Math Wars, GG Navier Stokes, Greg Brockman Joins | Dan Wright, Eric Seufert, Scott Wu, Sahir Jaggi, Andrew Borovsky, Harry Mellsop

You're watching TDPN.

Today is Monday, September 8th, 2026. We are live from the TDPN UltraDome, the temple of technology, the fortress of finance, the capital.

You know what time it is, Jon.

Ramp.com time. Time is money. Save both. Easiest corporate cards, bill pay, accounting, and a whole lot more all in one place.

I need my countdown. I need my countdown.

Doing some math.

There it is. Road to Christmas, 107 days.

December 25th.

It's flying by.

Minus today.

It's flying by.

Made only possible with math.

Time is flying.

107 days. Time is flying till Christmas.

Start—

Road to Christmas is going strong.

Start scoping out Christmas trees.

Yep. Now's the time.

Once we, once we get under 100.

Hopefully you've already planted them.

Open season.

Open season. Does math matter? That's the big question. I was debating this with Tyler today. Lots of, lots of, Lots of tension on the timeline over Navier-Stokes, the Millennium Prize math problem. There has been a Millennium Prize math problem that was solved before the Poincaré conjecture in the pre-AI era. These are very, very difficult math problems that all PhD mathematicians, all sorts of elite, the upper echelon of math has been grappling with for years, and AI is starting to knock these down. Remember, Last year was the year of the IMO gold medals from both Google DeepMind and OpenAI. We talked to Scott Wu about this, who's coming on the show later today. Huge race for cognition. And Scott made the prediction on TBPn early in 2025 that he believed that artificial intelligence would be able to achieve gold at the IMO, the International Math Olympiad, in 2025.

Yeah.

And he was right. He nailed it. Progress continued, but debate remained around cool calculator, bro, basically, is the critique. I mean, people just— they don't feel empowered by a system that's really good at math. They just— people already— a TI-83 can do more complex math than most people can do. And so, When you say, oh, there's another level of math that you don't know about and you don't care about and computers are good at it, a lot of people, their eyes glaze over. And so, the bigger question here for me is all about the public perception. What actually matters here? I think they're cool benchmarks. I think they're interesting stories. Obviously, there's a lot of drama over this race. It feels like it was neck and neck. Now, there's a bunch of debates going back and forth with a mathematician who works at Google and a mathematician who's independent who worked alongside an Anthropic researcher who are both working on it, and they were apparently texting back and forth and going back and forth on did they steal the data, did they copy each other's, do they have different approaches. And so lots of different back and forth. And I think someone summed it up well by saying that these math models have discovered the hardest problem of advanced mathematics, which is authorship contribution, who actually did the work, who gets lead left on the paper. But yeah, there was a drama last year, 2025, during the IMO. OpenAI and Google DeepMind, and I believe Harmonic, the math AI lab that we had on the show last week, were all working on solving the International Math Olympiad. Remember, there's 6 questions. Question 6, no AI model could solve. It was more complex. But Both teams got a 35 out of 42, even though they only got 5 questions out of the 6 right. That's good enough for gold. It's the bare minimum, but they did it. But there was a bunch of back and forth because OpenAI validated with former IMO gold medalists while Google's result was graded by the official rubric that year, which is internal to the organization. We called it at the time, it's like they ran the fastest 100-meter sprint in the parking lot while the actual Olympics is going on in the stadium. And so, a lot of these are just, you know, vibe wars, races, who does what, does it even matter? The interesting thing about Navier-Stokes is, does it matter, Tyler? You think it does? I'll debate you on this a little bit. It seems like it doesn't because the equations themselves are known. They're just not fully proven. Engineers and physicists already solve them numerically all the time for a particular situation. So airflow around a wing, you might have to use some of these equations, weather models, water through a pipe. These are critical things. They are boring, but you can imagine that there would be cause for optimism. You know, better weather models save lives during hurricane season. Increased airflow across a— an airplane wing that could lead to more efficiency, longer ranges, cheaper flights, lower carbon emissions. There's a bunch of reasons why you could solve all that and be excited, but that's not actually what—

One of the reasons I'm excited, I believe we agreed last week that if anyone solved Navier-Stokes, Tyler would shave his head.

Oh, that's right. Didn't we do that? We said something.

No, no, that was about the—

Oh, that was the task force. Details are coming.

Get ready to shave your head, buddy. But yeah, this is a big moment.

Yeah, I think it was actually more, It was, if any AI progress happens at all, Tyler shapes to say.

Yeah, yeah.

But do you have a steelman for why this matters?

So, I think I agree that, like, I don't think that the breakthrough, you know, this new stuff that's going on is, like, practically super useful. I don't think you're going to feel it when you're on an airplane. You're not going to feel, like, less turbulence now because of this. But I think, like, you know, these are, like, the most important open math problems. I think, like, These are more close to like, you know, a beautiful painting or something than like some like practical engineering breakthrough.

Yeah.

Like, this is like pure math.

This is like the, you know, this is the peak of the mountain of like—

Same month.

Math.

Same month that AI solved this incredibly difficult math problem, I saw someone use ChatGPT to book a haircut. So equivalent. No, I mean, in terms of the actual economy, the actual impact of these things, like practically usable, Using these tools for something that people enjoy, that's not just spinning their wheels, it's not just optimizing endless things, that's actually generative and moving things forward, not just defending against cyberattacks, doing something interesting.

Winsome says, if he gets a Millennium Prize, you all have to get bowl cuts.

Oh, bowl cuts.

Bowl cuts are interesting.

That's really, really ridiculous. You do get a million bucks, so, you know, there's that. That's not nothing. But in the AGI race where trillions are being deployed, it is very minimal. But it sounds like cause for optimism that Navier-Stokes would improve airflow over a wing, something like that, but that's not really how it will work. Proving these equations is unlikely to actually have any measurable impact, and that's agreed upon, and physicists have chimed in with this take that this won't actually move the discipline of engineering forward at all, but it is a cool demo. Engineers already solve the equations numerically, as I mentioned. So clearly, it matters for hype and vibes. Solving hard math problems is a good heuristic to show how quickly models are progressing.

But—

Yeah. And to me, it just comes down to humans are able to use machines that humans built to further humanity's general knowledge.

Yeah, yeah, it's good. But I feel like benchmarks, and if you're just trying to measure AI progress, benchmarks and raw inputs and raw metrics of the model sort of do the same thing. Like, the number of people who will be convinced that AI is progressing exponentially because they see this result, and then you look at the number of people that are that would be equally convinced of AI progress just by hearing, oh, there's 10 gigawatts coming online, or the new model has 10 trillion parameters, or it was trained with $1 billion of compute or something like that.

The thing is, the average person is probably more impressed by Sora than by this. Yes.

Totally. Totally. It's way more visceral to actually see something. And that's why the weekend was very much Blender and people showing 3D renders and remodeling their houses and building small games and web games. And that feels much more like a, okay, breakthrough. People are having fun, they're vibe coding stuff, but they're also doing 3D modeling, which is like a new skill. We've had image generation, video generation, but there's something much more, I don't know, like grounded in watching a— from just a few images, watching a Blender model come to life. You can walk around inside of it and it can be something that, you know, the model doesn't exist. We saw a few demos that were like a train. And it was like, oh, wow, I modeled this amazing train. But there are perfectly modeled trains out there that you can just go grab from the internet. That's not that impressive. When you actually see your specific house or your car or something that doesn't already exist out there perfectly modeled, you're like, wow.

Sorry to interrupt. John Anderson says, why are we only seeing the calendar days for Road to Christmas? Can we please get a weekdays and workdays breakdown for Die Hard? Yeah, we need a full-screen graphic. Let's work on that. Thank you. Thank you, John.

Good comment. Good comment. Let's— yeah. So, in terms of feel the AGI, computer use, definitely one of these. Astra, particularly faster at computer use for, like, really, really mundane things. Also, just interesting to just have a very different interface to your computer. like configurations, all the things where you used to have to dig through, dig through a bunch of settings, maybe set something on the command line, maybe find a control panel. You can just ask Codex or any AI agent to go and do it and it'll just do it and come back to you.

One of the use cases that I'm excited about is getting, you know, when you get a parking ticket and then you're like, I got this parking ticket, I want to deal with it. And then it's like, well, the ticket is not in the portal yet.

Sure.

Being able to go into chat and say, hey, Once this ticket is live in the system, please pay it.

Parking superintelligence.

And so you can just forget about it. Parking ticket superintelligence.

It is.

It feels like we're close.

So yeah, I think that the Millennium Prize will get solved, or proved, or delivered probably this year, if not this week. It feels like all of the different labs are like, put $1 million of inference on this. Put $1 million of inference on this. It's going to be a back and forth. And the labs are going to very quickly solve basically all of these problems. And then there'll be a search for new problems. But the conversation will move back to curing cancer almost certainly because that's the one that's so grounded. That's the one that's supposed to be tractable. It's been messaged from various labs for so long that people will want that. But there will be a longer flywheel from actually designing a cure for a specific cancer, running tests, seeing that it was effective. You can't do it with $1 million of inference over a weekend, I think. But my prediction is the conversation will go back to biotech after we basically solve math, which is a crazy thing to say, but it feels like we're there. What do you think?

Yeah, all the conversation—

Yeah, people say lab leads just stop.

Yeah. pharmaceutical industry has been curing forms of cancer or making progress against various forms of cancer for a very long time.

Gets zero credit.

Yeah, zero credit, worst reputation of any category of businesses on the planet. And so I don't think that that's the solution to sentiment around the technology industry.

Yeah, I know. I just think we get to saw someone—

And setting up this idea, like, imagine how underwhelming it will be if any AI lab comes out and makes, like, 20% progress towards one specific type of cancer, right? That doesn't— that's not gonna come across as like, oh, wow, they cured cancer. It's like, okay, they sort of advanced understanding and maybe the Yeah. Treatment process. And this type of thing is happening all the time. Anybody who's had a family member that's suffered from cancer has heard about, hey, there's this thing in trials that's showing promising results, trying to get you into it or whatever. So that's happening all the time, right? And it's not gonna suddenly change how people feel about the technology industry broadly.

No, it'll diffuse slowly and people will be back in like the, what have you done for me lately? I mean, I just saw a reel where basically the thrust was like, like, what has the pharmaceutical industry done for us lately? And I was like, GLP-1s sort of cured obesity.

Yeah.

It's sort of a big deal. And for a lot of people, GLP-1s have been a really, really key intervention in helping them fight diabetes, fight obesity, fight a whole bunch of knock-on effects that come from that. But it's not so much as the speeding train was coming towards them and the industry pulled them off the tracks at the last second. It's something where, okay, if you're living at a healthier weight from your 30s to 60s, you will probably be healthier at 60. But you're not seeing that like it saved a life today. Like a GLP-1 has yet to get that type of credit. So I don't know. It's going to be back and forth forever. But anyway, in the meantime, Somebody gave Astra a paintbrush, a robot, and a camera and asked it to paint the Golden Gate Bridge in real life. It figured out how to control the robot and progressively got better throughout its attempts. The time-lapse is sick. Here's a 1-minute time-lapse of— I don't know how to pronounce this, but his account painting the Golden Gate Bridge. How would you rate this, Jordi?

It appears better than I could do.

Better than you could do? How long are you giving yourself?

I'm saying at least by the later attempts.

Okay, the later attempts because it gets better.

Yeah.

The first one was not quite there in my opinion. Still pretty impressive. Yeah, somebody was talking about potential for a 3D printer boom with all the 3D modeling. The idea that you could— you can design. I saw someone else designed a system where you can, from a prompt, go from prompt to Lego kit basically in one go. So it takes your prompt, whatever you gave it.

Like, look at this, John. That looks pretty solid.

That's pretty solid. I think you could do better with enough time. I don't know how long that took, but it is impressive. It's fun. fun demo. And again, it, like, lives in the real world. It's not abstract at all, so you can just watch it work, watch it do the thing, and it's very, very interesting. A lot of times, what's interesting is that when you're using these models, when they have to do a task like that, they will write a deterministic piece of software to execute it. Like, I built a game where you— it was basically Tony Hawk, but you would play as a pelican on a bike, and you'd be able to do tricks and backflips, and the game worked really well. It was fun. It built it in Godot, the open source framework for game development. And then I had it go and try and get the high score. And instead of actually using computer use to go and play it, it wrote a script that would play the game flawlessly. And it looked really mechanical because it wasn't— there was no variation in it at all, like watching a normal person play. But it did get a very high score. So, I could see it doing the same thing where it generates the image and then maps the x and y axes in the plan into code and then runs that. So there's this process of what the model can do. The model might also write software to do that same thing. So there's this iteration process there.

Yeah.

Of course, people are looking at BachBench, which is the skill in writing a piece of music. And is the actual meme for my robot. Can a robot write a symphony? Can a robot turn a blank canvas— take a blank canvas and turn it into a masterpiece? And I don't know if people are calling this a masterpiece, but they're pretty impressed.

It was pretty good. I listened to this.

It was good? You're a fan of this Lily Pond 4-part chorale in the style of Bach?

One of the best reactions to Astra's release last week was Josh, who's over at Sentry. His company was acquired by Sentry. He said after Astra's release yesterday scoring 99% on Arc AGI-3, I think it's become clear that AGI is no longer some distant hypothetical. About 15 minutes after the benchmarks came out, I booked the first flight I could find to Jackson Hole. And he goes on to say he's also looking into purchasing a horse. At this point, horses remain one of the only major transportation platforms with no API, no OTA updates, and no realistic path to MCP support. So glad to see more people getting into horses, getting into the horse game. Bull market in the equestrian lifestyle.

Yeah, it seems like he was joking, but a lot of people were like this unironically. I want to, you know, touch grass, I suppose.

Did you see this? Bojan posted, it's over. Our last line of defense has fallen. Astra beat all 48 levels of the I'm Not a Robot game.

Yeah.

And this was crazy to watch. I wasn't sure if this was sped up. It seems like it's maybe sped up a little bit.

You have to play tic-tac-toe too. It's pretty good.

Yeah.

What?

Yeah, the capture defeat is going to be a whole new level of problems. At the same time, it feels like sort of remarkable that Pangram is as effective as it is, given how advanced text models are. And so if you— I mean, there's a world where it's indefensible. Was that Waldo bench? Was that find the word Waldo?

It was finding Waldo, not creating.

Finding Waldo. How are we doing on Waldo bench?

Oh, that's a good benchmark for Astra.

Yeah, we got to see. Because I bet that there's a way to have Astra generate the proper tiles. Because the problem when we generate—

Yeah, you can iterate on it more.

Exactly. The problem is that when we try and use an AI image generator to generate a Where's Waldo, it always gets the scale wrong, I feel like, or there's like two— there's like bigger characters in the foreground.

Yeah, perspective is wrong.

The isometric display is what makes a Waldo and— but there's probably ways to solve that with a little sprinkling in of deterministic code generation or Waldo generation. But Waldo Bench might be saturated now. I don't know. So it can't really be saturated because it's not a real benchmark. But lots of back and forth in the benchmark world. It does feel like we're getting to the end of it. And the results are either a big challenge, like a math problem, or just a cool demo like this.

Another good demo from somewheresy. Astra— he says, Astra and I are working on making my contact form as difficult as possible. So you actually have to fight a boss battle in order to—

This is hilarious.

Contact them.

That's amazing. This looks like a— I might try and contact some Wearsies after this. This is pretty fun. I like this. Yeah, the defeat of all of the CAPTCHAs is somewhat linked to the Instinct news. Did you see that this weekend, Jordi?

Oh yeah, the Resi bans?

Yeah, so Instinct is a personal assistant AI focused on consumer. You link it with your email, your text messages, And we've seen a lot of investors sharing posts about how they're using—

For what it's worth, I'd be surprised if anyone in the audience hasn't already heard of Instinct.

Yeah, and we talked about it before.

Thank you for reintroducing it.

But I think it's worth explaining how it works because it does seem like it triggered some bans. There was one report of an individual who got banned from Resy because they were Absolutely spamming to get reservations to restaurants.

And there was another example of somebody who was at the US Open, was on the fancam, and then wanted footage. Oh, that was really cool. And, uh, as the user experience would have been absolutely incredible, right? You're sitting there and you're like, hey, I appeared on the screen for a moment. Can you, can you go and find this?

I've had that idea so many times, like, oh, go find someone on the, on the, on the fancam.

But I think What Instinct was doing in the background was being absolutely relentless. And it seems like it was—

Yeah, especially annoying.

Reaching out to hundreds of people aggressively, you know, following up on it.

Like emailing the media contact.

Yeah.

But they did wind up getting the video.

Yeah, it worked.

And it was mutually beneficial. And you can imagine if the other side has agent intermediation and filtering and can process things that are spam-like, but there's actually a reasonable action to take, that could make it through a filter and not actually annoy anyone on the other side. So that was a very— that was probably a glimpse into the future and, like, a cause for optimism. I could see that being popular. But people are doing all sorts of stuff. The classic example is, like, go save money, get refunds, do that sort of stuff. But people are having fun with that. People are— so many demos. This 3D website that pulls apart the male anatomy into 2,200 pieces. 234 modeled pieces. I shared this with a friend who I was trying to like explain in one video what to get like their wheels turning. I think this is a good example. I wonder how much of this was like individually modeled versus pulled from an existing model and then just animated, but certainly works. Certainly works well. Yeah, I wonder. Oh, Anish made a game.

That's fun.

Astro is Opus 4.5 for games. Made Contra but photorealistic this weekend. Link to play below. And it's up on Vercel. I wonder if we will get an actual boom in Steam. It feels like Steam will be going through something similar to the Kindle Direct Publishing store where there will be flooded with inbound. Didn't we see this with the App Store too, where there were like a disproportionate— like the number of App Store submissions for Apple, like text—

Yeah, totally overwhelmed their systems.

But no real breakout apps. Like the top of the App Store is still like ChatGPT and like free money or whatever, Temu.

Yeah, it's evolved a little bit. The company that I'm confused that's been consistently in the top 10 is this company Vinted.

What's Vinted?

Pre-loved marketplace. Sell and buy secondhand clothes.

People are buying clothes. It's a shopping app. Everyone needs clothes. So big TAM, you make it to the top of the store. But yeah, like, you'll be able to vibe code a game. You'll be able to get it on Steam or, you know, in the App Store. How do you get the distribution flywheel going? How do you innovate to actually break through? How do you make something that, you know, goes viral, gets traction? Because the, like, the, the long tail is going to get a lot, lot longer. What do you think?

I feel like for a lot of these games though, you should just put them on— keep, like, keep them on web because it's much more accessible. Not everyone has Steam. You have to actually download, like—

I feel like web games are still, like, not that fun.

There were some web games I was playing this weekend.

How long?

I played one.

Did you play more web games this weekend or did you play more Oculus VR, Meta Quest VR, when you had the VR headset for an unlimited amount of time, but let's call it a weekend?

Uh, okay, I was probably playing more VR.

VR?

I was playing Call of Duty.

You played Call of Duty in VR?

Yeah.

How long? A couple hours?

Yeah, probably a few hours.

Not bad.

But I was playing— well, when I was making a game too this weekend on Astro, I spent a lot of time.

Yeah.

So does that count as playing the game?

No.

Can you drop it in the chat?

No, I think, I think, I mean, I do think that playing a game that's more you designed yourself or for a small group chat, it's the same thing as the images. Like, the AI images I generate are not broadly beautiful. They're not going in the MoMA, but like, they will often be funny to me and 3 other people, right?

Low TAM.

Yeah, low TAM. But that's the beauty is that, With lower cost, you can go after lower TAM opportunities. And so, I mean, this happened, what, a couple months ago? Somebody sent us— I mean, Jeremy Gaffan Simulator is the classic example. Someone sent us Jeremy Gaffan Simulator version 2, which was very low TAM in the sense that it was just in-jokes with me and a couple friends. And the long tail of these things will continue to flourish. And then, of course, people will use them in— You know, real businesses built on IP, built with distribution.

You know what time it is, John.

What time is it? Time for an ad? Console. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets.

It's trailer time.

Okay, it's trailer time.

We got 2 trailers that hit the timeline. Let's start with Nathan Fielder and his new Film featuring none other than Elizabeth Holmes.

Elizabeth Holmes, filmed before she went to prison. I can promise you that. A24's been on a tear. I don't have anything to deceive you on. Of course I'm not deceiving you. I'm engaging with you as a human being. Why would I deceive you?

There's no—

there's no reason for me to do that.

Okay.

The acting is so good. Okay, good shot. How long is this being real right now? Uh, I'm always being real. The team is stoked. Looks beautiful. Good sound design.

Uh, saw a post, somebody saying it feels like I'm living in a simulation because this media product is like perfectly with— yeah, you would think that you would You would think that a movie like this wouldn't get made by, I don't know, A24 specifically, because it feels like the TAM is just like 20,000 people. But in—

The Theranos story was really big. She was on the COVID of Time magazine, I believe, and cover of major magazines. It was all over the news when it happened.

No, it's a lot bigger than you would—

The John Carreyrou book is just a fascinating read. It's a real page-turner. And so I think a lot of people read the book. The adaptation of the actual play-by-play story, not the documentary, but the— what do they call it? Biopic, I guess. That, I think, did not break through in the same way. But it certainly continued to elevate her persona. And then the posting has kept her relevant.

Yeah, there was some speculation that maybe it was Nathan that got access to the account.

Completely disagree with that. I do not think that's at all— I think that would violate some tenet of what it means to do a documentary. I mean, Nathan blurs lines all the time in terms of how he interacts with his subjects, so it's certainly possible, but I just don't think that's reasonable at all. I think they're her poems.

How much do we know outside of the trailer though? Like, we don't know—

Well, this was leaked something like a year ago from a journalist who got the story, and it was widely denied that it was happening, and he was very upset that he wasn't getting enough credit for being correct here. He ultimately was, but yeah, all we know is that they filmed the documentary in the lead-up to her actually going to prison after sentencing. There was a pretty large gap between before when she had to report to the prison. She gets out in 2030, so coming up 3 years. 4 years away. And so I believe even in a week you can get a lot of footage if you're spending the full, you know, every day with someone. And then there were a number of visits to the facility by Nathan that were recorded, probably talking through glass or talking at a table, something like that. So I think some flexibility there, maybe some phone calls, maybe some correspondence in the aftermath, but We don't know what's up.

You thinking what I'm thinking?

What?

SBF, Nathan Fielder.

Ooh.

Sequel to this one.

Yeah.

This just becomes like a—

The problem is that he's already in jail. So the question is just what else does Nathan Fielder have in the can? Because he's obviously thinking years ahead since he filmed that exact scene in, I believe, 2022 or 2023. And so who knows? There's a rumor that he's secretly running a, like, a boy band right now. There's this rumor that there's a boy band that's popular on TikTok. This is real. That's popular on TikTok that's entirely invented by him.

BoyThrob.

BoyThrob, yeah. Are you familiar with this?

Really?

I thought you were going to say he's Nathan Fielder's SD kit.

Yeah.

So yeah, I don't know. I mean, these projects will emerge. And obviously, they're taking longer and longer because they're bigger films now.

Yeah, after the rehearsal, it was hard to imagine how he would one-up himself. And it feels like he may have done it.

Yep. He found a new tall mountain.

We're still— stay with us in the trailer zone. When we come back, we're going to be talking about Artificial.

Let me tell you about the New York Stock Want to change the world? Raise capital at the New York Stock Exchange. Just do it.

Pull up Artificial.

Artificial.

Trailer dropped just this morning. What is the image of the future that you see?

Wait, pause.

Was that a Valkyrie?

Yeah.

Okay.

Continue. So the visuals are already incredible.

They really are. I'm pretty sure that's a Valkyrie. It might be a Valhalla. Is that not a Zink Lucas Zinger?

No.

No, no, no.

No way.

No.

No, because you can see the suspension.

Yeah.

That front section.

Okay.

Anyways, continue. What is the image of the future that you see?

We've created a machine that will solve the world's problems.

We don't teach it. Gate check, gate check. You nailed it.

Gate check? Oh, the gate. Yeah, for sure.

Open Pandora's box. Together, the future is inevitable.

Countries will fall.

Industries are gonna collapse.

The voice is still Eduardo Saverin, but the styling— and the crazy thing is I have that exact backpack. Get to be the ones— I think I got it recommended by MKBHD. To shepherd people into this new world. Lots of cameos. Not cameos, but I'm so excited to see it.

Dramatic, dramatic.

Be interesting. It'll be interesting to see the the takeaway. People in the chat are back and forth. Is it too soon? Is it unnecessary? Is it going to be? something that, like The Social Network, is divisive, but also, like, it's oddly inspiring to a certain cohort of entrepreneurs.

We'll see.

We'll see what the reactions are.

Yeah.

Do you think that was— do you think Hollywood took away any lesson from The Social Network in that they were trying to frame Mark Zuckerberg as this, like, generally, you know, Troubled person, and then a lot of people came away feeling like, wow, I wanna be like Mark Zuckerberg.

Yeah, sort of an unintended consequence of it.

I don't know.

I don't know. I mean, they're going back for a second scoop of the ice cream with The Social Network, The Social Reckoning. Is there a trailer out or just images? I think we saw that too. So, it'll be interesting to see what happens there.

But get your tuxes ready. Going back to back. Artificial, Social Reckoning, and then capping it off with—

You Can See Everything.

Yes.

And then also, uh, the, uh, the To Catch a Predator movie. That's gonna be a good one.

Yeah, outside of tech, tech world.

Yeah, I guess you're right. Yeah, but still, uh, triple feature. Yeah, it just feels like it's a, it's a year of like niche movies that I expect to be breakouts. I guess, like, along with like Backrooms and Obsession and then these 2 movies and then the documentaries. There's a lot of things that are like oddly specific, but I would expect they do well. But I don't know, we'll see. The Social Network did very well at the box office.

And while we're at it, let's stay in the trailer zone and let's pull up The Social Reckoning, the official teaser. Tyler Culbertson is calling it pure cinema.

Pure cinema.

But I am—

I am here to help Facebook, not hurt it, okay?

You sent me a message.

What would you like to talk about?

The chairman gavels the session to order.

You'll read your opening Statement, which we'll skip past for now. That's a separate session. And which one is this about? I don't, I don't think this is based on the politics, right? It's not Cambridge Analytica. It's the Frances Haugen teenage girl whistleblower thing, which honestly looks a lot different in the backdrop of the addiction trials. I wonder if they were trying to get this out before the trials.

Aren't you a tech reporter?

Ish.

Ish?

These guys are counting on the next round of congressional testimony to make you likable, Mark.

I'm happy to lend a hand, but I think you're doomed.

This company and that guy are playing an unprecedented role in our lives.

The fire hose of bad information you are injecting into the air supply is becoming jet-powered.

I'm a free speech advocate.

I'm not the one who's lying, and I'm not stopping them from seeing someone who is. Anxiety, depression of teenage girls got worse as a result of time spent on the platform.

Senior leadership knows and is doing nothing.

I know there are easier enemies to make.

The mafia would be an easier enemy to make.

So what would you need?

To stand up the story, the internal documents.

This is a material violation of my NDA.

We're twice as big as the biggest country on Earth.

Earth.

We're not frightened of Congress or post-government around here.

Please.

Who is that supposed to be?

Please let me quote that.

We have 102 hours to get everything.

She's going to get sued into small pieces.

I don't want to be made an example of by a guy with unlimited resources.

Harm, I promise you, is imminent.

Yeah. So chat is calling it out, but I don't think that— I think this one— I think this one's going to flop for sure. Because of the lack of supercars? Well, lack of supercars. And this is just not top— this story is no longer top of mind.

Sure.

Yeah. People want to hear about like the talent wars. They don't want to hear about like this kind of stuff.

They do. The talent wars. If there's no Alex Wang character, I'm not, I'm not paying good money.

Sass Shane says, ooh, technology is so scary.

Yeah.

I don't know.

Yeah, I mean, they should do more tech movies. Maybe, uh, maybe SaaSpocalypse.

SC says it will flop because no Brazilian jiu-jitsu. Yeah, this was in the pre—

that's true—

jiu-jitsu era.

A lot of stuff making it in the script. Yeah, UFC.

Uh, if there's The Social Network 3 and all the scenes where Zuck is talking, he's in the middle of—

how is it 2026 and we don't have a Salesforce movie?

Seriously, like the, the comeback, the SaaSpocalypse, or at least a documentary on the relationship that he has with the dolphins.

That would be great.

And the wildlife generally.

Yeah, I'd like a CrowdStrike movie. I'd like a MongoDB movie. Let's do it. MongoDB, what's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it.

Here's a concept, Dolphin Force, how one man tamed wild dolphins.

It's maybe the best origin story of a tech company ever. I don't know if there's a better one. It's the best. It's just pure bliss, you know? Pure positivity.

And for those that don't know—

Just spite company.

Benioff famously conceived the vision for Salesforce in 1999 while swimming with a pod of roughly 100 dolphins off the coast of Hawaii during a sabbatical. Feeling one with the pod, he envisioned cloud-delivered enterprise software Leading him to quit Oracle and secure two million in seed funding from his mentor Larry Ellison. Yeah, incredible.

Be great. Be great. Anyway, in other news, lots of craziness in the AI world. Lots of new models. But the jobs apocalypse has been officially postponed by The Economist. An AI jobs boom is here, and people have. A wide range of takes on this. Is it just postponed? Is it permanently postponed? Kevin Bryan summed it up well. He said, "I am shocked, shocked." So he's being sarcastic, of course. Shocked to find out that a productivity-enhancing, investment-supporting technology is good for workers. Knowing nothing else, this should be your prior because this is how productivity and competitive markets almost always works. A new technology comes out; it supports investment; it enhances. Everyone hires to go after all the, all the different battles. But The Economist breaks it down in a lot more detail with a bunch of charts and a bunch of data. So they say perhaps AI will eventually make many humans unemployable, but there's no sign of it yet. On September 4th, the Bureau of Labor Statistics reported that the American economy added 162,000 jobs in August, far above expectations. The unemployment rate Just 4.1%, lower than almost 90% of months over the past 50 years. Young workers, often cast as AI's first victims, are holding up remarkably well. The gap between unemployment among 20 to 24-year-olds and the overall rate is close to a multi-decade low. Some companies and workers are being severely disrupted by AI. Hiring professional and business services is running 10% below the average from 2015 to 2019. Tech giants like Microsoft and Meta are trimming headcounts as they reorganize their businesses around the technology. Smaller firms such as Block, which is sort of a crazy— smaller firms, but it is smaller than Microsoft and Meta. Block, the owner of Square and Cash App, and Intuit, the maker of TurboTax and QuickBooks, are replacing people with bots. American companies have announced some 16,000 AI-related job cuts That's a month on average so far this year, according to Challenger, Gray Christmas, an employment consultancy. What a funny name for a consultancy. But AI-related layoffs get lost in the churning job markets where— so remember that number. So 16,000 AI-related job cuts a month. In a typical month, the US workforce just churns 1.7 million jobs. And then they add 1.8 million jobs. So there's net gains, but you're looking at less than 1% of the layoffs that happen in the economy or job losses or firings. Less than 1% is AI-related, which is what The Economist is pointing out here. And the evidence so far is that AI is already creating a lot of jobs to replace those as it has destroyed. The vast sums pouring into data centers and power generation have set off a race for construction and infrastructure workers. AI startups are hiring like there's no tomorrow. Incumbents are racing to keep up with new AI roles and making some workers more productive. AI may be increasing the demand for their services. Add it all up and The Economist estimates that AI so far has created around 1 million new jobs in America. That easily exceeds the roughly 200,000 layoffs attributed to AI. Since mid-2023, and it appears more than enough to offset weaker hiring in many back-office roles. America's AI infrastructure splurge has created many of them. The spending on the kit needed to make AI run, from chips and servers to data centers, cooling systems, and power, is roughly $500 billion a year above what it was in 2022 when the world got to know ChatGPT, calculates Goldman Sachs, a bank.

Thank you.

The Economist. Data center construction alone is proceeding at an annual rate of more than $75 billion, nearly 60% higher than a year ago according to Census Bureau data. That building spree requires armors of workers, electricians to wire them, HVAC specialists to stop racks from overheating, grid engineers to hook them up to the power supply, and technicians to install and maintain the machines. The hiring boom is visible in the numbers. The Economist tracked 5 industries at the heart of the data center buildout. From electrical contracting to equipment manufacturing. Since 2023, employment in them has risen by roughly 320,000 more than broader construction and manufacturing trends would suggest. The BLS expects utilities to be the fastest-growing big sector between now and 2035. Not all of those jobs owe their existence to AI. Grid upgrades and other factory building matters too, but lots of them do indeed. A jobs website finds that data center vacancies have more than doubled in 2 years as job postings overall have fallen. LinkedIn, a social network for strivers— these are so funny.

Just firing shots.

Goldman Sachs, a bank. LinkedIn, a jobs website, a social network for strivers, estimates that nearly half a million data center jobs were created between 2023 and 2024. 5 in America. The scramble for workers is showing up in paychecks too. Indeed finds that installation maintenance jobs at data centers advertise wages about 40% higher than comparable work elsewhere. Official wage data tell a similar story. In the year to June, average hourly earnings rose more than 13% in electrical equipment manufacturing and nearly 8% among electrical contractors. Even with the pay gains, workers are still not easy to find. On a recent visit to a transformer factory. Donald Livens of the National Electric Manufacturers Association asked the company's chief executive what help she needed most. Can you come in here and help run one of our lines for me? she replied. It's not just hard hats that are proliferating. AI is also creating a new class of white-collar jobs. Engineers build the models, data annotators— Tyler's clapping for the white-collar job creation. Data annotators label their inputs and judge their answers. Forward-deployed engineers adapt them for customers. Newly minted heads of AI decide what companies should do with the technology. And there's some interesting reporting in the Journal about Google and Accenture teaming up to do more forward-deployed engineering. They're deploying some— I think there's 1,000 people on this one team. And this is an example of like— New roles for AI diffusion. Some of these roles barely existed until recently, says The Economist. Many are quickly growing in number. Postings for heads of AI, AI engineers, and directors of AI have roughly doubled since 2023, 2024. The numbers are starting to add up. Preliminary research by economist Gad Levinon at the Burning Glass Institute uses the research outfit's career history database to identify jobs that would not exist without AI, whether at AI AI-native firms or because there are AI-specific roles at other companies. He reckons roughly 1% of professional jobs are now AI jobs, on the order of 1 million positions in America. In computer occupations and life sciences, including researchers using AI to discover new drugs, the share is 4 to 5%. LinkedIn's own analysis points to roughly 640,000 new AI-specific jobs between 2023 and 2024. So, Jevons paradox. The Economist tracked employment in professional occupations closest to the AI boom— engineers, software developers, mathematicians, and data scientists— and compared their growth since 2022 with professional employment overall. These roles have added roughly 730,000 jobs above the trend in recent years. AI will not have created every single one of them, but it almost certainly created a few. Third source of job comes from productivity gains. AI allows lawyers to draft contracts faster and analysts to comb through financial filings in minutes. If higher productivity lowers the cost of professional services, it is possible that demand for them can rise enough to create more work overall. Nakia Beer was talking about this this weekend. He used Astra to create 3 different potential deck remodels, redesigns, building expansions to his home. And it was interesting because it's a lot of work that you could say is like, oh, that's going to displace an architect. I don't think the models are at a place where you would trust a deck that's built in Blender from even the best model. I was finding in my Blender modeling tests A lot of misaligned beams and things that weren't quite right. But in terms of visualization, in terms of getting excited about a project and getting to that next stage of saying, hey, okay, I have a really solid vision here. I can imagine what I want to do. Let me actually go and do that. There's this odd diffusion that happens in the economy. I was talking to Sagar and Jetty about this, the credit card effect where— Credit cards had a big effect on the economy in a very, very boring way. They just slightly lubricated the machinery of the—

Global commerce.

Global commerce, basically. So you're just like, okay, if I send a wire, I'm not going to be able to get the money back. But if I put down my credit card and it doesn't show up, I can probably get a chargeback. And so that instilled just a little bit more confidence that what you're buying is accurate. And it enabled more commerce, right? And so the same thing is starting to be true here where people can say, okay, I'm ordering shoes. Let me find the shoes that fit me perfectly, that are the right price, that will come on time, that are from a reliable manufacturer, aren't going to fall apart. And if you get a very solid report back that gives you confidence, you make that purchase just a little bit Sooner. And if everyone is doing that, it winds up speeding up the machinery of global commerce, which is incredibly boring, but incredibly valuable.

Totally.

Because you're—

Yeah, the example— one of my favorite examples is you were looking at a number of different houses, all of which were not necessarily in the condition where you're like, great, I want to move in here. And you just dropped the Zillow links into Codex. And it made you an entire website that you could walk through with before, before the current state, and exactly how they would evolve based on your stylistic preferences.

Exactly. So oftentimes, you'll— like, a fully remodeled house will command a 20% premium, but you can get sort of sucked into, okay, well, the newer— the one that's already been through the remodel, it's just visually more striking. So you're like, oh, I want to necessarily pay up, when in fact you might want to buy the cheaper do the remodel, and then wind up with the same end product, and maybe capture some of that value yourself. And so there's a lot of different places, uh, you know, and a lot of times the, the conclusion is very underwhelming. I mean, we were talking to Nick about this where he was taking a picture of his room, and I saw a lot of people doing this online— take a picture of the room, make sure I've decorated it properly, that the bed is in the right spot. And I think the result for you was like, Yeah, it's basically as best you could do. Good luck, dude. Which is hilarious. But at least it gives you confidence that, okay, yeah, like I really shouldn't order a bigger bed. The bed I got is the right one. And if you do that on the way in, you wind up moving things faster. We covered Navier Stokes. I don't know if there's more to cover there. The chat's asking about it. We covered it at the top of the show. I mean, we also have Greg Brockman joining the show, so we'll ask him about advances in math, what's going on with Navier-Stokes, the back and forth there. He's joining at 1:10 PM Pacific, so stay tuned. And we have Scott Wu as well, who is an IMO gold medalist and a fantastic mathematician. He can, of course, comment about this because he, at Cognition, is in a unique position of both being incredible at math but also, like, Like, Cognition's whole business is not in the theoretical math space. It's in AI diffusion. It's in getting work done in the enterprise for real companies. And so I think he's in a unique position to both ground the conversation around what it actually means, how impressed is he by these solutions, and then also what does it actually mean for Boeing? Like, will they be able to make more planes more efficiently? more reliably. Like, that is a valuable thing that we actually want to see in the real world. So—

Something happening in the real world. Former F1 CEO Bernie Ecclestone was stopped in Portugal at the airport for bringing in a shotgun from Switzerland. He's 95 years old, and this is actually the second time he's been stopped at an airport. In 2022, he was arrested in Brazil for illegally carrying a gun while boarding a private plane to Switzerland and was ordered to pay a €1,000 fine to local authorities. So rough weekend for Bernie. But at this point, I, I think he's like, sorry, I'm 95. I'm not going to stop bringing guns everywhere I go on, on, on planet Earth.

I gotta do it. Well, let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish.

Just let Grandpa carry his gun. That's all. Um, I think in other news, Wimbledon does not plan on providing credentials to influencers next summer in hopes of avoiding the issues that affected this year's US Open.

What were the issues?

We didn't influence that. Players have called for spectators to follow the sports etiquette after matches were interrupted.

Did they specifically call out taking a picture of the court and putting logos all over it using AI and then posting that on X and getting a couple hundred likes?

Looking at you, Nick.

He's locked in. Looking at you. He's working. We're just making jokes at your expense. Don't worry about it. You're good. Was there anything specific? That was tied to this?

I think it was a number of issues. There were also complaints of— one of the players was complaining about the smell of marijuana in the stadium, which is interesting. But yeah, next year Tyler will be going to the US Open in disguise.

Full Hollywood makeup?

Yeah, you can look forward to that.

Yeah, I wonder what are they gonna do just Just not give free tickets to influencers or actually block people who are influencers?

I think what was happening is they were probably giving like media/press passes to influencers.

And then the influencers were like doing too much stuff, like getting in the way of fans, creating scenes, because they're sort of like running around with like gonzo filmmaking equipment as opposed to like something that's a little bit more cordoned off and like polished. It's different when you have a red carpet and you're doing, you know, proper interviews. That sort of lands a little bit better. Anyway, Shiel Monnat is calling for a revisitation of the Citrini piece. Agents become the default demand-side interface and transfer economic rents from incumbent intermediaries to whoever controls the agent, which does seem correct. And so, with instinct, who should you be long? Who should you be short? There's an interesting dynamic where Anytime something happens in AI, there's a lot of people that go— I think Buco pointed this out, that everyone will go and ask AI, okay, what are the longs? What are the shorts? And so, that creates an amplification in the market.

Yeah.

People are, oh, piece together this new trend. I want exposure to this trend. And so, things get overheated much more quickly. But yes, this idea that if it's a restaurant booking platform and your AI agent can call the restaurant and get a booking just as easily as they can go through a platform. Now, Resy has exclusives, so that creates more of a complex scenario, but by default, if you can just write an email or make a phone call or send a text message, an AI agent should be able to disintermediate those platforms. But that's not true for Uber. That's not necessarily true for DoorDash, where they're like, it's very difficult to find a person to go and pick your food up or pick you up personally because they need to be very close if you want the car right then. So, Sheel says, how effed are you in an agent disintermediating world? Completely depends on how much you control— how much control you have over supply. To answer your question, Buco, you may not have a choice in whether you allow bots. Expedia can block agents, but it doesn't control the hotel inventory. It If Expedia stops letting agents book, they'll go through Booking.com, Google, or direct. Resy has more leverage via exclusives, and I'm sure Expedia has some exclusive program, I'd imagine. But if agents become a major source of diners, restaurants will want to be bookable via agents, and reservation systems that support agents will win supply, someone will come along with a cheap direct booking sell. AI can disintermediate discovery and customer ownership, but Resy can survive. But can Resy survive as the restaurant's reservation infrastructure? But that's a much smaller rent pool unless the inventory network remains differentiated. Uber is much harder to disintermediate. It actually controls a valuable network of drivers, dispatch, pricing, and payments. Why would they allow agents is basically they can't afford not to. After an AI agent takes over your consumer interface, how much economically essential stuff are you doing that only you can provide. For Expedia, it seems tough to me. So, Sheila is worried about Expedia. Anyway, was there anything in the chat you wanted to run through? Because we have our next guest joining in just a few minutes. We have Dan Wright coming back on the show from Armada, building modular AI supercomputers in a shipping container. A supercomputer in a shipping container. Apple's coming out with new phones. We're getting a folding phone probably this week.

Finally.

It's gonna be exciting. I think it'll sell well. And it's been a while since there's been a visually differentiated Apple product.

Yeah, I think it's a pretty easy sell. Hey, you use this thing for 3 to 8 hours a day, or if you're Taylor Lorenz, 18 hours a day. Would you like a bigger screen? I think a lot of the answer for a lot of people. And would you like to have some differentiation versus your peers that have Just one simple screen. Yeah, I think a lot of people are gonna opt into that.

I wonder how it's gonna look. I've seen a lot of like leaked renders.

It looks—

a lot of the renders look squarish.

Yeah, not good.

Just throwing me off.

Yeah, not— I feel like they, they're gonna surprise us with something. They've, they've— it feels like the, the news and the facts and things that can go over text message sort of leak out, and Mark Gurman is Like, I saw a stat from Eric Newcomer that Mark Gurman has an order of magnitude more scoops than the next biggest journalist on Techmeme or something like that.

Yeah.

He's like the power law tech journalist. You hear that? Seriously. But it feels like people might— The most journalistic force has been sent live. Discuss things with Mark and say, oh, yeah, we're not working on a car anymore. Keep that off the record, but I'm letting you know, and he'll report it out. But people will stop short of sending him an image of the new phone. And I think he's been pretty good about that.

It was actually Josh over who says we don't talk enough about how Mark Gurman is the most cited, most market-moving tech journalist by nearly an order of magnitude. Yeah, the whole industry of Apple reblogs and day traders live off of his scoops. It'd be like if LeBron won 40 championships.

That's a really good post. We love Mark Gurman. Well, there's one more story we gotta cover before we go into our guests. A man drove a Ferrari Purosangue from South America to Alaska. This is a crazy, crazy story. This, the Transamerica Race, has been talked about a long time. It's really, really difficult with the Darién Gap. the gap of like wild forests that exists between North and South America. You usually have to put your car on a boat to get around that part, and I think it still counts, but it's still a very treacherous road, a lot of weather conditions. If you break down, you might be very far from anything. But a man named MJ has taken a Ferrari Purosangue far beyond where most owners would ever consider driving one. His goal was to travel the length of the Americas taking the V12 Ferrari from South America all the way to Alaska. He's now attempted the journey 3 times, putting roughly 800,000 kilometers on the Purosangue in the process. His dog joins him for the journey. The first attempt ended after an accident. He brought the car to a dealership in Santiago. On his second try, they made it all the way to Alaska. For the third, he reversed course with an even bigger goal, attempting to set a record from Alaska to Argentina before a major road closure forced the run to end in just 2 days. Through dirt, snow, and thousands of miles of pavement, the Purosangue has effectively become a long-distance expedition car, with Rico, his dog, riding alongside him through it all. What a—

Heartwarming.

Heartwarming story.

And a beautiful ad for this naturally aspirated V12.

For sure, for sure. Look at that.

Look at that shot. So good.

Oh, I didn't realize it was a Chihuahua. That's funny. That's probably a more convenient dog to have in a car for a really long— you don't want a big dog that needs to stretch its legs. You want the dog to be able to just run around the cabin, I think. If you're— it's a good road trip dog, I think. Anyway, we have our first guest of the show, Dan Wright from Armada, in the waiting room. Let's bring him in to the TVPN UltraDome. Dan, how you doing?

There he is.

I'm doing great. How are you guys doing?

We're doing fantastically. Welcome to the show. Welcome back. What is new in your world? Walk us through the latest and maybe just since it's been a while, sort of reintroduce the shape of the company, the mission, and where you guys have gone in the last year.

Yeah.

So Armada is the hyperscaler for the edge. We build the infrastructure for the 70% of the world that doesn't have the big hyperscale data centers today. The mission of the company is to bridge the digital divide and make sure that we have AI everywhere, wherever we need it. But first, you got to have the infrastructure there. And so that's what we're doing. We've been very busy. We just raised earlier this year $230 million at a $2 billion pre. We launched Galleon Forge 1, which is a factory with our partner Johnson Controls in Gilbert, Arizona, where we're now continuously manufacturing these modular AI data centers called Galleons that we've built.

Yeah. How is power usually solved for when you deploy these?

Yeah, so a lot of times we're deploying them where there's already stranded power.

Sure.

A good example of this is a couple of weeks ago I was in New York and we did an event at the New York Stock Exchange where we were sort of fast following on Jensen's announcement that compute is now an asset class. It's a new asset class. The New York Stock Exchange is actually making that now like something that you can invest in the same way that you can invest in electricity or other types of commodities. And we were there with our customer in Norway called Fossafall that has tons of distributed sites all over Norway. I'm actually going to be in Norway with them next week, but they also have them in Finland and in Sweden. And it is largely renewable energy, so it's 99% hydroelectricity. They can just plug in our AI factories and then scale up quickly. And one of the things that, you know, makes that easier is that we've recently announced some larger form factors. Last year we announced Leviathan, which is 2 megawatts per unit. And when I was in New York, we announced Orion, which is our newest form factor. That's 10 megawatts. per unit. When I was launching Leviathan, everybody said, that's great, but how do we scale up even faster? And so now Armada can credibly say, wherever you have power, we can plug in and we're the fastest from 0 to 200 megawatts anywhere in the world.

Talk about why, for example, you know, one of these companies needs to have that compute actually at the edge and why it matters across different industries.

So one is latency, and this comes up in a lot of conversations. I'll give you another real-world example. We work with the state of Alaska, and they don't have the big hyperscale data centers there. So the first data centers deployed were ours. We did this last year, and we're still working closely with them and kind of scaling up with them. They had 28 hours of latency to process data from drones for emergency response.

Mm-hmm.

for avalanches and floods. We also deployed last year with the Navy in the middle of the ocean, very similar types of use cases where if you have a large distance between the source of the data, where the data is being generated, and then where it's being processed, the data sort of becomes worthless because avalanches and floods and threats in battlefield scenarios, they don't wait days or even minutes. You have to be able to use the data in real time. Another big driver of this is sovereignty. The shorthand for our value prop is the 3 S's: speed, scale, and sovereignty. And the sovereignty piece is really important. There's this global trend that's going on around sovereign AI. Everybody wants to be able to take the latest models, but they want to be able to fine-tune them to really sensitive datasets that they wouldn't send to the cloud and then, you know, have like a sovereign AI infrastructure that they actually own. And that's what Armada enables.

So, yeah, speed of delivery. Yeah, that makes a lot of sense. Is hydroelectricity underdiscussed right now? Is there an opportunity there in America or abroad? I mean, we talk a lot about solar where it feels like there aren't as many like big winners yet or big like hot startups that exist in nuclear. That's very exciting. I've, you know, we've heard about wind, but no one is really talking about like, let's just do another Hoover Dam or something. Is that possible? Are you optimistic about that?

I think so. I think, I think it should be talked about more. And I think in general, stranded energy should be talked about more. We're doing these projects all over the world. As an example, you know, the Niners are playing in Australia this weekend. We're doing an event there with another partner called WindDC that we work with there. And they've got a huge amount of stranded wind and solar energy. Australia is big in wind and solar. Norway's big in hydroelectricity. But this is happening all over the world. There's these stranded pockets of energy. And they last year had to curtail 7.2 terawatt hours of energy in Australia because the grid's completely overloaded. But they have this stranded power that we're just bringing the infrastructure directly to, and then you can scale up into the hundreds of megawatts. And then ultimately, same with Fossil Falls, they'll scale to over a gigawatt over the next few years and we can scale with that.

Yeah. How flexible do you want to be around like the actual chips that go into the systems? Because when I hear latency, I feel like Cerebras, Groq, these like faster systems. There's Talus, which I think AMD just bought, where You're baking the weights on and you can actually get to an inference speed for certain AI workloads that would actually benefit from saving 100 milliseconds. Whereas if you're putting a bunch of NVL72s in a data center, you're going to be waiting while it's cooking.

Right.

And the last mile is going to be negligible.

Yes.

So again, it's all about speed, scale, and sovereignty, and then giving the customer the choice. So we say, okay, well, what workloads are you trying to run?

Yeah.

And then based on that, you can sort of right-size the infrastructure to the actual need and where they expect it to go. And a lot of what our customers want to do is they want to try different things. They want to try, you know, obviously the GBs, but now they're looking at the VeraRubins, they're looking at things from other chip companies as well. And so the nice thing about our galleons, again, we're manufacturing It's not like construction. And so as you want to try different things, we can just make, you know, quick iterations on the design and ship them out so they can try different things. And then what they like, they just order more of it and we scale up with the demand versus with the traditional data center.

Yeah.

The downside is not just as we were talking about the power, you got to figure out the power situation. We take advantage of the power that's already there. But what we also allow them to do is scale up with demand so that they don't overbuild or build the wrong thing.

How are you thinking about staying aligned to, I guess, why and how the company started and the focus around this edge compute versus the natural pull from the market where if somebody comes to you and says, okay, now I want the 1-gigawatt data center and we think you guys can do it, just running the numbers on that and resource allocation, I imagine there's a natural pull to go bigger and bigger when— And anyways, I think that's kind of a good problem to have, but it'll be interesting.

It was actually super simple for us because one of our company values is heal the customer's pain first. We're like obsessed with customers and what their pains are and how do we solve those. And so, the kind of evolution of our Galleon product line has all come out of conversations with existing customers. Where they've said, okay, I want to use your edge galleons, the smaller ones, say sub, sub a megawatt for inference. But then they came to us and said, I also want to do fine-tuning of all these models, latest open-source models, closed models, my own models on these proprietary datasets without sending them to the cloud. And so what we enable is sovereign AI factories where they can do both. They can take the latest models, open source models, you probably saw what NVIDIA just did with Hugging Face and with Poolside. I think we're going to see a lot more around open source. They can take models from the OpenAIs and the Anthropics of the world. We can help them fine-tune those to these sovereign datasets and then push them to all the edge nodes to run. And then you can do what's called federated learning.

Yeah.

Where you're actually fine-tuning that model on that data at the site. And then you're using it to improve your core model and then pushing the updated model out to all the sites.

Yep.

Which has the benefit of it's more cost-effective, it's more secure, and it enables you to take advantage of lots of different models, experiment with different things, and see what works.

Yeah. Well, congratulations on the progress. Thank you so much for coming on the show.

Great update.

We'll talk to you soon.

Great to see you guys.

Great to see you, Dan. Congrats on the progress.

Let me tell you about Shopify. Shopify is a commerce platform that grows with your business and lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents. Who do we have next, Jordi?

The Souffinator.

We got Eric Souffert, the founder of Mobile Dev Memo, live with us on TVPN, on YouTube, not on Netflix yet. Should we be considering switching teams?

And Eric, just for what it's worth, every time you come on the show, I call John afterwards and I'm like, hey, Eric, Eric's in my top 3 all-time guests on the show. It just fires us up to talk with you every time.

Good to see you.

And it's great to see you.

Yeah, good to see you.

So I told my wife, I'm going back on TVPN. And she said, is that the show with the 2 handsome hosts? And I said, yeah, but when I'm on, there's 3 handsome men. And it was like that meme with Natalie Portman where she's looking at the guy and I was having to say, right, right. It's like 3 handsome guys on the show, right?

So I think you're ready for Netflix.

I think Netflix would be happy. It would be lucky to have you.

But what would the— what would the economic dynamic be? Because Netflix and YouTube are both going for exclusives now. What's actually playing out? How much are they at each other's throats? How much are they going to converge in the way that like TikTok and Instagram and YouTube are all converging. Is there going to be another convergence or is there or is there actually a stress point there where YouTube is trying to be both Instagram and Netflix and they maybe can't do both?

Well, yeah, I mean, they are converging. So this is a really fascinating dynamic right now in streaming. I've been following it with a series called Netflix's YouTube Opportunity for roughly a year. Right. And so what I wrote about first was they brought Miss Rachel on.

Right.

And so what they did was they brought Miss Rachel on, but they didn't give her— they didn't, they didn't buy new content, right? They just paid her to bring her existing content over. So you've got this proven base of content that has however many billions of views. I know my children probably account for some several billion. But you've got this proven case of— you've got this proven base of content where there's essentially no risk. Like, you know, there's an audience, you know, it's popular, and you pay some amount of money to bring over this existing content. So what's actually really great for the content creator too, because they don't to produce anything new. They just have to chop— they chopped up the existing content into like a season. So they packaged it like a season. And they also did that with Danny Go and they also did that with Mark Rober. And so my point was, look, they've, you know, Netflix had been— they had been executing tremendously well for the past several years. They had commanded extreme pricing power. Right. The premium service is going at almost $30 a month in the United States now.

They've been—

they've been— the advertising tier price has been tracking with, I think, is the ARPU differential to make it equivalent to the next highest tier. So they've been doing a good job of maintaining that pricing power. My sense is they did too good of a job. Right. And so they made it really hard to compete in the space and that pushed everybody to the bottom. That made everyone take the exact opposite approach, which is like, okay, let's go fast, let's go free ad-supported television. Now Netflix has an even bigger problem on its hands, right? Because they may have reached a ceiling with what they can charge. Certainly premium. I don't know how much higher you can go than $30. There's not much. They've probably extracted all of the net additional subs they can from password sharing crackdowns. Now they're faced with, okay, we just have to bring in a lot of live events and sports, things like the Beyoncé Bowl, you know, and the January NFL games. It's really expensive. Right. And so are you going to compete on content or are you going to go the opposite direction, race to the bottom and try to get UGC on there? And I think that's probably what they're trying to sort of thread that needle, because if you listen to Netflix's leadership, they say, look, we can never go pure fast because that would have a deleterious effect on the brand. We're seen as a premium service. But how can you maintain that when you see Roku and Tubi joining forces? Right. When you see all of the pressure that Amazon is putting on your business because Amazon is the identity spine for advertising across all of FAST, all of FAST CTV is Amazon with these data deals doing all the— providing all the identity for the buying. Right. And so I think they're facing like this opposite problem now where they had too much pricing power. They pushed everybody in the opposite direction. To go fast. And now I think honestly, if you believe the reporting that The New York Times did, that they're going to be offering up these subscription bundles, how can you not go fast and do that? Are you going to be willing to accept paying $30 a month for a premium Netflix tier, seeing content in your Netflix app on your TV and having it say, and if you want this, you have to pay even more to subscribe to this other channel? I wouldn't. I think the only way they can do the bundling is if they adopt fast.

That ARPU gap. From the premium ad-free tier to ad-supported. It feels like— is there a reason why the gap doesn't make up the full $30? Is there some economic reason or did they just get stuck in this weird place? Because it feels like if they got even a little bit more juice out of the ad model, they could just offer a free ad-supported tier. But maybe that would create more churn because mentally you're going from $30 to $10 is different than going from $30 to $0. But do you understand more of like what they're grappling with? Did they get unlucky or is there something about the structure of Netflix that would, would, would really make it difficult for them to do just a fully free and ad-supported tier?

No, I think it's just a reservation they have. Maybe it's almost like a superstition, right? I mean, keep in mind, they were saying for the longest time they would never do ads.

Yeah.

You know, Reed Hastings said like—

They always do. Yeah, they always do.

They always do until they don't, right? Until they change it. But like, he— I mean, Reed Hastings said ads were like a blight. He said personalized advertising was a cancer. And so, I mean, you know, then they changed the tune. But I mean, I think he said just that, if I'm remembering the wording correctly. But the thing is, like, I think— so there's 3 tiers, right? So there's the— it used to be basic. Now it's standard with ads, standard, and then premium, right? And so the standard is what they have to make up the ARPU gap with. Premium includes a lot of stuff. And so maybe you can make the case, right, for a high-income household, it's worth paying for the higher streaming quality and the more, the more devices that you can onboard. My sense is with Standard, though, that, that revenue, that ARPU gap is just— they just, they just index it to the ads ARPU. But the thing is, like, the question is, like, just internally, what is the resistance to going fast? And my sense is it's that perceived quality that they would lose if they went fast. But I mean, what can the market move in that direction? They push the market in that direction. And so I think they're going to have to capitulate.

Have you been surprised that Amazon Prime Video hasn't seen more UGC? Because I was looking up randomly, what does it take to publish a film? And you can submit to Amazon Prime Video. It goes through a review process, but the cost to submit is de minimis, and you can get a video product up there, but it hasn't actually seen a groundswell. And I'm wondering if that's because Amazon specifically has partnerships that scratch that itch.

Yeah, I think that might be the case. I mean, keep it like I said, I mean, Amazon has all of these identity partnerships that essentially make it the data spine and the identity spine for FAST, like broadly, right? And so if you think about the— I remember people saying, I bet Amazon's going to buy Roku. And I remember thinking like, why would they? They've already got this identity partnership with Roku where they get access to the best impressions. If they bought it, they'd just be getting everything else. that they don't already buy, right? And so my sense was always that like, well, they're already picking over the stuff that they're not buying. Why would they wanna own it?

Mm-hmm.

And so like my sense is, yeah, maybe it competes with the fast channels that they're partnered with, or maybe they have similarly like, you know, some sort of perceived quality bar or hurdle. But I think like the UGC thing is a little bit of a distraction because, you know, I don't think Netflix is ever going pure UGC. Like they are very selectively curating the YouTube creators they bring over, but they brought over a lot and they went on a shopping spree this summer. They bought a lot of stuff this summer. And now, now YouTube's pushing back.

I saw some post about how the results of that, of that shopping spree that Netflix went on with YouTube creators had a significant power law where there were some creators that were really performing well on Netflix. Is that because of the way the Netflix algorithm works? Is that true? Or do you think that there's— okay, if somebody is getting 1 billion views on YouTube, we can pull them over and get X views and there's some multiplier, or is there something special about the way Netflix is actually rolling out these deals?

Well, I think there's a couple of pieces there. I think they've got— YouTube certainly has more Rexus adjacency, so they have more opportunities to push people into this content than Netflix does. Netflix has less stratified, less deep catalog than YouTube does. So that's one piece. But then you do just see, even with Miss Rachel, season 2 performed substantially worse than season 1 in terms of view hours per minute of content. And so, you know, it's maybe it's just consumer preferences or they had other stuff to prioritize. I mean, Netflix has invested a lot into its own recommendation algorithm. And so it just depends on like what they feel the need to promote at any given point in time. And they have to, you know, they just have to exclude stuff that doesn't fit that purpose. Yeah.

You have to imagine that someone like Miss Rachel, if she's doing a deal with Netflix, she as part of the deal. It's not just going to be money. It's going to say, hey, are you going to actually give us impressions so that we have the chance to be successful here and get real viewers that go into the rest of the funnel, show up to the live shows, buy the merch? Like, it's not enough for you just to pay and stuff us off in a corner. If we do this, we want to do it.

I wonder, I wonder if Netflix has an opportunity to just basically take all the best UGC from YouTube, because I think YouTube has a massive AI Yeah, that's that's seemingly what they're doing. But underappreciated maybe how big of like an AI like AI is probably like seemingly a net good for YouTube right now because there's more.

You mean slop specific?

There's more content for all the the long tail. Like something will happen and I'll get served a video and the voice sounds good now. Maybe the script sounds okay. And then you realize like okay this channel was just created two months ago and it's not really. And and YouTube is has an incentive. constantly be serving new creators.

In theory, Netflix and human review.

I generally want to consume content from creators that have been— are either new and extremely passionate and dedicated. I even saw a creator over the weekend that started years ago using like AI voices and just stopped and said that YouTube is so flooded with AI content now. He's just— and he's not American, but he makes all of his English isn't his first language, so he has an accent. But it's appealing because you're like, OK, this guy's actually making this content. It's not just fully generated, right? And so I think YouTube— or Netflix's opportunity is to try to carve out all— actually have a filter again and be a curator and take the best content that historically would have just been on YouTube and try to bring it over.

Yeah, but still be Well, yeah, but I mean, so YouTube absolutely does not want that to happen, right? So I mean, keep in mind, like, the dominant platform for YouTube in the United States is the TV. Yeah, right. And so they are— I've called it a CTV behemoth. That is their dominant platform by view time. And so they compete directly with Netflix for engagement. It's actually really problematic if Netflix is able to just poach their best creators, the top, like the sort of like the cream of the crop, and then push them into Netflix. And then to justify the subscription price. Right. So what they've done is they've— they're now inking deals to keep people exclusive for a period of time. And they're also saying, look, we're going to punish you if you move. We're going to deprioritize you in our recommendation systems. We can't promote you if you're on Netflix because we want to promote stuff that's exclusive to us. And so, you know, and you're not going to share in brand revenue. So it's actually what Netflix has done is they've forced YouTube to apply a lot of the curation pressure and incentives that YouTube always didn't want to do. They always wanted to resist being that kind of channel. They said, look, you know, it's just like this is just an open market, like people compete and you get, you get views if you outperform. But now they're having to sort of put their thumb on the scale in certain ways to keep the best creators.

Yeah, mask— very much a mask-off moment for the, for the platform that wants to position itself as like, we're just this friendly platform for creators and we just want to support creators. And they're like, Like if you even put one of your videos over here, you're not getting any more money and you're not getting any more views, and we will end your career. Like that's basically that's basically be a shame if anything happened to your audience. Yeah, be a shame if something were to happen to your reach.

What is going on at Apple around services, ads? Give me a little bit of the history there. Apple's obviously had some some sharp words about advertising in the past. Wound up building a great ad platform.

So John sent me a screenshot. It looks like it's from Mobile Dev Memo, and the screenshot says the Schiller exit is a bit more notable. Apple's app marketplace is loved by many consumers, but often criticized by developers and subjected to increasingly onerous regulations. But there's a bit more to the story. I'm told Turnus and services chief Eddie Q want to make even more money from the App Store and figure out how to ways to raise margins and squeeze. Additional recurring revenue from the platform. Schiller, on the other hand, seems to believe that such moves would only further irk developers and governments. And John said, "First piece of good news from Apple under Turnus: Cupertino will finally be focused on App Store margin expansion." I've been pulling my hair out about this for years. Feels good to be vindicated. Better late than never. So obviously joking.

Yeah, that quote's from German. That I was quoting German.

Got it.

So that was from Gurman's newsletter. But yeah, we got a new, a new era, right? Uh, Schiller's out apparently. He stepped away from the App Store. Eddy Cue is taking over with Ternus. Yeah, and apparently they want to squeeze more money out of the App Store, which makes sense because it's under-monetized.

But why not ads? Is that— is that— is that—

It can't be anything but ads. No, it can't be anything but ads. What else could it be? I asked this question on Twitter yesterday. Yeah, what else could it be? Yeah, what could it possibly be besides ads? What other opportunity is— are they going to increase the commission? Are you nuts?

Yeah, no way.

How do they increase the commission?

The commission's not going anywhere in any direction. Wait, so you said it's undermonetized, but you believe it's undermonetized purely on the ad side, the commission side, that people's totally at their limit? Totally.

Yeah, yeah, commission can't go anywhere. Certainly it's not gonna go up. They'd be lucky if they could maintain the commission. I mean, they're probably gonna have to reduce, you know, so basically I think what I think is gonna happen, so they replied with their, in the Epic v. Apple drama, which is interminable apparently, They made their latest proposal, which is that they'll apply a 15% commission on the link out.

Right.

So they're having— so if you remember, just go back a little bit, the Epic v. Apple case, Epic essentially lost on every 10 of the 11 points. But what they did win on is that Apple does have to allow link out.

Right.

So you're in the app. The developer can put a link in the app to a website that allows them to monetize there. Right. So they have to be able to do that. Now, what Apple responded with was saying, okay, yeah, but we're going to apply a commission that when you add in the Stripe fee, is essentially just as much as 30%. And also you have to do all this reporting. And also if you have someone clicking out and going to a website, they're probably not going to be as likely to convert, right?

Yeah.

So you apply all this commission, all these frictions that like, okay, it's never gonna— there's never gonna get any traction. Now what they've proposed is like, okay, we'll accept a 15% commission on this, but you still have to go through all the reporting. That makes it interesting. Now we'll see if the judge accepts that, right? Uh, uh, she might not. But like Epic was fighting this. They want zero commission on link out, but Epic wants zero commission on Yeah. But so like, if that happens and the commission drops even further, you're gonna push a lot more revenue outside of the App Store onto the web, which actually a lot of it's already fled to the web. I mean, like, go to any subscription app, go on the Facebook library, find your favorite subscription app, Strava, whatever, look up their ads and see where they link to. I guarantee you it's to the website. Almost every subscription app spends the vast majority of its advertising on web destination ads.

Yeah.

So they're sending you to the web, you register on the web, and then you download the app, you log in. And so all of the monetization has happened on the web. There was never any monetization in the app in the first place. Games have started adopting that too. And so they've already lost a lot of this in-app monetization to the web. The commission's already under a tremendous amount of pressure. And it's not, you know, you've got the EU with the DMA. They seem to have come to a resolution there, although, you know, Epic also thinks that's not true. But you've got Japan, you've got Brazil. I mean, you're going to see increasing cases where governments step in and say you have to offer alternative in-app monetization and you have to offer alternative app stores. And so if you think about what does the App Store have that can't really be taken away, that's probably undermonetized. It's a lot of engagement. It's— if I remember correctly, the last touchpoint where they released any data was 600 million. No, it was 900 million weekly active users. So they— ChatGPT reached parity with that at some point a couple of like like roughly a year ago, 900 million weekly active users. That's a lot of engagement. And what do you do when you have a lot of engagement, a lot of eyeballs, a lot of attention? You monetize it with ads.

I think they have no other option. But with the App Store, every time I search for something, I'm seemingly seeing a bunch of ads.

Yeah, but you can see way more. Like imagine you go to a website, you go to an app in the App Store, you're browsing, you don't purchase the app. And then the next morning when your alarm clock goes off, instead of—

Notification.

No, just replace the alarm clock sound with an ad that says, please download this app. Go retarget me throughout the day. I open up the iMessage.

No, but where I was going, where I was going with this is at what point does Apple, if they actually want to grow their ad business, not start to compete with an AppLovin and start offering like bringing their, bringing their ad network into the apps?

Right.

Well, I mean, they do essentially. I mean, they've got ads in the search results, right? They just added the second slot, the second placement in the search results recently, a couple of months ago. Um, you know, they've got like recommended apps where you can find the placement in an App Store page.

They could do like, uh, you have to pay to get organic results. Like you search and then it's just 10 ads.

That's not bad.

Unlock, unlock, unlock, unlock organic results. It's $1.99 to see what you're actually searching for. No, but what I was saying is like, like, you know, at like basically a user comes, they search for a game, let's say.

No.

Then they see an ad, but then they end up downloading a game. And then I'm saying, like, all the surface area that's potentially— like, the real opportunity is every time someone opens up that game, they're getting an Apple ad. And—

No, I think they will. I think they should.

I think they will.

I think they will. I don't know what else they could do. I mean, I think they're going to do that. I think they're going to have some sort of e-com ads offering. I mean, I predicted a year ago that they put ads in Maps because it just made sense. They've got a lot of engagement there, and they just did that a couple of weeks ago. But keep in mind, I wrote about this when they— so they took their ads service down for a while. Usually they do that when they're making a change, right? Right before they did that, and that's when they implemented ads in Maps, but they also created a whole new campaign optimization API to accommodate Maps now and all the other placements that they have. So this unified campaign optimization API. Right now, it's also— that's extensible. You could extend that to anything, right? So they're already putting in place the scaffolding to support that, but they're also putting in place the policy scaffolding. So they also made all advertisers like sort of recertify their approval of the Apple Ads Service Agreement, and that gives them permission now to serve ads on websites and apps they don't own. And I think that was the big sort of signal here. Now, they've done a lot of other things too. They rebranded SKAdNetwork to the Ads Attribution Kit. They renamed the whole thing from Apple Search Ads to Apple Ads. I mean, all of this sort of points towards a more generalized ads product. But I think opening up that services agreement to say we can place ads on third-party properties probably does signal that they intend to do that.

Yeah.

Yeah.

I mean, that seems like a huge way to grow services revenue. I don't know if you have the exact size of the product.

Ads come for everyone.

What about eventually? What about new Siri? I mean, looking at the ChatGPT ads hitting $1 billion, you've been very optimistic about the potential of ads and LLMs and that it's a logical end state. Do you think that they'll do Siri? What else are you tracking in the development of AI and ads?

Well, I mean, we talk about chatbot ads. You know, I have a lot of thoughts on that. But I mean, I could just go back to Siri. Siri is tough because it's only voice, right? And so I think you need—

there is an app where you can touch.

That's true.

That's true. That's true. I think once you have the chatbot interface, and they do have that, I think once you get a lot of usage there, ads become an opportunity.

Sure.

I just do think you need a visual component to make ads work.

Yep.

Um, it's also why I just don't think ads to agents is ever going to take off. First of all, I think there's like a lot of incentive conflict there. I just don't think that you can show ads to agents because you, you wouldn't know who to trust.

Yeah.

Um, and who's getting paid in that case and who's paying, right? So that's— there's like a lot of, uh, incentive conflicts there. But, um, I also just think there's a, there's a, there's a visual component to ads that's necessary. Right? Because that's, you know, that's what ad creative is. That's the, that's the principal way that you do messaging, that you do brand positioning, that you create affinity. And so I think if you're just doing voice, it's really tough. I think in the chatbot experience, you can. My sense is how they want to have— I think how they want to monetize AI generally on their hardware is, is, is through essentially a deal similar to what they struck with Google Search, right? Like, you have to pay to be the model that gets by default gets attached to these services. And I think that that could be very, very lucrative to them. And they've already sort of, again, they've already set up, they've already kind of created the environment for that to happen with this sort of the core AI framework, the access to the models going through private cloud compute. They've already created the conditions for that to happen.

Yeah. And I think Demis at DeepMind was saying that like there won't be ads in Gemini or like the core Gemini models. But he's out. And over time, you could imagine that the ad gets baked into the actual Gemini response and then that is monetized on the Google/DeepMind side, which then justifies the pool of capital that gets traded to Apple in exchange for that entry point. Makes a lot of sense. What do you think OpenAI needs to do to get from $1 to $10 billion on the ads product? They have 1 billion users. the technology does not seem that complicated to me. I know it's deeply complicated, but it feels like when you have the machine that can, you know, solve math and write endless code, like just writing a matching algorithm, serving up the UI, like that feels tractable. But is it a supply-side thing? They need to get more small businesses like what Facebook did where, you know, Ridge Wallet and every small company and medium-sized company is on there. Actually, do they need to go after e-commerce, gaming? Where do you see this going?

Well, first of all, famous last words that it doesn't seem that complicated.

To build? No, no, you're right. Because, because you and Ben Thompson were talking about like, oh, they, they've said they're gonna do this and then it still took months to actually get out.

Well, yeah, but they're doing it right. I think they just need to keep doing what they're doing. I think the opportunity is vast and immense, and I think they're executing at a blistering pace. I, I, so they just opened up to more countries. Now there are more than 40 that they're available, and they just opened up to 31 more countries. This was like 2 weeks ago. weeks ago or a week ago. So, I mean, this is like, I mean, so they're expanding. I mean, it's just, it takes time. Look, I think what they will be, I think you start to see the growth inflect when they integrate true conversion optimization. That means you're bidding a specific amount for a specific outcome. Like you're still doing CPC optimized conversions.

Sure.

Once you are bidding against a specific outcome, then the growth inflects. And I think, you know, they'll be off to the races. That's the gap from 1 to 10.

Yeah.

And then 10 to 100 is just continuing to onboard, you know, every SMB possible.

Yeah. Where do you sit on Instinct and the personalized agents?

Yeah, it's breaking news. So Meta just released an Instinct clone.

Oh, really?

Muse. And it's a new standalone app. It's not the Meta AI app.

Okay.

It's a new personal agent app that's in the App Store. Muse from Meta. It says your personal agent that takes things off your plate. Approve what track ticket prices and book reservations.

Mm-hmm.

Connect all your apps, get ideas for what your agent can take on.

Okay.

Which I think is smart because a lot of people just don't really fully understand what agents can do yet. But yeah, let's talk about Meta's current AI strategy in the moment. They're, you know, have started a pricing war with OpenAI. It's unclear what— how much the Meta AI app matters to the whole strategy. It's— they're doing Muse now, they're doing coding, they're throwing a lot at the wall. But what's your view? I think so.

I like to sort of just— I think I just— I'll hone on— I'll hone in on things. Yeah, they are throwing a ton of stuff at the wall. I'll hone in on the things that I think are like where there's, there's, there's like a true like thematic strategy. So a couple of things, right? Muse Agent, interesting. OpenCLAW kind of thing, like Instinct kind of thing. We'll see where that goes. I think there's gonna be a lot of those, and I think those get more and more domain-specific over time. But, you know, we'll see. That's interesting. It could get more consumer adoption on desktop. So we'll see. It's gonna be a standalone app. I was just reading Boz's tweet about it before we hopped on.

Yeah. And it's interesting to me that they own WhatsApp, which has billions of users, and they wouldn't just try to clone Instinct in WhatsApp. And just like try to get like actual crazy adoption there versus launching a standalone app.

Yeah, like should it be a button inside of WhatsApp as opposed to a separate app? This is always the trade-off.

You can interact with it that way. You can interact with it from WhatsApp.

Sure.

Totally, totally. But the friction of getting people to download a new app and set it up.

It was always different when it was just like, if you open Instagram, you get stories now.

Right.

Like there's no separate— there's no step. It's coming to you whether you like it or not. Not, because we believe in stories and we are going to get that to 1 billion very quickly. And they did.

Yeah.

I mean, you have to, you have to keep in mind that there's still a lot of reluctance to this on the consumer side and there's still a lot of mistrust from Meta on the consumer side. And so I think, you know, if you just, if you just integrated this as a forced download on WhatsApp, you might get a lot of churn. You might get a lot of Telegram adopters. My sense, though, is like if you look at a couple of different things, there's, there's a, there's a, there's a thread that you can sort of parse. Like one is Meta AI, right? So, okay, here's, here's an interesting enterprise use case. What if you had the go-to AI for managing all ad campaigns within your company?

100%.

Go-to AI app for managing all campaigns within your company. Because I'll tell you how people do it now. They use Codex. Codex is not purpose-built for that.

They use Claude.

Claude is not purpose-built for that.

Yep.

Meta AI now does that. They've added that as, as a, as a whole feature set. I think that gives you a strong indication of where they want to go with this. Who knows ad campaign optimization better than Meta? Why do they not have the right to win that? So even if you just said this only applies, the enterprise use case here only applies to optimizing Meta ads, that's still a massive opportunity for revenue. Not, and that's just on a first order perspective, not even considering like, well, that actually might result in more ad spend.

Well, that's why I was somewhat excited about Mannus, 'cause I was like, okay, they're buying this enterprise agent. They can point it, they can point it at Meta ads. And if the agent is only good and it's actually, not getting, like, if Meta's saying, yes, we want you to spend, we want you to optimize ads with this product and we're fully endorsing this, we're supporting it, it's not computer use or any of these other things.

Yeah, but I mean, like, I mean, yeah, that was exciting, you know, got rolled back, but I mean, Meta AI now does it. They've introduced that functionality to Meta AI. So now if you're a performance marketing team, you're saying, well, I need a tool to optimize my Meta campaigns. Which one am I gonna use? Am I gonna continue to use Codex? Again, it's not built for that.

Yeah.

Like, I don't think that they've devoted resources to making that a primary use case for it. And so Meta AI has done that. And if I, even if it was just for Meta ads, now imagine that it's the go-to for all ads optimization. Okay, well now it's an even bigger commercial opportunity for them, an even bigger enterprise opportunity, right? Now keep in mind, Meta launched the Robin Media Mix Model as an open source framework. So it's basically like a measurement apparatus that helps you do like probabilistic measurement across all of your campaigns that you're running, like online, offline, whatever, out of home. The reason they did that, I think, was because they felt they were being under-attributed right now. If you, if you thought, if you had this, if you had the sort of same mindset with the AI enablement layer, the tools that people were using to do the optimizations, Yeah. then this makes a ton of sense. This might end up with more revenue flowing to them because the optimization ends up preferencing them, not with the thumb on the scale, but just because it was actually under-optimized before. So that might be a reason to do this. If you genuinely thought people were spending less than they should be, to be optimizing their ad spend on your platform, then you would do this. And if that's the case, this could have dramatic second-order effects too. So that's one thing. The other thing is Business AI. And Business AI, I think, is underappreciated. I've been following this for quite a while. I had their VP of Business on the podcast maybe a year ago, and we talked about this. But they launched Business AI, which basically was a chatbot that you integrate on your website. They said, okay, that's kind of boring. Who cares? But what people were doing is they were clicking an ad, going to someone's website, and then they were using the chatbot to say, what's the What's the best-selling product? Stuff like that. That has a lot of opportunity there. Like, even if it's just surfacing basic data like that, there was no way for a lot of SMBs to do that. Now what they've done is they've introduced like an AI-enabled pixel which can automatically, you know, sort of tune itself, requires no— like very little optimization from the advertiser side. Imagine all the other stuff they could do if they have access to your landing page. Imagine landing page optimization. Imagine personalization is all driven by Meta's own systems. And the advertiser could say, look, I'm an SMB, I don't have time to A/B test, or I don't have the resources to do like constant A/B testing, constant experimentation on my landing page to optimize conversion. I know if my conversion was better, I'd be able to spend more money on Meta and get more sales. So if Meta can offer that to me, I'm okay to surrender that capability to them.

Yeah.

Imagine they'd penetrate even deeper into the customer experience on the website. Certain brands will say no way, but a lot of SMBs, which is that big bulk of their advertiser base, will say, Yes, please do anything you can do to optimize conversion for me because that's going to result in more revenue for me, which I'm then going to reinvest in more ad spend. So I think like business AI is really, really valuable to them and they've integrated that deeply into WhatsApp. So that is going into the WhatsApp experience where people are able to communicate with businesses directly, get a chatbot, you know, understand these things about the catalog, understand these things about the business and the core product offering. I think that stuff has a ton of potential and I think that's probably undervalued if you look if you look at Meta's AI initiatives, I think people focus too much probably on like the output of, you know, TBD.

Yeah.

But like, I think, and that's important too. I mean, look, the model that they just released is competitive. Like no one, people left them for dead 6 months ago, but now they're actually producing competitive models. I think that's important too. But if you think about the integrations in the surface area, they could just apply even commodity AI to, there's a lot of value left to gain.

Yeah, we never book an episode.

Very well said.

No, no, no. Yeah, and what you're saying to me, it's like, it would be— I think it would be thrilling for shareholders if they saw MSL actually focused on applying AI in the business and not like, oh, coding's a hot category, we should introduce a coding tool, or, oh, let's try to ramp Muse API revenue. to $100 billion, right? Or whatever these things are that they're trying to do. And so, yeah, it's like net new products that are consumer products like Muse that I think are interesting. And it makes sense that they'll take a crack at this category because it will probably be a multi-trillion-dollar category. It's aligned to their existing business. And then some of these other things of actually applying AI in ads.

Yeah, instead of personal superintelligence, intelligence, it's small business superintelligence. That would be a huge opportunity.

And small business owners, like, small business owners, I feel like, are people that generally, they may have a little bit of a love-hate relationship with Meta in that they're like, I'm dependent on Meta. If I turn off Meta ads, my business revenue drops 50%. But at least they're like, I need Meta. Like, they're dependent on it. Whereas average consumers, you look at the comments and people are like, OK, Meta is talking about privacy. with their new agent. It's like, nobody's like buying the Meta suddenly is privacy-focused narrative, right?

Yeah.

I mean, look, a lot of the SMBs, they understand that they wouldn't exist but for Meta. I mean, there's not a lot of animosity towards the company that provides your right to exist. It's like the oxygen that they rely on. I don't think there's a whole lot of hostility to that. I mean, what always bothers me about, you know, these discussions is like to just not recognize that D2C would not exist Absent Meta, if Meta's ads platform hadn't like evolved in the way it did, there would be no D2C category, right? And so it's like it's ridiculous to say. First of all, I think one one I mean let me know if you're we're up on time here, but like one issue like a lot of people look at the index like so if you look at like the advertising share of GDP, it's basically it sits within like a narrow band like over time historically going back really hundreds of years, and people say well look that proves that advertising is is not driving. the economy, it's almost like a drag, you could say, because it's just this cost base. But like, the thing is, when an advertising format, when advertising enables new business, the fact that I'm still maintaining a share of my revenue as my ad spend doesn't mean that that ad spend would exist or that my revenue would exist without the ad spend.

Yeah.

Right. And so if you create these new opportunities for commerce, like, it doesn't matter that the share that you reinvest back into advertising stays the same. That's not an indication of the value of the economy. It's creating that chunk of the economy. And so I think that you get— that gets missed in these discussions. Like, if you— it actually impacts economic growth. It's endogenous there. And so it's an input to the economic growth. And so you can't say that the fact that it remains within this narrow band as a percentage is some way to evaluate its importance to the economy. And when you are creating new opportunities to transact that are entirely derived from advertising, Then you are creating that section of that segment of the economy.

Totally.

Totally.

Do you think that— what is the state of affiliate marketing? Because when I think about a product like Instinct or this new Meta product, an agent that will go and buy things, it feels like that would be easier to bootstrap in theory than an advertising platform that you need scale and a demand side or supply side for. So you could, in theory, when even on day 1 of Instinct, I ask to go order a pair of shoes and it uses an Amazon affiliate link and they're making money. Is, is that going to persist or is that like a dying, a dying monetization path or is it just small?

No, I mean, I think there's some upper limit now, but I think it's something that'll persist, you know, indefinitely. I mean, you look at there's big companies that do essentially affiliate. Rakuten is one of them. It's a successful company. I think the thing, though, is like like, is that subscale? Yeah, affiliate makes sense. But this is the argument I made when ChatGPT introduced instant checkout. Like, that's not gonna— that's not the optimal way to monetize that attention. The optimal way is to do conversion-based, conversion-optimized advertising because that introduces the auction mechanic. You actually deliver the value that a person's bidding. And then so the more value you give to people, the more they bid and the more revenue you make, right? You don't get that with affiliate. Affiliate tends to do the opposite. It tends to preference It's the lowest cost but highest converting goods. And so it's kind of, you're promoting like the worst stuff, the cheapest stuff, right? The Shein stuff, the Temu stuff. And the thing is, when you unlock sort of like the latent value with the auction mechanism and bidding, especially second price bidding, then you get growth with performance. And so that's how you scale the platform. You onboard long tail SMB advertisers.

Well, Thank you so much, Jordi. You have anything else?

No. Always a pleasure.

Always a good time. Always, always a great time hanging out. Thank you so much for coming on the show.

Great to see you.

Have a great week.

The third technology brother.

And we'll see you soon.

Cheers, Eric.

Let me tell you about Public.com. Investing for those that take it seriously. You got stocks, options, bonds, crypto, treasuries, and more with great customer service. We ran long there, so we're shifting Harry Melcep from Antioch to the end of the show. We're going to be joined by Scott Wu from Cognition in just a minute or two.

Huge conversation there. While we were live with Eric, Meta came out with Muse Agent. So Tyler, download it.

Got it.

Already got it. Look at that. Look at that. Excited to see what you think.

Build me an equestrian simulator. I don't know.

I don't think it's focused on vibe coding. It's like, it's the whole book me a reservation, book me a flight, Okay.

I tried to say—

So, can you imagine—

Table at the Beverly Hills Hotel, please.

Can you imagine being like Resy or OpenTable and having, you know, trillions of dollars, watching trillions of dollars of CapEx, and people are like, one day, you're gonna be able to just tell your agent, you know, book me this reservation, and they're sitting there being like, it's 2 clicks, sir. Like, sir, it's 2 clicks, because you go into Resy. I mean, It's it's it really is. It really is quite quick.

Apparently, we have to ask Scott about the number 46. Do you get this reference? I don't know. Okay, we're gonna ask him because he's here in the waiting room. And let's bring in Scott Wu, founder CEO of Cognition. Welcome to TVCN, Scott. How are you doing? Great. Tell us about the numbers.

Thank you.

Thank you.

Slow news day.

I was gonna say. Huge news day.

You guys to talk.

Congratulations. Yeah, crazy, crazy news day. We're gonna talk about math, we're gonna talk about AI, we're gonna talk about fundraising business. But first, the number 46. I was, I was told to ask you about this.

Why?

Oh, that's funny.

That's funny.

Well, so my Twitter handle is @scottwo46.

Okay.

The reason my, my handle, my, my, I'm like @scottwo46 on basically all these platforms is because when I was in like elementary school, I, the biggest thing I knew of was this middle school competition, Mathcounts, which is now made a bit more famous because people have seen these videos of Mathcounts and stuff. And in the Mathcounts, there's like a written proportion—

portion.

And then when you do well enough on that, you go into the actual, like, head-to-head. But a perfect score in the written portion is 46.

46.

And so then what ended up happening is, with this particular round, we actually ended up pricing it at 46, partially for the meme, partially because it, you know, ended up being the right—

It did the meme.

—kind of thing.

Yeah. You have to imagine that certain VCs are like, OK, we've agreed to this number, but are there any— insider meme references that would save us even 3% here. Let's do a deep dive to see if we can get the founder to give us a little hint. Yeah, yeah. No, but congratulations. Tell us about the fundraising round. You got every investor in the world on the cap table at this point.

Is that right? Yeah. Look, you know, it's an exciting time for cognition, obviously. And I mean, I think the biggest thing I would just call out is like agents are just getting really, really good, you know, and I think 2 years ago, uh, by the way, we're not, not that old, like 2 and a half years ago. Uh, so, so 2 years ago, you know, was when we kind of did the Devin launch. And back then, as you remember, it was like, you know, it went pretty viral and so on. But, but it was really just like a prototype.

I would not say it was— And there was, there was so much skepticism.

People were like, there was a lot of skepticism.

They are coining a buzzword that will never exist. Agents, they're just hyping this up. And now it's like everything is agents. Yeah.

I remember thinking about this because I was just like, man, it's, it's great that Devin actually works because if it didn't, we would just look It looks so dumb right now for me.

Yeah, totally.

But so that was 2 years ago. And we didn't have customers. It was very much just like a prototype and here's what we're building towards and so on. 1 year ago, I would say it was very— the product existed. It worked. It was for very specific use cases that you would kind of have it work and have it do it end to end. But I think most people in the world hadn't really gotten across to the idea of using a background agent And now I would say it's, I mean, especially in our circles, it's like, you know, it's becoming a more and more commonplace thing. And a lot of it is just that the capabilities are just so good that obviously you should go and delegate to something that just does the task entirely, right? And be able to manage these and so on. So that's been the biggest thing.

Yeah, I mean, the agentic era arrived, the importance of harnesses like Devin arrived. How do you think about orchestration?

Yeah.

Like, we saw a glimpse of that with Gastown that went viral. Now, whenever you fire off a prompt in any modern AI system, you can see the harness and the lead agent sort of talking to subagents and delegating things. Is there— is orchestration a meaningful leg up? Is it another exponential in the sense that we went from LLMs to reasoning, that was an order of magnitude. Then we went to the agentic era. Have we always been in the orchestration era, or is this a new era? Is this a new meaningful change? Are there new disciplines that need to be explored in the world of orchestration?

Yeah, so it's a really important question. I think there's definitely a lot of, like, what's the word? There's definitely a lot of kind of, like, misconceptions, I think, out there about it. I will say that there's a sense that people get of, like, oh, like, if you just say the magic words to the model, it will suddenly become 15% smarter or something like that. That's much less of a thing, especially now because the models are all RL'd on very specific capabilities and these kinds of tasks and so on. I think what is much more the case is, number one, if you go and bring in all of the context and the information and the systems that you need, for example, in coding, a very simple example is a model and an agent that can go and test its own code and click through its website by itself and then go and look and say, OK, that was right. That part was wrong. Let me go fix that is obviously going to be way more capable than something that can one shot it on its own. Right? A model that can go, you know, an agent that can go look up in the data dog what went wrong in the logs is like much more powerful than something that can't. So that's one is just like being able to bring in all the tooling, the context, and so on. All this kind of like messy real world stuff. How do you navigate a big code base? How do you get through you know all the secure systems that are you know that a company will have on? Its software. And then number 2, I would say, is combining the strengths of the different models and the things that they're good at, right? And so, you know, I mean, everybody's talking about price performance of models. Everybody's, you know, showing the charts of the Pareto curve and so on. And this model is like really cheap but is good enough for this percent of tasks. And this model is like the way expensive one and it's, you know, this much percent better and whatever, right? But obviously what that means is that by combining the different models, as long as you know which use case to route to each model, then you can do a lot better. than anyone alone.

How permanent is that task, that job of picking the right model at the right price point? Because we see there's a lot of attention from the big labs. And it's very clear that every big lab is going after building beautiful slides, front end, back end, cybersecurity, bio, like math. They're all working on these. And there's this trade-off between And then a certain model might be really good at something else. And then the new big model comes out and it's better than everything at everything at every price point.

But how—

Yeah. And I think the risk is that you can wind up in a situation where you're like, oh, well, if I just wait 2 weeks, I'll just be able to use the major hammer 'cause everything will look like a nail. But it feels like there's also always going to be this cost optimization, speed optimization, So that optimization problem, is that sticking around forever? Is that what you want to build the core of Cognition around, nailing that?

I think it's sticking around. Yeah. I mean, I think it's, if anything, it's going to be a bigger thing as time goes on because for a lot of these use cases, the intelligence is really not the bottleneck anymore. Right. And so obviously it's been another hot thing for the last few months. But I would say it's like, you know, ChatGPT, for example, new, bigger, smarter comes out, you know, smarter model comes out. it doesn't necessarily change their retention metrics all of a sudden, because most of the things people ask ChatGPT, you know, it turns out the models are good.

Yeah.

You know, they already go and get them right, right? And like, now what you care about is, is, is, is not just pure IQ for, for any task, you know, whether it's coding or, or legal or, or customer service or whatever. Um, you know, it's, it's, it's, it's not just, okay, what is the raw logical intelligence of the model that I'm working with? It's like, okay, well, does it, does it know all of the details of what I'm dealing with? Does it do things in the style of how I want it to do? Is it fast? Is it cheap? Is it effective? Is it, Is it trained specifically on my subset of use cases? Or all of these things like that. Basically, what that means is I think there should be this frontier where people start to much more aggressively use all these different models rather than relying on the single biggest one. I think a year ago, that was more of a phenomenon of a year, year and a half ago because there were a lot of use cases that were just on the cusp of possible. When that was the case, of course, you want to go use the very smartest model that you have out there like Sonnet 3.7 or whatever it was, you know, a year and a half ago. But now because all the models are really good, you know, what that means is you can be a lot more judicious.

Kind of a question going around, I think, on different people's minds. Coding agents are now incredibly capable, but is the quality of software around the world actually increasing? And so I wanted to get your view on, you know, you guys are working with a lot of the biggest companies in the world. Do you feel like the quality of their software is increasing, or are they just doing more? Are they doing backend migrations that consumers don't even experience? Because you've had this, I guess, question for a while, which is like, okay, when I download an app for like a big airline—

Yeah, I was chirping at Rune about this. I was like, oh, if the AI models are so good, why is it annoying to use the United Airlines app? And he was like, And then he was like, have you tried to use it recently? It's actually pretty good. And then I did try it, and I was like, yeah, actually, maybe it is better. But I don't know. Where do you sit on this?

Yeah, I think short answer is definitely yes, it is better. I think to the extent that there's more to do, a lot of that is much more like— it's much more a function of these practical problems, like going out and getting distribution going and doing that. And I mean, I think people in our ecosystem understand this intuitively, but it's worth kind of calling out out loud. cloud, that, you know, a 50,000-person software org is not going to figure out how to use coding agents overnight the same way that like a 3-person YC company is going to.

Right.

And so there are steps in the process that need to go happen. There's like a lot of kind of like, you know, onboarding and education that has to be done. There's figuring out the right systems. There's, of course, getting through folks like, you know, the security guardrails and the reviews and the walls that people have to make sure all of that as tight and something that they want to have operating in their ecosystem. But once that's there, we very much see that that's clearly the case across all these industries. And I mean, I think it's like— yeah, go ahead.

So I can imagine if I'm selling Cognition, there's a few different pitches I could give. One is I go to a company and say, look, you don't have an ERP or you don't have an e-commerce system. We're going to come in and build that, and it's going to be done by the time we're complete, and you're going to have this. It's going to live forever, but it's a new capability. Others, we're going to give every person in your organization a Copilot, and we will help transform the organization. And then third might just be like, look, we're just giving you access to Dev and it does good stuff. Like, you know, go and you couldn't get it at this scale, but now you can because we're working together. What's resonating the most these days?

Yeah, no, I mean, I think folks are rightly, by the way, are very focused on at this point on just like what are the actual use cases that it's going to drive and what is it going to mean? You know, I think there was a, there was a period where it was very much, all right, we're in the token maxing world. Yeah. How many tokens are my engineers using? How— that was, you know, that lasted for all of like 4 months or so. You know, great times. But, but, but now, you know, I think there's much more kind of like this much more clarity of thought, I would say, from, from folks in the industry about, okay, well, look, AI is great. Let's talk about what use cases that we actually care about. Like, what are the top 3, 4 priorities we care about as a company? Like you said, maybe it's this, this killer new feature that we want to put out. It's this app that we want to make way better. It's whatever. And let's talk about how we actually go and measure that and improve on that. You know, I think the biggest thing, if anything, I think is kind of like Yeah, like, how can you show it in the concrete results, right? Like, the value should be there, obviously, because AI is so good, you know, and it's so smart. But, but like, unless you're tracking that, you're making sure you're using it for the things that are effective versus aren't effective, you know, you're, you're looking at the productivity on a case-by-case basis, obviously, it's like, you'll never actually know which things you're doing are working versus not, right? And I think that's been a big, big theme for folks.

How do you predict that the router market is going to evolve? We've seen a lot of moves recently. Ramp has a router. Stripe, you know, bought OpenRouter. There's a bunch of other players. Everyone wants to be in that token stream. But how do you see this sort of like category evolving?

Yeah, no, look, I mean, I think it kind of makes sense. I mean, it's, it's, you know, a lot of these companies that you're mentioning, for example, they think of themselves as like the center for finance on the internet. Right. And, you know, what are people going to be spending money on in the internet in 5 or 10 years? I think a lot of it is going to be tokens. models and agents and whatever you call it. So, so I think from that perspective, I think it's super reasonable. And I think the routing part is a big piece. But, but to your point, I think the entire kind of like infrastructure around how you do payments, around how you do spend management, all of these things I think are still going to exist and basically need to be redone in the world of agents. And so I frankly think that there are a lot of products to go and build in that space.

What has your reaction been to the recent progress in Max? When you, I think the very first time you came on the show, you predicted correctly that the IMO gold medal would fall. Google and OpenAI both scored just barely gold, not 46. I think it was 38 or something. It was like they missed the 6th question, but they did achieve gold. Now—

42, by the way, 42 is a gold in the Math Olympiad, which is slightly different.

Slightly different, yes.

You have to remember it's different numbers for every competition. Yeah, so I'll tell you what my honest thoughts on it, which are: Look, first of all, I think it's it's it's actually insane. There's a lot of controversy, and there's discussions, arguments, whatever. But I feel like the most important thing to call is that guys, we just solved Navier Stokes with AI. Like that is it's absurd, and and and I don't think it's going to stop. Obviously, I mean, I think the like you know what's funny? I actually have money. This is my this will be my only AI bear view that I ever expressed to you guys. I actually had money. On a bet that the rebound hypothesis will not be solved by the end of 2026. Okay, we'll see. As you know, three and a half more months. But I'm pretty sure if not 26, it will be solved in 27. I think it might get done in 26. Okay. But but I think it's less than 50%. We'll see. But but no, I mean, I I think all of this stuff is is just gonna get done. And I mean, I think it's like it's it's kind of insane. It's hard to put into words. I think if you're if you're not familiar with the the kind of scale of this, but like you know, Navier Stokes, for example, is like a very fundamental problem about fluid dynamics and such, and it's been around forever, basically. And lots of people have sunk lots and lots of time into it. And the fact that this can even just be done in a week—

Yeah, 88 hours.

—of models just going and working around in the system and orchestrating, it's insane. Yeah. I think on the point of the controversy, I mean, it's kind of funny, but similarly, I think a lot of people who aren't as familiar with the the math academia world might not know that this is actually just always what happens in math academia. Literally Newton versus Leibniz on the founding of calculus is still, by the way, is one of the biggest arguments that people have. This is back in the 1600s or so. And so in practice, I think I'm sure— look, I believe that both sides meant well. I think the accomplishments itself is going to be the biggest thing that we all remember from this. We could debate, you know, what exactly shapes up and what it means for folks. But it's, you know, the rest, I would say it's kind of the AI accomplishment itself is the biggest thing by far. The rest kind of feels par for the course, to be 100% honest.

Yeah. So it seems exciting for the world of math. It's certainly exciting as sort of like a, researcher recruitment. This is where progress is happening. It's also sort of a useful benchmark. Like, I could imagine in the future, like 2 weeks in the future, but just a few weeks in the future, if you're training, you know, MuSpark 1.5, you want to just throw Navier-Stokes at it and say, don't search the internet, don't look at the result. Can you solve it? Because that's a good benchmark, right?

Yeah.

But I want to know about your philosophy because you went viral with the Devin launch. Like, what matters to your customers, your recruitment? Because there's a world where you're like, oh, I want to put my— throw my hat in the ring and duke it out for different math challenges. But that doesn't seem critical path to the growth of your business. Certainly hasn't been because you're growing very quickly. But how do you think about the value of when you're talking to your actual customers, what is important to get across? Yeah.

Yeah, for sure. By the way, it's kind of a hilarious thought also that we're just going to be— I actually still think that one of the things that I've never gotten over is one of the token benchmarks that people do when they're going and training models and doing runs and stuff is AIME.

Yeah.

AIME. I think people don't necessarily even know what it is, but basically the AIME is a high school math competition. It's one of the hardest math competitions, and it basically selects the top few hundred kids in the US.

Yeah.

To go qualify for the next level to actually compete for the US team and everything. Like, I took all this, you know, every year as, as a kid. And it's kind of funny cuz it's like, yeah, dude, your model can't even get a 15 on the AMI. It's, it's out of 15, you know? It's like, what are you even doing? But like, no human can do that, you know?

Yeah.

And we're already, someday it'll just be like, wow, you can't, you can't even solve the Riemann hypothesis. Like, what did your model even do?

Like, you must have totally messed up your pre-training run or something.

I get it. You're trying to save money, so you're You're using a model that can't solve Riemann. Like, look, not everyone has token budgets right now. So you got to pinch pennies. It's cool. It's probably good at Photoshop.

But yeah, no, so for us, what I would say is, obviously, like, a lot of what matters is showing the real benchmark. Don't get me wrong, it'd be sick if Devin went and solved the Riemann hypothesis. I don't currently expect that that's what's going to happen, because it's much less, you know, this kind of like fundamental, you know, basic science research is much less our focus. I guess, as opposed to kind of going and doing real-world use cases. And so we see, you know, it's like when we show benchmarks, it's like, okay, here's a benchmark on how it does at finding security vulnerabilities in real-world codebases. Here's a benchmark on how it does. Yeah, right. And a lot of it is basically just like making sure we're speaking to the thing that folks care about and folks need.

So yeah, I mean, related to that last question, how big is the cybersecurity side of the business? How much demand is that driving? Obviously, that's been a huge story this year.

Yeah. Yeah. No, I mean, it's been a massive thing for us. Obviously, everyone, I think, is really thinking about this and thinking about— I mean, everyone's kind of freaking out, I guess, is like the honest way to put it, which is probably correct. I think cybersecurity is going to— you know, I think there's going to be real threats that happen. I mean, again, you know, when we say it's like these big orgs take time to adapt and to use new technologies and so on. Obviously, these hackers out there, like small teams that are going and using the best of the models or doing their own— who knows what they're doing— like, they're not waiting, you know. And a lot of these capabilities keep getting way better. And so, no, it's a small but meaningful part of our business. It's probably in the neighborhood of like, you know, around 10% of the Devin sessions and/or the Devin ACUs that get spent today are on security. But we see it growing pretty quickly. I mean, our security products are only like 2 months old.

So I love it. Well, congratulations on the fundraiser. I gotta ring the gong.

Amazing update. There we go. Great to see you, Scott. Please go, please go take out some open problems just for fun. Don't, don't distract the team, like don't, don't rope them into it, but just factor the next prime. Yeah, yeah, exactly. Spin up a swarm, just you and the swarm. Great to see you.

Great to see you.

Cool, cool. Thanks for having me.

Cheers.

Well, we were talking about security. No better time to tell you about CrowdStrike. Your business is AI, their business is securing it. CrowdStrike secures AI and stops breaches. We will be joined by Greg Brockman, the co-founder and president of OpenAI, in just a minute. In the meantime, the Descartes, acquisition. There were talks that Anthropic was buying Descartes, and it seems like they have walked away. This is an exclusive in Bloomberg, but—

Say it.

They put Descartes before the horse. That's the funniest quote tweet by Shashank Raman here on the timeline. We love—

Yeah, story— stories come out that maybe one of the— is totally, you know, I don't, I don't know how true this is, but it sounds like one of the sticking points was maybe Relocation. Piccart happily building in Tel Aviv, not wanting to move over to the US. Who knows? Dean at Piccart is incredibly talented.

It was a very fun demo. He came on the show and used his AI image model, his video model.

Yeah, his willingness to do just a live demo of technology that seemingly was better than anything else that we have seen.

I mean, it was low-res, but it showed you a glimpse into the future. He would, you know, be prompting it while he was on the call with us, talking about what's behind him. Now I'm in a wizard castle. Now I'm in a sci-fi, you know, cyberpunk city. And all of that was very, very fun to see. So I'm sure that they will continue cooking and we will have to check in with them soon. But we have Greg Brockman, the co-founder and president of OpenAI, with us. Welcome to the show, Greg.

Thanks.

How are you doing?

Doing great. Thank you for having me.

Thanks for hopping on. Huge day. Can we start with the, with the math advances? What is, what is happening? Why is this important? There's a lot of back and forth in the timeline, but I'd love for you to just set the table for us on what actually happened with Navier-Stokes today.

Well, it's always a huge day in AI and modern scientific progress, I would say. Yeah. Today, we announced that our model had solved the Navier-Stokes problem, that we found a counterexample or a sort of proof that you can actually— that these theoretical equations do have a singularity or kind of break down under certain circumstances.

Interesting.

And this is a problem that has been open for a very long time. It's one of the 7 Millennium Problems, which are kind of some of the deepest, most important problems in mathematics. And I think that the problem itself is important, right? This is new knowledge for humanity, that the proof itself is actually very elegant and beautiful. And I think that there's a lot to learn from it. Yeah. The equations have lots of application in fluid dynamics and other areas. But to me, what's even more important is about what this represents about where we are in terms of model capabilities and the fact that we can actually generate new knowledge, that we can learn from these models to have them help us solve problems that are otherwise outside of reach or would take us very long to solve.

Yeah. I mean, where should I actually go with this? Is this going to help me book a flight? Is this going to help me cure cancer? Is this just going to help you recruit researchers who are fascinated by this stuff? Because I think that there's— this has, you know, taken over the technology world, but I imagine that this will not be something that gets talked about at backyard barbecues with friends and family that are 3 clicks removed from SF?

Outside of SF.

Yeah.

Yeah. Well, look, I think that there's, first of all, the applications of this specific result or the equations themselves, right? Which are things that let us better understand phenomena from ocean currents to airflow around aircraft, around turbulence and things like that. But it's really about the broader insights and methods that can help scientists and mathematicians further accelerate their research. And I think that, again, representative of if we have models that can help solve this kind of problem, then what happens from here?

Yeah.

Like, what other problems that are immediately applicable? And I think that talking about curing diseases and new medicines we're going to be able to develop, all of that starts to become much more real when you have models that are at this level of assistance and capability. And I do think that there are going to be real changes to think about in terms We can have so much more ambition with the kinds of challenges that we can hope to tackle now.

Yeah. I mean, I think even if this doesn't break through to the broader world, the weekend definitely will because it seemed like everyone was talking about Blender, talking about Astra, building stuff. Take us through the launch of Astra, what the feedback has been, what you've learned. It seemed like there were a couple resets. The model scaled very well. How was this launch different than previous launches?

Well, first of all, I've just been blown away by the community reaction to Astra. It's been really amazing and very humbling, honestly, to see all the creativity and the different ways that people have been applying the model. And I think that it's very clear that we've reached a new threshold of computer use. So this model is able to I really work with different kinds of applications in a way that was not previously possible. And people are taking full advantage of that fact and really thinking about how to create. Lots of people showing off 3D creations and mapping out physical locations and turning them into, into these 3D models and thinking about, can you use this for design of physical parts? Someone talked about how they were designing some mechanic for catching hair in a shower drain. Yeah, that's cool. And that they were able to— it was super cool, right? That they're now able to actually manufacture that.

Shower hair, super intelligence.

Yeah. No, that stuff's so mundane, but it's so important. I feel like a lot of this stuff gets lost, right?

Exactly. And I think there's a core there that's very important, which is that we are talking about these grand challenges sometimes, or very esoteric applications, but really the everyday, the number of problems that you have in your life, Yeah. That you would love to solve.

Yeah.

It's now possible, right?

Yeah.

That we're really trying to empower the individual to make it so that you can be— you can have superpowers, that you can accomplish more.

Yeah.

And I think that, that, you know, really trying to benefit people, empower people, build tools that can really help you and help you in your daily life.

Yeah.

That's all part of what we're working on.

So, yeah, I mean, it seems like a huge number of people in tech effectively rebuilt their entire house in Blender and planned a remodel this weekend. Thanks, Dasha. But I am wondering about the merge and how you bring together Codex, ChatGPT Work, ChatGPT on desktop. I have a gaming PC now with an NVIDIA card in it. I have a Mac Mini. Like, I have all these things, and I can imagine that I'm just— I'm doing that unnecessary, or like, it's fun for me, but that early adopter work of going and unhobbling it a little bit here and there. But in the future, this will all just be tucked in one, you know, prompt box, and it might build a 3D Blender model for— to answer my question of, should I remodel my house or not? But how do you see the capabilities that we saw on display from sort of light power users over the weekend actually making their way into consumers who might not even know what Blender is?

Well, I think you're exactly right that we really want to shift these tools from requiring kind of low-level sort of access or guidance to really having the human be able to fly, right? To really be empowered, for you to be able to set the goals and the objectives, and that you still should feel like you can get into those details and you can understand them. You can provide that oversight. Because ultimately you should feel accountable for the outcomes. But you have this absolute amplifier, right? Like a trampoline or like a rocket ship for the mind, however you want to analogize it. And I think what that means at a practical level. So first of all, this year we've been really seeing this shift from just pure chat use cases to agentic use cases. But I would also keep in mind that chat is alive and well. I mean, we're now well over a billion users every week that You can see that the market share of ChatGPT is starting to climb once again because we've been investing so hard in, in so many use cases that are important for people in education and health and a variety of other areas. And then at the same time, these productivity deep knowledge work use cases, those are really taking off. We've had this like almost vertical wall of agentic adoption since we launched ChatGPT Work. And I think that the fact that these are 2 distinct modes, that that is actually a point in time that is something that we're continuing to unify and merge, and that we're starting to see that there's a new emerging form factor for how people want to consume AI. And I think that it's almost like that the promise of AI has always been that you have something that you can talk to and really delegate work to that's proactive and persistent. And that I— what we were promised, if you were to rewind 5, 10 years ago, was never a low-level language model that you have to think about context windows and you have to select thinking strength and you have to select different models. Like, none of that, none of that is the future. And so I think that we're moving towards real amplification, real giving you time back, real having computers that are able to operate according to your goals, to your desires. And I think that that is a core of it. We're developing this safely. That's, that's one of the core commitments that we make and how we think about this. But, but we really see the power of these tools starting to really, really start to increase in terms of what people are capable of. And that under the hood, utilizing tools like Blender so that that is almost a detail that fades into the background is absolutely the direction of travel.

It feels like in AI particularly, there's been almost like a first mover disadvantage in that billions of people have tried ChatGPT and some percentage of them tried it it for the first time and have a certain impression of the product and what it can do. And then, you know, even in the last few weeks, there's been new agents and products that come online and people try it and their mind is just completely blown. And I think it's funny because I'm like, well, as somebody who's like, you know, trying to get the absolute max out of ChatGPT, I'm like, well, I've been running like, you know, I've had like an agent running that, for example, will tell me every time a SpaceX launch is going to happen and if it gets delayed, right? these sort of like persistent agents that are running in the background. And but strategically, I feel like it's a new kind of challenge because you have this— as capability has been scaling, first movers need to be almost like constantly reminding the market of all these just like new ways to use AI.

Capability overhang.

Yeah, we think about this a lot. And I think that there's this discovery problem that we as a field need to really encounter in a first-class way. And we haven't done it fully yet, but I think we have a real shot at solving it better than any product before.

Mm-hmm.

Because the, the thing that, that right now we kind of rely on, you think about ChatGPT, ChatGPT Work, these are both text boxes.

Mm-hmm.

Right? And it's like, well, this new text box is way more powerful than the old text box. And, but oh, oh, there are some reasons that you still wanna use the old text box. It's not, you know, it's like, It's like far too confusing.

Yeah.

People just want something that can help them solve their problem. The whole point is to get your time back, not for you to have to go and become an expert in all these internal details. But at the same time, we also have a model that understands what you're trying to accomplish, right? That you're explaining to it, here's what I want. It has a lot of context on you. And so it should also be able to proactively say to you, hey, actually, if you ask me this other way, or if you added this connector, or if you authorize me to do this, Or if you hook up your credentials in this way, I can go and do this other thing for you. And so we're thinking a lot about that self-knowledge, that onboarding process. And I think that is a huge, huge opportunity. And I think that there is both the disadvantage that you cite of people tried it, they form an impression, and it's changed, that it's something new. But there's also an advantage. I mean, ChatGPT, over 1 billion users every week. That is unique. that kind of use on these models.

Totally.

And I think that the number of people who have tried ChatGPT before, I think it's another billion, billion and a half, something like that. And so that's a huge opportunity as well for us to go back to those users and say, hey, we can now solve the problem for you. We can now help you in ways that you didn't see before. And I think that it's true, it's real. If you look at how many people use Chat for Health, 300 million people every single week with health queries, right? And that that's really making a difference in people's lives. And that of their loved ones. And so we have such opportunity. Yeah.

Can you stay there with health and explain where this goes?

Maybe one note before that. Something that I think is really interesting, and I think something that OpenAI can do a lot better, is like when people talk about like everyone in AI wants to be like the Apple of AI from a marketing standpoint. And when they— when you think like Apple marketing, you're thinking like Mac versus PC, or you're thinking 1984, these big branding campaigns. But the The actual thing that Apple does really, really well with marketing is they just hammer really, really specific details about their products, right? They're like, they're advertising the new camera, they're advertising Memoji, right? They're advertising like one certain features in Safari, right? And so it's like with AI, the surface area of like things that you need to communicate is actually like an order of magnitude greater because it could do so many different things. And so I think there's such an opportunity for the company to focus advertising. Yeah, the brand campaigns are awesome. And like the launch video for Astro was amazing, but it's like there should be billboards running of like very specific things that ChatGPT can do to give you back your time. Yeah. Yes.

This, this has actually been a real sort of realization or just like something that I have really come to over the course of, of, of this year. And if you look at even, for example, the, you know, we just announced ChatGPT Images 2.5. And if you look at the launch video there, the thing that I love about it is it shows here's someone creating an image and here's like a bunch of different variations of it. And then here's them taking their favorite one and having it in the world. Like someone said, here's a cool, like, little sketch of a candle holder. You see an awesome visualization of it, and then you see the physical candle holder and you're just like, that's what you want, right? It's like you— it speaks to you immediately. And I think that, that really showing people here's a use case. And the thing that was also a little surprising to me is that we've sometimes highlighted esoteric use cases, something that appeals to someone in particular, and it's amazing for that person, but people don't then say, oh, because it's this powerful, I can also do this other powerful thing I've been waiting on. Like, that connection is, is, is something that is less sort of easy to make than I, than I'd realized. And it makes sense.

Yeah.

What you want is to— for there to be use cases where people say, I actually want that particular thing. Yeah. Now let me go try it myself. And then from there, you start exploring and you start to find, yeah, I actually do have this powerful use case that I didn't even realize was tip of the tongue.

Yeah.

Studio Ghibli model is probably like the best example of that working really, really well.

The other thing is, is reminding people to ask the AI what it's capable of.

Oh, yeah.

Like I was, I was having lunch with a buddy who's a real estate developer and he is using chat all day long for different like deal memos and to understand like projects that he's working on, all this stuff. And he'll ask me all the time, can ChatGPT do this or that? And I'm like, I'm happy to answer you, but you have the thing that will just explain exactly how to do the thing that you want to do, or maybe not, but it probably can.

Yeah, I did the same thing. I was kicking off this Blender thing, and I was like, should I run this as a local Codex thread or in the cloud? Let me know which one's better based on my system. And it gave me a good answer, and I was able to go for it.

On images, when Images 2 came out, I had some moments where I thought, okay, images is solved.

Yeah.

Like, where do you think images actually go as a category? Because it felt like this has been something that has maybe one-shot me more than anything else. Specifically with like when it released, I was spending hours like on a Saturday trying to design furniture Right? And just going through, like, you know, hundreds and hundreds of prompts. But how far can image models go, and where are they going, and what are the ways in which you think they can be applied? We were talking earlier, too, about the downstream impact of image models. If you can take, you know, a physical space somewhere and take a picture of it and imagine it as all these other variations, there's so much, like, real-world activity that will be driven from that because people can see this thing visually and say, like, now I want to go make that reality, which I think is really cool.

Well, I think that's exactly the right way to think about it, is as you hit new thresholds of capability, my experience has always been that fundamentally new applications become unlocked in ways that you almost wouldn't have thought about ahead of time. And so I think that within, for example, knowledge work, professional work, marketing, all those areas, you just need to be above a quality threshold. If you're below it, it's a cool concept, but you can't actually use the final material, right? That then means that you haven't really solved the problem and that having precise edit control, being fast and really being creative and having a diversity of different results and also being able to have this good interplay back and forth with a person, I think that really unlocks whole new use cases. And I think there's a huge market there. And even, for example, the kinds of things you may not think of naively but actually start to be really important applications we're seeing happening is slide creation or making awesome websites, right? Being able to have that image generation capability in the middle is something that's very unique to OpenAI relative to some of our competitors. And I think that you're able to then produce much better artifacts downstream. And so we really view images, we view voice, we view coding, all of these capabilities as one package that are going to come together to create an AI that empowers you. That means you can create anything that you imagine. And I think it's going to be something that's just unlike anything out there.

Let's go back to health. I think most people already are aware that you can synthesize some lab data with some sleep scores, but your vision that you laid out recently for I'm sure where that product goes is much more complex, much deeper. So tell me where ChatGPT Health is going in the future.

Well, I would think of it as there are 3 sides to what we do on health. There's the consumer side, again, 300 million people every week with health queries. There's the clinician side, which is bottoms up. And that's really about— think of it as ChatGPT that's really tuned for clinicians, that gives them Direct citations to medical literature, things like that. There's a third pillar, which is the enterprise side of selling directly to hospitals and them enabling it. And you can see things like we have an integration within Epic and really trying to bring each of these 3 pillars the best possible service independently. But you think about, as those really build momentum, that you actually are able to get synergies across them, right? That there's something that actually makes Yeah. The health experience and the ability to really transform healthcare in America and the world on the table because there are so many— think about how much work you, as a patient, have to do if you're talking to different specialists. You have to carry your medical record from one to the other. You have to explain again, here's the issue, and that, ultimately, you're on the hook. You're the doctor who has to make the decision whether you like it or not. Actually, being able to have just good sharing of that information across different providers, that becomes possible if everyone is on one platform. Or think about clinical trial enrollment. That's a huge bottleneck to drug development and finding people who are eligible and will benefit from being enrolled in a particular trial. And if you have that kind of data, if people are willing to sort of entrust you with that information, that's something that can actually really benefit them and benefit the world at the same time. And so what I view us as building is really trying to build the world's best healthcare platform to really be able to bring healthcare into the AI age. And I think that it's something that is going to be absolutely transformative to many people's quality of life, really uplift so many people. And we're seeing it already in such concrete ways. Some of my favorite stories about ChatGPT are people who say, hey, information I got from Chat helped me save my own life, helped me save that of a loved one, that there was this medical issue that someone had and that I— if, you know, the doctor told me one thing, I was able to double-check that and understand what they were saying. be able to push back on it and got to a good outcome. And that happens every single day. So I think that health with these AIs is something that we're still just scratching the surface of what's possible. And I think it's one of the most positive applications of AI that you can think of.

Yeah, I'm very interested to see how the advancements in memory intersect with health, because I expect that ChatGPT, with where memory has gone, where all the new thread and it will bring up just the right information or tie back to a thread that maybe happened 3 weeks ago. When you actually apply that to, like, health-related queries, it may be able to, like, pick up patterns that sometimes would take a human years to figure out, like a certain ailment or something like that, where it's like, hey, you're asking about all these different things and maybe you thought they were not connected. Turns out they actually are and you should go down this sort of, like, rabbit Last question.

I'm sorry. Please.

I was going to say, I think that's absolutely right. We're seeing that very concretely, and we've seen— just in my own personal life, my wife, we talked about some of her medical conditions publicly, but it was really this 5-year journey of talking to many specialists, each one who was kind of touching one part of the elephant and would try to address that one part, and it was only finally her allergist who said, hey, I think all these symptoms you're seeing are connected, and you have this genetic condition that affects all of your subsystems. And that's the kind of thing where it's really hard to say how many people have similar kinds of conditions and just never find out.

Yeah.

How many people have these areas where it's like if you just sort of are functionally specialized, that you're never going to bring together the whole diagnosis? And I think that is one of the powers and potentials of an AI that really deeply is able to help you across all parts of your life and also is a deep domain expert in all areas of medicine.

Yeah. I have one last question. What— how do you tell the story of Operator? It feels like it was a failure or a side quest, but it feels incredibly important now given the advances in computer use. Is there a clear lineage there? What was Operator? Does that still exist somewhere within ChatGPT? How How did computer use get solved?

Yeah, well, look, I would look at all these things as timing and all about iterative deployment, right? That there's a moment where you need to, where the capabilities aren't quite there, but actually learning from real-world deployment is very helpful, right? And I think Operator was just kind of below threshold in terms of the model capabilities. It was a cloud-based system that operated with computer use, but it was slow, it wasn't fully accurate, it was like pretty painful to use. Some people got value, but it really wasn't above threshold. And if you look at what's happened, that the team— like, one thing OpenAI does very well is we make long-term investments on things that really matter, and we do the grind. And that the team this year, I think, really started to build momentum, that we put in a lot of effort to go and sort of burn down a long list of issues. We're able to really focus on solving computer use. And I think that they delivered in a significant way.

Yeah.

And there's more to do. never done all those things, but it's a true milestone. I think people are really appreciating what's possible because we've been in a world with these agents using computers through connectors, right? Through these very painstakingly coded systems that are so different from how humans use computers, whereas humans can already use everything on a computer, right?

Yeah.

Everything's designed for people. So if you have an AI that can operate that way, and even from the very beginning of OpenAI, we had a dream that one day we could create such an AI. It becomes able to help you across everything that you would be able to do with a computer yourself. And so I think we're there with Astra. I think that there's just so much more that people are going to uncover in terms of applications where this can go. But it's an example of long-term focus, doing the work, and not giving up even when the going gets tough.

Yeah.

And it feels like Astra— my view is it really felt like all these different bets coming together at the right time. Right? The advancements in the model itself, the computer use, voice, all these things. And it's all making sense.

Yeah.

It's focus. It's something that I think this company does extremely well when we really put a challenge in front of us and think about how to accomplish it safely, well, and to really deliver the value. We're doing our thing.

Yeah. Put differently, it felt like when you look at last year, it felt like Yeah, I mean, last year it felt like OpenAI was operating like a big company and it was a big company. But, but the way in which like product experimentation was happening was, you know, when you think of like a hyperscaler, they'll launch a new product thinking, okay, if there's a 20% hit rate or even a 10% or 5% chance, that's okay.

Mm-hmm.

And then this year it feels like actually the entire company switched back into actual startup mode, which is like, no, Focus, focus, focus. All these things need to come together. The whole team needs to be rowing in the same direction, and then the difference in momentum and growth and all these things coming from that, and actually taking the company from operating and shipping more like a big company to shipping again and focusing like a startup has been— feels like an impossible task, and it's been incredible to watch. So—

Yeah, that's awesome.

Thank you. No, it's been a real effort for Many, many people at OpenAI to really bring together and something that I really value, that I think we really value as a company. And I think that we're just so laser-focused on our mission and really thinking about every piece of what we do should add up to helping us accomplish it.

Well, thank you so much for taking the time to come chat with us.

Great to see you.

We'll talk to you soon, Greg. Thank you.

Talk soon.

Appreciate it. Enjoy the rest of your day.

Cheers.

Goodbye. Let me tell you about Cisco. Critical infrastructure for the AI era. Unlock Seamless real-time experiences and new value with Cisco. And we have our next guest already here. We're running behind, but we'll bring in Sahir from Forus, the founder and CEO. Welcome to the show. How are you doing? Good. Good to see you guys.

Please.

What's going on?

Sorry we ran behind.

Tough act to follow.

Introduce the company. Tell us the news.

You're headlining for Greg Brockman right now.

There we go.

Yeah. No pressure.

Great opener.

Yeah, opener.

It's a lot of news.

Yeah. I mean, last time I was here in May, we had just introduced Forus publicly as the AI network for medicine and announced we had a $1 billion valuation. I'm back here now 4 months later, and we're announcing our $150 million Series C at a $3 billion valuation.

Wow. Congratulations.

And the biggest change for us is scale, right? We're now supporting millions of people across all 50 states. Forus is already used by doctors to treat patients in 85% of U.S. residential zip codes. And we're now working with 9 of the top 15 global biopharma companies to help them advance medicines.

What is the most helpful thing you can do for global biopharma companies? We were just talking about, uh, trial patient recruitment, but then there's also like the much more fundamental research. There's even just general— having a coding agent around can be useful to a But what are you seeing move the needle for them?

Yeah, I mean, so there's an enormous amount of capital and talent going into using AI to discover new molecules.

Sure.

And that is going to work. We're going to generate more potential medicines for more diseases faster than we ever have before. But discovery is really only the beginning piece. It still takes more than a decade and billions of dollars to turn a new molecule into an approved medicine. And you have to develop it. You have to launch it. You have to get the right doctors, secure coverage and distribution. And ultimately get it to patients. The network that we are building is really intended to help provide both the insight and visibility and the connectivity to kind of support all those steps, right? So once a new molecule is ready to test in people, you need to identify the right clinical trial sites and recruit the right patients.

Mm-hmm.

As you prepare to launch, you understand which physicians have patients who look like the patients that did best in the phase 3 trials. How should the medicine reach them? What coverage and distribution need to look like? Once it's on market, you need to understand who's receiving it, where adoption or access is breaking, whether people are staying on it, what side effects or positive results we're hearing. The way that our platform works, it gives us increasing visibility into how medicines are performing nationwide, as well as kind of the clinical and practice behavior of patients and physicians across the country.

Mm-hmm.

So for example, a new drug that came out earlier this year, back in March, for an autoimmune disease, 40% of all people who have ever taken that drug came through our platform.

Whoa.

And so you can imagine, like, that level of visibility is unprecedented, right? The FDA doesn't have that. The company that made the drug doesn't have that. The EHR companies don't have that. And so we have more clarity on what's happening on that medicine, how impactful is it, where is it getting stuck, than anybody has before. And that can help these pharma companies not just drive the process more quickly and efficiently, but actually make it more predictable, right? And I think kind of the bigger picture in our minds is that as you get this process faster, cheaper, but more predictable, you can actually change the economics of creating medicine. You can actually make it so these companies can afford to invest in more drugs for more diseases because you have this huge bottleneck between the number of medicines that are theoretically being discovered as molecules and the total number that are actually coming out to market. It's kind of like the movie industry where there's thousands and thousands and thousands of screenplays. But the number of movies actually get produced and become blockbusters at premiere is very small. And so that's really what our goal is, really. How can you increase the number of medicines that make it to market every year by an order of magnitude?

Yep. People within the tech industry talk about the AI labs just needing to cure cancer and that will fix the— or help fix the public. People love data centers then.

Like, put one in my back.

What does the pharma industry think about that sort of like line of thinking? Oh yeah, they're saying you guys are sitting over there actually doing the work, not one-shotting a blanket cure for all different forms, but at least making consistent progress towards a bunch of these different types of cancer. So, I'm very curious.

Yeah. I was literally talking to the CEO of a top 10 pharma about this a couple weeks ago at a conference where I was like, it's kind of crazy that you guys have such a bad rap. And yet, the people who have the worst rap in the game right now are thinking that being more like you is going to make them healthier.

That's a great take.

And he was like, yeah, it's super hard for them to process what these people are thinking. And at the same time, I think it's a little bit of a wake-up call. Like, maybe you need to reclaim the story a little bit. Totally. Because until we become immortal, medicine is going to become this permanent industry to invest in. has increasingly become the most important thing that's happening to advance society. And these companies are at the root of creating all these amazing cures, amazing treatments, are villainized because people don't understand their position. Even the way that people refer to them as manufacturers versus inventors, creators, whatever, is, I think, tough. But they're also at the same time feeling really eager to invest in AI. I mean, part of the reason we've seen such dramatic uptake is because most of these leaders are like, well, we need to reinvent our businesses because we are already actually doing so much of the complicated and game-changing R&D. And yeah, we're going to benefit from some of the tools that people are going to sell us to help us do some of that molecule research faster. We need to reclaim the story and accelerate how much we're pushing kind of through the market. I mean, even like GLP-1s, how much credit are they getting?

I was just saying this. Yeah, totally. Totally. Yeah. It's like you sort of cured obesity pretty close to it. That was a huge problem.

Yeah, but what have you done in the last day?

I'm just imagining, like, like, like in, you know, Amnesia, you wake up and you're the CEO of Pfizer and you're like, what do I do for a living? It's like, you help sick people. And they're like, I must be loved. Like, actually quite the opposite. Crazy times.

The story around GLP-1 itself is like, I think even still underrated. I mean, we've seen an influx in applications from bariatric surgeons wanting jobs in technology and trying to kind of move out of medicine because GLP-1s have basically eliminated that specialty as like something that matters in the country, which is so crazy.

Yeah, it's crazy. Last question. What is the biggest insight that you gained from working at Oscar Health that you carried into today? Like, what is the thing that you're like, okay, I understand the structure of the industry, or I understand this thing, or this—

Healthcare is easy.

Is that it?

Yeah, I mean, I think the thing that really clicked for me there is how much discontinuity and heterogeneity there is in the system, which prevents any individual player from really even tracking what's happening, let alone having control over the process. Right. It's actually what I think makes this problem so interesting and beneficial to focus on with agents.

Yeah.

Right. There's no standard case. In the system. Every doctor's office has different systems and processes. Every insurance company has different rules. Every drug has different clinical and coverage and distribution requirements. And even individual patients have their own financial situations, medical history, eligibility, and insurance circumstances. And so the level of difficulty in actually making progress before this technology was kind of hard to Uh, under, um, underwrite. And what we've seen is like, not only can you use the tech to kind of dramatically change how quickly and efficiently these processes occur, but you can actually, because you were the first company to actually see the full process and control the full process, like create new market power and change things that historically would have only been in the hands of the insurance companies and hospitals to really operate.

Yeah. Well, congrats on the progress. Amazing news.

Yeah, really wild progress.

Bain Capital, love them. They needed a win. No, we love them. Thank you so much for coming on the show. We'll talk to you soon.

Anytime.

See you soon.

Goodbye. Let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agents to deploy web apps, servers, databases, and more, while Railway automatically takes care of scaling, monitoring, and security. We have our next guest in the waiting room. We have a few other changes to the schedule, but we're moving on. Andrew from Split. Hey, Andrew, how are you doing?

What's going on?

Good, thanks for having me.

Welcome to the show. Since this is your first time on the show—

Great to finally meet you, by the way. I've heard a lot about you from—

We know a lot of people in common.

Yeah, I know. A bunch of people that Andrew used to work with. Used to work on Party Round.

Oh, that's right. Amazing. Well, good to have you here. Let's start with a little bit of an introduction to the company, and then I want to hear the news.

Sure. Yeah. I mean, I think there's, there's 2 parts to the story. There's sort of the what we do and then how we do it. And they're, they're very different. I think that what we do is quite simple. We like to say that banks move money and we move time. And the basic premise of the company is essentially to create like a net 90, but for consumers. So kind of start with this idea that, you know, what do wealthy people have first and foremost? It's time, time to make make, you know, good financial decisions and avoid bad financial decisions. Right. Everyone uses debt, but the wealthy can use it to their advantage. And, you know, people who are less, less well-off often get trapped in this death spiral. So the idea is how do we create sort of, you know, essentially room float around an average American whereby to the extent that they're paying for things and, you know, the biggest things people pay for is housing and autos and and, you know, insurance, student loans. How do we let them sort of pay it on their schedule? And, you know, right now we're at 30 days. Our goal is to get to 90 days of sort of float where they can shift all these dates around. And what we found so far, and we now have a million people using us, is that this is actually— all financial anxiety is downstream from this.

Hmm.

Once you have a little bit of room, and it doesn't have to be a lot, just a couple of weeks to kind of move your bill payments around so that it better sort of cycles with your paychecks and other sources of income, You just breathe better.

Yeah. What does actual customer adoption look like? What is it? Where does it come from? Is this direct response advertising? Are you partnered with mortgage lenders and rental buildings to offer this service? Is there some integration that you need on the other side? We've seen there's been companies that have done like pay your rent on a credit card or pay your mortgage on a credit card. And that's always felt like sort of crazy. Like, I can't imagine being like, okay, I have to take a 3% cut now. So how have you solved all of that?

Yeah, so this is in the sort of the what. There's 2 pillars to, I think, what we've built, and it's taken us some time to build it. So I think for the first 2 years of our existence, I think we're more like a lab, but we built our own foundation model first and foremost for underwriting. And it's a cash flow-based model. It's entirely trained on in-house data, and then we apply deep learning to it. You know, the performance is pretty stunning. I think we're the best cash flow underwriting model in the country today. And I think—

and with that, do you mix in credit card, like credit data, credit reports? Because I imagine that those will at least be helpful a little bit. They're not helpful at all. Or is that a cost savings thing?

We feel— we feel very strongly that it's not helpful at all. And we tried everything. We tried— we started with FICO. We tried every off-the-shelf model.

Wow.

Yeah, unless you put your own money at risk and you train your own model, it kind of gets into that sort of sovereign model world, right? You really don't get the alpha. FICO, I mean, it's crazy. I, you know, I didn't set out to sort of try and destroy FICO, but just increasingly you start to realize how ridiculous the whole— I mean, you know, it's a sacrifice we might have to make. But, you know, what it is, if you just think about it, it's a rating, right? It's like you're an Uber driver and you're just like, you're driving people around, in this case, like Like you're doing deals with lenders and if a lender likes you, they give you 5 stars. If they don't like you, they give it 1 star.

Sure, sure.

But because it's the only way that people get access to credit throughout their entire lifetime, it's really, really impactful. And I think increasingly what we find and we believe is that it's outdated and the people that suffer is anyone under 40. So our model is really, is really tuned to what we think is sort of like the core constituency these days, which is millennials. By the way, millennials are 36 now, right, on average. So our customers are sort of in the, you know, 35 to 40, their most sort of productive period of their life where, like, you know, they make more money every single year, which you'd think would be amazing for your credit score. But as you guys know, it isn't. Not unless you're doing deals all the time.

This company sounds like a working capital nightmare for you. Where is the money coming from? Is it venture dollars? wind up dealing and creating this float? Or do you have a lender or a bank?

No, no, it's— we have sort of a layer cake of facilities in that sense. I think our capital structure is very similar to, like, of a BNPL, like at Affirm, and Max Lechner is an investor. So, you know, we've learned— we have some great people that have sort of set us on this path. But I actually want to go back to just really quickly, I didn't realize I didn't address the distribution piece. The thing that we created that's really kind of fun— so again, I'll use Affirm as an example. You've probably heard of this concept of, like, credit-backed debit card.

Yeah.

Right, where you use a debit card, but it has actually a little bit of float attached to it, and it's sort of dynamic. You could swipe it even if you don't have money on it, it'll sort of, sort of stretch to fit whatever you're trying to bill. So we built this for ACH, and it's really nutty. So, so you're asking, do we partner with any? We don't. Uh, you can use SplitPay to pay any bill anywhere that takes ACH, and what it will do is we will in real time dynamically—

Do you think we need to scale investments? The next ACH, you guys said? Yeah, he's just like, who is writing all these checks? This is not what we built this company for.

I need to smooth out my bigger bets.

Well, it's funny, ACH, right, is 10 times bigger than a credit card network. It's the biggest payment network in the country, and it is like painfully old school and insecure and very, very slow. But we managed to augment it so it actually acts like a credit card. So we have a concept essentially like an off and a capture and a settlement over ACH. And it just breaks down all walls for us. So we, we're everywhere.

Yeah, that makes a lot of sense. Well, you have the credit facilities. You're also raising equity. Tell us about the latest round.

Yeah. So we, you know, really unleashed the product in earnest about a year ago. And then, you know, we did a million run rate in the first month and we're just about to cross $80 million today. And I think today is actually the team on it right now. Today is our biggest origination day ever.

Wow.

Oh, it's great to be here. Yeah. Um, let's do it. Um, yeah, so, so what's been really interesting is that no one scaled the lending business this quickly before, uh, because it's truly really, really hard because you have this interplay where you have to acquire customers in order to raise venture dollars in order to close credit facilities, and you have to be doing this basically basically permanently. You're constantly scaling. And so in our case, that meant that, you know, we did, you know, we launched the product with basically a venture debt facility and then grew it really quickly and then closed our Series A led by Khosla Ventures. And then, you know, within, within, you know, 5 months, we were 5x bigger. And so we closed the Series B with, with Khosla. And so they've been with us along the way. And, you know, I think I'm like, I've been in permanent fundraising mode for sort 12 months, I think I'm starting to feel, you know, I understand what Eric at Ram sort of feels, but I think he's like, he's sort of, he loves the grind. So I don't know, for me it's like I wanna cry sometimes.

Yeah. Well, get ready to do more of it.

I don't think so.

Oh yeah, I'm resigned to it now.

Job's not finished.

As Eric would say.

As Eric would say. Well, thank you so much for coming on the show.

Yeah, great to finally meet you.

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

Awesome.

Hope to be back.

Goodbye.

Can't wait.

Quickly, let me tell you about Figma. Agents, meet the canvas. Your AI agents can now create and modify your Figma files with design system context. And we have our next guest, Harry from Antioch. He was supposed to be on earlier. We brought him to the end of the show, but we're very excited to be joined by Harry, the co-founder and CEO of Antioch. Welcome to the show. Sorry for the switch up on the scheduling. Thank you for being flexible. How are you doing?

What's going on?

Doing great. Thank you guys so much for having me and excited to close this out with you both.

Yes.

Excited to have you here. Please break down— well, we'll get to the news, but I want to understand how you got into the business of training robots in simulation. A bunch of questions about the sim-to-real gap and all of this. what's going on in the broader ecosystem, but where have you been focused? Where are you focused now?

Yeah, 100%. So our co-founding team at Antioch all met at Stanford working on physical AI, applied AI. I spent some time at the Autopilot team at Tesla as well, and then kind of went into a bit of a company building mode with many of the same team that we have today. And so I think, you know, our observation really looking at the industries that have kind of moved the fastest over the last couple of years is that these are the industries that have sort of unlocked recursive self-improvement, right? Like earlier today we talked about— Yeah, exactly. Math. Navier-Stokes. Scott was on, I think, you know, Cognition and Devin have done a fantastic job of this in the world of software. And I think our observation is that, you know, automating the physical world is really the defining economic opportunity of our time, right? You look at the GDP that's tied up in that opportunity, but we don't have that recursive self-improvement. And so really, at Antioch, everything that we do is built around this idea of how we unlock RSI, how we unlock goal mode, you know, for physical autonomous systems.

So yeah, how do you do that? I mean, I don't know what the status quo is. We saw a bunch of people using Blender. I know you can make an inverse kinematic model in Blender. We have Unreal Engine. You can do some simulation stuff. I can play a video game and watch a robot walk around. It looks like you could learn from that, but clearly you need to go deeper. So what's missing from just build your robot in Unreal Engine, have it press a bunch of the keys until it learns to walk around?

Yeah, it's a great question. So I think, you know, broadly, the market today is a spectrum defined by 2 ends. One end is kind of exactly what you're describing. So it's these classical, you know, almost video game-like engines with really high fidelity. And on the other end, I think we're seeing a really promising landscape of world models kind of come to the fore. Right. And I think our observation at Antioch is that both of those approaches today experience a reasonably substantial sim-to-real gap. Right. There's a fidelity issue that means that if you're relying on one of those approaches in a singular sense, you're going to be missing some of the real stuff. about the real world, and you're going to experience a bit of a rough landing in reality. So our view is that, you know, the reason why that is in the sort of like classical world of video game-style simulation is it's just really hard to encode everything about the real world in software, right? Like, you go out to the real world, you find out, hey, actually the wind matters now. Now I need to add wind into my simulation. And it's this long tail of whack-a-mole.

Sure.

Whereas on sort of like end-to-end learn side, you actually experience a very similar thing, but it's about a data accumulation strategy. And in the world of physical AI, we don't have infinite data, not even close. And so, you know, we believe that these world models are going to be the right approach, but we just need to kind of play the game on the table right now to actually help, you know, real companies building real things in the present day. And so our approach is a bit of a hybrid where we use that classical simulation where it works, and we kind of learn the gaps, we learn where it doesn't. And it's helping us sort of move our customers along that spectrum, essentially, eventually towards these world models. We've seen some incredible launches from WorldLabs and others, and that's an incredibly exciting future. We think that's where the puck is headed, but we need to kind of help shift real companies in that direction over time as the technology kind of becomes completely ready.

Talk about your various robotic timelines. On the autonomous driving side, every time I've asked any automotive technology executive—

Someone came on the show and said like 2050.

It was more like 2040. For a major supplier of technology for pretty much every car, I'm not going to say too much because people will guess it, but he basically wouldn't give an answer. He was like, maybe, but—

There are people on the show that think we're getting Dyson Sphere before this guy thinks we're getting full self-driving cars. That's the range of predictions we're dealing with on this show.

Yeah. And so that's a category where we see close to full self-driving already with Tesla and Waymo and all these things. So the technology actually exists. And you have industry executives saying, yeah, it's going to— give me 15 years minimum. But I'm curious for you from your lens, I expect a lot of this stuff to come, you know, real advancements to come from, you know, new companies. And I'm curious your view.

Yeah, I mean, I think now it all comes down to the sort of data flywheel, right? And so I think the advantage that autonomous driving has— and by the way, I don't think it's 2050, right? I mean, I think we've seen Tesla and Waymo and also companies like Wave actually now doing these deployments that work extremely well. And so the reason why is these companies have built this incredible data accumulation flywheel, right? So Tesla's is particularly brilliant because if you buy one of those cars, you're essentially paying to help them train this system and get kind of more and more data into that engine. And I think now we've kind of got good line of sight on architecturally what do the models need to look like that sort of unlock that autonomous future. And so bringing that to new industries is really just a function of are you able to get the data at the scale and also in the kind of right categories, not only when things work well, but also really importantly, when things are not working well.

Yeah.

And so I think that timeline is going to be purely a question of how quickly can you build that flywheel. We've obviously got a bit of a chicken and egg kind of scenario in the world of robotics, in the world of industrial automation and these types of things where deployments are a little bit nascent. And so that's kind of really one of our key bets at Antioch is like, if you want to bootstrap that incredible opportunity, you need to figure out a way to be much more sample efficient than we are today. And so again, that's our hybrid simulation approach, right? That's like we can take a small amount of data from the real world, use that to train improvements to simulation. That simulation then trains a better version of the robot in the physical world. You can scale your deployments faster, and it becomes this kind of virtuous flywheel, effectively.

I can't wait for robotics companies to be Playing the benchmark game, right? Instead of the Pelican, it'll be like Mona Lisa. Juggle five balls.

Yeah, juggling. Juggle seven balls.

Juggle eight balls. Mona Lisa bench.

We'll be like, it still can't make me a good sandwich. Yeah, exactly. It put the mustard on the top, and I like it on the bottom. It's not here.

Possibilities are endless, right? I think we saw this with with Micro Duck in the last couple of weeks too, right?

Like, what's Micro Duck?

Ridiculous benchmarks of ducks balancing balls and things like that. We're going to see a proliferation and explosion of this.

How many customers are in your TAM? Because we've had a few humanoid robotics companies. There's a few names. Obviously, there's the self-driving car companies. There's some industrial stuff that's happening. But are you— at the absolute top of your CRM, are we talking about 100 companies? Are there thousands of these companies? Like, I can imagine building a great business selling to 20 robotics winners, right? But like, how big is this market and where is it going?

Yeah, it's a great question. So I think a couple of responses to that. So one is like, even present day, it's massive. Like, we're talking tens of thousands of companies. There's all the ones that you'd think about, right? So the humanoid type companies, self-driving, all the same kind of stuff that you described. But, for example, today we announced our partnership with Amazon, and in particular, the Ring team at Amazon, right? And so, this is smart security devices. And so, you know, essentially, it's any system that has this hardware, software, machine learning component where testing in the real world is really difficult and really expensive, and you need to kind of get that recursive self-improvement flywheel.

Sure.

And so, I think there are a ton of companies here that you probably wouldn't Typically think about as being physical AI, but really are.

Yeah. Everything from a robotic vacuum cleaner to like robotic drone, camera drone, sports. Like there's, there's just gonna be robotic pieces, even if it's not a robot, a humanoid in every category, there will be something that benefits from this. Fascinating. That's exactly right. Well, what a great, what a great industry to be in. Very excited. So glad that you have some fresh funding and good luck with the journey. We'll talk to you soon.

Yeah, great to meet you, Harry.

Amazing.

Come back on soon.

Thanks, guys.

Have a great rest of your day.

Appreciate it.

Goodbye. Up next, we have a surprise guest, Rohan, who connected GPT-6 Astra to a wearable he built for back pain, and it's giving him real-time back pain physical therapy. We're going to bring him in. I saw some people in the chat asking, chanting for Rohan.

Let's see if he can handle foot pain, because I destroyed my foot surfing this weekend. What's going on, Rohan?

How you doing?

Hey, how we doing, guys?

Nice to meet you guys.

How are you doing?

Yeah, great to meet you. Glad to have you pop on here. Busy, busy day in the world of technology, but you managed to break through.

Yes, huge.

Yesterday, a little Sunday ripper.

Yeah, absolutely. Yeah, break this down. What was your process? How long did this take? What was your goal? Set the stage for us.

Yeah, for sure. So like, actually goes back to when I was 13, had back pain for like a decade in and out of care. I didn't go to PT, but like, you know, like PT doesn't have the kind of classic modern like measure, care, measure again. It's just like kind of looking at you and they're like, all right, here's some exercises.

You know, see you next time.

Let's see how it goes. I'd back been at uni, decided like this is this is not it. Moved to my childhood bedroom, built something, found some like thread on Twitter saying who's writing angel checks for hardware? Just DM'd everyone. Luke from Boom Supersonic actually got back to me. Got on a call. First call I ever kind of took. He wrote me a check. I was like, oh wow, okay, this is something. That's me. Moved to China. I basically just took that check to China, just slept in a factory in Shenzhen, smoked. Wait, we met.

We we met. I just realized that we we met before. Yeah, sorry. I I. I didn't put it together. We had a phone call. We had a phone call. Yeah. Will introduced me and was like, you got to meet this guy. He's crazy. He's living in China. He's solving back pain.

That's amazing.

Very Chinese time in my life.

Yeah, yeah. Smoking cigarettes, I'm sure.

I would highly recommend. With AI, we will invent the smart cigarette that tells you exactly.

I saw a LinkedIn cigarette somewhere, which is, I think, hilarious.

Wow. Anyway, sidetracked.

Anyway, sidetracked, sidetracked. So yeah, built this MVP, shared it on Reddit, got like, you know, really, really good reception there, met our first customer. That's me now. There we go. And yeah, basically took that, came to the Valley, put together some money and like, was like, okay, like, let me try build this thing. And it was like, really kind of, it's just me. one-man team, taught myself how to build hardware, moved to China, build it all myself. Same with the code. And it's just honestly nice to see the fruits of my labors. I don't know, people have been coming here, like, text me, they'll be like, okay, I've had back pain, I've had ankylospondylitis, other congenital condition. How can that be used for their conditions? And it's just really interesting because I'm like, okay, well, this is great. I'm really just kind of humbled by it and really want to double down and really kind of bring this out to the world and hopefully Many, many more people than just myself.

OK, so quick timeline. You spent a long time actually building this wearable, getting it manufactured. So you effectively had a data feed, maybe an API, something spitting out data. And then over the weekend, you visualized it in this particular way and wired up a 3D model to the data that you were already collecting. But the plan is to sell the device that collects the data that then will be enabled enabled by AI and everything else.

Yeah, yeah, basically it's like, yeah, like it very fundamentally, yes, it's like for consumers, it's for people with back pain to understand their own care. 'Cause like Jordi, you said you have foot pain, hopefully your PT works out, but it's like, wouldn't it be nice that if you go to 1 or 2 or 3 PTs, you have like a ground truth? Like, okay, like my foot's getting better or it's not, but this PT's working for me and these 2 aren't. That's what I want to give people with back pain care, like in back pain care. Yeah. And then the kind of greater take there is really, I think that with what these models can do now, we can build pretty much whatever we want on the software side. And really the data collection that really rules because it's like, okay, what does the AI need to make intelligent decisions about my body? We'll have an AI PGP in our pocket, but it needs to understand what's going on with our body right now. And I'm really excited to kind of bring out this form factor of patches that you stick to your body, because I think that that can really nicely scale. Like, we've already done wrists and fingers, if you think about it. And like, now heads are kind of coming into play. We can cover the rest of the body and like kind of basically collect every biomarker that we might need for our AI PCP to make great decisions about our health day to day.

What's the plan to sell a lot of these? I could imagine everything from Facebook ads to working to infomercials and Late night infomercials or Shark Tank?

Rohan infomercial, smoking Chinese cigarettes.

No, I mean, like there's a lot of different ways to you can get reviewers and sponsors on your website.

Wonderful ironic sense.

Infomercials would rip, actually. Probably right. I mean, you're sitting there, your back might be hurting, maybe. And I bet you the inventory is really cheap nowadays. I mean, yeah.

And then also live selling is like infomercials before.

That's it. Sure, sure, sure. Yeah.

Like candidly, like one of my investors, Justin Mad. He wrote this great book, Traction, if you know. And I've like basically over the— this is actually how this came about, this kind of viral moment.

That's a Brath guy, right?

Hammering, sorry?

That's a Brath guy?

We love Justin.

He's a good friend.

If anyone knows anything about selling a lot of like a product, it would be Mr. Brath himself, right? And basically just been hammering out like a process for like, hey, let's just see what channel works. Doing a bit of UGC and like just testing it out. Tried paid, I was like, okay, Okay, let's now try Twitter. This is actually the results of me trying Twitter, which is kind of like snowboarding. And like, we'll just see how far it goes and really make sure that we are where our customers want us to be.

So people are joining the waitlist at yourbackhurts.com. Give me a timeline. When can people actually buy this? Are you going to take deposits and then ship, or do you want to go straight to order and then ship? Is this going to be the Tesla Roadster of back pain?

Good question. I think we're aiming to ship January, at the start of January. We'd like to make it a lot more affordable than Tesla Roadster. And I think we can. We've done like— I spent so long building this, like I've done a lot of engineering in the back to actually just shove down our unit economics. So basically, we— I'm trying to— I'm trying everything I can to charge less, basically, and still make it work for people. So yeah, like aiming to ship in January. Sign up your yourbackhurts.com and we'll basically and we'll open the waitlist to our first customers. Very cool.

Well, great to great to see you face to face. Yeah, and come back on. Come back when you launch. We'll talk to you soon. Yeah. All right. Take it easy, guys.

Have a good one, Rohan. We have the first results of Waldo Bench Type. Tyler put them together. Let's see. Let's see how ChatGPT Images 2.5 is doing. What'd you make, Tyler? You made OpenAI Dev Day. Wait, actually this is sort of hard. It looks good, but I can't find Waldo. I need to zoom in. Where is Waldo? Is there only one Waldo here?

Yeah.

Okay, I found him. Yeah. It needs to be a little bit more detailed when I zoom way in. If I zoom way in, faces start getting garbled, but man, some of this zoomed-in text is really, really good. It's pretty high fidelity, I think. We're getting close. I want to see— I want to see this paired with Astra, tiled, a lot more reasoning put into it. Let's finally solve Waldo bench. I think it's possible. With modern AI. What do you think?

Yeah, I can't find Waldo.

So you gave up?

That seems like a good—

You gave up?

I took a quick look.

What about in the second one?

Honestly, I was getting distracted by the beach one.

The beach one. Pull up the beach one. Everyone can take a, take a quick gander, try and find Waldo. I think this one's pretty easy.

All right, close it. Close it. You only had 5 seconds.

Okay, you fail. No, I think this one's not dense enough.

Yeah, too easy.

It's too easy. This one's too easy. This is about 1/4 of a real Waldo. For the real Waldo heads out there, they're not gonna be like, ah, this is not SOTA. Actually, I guess it is state of the art, but it's not super intelligence for Waldo generation. It's pretty good though. Looks pretty good. I like it. And then what is this animation? Explain the animation that you shared.

Yes, this is like stop-motion animation, but this is with the new image model. And I just had Astro make it into a video.

Oh, cool.

So it's much more consistent. That's one of the big—

Consistency.

Big improvements.

Very fun. Well, I look forward to getting Jordi sending me AI images at 3:00 AM when he's vibe designing the next piece of great furniture or something like that. No, it's a lot of fun. Anyway, there are other stories.

And we're going to get to them tomorrow. Tomorrow. But I'm glad we cracked the 4th hour. We're in the 4th hour. Yes. It's been a while. In the 4th hour. It's been a while. We got through summer. We did it.

Summer's famously a little slow news day, slow news season. We're back. It's September. We're going in the 4th hour. Get ready. 4th hour, 5th hour.

I got a text from a buddy who saw— or sorry, 107 days. He said he already got his tree up. I can't tell if he's messing with me, but I'm just going to pretend that he's being accurate.

If you put up your Christmas tree 107 days early, you have some serious botany to do. You need to keep that thing alive. You've got to be watering that thing regularly. There's a That's a lot going on. I don't know that a Christmas tree is meant to survive 177 days. It's sort of like a new level of challenge.

Just say you haven't cut a hole in your floor and gone down to the dirt.

That's true.

And you constantly are growing a tree, so you're cutting it and pruning it, whatever, but it's just a living tree in your home. Just say you haven't done that.

There are some houses that have little tree areas inside the home. Maybe, maybe this is the future. Design your entire house around, around being able to grow, grow a Christmas tree constantly on a never-ending cycle. Anyway, thank you for tuning in to TBP, and we will be back tomorrow at 11 AM sharp. Rolling Flashbang.

We love you.

Give us 5 stars on Apple Podcasts and Spotify.