'm like, whatever. Yeah.

People are going to accuse me of using AI. Yeah, I think you just have to use the minus sign. Just incorrectly use the minus sign. I do think it's going to be recognizable when a website is just like kind of bi-prompted, vibe-coded and put out there in the world. And you're going to want to feel that the developers spend more time considering that. Yeah. Amazing.

Well, thank you so much for joining. 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. Try out all the new toys. Yeah, thanks for stopping by. We'll talk to you soon. Have a great rest of you. Cheers.

Bye. If you want AI to handle your customer support, go to fin. com. The number one AI agent for customer service. Let's react to some of these Nanobanana prompts. They look fantastic. Here's one where someone took a map,

a Google map screenshot and just turned it into an RPG style map, a San Francisco monster map. And it's really is that reasoning model. 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. You get the sea monster in the water. And so all these things are like pretty logical.

There's of course some things that are a little bit repetitive. Like ogres in Golden Gate? Yeah, yeah. It's just very, very cool. I think this is going to be a lot of fun. Then there's someone else with a benchmark here. Angel. This is Nanoban 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 land, not, not, not. this next one's wild the make it Lego the Lego prompt is crazy this could be the next like Gibbly moment for sure if you can just take a whole bunch of photos and pipe them through we should take some of the our iconic photo take a picture of us and put it through Nana Banana Pro and make it Lego I did I tried doing this earlier it it works pretty well sometimes with the people it yeah it doesn't use like the mini figure like oh it doesn't 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 going to make us a mini-fig? I mean, maybe that could be worked into the prompt. Do you want to take some of the iconic 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 nanobanana and ask them to render us as mini-figures in this Lego world. I want to see the Ultradome in 4K Lego. while we're doing that, Pietro, Chirono, Shirano, shares that Nanobanana 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. I can already imagine, you know, when we got Chachapiti, it was like, oh, wow, you can take, you can take bullet points, and you can expand it 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, turn this diagram into an essay and then summarize it. Turn it into two words for me. 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 TBPN perspective.

D.D. shares that he literally fed Nanobanadpro raw graph viz of AI compute commits, generated beta 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 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. But it's like 70% of the way there? Yeah, yeah, yeah, I would say it's 70%.

I'm just excited that it actually puts, it gets the logos correctly because with the right direction, 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 would probably give it a lot of like scaffolding to build off of. That would be very cool. Look at this. Okay, let's see this. Let's zoom in on this.

So we are, Lego Us. We're not mini figures. We're just Lego, Lego people. Okay. Color temperature is 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. Have we done any others? This is the only one I made support. It's pretty slow. It's pretty slow. Well, this is funny. So on this next one.

Yeah, Gemini 3 Pro Image versus GPD. Okay, so I didn't read the caption. I just looks. at the image. I just assumed that the image on the left was an actual image. Yes. And then this was the output on the right. That is crazy. It looks terrible.

But it's actually that, no, this is a real image. Same prompt. Two different results. That's pretty remarkable. Yeah, V-O-4 is going to be a big, big moment. I'm very excited because V-O-3, I mean, such a huge leap over the original SORA. It was pre-Sora. What was chat Chb-T's video model, not Dali, but pre-Sora app?

Was it called SORA? Yeah, it was always called. It was always called SORA. Okay, yeah, that, because SORA 1 or whatever, the precursor was really kind of hallucinatory and crazy. V-O-3 got a ton of the physics down, but it still has this sort of like plastic-y look that you can just clock. but whatever they did with Gemini 3 Pro image is really pushing the photo reel is much farther. Very exciting.

What else is going on here? Nanobanana Pro edit this image and face swap with Sam Altman, slow show thinking, Nana Banana Pro. Is that a good face swap for Sam? It's just okay. That one's a five out of ten, I think. Let's see.

Someone is dropping chat GPT and saying, I don't want to play with you anymore. Wait, but this is... No, no, they're dropping Gemini. 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 GPD 5.1 Pro because it's rolling out to all pro users.

It declares clearer, more capable answers for complex work with strong gains in writing, help, data science, and business tasks. What is this? That's exciting. I made this one, but it only turned you into a Lego, Jordy. What did it do to me? Look at John. What did it do to me?

Look at John. What did it do to me? I'm just a human. It missed me entirely. What's going on here? We got Lego Jordy. 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. Wait, I know what you are. You're already a Lego. That's funny. That's funny. Well, speaking of Turbo Puffer, sign up today. Serverless vector in full tech search, built from first principles and object storage. Fast, 10x cheaper, and extremely scalable.

So GPT 5.1, have you had a chance to take its first spin, Tyler? What's the latest with GPT 5.11. Well, so the main new thing is the new model is, it's not 5.11 because that came out like what, two X ago. It's 5.11 codex. 5.1 Pro. Well, so there's codex, but there's also 5.1. Pro for like research tasks, I believe. Okay. But anyway, Codex.

How are we doing on the benchmarks? Oh, oh, and you have a take. You have a take. Give me your take. Yeah, I mean, basically. So I think the main graph or the main kind of benchmark that everyone is now kind of watching is this one from meter. Yeah. Show us where the goalposts have been moved to most recently. Where do we move the goalposts most recently? We still need to get goalposts. We do need goalposts. Okay, so we move the goalposts from like, you know, just surprise me with something that's remarkably human. No, because that's totally qualitative. You can't mention 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 MMLEU, like, there's math questions, and then at some point you just answer them all correctly. Yes, yes. So it's not like interesting. Okay. Where this is like, it, this is a benchmark that you could keep doing in 30 years, right? Because the time just goes up and up. Yes, yes. 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 time that a model can do, and this is just on coding task, but it's kind of generally applicable.

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

And the new model is like perfectly, basically on that line. Yeah. So I think it's like, this is just a great sign. Okay. So how long, like what time, what task duration measured in time would you, would you say qualifies as AGI? Yeah, I mean, I don't know if it's exactly, I don't know if that's my definition of AGI because I think there are a lot of tasks that take a long time, but it don't really require general intelligence. 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. Yes. I think if we get up to there and I guess you can, I mean, you can just chart it out to see if you follow this path. How long would that take? I mean, you said it's like basically human lifetime. Isn't that your? That's my, I think that's my correct benchmark. But I think that's wrong because if you think of like build a, build a company, that's not your entire life.

lifetime. That's like, for some people, it's only like... 30 years. 30 years? Okay. Four years maybe? Yeah, I don't know. 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 go create value. Just go exist. And then it goes and does whatever it needs to. That's when it's like truly, like, you know, embodied, I suppose. I don't know.

All I do know is that you can go to Profound. Try Profound.com. Get your brand mentioned in Chachipit. Reach millions of consumers who are using AI to discover new products and brands. I guess the question is this task duration thing is so odd because

like, does time move slower or faster in AI world? You would assume Alex Cart moved faster. You should be able to manipulate time. in the computer, right? So,

you know what I mean, right? Okay, so, so what I'm saying is that like, is it like, if, if GPT-5 can do two hours of, if it can work for two hours without losing consistency and still complete long tasks, 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 bringing this curve down. And so you have this weird countervailing force where, like, I would expect a computer to be able to do problems faster than humans. Right? Yeah, I mean, so at least over time. Compared to, like, how long it takes a human to do it, right? Oh, is that what this is? It's like 30 seconds. It's a two-hour project?

Answer a question. Like, five minutes is count words in a passage. Huh. Find fact on web. Yeah, it's like compared to how long it's. takes a human. Interesting. Okay.

So they have to... Because obviously you could just like, if you... How are they going to benchmark? I always thought this was like, come up with a prompt. Like, do they even have a prompt that can, that theoretically could take months to do? And it wouldn't just be sleeping? Build a massive company. That takes months, years.

Yeah, years. Is that where we're going to be with this meter chart in like, what, six more doublings or something like that? Yes. That's what they're, yeah? I mean, you think so? That seems... That would be comparable to, the time scales. Yeah.

I just, I just, I wonder how they're mapping that. Just 10 more doubling, sir. Yeah. I mean, it certainly does seem like good progress. And I mean, everyone, I feel it very much in the sense of like, just the amount of work that a single prompt can kick off just feels like it's doubling for sure. Yeah, for sure. I think this is just a good benchmark.

A lot of people, it's getting harder and harder to find good prompts that show a model is actually better. Yeah. 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 should be. So this is like a good model. Yeah. Like this is we're on track. Well, if you're looking for sales tax AGI, head over to numeral.com.

Let Numeril worry about sales tax and VAT compliance for you. Well said, John. meter says, what is my purpose? All you put new A-bi-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 update the graph.

There seems to be a mistake. I planned on assessing risks from automated AI R&D. That's funny. They're having fun. What else is going on 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&D automation feedback loop eventually speeds everything up. We've got to have the meter folks back on

the show and understand this a little bit further. I really wonder how they're actually developing the prompts. I really want to get their take on protein, the amount of protein in fast casual concepts and see and potentially get them to chart that out as well. Okay, and then someone took this chart and put it next to the AI 2027 graph, is this correct? So there's meters data, GPT5, Codex max. It looks exponential, but not super exponential. Is that what this read is? Yes.

Yeah, so this is still in the log graph. You see the blue line is the meter. Okay. And then the green is the AI. So AI 2027 was expecting like even more of an exponential. Yeah, and I think that's mostly because they thought. agents that would help develop the next day. I would come a little bit sooner. Okay. Interesting. But I think Gemini III seems to do very well in the kind of computer use stuff,

which you should imagine should like greatly help out kind of agents. Yeah. So maybe there's just maybe there's a month or two, you know, ahead. Oh, so you think we're going back to the green dots there? You're optimistic? You think I mean it's reasonable. 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-2020 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 GPT5. 0.1 Codex max triggered drastic AI acceleration or autonomously replicate. They concluded this was unlikely. Survey said, unlikely. But obviously, big growth in the capabilities. Sweebench, I wanted your reaction to this from vals.a.iValation, different eval. But with this company, they say Gemini III is number one on the independent sweepbench leaderboard.

Yes, this is their own sweet bench. It's also they did not test actually the newest open AI model. It's not codex max. Codex max. Okay. X high or whatever, like the maxed out. Yeah, yeah. I don't know, yeah, they named it poorly, but...

For the 10th time. Yeah. 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. Yeah. And there's also rumors of Anthropic releasing a model soon, and I assume it would be Opus 4 or 5. Right? Because that's like, they have like three tiers.

They have the Haiku sonnet and Opus. Yeah. So I'm very curious to see where all those end up. Very curious to see where Doug landed with his 100 gram protein. He got the 100 gram max protein bowls from Sweet Green yesterday. They got three of them. No, it's the core research. The core team across. 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. So if things go badly, it's listed out. If things go well, we can big bulk on Nvidia. A little update, I ate a little over half in my tummy hurts. What is this? It's not good. Protein Max is tummy hurting.

Protein max is the semi-analysis of... Phil Aaronstein said yesterday, you're telling me the CEO of Sweet Green is on TBPN that he's a chad with slick back hair, a golden tan, and a sick leather jacket, and his handle is Johnny Nemo. Annie's a vibes guy disregarding surveys, and he added seed oil-free,

106-gram protein bowl. Sweet green won me. I love it. Doug spoke a little too soon yesterday. He said, I survived the Great Bear Market of October 29 to November 19. Let me tell you about public.com, investing for those who take it seriously,

multi-asset investing, trusted by millions. NVIDIA emerges successful. And yet the market is selling off. Invita saw its shadow. Six more months of bull markets is high yield hairy. Although, who knows the market is tanking still? The NASDAQ is down 2.1% now. And Bitcoin is down at $86,000. They're 5% today. Significant sell-off. Let's check in on the sailor himself. Also down 5%. At least he's tracking the underlying asset. Meltem says, a Nvidia earnings call for 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. Michael Burry is still going incredibly hard. This is the circularity chart that I was calling out. I think that Nanobanana could pretty much one-shot this. I don't know if it would be as as overlapped and nuanced and like it's not quite there. You can't just say make it more overlapped.

No, I don't think you can yet. I mean, this is, I mean, we're really, really close. You got to talk to somebody that has been making graphics for media companies like this for 30 years. It seems so easy, but if you, especially. It takes a lot of, it takes a lot of. Yeah. And especially if it's like, it takes a lot of deep thinking and reasoning. It takes so much deep thinking and reasoning.

A model could never do this. No, they will be able to. They will be able to. 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 ridiculous demand. True end demand is ridiculously small. Almost all customers are funded by their dealers. If you can name Open AI's auditor in one hour, you win some pride.

What does he mean true end demand is ridiculously small? It'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. He's a desal with a crazy P-Doom. He's a decal with a zero P-Dume, I guess. I don't know. I do think he has, yes, if you're looking at the amount of investment happening now in comparison to the demand.

And you don't believe that the products will get better at all. If you don't believe that... It just flipped so much. 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. And then people were like, let's invest a trillion dollars in that. And it's like, okay, well, at that price, it's actually kind of crazy. I don't know. It's a lot to deal with. Yeah, but if you think about any industry on Earth, do we think every industry on Earth will be using 50 to 100 times more tokens within five years? 10 years? You don't even have to be that much of a... No. Of a...

Of a... Of a... Of a permable to believe that. Yeah. In fact, it's like hard to argue. Tyler... Tyler's permibulling. 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, Automate 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? Top Quiz, Hot Shot? Do you know what movie that's from? I do not. You got two quizzes. This one is from. No.

No? Goodwill hunting. That's it. That's the Goodwill Hunting image. You don't know that. Do you know what Pop Quiz Hot Shot is from? No. That's from Speed.

It's issued to Keanu Reeves. Is Speed worth seeing? 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.

Anyway, I'm in. Back to the Goodwill Hunting Image meme that I love this format. It's a very fun. It's a very fun way to illustrate and 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 20s with GPUs.

You will cling to that until next month when you hear Jim Chanos 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 core weave CDS chart. And you'll repeat that too without realizing that it was just dealers hedging credit portfolios, 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 CAFX is already contractually locked. None of that came from a crash narrative paperback or a Chano 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 stage of the largest infrastructure build out in decades. I love it. Very fun. Let me tell you about Figma. Think bigger, build faster. Figma helps design and development teams. Build great products together. Oh, Sunday Robots is coming on the show.

Very shortly. Maybe we should... We should play the video now. Let's play the video now. A little teaser. Pull it up. Very cool. So, this is kind of a combination of the R2D2

form factor with a humanoid. Yes. Look at that. Picking up two wine glasses is insane. I love the way it just, bounces around. So this is sped up, presumably. Yes.

I think it's at like a, I think this is at like a 10x speed. Something about the lighting here leaks it looks CGI to me. I know it's not, but it looks CGI-ish. I'm fascinated. So many questions. Says it's in autonomous mode. Sunday has speed up.

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 like 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, 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.

And Shulte Douglas over at Anthropics says this is insanely, insanely impressive. I agree, it is. I'm excited to ask Tony how much they had to speak. spend to get to this point. That would be interesting. I'm assuming it will be quite a bit less than many of the other players that are kind of competing here. The little telescoping pole is very cool.

Taylor says it's the hat non-threatening lid. I agree. Just throw a cool hat. 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 or, or weird, you know, like. Think about how scary, like, some of these humanoid robots would be to a one-year-old. Like, Wally looks kind of dumb. R2D2 looks kind of dumb, but it's still like a friendly, you know? You don't want it to be overpowering. Like the optimist or figure would be, like, traumatizing to a one-year-old. 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 anything. insights in seconds. Andrew Reed says every deal is a special situation if you're enthusiastic enough. Very funny. Let's move on. SAM 3 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 Sam 3? Is that Google as well? That's meta. That's meta. Whoa. This is segment

Anything. Yeah. Oh, okay, okay, okay, got it. Okay. Why is it on Google research then? That's funny. Oh, it's how to segment videos with segment anything, Sam 3, 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. Very cool. You can track all of our gong hits, potentially, for velocity. That would actually be. velocity relative to the audio volume. And you can understand how the production team is doing their job to lower the levels for you.

But if we had a live, like, speed tracker. Yes. Like, as you're swinging it. Yeah. Very cool. We could do it. Diet Coke tracker as well. We could automate all of this.

Shield has great coverage. Grock says Elon is more fit than LeBron and would win a fight against Mike Tyson. Fact check. True. You're absolutely right. I'm going to ask Grock if this is true. Is this true? Do somebody do that? I bet somebody did that 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 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 bowls don't glaze to this level usually. 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. Many people are doing this. I mean, you went in SORA and you said, depict me as a bodybuilder. Yes, that's true. And then somebody tried to hack it to give you small legs. They did successfully prompt engineer me. They got you good. They did. They did get me. But, Yes, I mean, I feel like at this point, like, 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 3. Very benchmark driven. They put out the model card. There were a bunch of demos that went out. There were some clear examples of next value coming from the model. Yeah.

sort of a buy-the-book launch and unclear how much this helps the GROC brand to have something like this leak out. But certainly funny. Kevin Weill, friend of the show, says, today we say, Hello World from Open AI for Science. We're releasing a paper across 13 examples of GPD-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 this type of posting has been heavily contentious in the past, but they are continuing to share their work. Yeah. And I think that this stuff will eventually be, you know, fully, you know, peer-reviewed. And also there's just this interesting dynamic where like the other labs, they won't really let you get away with anything. Like, they'll fact check you so fast. Yeah. If this is, if this is seriously impressive, like, you'll probably see some congrats from... There's also a dynamic where if you are using chat GPT to accelerate your own research, are you going to, are, is everybody going to stand up and yell, hey, I use chat GPT for this? Are they going to be like, my research?

Yeah. You know, who, you know, I'm not sure that a lot of people that are leveraging the tool are going to be quick to give open AI credit or an AI credit for something that they're working through. Yeah, yeah. But if you'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 spin up white label wallets, signed transactions, and integrate on chain infrastructure all through one simple API. Byrne Hobart says, Prussian New Yorker cartoon

that saw prediction markets coming more than half a century in advance. Wow. June 27, 1927. 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 a.m. They arrive at 10.20 a.m. And then there's odds listed there because, of course, you'll want to bet on when the plane lands. now you can maybe, you're close to being able to 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 chat chit. We mentioned this earlier. It's official. They're rolling out globally.

There was a successful pilot with early testers. Group chats will now be available for all logged in users on chat chitpT free go plus and propens. I didn't know there's a chat. GPT Go plan. That was the India plan. 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 it makes a ton of sense. Very, you know, we'll have to test this out and see if it's actually that use.

useful. But yeah, I think this is, I mean, this is the kind of thing that can help OpenAI build more of a moat outside of a brand and just general distribution mode. ChatGPT is turning into a social app. Sam pulled it off before Zuck could make the meta AI app good enough to compete with chat GPT. It says Yuchin-Gin. Axen Grock 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, okay, 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 how we... I send... If I do a deep research report, I'll send it around to people in the organization here at TV PN every once in a while. I'm wondering how much it makes sense to keep the chat running in chat. So they certainly do get value out of like putting together the queries, sharing that, sharing the whole theory. I don't know how much I'll be a DAU of this in a month. I'll need to test it out. But we have our next guest in the restroom waiting room.

Tarek from Stute here in the studio. Welcome to the show. How are you doing? We're saying it correctly, right? Yes. I mean, first off, we have to say we love the brand because we love day job. Legends. Thank you for supporting them. They did our brand?

Excellent taste in branding agencies, of course. The best. But please introduce yourself and introduce the business as well. Hi, I'm Tari. And you got it right. It is stew. It's actually played rugby after college. It means prop in South African.

Okay, after college. That means you were on the pro. track or? No, I was like just guy who wanted to drink some beers every now and then. Okay. Let's go. Let's give it up for those guys. But anyway, underrated guys. Thank you. I'm here to announce our series A led by Andreessen Harwitz for $29.5 million. Also participating is active and in coastal adventures. There we go. There we go. So we 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. Yeah. And I think there's a lot of sloth. and AI are 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 Perkin Elmer, Bishoplifting, 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. 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 kids game. And so we plug in. at five to nine. So you could punch out and go to that game. And that's really the, you know, what we're about here at Stu. And then on the messaging side, do you feel like your customers,

uh, want to know details about the technologies that you're implementing? Do they care about that? Well, you have everybody claiming AI. Like, I'm not Matthew McConaughey. You know, I got a brush on Matthew McConaughey. Yeah. And I have to compete with an AI technology versus Matthew McGahn.

It's almost impossible. And so, you know, our branding reflects our customers. You know, think Clippy, which day job did a great job with. Yeah, totally. And really helping, yeah, helping them be nostalgic. But it's like what software first promised you. It was going to automate things. Yeah.

But instead, 40 years later, you know, you need professional services, consultants. It just doesn't do the job. What's the best business model for this type of business these days? Consumption based, seat based. Success based. Success based. Success based. percentage-based.

It's almost like you talk to the team at day job to team me up with these questions. Yeah, we actually didn't. Yeah, I actually didn't talk about this, not with them. You all, I mean, you all buy software. It's so confusing. I'm not a smart man. And I go on and I 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 $60,000 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. Sure. That makes sense.

How does it, like, how is the actual, like, product design work? Is this, like, an agent that gets integrated into communication channels? Like, what does it actually look like? Yeah, I'm glad you asked. So we do audio, so we'll actually place AI phone calls. We'll do emails. We'll, you know, even do SMS and WhatsApp. app 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 IVR tree 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, 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 look. Yeah. 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? 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.

Yeah. 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. I know. So we want to make things as simple as our customer. Most brutal fall off of all time. I mean, look at me.

It's brutal, right? I love Chipotle. I know. But 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, Bishoplifting, reduced their invoices by 35% past due. And have been able to free up that cash flow for other things. 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 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. Yeah, that makes it sense. That's amazing. Well, I got to say the chat absolutely loves you. They love you. This guy is great. This guy is a man. Midwest sensibilities in Manhattan is extraordinarily powerful. He's even drinking your

Bermate, what a freaking legend. True king. I'm liking the sound of this stute. Stute. Let's give it up. No, I love it. I mean, I love an idea that that when you hear it, it's just

totally obvious. It's like applying, you know, there's like the capital war happening in like customer experience right now. They're using a lot of the same technology. You're applying it in a very clear way in a

different part of the org. And I'm bullish. So thank you so much for joining. Really appreciate you having a on and have fun out there. We'll see you back for the B. Yeah. We'll talk to you soon. Cheers. Cheers.

Let me tell you about adquick.com. Out of home advertising made easy and measurable. If you're launching a new company growing, get on adquick.com. Get some billboards. Our next guest is in the Restream waiting room. Let's bring them into the TVPN. There is. Nikita from Flection. Is also a day jump? Is this a day? No, no, no. You got it. This is Tony. Oh, Tony. Hey, sorry.

We got mixed around. Tony, so great to have you on the show. I'm sure your 24 hours, last 24 hours have been absolutely crazy. We played your demo on the show earlier today, and we're absolutely blown away. It's really tremendous progress, and we're excited to meet you. So before we talk Sunday, we'd love an intro on yourself, background, and all that good stuff. 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. How can we learn from human demonstrations? How can we learn to reinforce from 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. So this is why I co-founded this company Sunday with Chung, who is also a PhD student at Stanford, which leads to Memo, X-1, and all these new advances. Incredible. What has it taken to get to this demo that you released yesterday?

because I have no idea how much money you've raised up until this point, but it feels like you guys have accomplished a ton in a pretty resource-constrained way, at least compared to companies that you're competing with in the sort of like helpful humanoid in the home category. Yeah, absolutely. So we're functioning in a very efficient way. And I think as an early-stage company, we think about as a blessing that forces us to innovate. and finding out these solutions

that are orders of magnitude more efficient than like 20% efficient and 30% efficient. 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. Yeah.

John? Yeah, I'd love to know some of the tradeoffs. Also the chat is mentioning you forgot to mention, you forgot to mention, you worked at DeepMind, Tesla, and Google. So sort of a non-traditional background into robotics startup. Yeah, what are the key tradeoffs? I mean, there's a lot of focus right now on teleoperation.

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 teleoperation at all. other folks who say teleoperation is a really useful tool to pull forward some of the capability. Where do you stand on the issue? Yeah. So I think teleoperation 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 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 teleoperation to gather the amount of training data, it would take decades for sure. Because robotics is a harder problem than self-driving. 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. So what we design instead is actually have it here. is called skill transfer glove, skill capture glove. That is one to one to Memo's head. Oh, interesting. And yes, the idea here is that if you can wear the glove and do a task, Memo can also do it.

Okay. And that essentially decouples 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 cloud data. Yes, but, I mean, just to go back to the question of like capital intent, intensity, 1% of 8 billion people, that's 80 million gloves. If the glove costs even 10 bucks were back in, you know, you need a billion dollars to get your data set or something like that. You don't think Tony can raise a billion? I'm not saying you can't do it. I'm just saying like, 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? Yeah, I think robotics are at 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 have, of course, pre-training 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 role models. Whoa. Okay. Then how do you see the data capture from the glove scaling? 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 attractable 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. Yes. So I think this question is more about, like, for us, how can these data generate value so that we can keep this loop going on?

And it's kind of similar to the whole large language model space that we need to spend a lot of money into compute. But the model itself is generating a tremendous amount of value. And for us, we don't need to solve, robotics to ship a product. That's a lucky part. Sure. And in homes, there are lots of, like, it's one of the few places you can do relatively

simple tasks, but give people a huge amount of value, both emotionally and functionally. 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 home robotics is a harder problem earlier, if I heard that right? But at least it's lower stakes and that if you have an end. if you drop a dish, it's annoying and you want to avoid that, but there's not, like, nobody's

going to, like, die. Yeah. Yes. It's like the newer start of the company is to solve robotics. Yeah. But we don't need to solve robotics before we ship a product. Yeah. So, yeah, talk about timelines.

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 robots. And the general availability of memo will be either 2027 or 28, depending on the progress we've made through the whole beta program. Talk about form factor. Why not give it legs?

I'm sure you have a reason for that. and I'm curious because 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? Yeah, absolutely. So the way we designed this robot is super safety and they're 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 robots, can it stay stable or not? Interesting. A weird robot is actually like one of the few ways. So it can't fall over and crush your dog or even your foot, basically. Yeah, or just like damage the floor. All sorts of stuff.

That makes a ton of sense. And then also I imagine that 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 as 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. But there's a willingness to pay, at least from 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. What was the, what was the key design inspirations? What matters to you with design? Somebody in the chat was asking if you were influenced by Homestar Runner. Oh, yeah. It is sort of Homestar Runner. That's hilarious. 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, I think, detail that we decided to do is we do not put camera into the robot's eyes. The camera is actually right underneath its hat. Yeah, I saw that. 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. So we kind of intentionally avoided that. And, yeah. That's very interesting. Yeah. Yeah, the design, we were talking about earlier, it feels like it really just avoided, like, the uncanny valley, the creepiness.

Like, there's a lot of risk factors when you're designing humanoid robots. robotics right now. We've seen all sorts of them. They can look cool, sci-fi, but maybe weird or in certain contexts. I think this one came across very well. Well, super, super excited for you. Thanks for coming on and breaking it down. This is really fun. Thank you so much.

If we'd love to be in the, in the demo program. Yeah, we've got flat floors here. We got flat floors. We have huge messes. And we have a team of people that we will make wear these gloves all day long. And we will take care of, we will take care of Memo. We will. Because he's, because he's cute.

Yes. And we want to see him win. So thank you so much. Congrats on all the progress. Congratulations. Thank you guys. We'll talk to you soon. Goodbye.

You know what we got to do. We got to get memo a watch on getbezzle.com. Yes. Let's get memo a hitter. We said out. Lachery watched out. Fully authenticated. R.M.

Bezels, team of experts. I need an RM. He definitely needs a rechardt. He definitely needs a reshardt. Definitely needs a rechard mill. Why not? Why not? Next up, we got Nikita.

from Flexion, excited for this one. Thank you so much for taking what's going on. To talk to us today. Thanks for waiting. Good to meet you. How are you doing? Hi, I'm really excited to be here. I'm Biggian. I'm the CEO and co-founder of Flexi. Where we're building the intelligence player or the brain that powers all kinds of robots from humanoid manipulators. Yeah, I mean, fantastic.

We were just talking about the humanoid robotics. How do you see the market playing out? So 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, you know, 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 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 uh it needs to be able to identify objects figure out uh what it what is like what is something that is worthy of just throwing away what is something that like i don't want

thrown thrown away i'll be frustrated so i feel like there's like a lot deeper uh more levels of complexity to a lot of these robotic tasks than than just like object manipulation and kind of 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. Absolutely. Let me just say that Sunday is amazing. I think their videos are really, really impressive, probably the most impressive demo that I've seen so far. 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 bring it to a garbage can.

Yeah. 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 that knowledge is already contained in large language models. to think of it as GPT5 for 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 really good at understanding that.

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. Great, you know, can have a look at that video. Yeah, so say more, or maybe even just narrate the video.

Sure. So let me just quickly come back. We're bad on simulations. We train robots using reinforcement learning, not to imitate humans, but to solve specific tasks and just through trial and error. So we have robots trying millions and millions of times. I think 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. But in this specific video... 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? Yeah, absolutely. That's a great point. Usually we call it the cimterial gap. Yes. And once you train in simulation, the whole challenge is to cross that cintorial gap. 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 real. 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. A good example, I think, is if you want to train a robot to open a door, either you have to simulate every single possible door that exists in the world

with all the textures, the lighting, et cetera. Or you can use a model like segment anything. And then you paint the door in red and the handle in, let's say, green. Then all doors kind of started to look the same. Look the same. Interesting. And then you're basically training the motion against the segment-any-thing version of the door of the world. interesting.

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

All of this is super important. It's all about the timeframe. So today we're using physics-based simulators, just like Unreal Engine. And that's actually my take, this is enough for way more than what most people think. We can go a long way with Jodo simulators. And explain that. Is it that if you have a physics simulation that's running fine, and let's say Unreal Engine, you might use something else, but Unreal Engine, Do you think we're on a scaling curve where if you had a million GPUs running a million instances of Unreal Engine generating simulated data, that that would actually result in better progress on the robotics side, on the actual decision-making and planning side?

Yeah, exactly. That mix with one more thing, which is generative models that can create assets for simulation. Okay. So you don't need to have humans coming up with a million different versions of... all the things that the robot needs interact with. 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. etc., etc. But you're saying that generative AI can create even more variation. Is that the idea? Yeah, exactly. You actually have two ways to add more variation easily.

One is something like Goshen Splats, where you go outside, you collect real world data. Sure. And suddenly you have a lot of assets. And the second version is you ask, turn to the AI to do it for you. Yeah, yeah, that makes sense. That's cool. Any news? Yeah, what you got?

We announced this morning that for the first time they were raised 50 million. Oh! Congratulations. Who participated? Thanks. A bunch of investors. DST Global, Nvidia and Ventures. Versed David, Prozos, first, windfire.

Awesome. And then where are you building the company? You're in Europe, or have you moved over to the West Coast? So right now we're all in Zurich and Spitzern. This is why you see the robot walking in our nice Alps. Oh, yeah. 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 as I can office here.

Oh, nice. That's great. Well, good luck. Yeah, if I were you, I wouldn't be tough to leave. Zerick's pretty nice. Fizzling 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. Yes, peace. But congratulations on the milestone. Super exciting. And if you ever have hot takes on robotics, feel free to let us know. Oh, bye. Thank you. Thanks so much.

Awesome. Cheers. If you're planning to go to the Alps, Book of Wander with inspiring views, Hotel Great a Meny's, Dreamy Beds, top-tier cleaning in 24-7 Constellier's service. Transition. It's a vacation home. But better. Did you see this that we?

We don't understand why ice is slippery. I don't know if this is fully confirmed at this point, but Massimo Rainmaker, 1973, shares new research, misspelled there. 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, this is one of the bigger,

firm beliefs I held that got overturned. Like, I really, I like it. It's like, this is the one of my strongest beliefs. I knew that the world is round. The sun rises in the east and it sets in the west, and I know that. that the reason ice is slippery is because of a microscopic layer. It'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 can 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.

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 suddenly it feels like you got a banana under your foot. One of my friends was worried about getting canceled because the first tweet he ever posted decades ago when he first got on Twitter was, I just slipped on some iced.

Hashtag F-I-E-E-C-K-I-C-E. And he was like, am I being like rude or en-couth? Should I delete that post? Jackson Dahl pulled out 12 lessons. From our interview on Dialectic, his podcast. I don't think we'd ever written down a lot of these ideas. I think we've certainly talked. We were talking about the need for principles and the need for some sort of culture.

But that was more like operating principles within the company. Some of these are relevant. Yeah, this is more about the style of content. Good summary. You can't copy compounding. If you want to know more about us and how we think about the show behind. behind the scenes, you can go listen to Jackson Doll's latest episode with none other than yours truly. On our own very set.

On our own very set. Yeah, we filmed it here. The Dialectic Pod. Aiden says, just so we're clear, anti-gravity is a windsurf wrapper. Windserve is a VS code wrapper. Vs code is an electron is a chromium wrapper. Chromium wrapper. Chromium is a C++ wrapper.

C-plus is a C-Rapper. C is an assembly wrapper. Assembly is a machine code wrapper. Machine code is a binary wrapper. Binary is a physics wrapper. That's kind of a big jump there. Math is a logic wrapper. Logic is a philosophy rapper.

Philosophy is a humans wrapper. Humans are a carbon wrapper. Carbon is a star-forge matter wrapper. Stars are a gravity wrapper. Gravity is definitely not an anti-gravity wrapper. 19K like people enjoy it. This is very funny. This is funny.

Robin Hood had a post, trade the forecast, weather market predictions, and Augusta says, Ha-ha. This is how Augustus can, one way that he can monetize is just, you know, getting a hedge fund, betting on the weather outcomes. It's not going to rain. I'd like to say it's not going to rain with size.

I was like, where is Augustus? Is he in town? Is he betting on? He would be, I don't know, would that be in, would that be investigative? Would that be insider trading? We'll have to figure it out. Sotia Adela had a bad. banger. Barely AI says, never forget Satchinadell in 1993 as a Microsoft technical marketing manager

showing how Excel works. We can play this clip. This is funny. 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 going to 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- Call in questions now.

No way. He's doing a live stream, basically. It means it's on TV. 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 NTS and its server, connecting to the database. It sounds like agenetic AI. Sounds like a workflow that's getting automated for numbers. 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.

Suno raised more money. We had the founder on the show just a week or two before the rounds so we didn't have them back on. But congrats to everyone over there. And the regulatory deals are getting worked out. 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 kind of front running them 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 going to be pretty fun to play around with.

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 henhouse, 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 Deep Mind. And anyways, so this, I'm excited to play around with the product when it comes out. 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? I mean, the Minecraft Ferrari Gemini 3 actually looks something like a... I think I like the Gemini 3-1 too.

GPD 5-1 Pro doesn't look anything like... 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'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 that with? I'll close up by saying it's pouring rain so hard that I'm hearing it through our earbuds. Okay, fun.

So if you are in L.A., be safe. 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 Sager coming on. It's going to be a fun one. I'm sure he, I'm sure he'll have fully 180ed on AI.

Yeah. Our biggest AI bowl. We got semi-analysis and then breaking points. We're trying to bring you diverse perspective. We really are. We really do care about that. We don't want this to be an echo chamber. We obviously have strong views ourselves in many topics, but we're here to bring, to have real conversation.

So thanks everyone for tuning in. Thank you for tuning in. Thanks for dealing with the chaos in the chat. Yeah, very chaotic day in the chat. That was our first time being, like, raided. Yeah, we got rated a little bit. It was really funny because they made, they made, they made, like, seemingly like 20 fake accounts. Had 20 accounts.

And then they were really angry at Ben, for some reason. And then they also were really angry at Merckhor. They kept dunking on Merckor. Oh, yeah. Well, why are they made of Merckor? 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. Well, the chat was getting wild. But we appreciate you all, and we will see you tomorrow. I love you. Goodbye. Cheers.