OpenAI Launches Codex, Meta Delays Llama 4 Behemoth | Matt Grimm, Kevin Weil, Blake Scholl, Tim Fist, Chris Best, Sean Henry

28:07Yeah. Thank you to Figma for supporting the show. They should also get on Vanta. Automate compliance, manage risk, improve trust continuously. Vanta's trust management plan. Back to back. It takes the manual work out of your security compliance process and replaces it with continuous automation, whether you're pursuing your first framework or managing a complex program. Anyway. Thank you, Vanta. The other news is Semi-Analysis is talking about the AI deals that have been happening in the Middle East. Dylan Patel coming on the show in a couple weeks. U.S. strikes a deal with the United Arab Emirates and the Kingdom of Saudi Arabia, a five gigawatt data center, the Humane G42 Diversion and American AI wins.

28:45Dylan Patel breaks it down. He says, the US-Saudi slash UAE AI deals are complete wins. Everything is there to win over the Middle East and accelerate infrastructure spend. So American companies, AI companies are going to be making more money from this because we have a new customer. The main subtext here is that China is locked out of Middle East AI infrastructure investments. There's a different world. There's a different path where, yes, there were some controversial moments in Trump's tour of the Middle East. But there is a different world where we are watching Xi Jinping do that tour and we're watching from the sidelines.

29:18That's not what happened. What happened is our president went there and struck a bunch of deals. We would have cried for the first time on the show. Instead of Sam Altman and Elon Musk and Alex Wang, it could have been the founders of Huawei and DeepSeek and HighFlyer. all of these companies. There's an alternative timeline where, you know, the QIA is investing, you know, $100 billion. Not actually$100, but massive amount of money into Manus, right? Or DeepSeek or some of these other players. Are you talking about Manus? Manus? Manus? Oh, we can talk about open source AI agents. Manus? With Manus, you're going to make me put my mustache back on, John.

30:01No, no, no. We have a very important person coming on the show, Jordy. Put that mustache down. Put that mustache down. Put that mustache down. We have the chief operating officer or chief product officer of OpenAI coming into the studio. And so the UAE ruling family used headline grabbing investment numbers, deals with Trump's family businesses, and ties with tech executives to score an NVIDIA deal. And it's all looking good. Let's see if we have Kevin in the studio. Let's bring him in. He's here. Fantastic. How are you doing? This is wild. I just like join a zoom and I'm live with you guys. We're live.

30:35Yeah. It's great to have you. Welcome. I was watching you on X and thinking how weird it was that in 30 seconds, I was going to be on the screen. This is awesome. Yeah. We're leveraging this thing. It's called the internet and technology. But we'd love to get you up to speed on it since. It's funny. A funny story. So Tyler Cowen called into the show before one of your guys's earlier releases. And he was like, AGI is here. Like he was basically saying like, it's a couple of days out, but he couldn't get his camera working. And it was just like funny, funny dichotomy of like AGI is here, but like the internet is still actually.

31:10There was actually another day when we had to basically take the whole show down because Zoom and both Zoom and Google Hangouts just went completely down, like nationwide. And so we were like, well, we can't do our show now, I guess. The running joke is that audio video is like ASI complete. Yes. Yeah. Well, everything else in the world is solved. Yeah. So, I mean, I think long-term for the show, we need to go proprietary. We need to build our own video streaming stack. Can you help with that? Codex. Can Codex help with that? You know what can help with that? Yeah. Codex can help with that. Break it down.

31:45How would I use Codex? And by the way, you get to talk about AI and then you're going to have Blake from Boom here. So you get to talk about supersonic flight afterwards. You guys have the coolest lives ever. It's full stack. Full stack technology. But yeah, can you take us through the announcement, break down exactly what launched, when it's available, to who, what you're excited about, and then we'll go into some of the trade-offs in the product design and development. Yeah, let's do it. So we just launched this morning Codex, which is a cloud-based software engineering agent that can work on many tasks in parallel.

32:20Wow. So a lot of folks are used to the kind of cursor windsurf style, you know, GitHub co-pilot style of AI development, which is really more about augmenting a single engineer, right? You're writing code in your IDE and you can press tab, tab, tab, and it'll auto complete and you're, you know, 10, 20, 40 % faster. With Codex, you actually have a software agent that runs in the cloud that can do entire tasks for you. So you give it a task, it goes off and does it and suddenly you have a PR and you didn't work on it at all. And it's powered by a version of O3 that we fine tuned specifically to be really good at these kind of hard software tasks.

33:01We're super excited about it. It launched today inside ChatGPT. So you can use it if you're, it's rolling out now to pro, to enterprise, to teams. Well, it'll make its way to plus in the coming weeks. But the cool thing is it's a, it's an agent, it's a software agent in the cloud. So, you know, you're using it today from ChatGPT. You can imagine using it in the future from, you know, your terminal, your IDE, all kinds of other places. And even, you know, hooking it up via API to your bug queues and just having this software agent churn through every single bug that you have. You know, for each one, look at the context of the bug, understand your code base, and then suggest proactively how you would fix that bug and give you a PR that you can review.

33:47So the world of software is changing. We're super excited about it. It's a research preview. It's not perfect yet, but I think this is the future. Yeah, so I mean, I've already noticed that ChatGPT has been writing code for me for a while. It seems to be writing more and more code. I was telling Jordy about how I wanted to know the height of a desk in an image and I knew how tall a person next to it was. And I thought that it would just use images in ChatGPT to kind of one-shot this or guess, but it wound up spinning up and looking at the individual pixels with a bunch of Python code. I think it wound up writing like 5 ,000 lines of code or 500 lines of code.

34:26And it got it really right, unsatisfactory, because it was just an average size table. It was like literally the standard size, but it really knew it. But so I'm wondering, is this something that will feel like a deep research project or deep research functionality where I click a button to say, hey, let's use codecs for this. I'm giving you the hint. Or is this something that can be automatically triggered just from a text interaction like images in ChatGPT? Yeah, so it's a, I mean, by the way, you never know if that desk was actually like 38.2 inches. So yeah, it's worth it. It's worth it. Write the 500 lines of code.

35:02Why not? It's too cheap to meter. Actually, that's been one of the big, the coolest thing about O3, O3 personally for me has been a kind of feel the AGI sort of moment using that model, the things that it can do. And it's a lot of it comes from the fact that it can use tools while it reasons. So it's thinking. And in the process of thinking, it can do some web searches, and then it can take what it learned from a web search and write some code. And then after that code, it can do image analysis, and then it can write some more code and do another web search. And then finally, with all of that context that will output the answer, it's like, it's really been an unlock for a huge number of use cases.

35:41And that's the kind of thing that enables Codex here. Because you're right, ChatGPT has been able to write code for a while, right? You can just go to ChatGPT and type in like, here's a, you know, write me code to sort this array of integers that I have. And it can give you the code and it's going to be great at it. The difference here is Codex is built to work on big complex code bases. So you're not just saying, like, do this little task for me, write this function. You're saying, I have a bug and I don't know where it is. There's a, you know, 100 ,000 line code base. Can you please go understand my code base and try and fix this bug?

36:20Or I'm a new engineer at a new job. I'm trying to understand what the heck is going on in this code base. Can you explain to me where the code does X or Y or Z? And very quickly, it'll look through the code and give you an explanation of how something works. You know, it's funny. I've even seen examples where people go, hey, Codex, find a bug in this code base and fix it. Just find one randomly. Just go. That's amazing. Do complex stuff and do hard work on a huge amount of existing context that differs from what you've been able to do in ChatTBT for a long time. Yeah. Wild. From a personal perspective, I used to write Python pretty regularly.

37:03I haven't written much code, so I haven't really gotten into the cursor windsurf world. Obviously, I've been writing code via chat GPT now. I noticed recently I kicked off a deep research report and it prompted me to put it on effectively a cron job. It was like, do you want me to just run this for you every week? And I said, yeah, that sounds awesome. That's great. But if I'm like, I don't have a repo, but I could set one up. Is there a world where me as kind of like a prosumer, non-technical user should set up a repo to house the custom code that Codex writes for me to make a better experience for all the little custom software and tools and random stuff I use?

37:47Or should I just live in the O3 world where the code is pretty much ephemeral? I think it depends what you're looking to do. You know, for simple things where you just want to like quickly put together a script or something, writing it inside chat GPT and not using version control and all of that is fine. But if you're going to do something that you expect to be longer lived, like it's the basis for something you actually want to build for yourself and maintain, then I think setting up a quick GitHub repo and using codecs on it makes a ton of sense. By the way, so I used to be an engineer. I still dabble and screw around on the side and write code, but nothing major.

38:26And I hadn't written any code at OpenAI. I've been there about a year and haven't checked in a thing. Like Tuesday night, I think, I was doing the rest of my work. And I was like, I want to fix a couple of bugs. And so I went and found a couple of really basic bugs because I didn't want to screw anything up. And sent Codex off to work on both of them in parallel. Checked back in a few minutes. And I had two PRs. They looked right. So I submitted them, got them code reviewed by somebody, you know, who's actually a good engineer. And they were submitted. And now I've got a couple of commits in the code base.

39:06So it's just like this is stolen valor, though. This is stolen valor. You didn't write that code. But it's like I actually just kept, you know, other than just like getting to play around the product and offering a little bit of feedback on a couple of things. I was off doing the rest of my work. And I had this software agent working for me in the cloud writing code. Yeah. I'm so curious to hear about the kind of internal testing process and when you guys decided the right time to actually roll this out as a research preview. Because I imagine you've been using codecs in one form of another internally for like a very long time.

39:48Maybe it didn't have a name or anything like that. But I'm sure that ChatGPT has been contributing to effectively the ChatGPT code base almost since the beginning in some form or another. Yeah, for sure. I mean, we're big users of our own tools. There was a version for a while that was mostly about the kind of how do you come up to speed quickly in a big code base that was good at understanding our code base and answering questions about it. It was really popular with new engineers on the team. And then once you kind of get to understand it, you might not use that tool as often unless you're exploring a new area of the code base.

40:31So that was like a proto version of this. But we've been working a lot over the last six months at improving the ability of our models to code. Like you've got GPT 4.1, which we released a little while ago, which has very quickly become a really popular model. Now, I think, default in Windsurf, it's increasingly a large percentage of cursor users coding. That came from focusing on the things that matter in creating a really good coding model that you can rely on. It means really good instruction following, longer context, the ability to not just make the changes, but to make the changes a way a developer would.

41:12So don't add a bunch of extraneous stuff. Don't add weird comments. Make surgical, precise changes that accomplish the job. So there's a style element to writing good code, not just a correctness element. And we've been focusing on all of this. and then you kind of bring that together with O3 and all of the things that we were, you know, O3's ability to tool call and to reason and suddenly you can put together a really good coding model. So we've been thinking about this for a long time. This is the first time when we're like, okay, this is now good enough that we think it deserves being a product for the rest of the world and we're excited to see how people use it.

41:56Can you talk a little bit about product design, inspiration, product design? I noticed that the very first iOS app had these incredible haptics when the tokens were streaming through that I hadn't really seen anyone do. I just opened the app last night and saw that when you're using voice mode to dictate to it, it has a different modal now. It feels like there's a very strong design language evolving. at the same time there was you know people complaining oh i can't even i'm using chat gpt i can't even stay logged in that that bug obviously got crushed pretty quickly um but but what what is the actual product design inspiration are there people that are pulling from certain uh schools of thought or anything or like is there someone driving that internally or is it just like baked into the culture yeah for sure ian silver leads our design i was fortunate to work with him at Instagram.

42:49He's an incredible designer. Building for ChatGPT is a really interesting thing because we have well over 500 million weekly active users at this point. So it's a big scaled product. And so for products of that size, one of the most important things is to simplify, right? You're not just serving power users at that point. And so you're serving people that are just trying to get something done in their day. They don't care about the complexity. They don't care what the models are called. They just have a task and they want to complete it and you want to help them. But on the other hand, we have folks who are super deep AI enthusiasts and want to, you know, digest in every single new model that we use and try it out in all these different ways.

43:36And so we want to both and we want them to we don't want to sort of soften the edges of their experience. We want them to to be able to do everything they possibly can to experience all of the power of AI. And so we both want to like simplify the experience for for a lot of our users. And we want to provide the people that want it all of the bells and whistles. And so we try and we try and balance that. um you know so we you try and make it so that you don't need to worry about things like the model picker as much you don't need to like have a bunch of ai knowledge in the background to do what you want to do in open ai or in chat tpt but if you have that you should be able to expose the sharp edges and like test the different new features and stuff and so we really actually try and get both of those things right and it's a delicate balance how do you is that why is that why you guys don't seem to put too much emphasis into perfectly naming products just because in the on a long enough time horizon it doesn't really matter i just come to you know chat gpt and i work with it to get the outputs and the results that i want and i'm not regardless of my experience level i don't necessarily care the underlying sort of models doing the actual work wait are you saying are you naming isn't great i was no no no no no to be clear we think that you have the second worst naming after behemoth which is a very very untameable you know demon demon monster potentially a disaster so we are now long 4-0 and 0-3 and we just i just i just you know only six only six months ago a new model would get announced and people are like oh it's so confusing blah blah blah yeah but it just it seemed to me you know as an observer that you guys like it wasn't like oh this is a problem and we need to fix it it was just more so like let's just keep making really great models and make them easier to access in really intuitive ways yeah i mean in all in all seriousness it comes from our focus we have this principle of iterative deployment that we really believe in which is that these are new these models are new systems right they're Each one, each model has capabilities that we understand somewhat.

45:56And also we discover new things about it. And we believe that no matter how many smart people we have inside of our walls, there are way more smart people outside our walls. And the best thing we can do in a world of AI evolving so quickly is to kind of co-evolve with society, to ship stuff early and ship often and learn together. And so one of the reasons that we have this explosion of models is we're trying to build new capabilities rapidly. And sometimes the easiest way to do that is to kind of build it into a new model that's really good at one specific thing or a handful of specific things, but can't do everything.

46:35And so you end up with this like profusion of models that do different things well, like 4.1 is really good at coding and instruction following, but it's like not as chatty. and so if you're asking about other things you might prefer 4.0 for some things and 4.1 for others seems totally natural you'd go like well why don't you just build one that's you know good at coding when you're coding and good at chatting when you're chatting and we will do that that's what we're trying to get to with gpt5 where we're trying to bring more of these things together but if we tried to do that from the beginning we wouldn't have been able to launch as fast and so we've we've opted for like launching fast having a little bit of you know confusion that comes with it, but we learn faster.

47:16And then over time you sort of integrate and simplify. I guess like the, the, the meta question though, is like, is the future just, you're already using mixture of experts within the models is the future, like a mixture of mixture of experts models. And so I, I, I go to one, there's one command line text is the universal interface, not drop down model pickers. And, and, and it routes me, it says, Hey, this person doesn't want to chat. They want to write code. Okay. We're using 401. That seems kind of logical. And already this is kind of happening with the model picker becoming like just the UI is getting less and less in your face.

47:50And it's a little bit easier to just have a natural interaction. But how do you see it evolving? Yeah. I think over time, the capabilities that have existed for a little while, you've sort of learned how to bring them into a general model. But you're always going to have these new frontier capabilities that you're going to want to be able to iterate on really quickly and you might want to do specialized things to make the model really great at some new frontier capability. And so I think, yes, ideally you have a sort of model, you know, a layer above the models that's doing the choosing for you.

48:26But that is, you know, it's a hard problem, especially it goes back to the building for simplicity versus building to enable power users. As a power user, you might be the only one that knows for a particular question you're asking whether you want a 80 good answer immediately or you'll wait a minute for a 95 good answer or whether you want to do deep research and wait 20 minutes and get an amazing answer yeah i mean you could theoretically like train the user on that a little bit like like well it's because there are also like how you work with the comp that i or like my personal framework is like when you're working with people on your team or teammates there's certain people you'd work with that you would have to explain effectively like the exact tool set that they should use to accomplish the task you should get the crm and you should go in the crm and do this and that and then there's people that are maybe have greater intelligence or experience or context and you just sort of like discuss the task with them and you're not even thinking about the underlying sort of like toolkit to accomplish the task it's just sort of this higher level conversation.

49:31I want to talk about A-B testing versus personalization. When you choose a default model or the default prompts, when you open up ChatGPT, it says, create an image, write a Python script, make up a story, what's in the news. There's a couple options there. There's a lot of personalization going on, but you could also imagine doing A-B testing to understand what will drive turn down or retention up. How are you thinking about the balancing act between those two techniques of product development? I don't think they're really in at odds in any way we do both. So we definitely A B test a lot of things, because we're trying to learn what works and you know, how, how we can help people understand this new kind of strange world of AI.

50:17It's a funny, it's a funny product, right? You're we're used to products where you You have a UI, like computers before AI needed very specific inputs, like this button does this specific thing and that button does this other thing. And if you wanted to do a third thing and there wasn't a button for it, you probably just couldn't do that thing. Right. But then every time you hit the button, you got the same output. It was very consistent. LLMs are basically the opposite. You can give them input that has the full complexity and nuance of the human language, and you have no limits on what you ask.

50:54And then also what you get out is not the same from one thing to the next. It might be substantially the same, but the words are not identical. And so it's a totally different way of building product. And when someone comes to ChatGPT for the first time, if they just hear from their friends, hey, this AI thing is super cool. It can do all this stuff for me. And they show up at the front door of ChatGPT. They're a new user. What's the mental model? Because it flies in the face of almost everything that you've learned using computers over the last however long. So we really think a lot about how we get people going and how we teach them all of the different capabilities.

51:33Which, by the way, the capabilities are changing every month or two, too. So it's a really challenging problem. but something that we care a lot about because that's, you know, if you go from being a novice chat GPT user to being a power user, it can really change your life. It can save you a ton of time. It can accomplish a lot of tasks for you and that's only increasing. So the upside of us being able to teach people well is also increasing. Yeah. I feel like a decade from now, people are going to look back at this moment and realize that the people that fully understood the capability, like the full capability set of the models just had this ridiculous sort of extreme advantage.

52:17It was the same thing with social media. Like the people that really understood and took social media seriously early on are like famous now, like actually famous. Uh, interesting. Uh, what, what is your postmortem on the sycophancy thing? I feel like that made news because it was kind of blanketly, uh, uh, like it was kind of like everyone was experiencing or at least all the power users were experiencing it. But I could imagine a situation where some people really like that type of interaction and it was beneficial and it actually may improve their lives. And so if you go to the YouTube algorithm right now, and you only search and click on positive content that reassures you, you can have a sycophantic experience.

53:00And that can be good for everyone involved. So how do you what is your postmortem on it and how do you think that the personalization will play out in the future? Yeah, this was a really important issue. I mean, so the story for people who don't know, we rolled out a new version of GPT-4.0, which is something we do pretty regularly. We're always, to your point about EV testing, we're testing new versions of GPT-4.0 that are incremental improvements over previous versions. So we rolled out a new one and, you know, we may be tested in the past. So we, you know, the metrics look good. It looked like a really solid model, had some new stuff around personalization.

53:43And then as it got out there, we saw that a number of use cases, not like super widespread, but enough use cases where we saw the model sort of overly like, like just being, some of it was like glazing what people called glazing yeah we call it that i think we use that term yeah but then but there were other cases that were more uh more serious totally where someone had real problems and you know maybe they were having mental issues and the model was sort of validating them in ways that didn't really comport with reality and that's like a that's a real thing. And we took that super seriously. So we rolled the model back.

54:30And then basically have spent the last few weeks diving into where this is coming from and what we need to do to, to make sure it doesn't happen again. And we've tried to be super transparent about it. So you know, we tweeted as soon as we were rolling it back, and then immediately put out a postmortem, like, you know, within a day or so, and then put out a second after we had done a bunch of deep dives and so we've gone through we've like the team uh did some great work we've got evals now that measure this um we understand a bunch of the root causes from where this came from you know as always with these things they're not it's never just one thing it's like a little bit of this combined with a little bit of that and then this unexpected thing happened and together they created something that that you know wasn't up to the standards that we set for ourselves yeah so So I just have an interesting realization with the product.

55:29So we've been hearing this request for feature on social media for a while of like, I wish I could just reset my algorithm, start fresh, because I feel like it's funneled me in some sort of echo chamber. And I don't like that echo chamber. And I want to start fresh. And I don't know if social media feeds actually have that feature. It might just be buried. But I've noticed that with the Chachapiti memories, early on, I was really aggressive about prompt engineering and basically like prompt hacking. And so to get the best responses, if I was trying to learn about trains, I would say like, I am a world expert in trains.

56:03I own multiple train lines and railroads. Give me a breakdown of the market map of trains, basically lying to it. And then it remembered that. And so now it's like, well, as a train conductor, you'll probably want to eat this for dinner. and I'm like, okay, I have to back up. I wasn't being completely honest with you, ChatGPT. But the good news is that the saved memories are there and I can delete them all. And so I can kind of reset my experience. But was that a learning from the demand that people are seeing and the stated preference on social media for resetting? Or do you think that that's important?

56:42Or what other lessons are you learning from how social media has played out? because you obviously have a lot of experience there and a lot of people at the team have experience in social media. Yeah. Personalization is a really powerful thing that I think we're just at the very beginning of. In the same way that we know each other a little bit, you have best friends that you know super well who you're really comfortable with and then you have strangers and your level of comfort in interacting with them is very different. If you have a super assistant in your life in chat gpt you wanted to know you really well you wanted to know your habits and how you like to do certain things and you know even down to like do you want uh you know more flowery supportive language or do you want crisp analytical terse language things like that um i've i was uh messing around last night with chat gpt trying to do um trying to give my son some math homework.

57:40And he, I just said, Hey, can you design 10 math problems for Matthew? And the model knew Matthew was my son, knew he was 10 years old, and developed a bunch of like grade level appropriate escalating, you know, and it was like, that was super cool. And it even it was like, Oh, he likes Legos. And Kevin, you're a runner. And so a bunch of these programs, or a bunch of the questions were like Lego themed, and you're going on a run with your dad and this and so So like, okay, that's just really cool. It's truly making me emotional. It's like the coolest thing. As like a parent to be able to, as a parent, to be able to give your, to give your children like a truly bespoke magical experience is like almost priceless.

58:25Yeah. Even in the context of homework, like something like homework. Right. And so to that, and, and even knowing that like kids today, like both of our, all of our, you know we've got five kids between the two of us and um knowing that they'll grow up only knowing the sort of world in which this kind of technology exists where a parent can just like generate magic for them in like seconds is yeah just unbelievable i mean and think of the world this thing designed 10 math problems so i used oh three it designed 10 math problems for my 10 year old that were escalating in difficulty across a range of different things.

59:07It could easily, if he was entering in the answers, could easily realize over time where, you know, what, what concepts he understood and what concepts he didn't and just become a personalized tutor for every single kid. And remember, I mean, this is free, right? We don't charge for ChatGPT. You can, you can get a subscription, but you can also use it for free. Basically, you don't even need the count. You need an Android phone anywhere in the world for free and you can get this thing that's starting to be more and more of a personalized tutor. I think it's incredibly powerful. Totally. Every study I've ever seen says that when you pair traditional learning with personalized tutoring, it's like a standard deviations of improvement.

59:54So I'm with you. I I think the future is going to be very different. And there's a lot of reasons to be optimistic about what the next generation is going to be able to do with AI. I need to be more honest with Chet. How do you explain the rate of AI progress to, let's say, like a family member of yours that's not in tech? That's a good question. I think the only way, honestly, because you can talk about it, but it's like you can't get fit by reading about going to the gym. You can't use it. So I've been trying to get any of my family members who aren't using it, just try it. Start asking, you know, for everything that you're doing, ask, why couldn't I use ChatGPT for this?

1:00:42and you start to realize that there are more things that you can say yes to there and you can start using chat gpt and then you know the more you use it the more you realize the value and off you go it's really hard to explain in the abstract right people go uh agents i'm hearing so much about agents what is it and then you use deep research or you use codex and you're like oh wow that just saved me a ton of time or did something i couldn't have even done yeah yeah you know that's the promise of the future i think we're all going to be much more productive we get to not focus as much on the doing of particular things.

1:01:15We get to focus on the outcomes and what we do once some of the labor itself is taken care of. And that's a super exciting future for me. A couple more. Talk to us about HealthBench. Yeah. So a huge amount. I mean, people are increasingly using ChatGPT for health. I've done it any number of times. My son had a small surgery that was supposed to be 99.9 % innocuous, 0.1 % bad. And we got the results back from the doctor before I could talk to the doctor. And they look scary. And it's full of a bunch of medical jargon that even as a former scientist, I didn't understand. And I put it in chat GPT and said, this looks weird.

1:02:05what is this? Is this, should I be worried? And it said, Oh, no, no, no, don't, don't worry, it's fine. And I was like, Okay, explain it like I'm five. And it did. And, you know, I couldn't get ahold of the doctor for another 72 hours. That would have been a bad 72 hours for me if I was sitting there, like stressed out about my son and chat GPT gave me the peace of mind. So we always try and say with anybody, anybody asks about anything medical, you know, this isn't a substitute for actually seeing a doctor at chat gpt is not a doctor but here it is here's here's here's the understanding of what's going on there and it's really valuable you know for all of us for me it saved me 72 hours of of anxiousness for somebody else who doesn't have access to a doctor it might be a totally different thing so um anyways we care a lot about this we want to make if people are using chat gpt for this we want to make sure that the answers that it gives are really good answers.

1:03:00And so we're putting a lot of effort into improving ChatGPT's ability to act as or to answer medical questions. And the only way that you really know if you're doing it right is if you have a benchmark. You've got to have something to test against to show that you're getting better. And we figured if we'd done a lot of work to put this benchmark together, then others could benefit from it outside of us. And so we open sourced it. Last question. Can you give us the 30 seconds on why you were excited to join Cisco's board of directors? Oh, yeah, totally. So, I mean, Cisco is an incredible company, like iconic Silicon Valley software and hardware company.

1:03:43We all use it every day in 100 different ways. and they're at this really interesting point because I think AI can transform their business and they can either be transformed, like it can either happen to them or they can get ahead of it and build some really amazing software and tools and become, you know, sort of an even more powerful leader for the next generation. And I think that's not unique to Cisco. I think a lot of companies are at that turning point, but Cisco really realizes it. By the way, they're actually a launch partner today for us with Codex. So they're one of our early partners.

1:04:21They're looking a lot at how AI can help them get more done, you know, faster, more cheaply, etc. So it's just AI is going to really impact their business over the next, you know, three to five years. And I'm excited to be a part of that and hopefully help them navigate this transition gracefully. Amazing. Well, I have like five more hours of questions, but we will let you go. So we'll come up to us. Most questions can be answered by chat GPT, but there's certain questions. You got to go to the source. Organic farm to table. Yeah. Yeah. Well, thanks so much for having me on. It was good to see you guys.

1:04:58And great to see you, Kevin. Talking about supersonic jets. I wish I could stay for that one. Yeah. Yeah. We'll talk to you soon. Great to see you. Cheers. See you. Bye. Kevin is the man. But you know what else is the man? Linear is the man. Linear is a purpose-built tool for planning and building products. Meet the system for modern software development. Streamline issues, projects and product roadmaps. Linear. And also Numeral is the man. Sales tax on autopilot. Spend less than five minutes per month on sales tax compliance. Put your sales tax on autopilot. There was, Numeral was getting picked up by an anonymous account yesterday.

1:05:35Oh, yeah, we saw the news. They're maybe out there in the market. But Sam said, too low. Too low. You're not bullish enough on sales tax AGI. Did you tell him it's on record or not? Well, it's live. He posted. No, he posted this. He said too low. Okay, so we're not scooping anyone here. We're not scooping anybody. Anyway, we have Blake from Boom Supersonic in the studio. Welcome to the show, Blake. It's been too long. We've been wanting to do this for a while. Yeah. Ideally from a supersonic plane. We're aviation aficionados. We like traveling. but the hard thing with the show is that it takes us six hours to get across the country.

1:06:15We podcast for three hours a day. The math just doesn't make sense. So we're happy that you're working on a solution for us. Indeed. Well, thank you for having me. Yeah. It will be fun to do the first ever supersonic webcast. Maybe we should just agree to do that. I'm 100 % on board. Let's do it. Count me in. Can you give us just the general update? Obviously explain like what you're building, but what's the latest news in the world of Boom Supersonic? Yeah. Well, I think for anybody who's not been following the story, the goal is to really pick up where Concord left off, foment a supersonic renaissance, ultimately deliver faster travel for everybody from the president on down to every family.

1:06:54And so it's a decades long mission, ultimately to replace subsonic with supersonic for every passenger on every route. but building companies like Boom it's like building an iceberg from the bottom up and I feel like this year is the year that the iceberg has really emerged from the surface of the water in January we broke the sound barrier with our test airplane the XB-1 broke in the sound barrier we love to see it right and in February we did it again and arguably we broke it permanently fantastic that's great news You said you broke it permanently? Is that what you said? Yes, because we proved we could do it reliably with no audible sonic boom.

1:07:40Oh, there you go. Okay. They said it couldn't be done. They said it couldn't be done. You didn't believe it. By the way, it turns out that people have been talking to this for decades, and mostly the claims are it's really hard. You have to change the aerodynamics of the airplane and da-da-da-da. No, it's a software fix. Really? How? Explain that. Everything is computer, John. That's it. Everything is computer. If you fly the airplane at the right altitude at the right speed for the current atmosphere, the boom makes a U-turn in the sky and never touches the ground. No way. That's so amazing. But you need to be able to calculate that.

1:08:17That's right. In real time. You need decently good weather data. Yes. And ironically, algorithms that were developed for computer gaming. Wow. That makes sense. Yeah. It's a physics simulation. It's ray tracing. You're running Unreal Engine. I'm sure you have a proprietary system. Does that need to go through FAA approval to use that? Or is that separate from the rest of the aircraft? What's it called? Qualification or approval? Yes, so you need to certify an airplane to carry passengers. Basically, you've got to prove you meet all the safety standards. And this is separate because we have one of the dumbest regulations ever created.

1:08:59in 1973 we banned supersonic flight in the u.s like literally there's a regulation that says thou shalt not exceed mach one yeah and it's speed limit not sound limit right right it's really stupid it's really stupid and um uh and so no you know we could like make it play mozart when it flies over and you're still not allowed to do it and that's why that's why those coast-to-coast flights are still stuck at six hours it's why no one's done this already and uh and so we're we're working really hard to get that changed. We had something that like the EO would just drop and it would be like, Oh my God, he just tweeted it out or put it on truth social and it happened like love him or hate him.

1:09:38But like the guy definitely like likes to rip a crazy idea on short notice for everyone. Are you optimistic that there will be change? Does this need to go through the house and Senate or is this something that could just happen in EO? So we've had good conversations at FAA this week, a bipartisan bill dropped in the House and the Senate that does it. Cool. And Elon endorsed it. Jared Isaacman endorsed it. We've got bipartisan support. Great. Some people think this could get through the Senate unanimously. Amazing. I think that's probably a little ambitious. But even the idea that that's possible, I think it's pretty cool.

1:10:17Some lawmakers like traveling slower because it inspires them to grind harder. more time on the plane more time to prep their filibusters they don't want to go fast i guess that's right the the bear case for supersonic the fast plane is that i get so much email done i do get a lot of email done on a six hour flight i got there faster i'd do less i was texting jordy when we were flying to dc i was like this is incredible i've never been more productive uh but i think faster flights were just forced us all to be more efficient with our time right That's right. Yeah. But I mean, break down the actual scope of the problem here.

1:10:53How much is engineering? How much is regulatory? How are you staffing against that? Do you have a massive government affairs team and lobbyists? Or I imagine there's still a ton of engineers on the – you're not just taking some like white-labeled plane and then doing a bunch of lobbying and that's the end game, right? Yeah. I mean, the most surprising thing is that we don't have to invent anything fundamentally new. This is not a science project. It's not a technology project. It's really just an engineering project. Sure. In fact, we're taking 20-year-old 787 technology, basically reshaping the airplane, making it long and skinny, putting twice as many engines.

1:11:32And so there's a lot of engineering and testing that goes into that. But there's no science and there's no new technology. And there's only one regulation that needs to change. Just the speed limit. So we've got small numbers of great engineers. I'm a big believer in like tiny teams that are very focused. Like we built XB1, our test supersonic jet, with just 50 people. 50 people. Wow. And the overall company is not that much bigger, right? Right. We're like 115 now. And like we're growing very slowly. Like if a team does not complain about being understaffed, I know they're overstaffed. Yeah. Yeah.

1:12:06I mean, we were joking about like white labeling a plane. obviously you are standing on the shoulders of giants using some, uh, off the shelf technology, using some technology licensed from other firms. Can you take me through the journey of engine development, the decisions that you made? Have you changed course on any of those decisions? Uh, what would you recommend to the next generation of aircraft builders? Yeah. Um, well I'll go broader than that. Like if you're, if you're thinking about doing hardware at all, my, my, my advice, frankly, any startup, any startup, all startups are hard. I don't think any or easy.

1:12:38I think the difficulty level is set by founders because we tend to run at our own red line. And if somehow it gets easier, what does it make the job harder again? Like Brian just decided to double what Airbnb is because I guess the old thing got too easy. And so what I've found along the way is I'm far more successful if I pick a mission that deeply inspires me and makes it worth being at my red line. So I never get up in the morning anything, is it worth it? Okay, great. So now probably building supersonic jets is like, you know, the ultimate hard mode. And what I have found along the way is people around the company from the legacy industry will have all these stories about all these things that are impossible to do.

1:13:22You can't build your own jet engine. Like only a big company can do that. There's tons of proprietary technology, blah, blah, blah, blah, blah, blah. It's all bullshit. It's all bullshit. Like I'll tell you one story. Like we were trying to get high temperature super alloys for our engine. And, you know, we thought we needed to go license it from one of the big three. We get into this licensing conversation and they're like, oh, we can't give you this thing because there's a trade secret. I'm like, don't tell me the trade secret. I'll give you the part. Just make me the part and hand me the parts back and don't tell me the secret.

1:13:52And they said, no, no, no, no. You'll find the secret. So we can't do that. And for like two weeks, we're like, oh, shit. Like how do we ever get this engine built? but eventually we just went into the supply chain and we found the trade secret. We found it. And I'll tell you the trade secret. Okay. There is no trade secret. Oh, interesting. The thing that was theoretically proprietary is an open source material developed by NASA where all the specs are public. Well, the trade secret is that they're using open source. that was the trade secret. That's the trade secret. The trade secret was there is no trade secret.

1:14:24And because there's all this fake proprietary and that, So everybody's telling you why you have to work with them, why you have to use their stuff, why you couldn't create it yourself. And just nine times out of ten, it's just not true. And so we found, you know, we probably heard Elon talk about the idiot index, which is like how much a finished part cost is divided by the raw materials cost. We found a thing that's more important. It's called the slacker index. Slacker index is how long it takes to get something divided by how long it takes to actually make it. and so we've got these turbine blades we're 3D printing turbine blades for our engine and we go quote it out of the traditional aerospace supply chain it's going to cost a million dollars for one engine's worth of parts and it's going to take six months I was like well how long does it take to print a blade well it's like actually about 24 hours so why does it take six months to get one well they were like printing them like one at a time and then you got to wait for your turn on the machine all this not okay what's the machine cost two million dollars how long does it take to get a machine?

1:15:24Actually, they've got them in inventory. You get them in a couple weeks. For the price of two engines worth of blades, we got the 3D printers and the blades and we beat the lead time of just outsourcing it. That pattern exists everywhere in this business. Can you talk about the current fundraising market for hard tech companies? There's a bunch of companies raising a ton of money. There's other feast and famine out there. where the F-35 cost a trillion dollars. Somehow I feel like trillion dollar raise is on the table soon based on the current market. Yeah, we'd personally like to see you do one T on five T.

1:16:03I'd love to see that. I'd love to see that. One T on five T. That should be the new goal. But I mean, there is a world where a plane company comes out and raises just massive money and just dumps it all into subcontractors and takes a very different approach. Why wouldn't that work? And what advice would you give to the next generation of hard tech founders in aviation or otherwise? Yeah. So the key difference, so there's this mythology that hardware companies are more capital intensive. And if you go look at like how much money did Uber raise? How much did Lyft raise? How much did Stripe raise?

1:16:36How much did Airbnb raise? Like these theoretically capital-like businesses consumed billions in venture capital before IPO. So wait, what? And if you go look at SpaceX or Andrill, the theoretically hard tech companies, oftentimes they raise less money. So WTF. And I think that the difference is if you are building, say, Uber or Airbnb, you have an idea that sounds really counterintuitive, like invite strangers from the internet to sleep on your couch. Right? It doesn't sound like a good idea, but it's actually really cheap to test. and so what happens in most internet businesses is very small amounts of capital allow you to test whether you have product market fit and you test it by means of building and shipping the product we can't build a supersonic airliner as a means to testing whether anybody wants a supersonic airliner so what you have to do in a hardware business is find a capital light way to demonstrate that you're a product market fit And the answer can't be shipping the product.

1:17:41So, so, so, you know, our strategy was pre-orders. Yep. Um, so we, you know, we got United American to make like deposits, like non-refundable deposits on airplanes against a specific design. So it's not just like, okay, here's, you know, here's a dollar that says, you know, supersonic is cool. It's like, Nope, here's a significant multimillion dollar deposit, a whole further deposit schedule against a very specific airplane with very specific specs. And then so we can go to investors and say, if we build this, there's obviously a gigantic market, and this is going to be like a thousand extra money.

1:18:16We're already getting paid for it. You guys are already getting paid for it. That's right. Yeah, so all you have to believe is we'll actually ship the product. What was the, I want to ask you, what was the worst week building boom, and what was the best week? I'm guessing the best week was this year. Yeah, I think the, I'll give you the best moment. The best moment was the day that we broke the sound barrier for the first time. But it wasn't actually the moment of breaking the sound barrier. We'd done so much testing at that point. I knew if we flew that day, we were going to break the sound barrier.

1:18:51What I didn't know is whether we'd have the weather to do it. And I'd gotten up that morning and it was extra cloudy. We can't do it on a cloudy day. and so the team was kind of debriefing checking the weather and one of our safety culture rules is like no top management are in the safety briefs it's the team making an independent go no go decision with nobody senior putting their thumb in the scale and telling them they need to go and pressurizing it so I'm like in the hangar waiting and the team walks out and I could just tell from the energy like it was go time like nobody had to say anything it's just like i see him walking over the airplane they're like hooking up hooking it up to the hook the tow bars up like we're gonna go and at that moment i really teared up uh because i i knew i knew it was going to happen and then then we were and then like when it actually happened and everybody was jumping up and screaming i was sort of like yeah of course so that was uh i think that was the best moment the worst moment um we have near-death experiences like basically every year.

1:19:57And at this point, I just expected it. I'm like, okay, every year we're going to get the startup equivalent of a cancer diagnosis. And if I'm lucky, it's stage two, not stage four. But we survived stage four startup cancer. We had an extremely difficult fundraise. I think we got down to a week of cash. The lawyers had a plan to shut the company down. I remember calling one of our investors when we were sort of three weeks away. Um, and he's like, usually when people tell me you're three weeks away, you're telling me you're going to shut the company down. And I was like, of course, we're not shutting the company down.

1:20:32We're talking about how we get through this little knothole here. And, uh, like who have like 12 months of cash and they're like, I'm out of this. I've been doing this for a year. I'm going to go be a PM in big tech. I've done the PM at big tech. I don't ever want to do that again. No, I think I'll never go back. You could never make me go back no i can't go back the thing the thing with your with for you and your team and i'm sure the entire cap table it's like if we don't do this we have to be if we're not successful here we're going to be reminded multiple times a year every time yeah like yeah every time you get on a plane every time you know it like is is one of the very few companies on earth that can bend reality into into an, you know, bend time.

1:21:21Right. And I, and I just think, uh, I feel like humanity is fortunate that you are running this company because so many people would have gotten the stage four diagnosis and just said, all right, like we, we took a good shot. We did our best and that's it. Yeah. I mean, it's my biggest wish for other founders is pick the startup where if you got the stage four diagnosis, you say, screw it, we're beating this. And that, and that's that, you know, for me, that's supersonic flight, but every, for every founder, it's something different. But it's like, there's this founder market thing. And if you're a founder mission thing, and if you get that really in alignment, then like, then you can run through brick walls.

1:22:02Yeah. In many ways. I would run through brick walls for a note-taking app. I would too. In many ways, you could think of boom as being a very logical, obvious next step in commercial aviation, long haul travel, that type of thing. Yeah. What about the flying car market? I've often said we have flying cars. They're just helicopters. but the problem with helicopters is that they're not evenly distributed. Not everyone has a helicopter, but if everyone had a helicopter on their house, they could fly to work and yeah, you need air traffic control and stuff. We all need to be good pilots. But if a company could drop the cost of a helicopter by a hundred X and make it a hundred X safer, yeah, we'd probably basically say, yeah, we saw flying cars and yet that hasn't happened.

1:22:57So what is your take on the boom of helicopters, Whether or the boom of flying cars or yeah anything else They're great people working on that actually and it's actually way harder than supersonic flight way harder why but because Zillion reasons like one is you need a whole new set of regulations, right? Because these the electric vertical takeoff and landing so Joby and Archer who are really the two leaders in this There it's not technically or regulatorily a helicopter. There's a whole new set of rules Then you need a whole new set of infrastructure. But is that a mistake? What I'm saying is like, what if I started a flying car company that was perfectly regulated as a helicopter?

1:23:41Just like there are these companies now that are regulated as sea gliders, Billy, Thalber from Regent. Billy and Regent are doing great work. Yeah, it's like legally a boat, right? Yeah, I'm sure you love it, right? It's like legally a boat, but it's like a plane, but he's designated in one way. So it's a lot easier. And I feel like there's almost like this regulatory arbitrage, regulatory hack that the hard tech founders might need to think through. So if the regulations were done well, your idea would be exactly on point. The problem is we have too many regulations that are prescriptive about exactly how something has to be done rather than setting a safety bar or a noise bar.

1:24:20And so if you look at the regulations for helicopters, they tell you how you must build your helicopter. Okay. So if you want to build an electric helicopter where it's distributed rotors, well, it doesn't have the parts that the regulations tell you how to design. You know, or like how do you treat, you know, you're allowed to have like a wing on the thing too. So that like the better helicopter is that doesn't fit the regulations for helicopters. Yep. Yep. So yeah, it almost be better. It's a major bug in the regulations. Yeah, almost just to regulate against like, if flying things are crashing at a rate above more than like one in a billion, it's no go.

1:24:59It doesn't matter how you build it. Instead of saying, you have to have this screw in this place. That's right. And actually for large transport aircraft, it's actually a bit better. One in a billion is actually the regulatory safety bar. And yet there are also still some things that are like prescriptive. but there's a mechanism to get around the prescription called equivalent level of safety. So we can go to FAA and say, you say that our throttle has to be built in such and such and such a way, but the real thing is a reliable throttle. Let me show you my reliable throttle that's done differently.

1:25:37And you can get an equivalent level of safety finding. You can go forwards. But if you look at going from a helicopter to an electric vertical takeoff and landing, you know flying car uh like the whole thing is different what's your what's your timeline to e-vitols being in american skies who wins who goes first will i be in a boom supersonic or uh or a e-vitol yeah i mean just just kind of a wild guess is it like before 2030 seems unlikely to me Is it 2040? I feel uncomfortable speaking for other people's timelines. I think the technology is going to be ready before the infrastructure and the regulatory environment are laid flat.

1:26:26And the stated timelines for eVTOL are super near term. But I suspect it's optimistic. I don't know. Maybe we've made the mistake of being more realistic and now it just sounds worse. You know, so our timeline is 2029 for being ready for the first passenger. I mean, there's also like incredible things happening in eVTOL that are unmanned. You look at what Zipline's doing. We talked to the founder of Zipline and we were blown away by the progress there. And that was one of the things where they had to go to another country to get favorable regulatory treatment and kind of figure things out, but then have really been able to accelerate.

1:27:02Yeah, I mean, Keller's crushing it. Like, I love Zipline. It's crazy. Amazing founder, amazing company. great success story for how you do these things. They're brilliant. Keller is brilliant at finding the first market that was the soft target. The small and uncrewed are radically easier than large and crude. And the reason for both is safety. If you go big, even if it's uncrewed, you have a problem because if the thing crashes, it can hurt somebody on the ground. Yeah. Right. And so there's a threshold, you know, somewhere around, I think, 55 pounds, like above which like it's radically harder.

1:27:44Yeah. And then if you put a human on board, like obviously, you know, you don't want to hurt anybody. And so and so in building XB1, like the the the most surprising things were about the second order effect of our choice to put a human on board from day one and not to have an ejection seat. We basically put everything on hard mode, but we learned a lot more from that. Yeah. Is there an easy mode version of the boom story where you go to some country with lax regulations? You say, we're going uncrewed, we're going smaller. Maybe it's like a ISR reconnaissance application. You're only flying over the water, counting on fish or whales or something.

1:28:27And it's just much easier to actually get flying on a routine start to flywheel like Zipline did. I mean, you could go, you know, like there are people doing kind of what you're describing that I think are smart, like Hermia's Hermia's is smart. Yeah, totally. And, you know, but what is what is Hermia's doing? They're basically building a hypersonic bomber. Yeah. And, you know, and it's kind of OK if the trips were one way. Sure. You know, and yeah, and I think they'll have a great business and they're great people working on that. And, you know, I'm cheering for them. They'll probably get to Mach 5 long before I get to Mach 5.

1:28:59Yeah. But doing that and then later putting a human on board is extremely difficult. If you look at the history of development of aircraft platforms, you find that defense technology makes it into commercial technology. But there are basically zero cases of a defense product becoming a commercial product. But the reverse is not true. There are many cases of commercial products becoming defense products. So the Boeing 707 became the KC-135. The 767 became the KC-46. The 737 has turned into a whole bunch of command and control airplanes and anti-submarine airplanes. So I think the reason is when you're commercial, you have to worry about safety and you have to worry about noise.

1:29:50And neither of those things is retrofittable. They're actually foundational to product architecture. And so technology can go in either direction, but products go in one direction because safety and noise are fundamental to product architecture. You mentioned AJ at Hermes. When are we going to see the foot race? I want the two founders of the supersonic companies to race each other to see who's faster in the real world on the ground, and then we'll race the planes in the sky. I think AJ and I have been telling each other the other one's going to win the race. So I don't know. I think it'd be like the most, at least for me, it would be the most pathetic race ever.

1:30:24I am like not a runner at all. Well, we'll do it as a pay-per-view. We'll raise some non-dilutive capital. You guys can maybe split it based on. Let's actually make it a little bit less about the running. Let's just do the Murph. Yeah. We'll just do the Andrel Murph together. Well, we're going to hold you to being the first people to podcast Supersonic. Yep. And I would love to have you back on the show many times between now and then because this conversation has been extremely insightful. and we are just grateful for the work that you and the team do. It's great. Thank you so much, Blake. That was so much fun.

1:31:00Thank you. We'll talk to you soon. And in the meantime, we will tell you about public investing for those who take it seriously, multi-asset investing, industry-leading yields, trusted by millions. Trusted by millions. We'll also tell you about 8 Sleep, five-year warranty, 30-night risk-free trial, free returns, free shipping. How'd you sleep last night, Jordy? I got an 89. Did you smoke me again? 100! clinically backed sleep fitness. He's got a 100 sleep score. They got the new Pod 4 Ultra. Pod 5 Ultra. Pod 5 Ultra. With a blanket with the same heating and cooling technology. You got to pick it up.

1:31:35Can't wait for the blanket. Anyway, we got Tim Fist from IFP in the studio. We remember he was here when we were under attack and all of Zoom went down. That was devastating. That was devastating. It was such an interesting conversation. It was. Technically. Yeah, it was awful. So he's at the Institute for Progress. We'll bring him into the studio and talk to him. How are you doing, Tim? Welcome back. Thanks for having me, guys. We made it. We made it. Thank you so much for bearing with us. That was such a weird, you know, it's not the first time a nation state is trying to take down TVPN. Maybe we start talking and we get spicy and all of a sudden the connection gets fuzzy.

1:32:14It might happen again. We don't know. Yeah, we had Tencent was pretty upset at the H20 allegations. So prime suspect number one. Take us through the allegations again, break down what you guys published, and then we'll go through some of the reaction and the fallout from the piece. Yeah, so much has happened since this was the news. It feels like there's five different things in export controls that have gone on since. But yeah, basically, this was the H20 inference chip from NVIDIA, which some might remember was designed to be compliant with the export controls. So as this chip, they sold about reportedly 1 million of them into China in 2024.

1:32:55And yeah, there was a lot of outcry at the time for the Bureau of Industry and Security, who's the part of government that administers and enforces export controls, to do something about this. Because 2024 was the period where everyone realized that inference compute was perhaps the most important strategic input to frontier AI development because of test time compute scaling, reinforcement learning and synthetic data generation is the key things. And, yeah, you know, earlier this year, there was reporting that, you know, Nvidia had a huge number of additional sales plans to like big Chinese companies.

1:33:27And a bunch of people, including us, sort of said, hey, is this really what we want to be doing? Do we want to sort of allow, you know, China to get access to millions more of these chips? And, yeah, the government ended up taking action on this and issued some guidance, basically saying, hey, no, you can't make these sales. Nvidia reportedly was left holding the bag to the tune of about$5.5 billion. And now, you know, since then we've seen... It's a bag for ants for NVIDIA. It's barely a flesh wound. Well, they made up for it with deals in Saudi Arabia, right? Yeah, indeed. Yeah. So that's the new thing, right?

1:34:00It's the administration walking back on the diffusion rule, the speaking framework, and then the series of deals that have been announced over in the Gulf. Yeah. And what's your reaction to the diffusion rule and the deals in Saudi Arabia? Is this a step forward? Are you excited about this? Is this positive news? Yeah, so it really depends on the details of all this. So I think on the deals, I guess fundamentally, you know, what do we want? We need the US AI tech stack to win the global competition against China. And I think a big part of that is locking in these early adopters and big spenders, especially like the United Arab Emirates.

1:34:35But I think we need to be thinking about how to structure deals like this to sort of get the US the outcomes at once, which is, you know, US tech diffusion. the kind of like Chinese tech stack locked out in a way that, you know, we couldn't handle with 5G, like China won sort of like the 5G battle globally. And then appropriate National Security Guard rails in place. And it's pretty unclear, like what the trade-offs that have been made for this deal are. So I think like the high level specs that we've gotten is a five gigawatt AI data center campus to be deployed over some time period in Abu Dhabi, and then reports of 500 ,000 chips per year to be exported with maybe four-fifths of those for us firms who are building data centers over there and one-fifth for g42 this big tech conglomerate in the oe and kind of depending on how quickly that all happens and whether they're referring to you know uh 100 000 of today's chips or 100 000 of like chips in uh you know five years time this could be the difference between you know uh 1x to 100x in terms of like different uh differential in compute so yeah i think that really makes the difference.

1:35:36And I think the key question here is, do we want an AI lab in an authoritarian country to have the biggest clusters or close to the biggest clusters in the world? And this is a country that does collaborate with China in areas like drones and 5G and military technologies. And so you need to sort of be careful with giving them access to sort of frontier scale compute here. Yeah. So there's kind of like a diffuse... I'm curious if you have insight on capital flows from UAE and Saudi into Chinese AI. is how much investment activity is there? Do you have any insight? I don't have stats on this. I suppose the interesting thing to say on this, you guys might remember there was this$1.5 billion deal between Microsoft and G42 a couple of years ago that sparked a lot of this stuff.

1:36:22So this was essentially Microsoft making investment in G42 and starting to build data centers in collaboration with them. And the Department of Commerce got really involved in this And one of their requirements was to divest from Chinese companies in the kind of AI stack for G42, which is, you know, this huge tech conglomerate that is funded through the nation's sovereign wealth fund, which is the second largest in the world. So pretty serious money. We're talking trillions of dollars. And yeah, reportedly what they did is they had a bunch of passive investments in Chinese companies, including, you know, hyperscalers and AI companies.

1:36:56And reportedly, G42 did divest from those companies, but then basically moved the investments over to another fund called Lunate that was sort of also owned by this, you know, big sprawling tech conglomerate that's ultimately funded by, you know, the sovereign wealth fund. So the question marks about how much are they actually decoupling from China and like how costly is this to them? Like, are they actually burning bridges that are hard to reverse? Yeah, there's some element of like we have a smooth gradient of friendship with different countries. Obviously, there's the five eyes, like the closest allies.

1:37:24we sell nuclear submarines to some countries, but then there's countries that are more jump balls and could go either way. As the diffusion rule goes away, do we need firmer rules around, hey, we're willing to sell to you, but don't immediately set up a reseller and start selling, just passing these on to China? Because you could imagine if the natural economic forces take hold, and I was in some country that was just about to get 500 ,000 chips, it's pretty easy to just immediately start reselling these and just print money, basically. Yeah. And there's kind of two ways to do that, right? One is you can just on-sell the chips, so kind of smuggle them into China.

1:38:05And the other way that's pretty straightforward is set up your own data center and rent it out as cloud computing. Exactly. So yeah, I think that's what you're referring to. And yeah, currently there's not great guardrails around either of those things. And I think the administration wants to set up better versions of this. But yeah, I think this needs to be part where there's sort of like countries that are really willing to, you know, buy hundreds of thousands to millions of chips. You kind of want these structured deals that have these guardrails in place. And I think the Trump administration is really, really well positioned to strike these kind of smart, bespoke deals that get us the right possible outcome across each of these dimensions.

1:38:35And hopefully that's what they're pursuing. Yeah. Can we talk about the news today, or maybe it came out late yesterday, that NVIDIA to set up research center in Shanghai, maintaining foothold in China. Basically, they're opening this R &D center as almost like an olive branch to China is kind of how I would describe it. They say they're going to use it to understand Chinese customer demands and design US compliant products. And they're basically doing this to just kind of navigate export controls and compete with companies like Huawei. this uh to me i mean there's so much to unpack here i i love your kind of initial take and then i want to kind of maybe move more high level and to me this signals that that um maybe jensen doesn't take national the national security concerns about like an ai war as seriously as even some of the US foundation model labs talk about it?

1:39:41Yeah, I'd say that's an accurate assessment. And, you know, I think it's really hard to design, you know, sanctions, export controls, like these kinds of things in a way that it's not that sort of can't easily be escaped from. But it's kind of like the spirit of the rule in that, like, hey, we're worried about China, you know, beating the US in like the most strategically important technology of the century. And we don't want you selling to them. But then you can do all these things to actually be cooperating around the scenes, like design, compliance, or whatever. And to be fair to NVIDIA, I think they say, look, if the speed limit is 60 and we're going 55, we're not breaking the law.

1:40:15We should be allowed to do this. And if you think the strategic perspective for them is also a lot of their customers are these big US hyperscalers, right? And all of these hyperscalers are developing their own custom silicon. So, you know, we know Google has their TPU, you know, that they use for both trading and inference. We know Amazon has like Tranium and Inferentia that they're also using for trading and inference. Microsoft is developing their own stack. And so, you know, huge source of revenue for NVIDIA is potentially at risk. So it kind of makes sense for them to want to be, you know, diversifying to other parts of the world and not just be selling to U.S.

1:40:49hyperscalers. So they're certainly in like a difficult position overall. and yeah, what you think is right here depends really on how much you buy this kind of argument that, hey, over the next five years, AI as a technology that will like reshape the global balance of economic and military power is a thing and will be a really big deal. Yeah, it seems like the speed limit is getting lower and lower though. We went from the H100 to the H20. Now, NVIDIA is preparing to release a modified version of the H20 chip for the Chinese market after your piece and the changes to the H20 restrictions. Is there a point where the restrictions are so onerous that no one wants to buy a car that goes nine miles an hour?

1:41:36And at a certain point, NVIDIA will just lose the market share because Huawei Ascend chips will just be outperforming. Are you tracking any of that? Yeah, so here it becomes an interesting conversation. I think the kind of crux of the matter is if Huawei has better chips than NVIDIA in the Chinese market and is able to sort of capture that market, how bad is that? Like, should we sort of set export controls such that NVIDIA is always slightly ahead of Huawei, for example, and sort of like raise those limits over time? And here I think there's two sort of dimensions to it. One is that, you know, in AI, the quality of the chips matter.

1:42:11So how sort of individually performant each chip is, but also the quantity really matters. So, you know, you can have a million chips, each that are, you know, like half as good and sort of like substitute for like 500 ,000 chips, for example. And so, you know, this is a crude approximation, but you both want the sort of best chips and as many of them as possible. And where we're trying to squeeze China, like we being like the US government is across this whole supply chain. So not just on, you know, being able to procure chips directly, but also being able to manufacture their own. So having like the semiconductor manufacturer equipment and the fabs and everything there.

1:42:43And so I think the goal of the US government has to be really to restrict the quantity of AI chips that China, so like SMIC and Huawei as like the key firms here, are able to produce. And so by this logic, you know, even if, you know, NVIDIA has a chip that's only slightly better or slightly worse than Huawei's, you still might want to restrict it because you would prefer them not to have access to 10 times as many chips than they otherwise would have. Like you don't want them to have access to essentially like, you know, TSMC's production capacity of like being able to do many, many millions more chips than you could otherwise produce.

1:43:16So I think, yeah, these are like hard trade-offs to make. I think where it really matters is in foreign availability. I think, you know, where Huawei is accessing foreign markets and outperforming US chips, that's really bad. And we should sort of make sure that that is not the case. And it is US chips that are being used globally in countries that aren't China. Do you think Huawei will ever go public or do they not want people to know, they want people to know as little about their business as possible? Yeah, I'm clear. Depends what the CCP wants. It's an adversely opaque organization. Yeah. Interesting.

1:43:51What is your take on open source AI? We were talking to Aaron Ginn about this idea that if America does not provide a state-of-the-art open source stack to countries that want to build their own AI products on top of fine-tuned or post-trained LLMs that meet their definition of free speech or ideals or morals, the stack by default will be Huawei Ascend, DeepSeek, Manus. Yeah, I totally buy this. I think that being able to sort of have the best open source models in the world be American is really important for a similar reason as like the chip stack and the data centers and the cloud services.

1:44:38I think there's a question about how sticky is this ecosystem actually? So you talk to people who are like, yeah, we really need to lock in the tech stack globally. American open source models need to be the rails that the world runs on. But then if you look at how AI developers work, they are very happy to switch between different base models for their application, depending on which happens to be the best. You look at the revenue of Frontier Labs, and when they have the best model, it's up here, or when they don't, it's down here. Everyone is switching every day, depending on who has the best model.

1:45:07So it's, yeah, like you want to diffuse American open source as widely as possible, but like what about it makes it sticky? And my hypothesis would be that it's probably the security and reliability side of things. So, you know, everyone's worried about, you know, the fact that you can insert like backdoors to create slipper agents into open source models and there's no way to actually detect these. I think, you know, the U.S. wins if it can prove that it has the most trustworthy and reliable models over China. similar to how I think US cloud computing companies compete over, you know, Huawei and Alibaba cloud.

1:45:37Like often Huawei and Alibaba are coming to the market with a cheaper option, but the US is just more trustworthy in terms of, you know, data privacy, security, et cetera. So I think trying to figure out those technical problems around security, reliability, interpretability, and sort of proving that US models win across those dimensions is probably the way to make the kind of open source models from the US more sticky or just sort of be way better. And, you know, this is where sort of most of the effort is currently going and definitely support those efforts as well. Like, you know, building up more domestic compute, for example, and like finding more training data sets.

1:46:08Yeah. How do you think about the dynamic between the importance of the application layer versus the foundation model layer? We've seen efforts on the foundation model layer at the national level all over the place. But let's just use Mistral as an example. Mistral has a consumer app called LeChat. It is a direct competitor to ChatGPT. And I think that's great for the French and they could potentially have a fine-tuned model that meets their standards and guidelines. But if at the end of the day, the French consumers, 90 % of them are using Google and 90 % of them use OpenAI, well, then all of that is kind of worthless.

1:46:51And sure, maybe Mistral will be cheaper in the enterprise and be implemented in French businesses or European businesses. But in terms of like control of the population, that feels like the diffusion of American ideals in Europe, which doesn't seem extremely controversial because America and European ideals are pretty similar. But you can see how this would play out in other countries. Yeah, totally. And the economics are pretty brutal here, right? Like we're in a regime where the amount of compute being used to train a model goes up 5x every single year. We're moving from, you know,$100 million to train a model, you know, last year to rapidly approaching the billions.

1:47:30If you are a company committing to this and you're only sort of just slightly better or worse than, you know, one of these huge tech companies, you can't keep sustaining this over time. You have to check out eventually. Yeah, we've seen that with a lot of the early stage foundation model companies that have raised, trained something, but never been on the frontier. And now they're, you know, falling out of favor, more or less. Yeah, totally. Interesting. Has there been any conversation on the ground in D.C. or have you heard any chatter around the Manus investment that definitely kicked the benchmark made, kicked the hornet's nest a little bit in the hornet's nest of American Dynamis?

1:48:09a little bit deli and asperu of it's either fun but uh but but to to give uh benchmark a little bit of credit you you if you're going to be mad at benchmark for making investment in manis you also in some ways i think have to be mad at jensen for going up and setting up a research and design you know r &d center in shanghai right now explicitly to work on developing products for the Chinese AI ecosystem and in some way direct the CCP directly. Yeah. Yeah. It's a bit hard to evaluate. I think, you know, one, you know, one of the US's advantages is like very deep capital markets, right? And so being able to exert financial control over startups overseas by sort of like acquiring stakes is potentially a way to, you know, have you sort of a better, sort of fairer, sort of more aligned, like global system overall.

1:49:05I think this is hard when it comes of Chinese companies. But does America benefit from having American capital allocators and ByteDance? I don't think we have much influence or control over what ByteDance does. Yeah, exactly. So I think for big companies, especially those that are part of this, you know, Chinese sort of state industrial military complex, this is a pretty poor prospect. This is why, you know, the Treasury Department has outbound investment restrictions in a bunch of different industries, which have been expanded to AI over time relatively slowly though. So yeah, one thing that they've been sort of working on is trying to figure out whether we'll apply these outbound investment restrictions more solidly to commercial AI developers and commercial AI cluster operators.

1:49:51Like so like investments coming from VCs actually restricting investments of those kinds into China. And yeah, this is complicated by the fact that just like due diligence is really hard. Like let's say you're investing into an application developer who's doing something pretty innocuous, like, you know, automated code agents or, you know, search or something along these lines, you don't have any control over, you know, are they going to work with the Chinese military in the future? Are they going to be sort of like an instrument of state-backed surveillance over the local population? They're not going to tell you that and they might not have plans to do that, but they can certainly be compelled to do that in the future.

1:50:25So the due diligence question is super hard. Do you think regulators or the American government needs to be thinking about the pre-training scaling law potentially not holding or reaching some sort of diminishing marginal return. We've seen the data from GPT 4.5. It feels like GPT 5 might not just be 100x bigger than GPT 4.5. It might actually be some sort of mixture of experts, different models, and more of almost like a product challenge than just a scaling and just get the more chips challenge. At the same time, OpenAI is also investing in Stargate, and there's still a drumbeat of ever-larger data centers in the United States.

1:51:11But the overall tone of AI research labs in America seems to have shifted away from just the ever-bigger transformer, and there seems to be somewhat of a resignation to this idea that there might be more challenges on the path to ASI than merely scaling up the architecture that we have right now? Yeah, I think there's a few things here. One is that, you know, obviously these companies are still making huge investments in clusters and energy to get to the next order of magnitude of scale. So there's some level of just, you know, financial buy-in to the idea that, you know, pre-training scaling will continue.

1:51:55but also a recognition that, you know, we've got this other scaling law, this sort of test time compute scaling law, and now like reinforcement learning as a paradigm that seems to really be working for language models where a lot of companies are starting to put more of their compute resources overall. So I think the rough balance now seems to be around sort of 80, 20 pre-training compute and then post-training. And I expect what we might see as we see sort of, you know, companies who are building out these clusters and these energy sources to support them, where are they going to sort of like balance the compute allocation across those the calculus seems to be that um yeah pre-training compute is relatively less promising to the fore and putting more of your resources in a relative sense into rl which is sort of very much a kind of like the early stages of scaling up is the better strategy at the moment yeah i mean the other side of this is like as even even beyond post-training and rl as reasoning tokens and just more test time compute, more inference cost increases.

1:52:54Maybe the real way to get a GDP boost or a competitive advantage out of AI is just to make sure that there is an H100 or equivalent for every member of your society or every citizen, because everyone will need to be inferencing at a very high level, very large model, basically constantly. And so even if you train the greatest model, if you can't have every single one of your citizens constantly inferencing it all day long, you're not going to see the benefits of the AGI. I need my AI companion constantly on. I mean, we do need, we do need, you know, code gen and we need research and we need answers.

1:53:31And if we're, if we're timing out, it's not just studio ghibli. Code gen and deep research are going to be competing with the AI companions for inference. Yeah, probably. Just to, I guess, maybe push back on maybe a hypothesis underlying that. I am very confused about where most of the compute is going to be spent and how that is going to be distributed across people. So I can easily imagine a world like in two years where most people in the world still aren't really using simple tools like ChatGPT. Like, you know, my grandma still like never heard of it. And, but at the same time, you have some companies at the frontier who are deploying millions to billions of agents internally.

1:54:14So you had this really like unequally distributed use of compute overall. So yeah, I kind of buy the idea that there'll be just like massively uneven sort of like usage of these kinds of resources and like really like located in particular countries and within particular companies. Yes. Agree in the sum total of the importance of inference and compute allocated and available for inference, but potentially disagree on the distribution of that. And I think now that you hash that out, I think that makes a lot of sense. You already see that just in the prosumer versus consumer market. I'm probably kicking off like three to five deep research reports using a lot of tokens.

1:54:50I'm getting my$200 worth. And there's a lot of people that are, you know, just land on it every once in a while to toy around with it. That makes a ton of sense. What's next, Jordy? What should we talk about? How have you guys had success explaining the potential for AI, you know, AI's potential progress in the next few years in Washington broadly? Do you feel like lawmakers have fully kind of fully understand the potential? Right. Nobody has a crystal ball. We can't predict the future. But if you sort of extrapolate trends and even just use the products today, do you feel like Washington is pricing it in or is it still you know or are people going to be extremely surprised in the next few years Yeah, my bet is 100 % extremely surprised.

1:55:46I am I think consistently disappointed with how the lack of kind of level of AGI pilled uh people in DC are even with sort of like current capabilities and like where the sort of trend is obviously going I think like there's a lot that just isn't being priced in about kind of how weird the world is um you know in five years time there's also I think a sense of which you know there's a real loss here in that um if you look at technologies like the internet which you know as you probably know, like came from ARPANET, like this, with this like DARPA funded project, as well as, you know, early sort of genomics with the Human Genome Project.

1:56:20These were technologies where the government sort of saw what was coming and took sort of a really active role in shaping the development of the technology through basic R &D. So with ARPANET, for example, the focus was really on creating, you know, a resilient network system that could survive a nuclear bomb. But also they sort of like, they were able to take that sort of secure network infrastructure and apply the notion of kind of like openness and freedom of information to create like a scaled global network that really represented American values and was like very secure. We don't see like an equivalent kind of level of basic R &D investment in the United States around AI.

1:56:53And that would be really cool to see because there's basically a bunch of problems where, you know, right now, industry is focused on where the money is, which is, you know, B2B SaaS apps and chatbots. But there's like a huge space of just like, if you accept that over the next few years, we'll have these incredible new AI capabilities. There's all these massive important societal and scientific problems in areas like, you know, materials discovery, drug discovery, et cetera, where the government could be placing essentially huge bets that could pay off, you know, within a few years. And we're kind of not doing that because we're sort of in DC, at least failing to see sort of where the future is going and therefore what the role for this kind of basic R &D is.

1:57:25So yeah, there's a congressional coalition that's just started called the American Science Acceleration Project, which we're really excited about and try to build hype around. But they, I think I have the right idea around this, but you know, there's a deficit of this kind of thinking in DC at the moment. Last question for me. How is Meta's reputation changed in DC over the last few years? Famously, Zuck goes to Capitol Hill, they don't even understand how his business model works. He says, Senator, we sell ads. He was castigated for being too left, then too right. And a lot of political hot button issues around the Facebook app and the type of content it's servicing, obviously a big vibe shift there.

1:58:07But now it seems like Meta is increasingly a very important tool in the American foreign diplomacy AI tool chest with Lama. Lama is, of course, in their open source model that's, you know, there's defense Lama now. The DoD has partnered with Meta on this. And yet today we learned that their behemoth model is struggling to improve capabilities they're facing setbacks um has the has the tune changed in dc to say hey zuck like sell more ads please like do more stuff we gotta we gotta get you pumping because like you're a national champion now yeah i'd say like by and large you know there's different factions in dc who care about different things obviously sort of meta is going through this big like antitrust case at the moment as well uh like has been going through it um yeah on the ai side i guess it's interesting as well um like meta is very much seen as kind of like the darling of you know open source supremacy for the us i think it's kind of awkward for a lot of meta fans to see the latest batch of models and then really not be that impressive and also potentially a bit of gaming with uh leader boards and sort of what models they're releasing there uh obviously they've executed this strategic pivot to appeal to the current administration with you know getting rid of fact checking and bringing in like the community notes type approach and that seems to have been pretty effective but Yeah, I think they've certainly got a lot of backers here.

1:59:25Yeah, and the open source approach, it's really good to have a really well-capitalized company really pursuing the strategy overall. But yeah, I think they're copying a bit of heat for not actually delivering on the promise to some extent. And also, there's sort of another faction that's pretty worried about open source models with capabilities that could be significantly misused, especially in the cyber domain, being freely available to the whole world. But that's less of a concern while they're behind. Yeah, it was very interesting in the press release around this and the management of the news that they were delaying the rollout.

1:59:58They didn't say, oh, it's too dangerous to release. They could have easily said that. That's always an easy out for the AI labs to say, it's just too good. Trust us. We can't trust you with it. We're doing more safety research. Instead, it seems like there's a little bit of an admission that it's just not at the level we want it to be. this might be out of scope for you but do you think that ai is art is at a much uh is being used in nefarious ways in the context of social engineering attacks at a greater scale than people sort of realize today this was top of mind just due to the the coinbase news this week the leak that they had uh and uh i think people broadly anecdotally reported that they just felt like they had a huge uptick in like sort of inbound calls and social engineering attacks.

2:00:45My question is, AI is already, when you're using tools like Sesame and some of these lower, you know, Sesame is obviously state of the art, but even some of these lower latency video models should already be capable and better at social engineering attacks than like, you know, a PhD level person globally that that even even if english wasn't their first language and and and you know uh so anyways i'm curious if you have cyber security social engineering any sort of like insights there regulation around it yeah no insights into the true extent i will note that it is surprising to me that we don't seem to see more of this my impression is that from sort of GPT 3.5 onwards, we've had LLMs that can produce better text, more convincing text than the kind of median phishing email, which, as you know, is often like pretty poorly written.

2:01:44Maybe that's deliberate. But yeah, like, it's kind of weird to me that we haven't seen mass sort of spear phishing campaigns of the kind that have been possible for like several years now yet, or at least they haven't been widely reported. Maybe it's that, yeah, existing sort of filters and defensive approaches work. Maybe like we shouldn't expect actually that be that many people who are trying to pick this low-hanging fruit and they're not technically sophisticated enough or maybe it's like it's like happening and not being reported on but yeah i think it's i think there's potential that it's happening but it's happening to a demographic that doesn't even know it's happening right like i don't pick up random calls right i just don't i'm not going to answer that you know of maybe the last time i called you it was actually a bot yeah that's true that's true john you do pick up non-random calls but that's the point of them yeah i mean the same thing happened with the crypto stuff like there were a lot of the victims of cryptocurrency scams like weren't profiled in the new the new york times and so we just didn't really hear about them because like the people that with the loudest microphones didn't fall for the scams yeah tricky anyway thank you so much for joining tim this was fantastic i'm glad we survived the attacks from the nation state actors that don't want us to yap we kept the stream up a stream together we really appreciate it we'd love to have you back and this was fantastic yeah thanks for coming on we'll talk to you soon have a great weekend uh next up we're bringing chris best from substack but first let me tell you about ad quick adquick.com out of home advertising made easy and measurable say goodbye to the headaches of out of home advertising only adquick combines technology out of home expertise and data to enable efficiency by seamless ad buying across the globe and we should also talk about wander find your happy place find your happy place book a wander with inspiring views hotel great amenities dreamy beds top tier cleaning and 24 7 concierge service is a vacation home but better anyway let's bring everybody's getting distracted wanders doubling down wanders doubling down it's a knockout drag out fight in the rental home market and we recommend wander anyway let's bring in chris best from sub stack how you doing boom good to see you good that jingle is unstoppable unstoppable we got to get a jingle for sub stack what's the sub stack jingle what do you have a Landing page, hero, hero text.

2:03:54The app for independent voices. There we go. Same melody. Take back your minds. Take back your minds. Take back your minds. Put that in the jingle. How is the process of taking back minds going? It's going pretty well. Okay. Can't complain. Cost 5 million paid subscriptions. Wow. That's a lot. Awesome. Apps blown ahead of, I think, all of the legacy apps combined. Amazing. Chris and I were in the same YC batch. uh winter 18 another one another one so so my biggest thing with john is is actually probably good if john had just invested in in in his batch he would have been too post economic to start a podcast and we wouldn't be here like coinbase instacard zapier uh even uh what was the other one the the crypto exchange that absolutely ripped the nft marketplace open c was in our batch you know, Atlanta, Replit.

2:04:50Yeah, yeah, yeah. Remarkable companies. It was good times. So yeah, I mean, take me through the history and evolution of Substack and where you're going next. Yeah, I mean, we started, we were in winter 2018 together. The basic theory is we're building a new economic engine for culture. You know, we want to harness the power of technology, the power of world's internet scale networks combined with a business model that actually powers independence. The version of that, the simple version of that that we started with is basically make it dead simple to do a paid email newsletter. But as the years have gone by, we've grown that into multi-format.

2:05:29You can write, you can do a podcast, you can do live video. We've got this network and this app. You can get the Substack app from the app store and discover this whole universe of the smartest, best independent media and culture on the internet. And that's what we're building next, basically. I mean, we're building a bunch of video tools. We've got this live, you know, live tool that lets you do something that feels like a FaceTime call. And then AI magically turns it into like a produced podcast and series of clips. So if not everybody is the brilliant YouTube sensations that you guys are, if you're just someone that has something to say, we make it that easy.

2:06:07That's totally super powerful. Was Ben Thompson really an inspiration? He kind of tells that story. I don't know if that's true that Sir Techery and he kind of pioneered this like independent newsletter creator really scaled the business. But I mean, had you been reading him at the time? Had you taken lessons away from Sir Techery when you built Substack? Yeah, there was a few people doing it, right? There was like, you know, because we had this cockamamie scheme. We're like, hey, I bet you people would subscribe to, you know, things they deeply value if you had something great. And there was a handful of people that were doing it.

2:06:38Andrew Sullivan did this with kind of like the Daily Dish back in the day. Ben Thompson had Stratechery. And here was this guy writing an email newsletter from his bedroom in Taiwan, making, by our calculations, millions of dollars a year. We were like, that sounds pretty sweet. Why do more people not do that, basically? And the answer turned out to be, it's way too hard. so what if we made it way easier yeah i mean he's he's he's invested immensely in technology to build up passport and what he's done and now there's a few other companies that run on it um but yeah um where does this all go i mean i guess the big question is like ads product you know is is tied to the written media uh wall street journal has subscriptions and ads um new york times has subscriptions and even after you subscribe you still get ads we are an ad powered show ads feel very, they get a lot of negative attention, a lot of negative stated preferences, but the revealed preference is almost always people don't mind ads and scroll right past them on Instagram and sometimes even get value from them.

2:07:44We've been very pro ads, but what has your take been on ads generally in the Substack world? The thing that's actually different about Substack is kind of creator ownership and a model that rewards quality and value. So the fact on Substack that you, you know, people don't subscribe to Substack, they subscribe to you. You know, you can make, if you make something truly great and most people value it, they come, all those things are what actually set Substack apart. You know, the formats are similar to what you see elsewhere. You can write a long form article, you can make a podcast, you can do a video.

2:08:24Like those are, those are not novel things. The thing that's novel is this business model that powers independence. The fact that you own the, you know, it's your business, you're getting the upside, you want to invest in the quality. And I think there are methods of doing sponsorships and advertising that are, that, that, that work with that, right. The fact that people, you know, people are paying premium for you guys to advertise on this program, not because you're getting like the most clicks ever that gets sold at a market, like at some market rate for the demographic, but because you're making something special and you have tremendous, you know, jingle skills, but the audience is like, is, is magical.

2:09:03Anyway, there are a bunch of people today. We're doing all subscription. We're powering that thing that actually works really well, shockingly well, much better than people thought it was going to work. People are consistently surprised by how well that business model can work for them. But there are also people on sub stack who are doing sponsorships and doing tremendously well. And so I do think that model can and should coexist with high quality media endeavors. Do you believe the medium is the message? We've kind of felt this where we decided to go live for somewhat of like a technical reason, just that the way YouTube's organized, it kind of made things easier.

2:09:39But then once we were live, we realized, well, we're faster. We're not, there's no delay between us recording and uploading. And so we can react to the news. We could, we could open up X right now and, and read a headline that just dropped with you and get your reaction. And it's kind of changed the nature of the show. What's even your, uh, processing of like the McLuhan ideology. I'm a huge believer in live for that exact reason. You know, we're building this video product that you, that is live within Substack. And I think it's almost like a hack that sets the social expectation. The fact that you're going to be able to sit down, make this thing, ship it off.

2:10:15It's not necessarily the case that the majority of the audience is going to watch live. Totally. It's great that those of you are here, but it's going to be – there's going to be a VOD. There's going to be clips. Those might be the places that actually get the distribution. But as a hack to make the thing, the fact that the social expectation between us right now is we're sitting down having a live conversation makes it easier, makes it faster. It does feel different. Like I don't know if you have this, but I feel the feeling of being live is a little bit more energizing than otherwise. Definitely.

2:10:44And then also it just builds, I mean, we, we, we like it because it builds trust with you. Like, yes, there will be clips, but we're not going to edit. We can't edit out what you're saying. So if you want to make your case in some way, we can't be like, oh, we didn't want to put that in. Yeah. I could go back to the tape and say, well, here's, here's the full context. Exactly. Exactly. It's always there. Can you talk about this? So, so one, I want your help kind of framing something because there's, in my mind, there's stages of of substack right like our audience at least the initial core audience is this sort of like terminally online x you know tech enthusiast genius wealthy that that's a core audience but average net worth you could say terminally online uh you know yeah 20 to you know goaded definitely or like in the conversation definitely in the conversation no but but to me there's been these stages of substack where like i don't know if it was like a year ago at this point that i noticed that Substack had not become a household name, but it had become - Outside of tech.

2:11:43Outside of tech. Emily Oster, for example. But it had just become something that kids I went to college with that aren't in tech were subscribed to probably three Substacks. And it was like part of that. It wasn't just all like - It became like clearly part of popular culture online. Yeah, yeah. Was there a moment? Is that like the wrong analysis? and I just wasn't paying attention outside of our bubble? Or did you feel something like that too? I get this question a lot. And I think people, because it kind of grows in pockets, right? Like it'll be, there'll be, you know, this bunch of crypto people that all join at the same time, or a bunch of politics people, or a bunch of finance, like there's sort of like these, you know, next adjacent market pockets as it grows.

2:12:27And so all along, there's, I've sort of had people ask me this, but like, it feels like Substack is suddenly blowing up. Like, What does that feel like? Has your life changed? And internally, it's just been growing really consistently, really steadily. The curve just looks like a steady exponential curve. And so it is kind of always true. This is the most exciting time. I do feel it too. I get less blank looks when I tell people I'm working on Substack than I used to. But yeah, it's been a steady march. how has uh kind of the war with with x maybe the war is not the right word for it but obviously removing links changes yeah platform changes i obviously frustrated a lot of the core creators who had been building an audience in both places but for my view it seems like they maybe you know created a monster like in the sense that like it maybe made you guys react and be like uh and and maybe it was like a short-term win, but will maybe be a mistake long-term and that maybe it's kind of pushed you guys to think even bigger about what Substack is?

2:13:33I mean, the funny thing here is that Ben Thompson credits his early growth with LinkedIn and sharing links on LinkedIn. And then LinkedIn had an algorithm change and he had said that it wouldn't be possible to build Stratechery the way he did in the modern LinkedIn era. So there was kind of like a LinkedIn vibe too. But yeah, I'm interested to hear how you've been processing just all the platform changes really. So, I mean, this has always been my theory is if you want to build one of these businesses, you need a model that supports independence. You need people to be able to subscribe to you.

2:14:01And then you also need like internet scale networks to grow on. And in the old days of Substack, it would be, you know, the very start like trajectory, you'd have this newsletter, this website, but then you'd have to go on Twitter to promote it or go on LinkedIn or go on Instagram or wherever. We always want you to be able to do that. We want you to be able to publish. It's not a walled garden, like put it everywhere, put it on the RSS feeds, put it on YouTube, whatever. But you're totally at the mercy then of these other platforms that don't necessarily, you know, sometimes they go to war with you, like Elon got pissed off at us.

2:14:31But a lot of the time they just, they don't care, right? Zuck can turn around and say, we're not doing politics for a little while because people got mad at us. And if you're somebody that writes about politics, that's really bad for you. And so we always knew that we needed to make our own network, our own place. Fucking Zoom reaction. It's at the OS level. I feel like I've turned this off a thousand times and it sometimes still comes back. If we could just shoot whoever built that feature, that would be perfect. Anyway, we had to build a network. We knew we had to build a network. It doesn't make sense.

2:15:05It's not like Facebook or LinkedIn or X or anybody's TikTok's job to help you take your audience and own it and make something independent. It's never been their main thing. So we knew we had to do that. That's why we built, you know, the Substack app in the first place, which is why Elon got so mad at us. He felt like we were, you know, competing, but ultimately we just knew these independent places need their own network that actually wants them to grow and thrive. It was a big, painful thing for people that were on the platform. It sucked that you're like, links didn't work, super annoying.

2:15:39But it was a very small fraction of traffic. Like it didn't slow down the business or the numbers at all. There was a lot of sturm and drying. It was very like, you know, for people that are like had a big Twitter presence and we're trying to make a sub stack, it was very painful and stressful, but it didn't slow down the growth at all. That's cool. What's yours and the team's sort of decision-making process? It feels like at a bunch of different points with sub stack, you could make a product decision that might drive immediate, basically make the number go up, but maybe wasn't aligned with the kind of platform that you wanted to be and and maybe not even aligned with humanity and on that note there's sort of this like interesting thing where every platform maybe other than sub stack is sort of like converging on being the same app this sort of like short form slopification of social media where it's like you know doing a slot machine uh with information and it feels like sub stack is orienting around long form content we had jason freed on yesterday yeah that's sort of rambling but no no i completely agree i was gonna ask the same question i feel like if i land in a substack link it's going to be written by a human it's not going to be particularly sloppy i might not like the particular topic or something but i feel that it's going to have like this premium uh like vibe to it i guess and i'm wondering if there's is there are we just early and there's a coming wave of slop or that you're going to have to fight?

2:17:08Terrible. I mean, here's my take on this. Please. You know, I think if you put yourself in a position you described where it's like, hey, we've got this business, we're trying to make the number go up, and we either need to make the number go up or we need to make something that's good that we believe in and we kind of have to, like, make that choice. I think as soon as you have to make that choice, that already sucks because both of those choices suck, right? It sucks if you kind of give up your principles and make the number go up, But it also sucks if you stick with your principles and the business, like it hurts the business and the business doesn't thrive.

2:17:40Ultimately, the business of Substack needs to grow and support the thing we're making. And so the thing that we've tried to do is yoke the success of the business, set up the business model and the fundamental way that it works so that in order to make the number go up, we have to do the thing that's actually good. So an example of this is, you know, the way that we make money is it's completely free to publish on Substack to any size audience. You guys should cross publish on Substack, by the way, that'd be sick. And then we only make money when you make money, right? So we're trying to help you grow.

2:18:11We make money when you make money. And so, you know, we have an algorithm the same way that Twitter or TikTok has an algorithm. We even support short form video. You can see clips, you can read long form stuff, you can watch a long form podcast. But it turns out that when you're tuning the algorithm to introduce people to things that they deeply value and might pay for, the emergent effect of that is very different than if you take the same technology and point it at the goal of get people to spend as much time here as possible. Yeah, that makes sense. Because I might pay and then be satisfied, close the app, but that's a win for the algorithm.

2:18:46Yeah, our algorithm says great, right? Whereas Elon's been public about the links thing. It used to be just Substack. Now it's everyone. He's like, yeah, if you're scrolling your feed, you click into a long form post, you go and read that thing and get deep value from it. You just tank the metrics. Yeah. You're not seeing any ads. Like, what are we doing? Yeah, yeah, yeah. It makes sense. How is the health of the overall creator economy? There was a big boom where venture capitalists were investing in the creator economy. That's kind of died down. But how healthy is the creator economy? Well, we were joking yesterday that Erwan in many ways in L.A.

2:19:21is a product of the creator. There has to be double digit percentage of Erewhon's revenue. It's just creator revenue, you know, that flows in in different ways. A lot of influencers in there, for sure. A lot of G-wagons in the parking lot. It's the LA equivalent of like the dot-com bubble. Yep, Erewhon bubble. I've never liked the term creator economy. Even the term content creator kind of gives me hives. Here's the thing that I think is not going away is, you know, the media landscape is shifting. The legacy models have been eaten up by internet things that don't necessarily support the old businesses.

2:19:57And there's this shift in power to independent creators. People like you that can just set up, start a thing, make something that matters, earn the trust of people by going direct, building an audience. I think that trend is extremely robust. I think that thing is going to, it's sort of inevitably going to happen at this point. But I do think it's sort of undecided which version of that future we get. And so we see our, you know, the thing we're working on at Substack is could try to bring about kind of like the best and most valuable version of that future. But I think that, I think that shift is not going away.

2:20:36Yeah. I mean, you, uh, talking about the, the, uh, content creator as a bad term, do you prefer journalist, writer, like more specific scalpel-like terms? Is that what you're getting at? Or what specifically don't you like about the idea of the word content or the word content creator? It's pre-rational, man. It just kind of feels, it feels like a slop version, right? Yeah. You're a journalist, you're a writer, you're a podcaster, an analyst, broadcaster, filmmaker, you know, comic. There's a million things you can be. And listen, I've made my piece with creator. Like It's a good generic term.

2:21:17But I think there was a moment where we kind of like cargo culted the creator economy and everybody got really sort of hyped up about it in a fluffy way and missed the deeper thing that is actually still happening. What actually happened is that there was like a specific data point, which is that VCs realized that content creators were the fastest growing SMB category. And they were just like, okay, we should deploy like a billion dollars. We got to make money off this. We got to make, we got to find a way to make money on this. And the thing that, the thing that people missed is that that had been a sort of, they were sort of like a decade into that trend.

2:21:55And so people funded a lot of businesses that were like banks for creators, but a creator is like, you know, why would a creator not just use rent? You know, we use ramp, right? Like, or why would a creator not just go to bank of America and just get a bank account? Right. And so I think the sort of interesting investable opportunities were the sub stacks, which is like a new economic, you know, not to use your tagline, but like creating a new economic engine for media. They should have invested in Erewhon because that's like the grocery store for years. or they should have just bought stock in Mercedes because they make the small SUV for the creator economy.

2:22:41I want to talk about the value of curation versus instantiation of ideas. I'm not sure if you've been following Ben Thompson's erosion of the evolution of the printing press to the internet making distribution, zero marginal cost, to the instantiation of ideas with GPT and deep research and language models. It's become easier and easier to create a deep research report. The writing is good. It still feels like you can tell when it's AI written. It's at least fine. Yeah, it's at least fine. But I guess the question is, are there any sub-stackers out there that are openly using AI tools to write, but the value that they bring in, the human element is just curation.

2:23:29Because sometimes I feel like I should just share all the deep research reports that I put together because they're cool. And I'm asking interesting questions. And the humanness comes from the question that I'm asking. And very few people would think to ask that question to deep research. And the answer is less important than the question versus just the general trend in language models, if you have any takes there. Yeah. I mean, the way I think of it is, you know, we, even before AI, we already lived in a world of infinite content. You can't get bored. You can't run into stuff to see or watch.

2:24:04And so the limiting factor is like your attention in your life. Like what, what should you pay attention to? Who are you going to trust? What matters? It's sort of, it's, it's the human alignment problem. Like this is what culture is. It's not just getting what you want. It's figuring out what to want in the first place. And that's the thing that's actually valuable. Even before AI, nobody's subscribing to Substack because it's like, oh, I don't have enough emails to read and I want to pay money for that. You're subscribing for a perspective. You're subscribing for some connection, some piece of trust, some piece of curiosity.

2:24:35And so I think all of these technology tools that just give people superpowers kind of supercharge both sides of that. Now there's a thousand times infinity content. There's more than you could, even more, but also the people who who have those relationships can have so much creative leverage. And I think like literally writing for me is the, is one that's not that exciting yet, although it's not impossible, but like, yeah, help me do the research. Help me figure this thing out. Help me, you know, put the pieces of this together. I think there's, you know, there's even people that have, you know, Lenny Wachitski has his Lenny bot that people, subscribers can talk to.

2:25:09I think all that stuff is awesome. Talk about growth hacks. If someone out there is starting a sub stack today from zero, they have no audience. What can they do to turbo charge their business in the short term? The biggest growth hack I often have to give people is just start the thing. A lot of people I talk to are thinking about, oh, I should do this. Should I write 10 things? Should I come up with this plan? Should I do this? Everybody that succeeds that I see just goes, just gets going, just starts writing, starts publishing, try to make something good, try to share with people, and really just have a very strong bias towards action, towards thinking and moving in public, and then kind of like correct based on feedback rather than trying to come up with some genius scheme.

2:26:00All that said, you know, make great stuff, share with people. Have you been able to dig into any of the quantitative metrics around the most successful sub stacks like is there a correlation between posting weekly and and revenue or length of post can a deep dive ever be too long on sub stack or too short or is are there any patterns that you've seen amongst the the top performers you know there is a there is a trend that says look all else being equal being consistent and publishing pretty frequently really does help yeah um it's it's a lot easier you know if you're publishing multiple times a week consistently for a long period your chances of success really do skyrocket other than that though we sort of have the problem you see this in marketing too where sometimes the like the uh the opposite of a good idea is a good idea yep where it's like one thing works but also like the opposite of that thing also works you just got to find something that's sort of good and differentiated so yeah make something that's interesting, authentic, that you actually like, and then make a lot of it.

2:27:07Yeah. I mean, we found that 100%. I mean, 15 hours a week, three hours a day. And not only that, but I mean, I think we posted on X 20 times yesterday and like 10 clips and like some of them don't do very well, but you just set the quality bar where you set it. And then you just try and get those, you know, let the winners ride basically. Well, I'll tell you what, You cross stream to Substack. We'll do it. We're coming. We got AI auto clips. You can make your own, but also people will just find them. I love it. Let's hear it for that. Let's hear it for that. I like the sound of that. No, I'm excited to get over there.

2:27:41To be honest, we've been so, you know, the core challenge for TBPN is that we're a startup, but we're, you know, John and I are the founders, but we happen to be live for three hours a day and then we have to spend like a couple hours prepping the show. And so there's just like, and we've built out an amazing team now, but it's just about adding these other channels. But I see a ton of opportunity on Substack and I just love how thoughtful you guys have been about, you know, building the platform and staying true to your values. It's awesome to see. I have one last question and then we'll let you go.

2:28:18Lessons from Lulu going direct. What'd you learn from working with her? What have you learned from the most recent Lulu ideology and what it means to communicate as a CEO to an audience of investors, employees, customers, etc.? I'm a huge Lulu stan. I've learned a lot from her. maybe one thing that is non-obvious that I got from working with her that I wouldn't have necessarily picked up on just from the output is kind of like the, in many cases, the principles and the morals and the facts come first. The most important part of the go direct comms thing is not just how do I spin this or how do I posture it or how do I say it's like, are you doing the right thing.

2:29:03Are you, you know, are you, are you willing to stand behind the message that you're coming with? Um, I think that thing matters a lot. Yeah, that's great. Well, thank you so much for joining. This was awesome. Come back on again soon. We'll talk to you soon. We'll see you on Substack. We'll see you over there. Bye. Let's tell you about bezel, get bezel.com. Your bezel concierge is available now to source you any watch on the planet. Seriously. Any watch also So potentially creator economy startup, a lot of creators getting watches. Lots of folks in tech getting watches. Their cap table is absolutely stacked.

2:29:38It is. Lots of creators. We talk about numerals, benchmark series A. Yeah. Talk about a stacked cap table. Anyway, our next guest is here. We have Sean from Stored announcing a major size gong moment. Welcome to the stream. Sean, how are you doing? Good to be here. Thank you guys for having me. kick it off with the funding announcement what's going on what's new with you yeah we're excited today to announce that we've raised 200 million dollars across our series e congratulations series e series e been waiting for that long moment that's amazing let's go we're gonna hit more we're gonna hit more sound effects give me the ashton hall overnight success yeah how long have you been doing this I'm a few months away from my 10-year anniversary with stores.

2:30:29Exactly. Overnight success. Overnight success. Classic overnight success. That's great. But introduce the company, break it down for us. What do you do? For sure. So we're building a commerce enablement platform that's entirely designed to level the shipping experience for brands of all sizes with Prime. Over the last two decades, retail has changed where you don't walk in a store and swipe your credit card and walk out with the product. You swipe your credit card online and you walk out with trust. trust that you're going to get the right product. When the brand is said, you're going to have easy returns.

2:30:59And these massive giants like Amazon have realized that is what's driving today's buying behavior. And so they've invested tens of billions of dollars into building out this competitive advantage. Meanwhile, every other brand is kind of in the stone age where if it's take cloud computing, they're still building their own data centers, managing their own racks in their office. And so we give them a scalable platform that combines an end-to-end physical fulfillment network that ships over 30 million packages a year. Last year, we hit about 15 % of US households. We powered over 1 % of Black Friday, Cyber Monday.

2:31:34But then all of the technology, not only that runs that network, but that also speaks to the consumer. So there's a high probability to either yourselves or people listening, they've actually delivered you a package before, powered that tracking link you may never even have known. Wow. So talk about how asset heavy or asset light the businesses? What do you own? Do you own warehouses, trucks, planes, boats? What are you sitting on top of? Who are your key partners to make this happen? Yeah, great question. Our three pillars are really a network, software, and scale. We give economies of scale across a network of assets.

2:32:11Some of those we run ourselves. So we do now operate 13 fulfillment centers, employ about 2000 individuals across those fulfillment centers. There we go. Congratulations. That's huge. It's crazy. Keeps me on planes all the time, constantly going to a different city. Then about 70 % of our business is an entirely asset light network, existing warehouses, existing trucks, particularly existing last mile carriers where we manage a network of about three dozen of them. But then all of that overlayed with our technology. So the customer has one consistent experience, no matter where we deploy their inventory.

2:32:46We talked to Harley at Shopify. How important are small businesses to your business versus going after the scaled e-commerce players, obviously not Amazon, but maybe Walmart or, you know, Macy's and like the really big players? Great question. I'd say we're kind of squarely in the middle where we serve a lot of mid-market brands, about 500 of today's market leaders. So we power all of the deliveries for brands like Athletic Greens, Seed Health, True Classic Tees, Proactive, the skincare, Equip the toothbrushes, Billy the razors. It's a lot of today's kind of multi-hundred million revenue leading brands, the types of companies you'd see walking the aisles of a Target, for example.

2:33:31Yeah. Is that mostly founder-led? I imagine that a lot of the founders of those companies kind of in your boat, been in business maybe a decade, raised a bunch of money and kind of at the same conferences. Is that how you're doing biz dev? Do you have a massive sales force or you can outbound what's working, what's not? Yeah, we are one of the flattest founder-led sales cultures you'll find. We actually sell, I mean, last year we were nine figures of new sales. This year will be multi nine figures of new sales in terms of how we're growing. We have a seven rep sales team. And so we are on the plane all the time meeting with our different brands, meeting with these customers.

2:34:09And I think that's actually one thing that blew away investors in this round. If you look at our last four quarters of sales beats, I mean, we were 3X our Q1 goal this Q1 alone most of the year. we have a fraction of a kind of percent of revenue on sales and marketing and R &D to a lot of peer companies yet a lot of ROI. This is such a funny interview because not only is it live which is obviously a little higher stakes but there's also a very chaotic soundboard potentially throwing you off. Well I mean there's so many every single I don't want to throw you off. It's good news. That's my personal reaction.

2:34:44But if you can make it through a TVPN interview like Bloomberg or CNBC is going to be a walk in the park. It's two minutes, no soundboard. Really easy. You got a question? What was the dynamic? I mean, I imagine the last five, six weeks have been intensely stressful just because of what your underlying customers have been going through. Did the round get kind of done before that? Or were you simultaneously navigating trade? I imagine you were navigating a trade war and a new financing. Just sounds intense. But but what did the timeline look like? Yeah, we kind of come to the principle that everything crazy happens at once at Stord.

2:35:24So it's it's never a time time off. I'd say that we were laying the groundwork for some of this before the trade war really started. And it kind of threw a big wrench into the fundraise in terms of some people realized how good it was for us. And some people got really scared, thankfully. and I'm so fortunate to say, Stored grew massively at other issues, take COVID, take war breaking out, take UPS or Canada Post strikes. All those are a reason for a brand to say, you know what, I'm not going to face this on my own. I don't have the economies of scale to stay flexible. I don't want all the risk on my business.

2:36:00Let me go to a network like Stored. And so thankfully, very similar here, where there's really two tariff issues going on. One is anything that comes into the ports from other countries in large quantities. And that's kind of a standard tariff people are talking about. The other one is the de minimis IMAX section 321, all these e-commerce brands sending small packages into the US, not via a container on the water, just a small package where if it's under$800, they haven't paid taxes or tariffs on the import. That was really made for people like us, individuals traveling internationally, sending products back to families and it got exploited into this massive program where about half of shopify's top 100 e-commerce businesses were shipping from outside the us and so when that changed all of a sudden all these brands had to have this influx of volume into the us and these traditional providers again back to kind of the data center analogy are telling them oh we can get you live and set up a new fulfillment center for you in six to eight months we had a case study with that true classic t-shirt brand multi-hundred million revenue retailer took them live from meeting us and signing to fully outbound shipping in 18 days and that's only possible on a tech-driven network.

2:37:14How much of the business is international versus domestic? About nine percent of our revenue would be either packages leaving the U.S. or actually holding inventory internationally. Okay I have another question. I noticed that the raise was a mix of debt and equity. What are you using debt for? And how do you think about, is this the first time you've really included debt in a big fundraise? And then I have more questions that we can riff on after that. Yeah, I think for stored, we're at a late stage where part of what we announced in this round is profitability, which I think is a lot rarer in a category like ours.

2:37:55We've spent multiple quarters in a row now consistently profitable. And that's compared to venture times when you're in a rapid delivery business and people are wondering, are we using venture dollars to subsidize fast deliveries? Well, I think the proof is in our unit economics and in that profit. But at the same time, we still had a strong balance sheet from the 300 plus million dollars we had raised prior. And so when we looked at this round, we kind of said, let's raise the right amount of equity, but let's also use the scale, the profitability to complement the balance sheet with the right cost of capital and the right flexibility.

2:38:27And part of it comes down to, we actually have been acquisitive in the past as well. We've acquired three businesses over the last few years, all existing fulfillment centers, because we've just seen if we onboard existing infrastructure to our technology, we can multiply the success of those customers, the profits of the acquired business and more. And so plan to be on the lookout for more opportunities like that as we keep growing, which I think funny enough, we started that our first one in 2020, very not in vogue for venture backed businesses to be making acquisitions. Now with this kind of AI wave and more, there's actually some specific funds that are just being built to roll up traditional businesses and apply technology to them.

2:39:09What are you seeing that's exciting around actual fulfillment, automation and robotics? This has been obviously a tough challenge that, again, the major players have invested billions of dollars at this point into. And yet, oftentimes, fulfillment is still a very manual sort of process. It feels like if you're building a 3PL from the ground up, it's maybe easier to think robotics first, especially if you're a tech company young. You're obviously aware of everything that's happening in AI. But at the same time, hard to replace a human. They're pretty versatile. Yeah, we're very excited about both AI and robotics because labor and humans are one of the biggest costs in a business like Stor.

2:39:52And going all the way back to day one starting, even back then, we were seeing peer startups in different cohorts and accelerators more that were building drones for inventory scanning and all this robotics. And it oftentimes shocks someone when you step back and you say, hey, you realize over 60 % of U.S. warehouses aren't even using a digital WMS. They're doing pen and paper based picking. It sounds like it's made up. It doesn't sound possible, but it's true. And so then you kind of look at this gap of where the industry is. And if they can't even get to that kind of threshold one, getting to humanoid robots or AI deployed on how to slot in a warehouse is essentially impossible.

2:40:31And so there's such a fundamental advantage when we've built an entirely vertically integrated tech stack of we're really the only ones like Amazon, when we're making you that promise in the cart, hey, order now, you'll get it by 5 p.m. tomorrow or the next day. We're actually going through not only the front end consumer tech, the order management layer, the transportation management layer. We're looking at we actually have this unit in Las Vegas right now. We can get it to California by tomorrow if we ship it right now. And so we're connecting that vertically integrated system, which just gives us so much more opportunity to deploy robotics, AI, and more.

2:41:06And so a lot of it's on the AI wave right now, how do we use it to optimize demand planning, inventory placement, parcel selection, promise to the consumer, and more? And the next phase is we've been using some robotics, particularly around the conveyance, slotting, and picking in the facilities. But when you can go from static with moving arms to both moving arms and moving legs with a humanoid, it's really interesting. On other big tech trends, what are you most optimistic about across kind of the eVTOL package delivery that we're seeing from Zipline to something like automated trucking, self-driving trucks, to maybe even something just like really, really robust language model-driven AI agents just doing some of that paperwork, for example, but 100 % reliably.

2:42:01Like what technologies are you the most optimistic about? And if you have any timelines, I'd be interested to hear them. Ooh, timelines is always the question mark because even some of the ones you just mentioned, there's been heavily debated and now disproven timelines over the last decade already. So I struggle to make promises. But which one would have the biggest impact on the customer experience? Which one should we really rooting for in terms of just speeding up the time to delivery and reducing the cost? I think any form of autonomous delivery, whether the amazing teams over at Zipline and that kind of localized rapid last mile, some things that failed even at Amazon, like the driverless kind of sidewalk robots, anything that helps connect that last mile autonomously bends the cost and speed curve so dramatically that that's probably where we remain the most hopeful and excited.

2:42:56But with a model like store being the network, essentially what we're doing is aggregating the demand and putting it on one form of tech. But a lot of these businesses you're mentioning are actually key partners of ours where somewhere in our network, we're able to kind of test this net new technology. So same thing with one of the major grocery and kind of food delivery platforms. We have a pilot in a few cities for same day delivery for some of our brands, two hour or less type delivery. But you wouldn't assume if you looked at stores that we're working with that type of company because it's buried in the network.

2:43:30So anything autonomous last mile, anything humanoid in a facility and really anything around demand planning with AI, I think that is the most critical. Because if you talk to brands, demand planning is the thing that really only enterprises will say they do well. And I think most of them are a little self-impatuated when they say they're doing it well. Demand planning is the biggest struggle in anything physical supply chain and speaks to the problem our friends at Peloton had during COVID. It's really hard. I'm looking at the bottom of our ticker. It's Polymarket has the U.S. recession in 2025 dropping like a stone.

2:44:09Let's hear it for the U.S. economy. But my question for you is, are you seeing data on the health of the U.S. consumer? How is demand in the U.S. economy? yeah we have a really interesting kind of front line to the consumer across a lot of industries we've purposefully positioned to very macro resilient industries things like health and beauty you still take the same makeup and skincare in bad times things like nutrition and supplements very similar a lot of subscription orders that we fulfill almost 50 percent of the volume we ship and so thankfully our brands have been pretty well insulated but we actually saw an interesting trend in April, which was uptick for many of them.

2:44:51And we were questioning it saying, is this consumers buying because they think prices are about to go up or what is this signal? And so, so far we haven't seen a kind of bullwhip from that negatively in May. And so year to date, our metrics and kind of markers to the consumer have actually remained pretty strong, even though there's a lot of kind of fear and uncertainty when you watch the news and look at the macro. Let's go. Let's hear it for the American consumer. Endlessly relentless. undefeated uh this has been this has been fascinating thank you for coming on i remember the first time i heard about you and stored was from john at strike i think in 2021 and he was just so incredibly uh bullish on you and uh i can see why so thank you for coming on and congrats to the whole team on the milestone means a ton we're very proud to have strike led and uh i think something like 50 % of the store investor base has joined TBPN so far.

2:45:45I saw Nick Kleiner, Chad Sousa. Oh, yeah. We're thankful for not only their support. Well, come back on when you have interesting data, too. You don't need to come on just for fundraising news. If you're seeing stuff that you think would be interesting to us and the audience, we'd love to have you back on. Thanks so much. We keep the news flowing, so we'll reach back out soon. Thanks, guys. We'll talk to you soon. Cheers, Sean. See you. a little sound what a chat i need to elevate the energy at the end of a long week it's friday but that doesn't mean we can't listen to the ashton hall sound effect which i haven't got enough of we're going into the timeline john timeline okay welcome to the tbp and timeline review the best many people said oh it's almost five on the east coast they're gonna stop they're gonna stop they're gonna stop podcasting podcast for more than 17 hours a week we did it we did multiple four-hour streams this week yeah big big big week uh never podcast weekly always podcast strongly always podcast strongly and daily um scientists use crisper to rewrite dna inside a living baby fixing cps1 a rare lethal liver disorder no transplant no viruses just three tiny lnp crisper doses gene design dose to design uh to dose in less than six months personally personalized gene therapy is finally a reality today, says Didi, the world's first personalized CRISPR therapy given to baby with genetic disease.

2:47:11What a white pill. What a fantastic story. The footage from this is really, really sweet. So there's a whole video about this. You should go and watch it this weekend. But Didi goes on to say, right now, this is applicable to single gene well map mutations in organs that can safely be reached, which affect thousands of babies every year. It costs under five million dollars now so that's obviously expensive uh it'll need to come down further further and it was covered in you know all the mainstream media um but what a what a fantastic uh piece of news uh a baby's life was saved by crisper i remember learning about crisper um back when my co-founders were at caltech and it was kind of this hot technology there'd been actually a few previous uh technologies to edit dna zinc finger nucleases was one of them And all the hot PhD research being done in bio and biophysics back in, this was like 2012, 2013 at Caltech.

2:48:11And their bio division was all about CRISPR. There were a couple people that spun out companies around this technology. Jennifer Doudna won the Nobel Prize and has a fantastic biography written by Walter Isaacson all about that journey. it's a it's a very interesting technology it took so long to get here but it's finally having an impact i mean it already has in many ways but very very exciting to follow good story also out of the scientific community wonder if we could get gene therapy to make us closer even closer even closer to golden retrievers i knew you were going to go there absolutely just permanently change my dna so i can only be friendly yeah turn off the unfriendly gene entirely.

2:48:54Yes. And also make me hotter and dumber because that's key. You can't just be friendly. You also have to be hot and dumb if you're going to go full golden retriever mode. Anyway, I love this. I love this headline from the Wall Street Journal. Forget humanoids at MIT. Worms and turtles are inspiring a new generation of robots. Pull up this image. This is such a great image. Yeah. Look at this turtle robot. He's so proud of his turtle robot. It's great. So CSAIL, which is their artificial intelligence laboratory, envisions robots beyond humanoids, including soft, flexible, and even edible designs.

2:49:29Edible robots. Eat the robot. You're going to be able to eat your robot. Live in the pod. Eat the robot. This is what Joshua was talking about, right? Let's explore every potential magical form factor of robots. Developing soft robots like a sea turtle. Somebody acquired robot.com this week, too. Oh, yeah. I saw that. came out with it. Edible robots for non-invasive surgery. So you eat the robot, it crawls around inside you and cleans you up, fixes you up, sews you up. I mean, yeah. I mean, if you have internal bleeding or something, yeah, eat a robot, it'll sew you up, I guess. Russ's lab is also using new types of AI models inspired by the neural networks of worms to power robot brains.

2:50:09What a fun story. Anyway, and this is something that's actually gone on before. I found a different video from a different group. It says, we have created a robot based on a leatherback sea turtle, which is alive today, leisurely swimming in motion. Next, we plan to modify it to an ancestor of the leatherback sea turtle that remains as a mesozoic fossil to make it swim. And I think this is just like, I don't know, robo, robo, robo, I don't know. Very interesting. It's just swimming around. Just made a robot turtle. Fun. People like turtles, I guess. um we talked about a few of this uh what are the other funny posts uh i mean more more kit more fallout from google the design lead android auto this guy this is somebody's linkedin profile if you're on audio yeah this person says spent 40 of my time arguing with the worst pm i ever worked with 20 managing and coaching amazing designers and 40 on the inefficient overhead of simply working at Google.

2:51:08It's wild to see ideas we talked about in 2015 finally coming out in 2020, but they happened. Yeah. And this is of course a quote post from Daniel says, it's amazing how good Gemini is and how bad every product manager at Google is. I feel like the PMs aren't that bad. It's more just like you ship your org chart and they just can't move fast enough to position the products correctly. The products are good, they're just not positioned correctly or talked about properly. I don't know. Sundar Pichai did an interview with the All In podcast. He sat down with Dave Friedberg. I'm excited to listen to that this weekend and dig into how he's thinking about product design and AI generally.

2:51:51I'm sure there's a lot of good insights in that interview. So go check it out. We have a new landing page for Greptile. this is not going to come up i i just thought this was a fantastic website okay okay and the ai code reviewer catch 3x more bugs merge 4x faster 100 code base content we got to pull up the i'm going to share it it's a fun name these ai companies these sas companies sometimes they get wild all right team it's in the chat let's pull this up great great sas name you got to see the animation though because the static image so the so so the the greptile the the lizard is is catching bugs oh that's good a little on the nose the men we got it do we got it but we respect it oh look at this cute very cute that is a fantastic website i like it that's i'd love to see it that's great uh dan romero giving us a shout out little shout out i'm excited to have dan on the show yeah we're getting locked in tbpn format works because it generates daily content flowing out of the 48-hour timeline zeitgeist that can be cross-posted in video favoring algos on every major social network.

2:53:01Weekly podcast is too slow and stale in comparison. The only live competition is CNBC and they are too slow, boomer downstream compared to Twitter. Interesting. 48-hour timeline. We were, the Daily Show, we're on a 24-hour timeline. No, we sometimes catch stuff a little late. It's true. Right? Yeah. We gotta be more on top of it. We gotta spend more time on Max. But we're news maxing. We're news maxing, folks. Thanks for the shout out. Again, thank you for a wonderful week. I hope everybody has a fantastic weekend. Do you want to talk about the public.com vibe investing? Yeah, we should actually cover this.

2:53:33This is very cool. So public.com, obviously sponsor of the stream, has launched the ability to create synthetic portfolios around basically any investment idea you have. And then back test them against the broader market, which is where it gets really interesting. So we have a couple of these pulled up that we can share with you. The first one is Founders Fund funded companies that have graduated and gone public. How did they do in the public markets? And you can see companies that were backed by Founders Fund have done 659 % return over this test versus 150 % in the S &P. Sequoia's done similarly, a little up and down here for Sequoia, 123 % versus 147 % in the S &P 500.

2:54:19Yeah, a little bit rockier, but you have to imagine. Excel's done very well. 589 % versus 150 in the S &P. A16Z has also done well with 406 % total return versus 150 % in the S &P. And this is where it gets really interesting, John, the F1 index. These are companies that sponsor F1 teams. Yes. And they have dramatically outperformed the S &P. 317 % total returns versus 147 % for the S &P over the same time period. The best portfolio that they tested is the compensation size lords. These are highly paid CEOs, often controversial. 2 ,000 % return based on investing in CEOs that get paid a lot versus 176 % in the S &P over the backtest period.

2:55:15That's actually insane. All right. It turns out when CEOs get paid well, shareholders benefit. And also who else outperforms? Bald CEOs. Bald CEOs outperform. Let's give it up for the bald CEOs. 433 % versus 150 in the S &P. Fascinating. Bezos probably doing a lot of heavy lifting now. A lot of heavy lifting. Yeah. But you got the Brian Armstrongs. Yeah. Right? Strong. Yeah. Strong. He's got a good. I mean, this is insane. I'm interested to just continue tracking this over time because it really tells an interesting story. Yeah. Not investment advice, obviously, but... Never investment advice.

2:55:56Always entertainment advice. Steve Ballmer, another bald CEO. There's a ton of them. Yep. A lot of good bald CEOs out there. So shout out to your bald friends. Anyway, thank you for watching. Leave us five stars on Apple Podcasts and Spotify. And we will see you on Monday. Thanks for watching. I cannot wait. We'll see you soon. Bye. Cheers. have a good weekend