Loading summary
Transcript940 lines
- [00:00]
A
You're watching TVPN. Today is Wednesday, October 1, 2025. We are live from the TVPN Ultradome, the Temple of technology, the fortress of finance, the capital of capital, the stable of stablecoins.
- [00:12]
B
We got a real stable. We have a real horse behind us. Real horse. Statue.
- [00:17]
A
Not a real horse.
- [00:18]
B
Can we hand to something more like a Y?
- [00:21]
A
There we go. Look at this fantastic, wonderful new addition. Thank you to everyone that supported us. Thank you to everyone in the chat. You made this possible now. Thank you. We have a horse in the studio.
- [00:34]
B
Wouldn't have been possible without the chat.
- [00:36]
A
The slop could never. You have to. This is craft.
- [00:40]
B
This is an antidote to slop.
- [00:40]
A
This is an antidote to slop.
- [00:42]
B
This is craft.
- [00:42]
A
This is craft.
- [00:43]
B
This is craft. So if you're frustrated with all the AI slop that you're being forced to see, commission a physical monument to beauty, power. Get a massive statue, magnificent as the horse. Obviously, you know, those that watch the show know this show is about technology, business and the equestrian lifestyle.
- [01:05]
A
Yep. There is a. We're obviously going to be talking about Sora basically all day. It's really big. That's what you wanted, though. You asked for a big horse, you got a big horse.
- [01:15]
B
I asked Ben.
- [01:17]
A
It's not any bigger than a real horse. It might even be smaller than a real horse.
- [01:21]
B
I know, but. Oh, it's a bit taller.
- [01:23]
A
Yeah. I don't know. We'll have to get a real horse in here and compare it next to one another. There is going to be a huge arms race for making content that definitively cannot be made with AI. Right. There's something that's going to be. Oh, when you're shooting something, let's deliberately try and stay away from the AI aesthetic. Right. Don't you think that's going to be a trend? Yeah.
- [01:46]
B
Brent, as soon as there's one example of something you can just on almost.
- [01:51]
A
I don't know. I think it's fun to push the envelope and find stuff. Like, I was playing around with both Sora and Vibes last night, and I was trying to get it to generate me and Tyler walking around an M.C. escher type staircase with, like, you know, M.C. escher paintings where the staircases kind of go off at different angles and it kind of breaks physics. And both models were falling flat on their face. If you've seen, you know, Inception, there's a scene where Christopher Nolan uses CGI and some camera trickery to kind of recreate an endless staircase in a real film. And AI Would definitely struggle with that. There's a few examples of this for me.
- [02:32]
B
I use the second that I realize something is AI generated, I usually just scroll by it.
- [02:38]
A
Yes.
- [02:38]
B
If I'm seeing a startup launch video that's obviously AI, I watch like half a second in and keep scrolling. Because today, in general, if you could tell it's AI, it's usually somewhat of a low effort thing and it's probably not worth watching. Same thing with comments online. Everybody's been reading a response to a post or something.
- [02:59]
A
Oh, no, this is.
- [03:00]
B
And the second I realize it's AI, I just keep scrolling because it's just generally a good filter for quality. And I actually am worried that the day will come where I no longer can tell. And certainly it's probably happening already and I just don't know it.
- [03:18]
A
Yep.
- [03:19]
B
But.
- [03:20]
A
Yeah, but I think people will continue to be able to innovate and come up with different things. I mean, even just like this show's in three hours long, it's very clear that we're not AI generated because the models only go for 10 seconds. There's a million different things that you can bring to bear. One funny thing that I saw Sora fall flat on was it passes the liquid refraction test, where you take a glass of water, you put arrows behind it, and as you fill up the water, the water in real life refracts the arrow and reverses the arrow. And for a long time, AI video models fell down on this. They didn't understand physics at that level. Now Sora passes it. The one place where I saw Sora fall flat, another place, as opposed to MC Escher stuff, was, you know, those balls that go. The metal balls that go back and forth. You saw this, Tyler. Did you see the one that failed where it was just one ball bouncing back and forth and it didn't understand how the physics would transfer. From the left.
- [04:16]
C
Newton's cradle.
- [04:16]
A
Newton's cradle, that's right. And I noticed that our sponsor, ramp, ramp.com, time is money saved. Both easy to use, corporate cards, bill payments, accounting, and a whole lot more all in one place. They launched a publicity stunt. Massive, massive campaign. Heaven from the office in there. And in the video it shot extremely cinematically. A lot of people would probably be like, oh, is this AI? But just maybe they had foreknowledge of where Sora 2 would fall down, but they had a real footage of Newton's cradle and it's doing the real thing. And so it was just very funny that I saw this clip in that video, that was the one thing that you can't do with AI and it's in there and it lets you know that it's real. And so we will be, of course, following that campaign as it goes out. It's fun ramps, obviously, expanding the market very aggressively, going after tons of businesses inside and outside of Silicon Valley. And no better way to do it than with a household name like Big Influencer. This is what happened with Saquon.
- [05:16]
B
Household name and finance.
- [05:17]
A
Household name and finance, certainly Kevin from the office. Household name Everywhere in the U.S. honestly. Anyway, the slop versus farming debate. I got kind of messy with the metaphor in today's newsletter, but I was digging into the reactions to SORA and my question was, you know, slop is bad. We, the timeline, don't want to be pigs at the trough.
- [05:41]
B
And by the way, this last angle here, not AI, this is a camera that is on a. Oh, are we.
- [05:48]
A
Ready to do a.
- [05:51]
B
What is it like, effectively RC operated?
- [05:54]
A
Oh, yes. Yeah, this is real too. We have this animated camera now. Or motion, I don't know, slider. It seems like the boys are ready to give some love to Julius. Julius AI the AI data Analyst. Connect their data, ask questions in plain English and get insights in seconds. They are pumped up for Julius. Let's hear it. Everyone's going crazy in the TB Pan Ultradome for Julius AI, the AI data analyst.
- [06:27]
B
That came out of nowhere.
- [06:29]
A
Ridiculous. Anyway, we don't like when tech leaders treat us like farm animals. We. The timeline, of course, but we do love farming. I don't know. I was trying to map this because we don't like slop, we don't like pigs at the trough, but we like farm to table. We think farming is Lindy. There's this whole idea of like, get off of the Internet, touch grass. Where do you find grass on a farm? Obviously, and. And once you. And there's something about returning to a world where we're farming. But if you're a farmer, you're filling up troughs on a daily basis. And so, like, how do I put these two things together? This idea that, like, we don't like slop, we don't want to be pigs, but we do want to be farmers and we do want to create slop. Or is the farmer no longer noble because the farmer fills up the pig's trough with slop? Is that bad?
- [07:23]
B
Well, pig farming and farming, two different things.
- [07:27]
A
You think now it's like, oh, it's not enough just to be a regular farmer. You gotta be a Farmer of like crops.
- [07:33]
B
Well, when I think about a farm that is supplying a neighborhood restaurant that focuses on organic farm to table food, I'm not thinking pork.
- [07:43]
A
You don't think pork's gonna be on the menu? You don't think some pork cutlets will be served at some point, some delicious bacon and a breakfast burrito. Farm to table.
- [07:52]
B
You can go way down the pork rabbit hole.
- [07:55]
A
There are plenty of other animals that eat a trough eat slop essentially. Is it just pigs that eat slop? I don't know. But my question, my question, but I.
- [08:04]
B
Would just say put differently, I think a lot of people enjoy free range content.
- [08:10]
A
Yes, indeed. If you want some free range content, head over to Restream iO1 livestream. 30 plus destinations multi stream. Reach your audience wherever they are. So there are now three different AI video products that have all been received very, very differently. So Google DeepMind did launch with YouTube. Like you can generate slop videos AI videos with VO3 in YouTube and YouTube scales. Massive. And so we should be seeing a lot more of that. But YouTube's focused on the creator first. Meta Superintelligence launched Vibes in the Meta AI app. It's a purely AI video feed based on mid journey and OpenAI now has Sora and so understanding the three different strategies, who the customer is, how they think about different ways to bootstrap these new networks, all of that's very interesting to me.
- [09:01]
B
There is meta AI on the app store right now.
- [09:04]
A
Figure it out. So YouTube has number 17. Number 17 productivity. YouTube is typically beaten with slop allegations just because even though there's a lot of faceless channels out there, there is straight up AI content. I see some of it on YouTube where you click and it's like, oh, there's new cars coming out and it' just completely fake. Like nothing about it is real. It's just like, you know, Ferrari launched a new car and you're just like, you click on it and then it's just like this is just lies. It's not. It's beyond slop. It's like it's just completely fake. Clickbait. Just optimized for getting.
- [09:39]
B
Yeah, it's not entertaining. Yeah, it's view farming.
- [09:41]
A
Yeah, like I'm fine if it's grounded. If there's something that's grounded in reality.
- [09:45]
B
And then it's constituted meant to be entertainment.
- [09:47]
A
Totally, totally. But so those are clearly annoying. But in General, I think YouTube has dodge the slop allegations mostly because there are so many creators who are like crazy niche interests yeah.
- [09:59]
B
Yeah.
- [09:59]
A
What do you say?
- [10:00]
B
Well, it's so. It's so. I mean, YouTube went from being, you know, subscriber, subscriber focused driven to algorithm driven. And, you know, a lot of the slop that I'm seeing is, is friends of mine, people that I follow on X that are just sharing it because it's funny and novel and new. But I don't know if I'll be seeing quite as much in a week.
- [10:22]
A
We don't see Ghibli's that often anymore. And is this the Ghibli moment or Is this the ChatGPT moment? OpenAI was certainly framing it as the ChatGPT moment, but we'll see, we'll keep following it. Instagram also feels similar.
- [10:34]
B
What do they call it? The ChatGPT moment.
- [10:36]
A
For video.
- [10:36]
B
For video, for video creation.
- [10:39]
A
And so I wasn't able to come to a strong conclusion, but I just kind of. In the newsletter today, you go to tbpn.com, throw your email in there. Everyone's looking great. It's a good slop. I love the chat.
- [10:53]
B
Please, sir, just one more slop.
- [10:56]
A
I had five questions that I think we should debate. Tyler's back in a wonderful looking suit. Your new suit is looking fantastic. I like that dark green, the tepee green. It really works for you. I got a brown suit. I should have worn it today. I think it matches that very well. It brings us into the fall, into the winter, where we're, we're.
- [11:16]
B
First day of Q4.
- [11:18]
A
First day of Q4. It's a massive celebration.
- [11:20]
B
Happy Q4.
- [11:21]
A
Happy Q4. Happy Q 4 to privy wallet infrastructure for every bank. Privy makes it easy to build on crypto rails, securely spin up white label wallets, sign transactions and integrate on chain infrastructure all through one simple API. So the first, first, maybe we should cover Ben Thompson's take because he gave some extra context on Sora. He says some things are old, some things are new. File, iPhone, dominance of anything new and interesting into the former. The coolest new apps came first to the iPhone in the smartphone era. And if that is still the case in the AI era, can you still say that Apple is behind because the Android app for Sora is not live yet? Which is again crazy considering, like, if code generation is good, shouldn't you just be able to say, like, take our iOS app and translate it into Android and push it out?
- [12:07]
B
Rune, you said. Well, you were hyping up Codex.
- [12:10]
A
Yeah, hit Codex a couple times, buddy. Make it happen. Just clearly it should be one prompt by this point, there are economic considerations too, like the iPhone audience won't monetize better than the than the Android audience, most likely. But Ben Thompson continues. He says what is new is the reality that compelling AI video generation is very much here and it's widespread. Over the last three weeks we have actual AI video products from Google, Meta and now OpenAI. What is so fascinating is how diverse these products are, how different they are. Google's building features for YouTube. Meta is building the app, the aptly named Vibes to take you away into fantastical worlds. OpenAI is letting you make as many variations of Sam Altman as you can handle. Indeed, it feels like each company is an entirely different target audience. YouTube is making tools, Meta is making the ultimate lean back dreamlike experience. And OpenAI is making an app that in my estimation is easiest for normal people to use. And he outlines this idea of the 99.
- [13:08]
B
Entertainment value.
- [13:10]
A
Which one? Sora, that's not Ben Thompson's take, but.
- [13:14]
B
That'S your, that's my personal take is that I think they did an incredible job creating meme, basically memes, video memes. Yes, that's what they are.
- [13:23]
A
So quickly to level 791 rule. This is the idea that in social media or in technology, 90% of users consume. They're the pigs. 9% of users edit and distribute. They're the farm hands and 1% of users create. They're the private equity firm that rolls.
- [13:44]
B
Up the farms, the pork, the pork.
- [13:47]
A
Barrel politics who are redistricting farms and layering tranches of debt.
- [13:53]
B
I don't know the pork roll up.
- [13:55]
A
No, the creators are probably the farmers, I would think in this analogy. And the editors and the distributors are the farm hands. So he says if you were to categorize the target market of these AI video entrants, you might say that you've tube focused on the 1% of creators. OpenAI is focused on the 9% of editors, distributors. I don't know how much I agree with that. And Meta is focused on the 90% of users who consume.
- [14:18]
B
1% of users are the hog farmers.
- [14:22]
A
Let's really stretch the analogy as far as possible.
- [14:24]
B
Hog farmers.
- [14:25]
A
This is when we're at our best.
- [14:30]
B
Or the pork producers.
- [14:32]
A
Yes, yes. I mean farms do have multiple tiers of workers on the farm. It's not that crazy of analogy. So Ben Thompson, this is where he gives what it feels like, at least to you, as a contrarian take he says, speaking as someone who is, at least for now, more interested in consuming AI content than distributing or Creating it. I find Meta Vibes app genuinely compelling. The Soar app feels like a parlor trick if I'm being honest and I'm tired of my feed pretty quickly. I'm gonna refrain. I'm passing judgment on YouTube because he mostly watches very specific videos there. He says he has no idea if the evaluation is broadly applicable because who knows if he's like the median.
- [15:13]
B
The reason that I think Sora.
- [15:14]
A
He says he likes bots. Out of the three.
- [15:15]
B
Out of the three, I think Sora feels like the best strategy and that is reflected in the charts right now. YouTube VO3 is great, but not to the point where you're one shotting entire scenes or certainly not one shotting videos. It can pretty good. Yeah, but it can't create anything that's truly compelling.
- [15:40]
A
Is that just because it doesn't have the cameo feature and you can't put Sam Altman in there? I mean, I made BO3 videos that were fun. I made videos of driving a Ferrari through the Hollywood sign that was fun.
- [15:51]
B
But I think if you. I think if it was. You can only use VO3 to create viral YouTube videos. Maybe if you. You being like very talented and understanding YouTube, while you could do it, it would be very hard and probably harder than just doing things normally.
- [16:09]
A
Yeah.
- [16:11]
B
Meta vibes. The content is. Is cool, but not interesting at all. I haven't. I let you demo Metaverse. Right. I wasn't super interested in it. So that content in general wouldn't even do that well on Instagram. Right?
- [16:30]
A
I don't know. It does do well. I follow one of those accounts on Instagram that takes midjourney photos and slashes them up into 1980s vibe reels of Swiss bankers in Berlin in the 80s and they put a cool song over it. And I'm into it. I don't see it that often. Most of my Instagram for you page is bodybuilding content that's 100% farm to table. But.
- [16:55]
B
Vintage Arnold Gold.
- [16:57]
A
I just found I had the same experience.
- [17:00]
B
AI right now can create wild scenes. And I think Sora leaned into where the technology is today and that it can be. If it has the right sort of memeable moments, it can be viral by itself. The content can be entertaining and be standalone. You know, watching Sam Altman steal GPUs from a CVS is a human generated idea. Excellent sort of delivery. Very in the zeitgeist was able to. Somebody was probably had to do a few prompts to get ultimately something that was entertaining and went viral. Not even in sora I'm sure it did well in Sora, but going viral on X and started a conversation. Not a single thing that VO3 has, that I've seen from VO3 or Vibes has started a conversation or really gone viral outside of the. Wow, this model is good.
- [17:57]
A
Yeah.
- [17:58]
B
So AI, like video models today are good at generating short scenes and if you make them memeable and if you make it easy, these things can go viral. And so. And from my point of view, I think that Sora is the best product I got.
- [18:14]
A
372 likes, 20,000 views on VO3 is amazing. But there are some hallucinations. We're going to have to shoot this practically of us blasting through the Hollywood sign in a yellow Ferrari. It's not bad. It didn't. Yeah.
- [18:27]
B
But again, you have. You've been posting about AI daily for years. Yes.
- [18:33]
A
And it's a good demo of the product. So you're reacting to the model tied back into arc. And also you might like it to be.
- [18:38]
B
That's funny that you showed that to a stranger. They'd be like, thank you for wasting, you know, 90 seconds of my life. John.
- [18:45]
A
It was eight seconds. It was eight seconds.
- [18:47]
B
But if you walked up to somebody like on Melrose and you said, look at this video you use ChatGPT. Here's the CEO. Did you see him stealing GPUs from CVS? They'd be like, oh, that's crazy. More of a reaction than just, it's a good model.
- [19:03]
A
This is weird, but I agree with Ben that if I had to watch 10 minutes of one feed over the other, I would pick Vibes over, well, Sora, because watching so, so chaotic and it's so.
- [19:18]
B
I'm not saying I want to watch.
- [19:19]
A
It's more uncanny valley. That's what it is, more uncanny valley. It gives me more like jitters when I watch SORA videos than when, if I watch a mid journey video on meta vibes, I'm just like, okay, I'm listening to real music. It's clearly. It's trying to be illustrated. It's not trying to be photoreal. And then I'm like, oh, the leg was a little bit off. It's like, well, yeah, like, of course the. In the steampunk cyberpunk world, like the suns didn't line up perfectly. I'm not even. I'm suspending all my, you know, my.
- [19:48]
B
I'm not saying I want to go in the sewer feed, I want to go in the trough and slop.
- [19:52]
A
But if you had to go In a trough. Which one would you go in? That's the thing. Ben saying I would say vibes trough.
- [19:59]
B
What is going to get me to. What is going to get me to stop and watch a video if I see it on X or in Instagram or anywhere else?
- [20:06]
A
Right now, right now, it's Sora 100%. I agree.
- [20:09]
B
So I'm saying the output is better. It's more built for the Internet today. It's not just a research project or it's genuinely creating net new content for the Internet that people seem to be very entertained by.
- [20:22]
A
Yep. So Ben Thompson continues and kind of lays out a question about how successful this will be. And we'll debate this more as I get to my questions. But he says the company resolutely. This is OpenAI, resolutely claims that they had no idea ChatGPT would be popular, now feels confident declaring that they're about to repeat one of the most seminal moments in the entire history of technology. I'm not so sure. And that's about the OpenAI team saying that Sora is the ChatGPT moment of video. And Ben Thompson says Sora is to soar two is, to be clear, amazing. And the app is very easy to use. What matters in terms of creating moments that matter, however, is consumption. And while creation is obviously a prerequisite to consumption, it's not the main. What made ChatGPT unique was that LLMs immediately delivered perfectly customized content for each individual user to consume based on the most basic of prompts. In this. In this, the fact that text is cheap and easy, it was critical people create video, however, for others, and I'm just not sold that AI video that the AI video that Sora app, that the Sora app enables, easy though it may be to make, is that interesting to anyone other than the creator. And so you might get a laugh out of making a video for yourself, but how wide ranging is that gonna be? And that does seem to be their cameo strategy. Right. Like, it's like Tyler made a video of me bench pressing and then turning into a golden retriever, and it was really only funny for me and him. Maybe a few other people would think that's funny, but that's not gonna do well. Maybe people will come up with stuff that really does go viral within the app. I don't know.
- [22:03]
B
Yeah, I mean, the question that's most important to me is I think Sora is a great tool for AI content creation. I think it is the best tool to create AI content for the current Internet, which is algo4u video feeds. It's the best product for that today. If you went and imagine you give three creators. It's like you get VO3, you get meta vibes and you get Sora. Who's gonna have the most views after the end of a week. Right. I would say that Sora is like massively overpowered in terms of just generating attention.
- [22:37]
A
I think you're right.
- [22:38]
B
But scrolling the feed feels like it would make you go. Absolutely. It is not fun. I've made it.
- [22:50]
A
It might be fun.
- [22:50]
B
So far I've opened the. I took your feedback on not actually using the Meta Vibes app, just going off of the vibes and your experience. I've opened Sora a few times and I've made it about like three swipes. Swipes. Three.
- [23:09]
A
Three chugs of the trough.
- [23:11]
B
Three spoonfuls. And gotten to the point where it just feels really bad for you.
- [23:17]
A
It feels crazy. The editing too. Like when even just like the cuts from one scene to another really, like out of place and jarring. It makes my skin crawl. Like, a lot of the videos are not good in like, just like it feels weird and bad. Like, I don't know. It'll stay that way if you.
- [23:36]
B
It makes me. Yeah, that's the question. Will AI video always send a signal?
- [23:42]
A
Yeah.
- [23:43]
B
Almost a silent signal that you are unsafe as a human. That's what it feels like today.
- [23:52]
A
Yeah.
- [23:53]
B
Or can they cross the uncanned valley?
- [23:57]
A
Yeah. Uncanny valley. Getting out of that. And then also just what happens when you're like. If you. If I talk to ChatGPT as like a person and I try and treat it like a person, it does give me the same reaction where I'm like, this is weird. Why is it glazing me or whatever. It's like, it's sycophantic and what. But when I use ChatGPT for knowledge retrieval, I'm very happy and I'm like, give me the facts, give me the data on this company or whatever. And I could imagine a world where there's some sort of video that's generated automatically that I feel is much more. More tactical. It's not trying to fool me. It is upfront about what it is. And that is more satisfying. I don't know. Anyway.
- [24:38]
B
Yeah. I mean, makes total sense from a business model stamp, just like a business strategy standpoint, to create this. Right. They're going to be able to. I imagine if they can even get. If they could get a million people using this. Swiping, creating, et cetera. Consistently, that will give them. That will be very helpful in making the model better and better and better and better and better. And it will make the app better and it'll probably get more users. The question I was thinking about is, does this. Even if they create a Snapchat size business, it's sort of a rounding error on their market cap. Right?
- [25:15]
A
Sure.
- [25:16]
B
Snapchat. How many? You should know this, Tyler. How many? How many users? How many daus does Snapchat have?
- [25:27]
A
Let me think about that. Cluley churning. Not. Not being that you churn from Cluli allegations.
- [25:35]
C
63 million.
- [25:36]
A
It's a fair amount. I don't know. Well, the first question I had that I wanted to sort of debate is like, how did OpenAI do this? I ended the last newsletter asking that question. For the past month or two, there's been this vibe that Google D mine had a material moat around video generation because they had control over YouTube, MSL meta superintelligence. They have access to all of Facebook and Instagram video and a lot of people cross post all sorts of YouTube videos to Instagram and Facebook. What you got for us?
- [26:15]
C
I just want to fact check what I said. That's totally wrong. It's 460 million.
- [26:19]
A
460 million. Jordy. That's a lot of users. A lot of users. Yeah. DAUS. 460. That must be international. Yeah, must be everywhere. Wow. Well, what's interesting is that OpenAI didn't have a clear dance partner from which to source. Literally. YouTube has billions of hours of video. That's so much data. And so the OpenAI clearly found a creative solution. One of those things that was reported was that if you had copyrighted content, you had to tell OpenAI, I want to opt out. Pull me out of the training set. Otherwise they would train on it. And we're going to react to some SORA videos that are very clearly like Mario intellectual property. And so they figured out some workaround where they could train on and basically that's owned by other people.
- [27:09]
B
And they'll find it came out that they let various IP holders know you have 24 hours. And clearly a lot of them, I guess, didn't react quickly enough.
- [27:27]
A
I got hit with one of these. I found a reference that I couldn't. It said content violation. So I of course, was prompting a golden retriever with a man body wearing a suit doing a podcast because I'd seen this on meta vibes and I was like, let me bring this over and see what SORA can do with it. So you can imagine a midjourney with a. With a man in a suit, but with a dog head, golden retriever head on a podcast. But the song that was playing in the Meta Vibes reel was one of my favorite songs, Dumbest girl alive by 100 Gex and. And I. And so in my prompt to Sora, I said, golden retriever with man body wearing a suit doing a podcast. Dumbest girl alive by 100 gecks playing song playing over the video. Because I wanted to know, could it put the correct song over there, like I can do in Meta Vibes? Is it truly comparable video generation? And it hit me with a, hey, we can't do that. This is a content violation. What do you think?
- [28:27]
C
So I actually saw another version of this where it was sweater weather.
- [28:30]
A
Yes.
- [28:31]
C
And that was like, the lyrics were wrong, but you could very clearly tell that, like this, it was the same song.
- [28:36]
A
Yeah, it was the exact same. So what I imagine is that there are certain deals with certain record labels, some record labels, some IP owners, some IP aggregators have said, opt me out entirely. And then others have said, hey, just pay me if you monetize my work. Like what happens on TikTok, like what happens on YouTube. Others might say, don't put me in at all. Others might say, hey, I didn't get a chance. I didn't get the memo. Just train on me, do whatever, just go wild. Profit off of it entirely. You get a hundred percent. And then. And then I imagine that there's a second fee, like second filter that's happening on top of all the prompts and all the content to say, is this acceptable based on content policies? Is this explicit? Is this adult content? Ban that? If so. And then also, is this infringing on IP that we don't want to infringe on? Let's pull that down. And so I don't exactly know exactly.
- [29:24]
B
Yeah. One thing that OpenAI broadly is world class at is style transfer. This is why the Studio Ghibli moment happened. It was perfect. Very, very consistently. Right. And we saw this again yesterday with people doing their own versions of South Park. Right.
- [29:43]
A
Like, south park is one shot.
- [29:45]
B
It looks you could convince me, Rick and Morty, if the dialogue wasn't too crazy, you could convince me that that.
- [29:51]
A
Was a real scene, especially on the video fidelity. It looks fantastic. There are others that we're going to react to today that clearly didn't get opted out of the. Of the training set. And so they're in there. And who knows how that develops. But this is the story of all social media like YouTube was not really on top of copyright holders as they grew that business. When they got acquired by Google, Google had a massive settlement and I believe paid almost a billion dollars, like maybe more than the actual acquisition price. To settle those lawsuits and get in good with the we need a prediction.
- [30:30]
B
Market on how big the ultimate settlement for the class action.
- [30:34]
A
Yeah, there might be. There might be a one time. There might just be a rev share agreement. I mean, that's what happens on YouTube today is that if you go and steal my content and you upload this and you start getting AdSense revenue, Google will just route that to me. If you steal some of this video, you take a TVPN clip, you put it up there and if it starts getting views and starts generating ads, it'll be automatically claimed and come over. I will also have the opportunity to take it down, but by default I'll just get the revenue share over here. And there's a whole niche business of shuffling around these payments. And you can imagine the same thing. I mean, the same thing happened with TikTok where TikTok was abusing that 15 second preview in the Apple music preview. Do you remember this, Jordy?
- [31:21]
B
I do.
- [31:22]
A
So, yeah, in Apple Music, in the itunes app, you could preview 15 seconds of any song because they were like, well, you pay for that. Pay a dollar for full song credit.
- [31:32]
B
For that credit to ByteDance.
- [31:33]
A
Incredible.
- [31:34]
B
It was a genius hack.
- [31:35]
A
I don't know if it was credit to ByteDance. I think it was musical ly the company before.
- [31:39]
B
Okay, same thing.
- [31:40]
A
Eventually the same thing, yes. But that was something where it was very unclear how that would pencil out. And ultimately the record labels got in a good spot with TikTok, I think, and they're happy with the promotion and rev shares that they get. And it doesn't feel like there's active lawsuits. There's something going on here where either OpenAI is moving fast and breaking things and will pay a big settlement, or they already have contracts in place and will be sharing revenues over time with how much likeness goes. You can already see this with the cameos where if you upload your cameo and then someone uses your cameo to generate a ton of views, they go viral with it. There's a bunch of ads associated with that. It would be pretty easy to flow. Okay, this video generated $10,000 of ad value. We're going to. On YouTube, you get 5,000 of that. Basically half. It's like 45% or something. 55%. So you take half of that Put that into the creator payout. And then from the. Thanks nick, from the 50% rev share, you could also slice it up even more. So imagine if you upload a YouTube video. Yeah.
- [32:56]
B
And this is something that we were talking about earlier. The air. How does, how does open like right now? Sora, It's a great AI video creation tool that's undeniable. If they can build a network, they'll have to create durable incentives for people to build audiences, invest in making audiences on Sora and not just creating the content, taking it to other apps where they already have a big audience. Right. The incentive now is to make it on Sora.
- [33:26]
A
But that didn't matter on, on YouTube. Right. Or on Instagram. Like Instagram was a photo filtering app that bootstrapped a social network on the back of it.
- [33:35]
B
Sure. I'm not saying it's a bad strategy. I'm just saying they have to create an economic incentive over time. Like on, on, like sure, it's great that there's a. You need supply before you can have demand. Right. But I think that what TikTok did well, what YouTube did well, what all these platforms did is they, they ultimately became a place where creators could get attention, they could get money. Right. Torah needs to drive like user minutes in the app watching.
- [34:09]
A
I agree with that. I think it's maybe.
- [34:12]
B
But the last. What I was gonna. Yeah, because what I was gonna say is if, if IP holders, like if people can create John Coogan content and get a rev share or something and then you get a split, that can create an economic engine that will just drive more activity. Right now if you look at faceless accounts on other platforms, they historically are very, very difficult to monetize. So if you have a meme account on Instagram, good luck making money on that. You can have a million followers and it's really tough to get much value out of it at all.
- [34:47]
A
Totally.
- [34:48]
B
And same thing for totally faceless, non personality driven YouTube channels. Right. Same thing with X accounts that just share breaking.
- [34:58]
A
Right.
- [34:58]
B
And they're pretty much worthless.
- [35:02]
A
So yeah, I think act one, bootstrap the network on the back of a creative tool. You go there to create a video that you might share elsewhere, but it's shared by default. So there's liquidity in the system and there's content there for those who do want to scroll it.
- [35:17]
B
And then the challenge after that, the challenge right now is I go on the app. I hate it, I want to leave. Maybe I go on later today to try to create a video and I Stick around for a little bit. Maybe the content gets better and better over time, but they got to make it so that it's not disgusting to look at and can't be understated. The timeline. Yeah, yeah, the timeline absolutely hates it. They could not hate it more. I think people, this is a, this is kind of the first time that I feel like people are really impressed with the quality of the outputs from a model, but hate it. With Studio Ghibli it was like, okay, this is really impressive. You can one shot this incredible style transfer. It's magical. You look at it, it sparked joy. Some people are getting that a little bit, but at the same time it's rough. Right. Andrew McCallop said he was quoting William Fetis who's coming on later yesterday. He William announced periodic labs. Their goal is to create an AI scientist. William's coming on the show later today. We're excited about that. But Andrew said, I'm embarrassed for the existing AI players. Imagine sitting on compute empires yet not operating. A Bell Lab style effort to discover new physics or materials. This is infinitely more valuable to humanity then another video slot machine, slot machine, slot machine. And of course Sam Altman already responded to some of the various frustrations that people had. Let me try to pull this up.
- [37:00]
A
Yeah, it said so this came from runner Tushar said Sam Altman two weeks ago, we need $7 trillion in 10 gigawatts to cure cancer.
- [37:09]
B
I remember I was saying Sam actually.
- [37:11]
A
Said we have to cure cancer and.
- [37:14]
B
No, he was bas basically saying there's a world in the future where we might have to choose between free education for everyone and curing cancer. And basically implying give me a trillion dollars or we're going to have to make this call. Don't make me make the call. Turns out there's a third demand. There's a third demand. Right? So you might have to. He might be in a position where he has to choose between the infinite slot machine and free education, curing cancer. He's making the argument, he responded this and said, I get the vibe here, but we do mostly need the capital for build AI.
- [37:54]
A
This is a typo but you know, it's.
- [37:56]
B
This is good. This is good. He put a typo in here intentionally to show that he wrote it himself. He said we do mostly need the capital for build AI that can do science and for sure.
- [38:06]
A
For building AI that can do science.
- [38:09]
B
And for sure we are focused on AGI with almost all of our research effort. Incredible that they were able to create a tool this impressive with so little of their research effort. It is also nice to show people cool new tech products along the way, make them smile and hopefully make some money given all the compute need. When we launched ChatGPT, there was a lot of who needs this and where is AGI? Reality is nuanced when it comes to optimal trajectories for companies. Totally agree there. And yeah, a lot of people were speculating. Did Meta rush the Vibes app to try to be first to market? It'll be interesting to see how the meta's down 1.8% today. Hard to read too much into it. There's a number of other factors including that people now believe Zuck will spend all of their free cash flow on capex. So some various factors here, but yeah, certainly impressive that Sora was able to go number one so quickly.
- [39:16]
A
Well, over in the code generation world, Cognition has announced that they rewrote Devon for Claude 4.5. There's an interesting article we can break down in a minute, but you can check out Cognition. They're the makers of Devin. Devin is the AI software engineer Crusher backlog with your personal AI engineering team. Donald Boat is putting a lot of stuff on the timeline. You already mentioned this, but he said, guys, it may seem like their overt plan is to turn your children and grandparents into drool slurping morons for 551% year over year subscription growth, but they actually do need to make 5 million videos of an elephant walking into a McDonald's. McDonald's to cure cancer. The funniest take is that, like, maybe this is critical path. Tyler was talking about this a little bit where he was saying, like, the AI video might be good for robotics training data.
- [40:12]
B
I just don't know if OpenAI has the time to cure cancer. They got a lot of. They got a lot of stuff going on right now.
- [40:21]
A
Yeah. What are you sucked into right now, Tyler?
- [40:27]
C
I made a new video and it's not posting. Yeah, I think there must be too.
- [40:31]
B
Many GPUs are on fire.
- [40:32]
A
Yeah, GPUs might probably text.
- [40:34]
B
I texted Tyler, can you make a Sora of Sam Altman as a pig farmer at a trough putting slop in it?
- [40:41]
A
You don't. You don't even need to make that. I've seen that on there. Really? Oh, I've seen that like a dozen times.
- [40:45]
B
But I wanted him to say, oink, oink, piggies.
- [40:47]
A
I think somebody said.
- [40:48]
B
Really?
- [40:48]
A
I think somebody said that.
- [40:50]
C
There's no audio though, because I had to screen record it because I can't post it.
- [40:53]
A
Weird. Yeah. Okay. They must be censoring it. He probably. Well, that's the thing with the. With the collabs. What's it called when you scan your face?
- [41:01]
C
Cameo.
- [41:02]
A
Cameos. The cameos. You can put a prompt in there and say, never depict me as a pig farmer or always depict me with a gold chain. And it will just do that.
- [41:11]
B
But I could jailbreak it by saying, I want.
- [41:14]
A
You know, I gotta. I gotta do this swineherd.
- [41:16]
B
Swineherd is another. Is a historical term for someone who herds or tends to pigs.
- [41:22]
A
It's gonna know. It's gonna know.
- [41:23]
B
I can tell.
- [41:24]
A
Yeah. I need to get in my. In my cameo and say, always depict me as a mass monster. For sure. That needs to happen.
- [41:31]
B
Anybody can use my likeness as long as I'm a gigachad.
- [41:34]
A
For sure. Donald Boat says, Would it be terribly surprising if I told you that the geniuses behind their latest advertisement campaign are the same minds behind the seminal and profound visual televisual experience Euphoria? Listen to me now. Never, never take these computer people at face value. As soon as they build a machine of overwhelming might, they are going to send it against you and everyone you love.
- [41:57]
B
Black Pill 6.
- [42:00]
A
Wow.
- [42:00]
B
Yeah.
- [42:01]
A
And he says, attention, you are out of reply tokens for this month. To Continue Replying to Laserboat999, please visit donaldboat.comLink in bio and purchase either a yearly or monthly subscription. Failure to acquiesce to these demands will result in an authoritative block in 10 prompt minutes.
- [42:16]
B
Okay, I guess let's pull up the video Tyler made.
- [42:20]
A
Yeah. Let's see. Actually did generate it. Generated.
- [42:24]
C
I just can't post it. You can't post it and there's no audio.
- [42:26]
B
Yeah.
- [42:27]
A
Okay.
- [42:27]
C
But in it he says, oink, oink, piggies.
- [42:31]
A
This is incredibly photo real. There is another video in the timeline that's the same thesis for sure.
- [42:40]
B
Good stuff.
- [42:41]
A
Well, people are having fun with it. Santiago says Sora 2 is wild. I asked it to generate the greatest hype video of all time and it made me this. And it's of course, the handcrafted Founders Fund LP intro. The Pete Oxenham version with Alex Jones over it. We can play a little bit of this. It's just a classic. It's great. Instead of looking up to Thomas Jefferson. That's enough of that. The whole thing. It's kind of like putting pineapple on a pizza. You.
- [43:10]
B
And you watch that every morning when you wake up.
- [43:11]
A
Every morning? Yes.
- [43:12]
B
Every morning you throw on your meta quest and just fully experience it. No, no. Wellmanitis says, if you genuinely believed you were two to four years away from AGI is so really the thing you'd release, Brandon says. Jacoby says, sadly, yes, man was given the whole garden and we chose to just indulge in the apple.
- [43:35]
A
Interesting. Brandon Jacoby, one of the greatest designers in history.
- [43:40]
B
True.
- [43:40]
A
Of course, his tool of choice is Figma. Think bigger, build faster. Figma helps design and development teams build great products together.
- [43:47]
B
I've spent many late evenings in Figma with Brandon Cookie. Some of the highlights of my career.
- [43:55]
A
Okay, we got a timeline in turmoil. Boo. Andrew Wilkinson, the tiny man himself says, I think OpenAI just killed TikTok. I'm already laughing my head off and hooked on my Sora feed. And now the number one barrier to posting, the ability to sing, dance, perform, edit is gone. Just 100% imagination rip to theater kids. Hunter Weiss says, no shot. And Theo says, I can't stop thinking about how bad this tick was and gets 1000 likes on it. Almost ratios. Doesn't quite what do you think? Is this a material threat to TikTok? Is TikTok now overvalued at 14 billion? A minute ago you were saying, oh, 14 billion is too low for TikTok. Maybe TikTok's a $200 million company now. Maybe, maybe they just killed it.
- [44:48]
B
Well, I think it, I mean we got to talk about TikTok again because something I wasn't even thinking about, I feel silly now, is that they're probably just going to take TikTok public and it will immediately trade up. TikTok is not going to trade. If TikTok US is public today, it is not trade anywhere near 14 billion.
- [45:09]
A
Billion. Where do you think it trades?
- [45:12]
B
I think, I mean so they do.
- [45:14]
A
I think there's a.
- [45:15]
B
They do around 15. They're at roughly a $15 billion run rate for revenue if you assume they're going to be around for a long time and can get quite profitable if they stop doing this TikTok shop mumbo jumbo. Yep, that feels like it could trade a 10x revenue.
- [45:31]
A
Yep, that's where it was trading in the private markets beforehand. Although that was not just the US business broken out, but most people were expecting this deal to land at like.
- [45:41]
B
50, 60 ByteDance always was valued far less than the metas of the world.
- [45:48]
A
So the new entity is I think sharing 50% profits with the Chinese entity. There's some flow back of cash flow so you can't.
- [46:00]
B
Well, the public markets don't care about cash flow, John.
- [46:04]
A
But you know what is interesting? So Sam Altman is building Sora. He needs an endless trove of training data. Larry Ellison just bought TikTok. Larry Ellison runs Oracle, which has a massive deal. We need some red.
- [46:19]
B
Get the red string ready, Tyler.
- [46:22]
A
But Oracle has a deal with OpenAI for massive data centers. How are you going to justify burning all those GPUs? You need valuable services on top of them. There's a world where there's a TikTok OpenAI data deal a la the Reddit deal that happened. That kind of empowered chatgpt. Right. Like, what do you think, Tyler?
- [46:41]
C
I think it's also like, not that insane to just imagine that in one to two years when the tech kind of like trickles down, you just see TikTok release their own version that's just trained on their own feed.
- [46:50]
A
Yeah, maybe. I mean, at 14 billion. Also, like, it's an acquisition target for OpenAI. Like they. If they're at 500 or something right now and they're buying love from in the billions, why not pick up TikTok and then have a huge audience to bootstrap a new network with tons of training?
- [47:07]
B
Well, why not buy Snap too?
- [47:10]
A
There is an argument for Snap. TikTok just aggregate them all, probably. I mean, they're private, so it's got to be easier to get deals across the line. But you know that there's going to be some lawsuits and there's going to be some people fighting that in the antitrust world. But it's hard to justify if your two biggest Foundation Lab competitors are MSL and Google DeepMind and they have YouTube shorts and Instagram reels. It's not that crazy to say, hey, you know, I'm Sam Ullman, I run another lab and I want a play in vertical video. Give me this. I don't know. I don't think OpenAI just killed TikTok. I think OpenAI. I think I disagree with this, with this Andrew Wilkinson take.
- [47:52]
B
Yeah. I mean, he obviously structured this post to rage bait, so we shouldn't, shouldn't read too much into it.
- [48:01]
A
Yes. In general.
- [48:03]
B
I mean, I think just another reason why it's overly dramatic is that social networks, while scale is critical, are not. TikTok didn't kill Snapchat.
- [48:18]
A
Yes.
- [48:19]
B
Stories didn't kill Snapchat. Right. These are.
- [48:22]
A
This was like the One hard post OpenAI'd comment is.
- [48:28]
B
There was some reporting that ChatGPT is referencing Reddit less.
- [48:33]
A
Yes.
- [48:34]
B
Probably because Reddit is just becoming entirely chatgpt. Maybe generated. They are down 11% today on that study.
- [48:44]
A
Well, you got to get on. Profound. Get your brand mentioned in ChatGPT. Reach millions of consumers who are using AI to discover products and brands. Profound. My definitive take on the Sora thing was people vastly overestimate the amount of brain rot that will happen in a year and underestimate the amount of brain rot that will happen in a decade. I think that it's good to have these conversations about the slop farm and avoiding brain rot and searching out Lindy formats for content and not wasting all your time consuming sugar and candy, essentially. But I do think some people are debating me in the chat, but I do think that I would be very surprised if the sore.
- [49:29]
B
No, I think they're talking about. They're talking about Andrew Wilkinson.
- [49:31]
A
Oh, Andrew Wilkinson. Okay. Well, you're free to disagree with me too. I don't care.
- [49:37]
B
Healthy debate.
- [49:38]
A
But the idea that the Sora app, I think this was your point, was that the Sora app will not be like a major, major consumption vehicle for content in the very near term. Right?
- [49:52]
B
Yeah. That it could grow and it could see it. Now, remember, ChatGPT has an insane amount of users. They can continue to push users to Sora now. They're not having to acquire them all organically through people sharing content, et cetera. And we saw this with threads, where threads had an initial pop, fell off dramatically, and now has climbed and reportedly has more, you know, weekly actives and X now. So.
- [50:23]
A
And the only. And the only flywheel there is that they surface some threads on Instagram and then you just click one button.
- [50:29]
B
You can only see the first like 10 words.
- [50:31]
A
Yep.
- [50:32]
B
You're like, I want to know what this.
- [50:33]
A
And it just brings you over. And they just do a little bit of that every day with a billion plus users. And when you have a massive, massive user base, you can just grow it, grow it, grow it very, very slowly.
- [50:42]
B
Yeah. Akshay over at Notion says, why do we keep dedicating our brightest minds, billions of dollars in the most powerful GPUs on earth to build yet another app that optim for attention decay. I was hoping Chat GPT seemed to reclaim time from Tik Tok and I don't have reply reclaim time from Tik Tok and Instagram. It felt useful, even nourishing. But now we see disposable video, same engagement, Treadville and Path to Ads. If even our nonprofits can't resist this gravity, what does that say about. Of course, Soar was released by the for profit arm of the nonprofit.
- [51:19]
A
Yeah, Signal has some pushback on that, he says. Unfortunately, Ads fund research, Google Ads funded.
- [51:26]
B
DeepMind, I would say fortunately.
- [51:28]
A
Fortunately, yeah. Meta Ads billions poured into AR VR. OpenAI needs massive cash flow to bankroll AGI. The treadmill sucks, but it's also the only mechanism society has ever found to subsidize frontier science at scale. I say this as someone who despises ads. Eric Suefer, fantastic analyst, Mobile Dev Memo, you should go subscribe says digital advertising is the most positive sum economically expansionary technology summoned by humanity. No other technology has created more consumer surplus or stimulated more long tail economic activity. Digital advertising provides consumers with free access to products that would otherwise be out of reach to all but the richest segment of society. The existence of the Internet itself is dependent on digital advertising, and absent that commercial engine, it would cease to exist as a freely accessible commodity and would instead be a luxury good reserved for moguls and despots. I say this as someone who loves ads. Let's go. Thank you Eric Soufert for saying the hard part that no one else had the bravery to say you're on our side and we should have Eric on the show. He is a fantastic interview, fantastic analyst. Have not spent enough time reaching out to him to get him on the show anyway. Peter Go Speaking of ads, Vanta Automate Compliance Manage Risk, Prove Trust Continuously Vantage Trust Management platform takes the manual work out of your security and compliance process.
- [52:51]
B
If you think it might be stock two time, it probably is.
- [52:54]
A
Just do it anyway. Continue.
- [52:55]
B
Peter Gostive says what are the chances that Meta found out about OpenAI's AI video social app plans, got a partnership with Mid Journey Black Forest Labs and rushed out a similar app just before OpenAI did. Nathan says that's what happened. The Vibes app always felt rushed. It felt tone to death.
- [53:18]
A
I mean if you go to midjourney.com right now, I think that's their domain. If you just go to the Midjourney website and you don't log in and you scroll, it is just Meta Vibes. It is the same thing. It's like Midjourney images with Black Forest Labs video animation on top of it, like eight second clips just extending one frame and it's the exact same thing without music on top. And so I would love to be able to see inside what the Vibes like project team was because it does feel like it was just a couple developers really really fast, like just a few features in there. Tyler, you have any other context on Vibes?
- [53:55]
C
I mean even like the ui. If you're just looking at the UI of Vibes compared to Sora, Sora is like. It's so much better.
- [54:01]
A
It's so much better.
- [54:02]
C
Just like, the buttons just look nicer.
- [54:04]
A
Totally.
- [54:05]
C
It feels bad to use Vibes.
- [54:06]
A
Yeah. Sora, they actually have a few unique UI features. I mean, the cameo thing is something that we hadn't seen elsewhere. Makes a lot of sense. We knew that fine tuning on a particular person was possible, and we've seen that with those magic avatars. But have you noticed that if you double tap on a Sora video, it will automatically. It'll show like, a little, like, explosion of like, confetti, but then it picks an emoji that it thinks relates to the underlying content and shows that emoji. So if you're on a video about dogs and you double tap, it will show you a dog emoji or footprints emoji. And they do a really good job of kind of like mapping the videos to emojis. And that's just like a unique UI experience that I've never seen. Similar to when the ChatGPT app came out and there was. You remember when it was typing, it would like, kind of thump in your hand. Do you remember that? In the app, we kind of stopped.
- [55:04]
C
It would, like, vibrate a little bit.
- [55:05]
A
It would vibrate a little bit as the tokens were streaming in. And it was just like, okay, someone over there at OpenAI on the iOS team is doing novel things. And that's exciting to me.
- [55:15]
C
Yeah, I think the emojis are actually good because sometimes, I mean, the videos are like nine seconds long. I'm not going to sit through nine full seconds. I want to see what the video is about right when I go to it. So I immediately get the emoji and I can. I can see what the video is about and then I can scroll faster. Yeah, I'm consuming more.
- [55:29]
A
I think that's the big. I think that's the big criticism. The videos are just too long on the. On Sora. Someone should make a Sora competitor that's just like half a second every video. And it's just.
- [55:41]
C
I think people would, like, use it more if the videos. If they just basically sped up every video, like 1.5x. Really, I think people would.
- [55:48]
A
You want to watch the Soar app on 2X? That's so deranged. I hope we don't lose you, Tyler. It feels like one of these days we're going to give you an intern challenge that just takes you off the deep end.
- [56:00]
B
And to his credit, he's got to be ready for questions for three hours a day, randomly, randomly. So doesn't have the same time to slop it up as others. Let's pull up this post from Joe Weisenthal. In the timeline he's sharing the video of. Did we play this already? Of Sam Altman stealing GPUs in a CVS or looks kind of like a target.
- [56:21]
A
This is from Gabriel, who works at OpenAI. He was having fun with I think.
- [56:26]
B
The research lead on Sora.
- [56:28]
A
Yeah, he says I have the most likely. He was previously at middle. Oh, he was.
- [56:31]
B
Wow. Yeah.
- [56:32]
A
I will be enjoying this short moment. While at last CCTV footage of Sam Altman stealing GPUs as at a target for soar inference. Let's play the video. We don't have it. Okay, well, we do have an ad for graphite.dev, code review for the Age of AI. Graphite helps teams on GitHub ship higher quality software faster. We will go back to the timeline.
- [56:56]
B
So Joe says anyone who sees this video can instantly grasp the at least potential for malicious use. And yet nobody with any power, either in the public or at the corporate level, has anything to say, let alone do to address it or even acknowledge it. This isn't even a criticism per se. The cap may be completely out of the bag and it may be reality that there's literally nothing that can be done, particularly if open source models are only marginally behind my. The. The thing that I don't think people are ready for is for a, you know, fully jailbroken, a website that you can go prompt to Sora quality. Sora to quality with anything, no rules.
- [57:33]
A
Yes.
- [57:34]
B
Like John, but he lost 60 pounds and he's frail and he can't even bench a plate. Like if somebody said if somebody sent you failing to even rep out, you know, a single rep, you would be devastated.
- [57:48]
A
I would be devastated.
- [57:49]
B
And you got to be ready for that.
- [57:51]
A
Yeah. Tyler, what do you think?
- [57:54]
C
I feel like I kind of disagree. Like when you see the headline from like the Onion, it's like, you know, it's like a joke. It's not real because like you trust the source it's coming from.
- [58:03]
A
Yeah.
- [58:04]
C
So it's like a video. Like right now you can just trust a video if it looks real, but soon it'll just be like, oh, it's from a person. I don't know, it's like non reputable and then you just don't trust it. And I don't think It's.
- [58:14]
B
Yeah, but there's so many instances when you don't have a choice. Like, let's say you get a DM from someone randomly and it's a video. It looks like a normal account, and then it's a picture of your dog being run over by a car. How's that gonna feel? You're gonna. And let's say that only happens once every three years. That's still pretty. That's still pretty rough.
- [58:38]
C
That's pretty bad. But I'm saying, like, in the, in the sense of like, persecuting someone for, like, if someone was stealing from Target, like, you're not gonna use the video if you don't know that the video is from the Target store. You know, I'm saying.
- [58:49]
B
Sure. I. Yeah, and I don't think. I just think he's talking about the broad implications of not being able to trust.
- [58:55]
A
Like, you couldn't trust Text, right? Yeah, we've never been able to trust text because anyone with a typewriter could go and type up Jordy Hayes accused of murder.
- [59:06]
B
The chat just noticed Tyler's Marlboro's reds on the desk. You gotta actually.
- [59:15]
A
Stop right there. Yeah, yeah. The cigarettes are AI generated. Yeah, but we already can't trust Text. Right. Because anyone can go and get a typewriter and type up Jordy Hayes accused of stealing at Target. Right. And publish it on the Internet or print it out and post it on a billboard. Right. Or they can stick it to a telephone pole. People have still built up like these trust networks in institutions. Maybe this.
- [59:47]
B
I know. I'm just saying, like, video could be traumatizing for people. Yeah, I'm not. I'm not really a trauma, bro.
- [59:55]
A
Yeah.
- [59:56]
B
But imagine seeing some terrible event and it looks photorealistic and now that's in your brain.
- [60:02]
A
Yeah, I, Yeah, I agree with that. There is going to need to be more control for the networks and more control on. I mean, there are, there are like, if you go to Netflix, you can set parental advisories and say, I don't want to see any R rated content. Netflix by default doesn't show any adult content. But you can go in further and say, I only want to see things that are G rated and PG rated. Right. And I would imagine that most of the platforms allow you to do that. They do a lot of it by themselves. So I think a lot of the platforms will handle that also. I don't know, people have just built up trust in all sorts of accounts, institutions, identities. In the idea that if I'm following this person or this organization or this website for years. And they've built up years and years of trust that they will. Only that they will call it out when they. When they're using an AI image or something like that. Like, I think that holds for longer than we.
- [60:57]
B
Yeah. And remember I watched my first movie, Mountain Gate.
- [61:01]
A
Yes.
- [61:02]
B
Or my second after Borat, and I thought the plot was like pretty weak. It was entirely around this idea that the world is actually collapsing because. And going. And extremely chaotic because. So I'm less worried about the. I just think content is going to be made that should never be made. And it's going to be dark. Yeah, it's just going to be dark. I'm not saying it's.
- [61:29]
A
I'm not saying, most importantly, there's going to be more of that dark content because you can.
- [61:33]
B
Because it's just so much easier.
- [61:35]
A
You could have made a after effects right now. You can open up Photoshop and draw man stabbing someone else. But. But it's gonna look like made in Ms.
- [61:43]
B
Paint. And you could have spent a million dollars to make a photorealistic CGI of an event, but nobody would ever do that.
- [61:50]
A
Yes.
- [61:51]
B
And now it'll cost nothing to do at scale. Anybody will be able to do it at all times. And it's just.
- [61:59]
A
Yeah, it's gonna take a while until it's actually free. Every Sora generation is very expensive to actually marshal all of the GPUs. Even if you're like, okay, I'm just like neo cloud and I need customers. I'm gonna let someone do some inference like there's gonna be pushback on. Oh, you're the one that's hosting this sketchy site that lets you generate celebrities killing each other or something. Like there's going to be pushback across the whole stack, the whole supply chain. And I do think it will take another couple of years for Sora 3 to pass the Uncanny Valley. It took CGI a long time. It'll probably take a few years. Then it'll take another year, a couple years for like per generation cost to fall. And then it'll take another couple years.
- [62:44]
B
A couple other examples gave in the chat says the Ukraine war was all over the X feed. Right?
- [62:49]
A
Yes.
- [62:50]
B
I personally don't follow any of those accounts. Videos would pop up. I'd say I'm not interested. Right. I just don't want. I don't want to be reading about technology through my friends and then suddenly see somebody get blown up by a drone. Right. Yet I saw it a bunch. Charlie Kirk assassination. Same thing. I was. We were live at nyse. I opened my laptop and I immediately saw the video. A lot of people saw the video that wouldn't have. Wouldn't have, you know, if you gave them a choice, do you want to see an assassination or not, they would have said no. But those people still saw it because of the way that content gets services.
- [63:28]
A
So yeah. Quick Update on the TikTok sale. Polymarket has the announcement. The market only resolves to yes if ByteDance announces their intention to sell TikTok. It has to come from ByteDance. It's at 48% by December 31st. By the end of the year it jumped to 84% and now is falling. So maybe there will be some retrading, maybe the sale will be delayed. But we will keep tabs on the TikTok situation. There are some more videos. Do we want to react to more Sora videos? There's one of Minecraft plus GTA 5 with Sora 2. Sort of cool to be able to mix prompts. This is a great example of what you were talking about. 1,000 likes, 60,000 views on X. Who knows what it did in the. Also underrated that Sora 2 can generate horizontal and vertical video. V03 is only horizontal. Vibes is only vertical. And so being able to do both, I think it's more of a creative tool than we think. But let's watch this Minecraft GTA 5 video then craft them all what you got. Boom. That's one way to mile los Santos.
- [64:43]
D
Let's mind some distance before they respawn.
- [64:45]
A
Man, the city ain't ready for. Man, that is rough. That is rough.
- [64:53]
B
We're never going to get that time back. I'm sorry to everyone.
- [64:57]
A
I'm sorry. I thought we were going to be. I thought it was like, yeah, we'll do like a little reaction stream. We'll watch a bunch of videos and kind of give our reactions. And my reaction to every single video is like not quite there. And this is different than the Dall E or the Studio Ghibli's. I feel like that crossed the chasm, it crossed the valley and this hasn't in some way.
- [65:21]
B
I would love to understand Open ChatGPT's most active user types because it feels like new technology, new product. It's probably heavily in that, that teenager to 40 year old range. And maybe Sora is their way of going after the Facebook user. Yeah, you know, the 50 plus. Yeah, you know, maybe retired.
- [65:47]
A
Well, if you want to generate some video for something that's not slop. You have a business use case head over to fall the generative media platform for developers. The world's best generative image, video and audio models all in one place. There are a bunch of other interesting ways that you can use these tools. You can develop and fine tune models with serverless GPUs and on demand clusters. Let's play this video from Vittorio of Mario boarding a SpaceX rocket. And Bonegpt says a Japanese lawyer just made generational wealth off of this clip. And the question I want you to answer, Jordi is does this make you want to spend more money with Mario or less? Is this a substitute for Mario content or is this an ad for the real deal? Let's play the video photo real. It looks like Mario. It does look like a SpaceX rocket.
- [66:56]
B
When you see examples of videos like that. I do believe that somebody could string enough of those together to keep a five year old heavily engaged.
- [67:04]
A
Oh, 100%, there's no question. So bearish, bullish. I don't know. What does that mean?
- [67:13]
B
Yeah, well Mario bought that, bought his Tesla before Elon went crazy.
- [67:17]
A
I mean if you are running, if you're running a kid's YouTube channel right now, you have got to be so excited, right? Because you can just generate so much more content that you're putting ads for actual toys in. And then you can have an affiliate link to go and buy the Mario stuffed animal or something and you're getting a cut of that. Like this feels like huge, huge unlock for that crowd. I don't know. What do you think? If you're running like a Coco Melon.
- [67:45]
B
Type channel, I just think it would all be completely competed away really quickly. And the real IP, the Ms. Rachel's, the. What's the other one that, what's his.
- [67:57]
A
Name, the kid toys. I don't know. I don't know that stuff. Yeah, stick to the Lindy Pixar stuff, the craft.
- [68:05]
B
David Holes hit the timeline.
- [68:08]
A
What'd he say?
- [68:09]
B
We need Lockheed Martin to build us a truth seeking missiles.
- [68:14]
A
He's like the most in the conversation right now and he's like just posting about Lockheed anyway.
- [68:20]
B
And Matt Mullenwig replies to attack the truth. I guess that's not what we need.
- [68:28]
A
Who knows. Sterling Crispin says infinite high quality media generated on a per user basis will erode our shared world model and narratives. This is what we were talking about earlier. With trust in institutions and crazy stuff happening on the Internet, expansive yet hyper niche culture bubbles around. Handfuls of people, the next generation may only have a few ideas they agree exist a few nations and corporate entities. Everything else will be content produced just for them, unseen by anyone else. This will be accelerated by high quality AI bots that act as their friends and long distance lovers. These bots will form lifelong relationships with individuals, getting to know them in deeper ways than any other person could. They'll create online social communities around an audience of just a few or one person, reducing people's need to share their experiences elsewhere. The music, shows, movies, memes and in jokes shared in these bot based social communities will be generated on demand and entirely unseen by other people and vice versa. Each person literally in their own world. And these content bot streams will be hyper focused to each individual's impulses and desires, giving them exactly what they want to see next. The ultimate fire hose of dopamine. It will be both extremely isolating and pleasurable in unimaginable ways. It's going to be ultimate internal war between.
- [69:42]
B
You know, he posted this in 2023.
- [69:44]
A
Wait, what? Wow.
- [69:47]
B
So Sterling quoted this post from 2023 and said reposting this now that Meta and OpenAI have shipped the product. This tweet is about the end game is is just in time per user generated content, leaving everyone hyper addicted and isolated with little to no shared cultural references between us.
- [70:03]
A
He's not a fan. This difference in ability to influence and control these systems.
- [70:07]
B
You should have an affiliate link for.
- [70:08]
A
Black pills between those shaped by these systems and those shaping them. Training and fine tuning the models is one thing, but using them versus being used by them is another. Are you the pig or are you the farmer? I suspect this skill will become the primary focus of higher education somewhere between prompt engineering, pattern recognition, and being a kind of neural network whisperer who's able to shape their reality through sheer willpower. Interesting. What do you think, Tyler?
- [70:35]
C
I think like maybe a steel man of like going against this is like you still see with like, even though there's like, you know, insane brain rot. Like there's so many videos that there's like infinite videos you can watch. There's still like, even within brain rot, there's like distinct characters or something that people still like to come back to because a lot of like the reason people like these things I think is because you can share them with your friends.
- [70:57]
A
Yeah.
- [70:57]
C
And like, and like if it's. If everything is completely customized to you, you can never share it because it will be like completely unintelligible.
- [71:04]
A
Yep.
- [71:05]
C
So there's some. Maybe there's still some kind of like shared, you know, cultural thing going on. Despite it being completely like custom to each person, there's still some. Something you can kind of share around the Hayes Paradox.
- [71:18]
A
I think. Yeah, Hayes paradox. I think about Maslow's hierarchy of needs. This idea that like I, I have content and I basically have a social network that's just for my family which is like the imessage group chat and the photo feed that goes in there. It's like you can share family photos with other people and extended family, but it's not, I'm not really like my kid photos don't really hit with other people. I'm not really in that world. But then I have, we have Jordy, we have in jokes with our group chat of guy friends basically that's a very tight circle. And then we have larger group chats that are maybe a couple hundred people. There's Dunbar's number, 150 people. Then there's the TVPN community and everyone on X. And then there's. When we get something that really hits, it goes viral. And so there's like these layers of abstraction that you can operate on that I don't know, they keep some sort of semblance of, of relevance. And then I don't know if there's anything that only I find interesting. Maybe that is the true Hayes paradox. Maybe we're going into a world of more Hayes paradox content.
- [72:27]
B
Yeah, the concept, obviously we're joking around a little bit here, but I propose the Hayes paradox which is the more funny you find something as an individual, the less likely that the masses will find. Find it funny. It's inversely correlated which is when you maybe have a funny idea for the post that. For a post that's cracking you up and no one else because it's too niche. It requires too many niche references.
- [72:54]
A
Yep. We saw this with, we were joking about like Ben Thompson being on vacation a particular day and there was some joke about that and like unless you're, you know, unless you understand our context and then also the Ben Thompson context, there was like 100 people that would get that joke, joke, whatever we posted. So it just didn't do well. But I did notice that a few people who liked it were like my friends, my real friends, people in real life like. And it would hit a certain level.
- [73:20]
B
Well, I like finding, I like finding a video or meme that I know I'm not going to send to anybody but you.
- [73:26]
A
Yes, yes. Or just, or just the six person group chat or just the family group chat or just the you know, the small niche community on X or the bigger broader community and stuff goes viral all over the Internet.
- [73:38]
B
Brian Lovin says Sora 2 looks impressive. What's even more impressive is how little I care to watch anything made using it. The Notion team is just firing shots.
- [73:47]
A
They're going hard today. Well, is Notion. They are notions on turbopuffer search, every byte serverless vector and full text search built from first principles and object storage. Fast 10x cheaper and extremely scalable. I knew I had to lay up there because it's such a logical partnership between those two companies.
- [74:05]
B
Alex Albert, who we had on the show yesterday, says over these next few years it's going to become more and more important that you resist letting slop consume you. Keep creating, keep learning, keep thinking.
- [74:15]
A
There is a wild, wild divide between the slop farm video apps and then the coding agents. No one's really leveling slop complaints about coding agents. And it's something about, I don't know, you don't just consume it or I.
- [74:36]
C
Think there are some people who are like Vibe coding apps just produce like slop code. But there is definitely still a difference.
- [74:42]
A
Yeah, I guess there's still so much of a human in the loop and it's such a. And most of the code that will be written with Claude 4.5 is doing stuff that I don't know. It's just like it's automating manual processes. It's not something that you're looking for crafted necessarily either.
- [75:02]
B
Tom Osman in the chat says hog engineering.
- [75:07]
A
Post engineering. Post hog engineering.
- [75:09]
B
Hog Engineering. It's the Slop Vibe coding app.
- [75:13]
A
Yeah, I don't know. There's a bunch of great memes in here. Parham has a photo of an image loading in images in ChatGPT. We need more compute to cure cancer, right? It's like a Bart Simpson or Lisa Simpson image. Loading.
- [75:29]
B
Casey Neistat had a good summary of Sora TikTok, but all AI but with real people.
- [75:38]
A
Yeah, very interesting. Can we play the Casey Neistat video of him reviewing a giant nicotine pouch size slider? I'm gonna try it. Oh my gosh.
- [75:48]
E
It's like a pillow in my lip.
- [75:50]
A
It's crazy. It allowed. This is a loud contact because did he make this? Did someone else make this with him? Did he upload his avatar or his, his cameo and allow that? And then in general, I mean. Yeah, you think PMI is on top of this? Like, you know, how is our IP being used in.
- [76:10]
B
Well, I don't think. I don't think PMI would mind that at all?
- [76:14]
A
No, not at all.
- [76:15]
B
And so there's this weird free ad for Zen.
- [76:18]
A
Totally. And they can't advertise on typical social media platforms. So yeah, very, very odd times, odd trade off. If you're a CEO of some legacy company with some IP with a brand and you want to know, well, you know, what has Coca Cola thought about Instagram over the past three years? It's been like, how do we get Coca Cola in every Instagram influencer's mailbox? Like, we need to be on that platform. What's our strategy? We need to buy ads, we need.
- [76:45]
B
To get our own ad.
- [76:46]
A
We need to get a private plane, put all the Coca Cola influencers on a plane and have them post about it. Right. Like we talked with Emily Sundberg about this. Like the name of the game has been like, get your brand on Instagram. And. And now there's going to be a question for a lot of brands, like, how do they get their brand mentioned in sora? That'll be an interesting thing. Maybe James from Profound will chime in on recommendations for strategy for brands, for creators, for folks that aren't just deciding whether or not to consume.
- [77:13]
C
Tyler, maybe an interesting idea is once these models kind of get out, at least to some extent in open source stuff where you can make your own model, I think it'd be cool if some like big company basically does free inference, but they fine tune it so that every video has their product in it. So Coca Cola releases some platform where it's free inference. You can do any video you want, but every single video has the bottle in it. I think that could be an interesting idea.
- [77:39]
A
Yeah, I'd like that. We could definitely.
- [77:41]
B
That would be potentially abused to degrees never before immediately.
- [77:46]
A
You know what people are going to go and do? They're going to try and degrade the brand that's providing that free inference immediately.
- [77:53]
B
Or Chan would.
- [77:54]
A
So it's like when 4chan tried to send Justin Bieber to North Korea. It's the best. The Internet is creative beyond all imagination, which is maybe the bull case here. I don't know.
- [78:05]
B
Zeke says not even excited about the new Claude or OpenAI stuff today. Kind of lost its hype. Wow. Benchmarks are 2% better. Wow. Shopify can now be used to shill products. Who cares, man? Where is the cool stuff? We hit AGI months ago and nothing functionally has changed since.
- [78:20]
A
Since, wow. We hit major AGI months ago.
- [78:25]
B
And that's what I was saying. You show a major advancement in the technology innovative stuff at the product level and people are sad about it.
- [78:36]
A
Yeah, very odd. Well, people aren't sad about linear because linear is a purpose built tool for playing and building products. Meet the system for modern software development, streamline issues, projects and product roadmaps and start building anti slop. Geiger Capital says I've seen enough reacting to soar 2. I am raising my probability that the entire AI bubble is going to produce nothing but Facebook slop and TV commercials for to 20%. That's crazy because of coding and knowledge retrieval. I don't know that produce nothing but is kind of crazy because like people are going to be doing deep research.
- [79:14]
B
Anthropic comes out of the last week and a half looking great.
- [79:19]
A
Yeah, the race is on. They apparently Anthropic had an image model and they decided not to ship it because of like, you know, concerns. It was also weird someone in the chat yesterday was asking about like, oh, why didn't we press Anthropic on the hunger strike person. I still need to dig into that story because I don't understand how they picked Anthropic. It's like of all the labs, Anthropic.
- [79:42]
B
Seems like definitely like the most studied the lab.
- [79:47]
A
Am I crazy? I gotta interview this person Because I gotta know how'd you pick that one? When we have Annie over here and Sora over there and Vibes over there and VO3. The Moving Fast memes are not the strongest in Anthropic. They're very carefully, like, how do we write better code? How do we write better code? I don't know. What do you think? You think it's reasonable to go protest outside of Anthropic?
- [80:14]
C
No, it kind of doesn't make any sense. But maybe, maybe the reasoning is like, maybe they think Anthropic is the most AGI pilled company. @ which point then you should be protesting them.
- [80:23]
B
Right.
- [80:23]
C
If like other companies are going into consumer and doing AI video, like it's like, oh, they're.
- [80:28]
A
Do you mean most AGI pilled or most AGI likely? Like, like if there's a AGI breakthrough, it will come from Anthropic because.
- [80:35]
C
Yeah, I think both.
- [80:36]
A
Yeah, okay. Yeah, yeah, yeah, yeah, yeah.
- [80:38]
C
It doesn't. Yeah, it is very strange considering like count like save.
- [80:41]
A
It does seem like they're the least reckless of all the labs.
- [80:43]
C
Yeah, definitely.
- [80:44]
A
I mean, I don't know. They. That book settlement so you know, RIP to the authors, but they got paid. So good news for the, for the authors. Sign in with Chat this is an interesting underrated feature that OpenAI shipped. So when you, when you, when you, when you download the Sora app, you're prompted to log in. With ChatGPT, you get portable identity for all of AI. It doesn't feel like they bootstrapped the Sora feed. There isn't enough content right now. But in theory, there is a world where if we were a couple months down the road and there was a lot of content in there. The Pulse feature from ChatGPT I think is fantastic. I haven't been a dau yet, but when I went on Pulse, it was pulling really interesting articles about CapEx and AI labs and even some production equipment that we're using here. I liked it. I felt like it was properly fine tuned on me. I didn't feel like my Sora feed was that it was all the same Altman stuff. But you could imagine a future where you've been using ChatGPT for a while. You just find out about Sora in December, you're at Thanksgiving or something in November and your cousin says, oh, are you using ChatGPT? And yeah, I use that every once in a while at work. And they're like, well, you should check out this new app. And then when you go and sign in with ChatGPT, the first Sora feed is quickly looking at your ChatGPT history, understanding that you have a dog, understanding that you have a particular type of dog, understanding that you have kids, understanding that you like to travel to France. And so it puts all that together and routes you to a particular corner of the Sora feed and gives you more customized content.
- [82:20]
B
And you can imagine Sora going pretty hard over Thanksgiving just because it is optimized for entertainment.
- [82:27]
A
Yeah, Tyler, I'll let you respond. And then I have another thing.
- [82:30]
C
Yeah, I was just gonna say, I think it's even more natural instead of just like putting you in some like, corner of the algorithm, that it just like actually just writes the prompts for the videos.
- [82:38]
A
Totally, yeah. You can already see that that's what it's doing in Pulse. It's clearly looking at what I'm searching and then writing new prompts for John, search for AI CapEx, write a new prompt for what he might be interested in, go find new sources, piece all that together, generate an article, even generate an AI image as a header for that article. And you could just generate a video that summarizes that or entertains based on that. It is a further step because it's basically just firing off like an O3 Pro or like a GPT5 Pro query for me. And those are great. And whereas these are clearly still in the uncanny valley, not really delivering value. But you could imagine that I open my feed and if I look on my YouTube, like my YouTube shorts feed is like Dylan Patel clips and like, it is like meat and potatoes. Like, I like the content that gets serviced to me. And you can imagine that Sora is generating that stuff and that would make the positive reception. Like, the more customization, the more personalization there is. People will love that once that happens and it moves out of this just like it's all demo clips of like Sam Altman. Like, people don't. The people will get sick of that. They'll want something special anyway. Numeral hq.com sales tax and autopilot. Five minutes per month on sales tax compliance. If you're selling something online, you got to get in, got to get.
- [83:59]
B
Let Numeral worry about it. Meanwhile, in the real economy.
- [84:04]
A
What you got?
- [84:04]
B
Things are bleak. Oh, yeah, things are bleak. ADP revised down September payrolls is now negative 32,000 worse since March of 2023. And certainly a barbell out there right now.
- [84:23]
A
Where are we?
- [84:24]
B
Oh, I just pulled that up randomly.
- [84:25]
A
Oh, okay. Okay.
- [84:26]
B
Yeah, you're not lost really far. But I just think it cannot be overstated how detached tech is from the AI hype cycle is from the real economy.
- [84:39]
A
That's true. It's true. We will see where it goes. I don't know. Where should we move to next in the real economy? There are a lot of different things in the real economy.
- [84:54]
B
Let's just get out of SORA 2 land.
- [84:56]
A
Okay, sure. Our partner Adeo was running some ads in London.
- [85:01]
B
There we go.
- [85:01]
A
Nicholas Sharp posted some photos and they're beautiful photos of these billboards. We love billboards. We love Addio. Customer relationship magic. Adeo is the AI native CRM that builds scales and grow your company to the next level. This is interesting. San Francisco got inspired and built the elevated Salesforce Park, a downtown, a true downtown jewel. This was apparently inspired by the line. Wait, no. What's that one called in Manhattan? I forget what it's called.
- [85:31]
B
See, my immediate reaction is, is this real? Because it looks incredibly beautiful.
- [85:37]
A
It is real.
- [85:37]
B
It is real.
- [85:38]
A
Salesforce Park. This is Benioff going off credit where credit is due.
- [85:43]
B
Yeah, Benioff cooked.
- [85:45]
A
He did.
- [85:45]
B
This is amazing. This is not what this looks like, you know, this is what the future would look. You know that?
- [85:54]
A
Yeah. Yeah. Well, have you been to the High Line in New York? Yes, the High Line's interesting. Interesting. It's over a mile long elevated park. Former rail train went on it and they decided to it was a former New York Central Railroad on the west side, brought in some people, built it and now it's just a place where people walk around a lot of fun. Some sort of nonprofit. In 1999, Friends of the High Line advocated for preservation and reuses of public space or an elevated park or greenway. Celebrity New Yorkers joined in on fundraising and support for the concept. Repurposing the railway into an urban park began in 2006 and opened in phases from 2009 to 2011 to 2014. The Spur, an extension of the High Line that originally connected with the Morgan General mail facility at 10th Avenue and 30th street, opened in 2019. The Moynihan Connector, extending east from the Spur, opened in 2023. So they're still building stuff out there.
- [86:59]
B
Finally a white pill on today's show. We can still build beautiful things.
- [87:03]
A
We can.
- [87:04]
B
We can get out there and build looks like the world if meme yes.
- [87:08]
A
In other news, Brett Adcock is assembling a capital formation team. What does this mean? My Companies have raised 4 billion in the last five years and now it's time to re accelerate. Brett Adcock is the founder of of Archer Aviation. Correct. And Figure Robotics.
- [87:27]
B
Yes.
- [87:28]
A
And this team will raise tens of billions to bring Sci Fi into the present. Reporting directly to him. DM's open, no remote candidates. It's in San Jose, California. So if you want to go allocate some capital, head over to his dm, raise capital.
- [87:42]
B
I don't know if you're going to be allocating.
- [87:43]
A
Lomaz, should we go back to Soar?
- [87:46]
B
No, we got to get out of Soar land.
- [87:48]
A
You're done with it.
- [87:49]
B
I'm sick of it.
- [87:49]
A
You're done with it? With it.
- [87:51]
B
Let's get into so Dylan Patel went on fantastic episode. Best like the best.
- [87:56]
A
Highly recommend.
- [87:57]
B
It really breaks down OpenAI, Anthropic, AMD, XAI, Oracle, Meta and Google. On OpenAI, Dylan said Super awesome on Anthropic, he said, I'm actually more optimistic on Anthropic than I am OpenAI. The revenue is accelerating way faster because of what they're focused on. Is more relevant to that $2 trillion software market versus OpenAI's split between software, AI for science, AI for consumer. They're also doing hardware, et cetera, et cetera. Anthropic is definitely executing on the software side better. Amd. I love them, but they're pretty mid. Dylan's the best. Xai. They're in real danger of not being able to raise capital again. Xai had to come out and say, we have plenty of demand. If you have to say that, that to. To a journalist that's digging, probably, you know, probably something to the story. Of course everyone's going to give Elon capital, but the scale of capital required for him to keep up, he can get the next bet. Xai can get to the next stage of compute. They won't have more compute than open AI. They don't have more compute than any individual company, Google, Meta, et cetera. But they will have the biggest individual data center. They'll have a very focused team. And what they do with that, they have to do something really big. Otherwise they will fall behind in the race. And Elon will not let that happen. He can subsidize and fund this round, but as rich as he is, he can't go to a 3 gigawatt data center unless he gets capital, which he can't do unless he gets revenue and fundraising again, you know, Xai has government contracts. They have romantic companions. They have romantic companions.
- [89:36]
A
They have.
- [89:36]
B
There are people that.
- [89:37]
A
On X, like Grok. Is this real, that whole thing? I do wonder. It's funny to reflect on the mood around the romantic companion launch. Ani and Valentine, people were definitely like, this is gonna. This is bad, but it'll be popular. Which is the same that we heard about Vibes and it's the same that we're hearing about Sora. Right. And I wonder, is it popular? We were kind of debating and one of the ideas that was kicking around was like, okay, maybe this is weird, maybe it's Black Mirror, maybe it's dystopian. Maybe it should be avoided as a recommendation if you want to live a good life. But it's possible that this generates a billion dollars in free cash flow because people love talking to their AI girlfriends and then they wind up buying their AI girlfriends virtual items that are 100% margin, like the CS Go skin that I purchased for myself, which I still enjoy. And you can imagine a situation where you fall in love with your AI companion and your spending not just. Just $200 a month for Grok 4 Heavy Pro, but you're spending. Yeah. Millions. And. And you get the whales that you see on.
- [90:44]
B
Which sounds crazy. People have actually done that.
- [90:46]
A
It happens on only fans and it happens on mobile games. Right. There are whales that spend a ton of money because they have a ton of money and they love the product has that happened? Like, I, I don't know. I haven't heard any stories about Ani like whales or ANI companions or, or ANI even getting a lot of traction. Like, it's not like we've seen hopped.
- [91:06]
B
On the charts in Japan.
- [91:08]
A
Yeah, but how are we tracking? And then also like it's rolled into so many other products. It's like there isn't like a standalone metric that we can be tracking because it's not a public company and we're not, it's not pure play. And so it's not like we can be tracking DAUs and time on site and average ARPU month by month or quarter by quarter at this point. Point. But it'll be interesting. I don't know. But I mean it does feel like if, if they can get that flywheel going, that's something that could, that could be a catalyst for Marshall marshaling the capital. Regardless of what you think about it morally, if there's, if there's demand for those, for that inference to chat with Ani and Valentine and that's something that OpenAI and Google and Meta are staying away from, well, that's an opportunity. So we'll see where it goes.
- [91:54]
B
Yeah.
- [91:55]
A
Oracle is going to make so much money if you believe OpenAI is successful. Meta. Dylan Patel thinks Meta's got the cards to potentially own it all. The only company in the world that has the full stack from good hardware. That's what Meta just showed with their glasses, with the screen, plus good models, plus capacity to serve them, plus the knowledge and know how around recommendation systems to know what content to put in front of the user. It's all four of these that you need, plus the capital. I think Meta is close to being the only company that can do that. Google. Dylan Patel was pretty bearish on Google two years ago, but he's super bullish. Google now, they're waking up on every front. They're taking the TPA use, they're selling them externally, they're taking their models and they're actually competitive on them and they're training much better and better. They're being aggressive on infrastructure investments. There's a lot of dysfunction throughout the company, but they do have the hardware business that they can pivot into this. They do have Android, YouTube and all these IPs. They have search that can come together when we turn to the next interface of consumer, but they also can dominate the professional sense too. And I think Google's welcome position to capture both markets or a meaningful share of both. Well Interesting.
- [93:01]
B
They didn't cover Microsoft in here.
- [93:05]
A
No and they also didn't cover Fin AI, the number one AI agent customer service, number one performance benchmarks, number one competitive bake offs, number one ranking on G2 but we'll have to get Dylan Patel's take on that. In other news, we mentioned it earlier but the the team over at Cognition rebuilt Devin For Claude Sonnet 4.5 I don't think we'll have full time to dig into this whole write up but it's interesting that they dropped in Sonnet 4.5 into the old Devon and they did see a higher score but it took more time and so they rewrote it and they got both an even higher score with each even more time saved which is obviously valuable. They said the new version is twice as fast and 12% better on our junior developer evals and it's now available in Agent Preview and they're keeping the old Devon alive. It remains available.
- [94:06]
B
We didn't get into it a ton, but Alex in Anthropic yesterday was very excited about 4 or 5's potential on the agent front.
- [94:13]
A
Yeah, yeah.
- [94:15]
B
Nick Carter has a good little deep dive. I was going to call it a mid dive.
- [94:22]
A
Not it's pretty deep. You want to hop into this? We got a guest join in six minutes but we can try and rip through this.
- [94:31]
B
We can come back to it.
- [94:32]
A
I'd like to have another show to talk about this, but let's go through it. It's an interesting question.
- [94:36]
B
Few points here please. He says the stablecoin duopoly is ending. He's obviously referencing USDC and USDT. Circle Equity is currently worth 30 billion billion. Tether's Topco is reportedly raising capital at 500 billion which would become if they can get that done before OpenAI, that would be the most valuable private company in the world.
- [95:00]
A
Isn't that more than SpaceX?
- [95:02]
B
More than SpaceX? More than open AI?
- [95:07]
A
Why do bank robbers rob banks?
- [95:08]
B
That's where the money is tender at 500. But yeah, again this I think the most profitable per company in the world on a per employee basis.
- [95:17]
A
Extremely profitable.
- [95:18]
B
Together the two stablecoins boast 245 billion in supply, or about 85% of the stablecoin market. Since inception, only Tether and Circle have maintained meaningful market share. No one else has even come close. Dai peaked at 10 billion in early 2022. Terror's doom UST did surge to 18 billion in 2022, but accounted for only 10% of the market market and that was only Briefly, the most ambitious attempt to dethrone tether and circle was Binance's BUSD, which peaked at 23 billion. So he goes into some of the new market dynamics. 1 We had Zach on from Bridge yesterday talking about how anybody can now sort of quickly originate stablecoins. So expect to see a lot more banks do this. They're not going to just give up the yield that they currently get from dollars. So yeah, I'm interested to see what the legacy banks do, credit unions, et cetera, how quickly they'll move here because they're certainly all paying attention.
- [96:22]
A
Yeah, I still don't fully understand the full chain of value accrual because if you need cross chain interactions, there could be a tax there as you're trying to get from one stable coin to another. Why is this market not monopolistic? Why hasn't one just run away with it in the way that Bitcoin has with that digital gold narrative? Why are we seeing a proliferation of stablecoins? It is an interesting topic to dig into, but anything else you want to cover from the.
- [96:56]
B
No. Do we have our first guest yet? Not quite.
- [96:58]
A
Couple minutes. How'd you sleep last night? I got a new phone. I wasn't logged into my 8 Sleep app. I should have been because I got a 98898 baby 79 last night I got an 876 hours. So I still beat you. Play my song, Jordy Haydens.
- [97:16]
B
There you go.
- [97:16]
A
Let's go. Thank you.
- [97:18]
B
Buco Capital shares, Amazon, Google, Microsoft, Meta and Oracle are now spending 60% of their operating cash flow on CapEx.
- [97:26]
A
It's time to build, baby.
- [97:27]
B
It's expected to go higher. You think they'll. Where do you think they'll get.
- [97:33]
A
Yeah, why not spend 100% of it? Why not spend 100% of it and then go raise debt and then spend more? Like, I don't know, what's the efficient frontier? If the capex is valuable, if it returns, if it returns, profits go higher. Liz says it's going to go even higher.
- [97:48]
B
Well, we've talked about it, we've talked about it. But if you have the leaders of these companies that don't care about roi, that could take us into quite dangerous territory potentially.
- [98:01]
A
Yeah. What do you think, Tyler?
- [98:02]
C
I just have some other news. Thinking Machines has released a product. I mean it's not, it's not really a product. It's an, it's an API, it's an.
- [98:09]
A
AI video app and it's adult themed. They took the reins off Thinking Machines. What do you think the machines are.
- [98:19]
C
Thinking about thinking about, but it's basically, it's an API. You can basically do like training. They have a couple supported models. I think it's just Quen and Llama, a bunch of those series. But basically it's just like we handle the infrastructure, you just give us the data and then you can fine tune easily.
- [98:40]
A
I remember the narrative around thinking machines being like it's RL for business. So if I want the fine tuned version of the most cutting edge open source LLM, thinking machines will allow me the infrastructure to do that, to fine tune, to reinforcement, learn on top of my data.
- [98:59]
C
Yeah, I don't know if it's. Oh, I think it actually is rl. I need to read through the whole announcement. They haven't released a ton of it, but yeah, pretty interesting. I mean it seems like kind of very competitive with. I think OpenAI offers the same thing. They do fine tuning of the big.
- [99:13]
A
Models, but maybe there's value in focusing and being known for that specifically and then having a really clear device. I mean OpenAI with that product. Obviously they, if you're on the enterprise plan, they won't train on your data, but there's still like this, you know, you're dealing with a, with a shoggoth with multiple product lines. You go to mira, maybe you get a more focused team that's really, you know, not AGI built at all, I guess. Right. Does this update your timelines, your P doom?
- [99:40]
C
I don't think so. I mean this is kind of expected of what they were like they kind of said that they were going to do RL for business. They're doing fine tunes.
- [99:46]
A
This was our riff. What did Mira say? You know, everyone asks what did Ilya see before he left? But what did Mira see? Maybe she saw business value creation. And that's what I like. Let's hear it.
- [99:59]
B
Let's hear value creating.
- [100:00]
A
Value creation, we love it. Fine tuning, it's essential. Who else is going to do it? $6.2 million fundraise for no ZOMO. Why I cannot pronounce that. Arlen.
- [100:15]
B
We interviewed Arlen.
- [100:17]
A
Arlen, that's right. At YCDEMO Day.
- [100:19]
B
YCDEMO Day.
- [100:20]
A
Amazing. Well, Congratulations to Arlen. CRVs in box groups and local global, some good names.
- [100:26]
B
Hit that gong for Arlen. Congratulations on the progress.
- [100:31]
A
Congratulations. There's some other cool stuff. Mr. Haffner. Here Donnajar is announcing Dreamer 4, an agent that learns to solve complex control tasks entirely inside of its scalable world model. This is what Tyler was Alluding to with some of the value of world models besides just playing Minecraft, actually teaching, just serving as a reinforcement learning environment. And then there's a bunch of other stuff in here. Intern says hey man, just wanted to reach out and say how much I loved how much you drank at the networking event last night. You're not going to get that on Sora. You got to go.
- [101:19]
B
You're also not going to get that from Grok, apparently.
- [101:22]
A
No, not yet. There's also a wonderful ranking of energy drinks here. Notably missing is Andrew Huberman's Yer Mateena, of course. But Goth says the gold can is elite. If you don't believe me or don't get it, I don't have time to convince you. Sorry. And that's the caffeine free diet Coke. I, I do the caffeinated diet Coke but.
- [101:45]
B
Interesting.
- [101:46]
A
I don't even think this post is claiming that the gold can Diet Coke, the caffeine free and sugar free diet Coke, it's just Forever chemicals apparently is the drink of men who are destined to find eras, change history and who.
- [102:00]
B
Will be remembered for moment finally pure forever chemical, carbonated forever.
- [102:06]
A
Well, if you're bullish on Coke or bearish on Coke, head over to public.com investing for those that take it seriously. They got multi asset investing, industry leading yields and they're trusted by millions. We have our first guest of the show, Rhys Chowdhury from Concept Ventures.
- [102:22]
B
Welcome to the show.
- [102:23]
A
Reece, how are you doing?
- [102:25]
B
Good, good.
- [102:26]
D
How are you guys? Big fan of the show. Thanks, thanks for having me on. I really appreciate it.
- [102:29]
A
Thanks for hopping on the show. Give us the news. Break it down for us. What's happening today in your world.
- [102:35]
D
So today or yesterday we launched our latest fund. It's $88 million. Yes, you can ring the gong.
- [102:46]
A
But.
- [102:46]
D
We have a bell in our office and we many years ago and people, people would always ring it as well. So I thought that's. I love, I love the analogy.
- [102:55]
A
So let's go bigger.
- [102:56]
B
Let's get you a Liberty Bell.
- [102:58]
A
I don't, I just. From the way you describe the bell, I can tell it's not taller than you and it's got to be huge. I want a massive, well, 88 million.
- [103:06]
B
You have the management fee fees to have a bell. Guy just stands there ready to.
- [103:10]
A
People think 2 and 20. You get 2% for the figure. You don't get 2%. You get 2% over 10 years, 20%. You could spend $2 million on a bell tomorrow. And I think you should.
- [103:20]
D
That's the first order that's going on Amazon.
- [103:24]
A
Give us the pitch for the fun, give us the strategy. A bit of a contrarian take. You're hunting for value in Europe.
- [103:30]
D
Yeah, we are hunting for value for us. Obviously we're based in Europe, we're based in London. So contrarian to many guests you may be having Australian. Awesome that you've got some European founders on of course, and Carl and Victor. I saw that again, two contrarians there as well. And so we, we really focus on that earliest stage of investing, the concept stage or that pre seed stage. There's not many pre seed funds in Europe and you know we've got the largest as of, as of yesterday. And you know we're focused exclusively on investing in the people themselves. So we go very, very deep in, in you know everything about people from like their childhood to the how they met their co founder and you know what they spike in and you know, do they have like, are they like a chess champion? And we, you know, we break it down into kind of four kind of archetypes which most unicorn founders specialize in. So we have no sectors basically. So either you're the first, first time founder that maybe worked at big tech. You had a first time founder that you've you know, been at a unicorn. Like you know, we've got a company that spun out of Synthesia, a company called Anam which is backed by Redpoint, or you're a second time founder or a PhD spin out. And so like within the team we all kind of specialize in one group of people and that really gives us an insight in trades. Having like worked with I don't know, second time founders. Are they learning all their lessons before? So we often take these very contrarian bets in really rogue sectors from high frequency trading to robotics to commodities marketplaces.
- [105:00]
B
Let's give it up for high frequency trading. Like question, question for you. How does having less competition at pre seed? Obviously every stage is hyper competitive now, at least in, in the US Basically.
- [105:14]
A
Every pre seed allocator that we know is moving up the stack in the US at least that's what I can tell it feels like. I'm hearing that from so many American pre seed investors who are like well.
- [105:25]
B
People are moving down the stack too.
- [105:26]
A
15 years ago it was super uncompetitive. I was only one to explain what a pre seed firm was. How I'm doing series Bs.
- [105:32]
B
Well, well it's also even if you're still doing pre seed you oftentimes have you, you know Precede, lead. If you really want to lead a precede round, which maybe call it a 500k check, you often have to make a decision in 24 hours. Like, you need to move quick, you need to move oftentimes faster than you might be comfortable. And so having a little less competition at that stage, I don't know, does that change your, your process at all? Do you, are you, do you have more time to get to know founders? Are you still feeling, you know, that, that urgency?
- [106:04]
D
Yeah, look, I think, I think in Europe we're traditionally maybe a little bit more conservative than the U.S. and you know, we love, you know, when we started concept, everyone was like, you're crazy. Literally, you're crazy for doing this because, you know, you really needed a more, you know, market or a product or, you know, and in Europe, you know, traditionally we all come from outside. We're not insiders to VC at all. Like, we haven't come from those top names. We all the team are not, not from that background. And I think that gives us a real edge in kind of like evaluating, but also taking that risk. And I think what we see if there's no product or markets typically to go after the underwriting process is actually a lot quicker because you're literally evaluating people.
- [106:50]
A
People.
- [106:51]
D
So typically our process takes like, we call it like five hour energy because it's like, you know, five hours of founders time. And you know, we go through our process and we're looking for very specific traits and things in those individuals. And when we get that, we were happy to lead. We lead 90 of our rounds. We don't care what other people do. You know, we don't care about fomo. You know, it's a very kind of set deal. Typically we're investing a million dollars at kind of the star and you know, we don't need anybody else to kind of make that decision. And I think, think, you know, generally Europe needs, needs more of that. I mean, you had Matti on the show.
- [107:23]
A
Yeah.
- [107:23]
D
You know, obviously that was, that was our first investment out of the last fund.
- [107:26]
A
Oh, let's go. That's amazing. What a banger.
- [107:30]
B
Jordy, you got a. Yeah, it's just every, if you're getting a fund off the ground, just back one found.
- [107:35]
A
Yeah, just do 10 more though.
- [107:38]
B
@ least, you know, you know, one time a fund, you'll probably be pretty, in a pretty good spot. Where's your, what's your LP base look like for the new fund? Are we, are we like, you know, what, what, what's the mix European.
- [107:51]
A
Do they lean more Italian supercar owners or British supercar owners?
- [107:57]
D
No, we're only allowed to talk about golf now, given, Given the last. You know, obviously Europe's won something, so you got to focus on that. But no, but, no. To answer your question, we've actually got most of the institutions we onboarded in this fund. I think 80% of them are us, actually.
- [108:14]
A
Okay, interesting.
- [108:15]
B
Let's go. Keep it in the face.
- [108:17]
A
Thank you for.
- [108:19]
B
For our allocators.
- [108:21]
A
Secret American. Secret American over there doing work for us. Okay, thanks for supporting Lightning round.
- [108:28]
B
Question, please.
- [108:30]
A
I want to hear about sourcing. You have three places where you can go to source the next great pre seed founder. You can go to a Cambridge Student Union meetup, meet young new grads. You could go to a pub outside of DeepMind after they just failed a training run. You can try and poach someone from there. Or you can go to Monaco during F1. Which one are you picking and why?
- [108:55]
D
I'm taking the DeepMind. DeepMind, the pub.
- [108:58]
A
Hang out at the bar. Get the disgruntled DeepMind. Oh, the training run just failed. I want to start a company. I got to get out of this bureaucracy.
- [109:05]
D
Yeah, I think, I'll tell you that. In the last one, I think we invested in three, three, three ex Palantir people. So we love a good four deployed engineer. And so I think that's something, you know, we, we think that talent generally gravitates to places that have amazing cultures. And we spend a lot of time looking in the cultures of those companies, particularly in Europe. And I think you've had this one wave of, you know, the UI pass the Spotify. You're at the Kleiner ipo, obviously. And that was kind of like one wave which has happened and like, you know, gone public. And now you're having this new ways of. You've got Carl and Victor on the show, you've got Mattie and there's others like Granola in London. And I think these are like people are just going bigger, right? They're just thinking bigger because they're learning more from these kind of repeat successes. And that's coming from all parts of the ecosystem. If you're working in a company out of a company, you're seeing these success stories. And I think there's a lot more to come, personally. But having boots on the ground and that partnership with our American LPs, I find that typically the American firms eventually get in, right? Like every ipo, every exit, you'll see you win in the end. They're not always comfortable making that bet. So I think that's where we can play a role of making that conviction early and going forward.
- [110:25]
A
Fantastic.
- [110:26]
B
I'm super bullish on Congratulations. If I was an American multi stage vc be all over trying to take down a good piece of the fund. And yeah, I'm super, super excited that she accompanies you back.
- [110:39]
A
Thanks so much for hopping on the show.
- [110:41]
D
Thanks. Thanks for having us, guy.
- [110:42]
B
Cheers, Reece. Congrats.
- [110:43]
A
Bye. Cheers. I want to pull up this post by Sherwood. Sherwood says, may I suggest a different tagline? You can feel the diff posting an image of a graphite billboard. Graphite's of course a sponsor on this show. And the the billboard is like not even using all the billboard space. It's so over minus and then plus is we're so back.
- [111:07]
B
It's clean though.
- [111:08]
A
It got 5,000 likes. And this is the interesting long tail of less flashy out of home ads that wind up still going viral. That's the benefit. That's why you got to get on adquick.com out of home advertising made easy and measurable. Say goodbye to the headaches of out of home advertising. Only Ad quick combines technology out of home expertise, expertise and data to enable efficient seamless ad buying across the globe.
- [111:28]
B
Do we have our next guest here?
- [111:29]
A
We have our next guest. You can see him right there on the big word.
- [111:33]
B
There we go.
- [111:34]
A
Welcome to the show.
- [111:35]
B
Welcome to the show.
- [111:36]
A
How are you doing? Jim? Good to meet you. Hello, John.
- [111:38]
F
Hello, Yordi. Doing very well over here. How about yourselves?
- [111:41]
A
We're doing fantastically today. Bit of crazy day with Sora too, taking over the news. So we spent a lot of time talking about that. But what's new in your world?
- [111:50]
F
Yeah, well, we raised a series B which is of course the big news in our world over here.
- [111:55]
A
How much? How much? Yeah.
- [111:57]
F
So we raised bit north of $50 million to build new AI agents for AI factories.
- [112:03]
B
AI agents for AI factories. Let's go. We've been waiting for this. What does that mean?
- [112:09]
A
Yes, Jensen says AI factory is a token factory. It's just a data center. Or do you mean AI factory as there's robots walking around making iPhones, phone cases or something like that?
- [112:18]
F
Yeah, we mean it in the Jensen sense.
- [112:20]
A
Right.
- [112:20]
F
So purpose built mega scale data centers that exist to convert electricity into tokens.
- [112:27]
A
Right.
- [112:28]
F
They exist to generate intelligence. That's what it means. So concretely we make AI agents that are capable of autonomously operating and optimizing physical mission critical infrastructure.
- [112:40]
A
What are the top things that kind of break when you're running a large data center that normally a human has to step into. Normally it's time consuming. Now they're leveraging an AI agent to maybe just work while they sleep or kick off a problem solution so they can have the agent go and work and solve some of the problem for them. What are some examples of problems that you're solving concretely? Yeah, yeah.
- [113:07]
F
So you have to pardon me if I get a little bit nerdy. I'm a mechanical engineer.
- [113:10]
A
Go to town.
- [113:11]
B
Get nerdier.
- [113:12]
A
Go to town.
- [113:14]
E
Yeah.
- [113:14]
F
So I think one of the important things to realize is that AI factories today look nothing like traditional data centers.
- [113:20]
A
Right.
- [113:20]
F
Like when I started in the industry, I joined a much smaller Google back in 2010. Right. And our concept of a large data center were these 30 megawatt modular data centers. And at the time we thought to ourselves, who's ever going to use 30 megawatts of compute?
- [113:34]
A
Right.
- [113:35]
F
And now obviously the industry has changed a lot. Now we talk about gigawatt scale data centers that are purpose built for AI workloads, I. E. AI factories. The most important thing to realize about AI factories is that their orders of magnitude more complex. There's a lot of stuff in there to orchestrate. A lot of pumps, a lot of chillers, a lot of liquid cooling CDUs.
- [113:54]
A
Right.
- [113:54]
F
And they have very non linear interactions with each other. So what our AI agents do is they have a global view of everything that's happening across the AI factory. They can absorb tens of thousands, hundreds of thousands of sensor trends in real time. And they act like a general on the battlefield. They're issuing AI generated command signals that are sent to the local control system for automatic implementation execution. So it's really what we would call AI enabled supervisory control for these very large AI factories.
- [114:23]
B
Where do you think the. So if you're doing the orchestration and management, where is the physical work going to come from? From is that being met by somebody that's just on call in a data center today? Do you think that robots will eventually do that work? Data centers are great and that they're sort of a flat controlled environment. It's kind of a good setup for not necessarily a humanoid robot, but just a robot on wheels that can go around and make adjustments. But what's your kind of vision for once your agents decide what needs to be done within the system? If it can't be done purely with software, how does it get done physically?
- [115:04]
A
Yeah, totally.
- [115:04]
F
It's a good question. So you know to be very clear, the agents are not capable of replacing like the physical labor that needs to happen. So things like turning wrenches, right? Taking a chiller offline, cleaning it, that stuff that, you know, you need humans for today, right? Maybe eventually we'll see robots, right, taking care of that sort of stuff. But what we do is more of the software level orchestration of all the, of all the equipment, especially the mechanical cooling system, right, that happen in these large factories. But I think, you know, one of the important things to also recognize is that the industry is growing so rapidly right now that there is just a massive, massive, massive shortage of skilled labor expertise in the industry.
- [115:47]
A
Right.
- [115:47]
F
Jensen talks about how like electricians and plumbers are going to be two of the hottest trades moving forward because of the scale of the factory buildout in industry that I come from. From you see a lot of folks with white hair and you don't see a lot of folks our age in the industry. So there is a massive shortage. A lot of people are retiring. They're taking knowledge with them. Our hope is that these AI agents, in addition to providing optimal functionality, these AI factories can also augment the labor force. We communicate our AI agents as virtual plant operators. They're virtual members of the operations staff. You can't see them or hear them, but just like me, they're permanently working from home and doing a lot of stuff in the background.
- [116:27]
A
Can you explain to me what the software stack looks like for all the different pieces of a data center? We talked to a lot of folks who are building software for traditional manufacturing and they talk about Siemens has a control system that has some sort of archaic API and then they write a wrapper for that API and then they can create a dashboard or they might be able to trigger some things on that. Does a chiller have an API that can be accessed over Ethernet or something like, how do all these systems tie together? How open is it? Whenever you're building an AI agent, people will compute cursor, but cursor is a fork of VS code, open source software. And then all the code is stored in GitHub. Very easy to scrape in and download and work on. I imagine some of these tools are either closed source because. Because that's part of the business strategy of that industrial company, or they're just so old that they haven't gotten around to actually writing an API. So walk me through all that.
- [117:30]
F
Yeah, yeah, it's a great question. So frankly, the data science industry hasn't seen a lot of innovation a really long time, right? Until Nvidia came out with the latest generation of chips. And now all of a sudden there's real innovation, real creativity in the data center industry again, before is really just copy and pasting the same old stuff over and over again. Right. A large part of that is the closed ecosystem effect that you were just describing, right. If you look at what the traditional incumbents are doing, like the Siemens, the Schneider Electrics, whatever.
- [118:00]
A
Right?
- [118:01]
F
You know? Yes, right. There are protocols that exist like Modbus or you know, or OPC, UA, MQTT, whatever, protocols that exist that act as the APIs.
- [118:11]
A
Right.
- [118:12]
F
That enable these large mechanical equipment to coordinate with each other. However, that tends to exist within a walled garden. So each of these major automation vendors, like the Schneider Electrics of the world, they make it very difficult for you to integrate.
- [118:26]
B
Right.
- [118:26]
F
There's no such thing as an easy way for third parties to access the data and analytics that are coming off of these mission critical systems. So a large part of what we do is integrating with a large variety of the these building management systems or these SCADA systems to ingest that data and make intelligent decisions off of it.
- [118:45]
A
This seems like something that you can't just pick up from a couple of Google searches. Like what were you doing before? How did you actually learn that this problem existed? How did you get up to speed on all these different systems?
- [118:56]
F
Yeah, so in 2010 I was a bright eyed and bushy tailed college grad joining Google. And back then Google was still small enough and young enough that new college grads had outrageous amounts of responsibility.
- [119:09]
A
Right.
- [119:09]
F
So I was fortunate enough that I actually helped design some of Google's first large cooling systems that go into their data centers. Then in 2013, I became the person responsible for Google's energy efficiency programs for its data centers. The primary metric is PV or power usage effectiveness. Even back then Google was already consuming billions of dollars a year in electricity. So it is in our incentive to optimize the energy cost. And that's grown by orders of magnitude since then. In 2016, something called AlphaGo happened. If you remember, that's kind of what made DeepMind famous. My co founder Veda was one of the AlphaGo engineers. I saw what happened and I thought to myself, if we can create AI agents that beat the smartest, most intelligent humans at complex games like go, then surely we can teach these same AI agents to play other games like let's reduce the energy consumption of Google's data science power management. So yeah, so I reached out to a guy named Mustafa Suleiman who was the co founder of DeepMind.
- [120:09]
A
Let's go. Chad is loving you by the way.
- [120:12]
B
This guy's crack.
- [120:13]
A
This guy rips. Everybody loves you. Thank you for joining the stream.
- [120:17]
B
Almost like your entire life was leading up to this moment.
- [120:21]
A
Overnight success.
- [120:23]
F
So it was, I mean it was just a cold email. I was totally not expecting anything other than it. But I pitched this idea of like what if we take DeepMind's reinforcement learning agents and use them to control big ass infrastructure like Google's data centers. So we did it. It worked. It worked surprisingly well. I actually ended up joining DeepMind. For three years I led a business vertical called DeepMind Energy.
- [120:45]
A
Right.
- [120:45]
F
We applied reinforcement learning for all sorts of things like commercial building, hvc. Then my co founders and I realized reinforcement learning is actually really good at industrial scale control and optimization. So we started.
- [120:58]
A
Phaedra. Well, congratulations. What a run. Very excited for you and glad to see you. Have a nice series B to go.
- [121:06]
B
I got a feeling you'll be back on the show very soon and not too long. So congrats on all the progress. Great to meet you.
- [121:14]
A
Have a great rest of your day. Thank you very much.
- [121:15]
F
Have an awesome day you guys.
- [121:16]
A
We'll talk to you soon. Well, if you want to optimize your wrist, head over to getbezel.com your bezel concierge is available now to source you any watch on the planet. Seriously, any watch. We have our next guest from Moonlake in the TVPN ultra dome coming in from the Restream waiting room.
- [121:35]
B
Welcome to the show.
- [121:36]
A
How are you doing? Welcome to the stream.
- [121:39]
G
What's up?
- [121:40]
A
Thanks.
- [121:40]
B
What up?
- [121:41]
A
Not too much welcome. I. We got to be fast about this because I want reactions to all the stuff that's happening in the timeline but kick us off with an introduction on yourself, the company.
- [121:52]
G
Yeah, I'm fan Insai, tend to go by Sun. I'm the founder and CEO of Moonlake AI. We're really spun out of Stanford AI lab a couple around six months ago. We started this company really to build AI that can generate real time interactive content, think simulations and games.
- [122:08]
A
Yep. So yeah, I mean what are your reactions to soar2vibes? There's all these different paths that people are testing. Is it a tool? Then there's also this interesting take about maybe we won't even be consuming. It's all reinforcement learning data for robotics. How are you viewing the market play out just in AI generated content, video, virtual worlds, all that.
- [122:35]
G
Yeah. I think it's very apparent that there's a lot of people going through this space. And it's really just because the potential is unbounded.
- [122:42]
A
Right.
- [122:42]
G
Being able to model the world, let alone any other virtual reality is going to do the biggest, the next biggest wave in AI. No doubt. Now what we're doing is actually very different.
- [122:54]
B
Yeah, I was going to ask. Yeah. A lot of people playing an image and video. We haven't gotten a lot of pitches around creating sort of these interactive media. Something more like you could imagine a mobile game built on Moonlake. Am I going in the right direction or Totally.
- [123:13]
A
Yeah.
- [123:13]
G
You can imagine games being built, simulations and really the main difference when it comes to simulations and games compared to videos or images is the interactivity.
- [123:24]
E
Right.
- [123:25]
G
User outcomes change the trajectory of the world or the game or the environment that you're acting in. And that's what really differentiates us from videos or image generation companies.
- [123:36]
A
How do you. Yeah. How do you think about tool use? With ChatGPT just the knowledge retrieval app, we went from GPT 3.5, it was okay. GPT 4 was great. Then we added tool use and there was a moment when it was like, just keep scaling it up and it'll learn the answer to every math equation. And then the engineers OpenAI just realized like wait, let's just teach it Python and it'll just do the math in Python and. And then you'll get the right answer and you don't need to memorize everything into the weights of the actual underlying model. And when I see these virtual worlds, these generated worlds, I feel like we are in the GPT3 moment and we haven't yet given them a database to store your inventory or given them the ability to even a notepad to write down what happened in act one of the game. How do you think tool use use comes into these generative worlds and will it track with what's happened in the ChatGPT app?
- [124:37]
G
I think TOL use is exactly the right paradigm that we should go about thinking about creating virtual worlds. Computer graphics has been around for decades, if not hundreds of years and we have the tools that enable us us to create the highest fidelity virtual worlds already. Look at all the movies, look at gta. It's just that these things are hugely expensive and manual to create today like GTA costs $2 billion to develop.
- [125:10]
A
Right.
- [125:12]
G
So really what we believe is that we should use reasoning models to reason about the representation, the tools to use the models to use use and consolidate all these things into a platform that allows anybody to be able to vibe code their interactive worlds. These Are the things that are historically just barred by just complicated tools and domain expertise. But it doesn't have to be that way.
- [125:38]
A
Do you think?
- [125:40]
B
How do you feel the legacy game studios have been reacting to the opportunity in generative AI so far?
- [125:48]
A
Have they been?
- [125:49]
B
Yeah, like have they been. I don't follow gaming super closely but for example, like the EA take private, you could imagine they start huge opportunity. For sure, they have a lot of ip, they could, if they're private, they could have the potential to take a longer term view on AI and try to reinvent themselves. But I'm curious how you view it.
- [126:09]
G
Yeah, no, all the studios that we talked to are stoked about it, right. So we're actually talking to to a team that owns F1 and they were talking to us about the possibility of being able to create racing games that Ghost races with F1 cars in real time. And I think the way we should be thinking about this is that what we're building at Moonlake allows things like so many more things to be built or things that are historically cut or are just made impossible due to the time and resource that it entails to create these things.
- [126:49]
A
Right.
- [126:50]
G
And so really all the studios and folks that we work with on the gaming side are stoked about what we're building and a lot of them are looking to collaborate with us.
- [127:02]
A
What's most rate limiting to you right now? Is there enough data out there? Is there enough compute out there? Is it just. Do you need to just go and have really smart scientists think and think and think and come up with an entirely new technical paper and algorithmic innovation? What is on your critical path?
- [127:25]
G
There are definitely a couple of algorithmic breakthroughs that we are still working on. We have some of the best scientists here to work on those problems. But, but really even without just significant breakthrough, I'd say the technology today, if you just consolidate them with reasoning models, the upside is already unbounded. Look at Roblox. It's $100 billion business with really more traditional graphics approach compared to all the generative world model technologies techniques.
- [128:03]
A
I don't know if you're familiar with this Ben Thompson take, but he was saying that like ChatGPT is so successful that at this point OpenAI could stop training foundation models and it could be clawed under the hood and it would still be a fantastically amazing business because ChatGPT has won the application layer in knowledge retrieval. It's the app on everyone's phone, right? Like do you see yourself becoming an application layer company staying in a Foundation model, layer company, like doing some sort of blend, like how do you see your business developing?
- [128:34]
G
Yeah, we are actually an applicational company. So we are going to build products that allows anybody to be able to create interactive experiences.
- [128:42]
A
Yeah, right.
- [128:43]
G
The way we go about thinking about oh, whether we're product or model company is really that product is first principle. But a lot of the times, given what we're building is at the frontier. We need to do a little bit more research and some model training in order to power the best product experience. So the analogy that I like to make is Cursor.
- [129:03]
A
Right.
- [129:03]
G
Cursor didn't start off creating their own foundation model, but they had their tab completion model. They had these smaller modules that come together to form an amazing product that people love.
- [129:14]
A
Right.
- [129:14]
G
And that's what we're doing.
- [129:15]
A
How crazy is the inference cost right now? It feels like Sam ALTMAN Setting the GPUs on fire with Sora too. Obviously, per token, inference cost is Declining by by 1000x every couple years or something like that. But for all the demos that I've seen in world generation, they've all been somewhat limited. It hasn't gone super viral. I haven't seen, oh yeah, there's 100 million DAUs on Genie 3. It's like Genie 3 isn't even available. And so how GPU constrained, how crazy is it to actually go and release one of these things into the wild? Like, and what's the path for actually bringing down the cost of inference here?
- [129:57]
G
Yeah, so I would say inference is definitely something that we think a lot about and the computer definitely something that we use a lot for. So if you're a GPU vendor, I like to talk okay. But I would say our unique approach of using code to represent the world provides persistency and it consolidates all the tools such that we can use client side compute to calculate next state of the world. That actually allows us to not have to spend an insane amount of compute on just generating the next frame of the world. That's because of our unique approach of combining co generation models and diffusion models. The way we're building this product is is using a representation or a hybrid representation that allows models to choose what representation to use to represent this particular problem in this particular environment. And that actually allows us to be very computationally efficient compared to existing world model companies or video generation companies.
- [131:04]
A
Yeah, there's been a ton of pushback against AI video generation. Like the reaction to Meta Vibes on X at least was very negative. People were pretty amazed by Sora to the model. But then there's been a lot of negative takes about Sora the app. How do you think this will be received once we have a vibes or general availability? Just an app that anyone can download and start using really early. Do you think people will be more optimistic about it? Do you think there'll still be this pushback and then you have to get through it somehow? How are you?
- [131:38]
B
My view is the viscerally negative reaction that I have to a slop AI video is because of like the character inconsistencies and like something about it just feels like very wrong to look at. Whereas with Moonlake, I imagine you can create this like you know, if you generate a consistent world that's interactive and it's beautiful. Something like being in a mid journey prompt that I think would be. It's about the experience the user has when they actually are playing around with it. Like today Soar can be entertaining, but still you can have a negative reaction to it. Whereas I can see this done correctly.
- [132:17]
A
Yeah. How do you think people will react to when this technology, broadly, probably from you, hopefully becomes general availability to the point where. Yes, I like it. Bet on yourself. To the point where it's just an app, anyone can download it and we're seeing millions of people on board. What do you think the reaction will be?
- [132:36]
G
I think people will love it.
- [132:37]
A
Right.
- [132:37]
G
Because what we're doing is not like we're generating the content that people consume. We're just putting great tools into people's hands and allowing them create and craft the experience and worlds that they desire.
- [132:50]
A
Right.
- [132:50]
G
And we're just making it 10x easier, 10x faster, 10x more accessible.
- [132:56]
A
Take us through the round and I also want to hear about the uses, but give us the number. Numbers. What'd you raise?
- [133:00]
B
Also, is this your first company?
- [133:02]
G
It is.
- [133:03]
B
Nice. $28 million seed round. You're in the big leagues. Straight to the big leagues.
- [133:09]
A
Yeah. Who did the round? How much was it?
- [133:12]
G
Yeah. So we raised 28 million from Ventures, AI Ventures, Nvidia Ventures and alongside 10 + unicorn founders and angels and some of the most influential AI researchers too.
- [133:26]
A
That's great.
- [133:26]
G
We're super grateful for the support and.
- [133:29]
B
I can't wait to see what people build on top of Moonlake and play around with it and I'm sure we'll see you back on here soon. Thank you so much for joining and congrats on congratulations.
- [133:38]
G
Thank you for having me.
- [133:39]
B
Cheers.
- [133:40]
A
We'll talk to you soon.
- [133:42]
B
Up next, we have find your happy place.
- [133:46]
A
Hotel, great amenities, dream events, top tier cleaning and 247 concierge service. It's a vacation home, but better. You know, I. One day you're gonna snap when I cut you off. I still think it's funny. I think some people are getting sick of it with the. With the ad transitions, but we're having fun out here. Get some news. In the chat people, they said the. The circle table is growing on them and the horse in the background's growing on them.
- [134:10]
B
There we go.
- [134:11]
A
They say never.
- [134:12]
B
Can we get a. Can we get a shot of the horse, please? Our beautiful horse cam, this massive. I really have to stand next to the horse to get the scale.
- [134:23]
A
Well, we have Carl pay from nothing in the restream waiting room. We'll bring him into the TVPN ultra dome. Let's do it now and get the update on nothing. Carl, how you doing?
- [134:35]
B
Wow.
- [134:36]
A
Long time no seeing.
- [134:36]
B
Look at this view.
- [134:37]
A
Looks fantastic.
- [134:38]
E
Long time no seeing. It's good to be back.
- [134:40]
B
It's great to have you.
- [134:41]
A
Welcome to the show. Give us the update, give us the news. What's the latest in your world?
- [134:46]
E
Yeah, there's a lot happening over here. Last month we just announced a $200 million Series C raise.
- [134:51]
A
Let's go.
- [134:51]
E
We're very excited about that.
- [134:57]
A
We got Henry on the drums.
- [135:00]
E
Thank you so much.
- [135:01]
A
Congratulations. What was the biggest catalyst for that? Is that just growth? Adoption churn? Is this the AI narrative? What was the biggest catalyst?
- [135:11]
E
Yeah, I think it's just growth. Like last time I was on the show, I told you guys that we're approaching a billion dollars in revenue just this year.
- [135:17]
A
Wow.
- [135:18]
E
In the last couple of years, we've been really focused on building the foundations because we all know, you know, making hardware is pretty, pretty tough. There's so many things you got to get right. So we created this, like smartphones. It's one of the most difficult products to. To make in. In hardware, and we're seeing a lot of traction in the market. So I think this fundraiser, it's more about all the traction we have in terms of growth, but also with an eye to where we're going in the future because I think in the next couple of years we're going to see an explosion of new hardware and software products based on all the advancements we're seeing in AI. And I think we're just pretty well positioned to this opportunity because on one hand, we're not too small. Like, if you're too small, it's hard to ship, like, really high quality products. Products. We are a consumer business so people actually like real people, spend real money on our products. And we're also not too big. Right? If you're too big, it's kind of, you're already super profitable. It's a public company and it's hard to really latch onto these new opportunities. So I think we're in a sweet spot where we can still move fast and be super creative, but also have the infrastructure that we build from engineering perspective, supply chain perspective, and also go to market perspective perspective.
- [136:30]
B
How quickly are you and the team shipping features and functionality that you just want to try yourself that's maybe not ready for users? I mean I can't imagine there's so many people building an AI that wish that they had unlimited access to their like hardware and could do whatever they wanted. Right. Let's try a new model in there. Let's try this open source model. Let's try this new transcription.
- [136:51]
A
Yeah, he demoed that exact feature that I was begging for.
- [136:55]
B
And so I just feel like you guys have that incredible advantage of owning the full stack.
- [137:03]
E
Yeah, we're definitely cooking up a lot of products right now. And I think, you know, having built a team that can make phones, making other form factors is a lot easier. So we're going to be shipping our first AI devices beginning, starting next year.
- [137:18]
A
Let's go.
- [137:19]
E
But other really exciting.
- [137:21]
B
What can you give us? Give us stuff then. Form factor.
- [137:23]
A
Is it large than a bread box? Is it bigger than a phone? Is it smaller than a phone?
- [137:28]
E
Well, like, you know, we're in a phase for the entire tech industry. We're looking for a product market fit for the next form factor.
- [137:35]
A
Totally.
- [137:37]
E
We don't think anybody's going to get it right on the first shot. So we're really gearing ourselves up to iterate quickly, take feedback quickly and ship stuff with great taste. So I think that's the kind of winning strategy if you don't have products with pre existing product market fit to build towards.
- [137:57]
A
We were at and I heard an interesting take from their head of hardware devices. He was saying that he's a big fan of technology that mimics things that we've been wearing for like centuries. So he wants to do glasses, a necklace, a chain, a wristband, a watch, a hat, but not try and create something that no one's ever interacted with before. The exception's obviously the phone. We weren't really, I guess we were carrying around notebooks before the iPad or something. But do you ascribe to that belief that it's easier to bootstrap a new hardware device off of replacing a ring or replacing a hat or something like that. Do you like that framework?
- [138:41]
E
I do, because the product market fit risk and the user education risk drops significantly. But I think the real catalyst to these new devices gaining adoption is actually consumers finding value in them. And the reason why I believe they will find value in them is over the next couple of years, as people find that these new devices with these new sensors can capture more context on them, then the product will become more personalized and more capable for each individual.
- [139:10]
A
Is that just cameras and audio? I was laughing to myself about is Smell O vision a prerequisite to AGI? And I don't know if there's any progress on getting smell sensors, but it's.
- [139:26]
E
Probably not on the roadmap for next year.
- [139:27]
A
Okay, good luck. Hopefully you have some. We have one guy in R and.
- [139:31]
B
D, Carl Pei, says he's not entirely against smell ovision.
- [139:38]
A
Okay.
- [139:39]
E
The other thing I wanted to mention was our announcement yesterday. We announced something called Essential Apps. And basically it's a platform where every person, every creative person can just use natural language to type what kind of apps they want on their phone and then they click deploy and it shows up on their phone, on their nothing phone.
- [139:58]
B
That's the kind of thing that how many people that are building vibe coding products wish that they had had even the ability to do something like that without having to make it work within Apple's ecosystem. It's just, it's pretty. You guys have such an advantage in terms of experimentation.
- [140:17]
E
Yeah, that's the fun part about owning the full stack. And today we just launched. Yesterday we just launched the Alpha, so the capabilities are still quite limited. But we're already seeing just after 24 hours, almost 10,000 people sign up for the waitlist. And it's not just like any wait list because you have to pitch your app idea for us to accept you into the wait list. So I'm super excited to see what people build and deploy.
- [140:41]
A
If you were to build a humanoid robot, would you go with clear plastic so you could see the innards? And do you think that would be more dystopian or actually more like humans would be willing to embrace it because it's not pretending to be a human. I know exactly what it is. And I think would it get us away from the uncanny valley of a humanoid robot that's trying to look like a human and has some sort of skin mask on and looks kind of creepy versus like if you're just Seeing the guts, that's maybe better. What do you think?
- [141:16]
E
I think it needs to be even friendlier looking because all this new technology is very exciting for us in the tech industry. But out in the real world a lot of people are super scared of what's going to come. So it has to be disarming, has to look friendly. There has to be a warm feeling interacting with that technology. So did you see what doordash or something.
- [141:37]
B
Did you see what Doordash did this week with.it's such a cute little robot, they absolutely crushed it. So I think that's the way to go.
- [141:44]
A
I love it. More friendly robots, more friendly creatures, hardware, devices. Very excited for everything you're doing. Thank you so much for stopping by the tvpn.
- [141:54]
B
Always great to get the update and congratulations.
- [141:57]
A
We'll talk to you soon. Have a good one. Up next, we have Shazan from Lava coming into the studio with some exciting news. He's in the restream waiting room. We will bring him in to the TB pin ultradome right now. Let's play some sound cues. How you doing? Good to see you. It's been too long. Great to catch up.
- [142:18]
B
Great to see you.
- [142:19]
A
Give us the latest news. I got to warm this mallet up. What you got for us?
- [142:26]
H
Camera is always pretty horrible when I'm on TVPN for some reason.
- [142:29]
B
Yeah, what's. Yeah. We'll get Ben and the production crew to give you the guide to your first tvp. It's especially rough.
- [142:37]
A
Car will pay, maybe cinematic setup, wired in. So you, you're in, you're in 1980s.
- [142:42]
B
But you're too locked in. You're locked in.
- [142:44]
A
You know, there's nothing wrong with a laptop. Call in. We love it.
- [142:48]
H
Yeah, I'll have to get something better next time. So today we announced the race, but even more importantly, we announced the first way for people to earn yield on their USD backed by bitcoin. So basically historically in crypto you've been able to earn yield on your cash, your stable coins, but you had to engage if you really wanted to do it. You have to engage in all these complex crypto financial engineering schemes. You know Terra Luna, which was a big blow up that happened a couple years ago, or you have to lend your money on these platforms where the collateral is a basket of altcoins that you know could get. They're not as liquid, they're very volatile. Lava is the first way for you to earn yield on your cash. Cash backed only by bitcoin. So the way it works is you can deposit USD from Fiat from Stablecoins to Lava and you earn yield by Lava, then making bitcoin backed loans on the other side of that. And unlike banks, basically this is what banks do with your money. Right? You give them cash, they make loans. Lava's loans are all over collateralized by Bitcoin. Bitcoin is the most liquid asset in the world. It trades 24,7. It's very safe.
- [143:59]
A
Safe.
- [144:00]
H
All loans can be instantly liquidated in the case of default so that you as a lender are always protected and safe. Banks basically take 99.7% of the returns they get for themselves. Lava just gives that back to you. So that's why our yield is so high. It's double the next best savings account that you can get right now. And the nice thing about Lava is that it's available all around the world. So even if you're someone in Brazil, for example, example, you can use Lava, get cash and earn 7 1/2% yield on it. Obviously Lava has a lot of products for bitcoiners, but this is mostly for people on the other side who have dollars and are not all exposed to bitcoin but want to earn a safe and high return on their dollars.
- [144:46]
B
What's pushback? Have you gotten any pushback from users yet? I think when people see yield that's higher than the Fed's funds rate, they're starting to get a little wary intuitively.
- [145:00]
A
When I see 7.5%, I think 7.5% risk, I think 7.5% gain, but also 7.5% risk, I feel like risk and return should be directly correlated in a functioning system. But how do you think about it?
- [145:15]
H
Yeah, I mean that's why I'm here. There's a lot of education that we need to do and there will be a lot of trust that we'll need to build with users over time. Right. So I, I don't think that exactly. Just because it's seven half percent, that means it's a lot more risk than the Fed funds.
- [145:30]
A
Right.
- [145:30]
H
I think it also shows that on the other side, the bitcoiners are willing to pay a higher yield. Bargain against the bitcoin in the market.
- [145:36]
A
Sure.
- [145:37]
H
Fully efficient right now.
- [145:38]
B
Yeah.
- [145:39]
H
You know, the yields might compress over time as well the yield, you know. But here's some things I can talk about the safety.
- [145:46]
A
Right.
- [145:46]
H
Like one, the loan book has never had any losses when you lend your, when you deposit your money into cash, there's some level of FDIC insurance. But if you get past that, then you're also just taking on the risk with the bank. Right bank is making under collateralized loans. With Lava, you're making fully over collateralized loans, double collateralized. And because bitcoin is so liquid, as a lender, you know that the bitcoin can be sold at any time so that you're, the cash that you're lending it will never go underwater. So there's a lot of things that we're also going to review, release over time that can help people feel way more comfortable with the safety. So on one side, we've had a lot of bitcoiners that recently saw this product that are not all in on bitcoin. They totally get it. Bitcoiners have been saying for a long time that the best risk adjusted rate of return you should be able to get on your cash is by lending it against bitcoin, Bitcoin. So for a lot of these people, these are early adopter user bases. They've already just been like, we've had a ton of signups. They've been coming into this product, they already get it, they already understand the security, they understand Bitcoin's liquidity and how bitcoin market structure works, how selling bitcoin works, how liquidation works. So they've been really excited because there are bitcoiners that understand Bitcoin that might have 60% of their net worth in Bitcoin, but the other 40% in cash. And it's almost like this barbell strategy where your long term assets are in bitcoin. It's like the best long term savings asset. But then with Lava now you have this short term liquid way to deposit and earn yield on cash and you can always instantly withdraw as well.
- [147:19]
A
Yeah. Is your business helped by a lack of volatility in Bitcoin or are you hurt if bitcoin becomes really volatile? It's been really stable this year. It feels like it's just kind of hanging out around 110 or so, I think. And it certainly has felt like a lot less volatile time with bitcoin specifically, is that good for your business or does it not really matter? Walk me through the relationship of volatility to your business.
- [147:52]
H
I mean our business changes in different. Maybe I can just explain how our business looks in different worlds of bitcoin.
- [147:58]
A
Right.
- [147:58]
H
Like right now, I think if bitcoin were to go down in price over the next, let's say a few months, we actually might see a lot of increased borrower demand. Because when bitcoin is down. People don't want to sell their bitcoin because they're like, well, it's kind of down. I predict that it will appreciate over time. Right. So they actually will borrow against their bitcoin when bitcoin is up, maybe in a couple of years of bitcoin, really get it to where I think there might be a time where people might actually. I mean, I'll never sell my bitcoin. A lot of our users never want to sell their bitcoin, so they'll probably still bar against their bitcoin. But there might be some people who might say, let me take some, you know, let me. Let me sell some of my bitcoin. Then some of those people, instead of borrowing against their bitcoin, they might actually sell their bitcoin and then maybe even use that extra cash they have and deposit into this, this yield product. Right. So we've kind of built this system where, where no matter how you feel about bitcoin, Lava is a place for you. If you're more cash heavy, you can use our yield product to earn yield on your cash. If you're more bitcoin heavy, you can use Lava to hold your bitcoin and borrow against it. So we have a feature for both views that people have about the world and about bitcoin.
- [149:09]
A
Okay. Anything else?
- [149:10]
B
Jordy, how's the growth of the loan book been overall over the last few months? Few months? I know last time we talked it sounded like it was pretty insane.
- [149:22]
H
I mean, yeah, it's been growing exponentially. And I think over the next few months, this next month for us is going to be big. We're making a big loan upgrade as well and lowering rates. So anyone who borrows against their bitcoin today will see a substantial decrease in their interest rate this month automatically. So we're about to release a big announcement. Maybe I'll come on back on TVPN.
- [149:44]
A
For that as well.
- [149:45]
H
Well, love it, but it's been great. I think there's been a lot of people who have borrowed against their bitcoin to do a lot of things that I didn't expect. Buy houses, buy cars. The home buying use case has been one that I was particularly surprised to see.
- [149:59]
A
Yeah. Is there a rough way to think about the interest rate and over collateralization requirements? Is there sort of a back of a back of the envelope that you give people who are thinking about onboarding?
- [150:17]
H
So I tell people to start their loan with 50 LTV. So if they're borrowing a million, put up $2 million of Bitcoin.
- [150:25]
A
Yeah.
- [150:25]
H
Right. And with Lava, you will, we guarantee you the best rate. So if you come to Lava, you know you're getting the best rate across any other platform on the market. So confident on that. One thing is, is if you actually really look at the rates on Lava today and kind of consider like capital gains tax, you really have to believe that bitcoin will just appreciate 5% compounded for it to make more sense for you to borrow against your bitcoin versus selling your bitcoin, which I think anyone holding bitcoin probably believes it's going to appreciate more than 5% or compounded.
- [151:01]
A
That makes a ton of sense. Well, congratulations on the progress.
- [151:04]
B
Great to get the update.
- [151:05]
A
Thanks for stopping by. We'll talk to you soon.
- [151:07]
B
Cheers, dude.
- [151:08]
A
Have a good rest of your day. Up next, we have Victor from Synthesia, another AI video generation company.
- [151:17]
B
First AI company we've ever spoken to.
- [151:19]
A
Yeah, it is the first I've actually seen these ads. Synthesia is the world's largest AI video platform for the enterprise. And so we'll see how this is all taking shape. Welcome to the show. Victor, how are you doing?
- [151:34]
B
What's going on?
- [151:36]
I
I'm very good, guys, how are you?
- [151:37]
A
I'm great. Why don't you kick us off with an introduction on yourself? The business kind of give us a lay of the land as it stands now.
- [151:46]
B
Sure, yeah.
- [151:47]
A
I'm Victor.
- [151:48]
I
I'm one of the co founders and CEO of Synthesia. Started the company in 2017. So it's kind of a while back before any of this stuff actually worked. And we kind of endured three or four.
- [151:58]
A
Four, yeah, we endured three or four.
- [152:01]
I
Years of pretty painful, just like sitting in the proverbial garage trying to make this stuff work. And in 2020, we got the first company to launch Avatars Technology.
- [152:09]
A
Right.
- [152:10]
I
Which essentially is technology that allows you to create talking head style videos like you're kind of watching right now. But instead of recording into camera with a microphone, et cetera, you simply just type out the script and we generate an AI video of someone saying that.
- [152:22]
B
Right.
- [152:23]
I
Back in 2017, everything kind of started with the idea of like creating Hollywood films on laptops and going much more down the kind of creative route of, you know, just like you can publish and write a book from your bedroom, you should be able to do the same thing with Hollywood films since, since the founding of the company, you know, and kind of where we are today, we kind of took a fork away from that and essentially went into Enterprise Knowledge. So what we figured was that let's.
- [152:45]
B
Give it up for enterprise Knowledge.
- [152:46]
A
Thank you for doing that.
- [152:47]
B
They tried the big Hollywood creative. They tried to pull you in. You said no, I am loyal to Seth.
- [152:54]
C
I. I love it.
- [152:55]
I
I grew up dreaming about working cross functionally and making videos for the enterprise.
- [152:59]
A
Yes.
- [152:59]
B
Go hit that gong. The whole team's fired up.
- [153:09]
A
Fired up. Love it. Making corporate videos for employee training. That's why we do this show. Thank you so much.
- [153:15]
B
And it's, I mean, it's just really smart, right?
- [153:18]
A
Oh, it makes a ton of sense.
- [153:19]
B
AI is clearly good enough to make this form of video and. And clearly not good enough to one shot feature films or even most often, eight second videos. Eight second videos. Struggling.
- [153:30]
A
Yeah. Wait, so sorry, Jordy, do you have stuff or you wanna get.
- [153:34]
B
Yeah, I mean, I wanted to jump in briefly. Cause we spent the first half of today's show talking about Sora, kind of reacting to it. I thought it was a great product for entertainment purposes. Right. The cameo functionality, everything's very memoir. But something about when you're going through the feed, it's kind of hard to look at the videos. Like, yeah, I have this very negative reaction. Just something about these videos.
- [154:01]
A
Uncanny Valley.
- [154:02]
B
Yeah. That and then the distortions and character inconsistencies is still rough to look at even when you guys.
- [154:10]
A
There's a vibe or feel.
- [154:13]
B
And so you guys have somewhat of a cap. I'm assuming a lot of the use cases you have somewhat of a captive audience. People that need to watch videos to learn about their role and their companies and things like that. But still you want to avoid people watching, watching the videos that you guys generate and having that sort of negative reaction of oh, this is AI. Right. So how do you think about avoiding that? What makes that possible?
- [154:37]
I
I think there's fundamentally like two types of video, right? This kind of video you watch because you want to. And this is mostly stuff you'll find in feeds. So like when you're scrolling TikTok, you're on Netflix, you're selecting what movie to rent, whatever. That's something you actively choose to do. And generally you have a lot of choice, right? You could watch a thousand different action films, but you choose to watch this exact one. And I think that's your point before. I think AI will be a huge part in creating lots of this content in the future. But it still feels like the reason people mostly watch this stuff today is because it's fun, it's novel, it's cool, it's AI it's kind of weird. It's not because there are amazing content yet at least 99.9% of it, there's still some good content content out there. I think the space we operate in is much more around knowledge. It's not really around creating storytelling content. It's not advertisements. It's generally content that you wouldn't. It's not content that competes for your attention. Right. It's content that's very specific and helps teach you something that could be product marketing. As a company, you want to teach your prospects why your product is better than the competition's. There's only that one video to watch. And the reason people create these videos is because. Because it's better than text. So I think the first category of videos you're generally competing against, like high quality video recorded with cameras and people who really know what they're doing. I think the thing we really uncovered back in 2020 when we started building towards the enterprise was that what we're actually replacing is text or slide decks. And in 2025, most people, especially your average consumer, want to watch and listen to that content they don't want to read. And for some people, probably all the people listening to this podcast, I'd much rather prefer to read a page of text and watch an AI generated video.
- [156:12]
A
Video.
- [156:12]
I
But it turns out that that's not the preference for most people. Right? When people have a problem with their bank and they're trying to figure out like how to do something, they don't want to read a long page of text. If they're trying to figure out how their mortgage works or how much they have to pay in their mortgage, they don't want to read the whole page of text. They'd much rather watch a video. And so where we're going down, the path we're going down is just much more around like knowledge video.
- [156:32]
B
Right?
- [156:33]
I
It's much more of like replacing text with video than it's about replacing video with AI video. And that's a really, really big difference. Where I think, you know, when you look at Sara, Roger, way out of these other models, they're much more competing to actually replace real video with AI video. And I think that's also a big market. It's super exciting. But we think that there's a much, much bigger market actually in enterprise knowledge and turn that into video. And with that, I should also say, I think a lot of the talk around AI video is always very much about the models. The AI hype at the moment is like there's A new model every second week that's even more impressive than the other one. And that's awesome, right? That's a big component, of course, of making these technologies work. But I think we have this mantra called utility over novelty, which really is about like, you know, it's not about cool demos, about bringing value to the customer. And when you think about why do you make a video, the actual generation of the pixels is a part of that, but it's a smaller part of a much bigger process. Right. So what we've been doing over the last five years is really building out our workflow. We help replace the camera with the AI models, but we also give you kind of a Canva PowerPoint style edit editor to finalize your video. We enable someone who's a PowerPoint user, not an Adobe user, to make their own video collaboration platforms. A content management system, it's a translation platform. We have our own video player that's made to work with AI video. And so for us, it's always been about building for the entire workflow of like someone has an idea or some piece of knowledge that they want to communicate to an end consumer. How do we make that entire process as frictionless as possible? Right. And I think what it just turns out is that yes, the AI models are really important, but there's so many other components of making a product that does this workflow really well. I think that's what we're much more focused on at Synthesia.
- [158:15]
B
Very cool and very impressed that you were able to crack 90% of the Fortune 100.
- [158:20]
A
It's remarkable. Last question, we'll let you go. Do you see yourself as being a foundation model company and an application layer company over the long term, or is there a world where, where you partner with the foundation model companies to let them suck up all the capex to generate frontier models? Do you already do that? How do you think about the shape of the business long term? Sure.
- [158:49]
I
We try to be very intentional about what kind of company we're building. And I think every company should be. And I think it's increasingly clear that competing on the foundation model level is very difficult. Difficult unless you raise a shit ton of money and you hire extremely expensive people. And I think that's just a game that very few companies can play.
- [159:07]
B
Right.
- [159:07]
I
We still train our own models. I think there's lots of use cases where that still makes sense for us. It's again, if you work backwards from what are people trying to do with our models. Right. It's not just like an eight second clip, for example. It's not very useful. It's also not very useful that the identity doesn't get preserved across multiple. There's a lot of these asterisks that you don't see when you watch the cool demo videos of these new models that are coming out. But we just started also aggregating other people's models in. So that both means you can just use VO3 inside the platform if you want to create content that's outside of what an avatar model can do. We're also beginning to chain these models together in different ways. We create A roll, B roll, do lots of cool things with these models. But LLM is a huge part of our platform as well. We have a copilot product, help you write the script, put together the design the visual. So we train models where we think it makes sense and we think that it gives us a competitive advantage. But we're definitely also aggregating the market. And to the extent that we can avoid having to do massive training runs, I think that's the only rational choice to do for a company like us.
- [160:07]
A
That makes perfect sense. Well, thank you so much for taking the time to stop by the tv.
- [160:11]
B
Congratulations on the overnight success.
- [160:13]
A
We'll talk to you soon. Have a good one. Thanks, guys.
- [160:16]
B
Cheers, Victor.
- [160:17]
A
Our next guests are already in the restream waiting room. We'll bring in William and Dogas from Periodic Labs. I hope I'm pronouncing that correctly. Welcome to the show.
- [160:26]
B
Welcome.
- [160:27]
A
How are you doing?
- [160:28]
B
Hey, good.
- [160:29]
J
Thanks for having us. Appreciate it.
- [160:30]
A
Fantastic.
- [160:31]
B
We have been so excited for this. This feels like an antidote to the timeline today.
- [160:37]
A
Yes. There's a lot of negativity about certain AI products. Why don't you take a negative.
- [160:43]
B
A lot of positivity about Periodic what you're building.
- [160:47]
J
Yeah, I mean, first of all, that. That's very important technology too. So don't want to pile on the hate over there.
- [160:53]
A
Totally.
- [160:54]
B
Yeah. Yeah. I mean, we've talked about the good and the bad. The good is that it's very entertaining outputs. There's logical business reasons why you would do it.
- [161:04]
A
Clearly there's demand for economic flywheels and yes, world models for training, robotics training. You know, who knows, maybe a robotic surgeon in a decade will be trained on AI video in some way.
- [161:14]
B
Anyways, let's talk about you guys. Quick intros on yourselves and then the company.
- [161:19]
A
That'd be great.
- [161:20]
J
Hey, guys, my name is Dosh. I have a background in using machine learning to study physics. And after my PhD I joined Google to do deep learning research and that's where Liam and I met. So that was eight years ago. And at Google I did some deep learning research, but I continued to do my PhD passion to kind of move all the advances from deep learning into solid state state physics. And then in 2020 I actually started the team to just focus on material science because I felt like LLMs were getting really good and we could probably really benefit from porting these advances to the physical R and D. And then I've been doing that until, you know, six months ago when Liam and I left our jobs and started this company.
- [162:05]
B
Amazing.
- [162:06]
J
Yeah, and I go by Liam. Physics background, academic background, spent many years at Google Brain, worked on a mix of generative models, reinforcement learning, and created some of the first trillion parameter neural nets. These are sparse models. Really fun collaborations with Jeff Dean, Noam Shazir and others. Then in late 2022 went to OpenAI and a group of us took precursor chat products and we turned that into ChatGPT. I've told many, many people it was legitimately a low key research project review. I mean there's like the prior up to that point was there'd been no successful chatbot. So there's a poll as to how many people would use this thing and people were guessing numbers like 10,000, 50,000. Someone's like, oh, maybe a million people will use ChatGPT, which we exceed in like a week. And I mean reflecting back on that, I think obviously there's good luck, good timing, but also boring things done really well. Like we're really focused on, on training, details, data, evals. There was no novelty to it. Ultimately people have been building chatbots for many decades, but just crossed a critical boundary. Yeah. And so left six months ago to start periodic.
- [163:22]
A
What do you think of that Goern idea of you just need to generate random ideas and look at connections between them to find scientific insights. Do you think there's anything there is that directionally correct? Is that a path that people will go down? Are you going down that path at all?
- [163:39]
J
I mean I think it's incredibly promising. Like I think linking these disparate ideas is really promising. However, unless you actually have experiment in the loop, you're just sort of thinking, sure. So you can imagine like, hey, here's a really promising thing to try. But as every scientist know, it's like until you actually take that idea into like to try it to act, you're really no further along to like building up a scientific understanding. So for periodic we're it's that's our opinion, which is we need to have experiment in the loop. That's super key.
- [164:12]
A
There's so many different experiments. Are you selecting for a few different environments? You're going to get a large hadron collider or a wet lab. Like the science is very broad. So have you narrated down at all?
- [164:26]
J
So one way of thinking about it is, you know, the LLMs today are really good at math and logic.
- [164:32]
A
Yeah.
- [164:32]
J
So the next obvious frontier is theoretical physics.
- [164:35]
A
Sure.
- [164:36]
J
And there are different energy scales of theoretical physics. But we wanted to take the part that is very relevant to human life, which is quantum mechanics, you know, Schrodinger's equation. So this affects all the inorganic materials, organic molecules, drugs. So we're starting at the level where we're just designing how atoms come together and how their properties can be utilized in devices.
- [165:00]
B
Just start with the simple stuff. Yeah. So just drilling in deeper, what does success look like for you guys as a team over the next even I can imagine the 5 year old what success looks like 5 years, 10 years. Right. These sort of massive ambition. How do you sort of make, make, you know, immediate progress now that you guys are heavily funded and have the resources to really move on these opportunities?
- [165:32]
J
I think you can kind of describe it along technical and commercial goals. From technical goals, the highest level way to describe it is have a system that can intentionally design the world around us.
- [165:45]
A
Us.
- [165:46]
J
So right now we've built systems that understand the Internet. They have been trained on math and code. We want to take these same type of techniques and build system that can produce things of like, properties of interest. So you're trying to optimize for a new material for semiconductor industry or space company or defense companies looking for something for like, you know, a new heat shield for a missile or, you know, anything. It's like, how can we intentionally design the world around us? And that's sort of a goal. Some of our higher risk goals are around novel discoveries that DOGE can talk more about as well.
- [166:24]
D
Yeah.
- [166:24]
B
And even, even before you get into that, like, you know, you guys are coming from organizations that were totally okay with years and years and years of just deep research and not worrying about immediate commercial applications from you guys with the new company. I'm assuming you're open to having much longer kind of commercial timelines and just basically experimenting. Right. Some of this stuff is unpredictable. You talked about ChatGPT being a research preview and it turned into a hit product. So yeah, there has to be some takeaway from that process.
- [167:02]
J
I mean, one Comment on that is we can basically produce through this technology, through experimental data, a foundational understanding of physics, of chemistry, of atoms, just the world around us. And you can sit at different levels of abstraction. Some, some levels might be more kind of like scientific, like exploration. And others might be more towards advanced engineering, advanced manufacturing, engineering, engineering. And we're going to have both pieces as part of periodic. So the labs allow us to produce deeper understanding of different, like physics, chemistry, material science, systems. And our guidance there is like, are we able to produce new discoveries, are we able to advance science? But however, that data also gives us a better foundation when we're actually working with customers. So we want to make sure that our, our labs, our prioritization, the tools we're teaching, our agents do have some grounding in sort of the commercial.
- [168:05]
B
Commercial opportunities, Exactly.
- [168:07]
A
Yeah.
- [168:07]
J
Like technology and capital are very intertwined. It's not like technology doesn't develop in a bubble and we can accelerate science maximally when it's like a very commercially successful enterprise.
- [168:19]
A
Yeah. I went on a serious emotional roller coaster with the LK99 Saga. And then I. There was a proposed breakthrough in superconductors, room temperature superconductors. It was very excited on the Internet for a few days. Then it was found not to be such a breakthrough. And then I kind of fell out of that news cycle and stopped tracking it. Get me up to speed on where we are on superconductors. What's at stake if we can have a breakthrough there? Is that something you're interested in working on? Do you think that this is actually even a tractable problem that you can make any sort of prediction about where we are in terms of progress there? I'm super interested in that.
- [169:00]
B
Yeah.
- [169:00]
J
You know, we're very excited about superconductors. And being able to discover a novel, exciting superconductor requires us to do well on many dimensions of physical sciences, which we're very excited about. So one of these is being able to synthesize materials reliably, being able to characterize materials reliably, being able to predict their properties, design their kind of defects. And for us, you know, it's also a very important scientific goal. So Currently the highest TC superconductor at ambient pressure is about 135Kelvin. And under pressure you can make it a lot higher temperature. But then it's not very practical because you can't really apply that kind of pressure. And it's very interesting to wonder, you know, can we bring that up to 200 kelvin for. For example, and if we can, even before it impacts products. I think it teaches us a lot about the universe as we do this, a very large macro scale quantum property. So, you know, we have real good characterization tools in the lab. And that, you know, immediately prevents us from an issue like Alka 99 because we can actually measure its properties at different temperatures. And for us, the LLM being able to use the characterization tools and infer the, the relevant physical insights is crucial. So we're teaching our LLM agents the ability to synthesize the, able to characterize the able to hypothesize the next step. So it'll be really exciting. I think part of this is so key.
- [170:29]
B
Right.
- [170:29]
J
The lab is providing our grading function. So again that keeps us grounded, that makes sure. And it's sort of a very interesting objective to put a lot of optimization pressure against.
- [170:41]
A
Yeah. Have you stack ranked the impact of super room temperature superconductors or even higher temperature superconductors in terms of the impact? I remember people talking about quantum computing, but then I also saw a demo of just some like hoverboards and like skateboards that were using magnets and superconductors to, or super cooled magnets to kind of like float. Where do you think the impact would be if we get a breakthrough through there in the next couple years?
- [171:10]
J
I mean, the impact is huge. Like whenever you think about a futuristic technology like fusion, you know, low.
- [171:19]
A
Low.
- [171:20]
J
Loss energy transmission, quantum computing, as you said, and even, you know, when you talk to chip companies like think about the chip designer 10 years from today, superconductors always come up because it's such a important quantum mechanical property. But at the same time it's just very easy to understand. It lowers your resistance to a point where losses are very low. So for us, it's probably one of the most impactful things we can do. On the solid state physics side.
- [171:46]
B
Can you guys give us a white pill like a pump up speech? I feel like so much of, you know, there's been common chatter about, you know, all this capex, all this compute that we're developing as a country and a human race. A lot of it's obviously going to go to image generation and memes and things like that. And that's fine. But you guys are doing the thing that has been promised by the AI community broadly for a long time. How excited should people be about the potential? I think you guys are both very modest. But talk about the impact in your words, because I take, I think it means a lot coming from you given that this is your scientific discovery, is the entire focus and commercializing it versus labs historically that are maybe working on some of the same stuff, but they're also working on a million other things they're also working on. They could be working on Codegen, they could be working on new versions of Search, et cetera. But this is your guys whole thing. So I'd like to hear your. Yeah, just, just that, that optimistic vision of what's possible.
- [173:00]
J
I mean ultimately science is unbounded. That's one of the biggest drivers of progress for humanity. It's not like we optimize and we can now match the performance of a human on some task.
- [173:14]
A
It's.
- [173:14]
J
You're just creating new technology, new abundance by being able to kind of control the physical world around us. You and we think that the technology is at that stage where we've seen incredible things on models from mathematical reasoning, code reasoning, also the ability of some of these agents to start doing like real valuable work. But the world around us still kind of like largely looks the same. It's like not really moving too quickly. And I think as the cost of intelligence decreases, Doge and I are thinking the bottleneck increasingly becomes contact with the real world world. Bring data, bring atoms like into this end to end system that's going to lead to this new ability to just accelerate scientific progress, accelerate the technology around us. And I mean it's the, the applications of you know, scientific developments are, are endless. Right. Like you know, becoming more of a, you know, space traveling society. Like you know, new computation, it's unbounded. I think that's what excites us. Yeah, There'll be no end to this. Right. It's not like we can finish science.
- [174:21]
B
Actually the more job's never finished.
- [174:23]
J
Exactly. So it's a really fun place to be because of that.
- [174:27]
A
I love that. Well, congratulations, we're going to ring the gong for you. 300 million.
- [174:33]
B
Last question I have before you guys jump off and you can keep it to 30 seconds. What's your philosophy around building and sharing your guys work? We've seen different approaches from the labs. Thinking Machines has been experimenting, sharing some of their progress. SSI been very silent. Some of the other labs are also, you know, you know, constantly sharing, you know, just small products, etc. But how do you guys think you'll approach it given your focus?
- [174:58]
J
Yeah, so we're building tools in the physics space, we're building tools in LLM space and we'll definitely share whenever it seems like it will enable the community. We also have an academic grant program where we want to support academic groups that are also passionate about this space because we feel like, you know, science is endless. So there's so much to do, and we'll, you know, do it together.
- [175:20]
B
Amazing. Well, thank you, guys. Thank you so much for coming on Breaking It Down. I'm sure we'll have you back on again very soon.
- [175:24]
A
We'll talk to you soon.
- [175:25]
J
Thanks.
- [175:26]
B
Congrats.
- [175:27]
A
Our next guest is in the restream waiting room already.
- [175:31]
B
Legal AGI fundraising announcement.
- [175:35]
A
Let's bring in Jay from Eve. Jay, how you doing?
- [175:40]
B
What's happening?
- [175:40]
E
I'm doing great. Thanks for having me on, guys.
- [175:43]
B
Busy week for you.
- [175:45]
A
Give us the news. Give us introduction. Give us an explanation of the company. Give us an update on the news.
- [175:51]
E
Absolutely. And I can't. I can't wait for what's coming next.
- [175:54]
B
Oh, yeah, you know, you know, get that mallet ready.
- [175:58]
E
But we, we are Eve. We're AI for plaintiff law, and we just raised $103 million at a billionaire.
- [176:05]
B
Boom. There we go. We've talked to various players in AI Legal. We don't follow the space nearly as close as probably many of the investors do. And you do, but I. First time time hearing about you guys was yesterday. Give us the backstory on the company, because you're not obviously just coming out of nowhere. I'm sure you guys have been cooking for a while.
- [176:33]
E
No, Absolutely. So we. We took an unconventional approach, which is, you know, we basically took time to understand what plaintiff and defense law means and how they're different, which maybe a.
- [176:45]
B
Lot of people, A lot of people.
- [176:46]
E
Don'T know, but plaintiff lawyers, you know, these are the ones that you would typically find fighting, fighting on behalf of individuals. Right. So they typically don't charge billable hours. They make money when they win. For you. Think labor and employment issues, personal injury issues, you know, family law, things like that. On the defense side, you get the nice benefit of retaining a company and kind of cashing it in every year. Right. And then you have to. Typically, they charge by billable hours. So on the two sides, there's very, very different business problems and opportunities. On the plaintiff side, they care about how do I represent more customers with the best of my ability to win as much as possible for them.
- [177:28]
A
Right.
- [177:28]
E
Either changing conditions or money in civil litigation. And on the defense side, it's like, how do I do more for the client for the same amount of time or maybe more time and delivering value. So on the plaintiff side, what we found is there's actually a really good merger between AI and the efficiencies you get from AI and their business model because you're able to go deeper and actually benefit more clients.
- [177:55]
B
Is that because you're charging on a value basis? Right. You're taking a fee based on the value that you capture for the client or the, or whatever, sorry, the plaintiff, if there's a settlement. Whereas on the defense side, you're charging billable hours and your tech, you know, there's some. You can do a calculation as a client and say, like, how much value am I getting? Do I feel like this is a good deal? But it's less of a. Because it's based on billable hours, it's less of a. It's less like value based pricing.
- [178:24]
E
Yeah, you nailed it. Right? So we, we typically charge based on how many clients they represent. So there's a very clear alignment of incentives between what we provide for our customers and what they provide for their clients. So we're kind of.
- [178:38]
B
In plaintiff law, you're incentivized to use more and more and more and more AI so you can have less time to serve each individual client and generate more revenue. Which is, yeah, again, maybe different than on the, on the defense side where it's like, oh, I like AI because it makes my job easier, but I don't want to be perfectly efficient unless you just have overwhelming demand.
- [179:04]
E
You nailed it. And this is the really cool part about right now now is our customers are using this to level the playing field and flip it in a way that's never been happening before. Right. So if you think about the average insurance company, they employ so many lawyers. And when you have a personal injury attorney showing up, it's oftentimes one person in court versus like 10 or 20. They just have endless resources to run you down. Right. And I mean, you know, typically if insurance paid out, like the industry wouldn't exist in all. So that's what they're there to equalize and fight for on your behalf. And what people are doing with Eve is actually using it to now act as though they have a million dollar attorney backing every single step of the way, from intake to resolution of anyone. Right. So now you can actually do all these complicated legal workflows specific to plaintiff law. And we go deeper into these workflows than really any other tool can and deliver that to the high level of bar that you would want it to have as a client. Right. You don't want as a client to worry about how many cases are they working on? Do they have Time to spend on me and the law firm has the same problem.
- [180:18]
B
How far away are we from just AI on AI, just violence in every case. It feels.
- [180:28]
A
Sorry, I'll let you answer, but I've heard a story like 10 years ago about a law firm that set up a automated system just using, not AI, just normal software to run through the line items that they were billing their client on the corporate side and detect what the client would say, hey, we're not paying for that. We're not paying for your two hour lunch, so don't write that in. And so, so anyone who was writing their billable hours would know that it got flagged. They'll take it out. And then the other party got software as well to detect what was gonna be maybe debatable in the billing. And so they did have this software on software fight in legal specifically, but just at the billing level. But now we're going one layer up. But I'd love to hear your answer on how it all plays out.
- [181:13]
E
This is actually a really interesting question and one we're seeing play out live is again, the value alignment incentives. So on the plaintiff side, people are using it and they're coming up with tactics because they know the other side has to bill hours and the other side doesn't want to really reduce the hours quite yet. So what's happening is one of the common ways defense slows you down is they kind of start discovery requests and responses. These sometimes take up 40 hours for plaintiff law firms to go fill out out. And our customers using EVE are able to come back and return back right away. Right. And then the other side is like, what w. You know, what is. What's going on? And they want to spend the time to, to respond. And that's actually raising the settlement value of firms because you're basically eating into what the lawyers would be charging their clients.
- [182:06]
A
Right.
- [182:07]
E
On the other side.
- [182:08]
B
Interesting. What are, what are some of the downstream impacts of what you're building? You know, assume that, you assume that you can get to, you know, 50%, you know, or some meaning double digit percent of plaintiff law firms using EVE or comparable software, if there is a comparison, what are, you know, is that going to impact the insurance industry? Like are they going to have to, you know, if every plaintiff has this super lawyer lawyer, you know, what, what are kind of downstream effects?
- [182:42]
E
Great question. There's a few different downstream effects I foresee happening. And we're seeing a little bit of this already. So for example, let's take labor and employment, right? So think you get fired, harassed, etc, you're getting representation in, in this industry, typically a lawyer want to see at least $5,000 in fee before even representing you because again, they're doing it for free until they win. So they're baking that into how they think about it. And what we're seeing with Eve is actually a lot more pro bono cases or them going down the value chain to take on cases they otherwise would never have. Right. And now.
- [183:18]
A
We'Re getting more lawsuits.
- [183:19]
B
Yeah. So are lawsuits going to go exponential? Are we going to see exponential? I mean, this would be very American if, if we used AI to just 10x the amount of lawsuits annually. But it feels like.
- [183:30]
E
Great, great question. What's interesting, and maybe most people don't know this, this is very few cases actually get to a lawsuit perspective. Right. So in personal injury, about 70% or so of cases actually get resolved before a.
- [183:43]
B
Like at the demand letter stage.
- [183:45]
E
Exactly. They would get resolved demand letter stage or some form of mediation before.
- [183:49]
A
Might even just be like, hey, yeah, like you didn't pay this invoice. You know, we, but we both know that this could turn into a lawsuit very quickly with AI. So we're just going to settle quicker and we're going to get to like the median outcome where both people are happy.
- [184:01]
E
And that's exactly because once the lawsuit starts, you know, $700 an hour, they're counting hours.
- [184:07]
A
It's arbitration. Right, Arbitration. Essentially. It's like what's the, maybe the long term results of a lot of like these AIs battling it out.
- [184:15]
B
What's. Give us some numbers. How. I mean, you raised it a billion, so things are seemingly pretty good. But what, what kind of traction have you been seeing so far?
- [184:24]
A
How many times have you been.
- [184:24]
E
No, one of the, one of the interesting bits is, is usage as well. And we've been fortunate. I mean, one of the really crazy things about this industry is I was just in a conference probably three weeks ago and I asked this question every year. The last year I was there, I asked how many people have used AI every day or you know, in the last week? 10 people out of our 200 raise their hand. This time around, personal injury attorneys, 90% of them raise their hands.
- [184:51]
A
Wow.
- [184:52]
B
Wow.
- [184:53]
E
So the amount of awareness and traction it's seeing is crazy. But what they've always been scared about is like, how do I use it? And this is kind of the problem we're solving. Right. Because if you're just using ChatGPT or Gemini or whatever, you're spending a Lot of time fixing things at the end. And that's not good enough for lawyers. They lose trust very quickly. So we had to go.
- [185:12]
B
Yeah, if you have an associate that just constantly makes every single time they make, you know, one or two mistakes, those associates don't, you know, eventually they get turned out, they get let go. Right. So you want to hold the AI to a higher bar.
- [185:26]
E
Yeah, exactly. And then like when we hit that, you know, extra 9 of accuracy, which I'm sure you two are very familiar with, our usage and adoption rate kind of blew up. Right. So we went from I think around 13 customers or so, start of last year to now more than 450. I think we're approaching 500 now.
- [185:43]
A
Let's go.
- [185:46]
E
And you know, dollars, science are increasing very rapidly as well, which helps us invest even more behind R and D, which I'm super excited about.
- [185:55]
A
The Cognition team just rewrote all of Devin on the back of Claude 4.5. Are you in a similar world where a new model can cause a fundamental shift in how you think about building the product? Are you sort of model agnostic? How closely are you tracking what the foundation lab are doing?
- [186:15]
E
This is one of the key decisions we made early on in the architecture is we worked very closely actually. Funnily enough, we were one of the first productionized apps, I think on Claude 2 back in the day, and they actually expanded Context window just for us. So you have multiple modality of increases. Right. So model capability gets better, Context Window gets better, agent functionality gets better. All of these have a pretty drastic implication on how deep into a workflow can we actually go and what gets enabled?
- [186:47]
B
So somewhere, somewhere out there there's a, there's a defense attorney who isn't online, not using AI, just doing everything the old fashioned way, and he's probably just getting his world rocked like month over month. Like, how did you respond to that in an hour? I was expecting, I was expecting that to be, I was expecting that to be three, three weeks. Yeah.
- [187:11]
A
That's great. Walk me through this thesis. Tell me if I'm right or wrong. Anthropic's been incredibly focused on the flywheel of code coding agents. They're hoovering up tons of data, reinforcement, learning on top of the code, developing a really dominant code model. OpenAI has been less focused on API, but ChatGPT is dominant. I've talked to, I bet if you polled those lawyers, a lot of them are using ChatGPT if they're not on the enterprise plat plan. ChatGPT and OpenAI are going to be able to train on that data. And so you might have a data flywheel that actually makes OpenAI's APIs better for legal work. Is that a reasonable thesis?
- [187:55]
E
Well, I mean, according to the agreements we signed with them, they're not supposed to.
- [187:59]
A
That's because you're on the enterprise plan. What I'm talking about is, like, if I, a person, don't have an agreement with ChatGPT, and I can get, you know, a contract for an advertiser or something, and I just uploaded that to ChatGPT, they're going to train on that. And I'm not saying that's your company, I'm saying me as an individual, I'm just putting extra data in ChatGPT. They're getting more data, it makes the bottle better. And then you benefit on the API side, even though they're not training on your data.
- [188:26]
E
Yeah. I think one of the interesting bits for us is we benefit from the general training. OpenAI, all these providers do.
- [188:33]
A
Yeah.
- [188:33]
E
Mostly because actually, law is weirdly touches everything.
- [188:38]
A
Yeah.
- [188:38]
E
You know, when you get into a lawsuit, sometimes it's medical, sometimes it's technology. Right. Sometimes it's like quantum computing or something. And these lawyers have to somehow be up to speed on these things. Right. And the general training actually really helps. So that's one part of Data Flywheel. There's another part that I don't think people are thinking enough about, which is the one we benefit from from in how we build eve, which is all these intermediate tasks are actually not captured by anyone. An example of this is like, you actually can't even go Google how do I write a really good demand letter for a moving vehicle accident for, again, State Farm or something? It just doesn't exist. And what they're doing internally is they're coming up with multiple artifacts to produce that final output. Right. So everyone has a workflow, workflow. Produce this document, this document, this document, and pull information from these three to make this final document. And they've all come up with their own workflows. One of the things that we've been, I think, fortunate to do in EVE is we actually allow them to kind of teach EVE to do their workflows, kind of like how they teach a new associate or a new hire. And then after that point, it just follows it diligently, without fail, at a high level of bar. Right. This does. It doesn't miss things like a person would, because it finds everything instantly. So that's one of the ways in which it's changing and maybe I'll be a little spicy. Like what I kind of predict will happen here is I actually think these law firms will basically transform into software companies over time. Right? So think about early Amazon or something. Everyone started being able to offload their hard disks and all these things into APIs. And what is happening with EVE is now, instead of relying purely on human labor to scale, you can start offloading these very crucial tasks that were previously not apiable into eve. Right. And you can view yourself as now, hey, I'm not just servicing one client one on one anymore. I'm figuring out a way to service everyone that I'm an expert. Expert in. Right. So if I'm an expert in dealing with traumatic brain injuries, I'm going to take on every single traumatic brain injury case and do the best I can for them and be a rock star at it. And this is like the main benefit that software companies benefit from. They've consolidated the R and D efforts to make a really good product and sell it at incrementally lower additional cost for a net new customer.
- [191:11]
B
I think it's super exciting. I think the best lawyers are your friends. They're expensive friends. Friends. Right. But they help you. You know, they're just, they're. They're there to think through problems. And I think that's just going to continue to be. I think that's going to continue to be valuable for a long time, especially for the market that's paying for legal services today. But they're going to be able to be a one person. You're going to have a one person, $100 million revenue, you know, plaintiff law firm, I imagine. Right. Because you just have this massive, massive efficiency and one person who's just great at. Yeah. What's the doctors, you know, bedside manners. Basically like guiding somebody through a, you know, potentially challenging part of their life. So super bullish on what you're building. Congratulations and thanks for breaking it down for us. Gotta come on anytime. There's stuff in law. This was a fun conversation.
- [192:01]
A
We'll talk to you soon.
- [192:01]
E
Thank you so much. Appreciate it, guys.
- [192:03]
B
Cheers, Jay.
- [192:04]
A
Have a great one. I want to close out. We gotta hop on with Taipei soon. But first I want to close out with this video. Video from Elena or Elena. She's introducing Taya AI. You'll actually want to wear. I want to get your reaction to this launch video. Another AI wearable. They said it couldn't be done.
- [192:28]
B
Jewelry.
- [192:29]
A
We worry so much about tomorrow, we forget about today. It's an aura ring of your neck. It is remarkable how much you can pack into a ring these days. I've always did an aura ring though.
- [192:42]
B
Is it give you bio?
- [192:44]
A
No, no, no. It's not technically an aura ring but it's just the oura ring form factor. It's a ring Lindy form factor. People wear necklaces. I haven't seen a device company go after this market at all. It seems obvious now that you're seeing it and it.
- [192:59]
B
I noticed you can turn it on or off so it's not just default.
- [193:01]
A
On recording and so it is an AI pendant. It is in the lineage of humane and friend.com but it's less about saying mean things to you if you're a journalist and instead it's more focused on note taking and remembering things and being an opportunity. You pull out your phone to make a note, you get lost in the slop feed. Instead you can just make a little note and it will remind you to go to this restaurant or that you want to, you know, get pick something up from the grocery store.
- [193:34]
B
Go eat. Brady in the chat says I pre ordered a tie. I will probably give it to my mom. It's like Avi's friend thing but actually novel and framed as jewelry.
- [193:41]
A
I imagine this would go. Yeah, they should ship this before Christmas because this would be a great Christmas gift for sure. I agree with.
- [193:50]
B
Obviously a more refined. You can see even at the software layer it seems like they've done a bit more to not. They're not trying. They're. They're what. What I'm not clear on is how much are they trying to be your friend. Do they care about that at all? Is it just about function?
- [194:06]
A
It does have a human name. Taya. Sounds like a name that you could wind up interacting with as a, you know, an AI friend or AI companion.
- [194:15]
B
All right. Fused in the chat says might have to propose with a Taya. I don't know about that. Settle.
- [194:22]
A
I think the non IoT device might be Lindy but I mean it's cool. I think we've seen so many shots on goal with the AI devices and it's good to see another one.
- [194:37]
B
We should work on our own. An AI device that's a full size life size horse statue.
- [194:45]
A
It just has. It does an entire H100 rack inside of it.
- [194:49]
B
You need a forklift to bring it around.
- [194:51]
A
It can inference GPT5 locally horse cams looking great. I was gonna say what about Jordy? What about a smart cigar that you smoke to the filter and. And as you smoke it. It gets less and less compute that.
- [195:07]
B
I love this horse so much.
- [195:09]
A
I can't wait for the literal unboxing of the horse. We need to make more videos. In other news, Elon is building Grokipedia with Xai. He says it'll be a massive improvement over Wikipedia. People have been going back and forth on how valuable Wikipedia is. I think the founder of Wikipedia was not a fan of it anymore. Kind of the creation. Sort of a Frankenstein's monster situation, I suppose.
- [195:31]
B
Well, I have a little white pill to end the show. Let's end on a positive note, please. Gwart says doesn't matter who you are or where you come from, crypto provides the opportunity for anyone in the world to make. It's true, of course. Forbes reported yesterday that Baron Trump is worth 150 million and has made an estimated 80 million from token sales with an additional 2.3 billion in locked up tokens. So he has been busy.
- [196:00]
A
Get him on the show.
- [196:03]
B
You know, you remember when people got mad that Hunter was selling his paintings for maybe more than they were worth or being on a board here, there, you know, he's so much more efficient.
- [196:12]
A
To remove the paintings. Just eat tokens.
- [196:14]
B
Hunter Biden's got to.
- [196:15]
A
It's way more efficient.
- [196:16]
B
He didn't. Didn't get into cryptocurrency.
- [196:21]
A
Yeah, missed opportunity. You missed 100% of the tokens you don't want.
- [196:25]
B
Really missed out on, you know, Bill. I don't know. It's hard to imagine. I guess Hunter Biden token launched by Hunter Biden would probably.
- [196:33]
A
Might move. Who knows?
- [196:34]
B
Probably.
- [196:35]
A
Be safe. Be safe there.
- [196:36]
E
Out.
- [196:36]
A
Out there.
- [196:36]
B
Yeah, be safe out there, everyone.
- [196:37]
A
The chat is asking us to name the horse. They're also asking for the backstory on the horse. We'll bring that to you tomorrow. We can do a whole deep dive on equestrianism tomorrow. We've long been fans of equestrianism, but we got to hop on with Taipei. Thank you so much for tuning in. Leave us 5 stars in Apple podcasts and Spotify and stay safe out there in the trenches, folks. We'll see you tomorrow. Have a great rest of your day. Goodbye.