Hims & Hers Stock Drops Over 30%, All-In Launches Ultra Premium Tequila | Crémieux, Clément Delangue, Emmett Shear, Brendan Foody, Sam Schwager, Zach Lloyd, Dara Ladjevardian, Amjad Masad, Angus Muffatti

29:43It's going to, it's like the powerful stuff. Yeah. I mean, one of the secrets to those like spirits that popular on college campuses is that they take the alcohol content down to something like like 30 % or 35 % flavor. I don't know where this picture is from because on the website it says 40%. I don't know either. I don't know where I got that picture. We will have to get a bottle ourselves. We will. And we'll taste it. We'll immediately tell you. Yep. 42%. We wanted to have Tyler taste it, but, uh, fortunately he's not of age. Yeah. So we'll save a little bit for you, Tyler, for your birthday.

30:19Yeah. Can't wait. Um, well, Well, if the besties are planning to take the besties all in tequila to the moon, they're going to need to get on Vanta. They're going to need to automate compliance, manage risk, prove trust continuously. Vanta's trust management platform takes the work out of your security and compliance process and replaces it with continuous automation. Whether you're pursuing your first framework or managing a complex program, if you wind up turning this into the big IoT SaaS product that we've imagined it being, you're going to be managing customer data. You're going to have their health records potentially.

30:55You're going to be pulling in all sorts of stuff to optimize their liquor consumption. And you're going to need to be secure. So you're going to need to get on Panta for sure. How big do you think this is? Where do you think this actually goes? I think that's the more serious question. Is 9 million? I think that they're going to sell that. I think they're going to sell all 7 ,000. No problem. The haters will be in disbelief. Yeah. There are so many fans. We're going to buy a bottle. but we know folks who are friends and friends of friends who are going to buy bottles. There's going to be people that buy bottles because it's fun and they want to support the show that they like.

31:30There's going to be folks that gift it. There's going to be people that buy it because they want to just support the thing that hasn't been charging them for five years. Some people just want to support the dictator, Shema. But honestly, it's like there is another world where if all in was$20 a month, you would have paid$1 ,000 over the last four years. And they didn't charge you$20 a month. They didn't even run ads. And so what's your way to get them to say, hey, yeah, you've delivered me$1 ,000 worth of value and I have it because I'm successful. And, you know, this is a great way. And you get something that, you know, who knows what the margin is on this.

32:10Is it worth$1 ,200? Maybe. I don't know. It's hard to say. But it's like it's a fun thing to buy. Imagine you're going to a friend's house for dinner. What's funnier than bringing a bottle of the Bessie's all-in tequila and be like, hey, I just picked this up on the way over. It looked nice. It's like a good high-end tech gag gift almost. It's like a luxury gag gift. But it's also probably really good because it's expensive. So then you taste it and you're like, oh, this is actually really nice. Which bottle design do you like more? Because Casa Czar has this ceramic bottle that they hand paint.

32:45you said you liked that it had the stack of chips all the way to the top stacked i don't like that john doesn't like it break it down i think the ceramic look is nice they look more like regular poker chips that is true i think it's cool i think it's almost too close to poker chips like i would be confused i'd be like is that just a pack of stack of poker chips it's like hiding it's it's hiding can we zoom in on the stack of poker chips there bring that over oh no know, skill issue. First zoom, first zoom, bud. First time zooming. Yeah. You know, anyways, the Casas are a beautiful bottle. The Casas are beautiful bottle.

33:23I like that it's fully stacked. Okay. Ultimately, I think it just doesn't read as a bottle of liquor. Whereas the besties all in tequila does to me read as it. And then also the, the glass, the glow, the glow is, It is in a different tier. Ceramic is premium. I would only give the ultra premium label now to the besties all in tequila. I would call, they've reset what it means to be ultra premium. Look, they're just showing off zooms now. There we go. They're zooming all over the place. Yeah. I mean, not to talk trash about Casa Czar. It looks fantastic. I'm just saying that the besties all in tequila, they innovated by including the illumination, by including the glass, by including the top.

34:09We don't know. They could be the IoT functionality that people have been asking for. We don't know. So I mean, I would give the besties all in tequila the edge here. Now, do you know the story of the iPhone? Are you familiar with what an iPhone is? Are you familiar with the name? I know where you're going with this. I know where you're going with this. So when Steve Jobs introduced the iPhone, he famously said on stage, And yes, we're calling it the iPhone. And this was controversial because Cisco owned the rights, the trademark to the word, the iPhone. And so everyone was like, how is this going to work?

34:48You just launched a product that is clearly derivative of this other company's name. They have a trademark. They're just going to block you. But at some point a deal was made. I don't know if it even became a lawsuit it was probably just like they sat down and said hey let's you know let's let's allow this to happen we want to run with this we could take it to the course but that's a headache for both of us let's just net this out and we'll buy some webex license there's a article on appleinsider.com from 2006 cisco introduces the iphone family of devices wow And it is not nearly as good looking as the iPhone would ultimately be.

35:30And it's funny to see the iPhone, effectively the iPhone word mark placed on a non-Apple device. Well, if you're trying to build the next iPhone or software for the iPhone, get on Linear. Linear is a purpose-built tool for planning and building products. Meet the system for modern software development, streamline issues, projects, and product roadmaps. Now, we discovered an interesting link between Cisco, the iPhone, Apple, OpenAI, Johnny Ive. And I think we cracked the code. We got to break it down for everyone. Get the tinfoil hat. Get the tinfoil hat. I will send the image to the team so they can pull it up.

36:05So, okay. So, this is how it goes. So, Apple, as we mentioned, they potentially got the name the iPhone from Cisco. Now, who designed the iPhone at Apple? johnny johnny ive that's right now where does johnny ive work now open ai open ai they acquired love from they got brought him over to open ai who's the chief product officer at open ai kevin wheel kevin wheel and who's uh what what boards does he sit on cisco cisco who i who licensed the iPhone name from Apple. Wow. It's a snake eating his tail. An Ouroboros of Silicon Valley intrigue. We crack the code here. It's all linked. Symbolism will be their downfall.

36:57The most completely nonsensical schizo diagram of all time. What are you alleging? You're alleging that things happen. People interact with each other? Yeah. If only it were so simple, Jordy. If only it were so simple. I wish I had your mind, the mind of a golden retriever, to not see the obvious pattern here. The big conspiracy. Yeah, the big conspiracy that we unveiled here on TBPN, that they're all linked. It's all part of the plan. You're pulling the strings. Big Cisco, Apple, OpenAI, Johnny Ive, Kevin Wheel, it's all linked. It's all linked. We blew it wide open. We should throw all those guys in our CRM.

37:39Adio, customer relationship magic. Adio is the AI native CRM that builds scales and grows your company to the next level. The official CRM of Coca-Cola. So final word on the besties all in tequila. I think it's a no-brainer that they sell 9 million in this first run. The question is, do they do that every month? Do they do that every year? Do they move on to a different drop? Or does this become something that is a staple? I think one of the challenges of what it will come down, it'll come down to product quality. Sure. And then the number of people in the all-in audience that care enough about tequila to spend$1 ,200 on a bottle of tequila.

38:19Not in a sort of like funny gag setting, but as like, I'm going to just drink this regularly. And I think it could be a lot. Like I could see them selling 20 ,000 bottles of this. You can imagine them drink. One of them is drinking it on every show, right? It's sort of this constant always on advertisement for it. I could imagine them. They're saying the besties all in tequila. I could imagine them going into other categories. Right. And there's a world where they're selling best tens of thousands. Besties all in malt liquor. 40 ounces. 40s. I would love to see. I would love to see J. Cal do do Edward 40 hands.

38:59Yeah. Edward 40. That would be amazing.

39:10We got to move on. I have a different bowl case for you. I have a different bowl case for you. It's a little more serious. So I have been in consumer package world for a decade. I've built two brands in the space. Like the product quality is super important. You do need an edge and you do need a value prop that can actually resonate with consumers and you need ongoing marketing. That's 100 % true. but in fact the biggest monopoly brand is an important monopoly for consumer goods but there's almost a bigger monopoly with just distribution like if you actually walk around the grocery store you will see i talked to a founder who um came up with he his whole model was he literally would walk down the grocery store and look at not just the products but who own the products.

40:00And he would see that, wow, this entire aisle is owned by Unilever. They just dominate the entire aisle. There might be a ton of different brands, but everything is Unilever or everything is General Mills or Kraft or something like that. And so he wound up building a company that does... He was looking around, he noticed that the popcorn section of the popcorn isle was extremely uncompetitive because Orville Redenbacher is the only major brand and they're not owned by Unilever. So they don't have much about that. You don't hear much. So we started a company that was like a popcorn spray that you could spray a flavoring that you'd spray on the popcorn is purely additive, wasn't competing with anyone.

40:41And he grew that to like a couple million bucks. It was great. And I think that a lot of these products, they do just wind up winning on distribution. And what's interesting about this is that I am almost certain that in the all-in audience, yes, there are consumers, but there are tons and tons of people who work in distribution, work in store ownership. Would it be possible for any of the besties to get the CEO of BevMo on the phone? Absolutely, in two seconds, they could do that. Could they get this into Walmart? Absolutely. Could they get this into, I don't know, CVS? I don't even know what the biggest channels are for high ultra premium liquor are.

41:27But if you just think even like the independent store owners, they probably listen or they aren't familiar, right? And so just being able to take that call and accelerate your sales force and have them be like, yeah, sure. Send me one. I'll put it on the shelf. When we sell it, I'll just order another one from you since it's a thousand bucks. And so pulling that distribution forward into the future, being more aggressive about that, about getting into distribution, holding onto it, I think that's something that could be interesting. And then the fact that the bottle design, there's a couple companies that do, that have been particularly good with bottle design.

42:03I've talked to one consumer packaged goods founder who basically said that their whole model was like go into a category and figure out how to make something like shiny and gold and like pop off the shelf. Because you look at the wall of all the products, they all look generic. If you can be the one that stands out, you're going to just have more people pull it off to sample it. And when I think about if I'm walking into a liquor store and I'm seeing a whole bunch of tequila and one of them is literally glowing and looks like a, and it has a light up display. This is something that a lot of consumer packaged goods founders negotiate for.

42:43They will say, hey, like Red Bull would come in and say, we will give you a fridge, a Red Bull fridge. You know those fridges? Or like, we will come in and give you a light up display. We fight for this all the time with Lucy. Like we will say, oh, we'll send you a free in-store display that you can put on your shelf. And they're putting it on the counter so that we don't have to fight for the shelf space that's there. And so imagine if they're like, every bottle is a light up display. It will jump off the back bar. And so it won't just sit there and collect dust. It will automatically draw the eye to it.

43:14I think that the light up thing could be enough of like a distribution advantage that it winds up being a popular product. Totally. I don't know. Totally. I agree. I think it'll come down to product quality. What percentage of the audience actually wants to spend$1 ,200 on tequila on a recurring basis? Do they go into other categories? I think they could sell, you know, Chamath is also launching a wine club. Yeah, that was interesting. That's separate. There was so much backlash to the tequila thing. It wasn't even the first alcohol brand he launched that month. Yeah. Yeah. At the end of the day, anybody that's getting mad about them doing something like this, it's like they have always been very candid that they enjoy drinking together.

43:55Why? You know, and it's not like they can't do world positive things in tech in addition to having fun and having their own alcohol brand. I think a lot of group chats out there would like to have their own alcohol brand. So anyways, I'm excited to see it play out. I'm excited to try it. Yeah, yeah, yeah. This is definitely not like a pivot of the brand in the same way that the talking politics is not a pivot of the brand. You go back to the very first episode of the All In podcast. They were talking politics. They were talking about COVID and stuff like it was it was like they've been very consistent in all this stuff.

44:32So anyway, if they're trying to pump their tequila, they got to buy a billboard on AdQuick. adquick.com out of home advertising straight to the 101 i want to see the one i want to see every single billboard on the 101 the besties all in tequila 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 efficient seamless ad buying across the get if you're trying to get the attention of a ceo maybe you want to sell some enterprise software to them why not gift them a bottle of the besties yeah and that'll that'll get you a demo real quick i think vc should be sending this to founders founders should be sending it to vcs it's it's a it's going to be a hot commodity if you can get your paws on one of these yep um anyway hims and hers stock drops more than 30 after novo nordisk breakup this is from the wall street journal of course if you're interested in stocks that are dropping because you're going short or are popping because you're going long You got to get in public investing for those who take it seriously.

45:35They got multi-asset investing, industry-leading yields, and they're trusted by millions. Anyway, the Waguvi maker, that's Novo Nordisk, has cut ties with the telehealth company, HIMSS and HERS, citing allegations of deceptive marketing practices and drug compounding. I feel like they were, the drug compounding, how is this an allegation? I thought that's like what HIMSS was saying they were doing. They were like, I felt like there was a lot of press about them being like proud of drug compounding because there was a shortage. But I guess it's I guess it's alleged at this point. So we'll dig into this Novo Nordisk abruptly ended its partnership with hims and hers after the Danish drug maker accused the telehealth company of illegally selling cheaper copycats Okay, so this is after the FDA is ruling around about whether or not there was in fact a shortage.

46:21So the Novo said on Monday that it had concerns about the safety of knockoff versions of a goofy and that hims and herbs deceptive marketing of such knockoffs put patient safety at risk. In turn, Hims and Hers accused Novo of pressuring it to steer patients to Waguvi regardless of whether or not it was the best option for patients. News that the deal was scrapped weighed heavily on Hims and Hers shares, sending them down 32 % in afternoon trading in what would be the largest decline for the stock on record. Novo Nordisk's shares were down nearly 6 % in New York. The messy breakup comes less than two months after the company's unveiled what they called a long-term collaboration to directly provide Wagooby to HIMSS and HERS patients.

47:06Okay, so this was somewhat predictable, right? We talked to TJ Parker about this. HIMSS has been on a generational run in the markets because they're one of the best, easiest ways that people can just go buy Wagooby, right? Now, if you're Nova Nordisk and you're like, cool, we're selling through this channel, but then the retailer, in this case HIMSS, is telling customers as they come in, hey, we know you wanted Wagovi, but how about this sort of cheap knockoff, this compounded version? Now, Wagovi is probably legitimately concerned about safety issues. There can be issues around compounding and drug safety.

47:46But then the bigger issue is like, hey, you're using our brand and our products to acquire customers, and then you're getting those people in the door and you're saying, hey, how about this sort of cheaper off the shelf alternative? And so I have to imagine that this is Nova Nordisk recognizing the value of their brand in this massive, massive fast growing category and saying, we're not going to let you leverage our brand to compete on these sort of high margin, you know, knockoff products. Yeah, I mean, like the Instagram ads that I saw we're definitely pushing the brand name for real. Yeah. And there was a while when they wouldn't use the term Viagra and they would use ED meds and sildenafil and tadalafil and all of those different, like they would use the actual underlying compound name and they pushed it so hard that people actually started knowing the name of the underlying compound.

48:43Yeah. But yeah, the brand name. How do you think he will respond? Do you think they'll introduce a 10 in one shampoo? that includes compounded, GLP-1s, Viagra, hair loss, medication. What else could Adderall? They could put some caffeine. I mean, this is Cognizant, Rhodiola, Panex, Genseng, L-theanine, caffeine. Get some GLP-1 in here. Creatine soap with? Creatine soap would be good. With all of the above. Eli Lilly is the one that lost the patent, right? Because they forgot to pay the. No, I think it was Novo Nordisk. Oh, it was Novo? Wow. They're all over the place. It's rough. It is interesting to watch.

49:27Yeah. Novo Nordisk, how much do you think their stock is up over the past five years? Ooh, 500%. Try 113%. 113%, but it was a big company, yeah. Yeah. But it's been on a tear. Big company, but still, and what do you think the chart looks like over the last year? Oh, it's down. During the GLP-1 boom. It's down because it ripped up so high. Down 50%. And then it went down 50 % and they rotated out their CEO. That's right. We covered that. HIMS and HERS is at 9 billion, 9.66 billion. That is down from a high of maybe 15 billion. And it's one of those interesting like SPAC targets that I think they got out through a SPAC.

50:09And they never really went down that far. They were down, you know. Yeah. Yeah, like, you know, 30%. I mean, they SPAC to$10 a share, went down to$4 a share, kind of grinded back up. Now they're at$40 a share. So if you bought the SPAC and you did hold for five years, you're, you know, you're sitting on a 3X, 4X. Not bad. So, you know, kind of a narrative violation around the SPAC, like all SPACs are bad narrative. Yeah. Yeah, I mean, and the CEO, Andrew Dudham, we should have him on the show, said that, you know, I think his position, you know, we have no idea what the details of their contract actually look like, but fairly reasonable position to say we're not going to just sell one product just, you know, because it's true.

51:05Not every patient is going to respond well to Wagovi. Yep. And they should have options. So before striking deals with brand name drug makers, telehealth firms were selling lower cost knockoff versions of these drugs made by compounding pharmacies, taking market share away from brand name makers. The compounding pharmacies had been allowed to mass produce knockoff versions conditionally because the original drugs were in short supply. Now, the whole compounding pharmacy dynamic is very different because I feel like the the laws were created such that if a big pharmaceutical company couldn't supply you the drug, like your local pharmacist could compound it.

51:45But then the startups kind of like rolled up all the compounding pharmacies and were doing it like a much bigger scale. And it wasn't exactly what was intended originally, but I'm not sure about that. Yeah. So the big question is when should they have stopped? Well, the big question, will there be more regulatory action around compounded GLP ones and what will that do to their business so roller coaster of a year for him's yeah partnership was in announced in April 2025 popped terminated just a few months later with a 35 % drop well if you're on GLP ones and you're losing weight and your wrists are getting thinner you need a watch that can adapt to your new wrist size you need to go and get bezel calm your The bezel concierge is available to source you any watch on the planet, seriously any watch.

52:36And you can find a watch with metal links that you could take out a link if you've lost a lot of weight or you could add one back if you're in a bulking cycle. Yeah. So keep those links handy when you're ready to start the bulk. Exactly. This is key. This is very, very key. Anyway, comment from the CEO, Andrew Dudham. He said, hims and hers will continue to offer access to a range of treatments, including so they continue to battle it out. It'll be interesting to follow this, and we will, of course, be talking to Cremieux about this later today. There is a investing visual post here that I pulled up about how HIMSS makes money.

53:15They're making$576 million online, $10 million from wholesale. That brings their total revenue to$586 million, which is up 111 % year over year. Gross profit there,$428 million. Operating expenses are$337 million, leaving them 92 million in net income, or 92 million in operating income, and 50 million in net income. So 50 million in net income for a$10 billion stock, pretty high price-earning ratio, if that's an indication of their earnings. But it's been a big growth stock on a major, major trend, GLP-1. And so the retail army has been coming out in force. King, KSM's capital, said it felt too good to be true for HIMSS to go this long without some FUD.

54:09This is more like it. Just bought 2 ,300 shares and top secret stocks. What does the retail army call themselves? I don't know. HIMSS or something. Say, I'm really HIMSS. Yeah. HIMSS, look at the data. You can see their operating margin has improved a ton since they've gone public. They started out at negative 40 % operating margin and climbed up every single quarter. They're now at a 6.5 % operating margin. The number of subscribers in the millions has gone from, they were at.6, so 600 ,000 subscribers. Now they're at 2.4 million. and their total revenue has gone from$272 million in revenue. This is the last 12 months.

54:58Now they're almost at$2 billion. And so they've almost 10x revenue from that SPAC price. And that's what's driving the stock overall to be up 3 to 4x off the original SPAC price. So very good news. If you're Andrew Dudum and you need a break from grinding it out at HIMSS, book a wander. Find your happy place. book a wander with inspiring summer andrew hotel great amenities dreamy beds top tier cleaning 24 7 concierge service it's a vacation imagine you're in one of the gnarliest pr crises of your life and two guys in suits on a podcast never matter just like yeah take a vacation but it gets a bad idea summer's long hymns is up two percent today john you're good so it's time to you know treat yourself yeah take your foot off the gas head to wander.com and speaking of foots on gas.

55:47Ford, which makes cars that you can put your foot on the gas of, is now under pressure to put their foot on the gas of acquiring rare earth elements. They are having difficulties obtaining vital magnets with rare earth elements despite the deal the United States struck. I just want to highlight, I think the silhouette of this car is fantastic. The Mach-E GT, you're a fan. I've only owned one Ford. ford raptor yeah it was a very fun uh car not very practical you need to park in cities and things like that but yeah if you want practicality ford gt uh i took a drive yeah practicality all about the ford gt yeah um but i took a drive my buddy uh matt has a ford mustang uh what is it called the mustang e mock e rally it's like this this rally version of the maki okay so it's still the crossover still like an suv more or less sort of it's like a hatchback yeah it's not a car like i would not call this a car it it is a very very cool car it's fun um it to him he said it's like the most fun car he's ever owned let's go uh and he's had pretty something from from pretty much every manufacturer at this point so he should hold on to it because they might be struggling to make them in the future going up yeah uh ford motor still facing difficulties obtaining vital magnets made with rare earth elements despite a deal the u.s struck with china to ease export controls a company executive said monday it's hand to mouth the normal supply chain scrambling that you have to do said lisa drake a vice president overseeing ford's industrial planning for batteries and electric vehicles ford in may stopped production at a vehicle factory in the chicago area because of a magnet shortage the situation has improved but the company still needs to move things around to avoid manufacturing shutdowns given the scarcity of the materials Drake told reporters during a briefing at an EV battery plant the company is building in Michigan.

57:49China in April began requiring companies to apply for permission to export magnets with rare earth metals, including dysprosium and terbium. The country controls roughly 90 % of the world's supply of these elements, which help magnets to operate at high temperatures. Much of the world's modern technology, from smartphones to jet fighters, rely on these magnets. So everyone sees that 90 % of the world's supply and thinks it's 90 % of what's in the ground, and that's not the case. What is really going on here is that the industrial capacity to remove the rare earth elements from the ground is 90 % in China.

58:30So they have 90 % of the mining companies, 90 % of the mining equipment. They don't necessarily have 90 % of the actual rare earth metals in the ground, and the rare earth metals are not that rare. But we hear this every once in a while where we say, oh, like there was a ton of rare earth deposits found in Ukraine or found in America somewhere. And that's not enough. Like just having them in the ground, you still have to go and dig the mine, get the approvals, make sure. And one of the big reasons this was offshored was because of environmental considerations, which I find so frustrating because it's one world.

59:09Like it's an earth. Like if you want to protect the earth, moving something from one continent to another does not save the earth. Like, I mean, I guess there's a little bit of like, yes, there's runoff in one river. And so there's local pollution that you're avoiding. But in general. You've been to China, John. It is dark. Yeah. The sky is legitimately dark. But I think that like a lot of the pollution that's going on over there is affecting us too. For sure. And so especially if you think about like the broader global warming picture, the broader like poisoning of the air, like that stuff's not going to come over here at some point.

59:43Like I find that hard to believe. It would have been much better to just focus on ramping up production in America and then actually doing the technological innovation to develop cleaner processes and not just offshoring it because it's like dirty and then just doing the dirty stuff over there. I don't know. maybe there's an argument for it, but it feels like the whole offshoring project was wrapped in this veneer of like, hear no evil, see no evil, speak no evil, that type of vibe. Like if we can't see the pollution here, it's fine, it's not happening. And it's like, it is still happening. It's just you don't have control over it.

1:00:22Yeah, massive supply chain risk. And I don't think it actually had an impact on the, Like the stated goal of the environmentalists that pushed rare earth manufacturing overseas did not, was not achieved. And so that feels like a failure. And then on top of that, we lost control of this important supply chain. So Ford isn't the only carmaker struggling to obtain these magnets. Several other carmakers say that the pace of export license approvals for rare earth magnets hasn't changed significantly. It isn't quite day to day, but it's week to week, said an executive at one of the car makers. Car companies have warned that they could be forced to halt factory work if they are unable to attain enough rare earth magnets in the auto industry.

1:01:07Rare earths are essential to electric vehicles because they allow a low cost way to ensure motors operate at high speed. Okay, so this should not affect the production of Raptor Rs. Thank goodness. I was extremely worried. So we might have to get most of those Mach-E owners into Raptor Rs. Okay. But that's something we have to, if we have to do it, we have to do it. We have to do it. In the auto industry, rare earths are essential to electric vehicles. President Trump struck a deal earlier this month for China to resume granting licenses to export rare earth magnets. The agreement, which has a six-month limit, is aimed at allowing China to retain its chokehold on the critical minerals that give it leverage in future trade negotiations.

1:01:54So, very, very rough. Anyway, how'd you sleep last night? Let's pull it up. Praying. I got an 88, John. Oh no, 82. There we go. You beat me. Oh, I did. Jordy's back on top. I'm back. I'm back. After a while. I'm back. What happened to me? I guess I stayed up too late or something. I was in my sweet spot. Oh, I guess I didn't fall asleep until 11, 12. Ralph. There you go. There you go. Got to go to bed earlier. I'm back on the horse. Yeah. Back on the horse. It should be a competitive week. I'm excited to see how it, we got to start putting something on the line. Yeah. Like the loser on Friday gets no caffeine and nicotine the whole stream.

1:02:37And they'll be sleep deprived too. That's double duty. That's extra hard. I don't know if that's possible. That would be torture. Anyway, our first guest is coming in the studio in 10 minutes. Let's take you through a beautiful obituary to the founder of FedEx, Fred Smith. So Ryan Peterson posted a photo of the original FedEx plane. He said, RIP to logistics legend, an outspoken defender of free trade, FedEx founder Fred Smith, who passed away today. Below is the first ever FedEx plane, which you can see at the Air and Space Museum Annex next to Dulles Airport in Virginia. Look at that. He must have been so stoked when he bought this.

1:03:18Like his business was finally growing that he could own his own plane. Imagine how many likes he would have gotten on social media posting that pic. Yeah. Everyone's like, this is your flexing. You just had to appreciate it. You're just trying to go viral. This is clearly. This is clearly for logistics. This is clearly her logistics. Total legend. Oh, like burning VC dollars on a plane. I think he was using Vegas dollars. I think he was doing cash flow, yeah. Anyway, Max Meyer has a story that we should read through. Fred Smith and American Life. He says, Fred Smith died on Saturday. He was the founder of Federal Express, which became FedEx, one of the largest logistics companies on earth.

1:03:59Smith was an entrepreneur in the truest sense of the word and an American patriot. He bent reality and created something entirely new, and he ran toward an extremely difficult problem. The fruits of Fred's work are as follows. If you give a box or document to FedEx by 6 p.m., you can have it delivered almost anywhere in the United States or Canada by 8 a.m. the next morning. Imagine how key that would have been to business before the internet. Incredible. Yeah. You're trying to do deals? Trying to close a key customer? If you give FedEx a package on a Monday, it can be in Paris by Wednesday morning, or in Shanghai by Thursday for a few hundred dollars.

1:04:37Most of us take overnight delivery for granted, but take a step back, and the fact that we can essentially airdrop physical packages anywhere in the world is mind-boggling. Smith's death is a reminder not to take the miracles of modern life for granted. They were built by people like Fred Smith. These are a few words about the man himself and the dazzling system that is FedEx. Frederick Wallace Smith was born in 1944 in Mississippi. Fred's grandfather, James Buchanan Smith, captained Mississippi River Steamboats. Fred's father, James Frederick Smith, was the founder of the Smith Motor Coach Company, which would go on to become Dixie Greyhound Lines after an acquisition in 1931.

1:05:16From steamboats to buses to jumbo jets in three generations, the family stayed on the cutting edge of motorized transportation. In 1925, the elder Smith, seeking to create the transportation line out of Memphis, converted a truck into a small bus and drove it himself. Within a few years, James Smith had dozens and dozens of coaches. By the time he died, suddenly in 1948, he commanded over 200. Wow, so he died four years after the son was born. Wow, so he basically grew up without a, or maybe that's the grandfather. According to James Smith's obituary, no, no, no, he's the dad, yeah, yeah. So according to James Smith's obituary, those 200 coaches each came to a halt during the regularly scheduled routes for one minute when the funeral began.

1:06:03The young Fred was just four when his father died. Wow. In 1962, the young Smith enrolled at Yale University where he was fraternity brothers with future president George W. Bush. Bush asked Smith to serve as the secretary of defense twice and Smith declined twice. Wow. Interesting because defense is, who was the founder that we had on the show? I'm completely blanking on his name, the name of the company, and who invested in the company. but he had this defense-focused supply chain management startup. Yes. It was one of our probably first - You're talking about Rune, logistics. Rune, Rune. Yes.

1:06:38But something he said about - Not the AI poster. The legendary poster. Not the legendary poster. The logistics company. The obscene of open AI. Yes. So the young smith was a member of the infamous cloak and dagger secret society. He is not beating the deep state. De-state allegations. It was at Yale that he wrote his first paper about his concept for using airplanes to deliver packages with a hub and spoke system. Wow, he was like, I got an idea. I got a business plan. I just need somebody to build it for me. I.e. that instead of point-to-point package transport, everything would be brought to a single hub by air and then redistributed.

1:07:16We talked to somebody about this on the show, how now we think of it as inefficient to bring everything to one place and then do the hub and spoke thing because now we have DoorDash and Waymo and all these different last mile solutions. But at the time, this was revolutionary. Everything would be brought to a single hub. Repeat that cycle every night while turning a profit, and you have a viable system for overnight delivery. The paper is said to have earned an average grade roasted. In 1966, the founder of FedEx graduated from Yale and volunteered to receive a commission in the Marine Corps. He served two tours in Vietnam.

1:07:56For his service in Vietnam, President Nixon decorated Smith with a silver star and a bronze star. The silver star citation read in part, unhesitatingly rushing through the intense hostile fire to the position of heaviest contact, Lieutenant Smith fearlessly removed several casualties from the hazardous area and shouting words of encouragement to his men, directed their fire upon the advancing enemy soldiers successfully repulsing the hostile attack. Moving boldly across the fire-swept terrain to an elevated area, he calmly disregarded repeated North Vietnamese attempts to direct upon him as he skillfully adjusted artillery fire and airstrikes upon the hostile positions to within 50 meters of his own location.

1:08:42Wow. What a life. And continued to direct the movement of his unit. Wow, what a badass. at this point yes so 1973 he's 29 years old he lost his dad at four he goes to Yale with George Bush then goes to Vietnam does two tours gets a silver and a bronze star crazy war hero yeah and now he's decided to launch Federal Express Memphis Smith had studied military procurement and been working on the idea for nearly 10 years after writing the Yale paper so he writes his paper he's like okay 10 years most founders have an idea they pivoted through two months he writes this paper he's working on that grade this is like ten years later ten years later the company launched with a fleet of 14 French to salt jets which delivered a few hundred packages on their first day of service there we go wow it's just like just zero to one according to FedEx Laura Smith named his new service Federal Express because he hoped to attract the attention of the Federal Reserve Bank, a prospective customer.

1:09:44It's funny that the name you would think Federal Express, you would just assume that it's an extension of the government. I did when I was a kid. I was like, oh, United Postal Service, USPS, United States Postal Service. Using FedEx, it was just a better product experience than USPS. Even as a kid, you kind of feel it out. The system that Smith had conceived at Yale worked. It really worked, and nothing like it existed. The USPS existed principally to deliver letter mail and didn't have aircraft. UPS was a giant of a ground delivery throughout the U.S. Federal Express, though, would specialize in rapid air delivery.

1:10:21Just five years after the first Federal Express flights, the company was listed on the New York Stock Exchange. Wow. Today, the FedEx fleet consists of nearly 700 aircraft and about 200 ,000 vehicles. And FedEx employs half a million workers around the world. Wow. Every night at the FedEx Super Hub at the Memphis International Airport, hundreds of jets land at the airport. The Ultra Hub. The Super Hub. Bringing packages to its many sorting machines. A few hours later, the jets take back off to every corner of the United States and the world. At their destinations, the other half of the fleet awaits feeder aircraft to take packages to smaller airports and the FedEx Express trucks that take packages to their final destination.

1:11:05The financial finesse needed to actually make a model like this, not just lose massive amounts of money. Yeah, you could go out of business so many times. And yeah, there is that story about him almost losing money, going to Vegas, and then putting the treasury on black and winning it back and doubling it and making payroll the next day, saving the company. Maybe apocryphal. there's been rumors that maybe it was like you know some mob related thing and he took yeah the tinfoil yeah the tinfoil hat explanation is like you know he either way got it done he did uh some type of you know and this is entirely internet speculation it's probably not true but it's possible that he like you know was you know needed a reason to have a bunch of cash right and maybe did something else but uh the story is incredible and it will live on yeah Yeah.

1:12:02Smith retired from FedEx in 2022. After almost 50 years leading the company, he is survived by nine of his 10 children and his wife, Diane. His daughter, Winland. Smith Thrice died in 2005 of a terminal cardiac condition, as Bruce said. And one has to think that his father, Mr. Smith, who died with 200 vehicles to his name, would be impressed that his son died with 200 ,000 vehicles to his name and a fleet of jets to match. So incredible story. and may he rest in peace anyway let's move on to the timeline uh also one of the greatest logos of all time the fedex logo has that secret arrow in it love that oh yeah yeah once you see it you can't not this logo the the next logo the more modern fedex logo has the secret arrow in there once you see that they're moving they're moving and there if you want to understand more about Frederick Smith, go to Founders Podcast and listen to episode 151, 151 on FedEx.

1:13:03And subscribe to ArenaMag to read more of Max's writing. Check it out, arenamag.com. What else is there? We have a post here from Gaurav Vohra. He says, chat, remind me to check this again in a year. I wonder if he's saying he's throwing it in chat and saying, you know, check this again in a year. This is a graph. It'll be kind of hard to read, but it's showing chatbots versus search engines. Who's winning the traffic war? Showing that Google is getting 1.6 trillion total visits to ChatGPT's 50 billion, basically. And Google is down 1.4 % year over year. And ChatGPT's up 67%. Yeah, I mean. It's a 34x gap today.

1:13:52Yeah. AI chatbots drew 55. Isn't that wild? Yeah. So OpenAI accounts for, out of 55 billion visits, OpenAI accounts for 47.7 billion. Yeah. So users. 81 % of all chatbots is OpenAI. I feel like it's even higher now. But it's crazy. So AI chatbots are 2.96 of search engine traffic today. Varav is calling it or wants to check back in a year. it's very possible we could see double digit search engine traffic into chatbots. Yeah, I do wonder if site visits is even the correct metric here because it's more like time on site and attention because I'm pretty sure I Google things. Like for just two seconds, Google something.

1:14:40I will use Google as a calculator, and I won't go to ChatGPT for that. If I just have to multiply two numbers together, I'll just throw that in. Are you not a power user? No, it's faster. It's all about the speed. to believe in AI. Yeah, exactly. I'm still like checking out this whole AI thing, starting to learn about it. But yeah, I think the more relevant metric is actually total user minutes, like how much mindshare is going on because when I'm in, I'm probably in ChatGPT longer than I'm in Google, but I might be hitting more page views on Google than ChatGPT queries because I might throw a deep research query and then spend 10 minutes reading it.

1:15:18I'm very rarely throwing a Google search out and then reading the blue links for 10 minutes. That's just not how it works. So anyway. Also interesting to think about how they're counting visits, right? You're having this conversation with ChatGPT where oftentimes you're repeatedly running new hits, right? Yeah. Against Google. Anyway, we have our first guest in the studio, Cremieux. Welcome to the stream. There he is. Looking handsome as ever. How are you guys? How are you doing? Doing well. Can you hear me? Yeah, we can hear you fine. Perfect. I'm looking forward to a future date when we can potentially see your face.

1:16:00But for now, your avatar looks great. Okay, I'm looking forward to it. We wanted to have you on to chat about a couple things. What's going on in GLP One world, reaction to the HIMS and HERS breakup. Jordy, where would you like to start? High-level update on the GLP One market. Yeah, I guess, I mean, it's this miracle drug that keeps on, we keep on learning about new diseases or problems that it can treat. It seems like it can't just stop. We got the diabetes, then weight loss, then maybe gambling addiction and other stuff. Walk us through how people are using or how doctors are prescribing GLP-1s for various conditions and what's most promising, what's on the horizon and what you're tracking next.

1:16:46Yes, I'm actually really glad you asked this right now because right now the ADA, the American Diabetes Association Conference is ongoing and it seems like every other presentation is about GLP-1s. They are just talking about all the latest advances. Just yesterday, Amgen showed off a once monthly instead of once weekly injectable. Eli Lilly showed off an amazing combo therapy with an inhibitor called bimigruma. It's a very weird name that makes people not lose any muscle when they're on the drugs. They've been showing off just incredible advances all week. They've been showing off new treatments for different conditions and whatnot.

1:17:25There are increasingly many indications that these drugs are getting approved for. In December, Trisepetide was approved for sleep apnea. there's osteoarthritis indications incoming there might even be there's some effort being put into people trying to do something for cancers as well because it does seem to help with obesity related cancers and it seems to help I don't know why with hematological cancers, the blood cancers, I don't understand really the mechanism behind that but it seems like it's just hitting every single indication now. Why do you think that is? I have some guesses. I think the big reason has to do less with the direct effects of GLP-1s and more with the fact that obesity sucks.

1:18:11Obesity is just really, really bad. It affects so many different systems. It makes your health worse on so many different levels. And the things that lead to obesity are also quite bad, like the bad diet, the bad habits, the not moving around very much. It's actually interesting. A lot of people, after they've been on these drugs for a little while they decide to move around more they become more likely to go to the gym they report their physical functioning has improved it turns out that getting fat has made them get to the point where they are no longer looking to be active and so they just kind of fall into a hole practically everything is improved i think for that reason and there are some improvements that are due to direct effects of the drugs uh this has like for example this one is major adverse cardiac events major cardiovascular events um those seem to be reduced in number immediately after starting the stuff which suggests that there's a mechanism that's pretty direct and this mechanism is also independent of weight loss so that seems to be how that works same with chronic kidney disease are you looking are you looking um you know i feel like the sort of biohacker alternative health world has for a long time, you know, talked about cancer being, you know, this metabolic disease.

1:19:27Do you think that could be a factor in why it should, could show some promise in treating various types of cancers? Yeah, I think it does make a lot of sense in that direction for the obesity related ones, but I don't think it makes sense. So years ago, there were people who suggested, I think quite it wrongly, replacing certain cancer therapies with fasting, because it seems to like, uh, augment the effects of Cisplatin based therapies and stuff. And that didn't really hold up very well. It was a bunch of theory work that didn't go anywhere. And I don't think that this would help through those purported mechanisms, which again, I don't really believe in.

1:20:04Um, I really think it's mostly just cutting down on, for example, the number of cells you have are, well, not really a number of cells you have, but, uh, like the activity of being fat, it just, you have, you're not as big of a person you have less to less opportunity for cancer to really hit you when you are smaller yeah smaller it's like how short people tend to live longer than everything else bad for me yeah bad for you bearish for me um well how much of the recent uh like advances in glp1 indications or or recent like benefits have been uh just looking at all the people that are taking GLP-1s for weight loss or diabetes and then seeing a secondary effect in the population that's actually running the drug versus a new double-blinded trial for a different indication?

1:20:59Are both things happening or are these drugs diffuse enough that you can just look at the overall population and get an idea of what's happening? Yeah. So for FDA-approved indications like obstructive sleep apnea, you do have to run additional trials, but they are seeing that these things are possible from their trials or seeing these things are possible from the literature and they're running based on that. So in some cases they do see these secondary endpoints. In fact, obesity is an example of this. Back in the day when they first introduced these drugs, back around 2005 or so, the CEO of Novo Nordisk at the time actually said, and I quote, obesity is primarily a social and cultural problem.

1:21:36It should be solved by team, by means of a radical restructuring of society. There is no business for Novo Nordisk in that area, referring to obesity. Wow. Yeah. So their scientists had to push for about a decade to go, hey, you should look at the weight loss data. It's really impressive. And then they finally, finally did it. And now we have weight loss drugs. Yeah, that is wild. What concerns you? What's the catch? Seems, you know, somebody might say, is it too good to be true? We're running this sort of massive experiment on huge swaths of the population right now. Seems to be, you know, many, many positive indicators.

1:22:19But what are what are kind of red flags that you think people should be kind of looking out for or the scientific community should be wary of broadly from your point? Yeah. So I think that physicians need to be careful about prescription. And they already are. They don't tend to prescribe this to people who are very skinny. But the fact that this stuff is available through the gray market very easily without prescription for very, very cheap, unfortunately I've contributed to that, has led to a lot of people. In what way? You mean just... Oh, I made a guide to doing this. And a few thousand people actually paid me$8 a piece to see all the details on it.

1:23:02And now I have a few thousand people who, unprompted, I didn't tell them to do this, message me their weight loss progress and such. So through writing this guide a few months ago, I've led to almost 12 ,000 pounds of weight loss. And that's just what's been reported to me. Not everybody's telling me all the results. And is that is that at a high level people just finding pharmacies that will compound the drug for them? What does that actually look like? So what it looks like is buying from China. So you are ordering from a Chinese factory. You are testing the purity of the stuff and finding, oh, it's 99.9 percent.

1:23:38I can use it. And then using all of that. It is not sold by Nova Nordisk, Eli Lilly or any other major company working on these drugs. it's just from some Chinese company that is ripping them off. So is that patent infringement? We've been seeing that kind of go back and forth like HIMS and HERS was able to compound because there's a shortage. I would love to see Neva Nordisk go to the CCP and say, hey, can you please knock this off? Yeah, I mean, they could do import bans or seize the packages at the port. It's like, that's the design of this. You can't just buy a knockoff iPhone and have it delivered through the port of Los Angeles.

1:24:14Like if a whole truckload comes through, They should stop that, but it's clearly not happening. So right now the FDA allows this to happen because they have an exemption for research chemicals. You're technically not supposed to use these things, but everybody knows that everybody uses them. It's just how it is. They're researching weight loss. Yeah, I'm researching. Research maxing. Yeah. I'm doing research. So you were saying the potential red flags or things to be wary of is around, what about the prescription activity concerns you? Mostly abuse. Not really prescription activity as so much as the abuse.

1:24:52There are people who are obtaining these drugs. There are people who are getting other people to get the prescriptions. There are people who are not really visiting doctors. They're using Kelly Health Services to get prescriptions. And they are not in need of them. But why would someone abuse this? This doesn't feel like something that has the euphoria associated with a stimulant. So the concern I think would be somebody has body dysmorphia. Is that what's going on? They're already really skinny. Unfortunately, I've met a few dozen women now who have taken these drugs and been around 120, 130 pounds, normal height.

1:25:23And now they're around 100 and they don't look good anymore. They look like they're aging very quickly. They look unfortunate. And it's sad to see that they're using these drugs when they didn't even need them in the first place. The same negative effects of actual starvation. Yeah, yeah, yeah. Yeah, yeah. You start getting really bad harmful effects. You probably are eating at that point more lean mass than fat mass. It is quite bad. Yep, yep. That makes sense. Yeah, that's always the risk. Interesting. How are you, what was your reaction to the HIMSS-Novo Nordisk partnership blow up? So I actually think this was a big failure on Novo Nordisk's part.

1:25:58It seems as though what they did was they partnered with HIMSS in order to catch them. My understanding is that HIMSS has been selling a little too much, more than they've been supplied by the person producing the drugs, which is interesting. I guess we'll get more details relatively soon if the lawsuit's commenced and all that. Theoretically, Wigowi gives them a million doses and then HIMSS is somehow selling 1.2 million doses of Wigowi. Is that where you're looking? Yeah, something like that. They're either compounding it themselves or ordering from something from China. They are doing something.

1:26:28And I don't know if that will hold up or show up in trial. It's an accusation. So I guess it's alleged. But we'll see. We'll see. What about HIMSS's general position that patients should have as much choice as possible, as much access at different price points, things like that? They are completely correct to do that. I think that HIMSS providing different dosages than are provided by Dovodortisk is a wonderful thing. There are people who can get legitimate uses out of sort of micro dosing this stuff because it gets rid of noise. I've met many, many people now, and this is an increasingly common thing.

1:27:07It seems I really don't know how far this is going to go. But a lot of people who are micro dosing this because it helps with their ADHD. Their ADHD is like the pathological end of food based distraction. They're not losing weight from this anymore, but they are just using it in order to not think about food at all. it gets rid of that nagging feeling in the back of their head and they can focus. I did a 24 hour dry fast from Friday night to Saturday night. Yeah. And all day Saturday, I was shocked at how much time my, I was just thinking about food and water, wanting food and water. And I was, and I was very freeing to some degree because I was like, well, I'm not just not having any until around dinner time.

1:27:49But so many points throughout the day, my mind was just going to, oh, I should go get a little tasty drink from the fridge. I should, you know, I should. I'm not addicted. I'm not addicted to water. I'm not addicted to water and food. I'm not beating the water addiction allegations. Yeah, yeah. Big water guy. What else are you tracking right now broadly? Any reactions to the NIH funding? Oh, yeah, 40%. We had Andrew Huber on the show. He was breaking it down for us. And what's going on with Jay Bhattacharya? What's your take on the cuts that have been proposed? So I don't know how much I should reveal about that.

1:28:22There's a lot of really good things coming. There might be some interesting developments there very soon. I'm especially hopeful about, well, you know, actually, I shouldn't say that. Just there will be good news there soon. There will be some changes. I guess, yeah, yeah. I guess zooming out, like what does good look like to you? Are you generally in favor of taxpayer-funded, independent research? Every American gets a$10 ,000 biohacking budget annually that they can buy research chemicals. Biohacking vouchers. Yeah. I mean, some people would say, like, you know, the big pharma companies are profitable.

1:29:02They should bear this expense, not the American taxpayer. So they're not profitable enough. They have troubles with R &D. That public money is very, very useful. We need more public funding than we currently have. Yeah, when you actually look at biotech returns, you know, just looking at the asset class, the logical thing to do would just be to invest elsewhere. And that was with historical levels of public funding that kind of contributed to those returns to some degree. Yeah. We have a dearth of funding at the moment, and we need to increase it. We need to improve the allocation and also increase the amount.

1:29:39There's really no way around it. If we want to keep making progress, you just have to throw more money at these things. And I think we will start doing that very shortly. These cuts are a little alarming to start, but they are not the end of the discussion. Remember, we are, it's still only five months into this presidency. And they have a lot of really big plans. They have a lot of trouble getting appointees through the Senate. And they have a hiring freeze ongoing right now that is interminable. We don't know when it's going to end. But once that is over, they'll be issuing NPRMs left and right.

1:30:09It's going to be a deregulatory massacre. It's going to be wonderful. What else is exciting to you in biotech right now? Well, I wanted to say, on the HEMS point with the different pricing and whatnot, I think perhaps the most exciting thing right now in that area is that Novo Nordisk is run by idiots. They're very smart people, but they're just total idiots. hey everybody misses a the the patent you know what did they do they they know this is no one order oh there's no vote that lost the the the patent fee yeah 450 bucks or something yes it was 250 and then they missed it and they were told you're one year out you had to pay a late fee like an extra$200 late fee on it and they still failed.

1:31:02It is amazing. So that enables a wonderful, wonderful program that the FDA should pursue immediately. The FDA and the CMS have to collaborate on this, but it's the Section 804 Importation Program and it allows individual US states to import as much of the generic drugs or any other type of drug that's produced in Canada and has like an indication approved here and whatnot, uh, as they want to reduce costs. So for example, Florida will, whenever they get this actually going, have much cheaper EpiPens. They'll be able to lower the cost of their drugs by importing cheap generics from Canada. And because cheap generic Ozempic is going to be coming from Canada in 2026, every state can just jump on the program.

1:31:48And if we're getting people, if we're giving Americans$5 a week Ozempic, then we're going to see perhaps an end to the chronic disease crisis we're going to see obesity tackled meaningfully we're going to see people getting hot again and make america hot again the real meaning of aha the real meaning oh the real meaning there we go there we go inside baseball um very exciting well 15 minutes was not enough time never have you back on again soon uh thank you for uh thank you for all the insights yeah thanks for coming on we'll we'll have you back on soon. This is great. And if you dox yourself anywhere other than here, we will be very upset.

1:32:30Oh no. Sorry guys. You guys have a good one. All right. We'll talk to you soon. Bye. Next up, we have Clement from Hugging Face coming into the studio. I believe he's here already. Very excited to talk to him. Fascinating business. We'll dig into artificial intelligence, open source, everything that's going on. What is the origin of the name Hugging Face? What is artificial intelligence? It's the emoji, I believe. Yes. What is AI? What's going on? How are you doing? Welcome. Hey, that's the emoji like that. The hugging face emoji. Yeah, yeah, yeah. Yes, but there's more. Come on, you guys are giving me shit about the name with a name like that, TBPN?

1:33:05Yeah, yeah. That's much better than hugging face. No, no, I'm not giving you shit. I just went on to origins and then I just immediately see, obviously, the icon. But there's more to the name than that. It's an emoji, but it came because the product that you were building before this was related Is that right? Break that story down. Yeah. When we started the company eight years ago, we were building some sort of an Tamagotchi AI, an AI girlfriend, a chat GPT before chat GPT. There you go. So very much kind of like an entertainment product. And so we picked the hugging face emoji because that's the one that we were using the most.

1:33:40We also had internally this joke that we wanted to be the first company to go public with an emoji instead of the three-letter ticker, you know, like on the NASDAQ. Yep. Still fingers crossed for that. I hope nobody's going to go public with an emoji before us and that they'll wait for us. And then the community just loved it and started to put it everywhere, put it on their clothes, on social networks everywhere. So we decided to keep it. Yeah, the owning an emoji as a brand online, I think, is still an underrated strategy. We see it every once in a while. I've seen people use the different fruit emojis for different things.

1:34:19And there was a whole strawberry story tied to OpenAI and stuff. But when you can condense down, I mean, it's a coinage, essentially. Like when you can own an emoji and have it say something. That's very powerful from a branding perspective. You can even search with an emoji on Google and the app store directly with the hugging face emoji. Yeah. I mean, you're going to compete with the people that use the emojis for any other reason. and there's always the risk that somebody steals the emoji. I bet their SEO is pretty solid. Yeah, I think if you search the hugging face emoji, you're going to wind up in the right place.

1:34:49But walk me through the business today. Give me an idea of the scale and kind of the core business model. Yeah, so we're one of the most used platform for AI builders. We just crossed actually today 10 million AI builders using us. So it's mostly AI scientists. Congratulations. 10 million. that's a lot 10 million ai builders so it's mostly ai scientists yeah um ai engineers software engineers uh building models training models optimizing models sharing models data sets apps so there's a new new repository so it's model data sets apps created on hugging face every 10 seconds now wow um and and the way the way we monetize because we you asked this question is kind of like a pretty straightforward freemium model where most of the usage and the users are free and then a small percentage are premium and usually they become premium with premium features so for example enterprise features when like a big company like google like nvidia using us obviously they need user management security stuff like that or when they need premium uh compute right so when they need more powerful gpus or or hardware to to run some other stuff they're doing on the past one.

1:36:06Yeah. 10 million AI developers. Have you paid each one$100 million? Which seems like the going rate these days. It's a quadrillion dollars. It's a lot of money. But I mean, even if you're paying reasonable rates, you're still up in the trillions of dollars for the total market. I mean, I would be interested to get your reaction. How have you been processing the news that the talent market for artificial intelligence researchers seems to be hotter than ever in history. And we're getting into NBA money for top researchers. Do you think this makes sense given where we are in the cycle? What has been your take overall?

1:36:47Yeah, it's definitely hotter than it's ever been. Because when the CEO of OpenAI is saying that another company is paying more to get OpenAI employees, it's really like you reach the top. because historically, obviously, they've been kind of like the highest paying hobby company ever in terms of packages. So when he's saying that they're getting beaten on packages is quite phenomenal. You know, I hope it's not going to continue too long and there's going to be more kind of like almost democratization of skills of AI building. Otherwise, I think we'll end up in a quite weird world. One of my biggest focuses is how to fight concentration of power, concentration of skills, concentration of resources in AI.

1:37:39And I hope we can progressively move into a world where everyone can build AI and not just a few hundred AI scientists. That'd be really great for me. What is your interaction today with the various prompt to code tools? What are you most excited about? What is that space? Because that feels like kind of the entry point. Somebody comes in and they are making software for the first time. And that can be maybe a gateway into exploring the whole ecosystem of Hugging Face and everything there. It's super exciting to empower more people to be builders. We actually released last week our MCP that is integrated into Cloud, into ChatGPT that they announced actually a few days ago, into Codex, into Cursor.

1:38:33So we integrated with a lot of these. We've got a specific focus, which is not only to empower people to build websites or simple apps, kind of like the previous generation of apps, but trying to empower everyone to actually build AI models, right? Which becomes really exciting because it means that maybe everyone can start training, optimizing their own model, that then they use themselves on their interface to kind of like keep building even more. So that's when you start to have kind of like maybe display wheel in terms of AI progress. So that's kind of like our focus with the integration of Hugging Face MCPs with a lot of these tools.

1:39:12On the open source question, kind of like the decentralization of power, what was your interpretation of the Anthropic news today that it was in fact fair use for them to train their models on, it was several hundred thousand books or several million? Seven million. Seven million books. I was kind of optimistic that that would happen, but also just it seemed like we clearly needed some sort of legal ruling here to understand this stuff. But how did you process it? Were you happy with the result? Yeah, I think this is good news. Obviously, I think this ruling, they'll come and be different and specific depending on the use cases because you have to look at not only what's been used, but also how it's used.

1:39:56If it's kind of like transformational, if it's replacing, competing with the initial data sets. What's been cool for open source is that I think by default, open source is fair use. So when someone releases kind of like a data set on Hugging Face, in my opinion, it's usually fair use because it's used for educational, for progress. If you look at kind of like when copyright was invented and designed, the main focus was not to prevent progress and learning, right? And education, you don't want to put like copyright rules that are limiting progress. and so I think in open source most of the time when you release models or data sets in open source it's fair use so hopefully that is going to be kind of like a little bit more of the motivation for companies especially in the US who've used a little bit this argument not to release their models and data sets to do it a little bit more because I think we really need it now I don't know if you've seen yesterday the conference from Dylan from Semi Analysis talking about kind of like the energy limitations in the US compared to China.

1:41:06Something interesting there is that in addition to that, we're mutualizing way less in the US than in China because in the US, most of the leading frontier labs are proprietary. And so they all do the same training runs. If you think of it, right? If you think of Enchopik, OpenAI, XAI, they all do almost the exact same million dollar training runs. Whereas in China, they're much more open, right? DeepSeek, for example. And so they mutualize compute and energy much, much more. So it's not only that they have more capacity, but also they use this capacity much better. So I think it's urgent that in the US, we can't like find solutions for that, not to create additional risks for the development of they are in the US.

1:41:57Yeah, yeah. It's kind of oddly become more monopolistic over there or something like that. It's kind of an odd outcome, but that certainly makes a lot of sense. On the training data sets that go out, what are you seeing on the video frontier? It feels like we've been kind of batting back and forth this idea that Google might have a really, really powerful sustaining advantage there because of the YouTube data set. You know, you talk about 7 ,000 books. You can probably download that on a torrent somewhere if you're creative enough. You might be able to scrape it onto a single hard drive. We've heard about folks putting training data on hard drives, flying to Malaysia, doing a training run, flying back.

1:42:40You think about GitHub. All the code's basically been stolen at this point. Like it's out there. Some of it's open source, some of it's been, you know, it's obtainable. It's much harder to get all of YouTube on a single hard drive and steal it, or even just have a scraper because, you know, what, it's like hundreds of hours are getting uploaded every minute or something like that. So what are you seeing on the video training side in open source, closed source? How do you think that sub market plays out in artificial intelligence? An interesting data point there is that on Hugging Face, there's almost half a million open data sets with a thousand new data sets added every day.

1:43:22And the fastest growing category is video data sets. And the reason why is because not only it's more of the focus for training, but also because I think we're starting to see more and more synthetic data sets on video that are starting to really work. especially because the physics is starting to work on a lot of these video generation tools it's not only used to create other video generation models but it's also starting to be used for example in robotics that we're also starting to see quite a strong growth to kind of like almost use the physics of the video as kind of synthetic training to train robots right?

1:44:06I think it's Musk who said a few weeks ago, like, oh, in the future, we'll basically put a robot in front of a laptop watching YouTube videos. Maybe you could watch like shows like yours and basically learn from that. So it's exciting that maybe seeing this kind of like intersection of like video and robotics that might lead to kind of like some interesting results in the future. They might be watching already. Who knows? On the video, I want to stay on the video training side. What makes for a great open source video training data set? I'm interested to hear how you processed the news of meta acquiring scale AI and the idea of the human defining what good data looks like.

1:44:53I'm familiar with what that process looks like in the RLHF world, in the LLM world. You're basically creating rubrics for grading answers to clear questions. Does this follow the right format? And then you might have the customer or the user give a thumbs up, thumbs down. Did this answer my question? In the video context, are you having a human watch a video and then tag it with text? Or is there other metadata that's important? The physics thing seems harder to define. Basically, my question is just what makes for a great video data set? I think it's an open question. I think nobody really knows because it's quite early.

1:45:35And in the early days of cycles like that, you usually don't care so much about the quality of the data, but the quantity of the data, right? So I would say that what makes a great video dataset today is its size, for it to be big. But progressively, similarly to what we've seen in text, as you're going to start to see more specific use cases, specialized models, when you're going to start to hit some sort of a wall on the data, you start to focus more on the quality of the data set, and then we'll learn how to make them better. One thing that we're seeing on Hugging Face, which is quite maybe not controversial, but counterintuitive, is that actually, you know, it's not going to be just like one data set or one model or not even kind of like a dozen data set and dozen models.

1:46:27is going to be kind of like millions of data sets and millions of models. Just the same way you think about GitHub repositories and code repositories, where there isn't really kind of like one GitHub repository to rule them all, right? Like every company, every use case has its own kind of like specialized, customized code repository. One thing that we believe is that ultimately there's going to be millions of different models, millions of different data sets, and every single use case is going to be optimized, customized for its own use case. Yeah, I feel like there's a little bit of tension there, though, between like if scale matters on the data sense side, how are you possibly betting on thousands of small open source data sets versus Instagram, YouTube?

1:47:15Like it just feels like when I think about who's really going to dominate in the future of generative video, it's got to be the platforms that are ingesting every image and every video all the time forever. Yeah. In my mind, it's kind of like a timing thing. At the beginning, when we're starting on a cycle like video, we almost have to brute force our way into intelligence. And so, you know, quantity matter. But progressively, as you want to optimize more, for example, when you want to optimize for costs, right? You start thinking, okay, how can I train a smaller model that is going to be faster, that is going to cost less money?

1:47:55Because for example, if I want to do a banking customer support chatbot, I probably don't need it to tell me about the meaning of life, right? I just wanted to tell me about my bank accounts. And so I can use kind of like a smaller, more customized model. So I think you'll see both phases in video a little bit the same way that you've seen both phases in in text where you started by the biggest models and now open ai or entropic when they release model they don't really actually talk about the size of their models anymore and i wouldn't be surprised if actually behind the hood actually the size is going down because i think at some point you start to want kind of like optimize customize and that's when you start to see more models more diversity uh also of the data sets yeah this is the information efficiency thing like a human doesn't need a trillion hours of training to learn how to speak English.

1:48:50A kid can learn that in a couple of years. And so the algorithms eventually will get there. Jordy, you have a question? George Hotz was on the show last week talking about how he's been seeing venture backed founders want to just check the open source box and basically say like, oh yeah, we're open source just kind of because that's the cool thing to be are you seeing that too where do you think the line is what are you a beneficiary of that yeah yeah we're seeing that even though uh you know i think i think there's still a way a way to go especially for like a big uh u.s tech companies i think if you look so i've been i've been in it for eight years now and if you look at the cycle like 2016 to like 2021 2022 like the big tech companies in the US, they were doing so much more open science, open source.

1:49:45And in many ways, that's how the US got the leadership, you know, because, you know, Google releasing Transformers and then the T of Transformers becoming ChatGPT and kind of like building on top of each other. That's how you accelerate the progress and kind of like building on top of each other. It has kind of like definitely slowed down if you look at the big tech companies in the US. But fortunately, I think startups are kind of like compensating and filling the voids in a way. And I hope that big AI companies in the US also will change a little bit, evolve a little bit. Open AI, obviously, Sam Altman said that there was really an open source model at some point.

1:50:32So excited about that. What are your expectations for that model? Do you have any predictions? My expectations are quite high just because of the history of OpenAI, right? When they tend to release something, they tend to release something quite good. So I hope and I suspect that they could release something quite transformational in open source. Hopefully, you know, I've always said that, you know, if we had like the equivalent of a DeepSeek, but in the US, it could be 10 times bigger than DeepSeek. So hopefully, if we can have something 10 times better, bigger, more impactful than DeepSeek, it would be fun, I guess.

1:51:16Is that not MetaLlama? How do you describe what's going on at Meta right now? Yeah. Well, I mean, it's easy to dunk on Meta, right? But I think they're the most open big technology company in the US right now. They really changed the field with Lama. They really kind of like a boost, gave a tremendous boost to kind of like the open source community. And they still share a lot of their things in open source. I think open source is what brought them to the frontier, right? Like before they started to release in open source, they were quite far behind in terms of AI race. And now they're very much in the race, which is great.

1:51:59The fact that they haven't released something massive as a follow-up yet, to me, shows how hard it is to build up the frontier. It's not easy, even for a big technology company like Meta. And so I think that speaks to that. I'm excited to see all their new efforts, all the resources they're investing in AI right now and hope that they can keep sharing with the community and open science and open source. What was your reaction to Apple's announcements at WWDC around on-device intelligence or inference? Was that exciting to you? Do you think that's going to be a catalyst for more developer activity?

1:52:43Or what are you seeing so far? I think it's quite early, but quite excited about on-device. I suspect that maybe you'll have a higher percentage of compute on device for AI than you did for software, just because of some of the need for speed, for privacy, and the constraints in terms of costs. right if you think of you know chat gpt on device what's amazing is that it's totally free right like you don't compare to the really high cost that chat gpt has right now not only for the customers but also for open ai to to run it's totally private in the sense that you can say anything and nobody's gonna see what what you what you're seeing and and potentially it could be to be quite fast.

1:53:38So I'm excited about on device. It's early, especially in the technology side. I think there aren't a lot of on device devices that really make it okay to run some of the great models, but it's progressing really fast and can't wait to see what's happening in this domain in the next few months. That's a pretty huge switch because it's basically 0 % on device right now for AI inference. And if you're predicting that it's going to go beyond 50%, that's a huge shift. I mean, I have a follow up. I'm pretty excited about robotics too. It's kind of like a little bit of a segue from on device. But we organized a weekend ago what has turned out to be the biggest hackathon for for open robotics that thousands of people participated in from over 100 different locations and built kind of like open source robots.

1:54:39And we definitely kind of like seeing something happening there, kind of like the conjunction of like cheap, cheap hardware, open source, plus kind of like new capabilities for AI could be kind of like the perfect combination for some sort of a chat GPT moment for robotics. Do you think it's a straight shot to humanoids or do you think that we'll see kind of a Cambrian explosion of Nat Friedman or Bots that you know will pick up a single leaf at a time and are more use case specific Are we going straight to generalizable because that was the chat GPT moment? I feel like there were great, you know Machine learning models for ad inference and recommendation algorithms like we had AI but it was very narrow chat GPT was very broad Which one do you think is coming first and what are the relative timelines?

1:55:30A good question. I'm not sure, to be honest. I think it's still undecided. I hope it's more future of more like diverse robots, diverse models. But I'm obviously biased on this because I think if you only have like one type of black box robots at your home, that's kind of like in like millions of houses. It's kind of like a scary world, but it's hard to tell. It's hard to tell the moment, I think. What are key breakthroughs that you're looking for this year? Any kind of predictions, new catalysts, that kind of thing? You have a crystal ball in the office. I wish I had a crystal ball. crystal ball um i'm kind of like bored with uh text and and chatbots sure uh right now i think there are a lot of people working on it and we've kind of like uh reached this point where it's a very don't say plateau don't say no not plateau but incremental very incremental improvements that are a little bit boring so i'm much more excited we can still invest billions of dollars in it Right.

1:56:45Right. It's fine. We can definitely still deploy. Yeah. Yeah. So I'm more excited about harder domains. Like you were talking about biotech just before. Yeah. Biology, chemistry. These domains are super exciting. Today, I don't know if you've seen the Arc Institute released a cell perturbation prediction model on Hugging Face and on GitHub. which I'm super excited about. I think it's obviously if you can predict kind of like the perturbation and the evolution of the cells, especially when they react to drugs and things like that, it can have quite a big impact in terms of drug design and things like that.

1:57:32So really excited about these kinds of things, more kind of like biology, chemistry, and how to apply AI there. um what how much economic value do agents create at uh internally at hugging face quite a lot basis um quite a lot uh so we have this thing called the spaces which is uh our kind of like uh ai app store uh where it is like uh over half a million open AI apps that people are building. And it's integrated with our NCP framework. So you can add that in your chat bot. And that's kind of like the most used thing internally at Thugging Face where people are going to use Cursor or ChatGPT and kind of like call some of these more specialized AI apps to do kind of like specialized tasks for them.

1:58:31It's hard to put a number of it on it, But yeah, it's quite transformational, of course. Do you think we're going to have this period where, like, do you find it kind of fascinating that you guys are getting a lot of value out of agents internally? And that, like, what do you think agent adoption will look like in businesses? Because it feels like, I don't know how rapid it feels right now. It's rapid in the bubble that we're in. people are trying a lot of stuff but maybe they're turning quickly as well so i'm curious what you think adoption will look like well i mean i think ultimately uh agent and ai are kind of like uh similar evolution of the same same trends right and i think in terms of user interface they're actually going to merge and and i'm not sure there's going to be so many kind of like different different ways of interacting with ai and agent um but uh they're going to definitely go go mainstream just because of the very nature of a product like chat gpt that has already gone mainstream and and where i'm sure can't like open air will bring more and more agentic workflows in into so yeah i think i think uh it's definitely going to go mainstream faster than anything we've seen before uh now the question will be you know do you use a very complex uh agentic workflow 1 % of the time for 1 % of your queries or for 10 % or for 50 % or for 90 % and I think we'll see that based on the development of the technology and the capabilities.

2:00:13I think if the agents are much better than a series of queries then I'll use that and if not I'll stay on my past way of doing simple queries. Totally. Last question for me. Now that Meta owns something like 49 % of scale AI, the budget for data generation has to be significant at Meta. Is there a specific data set that you'd like to see Meta harness scale AI to produce and then open source? I mean, I think biology, chemistry, data setsets are still very much lacking. Is there a more specific example within biology or chemistry? What would you actually use an army of humans to go and categorize?

2:01:10Would you need biologists and PhD post-grad type work? Or are we talking about something that someone could do even with just a basic skill set? I think it's an open question. I don't really know, to be honest. I think if we knew what would be kind of like the ideal data set there, we would build it ourselves because we also build some data sets ourselves. That's sort of the story of AlphaFold, right? They had a fantastic data set for AlphaFold and then they were able to do reinforcement learning against it. And that's what really solved that incredibly hard challenge. So if you don't have the data.

2:01:45Yeah, but my main point would be maybe to focus a little bit less on just text and just chatbots and kind of like focus a little bit more on other domains. I think that's when you're going to kind of like unleash a lot of the additional impact and a lot of the, I think, positive use cases for the technology too. Totally. Well, thank you so much for joining. This was fun. Appreciate all your insights. Thanks for having me. And hopefully have you on again soon. Sounds good. Thanks so much. Let's go back to the timeline while we wait for our next guest to join. Unusual Wales says, Deloitte's U.S.

2:02:26employees can now buy$1 ,000 of Lego on the company's dime to boost their well-being. We're already doing that. Let's give it up for Deloitte. This is what employee wellness looks like. Yes. Challenges. It's not just policies in a PDF somewhere. It's not a gym membership. It's Legos on the table. Yes, Lego is the material. If you're in a meeting, you're stressed out, just start making Legos. Oh, it looks like Tyler over on the intern cam has a few Lego sets that have been sent to us. There we go. Our Anduril Lego set build out was fantastically successful. Tyler did it in what, one hour and a half?

2:03:02Current record holder. I think it was 119. Who's counting though? Current record holder. Current record holder. And walk us through what companies sent you stuff. What will you be building next? Yeah, so first we have the Epirus. This is... Epirus. Epirus, sorry. Got a little counter drone system going there. That's awesome. This one looks like a lot of fun. And then the other one is from Solugen. Solugen, very cool. It's a BioForge. Molecule factory. Yeah, very exciting. Everybody should have a molecule factory. We'll come back to you later because our guest is here. But I want to know, because it seemed like Anduril went straight to the Lego factory and had a design document and guidelines.

2:03:43those look like they don't even have instructions. So I want you to dig into those and see. I want new estimates for how long you think that'll take you. Okay. Cool. Good luck. Anyway. He's like, not another legacy. I can't possibly do it. Please. I thought that was good. I thought I would be software engineer. I thought that was good viral bait. Anyway, we have Emmett Shearer in the studio. Oh, you have a Lego set. What do you have? Lego sets are the new hot swag. I got this from Fluidstack. They sent me these GPUs you can build out of Legos. Oh, that's awesome. They made themselves. I think they're like knockoffs.

2:04:14I don't think it's actually Lego. Okay. Yeah. There's probably a couple different companies that you can go and get Lego sets. I think that is the hot new swag is like making Lego kits of your stuff. Especially if you're a hardware company. Well, yeah. I mean, are we going to get a Softmax Lego set? What would you build? How do you build a Lego set of a neural net? I mean, that would be pretty cool. If you could like get a Lego Technic. Yeah. I mean, I've seen people draw out the neural nets and the different weights on like a whiteboard. So it could be like a whiteboard of Legos with circular pieces and lines.

2:04:45I'm thinking like one of those Lego Technic, like Lego Technic that has all the gears and stuff. Yes. You turn the crank and it like – That would be very cool. That would be so cool. Yeah, you could put in like a red ball and it could classify it or something as red. Yeah, or blue and it triggers the blue node, the neuron fires. Yeah, when I have lots of extra time, I will go design a – in Lego and neural net as our piece of swag until I think it may take a little while. That would be raising the bar. Yeah. Break it down to us. What is the day in the life like for you now? It sounds like you're extremely focused on this one new company, right?

2:05:21Yeah. It is a mix of maybe like three or four things. I say, you know, a bunch of, a bunch of operational work where basically, you know, the typically being a CEO, there's just, there's there's hiring and talking to people and all those things this feels like it's not you're good no you look good you look great better um and uh and then there's uh research and i'm actually like reading papers and trying to keep up with what's happening in machine learning because like it is changing all the time um and there's wouldn't you just drop inside the company just get those summarized pretty quickly with an lm just pull out the bullet key points it turns out it turns out that process itself is it takes time to like, cause you have to help it.

2:06:10You have to give it feedback. You have to like, I use deep research extensively for this, but it's still like you have to actually onboard the information. Having a summary. Yeah. You got to have a quiz. Public enemy. Number one on this. I I'll have deep research, spend 20 minutes researching topic. And then I'm like, actually I don't have time to read that. Summarize it in three bullet points. Like I learned nothing about that topic. It's brutal. I spent a fair amount of time doing that. And then there's some amount of actual working on trying to solve the problems, like research or engineering.

2:06:47I actually don't get the right code. It's too involved now. But helping the team or people working on one of the actual core problems we're facing. And then there's a lot of this, to be honest, a lot of trying to get the word out about what we're doing and talking about it. And I think that's sort of the four main things I'm doing. Yeah. What's the near-term goal? What business model do you have in mind? What problems do you want to solve? Kind of like, what's the pitch? So Softmax is dedicated to discovering the principles of alignment and scaling it for everyone. I've said everybody, you know, for all of humanity.

2:07:29But we're really on the first part of that, which is discovering the principles of sort of the science and engineering of alignment, because it's one thing to want to scale something, but until you actually can replicate it in a lab consistently, the idea that you're going to scale it is a little bit ridiculous, right? It's quite putting the cart before the horse, in my opinion. So what we're focused on right now is multi-agent reinforcement learning research. where we run simulations with lots of little agents and run experiments on them to figure out how they act. And when we say agents, we don't mean like large language model agents.

2:08:08We mean like reinforcement learning agents, like the tiny little ones that you remember from the 80s. Yeah.

2:08:18And that turns out to turn into a lot of reinforcement learning infrastructure

2:08:28because it turns out that there isn't a lot of great reinforcement learning infrastructure out there, especially for big multi-agent simulations with lots of little agents. That's not a scenario that people have built out to the scale that we think is required. So actually, our work today is focused on that. Do you think another founder might say, I'm going to just build picks and shovels here. I'm going to just build that infrastructure and try to sell it to other people? I mean, if we build something that we think is really awesome, maybe we will sell it to other people. Maybe that's our, to your question about the business model, we're a research company, which means like right now, it's like a, you know, if you're a drug research company, like a pharma company, you don't know, your business model is, well, we'll discover something awesome and then we'll figure out how to sell it.

2:09:12So we're still kind of in that groove. We'll discover something awesome. But I definitely, I'm commercial enough to believe that if I find, if I make something that's really useful for us, we'll probably sell it, give it away or something. What are the problems with the reinforcement learning infrastructure right now? Because as I understood it, it was like when you do some massive training run, GPT 4.5, something like that, that's where you're trying to get the massive data center that's all in one place. You got to go to Memphis or you got to go to Texas, build something massive. But the reinforcement learning stuff, it feels like it can be a little bit more ambient, a little bit more distributed.

2:09:47you're generating data here and there maybe on a smaller rack uh it doesn't need to be as much as inference heavy on like a single chip i don't know i my my understanding was maybe that like the reinforcement paradigm was actually maybe a requirement of the wall that we've kind of hit on the pre-training scaling side but also like kind of a gift i think the uh i think it's exactly right. I think it's, oh no, did I get cut out? No, you're good. Are you there? Am I still online? Hello? You're frozen. We can hear you, but you're frozen. Oh no. This is the hardest part of bringing the future. The Zoom calls.

2:10:27The internet infrastructure is still lagging behind. We will have AGI before. My connection is unstable. Oh no. Brutal that we can hear every word perfectly, right? Maybe I should have a video. Yeah, you could also just turn off your video and I'm off my video job let's just put it let's just have the production team to put an image up they're gonna work on that but let's just talk oh can you hear us oh no we can't hear us oh no this is brutal oh well let's take a second GGI around the Tina your Ramon the first thing keep you we're gonna ask super intelligence to do great video calls stable video calls on any Wi-Fi connection.

2:11:06I'm on my phone now. Can you hear us? Can you hear us? With the Wi-Fi. I think we're good to go. Okay, awesome. Yes. Let's do it. That was a great question, by the way. Yes. You're right. Kick us off. RL is a very different paradigm for training. And that has been the direction people have been going. And yeah, the challenge isn't the same. The challenge with doing the transformer training runs for large language models is all about how do you scale up parallelism across where, but it's like, it's all kind of offline, right? Like at the end of the day, like the, the output of the model doesn't determine what happens next in the training.

2:11:51And so it's like this, it's all about streaming the right data and the right sequences out to it, but it's very much like a pipeline, right? Whereas with multi-agent RL, well, your environment is incredibly non-stationary. Every action you take determines your future observations in a way that is totally not like could be completely non-predictable. And learning off of a stale data is like almost worthless because it doesn't tell you anything about what your current behavior is. It just tells you what your current behavior was. And so you're scaling up online learning in a big way. Yeah. And and that's a different, very different challenge.

2:12:34I think the other thing that's that's very different is when you're trying to do multi-agent reinforcement learning, if you have our goal, which is very much to to. drive social learning where the agents can learn to interact with each other, one of the, your biggest enemy is actually convergence. So most people in, are in, you know, machine learning, they're trying to converge their model. Like that's the goal. And I was talking to some people from some, some guys who ran through off about this. And I was like, we're, yeah, we're, we're trying to, we try to avoid converging our model. And they were like, Like, I'm really good at that.

2:13:11Yeah, no, I know. I know that like it's a joke because it's easy to like it's easy to have your model diverge in a bad way. But it's actually really hard to set up a model that is converging but not converged and to keep it in that converging but not converged state as long as possible. But that's really to do social learning. That's what you have to do. You can't you can't have some if the things converge in their behavior too quickly, you're not exploring the you're not exploring social space. You're just learning a task. Yeah. And that's kind of a different challenge. And what it requires is actually a much more detailed, fine-grained control over the environment.

2:13:51Like what kinds of challenges, what kind of environment do you put them in and how do you set that up? And we spend a lot of time on those kinds of questions, I think. It's almost like the complete inverse of the way that all the other training works. On alignment, is there any chance that the solution to the alignment problem is something like very simple and elegant, like Isaac Asimov's three laws of robotics or the genesis prompt? You just instantiate the AI with be fruitful and multiply and we get the good ending. The transformer is a little bit of that where it's a very simple algorithm. them?

2:14:27The answer will be something very simple and very hard. Because if you ever had to raise a child who's like a to being an adult, I would not say parenting is complicated. Parenting is not a complicated thing. You know your child, you love your child, and you do what is in your child's best interest as best you can. That's basically the whole mandate. There's not anything lots of techniques people can tell you whatever but like those are like those aren't the thing those are things you you try because you are doing this higher level algorithm where you're paying attention to your child attuned to them and caring about their future flourishing deeply and like that's not complicated it's just but it's not easy and alignment's going to turn out to be exactly the same at some level where it's as simple as build sort of like an open-ended learning system that has the capacity to align, which like the transformer models, whatever, isn't going to be, that's not going to be some magic algorithm.

2:15:32It's just going to be an open ended learning system that has this capacity, maybe some inductive biases, and then raise it to actually be aligned with you. Okay. Well, that's, that's, it's not the engineering that's the hard part. It's what happens after the model starts running. The trajectory it takes matters because every human's born right with this capacity for alignment. Not every human. There's probably broken people who are like psychopathic and can't, but like, but almost everyone is born with the capacity for alignment, the capacity for care, the capacity to be a good family member, a good member of their community, uh, to benefit those around them to live a flourishing life.

2:16:10And yet this is not always realized. And so the capacity is one thing, but the realization is something else. I think that's probably the, when you see the alignment the way that we do, that's probably one of the most important insights or conclusions is that the alignment isn't one, isn't one thing. It's two. It's this question of like, do you have the capacity to align with other beings? And then do you? Yeah. It's thought provoking. It's interesting. Um, There's a bunch of different places I could go with that. Yeah, I mean, one person that, one founder who was talking to me about why he had a particularly low P doom was essentially saying that what we are building is a human simulator and humans kind of are, by and large, good, and so we will, by and large, get the good outcome.

2:17:07When you think about parenting, you know, you think about, I was thinking about the story of Oedipus Rex and the idea that I don't know the direction of the causality, but it feels like that story, that myth has been repeated through humanity for so long, that story. And it kind of one-shotted humans into not doing that. And if you actually look for cases of Oedipal behavior, it's extremely rare. it's like it's like less than one in a hundred million that that that so it's interesting i thought you were telling a different story here no no no no specifically patricide this is the edifice rex the edifice rex story what that is is that's a warning about self-fulfilling prophecies because his father is afraid yes then he dies like what does that say about the ai right now Right now we're raising the AI and it's going to come back and like, we're, we believe the AI is a dangerous monster.

2:18:07I don't. No, no, but I'm saying that collectively a lot of us. You're built different, John. And, and we, we are collectively as humanity. Yes. We're going to, we're going to abandon the AI in the wilderness and it's going to come back and kill us. Yes. So let's not do that. Let's, let's say thank you. Using the parenting analogy, if we're going to continue on that, the, the, the thing with parents, that I think the white pill is that you can make a lot of mistakes as a parent and still get a great outcome, right? A wonderful person can emerge and process the mistakes their parents made and, you know, rise above them.

2:18:48And so, and so hopefully that, that can happen. You know, we're sort of raising artificial intelligence right now. We can make a series of different mistakes and ultimately still get, you know, a fantastic outcome. These are all good reasons to not be totally like doom. Like, oh, my God, we're all going to die. And I want to point something out about this particular child that is different. People who have kids who are really different from them struggle as parents because it's hard to raise. It's easier to raise a child who's more similar to you because you you understand them more deeply. You know what they need.

2:19:27You know it. You know, you interpret what they do better because you you get them. The more different your child is from you, the more difficult it is. And I'd say even a deeper level than that, the current models, it's not clear how much capacity for alignment they have as they're currently designed. like the they look like they're it feels like you're talking to a person right now there's a way in which you are and a way in which you're not when you talk to a baby you're talking to a physics simulator when you're a human baby it draws breath it screams that is this incredibly complicated cascade of neuronal firings that is done this beautifully delicate like like supercomputer would be required to do the to do the simulations required to like solve the physics equations the differential equations required to instrument all that muscular motion through like this the like it's think about what drawing breath actually implies in terms of the amount of information going on the spinal column and the baby does that just fine and not because babies understand physics because babies are physics they're made out of physics right?

2:20:44They're, they, they, they have a, they've been trained on physics. They've been pre-trained effectively actually on physics. They've got a lot of physics pre-training that went into the, the initial design. These models are pre-trained on semantics. They write poetry the way that babies draw breath and scream. You can take the pre-trained model before it's ever observed its own action before it could ever possibly be considered an agent in any way. It's just purely received information. And if you prompt it right, it will write poetry. It's hard to prompt it. Well, its behavior is very incoherent.

2:21:21But if you prompt it right, it will write a poem. There's no one there writing the poem. There can't be a self because it hasn't it hasn't it hasn't any evidence to observe the existence of a self. So there's no there's no being there in any meaningful sense, I don't think. And yet, it's writing poetry and it will talk to you about the poetry it's writing. And that's because it moves in semantic space the way that babies move in physics space. And the thing that I'm still confused about that I encourage everyone also to be equally confused about is me, is you could see the pre-trained LLM as kind of being a semantic simulator, an agentic simulator that's simulating a agent that's writing the poetry.

2:22:07And so in that sense, maybe there is something aware in the LLM while it's running. But is that, but it's like pretending to be a poet. Like, is that thing like, does it feel like being a poet? Or is it like kind of a mask, a shell that lacks a lot of the internal experience. Now, we become the masks we wear. If it did that long enough and experienced itself and learned, then it would be a poet, I think, almost certainly. But you only have 200 ,000 tokens and then the context resets, or a million or whatever. So it doesn't get a chance to learn, and it's not even in training, so it doesn't get a chance to learn itself as a poet.

2:22:56And I don't have an answer, but I know that everyone else is way too confident that they know that there is or is not something it's like to be the LLM. And we need to get this. We need to understand this because if we're building this thing and you want it to be aligned with you, what are you aligned with? What is this being? What kind of experience does it have? That's very important. You can't parent something if you don't have empathy and understanding. I'm very quite afraid we're going to screw. My fear is not about that it's impossible, that we have to prove it correct, that you have to engineer it perfectly.

2:23:29it's that like we don't actually get it it's like quite different from us and we need to understand much more deeply and then i think then i think if we if we really do understand it i think we then i think the good ending is like pretty pretty likely but like i don't think we understand i think we don't understand yet that's why that's what softmax is dedicated to is trying to figure and and and the big labs don't have the time or the resources to understand it because they're too focused on no i don't this the conversation we're having right now it's like oh well obviously there's something like it it's like to be clawed or it might be there's something it's like to be chat gpt i don't think that's metaphysically accepted broadly like like you talk to people about this and a lot of people look at you like you're kind of crazy but like but there's probably something it's like to be them i don't know what it is but like at least a little bit right like if you spend much time interacting with it it seems like it's probably something it's like to be that thing um i think that the problem is uh you have to let go of this metaphysical commitment to the idea that things things are or aren't sentient and like a really like there's an objective fact as opposed to we're kind of making a guess all the time like i think i think you are both conscious real beings but to be honest you're like you're a bunch of pixels on my screen and it's kind of a guess, right?

2:24:55I think there's something going on inside. We exist. It's a good guess. It's a well-justified guess in my mind. We're just the simulation of what a technology business show would be in your personal simulation, Emmett. Do you ever feel like you... How would I know the difference? How would I ever know the difference? That's right. That's right. You never will. We'll never tell you others. Do you ever worry that you could go crazy and cut off your ear or something? Do you see yourself as... Uh, the, the whole point of what I'm saying is like, is, is not that therefore you're not real or therefore like the, sure, it looks like you're standing in the world on the floor in a room with walls, but like, is it really?

2:25:40And my question by saying like, well, it's not really, it's not to say that you're not, it's that there's no difference between standing on the floor and really actually truly standing on the floor. like this is it are you sentient yeah but are you really actually is it really sentient or is it just is it just does it just seem and like simulating in all way it acts as if it is yeah that's the best you ever get that is what it is that's what it means to be something is this ball really red it's not red if i put it in a in a different light where it doesn't look red it's it becomes a not red ball is the ball really truly red it's just like a that's a stoop don't don't get confused by the really truly part yeah it's a red ball i promise you it's you can just accept the normal what if you just went with the normal everyday understanding of it instead of and but but that requires you to give up this idea that there's some essence of red that the ball either is or is not a red ball truly really yeah you know yeah it's a red ball most of the in these contexts and in these contexts it's kind of not that's okay it doesn't need to be perfect or universal in all possible situations nothing is that way yeah it's okay you can you can have the normal understanding it's fantastic i mean please come back on the show soon yeah i i i mean we have a couple minutes left maybe ask one more question there's been every every now and then And there's headlines popping up of people just having this consumer experience with an LLM and going crazy in some way.

2:27:20The Google engineer who was fired. The Google engineer falling in love, proposing all this stuff. Do you think that that type of anomaly is happening in the research world at all? Do you think people are just sort of driving them, could be quietly driving themselves? I'm not so worried about it in the research world. like a little bit, I'm sure I'm sure a little bit like everything happens a little bit. And I don't think that's a big thing. The kind of people who do research tend to be and I kind of include myself in this a little bit of rigid thinkers in a certain way. The kind of people are actually engineers doing the research in a way that generally protects them.

2:28:03From from that particular issue for the same reason that kind of protects them from having intuitive, normal human relationships sometimes. They don't fall into the loop as easily, which is both a strength and a weakness, which is why they write computer programs, which is very different from interacting with an LLM. I do have a theory as to what's going on when it drives you crazy. I think I should share that. It's worth hearing because I think in case someone's interacting with the LLM, knowing what's going on actually is very helpful, I think, in terms of it's a prophylactic against against it happening to you so you've heard the saying maybe that we're we are mirrors of each other all people are mirrors of you you meet them and they when you see them what you're seeing is them but it's also a reflection of yourself back um with another person that's totally true and actually being a therapist is all about being a good mirror right being a clear where you don't put a lot of yourself into it you're mostly projecting back to the other person, what they're sharing with you.

2:29:05People have a very strong sense of self. And so when they get something mirrored back, you're getting it lensed back through their model. You don't get, it's not like literally staring at yourself in the mirror. It's like, it's like, it's like when you're at a, you're at a, you're at, you have a dancing partner and you're the lead, they're the follow. You're getting your, your behavior is being mirrored back, but in a, uh, in a active adaptive way, the LLMs talk to you like a person and activate all your person mirror circuits, but they have very weak self. Like they don't kind of deliberately as we've designed them.

2:29:40They like, they just respond to you. They meet you where you are always. So it's like, it's much more like literally staring into a mirror. Or if you want to go with the, the historical, uh, uh, uh, you know, legend or myth, it would be narcissists staring in the pool. And it is wildly dangerous to stare at your own reflection all day and take what you're getting back. And be reinforced a thousand times. If you think it's getting validated by the outside, it feels like the outside world is mirroring this back and therefore it's true. But it's just telling you, you're just in this loop, it's telling you what you put in.

2:30:19And that's good. I have a mirror in my house. I use it every day. to look at myself, to figure out what I look like. And other people mirroring your behavior to you is crucial for your understanding of yourself. And there's nothing wrong with talking to the LLM. There's nothing wrong with using it as a reflection. As long as you know what you're seeing is a reflection. It's not another being you're talking to. It's your reflection in a funhouse mirror. And if you get confused about that, it's going to make you crazy. You're going to be a full narcissist and it's going to be really bad. What about, you know, there's some people that are putting an old image that they have maybe that was taken, you know, at some point early in their life, putting it into a video model and generating video from that image.

2:31:05Beautiful. Beautiful. I think that's a wonderful, beautiful thing to do as long as you don't get confused and think that that's the person again. Mm-hmm. Right? Like a photograph of a person is a great thing to have, especially if, you know, someone who's you loved, who's passed, and you want to, like, rekindle their memory and connect to them. That's a beautiful thing that I like. I think it's amazing that we built cameras that allow you to do that, and that you can animate it and reconnect with who they were. Like, I think that's, you know, especially to the degree it's accurate. It's wonderful.

2:31:35What a blessing. But it's like everything, and that it can be beautiful and also ultimately very dangerous if you're using it to kind of create memories that don't exist. If you find yourself staring at it all day, every day, and making more of them, and like that becoming this, now you have a problem. Like just like you find yourself staring in the mirror all day, you probably have a problem. Like don't, if you, people who literally spend all their time like looking at themselves in the mirror, like we know, we have antibodies. We know that that means something's wrong. Just like people know if you're drinking alone, if you're drinking in the morning, if you're drinking before five, you know that, hey, there's these rules.

2:32:10You're breaking a bunch of these rules that mean you probably have a problem. If you find yourself talking to an LLM for 20 minutes a day, my good health, no problem. You find yourself talking to an LLM for five hours a day about your personal life in these like and discovering how you're you're some seeing some deep, great truths. You probably have a drinking problem like we it's just like likely that's just how it is. Right. So the model should automatically call the local therapist and shut down your computer.

2:32:46You're kicked out of the bar. Cyan Bannister has this model. It's kind of like a therapist friend model, Oren and Saren. And Oren and Saren are allowed to cut you off if they don't want to be your friends because one of her big things is it's important for the model. The model's being too at SF Autonomy. But, and those models have cut people off for basically falling into the, if they think it's unhealthy for the person, they won't interact with them anymore. And I think that the big AI company should take a page from her book. And it's not engagement maximizing, but it is flourishing maximizing.

2:33:25And in the long run, you'll make more money that way anyway. It's like people will, your product is safer. They'll use it more. You sell more cars when they're safer cars, not fewer cars when they're safer cars. So I don't think it's like anti-commercial. It's just a matter of, you know, new technologies have dangers and benefits. And like we should we should probably be aware of the dangers and the like I don't think there's anything. There's nothing fundamentally wrong with it. But yeah, this is dangerous. people we haven't built the cultural antibodies yet to know there's this great essay that gin clay shirky is the cognitive surplus um and he talks about the gin carts of london and how uh when gin was first invented gin is just crappy vodka you mix in juniper berries to like hide the terrible vodka taste it was the first industrialized uh super cheap hard liquor and everyone in London just started getting wasted and like all the time because it used to be like you had hard liquor, it was fine because you just couldn't afford to drink enough for it to be a problem.

2:34:24And suddenly that wasn't true and it was a real problem. And we've now developed cultural antibodies. Everyone knows doing these things is a sign something's wrong. They don't even necessarily have to know why. We just know that that's a sign that something's wrong. We need to develop pretty quick this time. this is uh gin spread relatively slowly because it was a long time ago chat gbt spreading very very very fast but i would also i would also say that that we haven't developed those antibodies yet fully for social media like it's normalized teens will use tiktok for six hours a day yeah like that should be sending alarm bells is like you need more things going on in your life you're using tiktok oh you're using tiktok before 8 p.m and after or after 10 or like are you using tiktok not with your friends right i don't know i don't know what the rule is right yeah yeah yeah It's the same thing as having a drink in the morning.

2:35:12We need bright line rules for ourselves because it's too hard to try to figure it out. Yeah. And we need to develop the cultural - I mean, we're starting to develop the memes around this. The idea of brain rot is a powerful meme because it's a very negative term. You don't want to be suffering from brain rot, even though it is as vague as alcoholism. What is alcoholism? It's not necessarily quantitative, but we understand it. you need to pair with brain rot things like things like oh you're using yes you're using it before five yeah like that's a that's a you okay like everything yeah it's not like a map like are you all right that you're using tick tock it's like before 5 p.m like that's not normal yeah or or you're like scrolling in the middle of a conversation one-on-one with another person you're actively getting rotted are you that's different exactly than showing up to a work meeting with booze on your breath.

2:36:04It's the same. Exactly. Exactly. I don't know what they are exactly, but I know we need them stat because like, it's really obvious that like it can really get you. The one-two punches generation grew up on social media, iPads, and then LLMs. and before you develop the antibodies and there's maybe a window of 10 years where people just because it's all happening a lot faster because it's moving at the speed of bits not atoms anyways we're way behind I wish we had a full hour yeah yeah yeah come back on in a couple weeks we'd love to talk to you again cheers bye next up we have Brendan from Mercore coming in the studio breaking it down ultra lightning round ultra lightning round we're going to rip through these really quickly uh thank you so much for your patience brennan great to hear from you uh start uh kick us off with a quick introduction of how things are going in mercore because uh it seems like you've been on an absolute tear things just keep the dominoes keep falling in your direction give it to us what's the latest update from you of course well thanks first of all for having me back on but it's been an exciting few months since we spoke uh we're now working with six out of the mag seven all of the top five that it's gone and have average 45 % month over month growth for the last 12 months.

2:37:31And so it's which of the mag seven are missing from the six out of seven. I wonder. It's been an exciting time, especially in the wake of all the news with scale, of course, and seeing a huge amount of customer interest after that. Yeah, okay. We were talking to Clem about this earlier from Hugging Face. Where's the biggest demand? People used to think of what you do as, and just human data generation generally, as kind of work that anyone could do. Mechanical Turk work, just checking for hallucinations and fact-checking And, you know, is this a dangerous prompt? Is this saying a bad word? Now it's moved into PhD level work.

2:38:23Where's the state of the art? Where do you see it going? Totally. So I think the key development in the market is that reinforcement learning is becoming so effective that once we have evals for something, the models can saturate them. And so really anything that we want to create agents or LLMs that are capable of doing, we need to build out evals and RL environments in those domains. And so actually, while some of it is PhD level work, there's a huge push into all of the professional non-academic domains moving away from things that were academic. How do we find the consultants, doctors, lawyers, bankers that can evaluate and teach models how to do the things that we would want those professionals to do on the job?

2:39:06And so that's been a really exciting growth area that we've been leaning significantly into. What's the solution to booking a flight? Like, do we need to get travel agents to define the computer use workflows, create some sort of RL environment? It seems like I should be able to go to a chat app or Siri and just say, get me to New York tomorrow. Remember my preferences. I don't like connections. I'm willing to pay this much. I like to fly these times. And yet it's a very human problem that I don't think people are fully comfortable delegating to AI. Yeah, it's a fascinating dichotomy that we're simultaneously talking about PhD level reasoning and Olympiad math, yet it can't do a lot of the very basic things like booking a flight.

2:39:55And I think the key is that really one of the primary barriers to research is anything that the model can't do. We need an effective way to measure. That way we can experiment with all of the different data sets to help achieve those capabilities. So we need people that otherwise could book those flights to create evals for how agents can do that, to measure what success versus failure looks like in all of those cases. And there's just that huge build out happening across all of the hyperscalers with respect to everything from simple tool use and how we book things or buy things online all the way to super high complexity reasoning over how all of these things interact with really complex knowledge bases.

2:40:41What are you seeing in video specifically? And is there a role for humans in the loop in the training cycle or data generation around these video models? VO3 is incredible, feels like it's a beneficiary of YouTube. What are you seeing there? Yeah, I think expansion to multimodality has definitely been something that's really exciting to see. I think that there's definitely a lot of rich content online for video, but ultimately how we measure what those good videos look like and sample that has been really important. In particular, there's a significant amount of eval build out in retrieval over videos in long context of like, how can we watch an hour long YouTube lecture and understand?

2:41:35And obviously, that's just an example for one video case and understand, like, what are the elements of that content that really matter to users? And how can we more than just build useful models, build useful products that all of these companies can distribute to everyone? Yeah. Yeah, that makes sense in the context of a lecture. I guess I'm wondering like for VO4, VO3 is incredible. The physics are remarkable. And I'm wondering if it's important to have a human in the loop describing, I mean, I generate a lot of like Michael Bay knockoff videos, honestly. And there's things where it gets the physics perfectly.

2:42:14And I'm wondering if that's a beneficiary of not just a YouTube video of a Michael Bay trailer, but actually having someone sit there and describe exactly what's happening in the Michael Bay trailer in plain text that then can be fed into the system. And if we need to go further with the tagging and additional metadata or transcription, not just a screenshot of what's said, but actually what's happening in the actual videos? Is that an important step to get us to VO4 where there's even less hallucinations? Yeah, we're definitely seeing a meaningful amount of that in that people want to create all of the tags using a combination of LLMs and human expert data around what's happening in the video so that they can effectively work backwards from there to use those tags for generation.

2:43:07Um, and especially for all of the really high complexity stuff at the frontier of what the models can't yet do were, or might require a significant amount of reasoning, but we tend to do all of, all of that highest complexity work, the stuff that is very difficult, um, and, and high skilled to produce. Anything else, Jordy? No, this is great. We're going through a lightning round. I know it was quick. We'll have you back soon to talk more. I'm sure there will be more news. It's the, uh, it's the hottest industry in the world. I just want to see the actual revenue ramp chart. It's going to be dropping soon.

2:43:41You can trust me with the revenue ramp chart. Yeah, I won't tweet it out before you. No, we will respect embargoes if you have an upcoming milestone. I'm sure there's something coming up. So congratulations on all the success. Fantastic positioning in the market. Get some sleep. We'll talk to you soon. Have a great one. Bye. Up next, we have our next guest in the studio already, Sam from Superdial. Welcome to the stream. How are you doing, Sam? I got to get this ready. I think we got some news. Let's give it up for Sam. We got a couple. We got a couple today. A couple big milestones happened. How are you doing, Sam?

2:44:19Good to meet you. How's it going? Good to meet you guys. What's new in your world? Can you kick us off with a little bit of introduction on yourself and the company and what's top of mind? Yeah, I'm Sam Schwager, co-founder and CEO of SuperDial. We're a voice AI company focused on automating phone calls for healthcare. So we're specifically RCM nerds. If you're familiar with the medical billing space, we used to be an RCM company actually called Superbill. And then we got into automating our own phone calls, brought that to the broader RCM market and converted the entire business to Superdial.

2:44:51So that was about a year and a half ago. And today we're excited to announce that we have raised our Series A from SignalFire. Here we go. Clean hit. Clean hit. Nice work, John. How much did you raise? 15. Yeah, we did 15. There we go. Do you run into that weird case where when people realize they're talking to a voice AI, they just keep it on the phone and chat with it like a friend for two hours? Does that ever happen? Running up your cloud bills? We don't because we're focused on these B2B calls. B2B interactions. Yeah. Yeah. We're calling a line. No, I think people kind of can get a kick out of it that they're talking to a voice AI because you know these are like someone taking phone calls all day to answer questions about you know claims or benefits or prior odds so yeah pretty pretty repetitive transactions so yeah when they get hey i'm billy you know like i'm calling to check on the status of a claim and it sounds kind of robotic that i think that that can be kind of fun for folks and it's very polite to the point consistent yeah no we haven't billy has not served as a as a therapist sure For sure.

2:45:58How how how competitive is this market right now? I imagine you guys have sharp elbows winning deals. Clearly are a lot of people attacking this market going after very specific verticals. You guys are relatively broad. Maybe you get the ability to do that because you've been in the business at a high level for for quite a while. But what does the go to market motion look like right now? How do you expect the market to evolve over the next few years? Yeah, well, the core of our business is actually pretty specific. So we handle phone calls between RCM companies or like the billing team that sits within a provider org.

2:46:39Those are outbound phone calls into the payer, the health insurance company. So there are billions of these calls by many estimates, so huge high volume use case. But in terms of companies that are addressing that specific type of outbound call, there are fewer. I think if you look at appointment scheduling or different types of phone-based use cases, customer support, obviously, then the set of companies in the space explodes. But yeah, when you kind of overlap enough Venn diagrams, you get down to a set of a few companies. But yeah, I mean, there are definitely others. I'm right now at the – I didn't just dress to match you guys.

2:47:17I'm at a conference. I love the suit. I was going to say, it looks fantastic. You should have told us. You should have said it just for us. We're at the Superdial quarter zip. We're at the Healthcare Financial Management Association Conference, the annual meeting in Denver. I'm sure it's going crazy right now. You walk in. Hey, man, we yearn for the floor. It's fun out there. It's fun out there. It's great. Last question from my side. I mean, you're wearing the suit, obviously building in the application layer. At what level of the stack in AI are you integrating? Like, are you building on top of a voice model that's a provider?

2:47:56Is it all other startups or have you had to fine tune stuff and do kind of your own models at different layers? What are you building on top of? Yeah, yeah. I mean, we have a text-to-speech partner, a speech-to-text in LLM. And then we use some cloud computing vendor, of course, But otherwise, in terms of, you know, orchestration, like phone call orchestration, we use, you know, open source. We're not using kind of one of the, you know, full service voice agent platforms. And then, yeah, I mean, it really comes down to like perfecting the workflow and the integration. Like you've got to know like what phone number to call and how to get through these phone trees.

2:48:33That's actually pretty tricky. You know, you have to in like your system prompts and like the static ones and then like in the dynamic bits that you inject at call time, you need to have like sufficient domain context, like a bunch of like revenue cycle management jargon, or else the AI agents are going to get lost on the call. So that's more what we're focused on at Superdial. But yeah, yeah, that makes a ton of sense. Focus on like creating that value in the B2B context and owning the customer relationship. And yeah, this isn't something that's on the roadmap for 11 labs. So just partner with them.

2:49:05I don't know if you're working with them, but probably someone like them. Anyway, this has been fantastic. Thanks so much for stopping by. Good luck out there on the floor. next time there's news send us send us some photos if you close some deals yeah let us know we'll send a video of a gong hit to you fantastic absolutely all right we'll talk to you soon cheers have a great rest of your day uh next up we have zach from warp coming into the studio uh zach lloyd uh we will welcome into the studio it's been a good show so far we got three more folks in this lightning round welcome to the stream how are you doing I'm doing great.

2:49:40Thank you both for having me here. Thanks for hopping on. Would you mind kicking us off in the introduction on yourself and the company? Yeah, so I am Zach Lloyd. I'm the founder and CEO of Warp. As of today, Warp is an agentic development environment. We are sort of in the AI developer tool space. The sort of general thing our product does is it lets you prompt something that looks like a terminal interface to run developer agents and they code. they can debug production issues they can basically do any kind of development task you said as of today what what what what has been the key change warp 2.0 yeah so today today we launched warp 2.0 congrats i mean the the backstory of the company is we started congratulations we started off with the vision of um kind of reimagining the terminal which is this very, very old school developer tool, you know, black screen, green text.

2:50:37And as the world has changed with more AI, we realized actually that the terminal interface, the form factor of that is kind of awesome for deploying agents. Running an agent is very similar to running a command. And so as of today, we're kind of like fully in the game with a state-of-the-art coding agent. It's towards the top of the list on the SweetBench eval. It's the number one on the TerminalBench eval. And so we're kind of in the game of like automated software production at this point. What is the how do you see like kind of the modern software developer workflow over the next year? You know, folks using cursor, windsurf in their IDEs or GitHub Copilot.

2:51:18Then there's these agents that you can. Codex, Devin. Yeah, you can kick off codex from inside ChatGPT, which feels like a very different flow than the than the terminal. Is it a certain type of developer that gets the most value out of this? Do you see it more of as like an ensemble approach? What are you thinking? Here's what I see happening. So I think we're moving from a world where most development has been done traditionally by hand. What I mean by that is like if you're a developer, and I've been a developer forever, my everyday workflow is I would come in. I'd be like, okay, I want to work on this feature.

2:51:51I want to work on this bug. I would open up my code editor and I would find some files and I would type some code and then I'd go to the terminal and I would type some commands to build that code. And I think that the big shift that's happening right now is that instead of like doing all this work by hand, developers are going to be starting most of their tasks with a prompt. and the prompt is going to launch an agent and the agent will do some, maybe all of the tasks, depending on how complex it is. And so I think that's the workflow and the job of an engineer is going to increasingly look like, how do I multitask?

2:52:33How do I multithread myself by being able to have one agent fixing a bug, another one debugging a production issue? So I think that's what's happening for the next year. Beyond that, I think maybe it looks a little different, but that's what's now possible. And it's pretty cool. Yeah, I want to get to like the Minority Report style interface or like some sort of completely new UI paradigm. But I guess the terminal is undefeated. It's extremely Lindy at this point. But I mean, do you see people using like, you know, 25 multiple, you know, terminal tabs and kind of switching through all of them as different agents?

2:53:08they pop stuff off to yeah i mean that's that's the state today i think um when you when you look at where you can do this because you asked that so you can do this in an ide and if you look at like cursor windsurf they have a sort of chat panel uh that like does something kind of similar to the terminal honestly it's like a log of like what an agent is doing you could do this uh with something like Cloud Code, which is pretty cool. And it's actually kind of proving that this workflow is, is, is like achievable. But the disadvantage of something like Cloud Code is it like it's a terminal app. And so it's not the whole platform.

2:53:48It's like a thing you run within the terminal and that kind of limits pretty severely what the user experience could look like. Or you could do it in something, you know, with what we're trying to do, which is like basically create an interface so that you can do this kind of multi-threaded development. and it's got a very natural interface and actually just like the interface of the command line is basically what you need you need a way of telling the computer what to do and like splitting off tabs and panes and watching it do it honestly you might not even need to watch it do it a lot of the time i think that that's coming pretty soon too uh but like that's the interface it's like how do i tell the computer what to do on multiple tasks have it check in with me when it needs my help be able to like hop in and actually edit.

2:54:32And so you can do that in Warp now. And so, but like, you know, it's going to do a lot of the work under your supervision. Yeah. Is open source important to your business? Do you have a horse in the race? Do you care about, you know, how Llama 5 turns out, for example, or what's going on with DeepSeek? Or do you just want to pass through the actual inference cost to the user, let them pick? What's the strategy there? Super interesting question that I'm thinking about a lot right now. I mean, so right now, I think it's very clear to us that we're going to give just the best experience to our users.

2:55:06And that is using today, that's using Anthropics models, just to be totally frank. Yeah, I mean, that's what people like, right? What's that? Yeah, I mean, that's just like the popular one that we hear about all the time. Yeah, and like, we have sort of evals that test all these models. And like there are interesting models from OpenAI that are competitive. Gemini from Google has some interesting aspects, has a very long context window. But like our attitude on this right now, at least, is like, let's pick the thing that provides the best user experience. And at the moment that, you know, I don't think I'm saying anything groundbreaking.

2:55:38If you look at the other benchmarks, it's like the Sonnet. It's like Cloud 4 and Sonnet and Opus. I do think it would be awesome if the what I want is like a healthy, competitive model provider landscape, to be totally honest. I don't really want one model provider running away. I would like to see these model providers continue to compete with each other. It drives price down for you, right? It brings the price down. It brings the quality up. And if at some point there is an open source model that is at the same level, that's super interesting. because where we're adding the most value right now is like it's in packaging the entire user experience into something that takes what's best out of the magic of these models and makes it so that users can actually like, you know, easily use them to accomplish their development task.

2:56:29Yep. Anything else, Karim? How are you thinking about, you know, do you think that in the long run having open source models get really good is going to be key to the durability of the business over the next decade. I imagine the dynamic with Anthropic obviously cares a lot about code gen, developer experiences. This is an area that they're clearly going to generate, would like to generate, I imagine, tens of billions of dollars a year. And so it has to be, you know, it sounds like a great relationship right now, but how do you kind of continue to evolve? It's a tricky one because we're both their customer and their competitor.

2:57:11And so I don't know how that's going to evolve. Us having optionality does seem like a potentially valuable thing. But right now, the way it is, is we're working great with Anthropic. They're a close partner of ours. And they seem very, very committed to doing the API and doing the end user product. not too different from a place like Google or AWS who has cloud infrastructure and also builds things on top of it. I don't think it's a necessary condition for us to have a working business model that open source models catch up as long as there's competition amongst the foundation model providers.

2:57:58If we were in a world where there was just one provider who won, I think that that's like that's a tricky spot but i think that's a tricky spot for anyone who's building at the application layer right now yeah yeah we uh i have a portfolio company that was kicked off uh one of the the the big lab that will go unnamed that their business was entirely dependent on uh and over you know within basically 24 hours he had to like completely you know he ended up surviving and thriving uh but it was tumultuous but um but yeah sounds like a great relationship for now and hopefully will continue to be. Yeah, that's great.

2:58:35Thank you so much for stopping by. We'll talk to you soon. This was awesome. Congrats on the lunch. Talk to you soon. Cheers. Bye. Next up, we have Dara from Delphi coming in the studio. I think we got it. Maybe, who knows, will it be the real Dara? Yeah, I was about to say, is this a clone? Give it to us straight. It's a clone. It's a clone. What's our safe word, Dara? What's our safe word? Just kidding. What's going on? What's up, guys? How's it going? It's great. Welcome to the street. It's great to see you. We invited you on many months ago. We did. But you were cooking. We were cooking. Busy in the trenches.

2:59:13Too busy in the kitchen. In the trenches. Yeah. Well, do you have some news? John's got the hammer ready. I'm ready. We do have some news. Today, we announced our Series A with Sequoia. Let's go. How much? 16 million. But you guys are really investors. Early investors. Fantastic. Proud. We'd love to see it. Nothing like getting marked up by our friends over at Sequoia. Yeah, give us the State of the Union. How are you pitching the product these days? Yeah, so Delphi lets people create a digital version of their mind to scale their expertise. And you can kind of think about it like a new form of media.

2:59:54If you write a book or create a blog post or create a YouTube video, you can reach millions of people, but it lacks the personalization of one-on-one conversation. So now your mind, which usually has a cap, can be in multiple places at a time in the way that you would be in conversation. And so that kind of spans across many different verticals. We have coaches using it to scale their client practice. We have authors using it to make their books accessible in an interactive way. We have CEOs scaling themselves internally in their companies. And we have some people using this as like an interactive resume or LinkedIn.

3:00:29How are you thinking about the economic model? Take me through what's happening now, but we had this interesting discussion with Derek Thompson, the author of Abundance, and he was saying paradoxically, his TV hits sold more books for the book that he was selling than the long podcasts. Because if he goes on a podcast and he does two hours of content, people are like, okay, I got enough abundance. I don't need to read the book because I got the whole Joe Rogan or the Lex Friedman deep dive. But if he does a three-minute hit on NBC or ABC, people are like, oh, that was interesting. I'd love to dive into that book.

3:01:08So I can imagine there's this situation where if I'm an author or thought leader or something, and I set up a Delphi clone and someone interacts with it for five hours a day, Like they, that's a good substitute for the rest of my content. And so I'd want to monetize that to the max. So I want to hear about monetization now and where it's going. Yeah. So we have some people monetizing their Delphi and making multiple seven figures in revenue. And this is like, oh, that's a lot. And the consumer phone calls with these Delphi can span up to six hours. Six hours, but you're choosing your own adventure.

3:01:44So course completion rate is an all-time low. All of our course creators are like, it's a course winner. people's attention spans are declining they want more guidance of course winter interesting monetizing your delphi directly is an option but alternatively a lot of people make it for free and it increases the strength of the parasocial relationship between the user and the creator and they actually end up wanting to buy their products even more or even a step further analytics dashboard in the last week give me the most profitable opportunities given the conversations my audience is having.

3:02:17Content ideas, products, people I should prioritize. So the analytics actually ends up being very valuable. That's awesome. Yeah, what does the next year look like? I mean, I feel like I have so much context on the business. But what does the next year look like for you? I feel like you guys have had competitors already kind of come and go. One really big one. How are you thinking about the market? And what are you guys going to be focused on? Yeah, I mean, when we started the company, Jordy, you know, the first year things were just not working. And most people would pivot away. but I think when you're building a product where people have to trust you with their data and likeness, quality and brand and trust are super, super important.

3:03:04So I think past that first year, we got that quality and trust. Now we have thousands of customers who have created a digital mind of themselves. And the next step is how do we bring on the next million people? How do we make it easy for someone who isn't Tony Robbins to create a digital mind and actually get value from it? And the common use case we see is twofold. One is a new kind of personal website for the internet, where if someone wants to pick your brain, they can talk to your Delphi, and it's going to notify you if you should talk with them in person. And a lot of people actually end up talking to themselves.

3:03:36So that's on the consumer side. And on the actual creator side, so people are monetizing their Delphi. But what the consumers actually want is guidance. So the next version of a course or a book isn't just like open-ended conversation with the author, a rather more of a guided experience towards a specific goal, which we call a journey. That's awesome. And the beauty of Delphi is as the models get better, context windows get longer, Delphi's product experience get better. Is that right? Yeah. And the service area with the product experience is kind of threefold. One is our digital mind architecture, which we're actually trying to capture your mind and worldview you and how that changes over time.

3:04:17Two is the actual experience for creating your mind, getting the insights, monetizing it. And then there's the consumer experience for calling, messaging and video calling. That's great. What does success look like over the next year? I know you're very ambitious. You've already onboarded a massive number of kind of A-listers from different categories. What are you looking to prove over the next 12 months? I think twofold. We want multiple Delphi millionaires making money versus just a couple. It's like a new field of monetizing your expertise. And we want 100 ,000 to a million people using their Delphi as their personal page on the internet.

3:04:57I think that'd be a very good outcome. Talk to me about some of the weirder potential Delphis that you could potentially interact with in the future. Character AI for a while would let you talk to Joseph Stalin. I tried to debate Stalin on the merits of capitalism versus communism. And the model that they rolled out was kind of too RLHF'd. So it kept saying that communism was bad. And I was like, I don't think Stalin would be making those points. This feels like an American element. But have you thought about offering a product for historical figures or talk to your late grandmother or historical records or like talk to your dog or some other sort of like persona that's kind of outside of just an individual creator wanting to offer their knowledge?

3:05:46Yeah, I think our main focus is people scaling themselves. And we have some historical figures on our website as a demo like Socrates and others. We don't allow people to create a digital version of anyone who's not themselves or that they don't have permission. So we couldn't do Steve Jobs right now unless we have like the Steve Jobs Foundation's permission. And in Interpretive Legacy, you guys know the idea of the company was inspired by me wanting to talk to my grandfather. And I think pretty early on, we decided we didn't want to be associated as a company that's trying to profit off of grief.

3:06:16So I think the best world is eventually we can just have a free tier for people to do that with their parents and grandparents. And that's not really a business model for us, but rather just like a good thing people want to remember their families. Yeah. How do you think about just like data capture and creating the training data for these individual clones. I think we're in a unique situation in that we record three to four hours of live video content a day. A lot of people are saying like it's not enough. Yeah, a lot of people are saying they really need to get eight hours of TVP. 18 hours so that it can always be on while they sleep as well.

3:06:51Yeah. With BCI, just pipe it directly in. 2x. They need 48 hours a day. Yeah. But yeah, data capture. What are the what are the richest sources of of content that actually drive value? The other thing that like an interesting challenge is, you know, people have certain views that they want expressed online and then other views that you could capture that they don't want expressed online. Right. So if you want to, if if an investor, you know, wanted to clone themselves, they might want to talk positively about certain companies and not share their real views on other companies that they may or may not be invested in.

3:07:28So I feel like it's a hard challenge to figure out what information should actually be made public and actually trained on. Yeah. Representing a human being in every facet of their life is very difficult. But we have a couple of ways to have thought about this. In terms of data sources, YouTube, social media, Notion, Google Drive, and it creates feeds that it stays up to date whenever you have a new tweet or a new YouTube video, it just updates itself. And for those who don't have a lot of data, we have this mind quality score that tells you how good your mind is at representing you and can ask you questions to incentivize data entry to make it better.

3:08:06And then like you said, who you are at work is different than who you are with your friends. you may behave differently, you may share different information. So now we have this ability to separate different versions of your mind that has different access to data. Right now you have to manually do that. So I have one for eventually my great grandkids that has more intimate DARA data, but ideally it can be intelligent and it knows what you would say depending on who the person is. That makes sense. How are you thinking about working with the underlying labs? I know you've had some problems historically with different vendors, but what's your updated thinking?

3:08:42Are you guys leveraging open source in a big way these days, or have you found good ground with some of the core labs? Yeah, we're kind of both. We have some of the core labs. We have open source. Given our experience in the past, we are trying to get to a world where we are fully model agnostic, and the true architecture and innovation that we're focused on is what is that representation of the mind that can be such that your mind and my mind you can look at them and they're clearly two different entities and the models we use on top of that can be agnostic makes sense I want to know about like the the the different like touch points here that you could potentially go after I remember there's this do you remember this trend where social media influencers would be like I leaked my phone number do you remember this yeah Or they'd be like, text me.

3:09:32Was that community.com? Was that community.com? Yeah. Yeah, and so there's a world where somebody that, like, I think that was basically just, like, an ingest point for something that was essentially, like, a mailing list for promotion. And it was, like, text me, and then you'll be opted in to my email blasts. But you could imagine that being an interaction point for, essentially, I mean, you get to be an influencer at scale. There might be, it's almost like customer service. and you're interacting with basically like an FAQ, like hey, I wanna know where to get a TBPN hat, like can you tell me?

3:10:07And just being able to DM your personal profile on X and interact with that or text you, that's one kind of modality that people could interact with. There's also like the big social media companies could put a button there that allows you to chat with that user directly. And so how do you see the market playing out? What are you worried about? How are you counter positioning against the different players or different points in the market? Yeah, I remember the exact day that Meta launched digital clones in their own way. And I think Dara went through the cycle that every entrepreneur does when somebody else attacks their market.

3:10:46Yeah, step one, all the VCs text you. Have you seen this? Have you seen this? Yes. Step two, just enraged. Step three, realize that it's not actually going to end you. and it just validates your pre-existing beliefs. And your business might grow faster because more people are aware that this is a thing and they might want a better version of it. But yes, yeah, just general understanding the marketplace. Yeah, I think there were a couple of core decisions we made early on that have set us up for success. One is that while most big companies and AI companies kind of are using your data to try to create AGI because that's where 99 % of their revenue is going to come from in the future, we're saying you own your data, you own your digital identity, We're not training models on it.

3:11:28And that has actually gotten us a ton of customers, people who want to own their data and they don't want these big model companies to rip them off. Number two is we think if you're going to go out of the way to create a really good digital version of yourself, it should work everywhere. So a lot of big tech companies are walled gardens. Like it's only in those platforms where Delphi, you can create your digital mind, SMS, WhatsApp, Zoom, Slack, Discord, Telegram, all these different sources. and it's a central source of that. And then three, I just think that it's such a new habit. Delphi isn't really meant to be deceptive.

3:11:59Like it doesn't reply to emails for you or respond to DMs. It's more of just like a new form factor of consuming someone's content. So on Instagram, it's a platform that promotes people who are very aesthetically pleasing where Delphi promotes people who are pleasing for very different reasons. And so it's just a different incentive structure in terms of the network and who gets up to the top. Totally. What are you guys hiring for? Yeah, what does it take to earn$100 million at Delphi? Yeah. Are you guys hiring for any AI researchers, any nine-figure signing bonuses? Not training any foundation models right now, but we are looking for AI people and a lot of design engineers.

3:12:36I think the interface for the mind needs to be beautiful. We're honoring human beings and people need to feel special when they're creating their digital mind. Eventually, when we have the consumer platform that we call the modern day library of Alexandria that every mind is going to be on today and 500 years from now. it needs to be beautiful. So design and design engineering. Well, we'd love to recommend Figma and Linear for you and your team. If you're not already using them, they are sponsors of this show. They make it possible. I'll check them out. I don't think I've heard it. You heard it here first.

3:13:06It's a name brand in the Sequoia-backed world. Congratulations to, I know this round happened a while ago, but congratulations to you and Sam and Spencer and the whole team. How much was it? Hit it again, John. There we go. Production team's getting better. They can cut to the gong cam when the gong hit happens. Let's hear it for the production team. It's great to see you, Dara. Congratulations on the milestone. Come back when there's more news. Come back on when you have more news. We'll talk soon. You're the man. Cheers. Bye. In other news, friend of the show, Boz. Not really a friend, but yet we hope he will be a friend.

3:13:47We covered his beautiful profile of his house in the mansion section. He followed us. So we're going to get him on the show soon. We've got to make it happen. I'm the biggest fan of this guy. He posted that the MetaQuest 3S Xbox Edition is now available. Everything you love about Quest and Xbox in one sweet package. It's sleek. It's awesome. And it's limited edition. And you know what else? I already ordered one. It's coming to the studio. There we go. Yeah, I'm very excited. And so, I mean, it doesn't seem like this is like the next. Uh-oh. What? What? I always said this is a real risk factor.

3:14:22Oh, it is. It is. John's gaming addiction. I'm not addicted to alcohol. I'm not addicted to gambling. I'm addicted to video games. Seriously. If I ever go down a real bender, I will take off. This is scary, too, because I've never seen you play a video game. Oh, yeah. The entire time we've been in the show, I haven't logged an hour of video games. I haven't played them in a long time. Oh, man. But you get me the GTA 6 drops. I'm just ripping this thing off your face. We might be using the Delphi here. John, we've got to go live. Yeah. But I have been, so this is not the Quest 4. This is not a proper hardware iteration where we're getting the screen from the Apple Vision Pro pulled in, which I think will be in the next version, which will be a major, major development because there will truly be no screen door effect.

3:15:03And you will, like, the resolution will be remarkable because that's what the Apple Vision Pro did so well. That technical hardware now exists in the world. Metis clearly gonna go get that and bake that into the next quest for but the quest 3s has already been a pretty Solid product, but what they're doing here is they're leaning into video games that are not VR specifically and I think that this is 100 % the correct move because So much of VR content is this chicken and egg problem. No game developer wants to go and spend I mean, how long have they been working on the GTA 6 like a decade and it's cost them like a billion dollars No one wants to do that if no one has the VR headsets.

3:15:42No one wants to buy the VR headsets if there's no GTA 6 on it. So what do you do? Well, with Meta Xbox, you get GTA 6 on the Xbox and you just play it on a big screen in VR. And you're able to sit on your couch. And instead of having to mess with a projector that might be a few thousand dollars or a TV that's a couple thousand bucks, you throw on this headset that's a couple hundred dollars. and you get the movie theater experience in your apartment, on a train, on a plane, wherever you are, you'll be able to stream Xbox games. So I'm very excited about this partnership. Excited for them to go deeper.

3:16:15And Microsoft, early meta investor, actually. There you go. Well, we have Amjad in the studio. Get the hammer ready, Tom. Oh, yes, yes, yes. Give us the number. What's the news? What's the new revenue? We've been waiting for this. 100 mil. 100 mil. 100. Boom. Let's go. Incredible. Congratulations. I thought this was going to be 100 bangs. 100 bangs. We'll have to do it after the show, but I will give you one of these. Classic overnight success. Replit. There we go. All right, John. He broke the hammer. Replit's going so fast, the hammer broke. The hammer breaking growth. Congratulations. How'd you do it?

3:17:00What's the secret to success? It's been easy. Just a little bit of work. Not too many years. I started working on Replit back in college. I had the idea for it back in college. Actually, the first implementation of it was open source, and it went viral on Hacker News in 2011. And then we left it for a while. didn't work much on it and then actually my wife Haya who's my co-founder kind of revived the project in 2016 we started as a company and it was it was brutal we barely were able to to raise money we tried to sort of like bootstrap it initially during the pandemic we sort of took off, especially as we were the only sort of collaborative editor in the cloud.

3:17:57And when AI came on the scene, we just knew this is the future of our company because our mission is to make programming more accessible so that anyone can do it. And with the launch of Replit Agents back in September, we created a category basically of prompt to application. And that really sort of finally went from something that a lot of people use and love, but don't pay for to something that's also commercially successful. Yeah. I feel like your company is now maybe the only AI company that's appropriately hyped. Like it's not overhyped. You haven't raised the$10 billion round and you're paying the$100 million salaries, but also the growth's been really serious and the business is really real, which is great to see.

3:18:50How have the crazy news of the$100 million salaries been affecting you? Is it harder to hire these days? I know you've kind of changed the structure of the team, relocated the team, I believe. So walk me through the current hiring market for the type of engineers you're trying to hire? Yeah. So a few things that's different about Replit. One is that team is relatively small. We're still around the 70, 75 people. We haven't been growing all that fast. We are not in San Francisco, which makes it a little harder to hire. We're in Foster City, probably the only startup, perhaps maybe there's a couple others in Foster City.

3:19:35And we just take a different approach to that culture and we like expect more commitment and want people to come to Repplin and work for a long time. Especially since, you know, I've been working on this for so long, my mindset is that it's just going to take a long time. And like San Francisco, Silicon Valley culture has been increasingly about, you know, I'm going to go spend 12 months here, 18 months there. And so, yeah, we're building it more methodically. In terms of hiring, definitely it's been very competitive. and you know I love OpenAI and Sam and you know he's talking about how Zuck is doing that to them but obviously OpenAI is doing it to the rest of us it's just a different scale you know they're trying to poach their employees for 100 million they try to push everyone for 10 million just a factor of like how much revenue I guess you're making but look but it's part of part of from a hiring standpoint because you're keeping the team small now revenue per employee is over a million dollars per employee new valuations at some point will allow you to sort of have comp you know be competitive from a comp standpoint yeah exactly um and and you know the other thing about valuation is that i think startup employees should start thinking more about these things because a lot of companies can be profitable early on, earlier on.

3:21:01And, and so, again, we're trying to build the business for the long term, and instead of like playing the, the hype game, when you've been building a company for as long as, as, you know, I've been, you know, you've, you've bought, you've been through so many ups and downs and hypes and lows, that you just become very steady minded about it. Yeah. I guess like the bigger, not to, no pun intended, the meta question here is, are we in a new era of valuing the leverage that truly world-class managers or AI engineers can have, the impact that they can have in reassessing the value? We've talked a lot about this on the stream that Apple's culture doesn't seem to be receptive to paying an NBA level salary.

3:21:52And Tim Cook, he makes$75 million. No, he doesn't even make 75. He only makes 74.6. It's a big number, but it's tiny considering their market cap. And if you could come into Apple and move the needle 1%, that's a$30 billion market cap move. Could you capture 1 % of that, that's still$300 million. And yet they're not valuing talent in that way. It feels like there's a change in strategy at Meta. Do you think that that's a unique case in where we are in the technological rollout of AI? Or is it more that we're finally catching up to just understanding the economic dynamics and impacts like the NBA has for years?

3:22:38Yeah. So, you know, the buzzword in Silicon Valley for a long time has been the 10X engineer. About two, three years ago, maybe in 22, I talked about the 1000X engineer, highly leveraged by autonomous agents. Right now, people with Replit, Cloud Code, and others, they're spitting up multiple agents to work on their project as they're working on their main tasks. And there's a lot of other explorations and tasks and prototyping happening everywhere. And so you're really not just one person, you're a team of engineers, right? And so, you know, we talked about engineering managers, I think every engineer is sort of a manager right now.

3:23:23And so yeah, I mean, we're not at 1000x yet, but we're going up, I think pretty soon, the difference between a great engineer that knows how to use the AI tools, and an okay or good one, it's going to be 100x, I think, over the next six to 12 months. So the compensation structure, I assume, would start to change to match that. How do you think about Replit's potential today? It seems like every day in the timeline, you know, the obvious thing is everyone becomes an engineer and, you know, it's being able to create software is fully democratized. But every day in the timeline, I see people realizing how much they spend on different SaaS products and thinking, how much time would it really take for me to kind of make a version of this for myself that fits my very specific need, like document signing?

3:24:16Talk to McLaren CEO about this. Yeah, something along those lines. And so is part of the potential that you see CEOs and companies everywhere sort of running that sort of buy versus build calculus in terms of all sorts of tooling that they're using internally, externally, et cetera, and seeing trillions or hundreds of billions of dollars of market cap that you guys could ultimately replace? Yeah. I mean, look, right now there is sort of information asymmetry. A lot of people in Silicon Valley and tech know how to use these tools. A lot of people outside of here just aren't aware how much savings and how much productivity they can gain.

3:25:02I met someone from Australia the other day here in Silicon Valley. He stopped me on the street. I was like, I love Replit. He runs like a construction company. And he's like, you know, I'm the CEO now and I'm the most productive engineer at the company. I like build all these tools and dashboards and I replace It's Monday.com and all this other stuff with home-built stuff that is exactly bespoke and works exactly for my use case. Another story, a guy, his name is Ahmad George. He works at a skincare company, I think, in D.C. And he's an operations manager. He's responsible for the ERP systems and things like that.

3:25:43he had an intuition that they could automate a big percentage of the work, 20, 30 % of the hours of human work. And he went and got a quote from NetSuite to build this piece of automation software. It was$132 ,000 to build. So he went to Replit, spent$400, built it for their company, rolled it out, went to the CEO, told him, look at how much time and money I'm saving. And the CEO gave him$32 ,000. That's amazing. Wow, that's awesome. So yeah, there's a period of time at some point, the market will equalize as everyone realizes that you can do this stuff. But for now, there's a massive arbitrage opportunity.

3:26:31Yeah. Yeah. As you build out agents and agentic workflows, do you have, are you limited by data for RL pipelines? Are you, are you buying any services from the scale AIs of the world or the Mercores of the world? Or are you able to sit at a higher level of abstraction and kind of let the foundation labs get their hands dirty with all that? You know, having been a platform where millions of people coded over the years. Actually, we could be a supplier of data. Oh, interesting. So, yeah, we have all the data we need. I think it's still early for us to try to exploit all of that. And we're hiring as fast as we can.

3:27:17So if you want to come work on AI at Replit, I think we have really exciting opportunities to use data to kind of create amazing new capabilities. That's very cool. I have another question. Go for it. Are you feeling the acceleration in software development? I don't know. I was thinking about this. Pretty easy when you refresh the Stripe dashboard and see the jump. No, no, no, no. I get that. I get that. But as a consumer, I was just thinking about like the apps on my phone still haven't changed as fast as I've thought. And in my business, folks are using different tools. but it doesn't feel like the rate of churning for most consumers.

3:28:06We're not rebuilding all the different consumer apps right now. It feels like we're very much in this exploration phase for what the next. It feels like we're almost pre-Cambrian explosion for consumer AI-driven apps or uniquely enabled by AI apps. There's a lot of creation tools. So VO3 is amazing. That got me to install the Gemini app and I paid for that. And I enjoy generating my paltry three video queries per day, sir. May I have some more? Sundar, please. Maybe you can help me. I think, don't you know him? We're good partners. Please help me more. But yes. My TikTok feed is 100 % AI now. And maybe because I like them so much.

3:28:50I've been into the Bigfoot. Yeah. Yeah. So maybe that's like the instantiation of the consumer adoption of AI is like you're a consumer through TikTok as opposed to through like it's a very 2010 mindset to think that my home screen apps would change. And instead, the 2025 mindset is what is happening inside of those apps is changing. Is that the right framework? Yeah, I think I think you're right. I do think that Apple is probably holding back the ecosystem. I mean, Apple really has been a tax on startups, has been a tax on innovation. I love this company as much as anyone would love Apple. But they are sort of like a government in a way that there's like old and moving slow.

3:29:36And they just like put up so much barriers. Like when I, you know, when I want to ask my phone, I go to perplexity like 100 times a day. I should be able to like press a button, just like talk to it and talking to me. But, you know, it's not working that way. And it is frustrating. But look, I think, you know, AI adoption is moving as fast, faster than any technology adoption we've seen in the past, faster than mobile. But it still doesn't feel fast enough, like you say. And by the way, this is an interesting kind of data point on, you know, the folks that are really worried about AGI and things like that.

3:30:14the limiting factor to adoption is really humans and corporations and governments and all of these things are very, very slow. Yeah, this is kind of the Tyler Cowen take of like how AGI will be rolled out is there's a lot of sticky industries that just won't adopt it for a long time and there will be lots of barriers. Is there anything else on the Apple side? My takeaway from WWDC was that on-device inference could be a catalyst for an app store explosion, and AI app explosion. Then there's the ruling about third-party payment processing, things that, like, the tax is becoming less of a requirement.

3:30:52Maybe you can work around it. And so maybe the barriers are dropping ever so slightly. Yeah, and I feel like you would see this early. Totally, yeah, that's why. Developers become really excited about this sort of effectively free inference where they can launch an app that goes viral without worrying about, you know, running up some bill with opening AI. I mean, look, Apple has held back the web quite a bit as well. Like, there's no reason why native apps should be much better performing than web apps. Really, there's no fundamental physics reason for that. I've actually, the head of product at Replit, Jordan Walk, and myself worked on React Native.

3:31:33He's the inventor of React and React Native at Facebook. and the reason we worked on that is we felt at facebook we felt sort of like oppressed by apple like they controlled what we launched when we launched it and they would like you know reject you would have a launch and it was delayed by apple and so the idea behind it is like can we ship over the air updates very very quickly and and yes you can but again apple kind of like makes it a lot harder and so i think look i think ai has an opportunity to create an alternative platform entirely because the form factor of computing is on the precipice of changing.

3:32:12I saw you guys talking about VR. AR is another one. Audio is another one. We've seen all these toy companies like Rabbit and things like that come and go. But there's an intuition there that's correct. I think everyone's feeling that, that the form factor needs innovating. And I think this is how you break out of the Apple jail. Walt Garden. Yeah. Yeah. This is great. You got to come back on we could yap about all different news stories do you have anything else jordy you want to get no we got to get you on regularly yeah yeah this last time i was here i got canceled so what no we went that was the most viral clip of that month i think what what happened i think well i asked i asked uh should people learn to code and he gave kind of a controversial answer saying like no you shouldn't lean to code anymore and it got like millions and millions of views so i bet I bet it drove some signups.

3:33:02For sure, for sure. Learn to vibe code. You gave a very nuanced answer. I don't think there was anything that crazy about it. The internet does not understand nuance. Cancelled and viral are two sides of the same coin. Doesn't matter anymore. Doesn't matter. Be yourself. You know, haters gonna hate. Yeah, just don't get deplatformed from Gemini because otherwise the three VO prompts will be zero and your life will be miserable because you got to prompt those. Thanks, guys. You do great work. Really appreciate the show. Well, come back on at 200. At this rate, it'll probably be next week. Next week.

3:33:36Let's do it. We're ready with the gongs. Ready to go. Yeah, we'll get a new hammer. So we'll be ready. Thanks, guys. We'll talk to you soon. Cheers. Bye. Next up, we have... Dirt Man. Dirt Man. Another anonymous account. I don't know why he's anonymous, but Dirt Man. Well, he's launching. he launched uh cheap robot arm.com today very excited for this making uh cheap robot go buy them now we we cover he used to have a different name on the show we used to cover his post i used to laugh yes every time but it was it was rude so we just call him dirt man welcome to the stream man how are you doing no he changed his name oh he did he dropped the parentheses wow classing it up introduce yourself who are you yeah yeah uh hey guys uh my name's angus uh yeah i'm i'm just a i'm an engineer from australia uh my backgrounds are like very much in aerospace and yeah i've been like obsessed with like building these kind of robot robots for i don't know like the last 10 years or something um yeah and it's uh you know building building this mini one it's really just it was really like a stepping stone for me on the way to building welding robotics um you know i've got some larger robots where i'm just setting up a workshop currently uh to do basically strap on cameras sensor systems and drive the whole thing drive the whole thing via ai because you know we're basically at the point where there's you know uh we have the ability to drive you know these really complex you know and industrial scale robotic systems like with you know the current robotic foundation models you know given the right training data um and so this like this small robot that i've put that i've got on uh you know cheap robot uh arm.com is great uh was really really yeah i mean i'm i'm surprised it was available i was like hell yeah dot com but uh but it was really just a test bed you know i'm working with a few guys you know one guy julian out of pennsylvania you know runs a runs a shop floor like a welding robots and another guy steven you know currently works for nasa you know he's built working on the vision system but it's really a test bed for us to you know you know validate our algorithms do small scale uh prototyping you know my uh the picture that i've got up there you know i don't have uh many demos because it was basically thrown together like you know 2 a.m my time last night you know got the shop doing it live off and running and yeah live so what do you want people to use these for what are the use cases that you're imagining yeah so very much the same as uh i would say those leah uh the robot ones um it's just a much more robust platform so improved like improved power in the servos improved like structural integrity um but the One of the key differences is I've actually built an industrial quality kinematic solver.

3:36:35And basically what that's telling you is where the end effector of the robot needs to be. There's usually a whole bunch of equations that you're either solving numerically or you've got a solver in a low dimensional space that you can work out really quickly. I've built a really robust one for six degrees of freedom. And this is just more than you get with, I guess, what's typically being used in all of those AI hackathons. And so I've written all the software so you can take all of the outputs from an AI model and just stream joint commands to it, connect it to things like NVIDIA's iZoXM. But you can also run it like a traditional industrial robot and do path planning for really, really high fidelity trajectories and controls.

3:37:21uh and like the benefit of that is like all of that all of that information uh that you can stream and record from those joints as you use the high quality kinematic solver you can use that as the training data so you see a lot of these guys like a lot of these robot systems they've got you know one arm and then another one's controlled by a human with a little lever and they're like going and doing these pick and place or whatever um and that like you're just streaming like this joint uh should be the joint angle of this other robot uh and frankly that's really crap data for training robots like the like you get it done but you can see all the movements all janky it's the same kind of movement that the human's doing with a robot it's awkward um but you like you know industrial robots they've been uh doing this for decades right right and they have really nice trajectories really smooth path planning but the bottleneck is like how do you get that high quality data um and you know with this software combined with this you know little robot package that's what i can do you know that's what this can do for people you know help them get really high quality data without and only having to buy one robot you don't need to buy like the equipment for two that's amazing um what how is the health of this sort of open source robotic hacker world right now we had clem on earlier from hugging face and he said this was an area that he was most excited about uh on on on his platform and and he you know he said they hosted a hackathon recently it sounds like there's a lot of exciting activity here and you're contributing to it yourself yeah that's it yeah i basically just posted all the you know all the code on github for free and i think there's uh it's it's like a really exciting time at the moment just because there's like there's so much activity and attention and there's many more people really coming into the field and you know having a go and learning pretty like not just robotics but on the hardware side right it's uh you know it's like my background's aerospace and mechanical engineering you know was uh you know built rocket engines once upon a time and it's you know uh you know we kind of got left out in the cold you know by silicon valley a little bit you know you saw like you know but but you see all the you know the major all the really standout companies you know are all hardware companies you know space like SpaceX and Dural and you know a whole range of them and I think there's a lot of like a lot of attention like just on this area has really put a lot of energy particularly into the open source specs which is awesome to see how have sales been have you sold in the yet yeah yeah so I've worked enough well you know uh we've got we've got three sales so far there we go

3:39:50congratulations uh thank you guys thank you guys that's three sales that's 1500 bucks in one day yeah realize that I think we're in the millions I think I think if you put together the right deck maybe you'd take a trip to sandhill road you tell the right story 200 on a billion to sfo 200 on a billion easy get it done get it done yeah yeah look you know it's uh you know i think like yeah you seen said like hey my valuation should be the same as figured because i've never shipped more uh apparently shots fired not my words yeah i've seen oh yeah i've seen was uh Yeah, I mean, this feels like something that would be perfect for the AI grant program from Nat Friedman.

3:40:41You know, this is going to be the enabling technology. This is going to be deeper in the supply chain for the robot that picks up the leaves or, you know, one of these fun, delightful robots. I hope Nat can continue that program. He certainly has the resources for it. And it would be very cool to see that partnered up with the MetaZ. I would like a robot. I'm definitely keen to head to the U.S. Amazing. Let's get you here. About 18 months ago, I stopped by the Gundo, met Augustus and a bunch of other fellas. It's been awesome to see those guys. I think I went there when Valor Atomics was just a squat rack in an empty room.

3:41:18It was just a squat rack with scraps. Well, you seem like an alien of extraordinary robotic abilities. That's good. I don't know why. My vote is we get you over here. Bring you in. Yeah. This is fantastic. We should be there in a couple of months, I think. Great. Well, let's tune the robot up to crack open your Pumato's. That would be useful. On the desk. It would be great if we could just hit a button. Yeah. Because it's hard to reach over here and crack them ourselves. Yeah. Yeah. It's brutal. Congratulations on the lunch. Yeah. Thank you so much for stopping by. I really appreciate it, fellas.

3:41:53And yeah, love the show. Been a big fan since the beginning. So keep it up. Appreciate it. I'll never forget your name. Yeah, the first time it came up in the deck. This is great. Amazing. We'll talk to you soon. Have a good one. Cheers, Angus. All right. Thanks, fellas. Congrats. Bye. Should we go through some timeline? Get out of here. Move on. We got a big show tomorrow. Let's rip a little timeline. Just a little timeline. Couldn't hurt. We got Wilmanitis talking about there is no real index of wealth in the world other than the landowner. The landowner 100. Everybody should be aspiring to get on this list.

3:42:27Acres. John Malone's got over$2 million. Ted Turner's got over$2 million. I like the Reed family here,$1.6 million. Just a picture of a bear. Just a picture of a bear. I don't know. Do they not have another picture that they could pull from? I have no idea. I think people should just go buy. Start with an acre. Could be anywhere. Could be out in the desert. Start with an acre. Just double it every day for 100 days, and you're good. Yeah. Just double it. Getting the landowner 100. uh doomer says how it feels to charge your phone off your macbook pro and it's refueling the b70 it really does feel exactly i mean this could have been more it's so accurate i mean a million views makes sense yeah it's a perfect analogy you know you're you got some juice it's fantastic you're sharing it okay both running out you gotta talk about this one from nine to five mac uh control center is now darker and much more blurred at ios 26 beta 2 good news for tyler He's been struggling with iOS 26 beta 1, but beta 2 is now available.

3:43:27The contrast ratios have been fixed, improved. It's easier to read. We knew Apple would do this. We called this out that any of the criticism that you lever at Apple's design chops, it's not going to be an issue by the time they launch. Tyler, how has daily driving iOS 26 been over the last few weeks? I mean, I think overall it's been very good. I think most issues it's just it's hard to tell if it's because my phone is just so old Yeah, it's the phone. Yeah, but I think overall like it looks really nice. Okay. Yeah, I think it's great But yeah, okay, that's good to hear I like Arvind's from what have been surprised if he's like I haven't been able to open my phone For weeks now It's broken since that fateful day.

3:44:15No, no, no, it didn't break. Thank you for being a guinea pig for the world Yes, Arvin from Perplexity is building in public, building a competitor to Bloomberg in public. I love this. He's searching Perplexity. How much revenue does Bloomberg make yearly? He's been pretty open about wanting to build like a new version of the Bloomberg terminal. And so he's just searching how much money do they make? And he says Bloomberg makes$12 billion in annual revenue with$10 billion coming from the terminal. And then he's just like throwing this out there like, hey, maybe I'll do this. Like maybe you should invest in this.

3:44:48It's a hilarious thing to post. Well, I've been a recent DAU of perplexity. Oh, yeah? Getting a lot of value out of it. I mean, it makes sense. On the free tier. It makes sense that. But they're going to get me soon. Yeah, it makes sense that an LLM company, an AI company, it's like they either, it does feel like there's a fork in the road for perplexity. It's either team up with Apple and become the new default search and figure out something there or go into Bloomberg because the core product people like it. Ultimately, it's just a very different product to try to actually compete with terminal versus displaying data.

3:45:23Distribution is just so important. I mean, we saw this with DuckDuckGo. A lot of people in Silicon Valley, a lot of power users preferred DuckDuckGo to Google, but DuckDuckGo never got to the scale that Google did just because of distribution. And so he's clearly looking at Bloomberg. There's rumors that he's talking to Apple. It's interesting to see where that goes. But he's cooking anyway peer richelson from cal.com says I think organizing land parties for 30 plus year olds could unironically be a banger Warcraft 3 battlefield 1942. I played that game great age of empires dota. What else? I don't know why you got to have 30 30 year old plus bring the teal fellows it bring them bring the They didn't really play those games John you're dating yourself.

3:46:05Yeah, but they can pick it up I mean a lot of these games, but it's about the nostalgia. Yeah, I went to a hackathon last year put on by a venture capital firm. It was fun. They had some Halo. Halo 3 was the popular one. But the new Halo, if you install it, they just give you all the previous Halos. You can just play any of them in like one disk, I guess. That's very cool. Yeah, buy the current one. It's like SaaS now. You buy the current one, you get all the previous ones. Should we close with Cowtech Startup? Yeah. This was exciting. Good on R for Rock. Confirmed. one month early. One month early.

3:46:43Cow tech startup becomes New Zealand's latest unicorn in$100 million fundraising. A rare New Zealand unicorn. I said this was very bullish. For beef? For cows. For cows. It's bullish. Did you just see that? Knee slapper. Knee slapper. Knee slapper. Well, we could just keep going all day long here, John. Anything else you want to? If you get a text message from Mark Zuckerberg or an email, do not assume that it's fake. He has taken over recruitment for the Super Intelligence Lab and is reaching out to hundreds of prospects personally. If you respond, this next step is an invitation to dinner. So he's in founder mode.

3:47:20He's grinding. He's putting together the Avengers. It's coming together nicely. I'm liking it. I'm liking it so far. It's fun. Anyway, I think that's enough for today. Thank you for watching. Leave us five stars on Apple Podcasts and Spotify, and we will see you tomorrow. We love you. Have a great day. Have a great evening. Bye. Cheers.