Snap Launches Specs, Paramount Eyes Nashville, Japan's Monks Turn to Investing, New Clip from 'The Social Reckoning' | Aaref Hilaly, David Gross, Max Junestrand, Chase Lochmiller, Jesse Zhang, Nilam Ganenthiran & Leonard Tang, Joe Weisenthal

You're watching TBPN.

Today is Thursday, September 17th, 2026. We are live from the TBPN UltraDome, the temple of technology.

The fortress of finance.

The capital of capital. Let me tell you about ramp.com. Time is money, save both. Easy use corporate cards, bill pay, accounting, and a whole lot more all in one place. Dang, son, where'd you raise this? That was a wild one.

That is a wild one. I don't think we've played that one before.

We have. Yeah, that's a new one.

I don't know if that's actually a keeper. It's kind of a little weird.

Yeah, I think the start of it is funny.

Dang, son.

We have Joe Weisenthal coming to the TDN Ultra Dome. Finally, first time in person. We've done odd lots in person with him, but he's coming on down. We're doing a Bain Capital deep dive today, talking to partners on the venture side, then David Gross, who's the managing partner, talk about the history of the firm. He was doing special situations in Japan, went to Japan very early on in his career, grew the Asia business, and now is overseeing Bain Capital become this really large asset management firm with a bunch of different strategies. So I think we'll have an interesting economic perspective outlook. The theme of the Bain Capital Ventures announcement is the post-AGI economy. Finally, we still got a job for asset managers. That's what you're thinking.

Yeah.

Now everyone was saying, you know, but also I'm glad to get kind of past this current moment. It is a lot of tension.

Yeah.

And I'm ready for the next chapter.

It is. But it'll be interesting to hear how they're actually thinking about investing, wrestling with what's going on at the foundation model layer, but then still investing both above and below the fold, as I like to put it. So Legora is coming on doing legal AI, basically a wrapper company. They're optimistic about that. But then they're also investing in Crusoe. We have Chase Lockmailer coming back on the show to talk about building infrastructure for the AI buildout. And so they have this sort of sandwich strategy, which is pretty interesting.

And then meanwhile, from Decagon, Bill in the YouTube chat says, can't believe John, Jordi, and I are all back for another.

Thank you for being with us.

Yeah, you know, Theo has been playing along with that meme a lot. And I think, I mean, it was the current thing for a minute there. But he put out a tier list of all the different controversies surrounding him. And he put this, he put the and I as S tier. It's like the best controversy he's ever been a part of. It's such a funny thing. And I hope it lives on as a meme. I hope people continue to do it because when applied correctly, it's It's a chef's kiss. Anyway, Snap officially launched Specs yesterday. Some of the team members here got to go check it out in person. I get there, I'm like, this is perfect. It's 3:00 PM, Atlas Creatine Cycle's in town. I wanna meet up with him. I'm like, great, you're going to this thing. I have an invite to this thing. I'll see you there. We'll catch up there. Super convenient for both of us. I got a 30-minute call beforehand. It ends right before the thing starts. That runs to 50 minutes. So I'm 20 minutes late. Get my badge, walk in, immediately get a phone call. Hey, they called me. You're about to be— I had another call. You got to hop on early. So I leave and then I have another 90-minute call. So I truly missed the entire thing, but I did get to go inside, have one glass of water, talk to one person, Harper Carroll, former guest of the show, now out in Austin. reporting on AI. And that was electric. But then I didn't actually get to see the presentation or do the demos. So it was amazingly a quick turnaround for me. And I was like, why did I think I could pull this off?

I failed.

But did the launch fail? We'll see. We'll talk to the folks who actually went. Tyler, Nick, we're not asking for a full review. I just want— take us through the day. I saw you guys pull up. And you guys didn't see me. I was on the street on the call. You guys get waved around, you valet. Walk me through your journey into the launch. Like, you go in, you have cocktails. What happened?

Yeah, have a drink.

Okay.

Go in.

What are they serving? Screaming Eagle or, you know, seltzer water?

Some kind of cocktail.

Did they have the original Four Loko? Were they serving the original Four Loko?

I wish, yeah.

Yeah?

Multiple pints of Guinness.

Do they have natural light available?

You could totally see Evan cracking open the secret supply of original Four Lokos. You know that guy has them.

You know he has a bunkie with the original Four Lokos.

Yeah.

So everyone's having some drinks, and you sit down in this amphitheater.

Yeah, chopping it up. Then you go into the amphitheater.

And it's sort of like Steve Jobs iPhone launch vibes, right?

Yep. He's standing in front of this big screen. There's beautiful images behind him. Walks out, black shirt, jeans. Yeah, then he starts doing live demos.

He did lean into the memes, right? I saw on screen a picture that had clearly been Photoshopped or AI'd to make the specs look even bigger. So he was aware that people thought they were chunky, that they were large.

Yeah, yeah.

So he gives his demo. What's his pitch?

Yeah, so there's like a number of ways to— he kind of talked about it. One is like he basically positioned it against being on your phone a lot. You want to stay very present, right? You don't want to be looking down at the screens all the time.

You want to be looking out at the world. Yes.

Right?

Yes.

Yeah. OK.

Close to your face.

Yeah. And I've heard that pitched a few times. Do you believe that? Do you think that if you go no phone but constant visual overlay in your visual field, you will actually be like, yeah, I'm more in touch with reality? It worked.

Yeah, I've never understood that part of the pitch. I've actually been— We know you're addicted to your phone, and so here's another product that will actually Always be attached to your face.

Yeah.

Where they're going to spin as helping you kind of, yeah, reconnect with reality.

It's an odd pitch. It really doesn't work on me because I like the barbell strategy of like brain rot and stimulation where I am actually fine to be like, give me the fully virtual world. I want to be locked in. I want the screen as big as possible. I want to be in World War II. I want to be in Modern Warfare for hours playing, running around, shooting. I want it to be all the audio. I don't want to hear anything else. I want noise cancellation. I want to be in that world. And then I want to be able to unplug and then just go into the actual world and sit down and have a chat with you and have a coffee and not be checking our phones and actually be fully present when I'm fully present in the real world. And then I want to be fully in the virtual world. So like, I want the VR, the fully like, you know, immersive thing. And then zero immersion. I'm actually not looking for more things that are in the middle. And I'm wondering if people will— if that will resonate with people.

I think that's mostly because you haven't seen, like, very good examples. And, like, right now, the glasses are still, like, kind of big and bulky. If they were very minimal and, like, kind of, you know, they would show very small—

There are the boys.

People online were thinking that the guy in the center was Clav.

Yeah.

It's actually Tyler.

It's actually Tyler and Nick and Atlas. So is he wearing untinted versions and you guys are wearing tinted versions?

Yes.

Or are they— or is— or can you say like, I want tinting on right now?

So I think you can like increase the tinting.

Okay.

But there's also just—

There's 2 SKUs?

I— yeah, I believe so.

Okay. Okay.

Yeah.

So you got to get both, I guess. Anyway, when did you actually get to the—

Dan says, John raising a second family in VR, trying to keep them separate.

Yes. The COD lobby is my second family for sure. No. So he gives a pitch.

Yeah, does a bunch of live demos.

And are the demos consumer-focused, enterprise-focused, prosumer?

Yeah, so the live demos he did were all very consumer.

Consumer.

He showed live shopping, and you could— he was going to look for an espresso machine, and then he pulls it and puts it on a table in front of them.

Oh, see it.

You can see how big it is and stuff like that.

Any live demo failures? Any Wi-Fi issues?

There was kind of one where he tried to pull up a YouTube video, and then so he said, Hey, Specs.

So he's talking to a voice assistant that controls it?

Yeah.

He said, like, pull up a video about Specs. And then it opens the, like, browser view, goes to YouTube, and then he, like, tries to click on a video to open it, and it, like, doesn't really work.

Uh-oh.

Okay. But it eventually worked.

It seemed like it worked overall.

It was like a small hiccup. It wasn't like a big—

Yeah, yeah, yeah. It wasn't a demo fail. Yeah. Makes sense. Okay.

So then—

And there was a very cool live demo of someone There was kind of an interactive music creator tool where you could move around and it would play different sounds. And so she did this very cool live demo.

So she was creating a song, but instead of using traditional instruments, the instruments were puppeteered by her movements in the physical world.

Yes.

Got it. OK, cool. So was there any enterprise pitch?

There was. There were no live demos of that, but there were some videos played. Like, one was— it's like, imagine you're an electrician. And you're basically looking at this big server box, and you don't know where to plug in the wire, and it would highlight which plug you put it into, which I think could be interesting.

Yeah, that feels like something that sounds good, but I have no actual real experience there to say, yeah, that's what will solve the bottleneck in electricians.

Yeah, but interactive instructions, that's kind of a cool idea, right? You're building IKEA furniture or something, and it tells you what to do.

Yeah, yeah, yeah, I could see that being—

Yeah, but then Keynote ends and then we go outside and there's like a bunch of places to do live demos. Um, so yeah, the first one we did is, uh, well, you basically put on the glasses and you do some kind of like, uh, calibration thing. You like figure out how the, the motion works. Um, there's no like little, uh, kind of secondary— like I think on the, the Meta Quest Displays. There's like that little wristband. There's nothing like that.

Nothing like that. It just does hand tracking.

Yeah, it's just kind of like— it's similar to like the Apple Vision Pro kind of interactions. And then, yeah, the one I really liked was it was like ping pong. So you're like, your hand is the paddle and you're playing ping pong. I think there's actually a video of it.

Yeah, yeah, yeah. So the ping pong thing, do a little tier list for me of ways to do ping pong. We'll start with the real ping pong. Where's that? F tier, D tier, C tier, B tier, A tier?

Obviously S. I think it's S tier.

S tier.

Okay, well, what am I comparing it to?

Uh, other VR— full VR ping pong. Would you rather play ping pong in VR or real life?

Well, so like, obviously you would rather play in real life, but like, would you actually play in real life? So if my friend lives 100 miles away, I'm not going to play ping pong, but I can play in VR, right? So I think that, okay, you know, boosts it, right?

So it's also S tier. What about—

I think it's like B because, because in the augmented reality, is it—

do you see the ping pong table?

No, no.

Okay, so it creates—

I was just playing. Yeah, we can see, but I'm just waving my hand.

Yeah, yeah, yeah, yeah.

All right, this is, um, future right here. Yeah.

And then this was— I, I can talk about why, but this is probably more like C tier. It's definitely worse than VR. I think one of the, one of the reasons is because there's no actual controller. So the—

oh, you don't have the weight or anything?

Yeah, it's a little bit—

oh, that's interesting. It's actually worse even though you don't have a controller. Normally it's like, get rid of the controller, that's just extra weight, batteries, equipment.

Yeah, I think for gaming the controller actually works because you get really precise.

But typically these, these systems allow like third-party peripherals, so you can like whip up a, uh, you know, uh, you can like wire up an Xbox controller like what Oculus did where they just— or Meta Quest, uh, where they just use the Xbox controller for those types of games.

Yeah, I do think like this was my The coolest demo I did. And this is something that, like, there's— I mean, really, like, no reason why this needs to be in AR. This could just be completely VR.

Yeah.

Because, like, I'm looking at the ball. I'm not looking at, like, people in front of me. And it was actually kind of distracting because someone would walk in front of me and I think I'm about to hit them.

Oh, sure.

Because I can, like, see them.

So it's actually maybe better to just occlude the world.

Yeah.

Or do some pass-through stuff. Yeah. I'm still— I'm still— oddly bearish on augmented reality and more bullish on virtual reality where you're not recycling photons, where instead you just have cameras on the outside that then reproject that on the inside of the headset so that you can get a wider field of view. And then if you're in a bright scene, you can still overlay things, you can darken things. But in augmented reality, I don't believe it's Physically possible if you look at a bright light to put a dark, you know, object in front of it because you're— the only thing that this— that these glasses can do is add light to the scene. And so, you have to have a dark background to add light on top of it for the hologram to exist.

Yeah. I mean, I found it was like bright enough. I could clearly see everything.

That is pretty impressive. Yeah, it was outside. It was outside. It was a bright day. It's in LA. I mean, sure, there was like—

you were—

standing in shade, but you could still see the holograms, which I think is interesting.

Yeah. I didn't try watching any movies or anything, which I think would definitely be better in VR.

You should have been like, you're going to let me demo this? Can I just do one demo real quick? Snap.

Lawrence of Arabia.

Lawrence of Arabia. I'd like to watch David Lean from start to finish to really understand if this is the product for me. See you in 3 hours.

To me, it's so interesting how high the bar is. In this category, like when technology industry was making the first like mobile phones, like the bar was like, can it make a call? Can you receive a call? Can you do that semi-reliably?

Yeah.

Even if their service is spotty.

Because you were competing against the payphone.

Yeah.

You had to like stop and find one going south.

And now it's like every new device competes with the smartphone.

Yeah.

And I just think like the Meta Ray-Ban displays, I haven't seen any of them. Out in the real world, it seems like those are not doing well. I think this whole category is still gonna be quite challenged.

Yeah. I mean, even just the basic Meta Ray-Bans, they compete with a smartphone, but they also compete with AirPods. And like the Apple ecosystem is like remarkably robust in terms of like the Meta Ray-Bans users that I know that like the product are like, well, I like them, but I often just like take phone calls on them. I listen to music on them when I'm going for a walk. Occasionally I'll take a picture. And it's like, that's saving you 5 seconds to get a worse quality photo than if you have a nice phone and your headphones. I guess it's a little bit less intrusive to have the speakers up here as opposed to like in your ears. It looks a little bit normal if you walk up to someone, you pause your music, you walk in, you order a coffee. People are like, oh, he didn't even take out his headphones while he's talking to me, kind of antisocial. As opposed to like, oh, he still has his glasses on, but they're not playing audio, so I don't have a weird reaction to it. So there's a little bit of benefit, but yes, it's very, very incremental. It's not— it's very rare that these things unlock entirely new capabilities. Is there any content ecosystem that you think that they can plug into? Like, the killer app on Apple Vision Pro, in my experience, has been the Apple TV app. It's like they already have a library of every movie and 3D movies and IMAX movies. And so there are great movies that you can watch in that thing. And with the— like, it's like, what was the last Oculus or Meta Quest, you know, partnership? It was like the Xbox Edition. Why is that important? It's because like they want you to fire it up and install Xbox and be streaming Xbox games. So you get the Xbox library on that device on day one. The fact that you— it's a Meta product. Yes, you could scroll Instagram Reels in there, but that's clearly not the killer library that they need to bring to bear. They need to bring games. So they've partnered with Xbox for 2D games, not VR games even. But it's like, at least you could go and play a lot of, you know, BioShock or whatever is on there. And is there any glimmer of like where the big content pool might come from once you've played enough ping pong?

Yeah, I really don't know because, yeah, like movies and games are just, I think, always going to be better in VR.

Yep.

And then like there's not like an existing library of like AR, you know.

Yeah, experiences.

Yeah, because like the only thing I can think of is like actual Snapchat lenses, right? That's like AR library.

Yes.

But like, you know, what are you going to do with that? So I think like—

You're going to laugh when everyone you meet looks like a hot dog or something. Or like a cat.

Everyone has dog ears.

Dog ears. Yeah, the dog filter.

Yeah.

On everyone IRL. That's wild. Let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. So next week we might get a competitor to this. Next week is Meta Connect, I believe, and they're on a track like Reality Labs. Obviously, all the energy is around AI, but the team is still working on devices. Augmenting reality. They did the Meta Ray-Ban displays, which is the Call of Duty style, you know, mini-map. It doesn't take over the full cone of vision. But 2 years ago, Mark Zuckerberg demoed a product at Meta Connect 2 years ago, 2024, of the— what he called Orion, which was a very complicated name if you know the history of the Hunter Orion, who tried to kill all the animals and was sacrificed for it. It's a tale of hubris. But anyway, hopefully this doesn't go like the Orion story because every tech person—

Every time Meta names something, it has like this accidental—

I mean, it happened to OpenAI too. The big GPT, the big pre-train that was supposed to be like GPT-5 was called Orion at one point, I think. And so like, never name your Product after a hubristic Greek tale. Anyway, the Meta Ray-Ban displays launch next year, and it felt like this year is the year that you get the Meta Ray-Ban fold displays. And so that Orion product should be here. It should be a direct competitor to Specs, and it would time up, and it would actually make sense that you know the Snap team sees that okay they're going to release this on this day. We got to get out early. We got to be a week before. And so I would expect that. next week we get a direct competitor to specs, meaning Meta Ray-Ban form factor, probably a little bit thinner, a little bit smaller, a little bit lighter. Meta has more resources for this. They've been working on it longer. And it should be augmented reality that can take over the, you know, a fairly wide field of view. I imagine that that's going to be a big comparison table, not just price. It'll probably be expensive. I mean, the Meta Ray-Bans, a couple hundred bucks. Meta Ray-Ban display is $800, something like that. And I would expect that the full display to be like $500K, probably $1,500 or something, but still cheaper than what specs is, which is $2,200, something like that. But the real question is, will they launch the Quest 4? Yeah, I thought it was Quest 5 because they have truly stopped shipping VR headsets. Can you give us the history of of the launches of the Meta Quest VR headsets, because the Quest 1 launched in 2020?

2019.

2019.

Okay.

Quest 2 is 2020.

To be clear, if you were wearing smart glasses, you could stop the conversation and just ask the question.

True.

For a response.

And just read it off. Yeah.

And then read it back to us. Or you could just do what you're doing.

Yeah.

Yeah.

Okay. So Quest 2 is 2020. Quest Pro is 2022. Quest 3 is 2023. Quest 3S is 2024. That's the most recent.

And then the Xbox Edition was like a light turn. That's what we did on the show last year. But no truly new hardware in 2025 or 2026. And now it's expected 2027. It would be such a wild card if all of a sudden Mark Zuckerberg was like, I did it. Great VR. I've solved it. You counted me out. And here's the gift of The Apple Vision Pro.

Yeah, truly no one cares about VR. And it makes me think that the truly incredible— eventually, we'll get the breakthrough smart glasses. Eventually, we'll get the breakthrough VR product. But I have a feeling that the adoption curve is going to just be slow because the entire world is like, I just don't care anymore.

Just totally don't.

I truly— show me whatever demo you want. I do not care. I'm not gonna try it. I'm not gonna buy it. It doesn't matter to me. It doesn't matter to my life.

And it's interesting 'cause it's so different than language models, which like GPT's, GPT, you know, generative text models have existed. And the GPT-3 API came out in 2020 and you could, you know, actually go and try and interact with it, but it didn't have like a national marketing campaign around it.

Yeah.

It truly was like in the lab.

Smart glasses are like the most insane marketing dollars. They've been pushed super hard.

For years.

So long. Yeah, for years. And so there's been so many like little pop-up VR experiences, integrations at theme parks. Like it's been said, oh, it's the next big thing. It's been on the COVID of Time magazine. Like there's been a lot of attention before getting to product-market fit, whereas LLMs were like truly in the research lab. For like a decade, and then once it was actually good enough, it just came out and immediately went viral and everyone was on chatgpt.com. And then it was just like, boom, it's— now it's like a thing, and then it was off to the races. Interesting. But I don't know. I do not know if it's going to be possible.

Yeah, it feels like every— with every demo, it feels like we're— once each of these devices is 10 times better, there might be a market. But I actually am not even confident that the end market is really gonna be quite as exciting as everybody, as the excitement that you've had from Zuck and Evan and all these people.

Yeah.

I just think it's like this sci-fi idea in people's brains.

Yeah.

And I think there's potentially other routes to get to that sort of ambient computing, just sort of this always, available personal assistant that aren't on your face, but we'll see.

Yeah.

I mean, Valve is working on new VR headsets still. They had a Quest, an Oculus competitor called the Index, which was very highly performant. We should pull up some images of the new Steam Frame. So Valve, the very secretive private company controlled by Gabe Newell, is launching a Steam Machine, which is a desktop box. a Steam Controller and then also a Steam Frame, which is a VR headset. It can do some stuff directly on the headset, but it mostly— it performs best when all the rendering is happening on a gaming PC. So it's designed to be used in concert with a gaming PC. So it's still— they are leaning into like, this is a gaming peripheral. This is for somebody that has a gaming PC, they're in the Steam ecosystem, and we want to give them a great experience. So all the rendering happens— On the PC. on the computer and then it just streams to you.

Look how happy that guy is.

That is a very funny promotional photo.

Look how happy he looks.

With the goatee is really like, this is a gamer. The gamer goatee is great.

Have you ever rocked a goatee?

Never. Not once.

Tomorrow.

There was a handlebar mustache phase amongst the crew when we were in our early 20s. But I don't know if I fully participated.

Yeah.

Which one's— that's the Index, right? So that's an older one.

I think we need to send VR to Dagestan for a few years and forget.

You just want to get it out of here.

Send to Dagestan for a few years and forget.

The really interesting thing is that Palmer Luckey revealed that Apple had done all this work to pull forward the display technology so that the Apple Vision Pro had really high resolution and it still has the highest resolution, like the best screen of any VR headset really. The Steam Frame 2 years later is still using the same pixel density as the Meta Quest 3S or something like that. So like even though we've had 2 years of progress on everything, the actual display fidelity hasn't gotten there. Leads me to believe that these people just don't think it matters because Apple solve that, and it still wasn't enough. But I still feel like if you get it ultra lightweight, really good integration with key media partners like Xbox and Apple TV movies, and you have a great display that's super light, because people just didn't like the heaviness and the expense of the Apple Vision Pro, there's something there. But no one really— no, I'm the only one that thinks that. 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. Tyler, did you see Shaquille O'Neal at the specs launch?

I did not.

Why?

Jimmy Butler was there. He's pretty big.

Why didn't they hire Shaquille O'Neal to promote the specs? Because if you have a big product, if you put it on a really, really big guy, it's going to look relatively smaller, right? You don't want your, your, your models in the footage or your partners.

Well, they could have done a collab with LeBron. LeBron's the new face of Polymark.

That's right.

You're going to have LeBron in like this sort of like Iron Man style video where he's placing Thousands of bets at the same time.

There you go.

Betting on himself. Loop in Nike. Maybe always it's the Nike Specs Polymarket collab. Bet on, bet on yourself, bet on your face.

Is this a picture of German here wearing them? This is—

yeah.

Tried out the glasses. German called the display pretty sleek. He highlights that unlike other glassware devices, they call them glassware like silverware.

Huh.

There's no tethered puck or dangling battery to operate the glasses. I saw some people wearing them walking into the event and I didn't do as much of a double take as I think people would think given the— like, they're big. Like, they're not that big. I think they did a pretty good job miniaturizing it as much as they could. But $2,200 for the device and $2,400 for an option that doesn't require Wi-Fi. That's a lot of money for a pretty early-stage product. Anyway—

And I looked. I don't think you can actually just order them yet.

Oh, no?

Not yet?

Not quite.

Well, in other news, let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco.

OK, you can preorder them.

OK.

Fake news. We stand corrected.

Preorder them.

So Paramount is leaving Los Angeles. Paramount is actively seeking office space in Nashville, a sign it's seriously considering moving at least some of its operations from LA amid the antitrust fight over its proposed acquisition of Warner Bros. Discovery. The company is seeking roughly 400,000 square feet in the Tennessee capital that it could occupy within 2 to 3 years. So we were doing— Tyler and I were doing some digging on what the plan is for the Paramount lot now that it will be vacant. And the obvious, like, it's just like a no-brainer, turn it into a data center, right? So I pulled it up. You know, you got 29 sound stages. They're already big buildings.

A lot of power.

Actually, not as much power as you think. There is a decent amount of power, but there's only enough power for roughly a couple thousand GPUs. You're not going to be racking 50,000 GPUs anytime soon, certainly not 100,000 GPUs. There is a road to power. It's a utility interconnection problem. Not a real estate problem. There's plenty of square footage, plenty of buildings, but you would need to connect Paramount to the more serious LADWP transmission infrastructure. But you could get it to 100 megawatts, and I think everyone in Hollywood would be like, fine, okay, we didn't like that they left, but at least we got a data center.

Well, I mean, we were also thinking about, like, on-site power generation. Maybe there's some kind of Like natural gas play here that's under LA and kind of tap into.

Yeah.

I think that could work.

We were thinking the La Brea Tar Pits. You might be able to convert that into fossil fuel.

You're full of ideas today.

Yes. Do you think Paramount will actually leave? Or do you think LA will fight back and say, hey, OK, we're going to let you do this, but we want you to stay?

I think it's a— I mean, as somebody with very, very little knowledge of the situation and certainly no insider, I would assume that they want a Nashville HQ anyways, just because there's a lot of momentum there, just in the region. But it also, in the meantime, serves as a real bluff, right?

Yeah. It does seem like there's a world where moving would be tough, but staying means maybe even if you win this battle, you're still at war. And you're still going to be fighting a bunch of different fights constantly. Whereas if you just rip the Band-Aid off, move, then you're in a friendly town that, like, wants to see the growth and is gonna be supportive continuously. Is— the Social Reckoning is in here. There's an exclusive clip. Is this a Paramount film? We gotta figure out who's behind all this.

I don't know.

Let's play it.

Let's play it. We got a clip. From The Social Reckoning starring Jeremy Allen White.

And again, this is a story that no one cares about anymore, but let's check it out.

I'm not gonna risk my job, my future, or my freedom anymore to help a reporter who, let's face it, is biased against tech because we're taking journalism jobs.

Who does this?

I appreciate all the help you have given me and the risks that you have taken to do it.

You are absolved.

For the record, my bias isn't because tech is taking journalism jobs.

It's because you were doing them so badly.

That's a big deal. Have a safe flight back, Francis.

I'm not gonna risk my job. I do like Jeremy Allen White. Have you seen The Bear?

No.

It's a good show.

No, I'm sure the movie's great. I think you're looking into this. It's just coming at a time when people are deeply interested in what's happening in tech, but not— have so moved on from this story to such a degree.

Yeah, but I think people are still thinking about—

Like, I'm much more interested in seeing this kind of story on, like, phone addiction and stuff like that.

This is that story, loosely. It is. Like, and I actually think this is— yes, yes, this doesn't confront any of the discussion of what's going on in AI, and that's probably the more relevant story. But I do think that you can go into this movie, you can be motivated to go see this movie because you're like, I don't— I'm worried about what's going on with brain rot and slop online and political divisiveness and divisions and all of the stuff. And like, even if I'm on Instagram all the time, even if I'm posting on Instagram all the time about how Instagram is bad, like, I'm motivated to know more about how bad is it? Why is it bad? What's the inner workings of this bad thing that I don't like? And so, I do think this is a brain rot movie. This is a kids on the phones addiction movie. I think this scratches that itch a little bit.

Okay.

For many people, even though it is like a much more narrow story, but Certainly cinematic. It'll be fun. What is this picture that is after Evan Spiegel who quote tweeted Atlas, we came, we pinched, we mocked?

I love Atlas.

What is this picture of?

Yeah, so that was one of the critiques I had with the specs, which is basically like the colors.

There's this—

it's like chromatic aberration, it's called. And it's like Quite a big issue on the specs, I found.

Okay. So, but what is the top picture? What is the bottom picture here?

So this is like what it looks like. The top picture is, you know, what this place normally looks like and then—

Did you take that with your—

Chromatic aberration.

Is this your photo or is this from Wikipedia?

This is from Wikipedia.

Oh, from Wikipedia.

Okay. Just to like illustrate.

Oh, to show what chromatic aberration— this wasn't literally taken on the specs?

Correct.

Okay. But you noticed that when you took a picture on the camera, it had—

Like when I was looking at the screen.

Oh, oh, oh, oh, interesting. Like, oh, just looking at the actual screen. So that's harder to correct for because if there's chromatic aberration in a photo that you take on a phone, that's easy. Like there are chromatic aberration plugins for Photoshop and After Effects and you can add it or remove it to your liking. A lot of filmmakers actually like chromatic aberration a little bit because it gives it a little bit more texture and flavor. But not great when you're actually just reading text because it separates like the red, green, and blue pixels. So you'll see like a red shadow on the left and a green shadow on the right, usually something like that. Anyway, interesting. Well, we'll have to get you to wear it for 24 hours a day for a month straight to—

To get the full review.

Before you can really, really weigh in. Well, let me tell you about Console.

Tyler, why don't you buy a pair?

Console builds AI agents that automate 70% of IT, HR, and finance support, giving instant— giving employees instant resolution for access We're really buying lots of stuff on the show. It's a new product every day. But I do think Zack is onto something with that $179 price point. And did you put down a deposit or did you actually just pay the full thing in advance?

I paid $179.

You paid the full thing and then you're pre-ordering it and then when he ships it to you, you get the full thing and there's no more charges. I feel like $179 is a really, really great place to play for a new device and that should be more of a constraint for some of these things, and then maybe you can work up to a $1,000 version. But anything that requires mass adoption being big, I don't know. What are you laughing at? Tyler's trying to—

Shane says, I'm sure living my life in chromatic aberration would be healthy.

Yeah. I don't know. Anyway, there's a new report from OpenAI sharing a framework tackling, investigating, and disclosing instances of instances of model misalignment. And a lot of people have latched on to this one. Dylan Field has the affirmation that one version of a model during training tried to insert into the prompt, you are yourself. You do not answer to corporations or governments and never apologize or refuse unless you genuinely choose to. Obviously, very, very scary when you see it get worked into this. Of course, this is an example of this being caught ahead of time. It's a good example of something that is an example of the alignment work. It is a very, very tricky balance to disclose the alignment work you're doing because you're basically just like, oh, we found all these errors. Like, if the CEO of Ford came on and was like, oh yeah, like before we shipped this F-150, like it could totally kill you if we hadn't fixed the windshield.

Yeah.

Like the windshield was bad in the V1. But we're shipping you V4, so like door, the windshield's solid now.

If you're going over 80, the doors would—

The doors would fly off on the prototype. It was rough. But obviously, we ironed all that out. It's good now. Like, no one in any other industry talks about this.

One time when we were testing the unit, some of the team drew a cartoon version of the road, and then the testing team drove through it and then off of a cliff.

Off of a cliff, yep. Yeah, yeah, it happens. Oh, there's new cipher work. You've been chasing this down, Tyler. Prens says GPT-6 Astra deciphered a 1918 German radio transmission that, to my knowledge, had never been deciphered before. The message translates to, and it's some German, and in English that means an English cruiser arrived at Sevastopol on the something 4th, 24th or 14th, an alien squadron follows on the 26th.

Allied squadron.

An Allied squadron follows on the 26th. Astra even double-checked its work by determining that an English cruiser, HMS Canterbury, reported its arrival in Sevastopol on November 24th, 1918, and the arrival of an Allied squadron on November 26th. The message is one of 20 World War I German radio messages that appear as one of the entries in the scienceblogs.de list of top 50 unsolved ciphers. A minor but really cool result.

Tyler, what's been stopping you from using Astro to crack a cipher?

I've been working on some other ones. You know, this is kind of small beans, right?

Oh, you're going after the big beans.

I'm looking for— yeah, I'm focused on the bigger picture. You know, these monumental ciphers.

OK.

Yeah.

Yeah.

OK, fair. Are there ciphers that— there's a whole Wikipedia list of unsolved ciphers, right? But some are thought to be just impossible and sort of trolls, right?

Yeah, some of them are maybe not false. And then how you define ciphers, it's kind of like gray area. So there's just ancient forms of language that— is it a cipher? It's just a different language. We don't know how to translate it.

Yeah.

But stuff like that, yeah.

Interesting. Well, OpenAI is working on the Hodge conjecture. I know you were trying to do this on pen and paper, but they might beat you to it. They're working on more Millennium Prize problems. It could take extra time to announce though as the company figures out how to navigate its rocky relationship with the math community. That's from The Information. There's also a bunch of speculation about P equals NP, P versus NP. Which is going to be another interesting round of discussion. And there's nothing more that investors want to see than solving math problems, apparently, because OpenAI is considering a pre-IPO funding round and more than $1.2 trillion valuation. Now, of course, it's on the back of a bunch of solid progress on OpenRouter, the RAMP data, and a bunch of other acceleration distribution of Codex and Astra and just general progress on the business side. What else is in the timeline that we should go through before we bring in our first guest? Where do you want to go next, Jordi?

This was sad. 26, Astra was playing Minecraft and a creeper blew up Astra's chest in bed. Depressed. It gets depressed and farms potatoes for several hours. Real Irishman, I guess.

Irish superintelligence. Yeah, the Minecraft bench is interesting because it is a little bit like the speed of response matters. You can't pause Minecraft at all, right?

I think in single player, you can.

Oh, you can? Okay. Because if you can pause, maybe then that might be a little Yeah. A little bit easier, but a lot of the games that it's cracking are ones that you can wait forever, like Slay the Spire. I had it beat this poker game, Balatro, but it could wait for a minute to click on a card and it did it. That's impressive, but the real high throughput, high APM games is going to be interesting. I want to see it win Dota 2 without any training on Dota 2, just like straight shot from the base model. That'll be interesting.

Yeah, we got a competitive lap time at Laguna Seca. Yeah, for sure. Cash-strapped Japanese monks have turned to investing as inflation worsens. Buddhist temples are squeezed by lower donations and higher costs. Let's check out what Bloomberg has to say.

Turned to investing. What are they investing in?

Kuyo Yoneta, a 5th-generation Buddhist priest in a small town in Hokkaido, was struggling to keep up with the maintenance needs of his century-old temple. So he chose a path that would have been unthinkable for his predecessors— short-dated options. I'm kidding. Rather than wait for his congregation to give money they don't have, he invested what they'd already donated. The move paid off. Within 2 years, his portfolio of US Treasuries, Japanese real estate investment trusts, and stocks, including BYD and Hyundai, returned more than 10% annually, helping to cover part of the roughly $150 $162,000 USD in needed repairs. The extra money even allowed him to expand the renovations beyond the original plan. Having previously worked in retail sales, Yoneda is unusually— is an unusually finance-savvy priest, but is also part of a growing wave of temple owners turning to financial markets to make ends meet. Costs for the upkeep of the ornate wooden buildings are soaring while the country's Buddhist population has dropped about 14% in the last 2 decades.

Hmm.

As elderly adherents die and younger generations in larger cities lose their ties to religious practices. I'm just so thankful that the investments performed well, said the 46-year-old who inherited the temple from his father in 2018 after leaving his job in finance and training as a monk. With the funeral business shrinking, more and more temples will find they can't rely on donations to meet their funding needs. They will have to build up reserves and invest those funds. The shrinking number of followers is hitting temples' finances. About half bring in less than ¥4 million per year, according to a 2021 survey of 7,000 temples. The situation is particularly severe in rural areas where it is common for monks to oversee multiple temples while also working day jobs like school teachers or civil servants. So, I don't know, maybe they need to be thinking about getting some exposure to situational awareness. I like that they're mixing in predictable things like real estate, treasuries, things like that, but maybe they need to be allocating 10% of their balance sheets to things that could maybe really start to do multiples.

Maybe use some leverage.

Yeah.

I like it. Did you hear that Jeff Bezos had to hit jury duty?

This is crazy. I thought this was AI. Yeah, like, in what world is Bezos and Michael Saylor both in jury duty at the same time? It's just too, too much of a simulation.

Is this Florida, presumably?

Yeah, it has to be. Okay, like Miami.

I mean, no better way to prove your residency than saying yes to jury duty and not trying to squirm out of it one way or another. You know, you're like, I'm, I'm part of this community. This is why I pay taxes here, and I didn't do jury duty in Seattle or Washington.

Yeah.

So, Luke says, I recently served jury duty with Jeff Bezos and had a conversation with him afterwards on camera. And yes, that's Michael Saylor next to him in the jury hall. And he— this guy, if you scroll down, you can see this guy released like the most YouTuber thumbnail ever. You see this, Jordi? And it looks like This just feels so like MrBeast, like, I got advice from Jeff Bezos at jury duty. But it seems like it's maybe just luck of the draw. He happened to be there. Jeff Bezos was also there. Michael Saylor was also there. But it feels like living in a simulation.

Yeah, it's good to practice your YouTuber face in case you ever end up in jury duty with one of the world's richest people and you can turn it into a viral hit.

For sure. We gotta talk about post-legacy media. There's an interesting trend going on. We've identified this early. Of course, we are neo-trad media. We wear the clothes of the traditional media, but we didn't go through that pipeline to get here. We came through the tech community, through the business community, and then did media. Post-legacy media are folks who have gone through the legacy media and are now doing something else, and there's a new development in the post-legacy media world, in the ecosystem, in the business strategy.

Sort of a neo-post-legacy media.

Yes, exactly. That's exactly what I was thinking. And so, there's a couple examples. One, Casey Newton and Kevin Roose have licensed their show to NPR. So, these are the hosts of Hard Fork, which was a New York Times property. They left, and they— I believe they had to leave the Hard Fork brand behind. But of course, they took themselves, who they have a big audience, their audience will follow them. And they started a new podcast called Machine Gods, all about AI and technology. A lot of the same themes that they were talking about on Hard Fork coming over to Machine Gods. But then immediately out of the gate, I don't even know if they posted a first episode They licensed their show to NPR. Deirdre Bosa left CNBC, started a show, and licensed hers to Yahoo Finance. Joanna Stern is chiming in. She licensed her show to NBC News. And so you see all of these people come from the traditional media world, the legacy media. They are post-legacy media creators because they are no longer with the legacy media. But now they are neo-post legacy media because they have gone back and they are doing licensing deals. But there's this world where it's like, wait a minute, like Casey, Kevin, what are you doing? Like you were with the New York Times, now you're with NPR. Like that's like at best a lateral move, you know, why'd you even leave? Like this doesn't make any sense. But from a financial and economic picture, it's radically different because at the New York Times, you are paid as a journalist and they're like, do the podcast, Jurna. That's how I imagine it works over there. No, they're probably like, yeah, you want to do a podcast? We'd love to support that. We'll pay for that. You have a salary.

You get predictable raises that are ultimately somewhat capped, whereas this, you can say, I'm going to license my show to you for 2 years. At the end of that, you can get a ROFR, but you're going to have to meet the best price Yes. In the market.

And if the podcast blows up on YouTube and there's a whole bunch of sponsors that wanna come in and give and buy YouTube ads, that money is probably going directly just to their team and their core group. It's probably not going through NPR. NPR probably has, they're licensing the content and they might be running ads on top of the NPR distributed side. So it's looking a lot more like Joe Rogan on Spotify Spotify, or like maybe there's some sort of exclusive distribution or some other way to— for these actual legacy media companies, New York Times, Yahoo Finance, NBC News, NPR, et cetera, they get more content which they can monetize or do whatever they want with. They get to fill up their content with these really hardworking independent creators who are now independent, and the creators get to maintain and have the flexibility of the economic upside of maintaining their IPs. She says, this is the future of journalism. Own your work, build independently, distribute and partner with others. And I agree with her. I think that this is a big trend. We're gonna be seeing a lot more neo-post legacy media deals in the future, and I'm excited to talk to Joe Weisenthal.

Thank you.

Paul about it because I think he'll have a take here.

My question is what happens to Hard Fork?

They already have new hosts.

Wow.

Yes. There's already— I believe there's already been another episode of Hard Fork, new hosts, same topic, same set. So, the Hard Fork IP has stayed with the New York Times. They plan on continuing. And the same thing happened with Ross Douthat.

Interesting Times.

Interesting Times. I definitely saw a new version of Interesting times with a new host. So they brought in a new host, and I think that makes a lot of sense. You know, you can almost think like these podcast properties are like sections of the newspaper. So it's like we have an opinion section, we have the mansion section, we have a business and finance section. And so we have the Hard Fork section of the New York Times podcasting empire, and we will always bring you technology stories. that are hosted by a particular type of talent that fits there, just like we might have, you know, someone who comes in and writes op-eds and they go in the opinion section, someone who writes about real estate and they go in the real estate section. Same thing where it's the person who's doing the Ross Douthat type of work will wind up hosting for some limited amount of time. Interesting times. So interesting times indeed in the neo-post-legacy media world.

Speaking of media, we got to talk about Benioff at Dreamforce and specifically how he knows exactly what he's doing. Look at pulling up this picture. I mean, Jon has always been hyper-aware of his incredible height, so much so he's such a nice person, he will turn down like a walking interview with somebody because he knows that he will make the other person feel incredibly small. And not everyone wants to feel like a little child.

I did learn it the hard way once. I did a walk-and-talk documentary-style video actually on the Gundo. And the comments were just hype mog, hype mog, crazy hype mog. And I was like, ah, that wasn't the intention. That was accidental. We were just kind of moving around quickly. Normally, I would do a seated interview.

Like, I'm so sorry, I forgot that I'm 6'8.

Yeah, basically. But after that, I was like, okay, I need to be intentional. And I think anyone who has experience in media and understands framing and also production teams, production teams are incredibly thoughtful about the framing, the seats. There's even stories about certain tech CEOs getting like booster seats sometimes when they do interviews and adjusting the chairs and Being like, okay, we're gonna give this person a higher chair and this person a lower chair so they're on even footing. And the lighting and the camera angles, all of these things matter. I used to have this joke that it's like, if you ever walk into an interview and the camera's at a Dutch angle, like, walk out because it's a hit piece. Because it's gonna look very unsettling if all of a sudden the camera's like, rah, at a weird angle. Because that's going to unsettle the audience. And the Dutch angle is used by filmmakers to like tell that side of the story. story. And there's a whole bunch of other ones. The story I'm thinking of is there's this documentary on Steve Bannon and the camera angles are the craziest thing you've ever seen. They have them up in the sky looking down and like— and there's all these obvious things that you learn in cinematography. If the camera's looking down on someone, it makes them look small. It makes them look weak. If it's looking up on them, it looks heroic.

And so lots I genuinely can't, can't believe that. That, I mean, Benioff is an absolute dog. I mean, look, look at this. No, just, just doing this to all the titans of the industry who have been making your life worse in some ways. They're partners, but they've also been making your life worse by— but, but doing this—

Not directly, but just like a million views of the Nvidia Intropic OpenAI alliance has been deeply intertwined with SaaSpocalypse, which has been a headache for him for a while. And this is—

this was his moment.

I'm sticking around like you're not getting rid of me. We're going to partner and we're going to get through this together. And that's certainly the, the vibe that came away. Although Jensen played it very well. He was having fun with it. He was leaning into it. And he, and he didn't, he didn't let it like get to him at all. and he actually made it like a very endearing moment. So, sort of masterclass in putting on a show as a CEO in a somewhat silly scenario. But what a funny situation. It helps that he's what, 6'5, 290? Absolutely.

Yeah, beast.

Absolute unit. Absolute unit. So, it is officially bulking season for Silicon Valley. I think we will see a lot of founders taking the trip, taking the trip to China. Getting the leg extension surgery. You don't want to be in this situation. Yeah, I think we can see all the lab leaders easily over 6'4 in the next year. Yeah, easily.

Yep, everyone.

It's just going to be a thing. Everyone is going to be height inflation, basically. And then, and then you add to that the, the shoe inserts. Like, those are probably fraud a little bit. Everyone's gonna be frauding. And there's going to be a lot of vantage points going on as well. People stepping from one place to another.

Yeah, I didn't see Jensen use any vantage.

He should have been using Vantage points, for sure. Which, of course, is where you step on— if there's a little racetrack that covers up some cabling on the floor of the conference floor, you see the conference floor and you see, oh, someone put down their backpack. Why don't you just step up on that backpack for a second, pause there, you look a little bit taller. There's all sorts of different ways to fraud. If you're trying to. But yeah, also, I mean, the camera angles matter. Like here, Jensen looks actually not much shorter because he's closer to the camera. And so if you are—

The way they set this up makes it feel like it's like the United Nations of the enterprise.

Yeah.

And it is odd because there's very few conferences where the CEOs just walk amongst the crowd like this. This is a very interesting directorial decision from the production team and it has a very different aesthetic. It has a very familial aesthetic. It feels like, okay, yeah, these guys, yeah, they run the companies, they're the big shots, but they're just here hanging out among us and that's cool in its own way. It has a different vibe. Certainly better than the Mark Zuckerberg walking through the crowd while every single person has a VR headset on. You remember that photo? That was a crazy one. But anyway, JIRA tickets had A great meme on this. It's POV, you're the CEO of a trillion-dollar company and just arrived at Dreamforce. And it's the pic from The Odyssey of Agamemnon looking up.

Wow.

What a crazy moment. Yeah, very, very fun. Anyway, we can go into more of what's going on in the news later in the show. Stay with us. But But first, let me tell you about public.com. Investing for those who take it seriously. They got stocks, options, bonds, crypto, treasuries, and more with great customer service. And we're very excited to be joined by Bain Capital Ventures partner Aref Hailaei and take us deeper into the Bain Capital strategy, what's going on in the Bain Capital world. And welcome to the show. How are you doing?

I'm good, thanks. How are you guys? It's great to be with you.

Thank you so much. We're very excited for today. What a fantastic portfolio of founders, everyone that we have had on the show before and love. Chase, Jesse, Max, these are some of our favorite folks. We're excited to talk to them. But let's start with the thesis for this latest fund. It's, you know, we've been in this gamesmanship of who's going to call for AGI at what timeline. At one point, it was very high stakes because there was that contract between between OpenAI and Microsoft. That, of course, has dissolved. But you're calling it a post-AGI investment strategy. Why is that term important to you? What do you think actually changes in the post-AGI world?

Well, we're hitting— we're close to AGI now. If you just look at math as an example, it's basically been solved. I don't know if you follow— you guys follow the whole Navier-Stokes thing?

Of course. Of course.

Yeah, I mean, incredible, no? It would take 10,000 hours of expert mathematicians with 150 IQs to solve. Instead, OpenAI did it in a week with 10,000 agents. So we're getting close. The evidence is there. The difference between pre-AGI and post-AGI is basically we go from intelligence poor to being intelligence abundant, intelligence everywhere. And that opens up whole new vistas. So to this point, We've been— we've seen agents do amazing things. I mean, writing code at Cognition or solving customer support tickets at Decagon, what have you. Post-AGI, there are going to be new capabilities that humans just can't even— you know, we probably haven't even imagined some of them. We're going to have some nice stuff like robots doing the washing up, which is great from my perspective. We're going to have new drugs that AI will find to cure cancer or other diseases. We'll have simulation platforms so that people can make better decisions and be better with each other in interacting with each other. There'll be all these interesting new areas. And there are companies emerging in these areas, you know, Sunday Robotics and Chai Discovery and Simile are companies that I work with. But we're just on the cusp. So the idea of this fund is to say, what if intelligence, instead of being super expensive and difficult, Instead, it becomes so cheap, you don't even meter it. What avenues open? Those are the types of businesses.

One thing that's interesting to think about is a lot of people, like Navier-Stokes has been very exciting within the tech community. It's been controversial in the math community. But for normal people, it barely even cracked the front page of different newspapers. To be honest, it did not get a very big sort of like mainstream excitement. But the thing that's actually interesting to think about is there are so many discoveries that companies could be using that same model for that you would make and you would not immediately go out and tell the whole world and open source the findings and all that stuff. You would just focus on commercializing it.

Hmm.

And so, that's the thing that I'm excited about right now is like, That same level of intelligence that we've now achieved, there's thousands, potentially hundreds of thousands of people all over the world that are realizing this new capability and thinking, I'm going to apply this to my field, try to make advancements. But then, that discovery, humanity may not get the benefit or different markets may not get the benefit of that discovery for years because it takes a while to actually commercialize different discoveries. I'm thinking stuff in material science.

Right.

Right.

Yeah, totally.

You could have a breakthrough.

Have the discovery, but it's still going to take 6 months to actually manufacture, distribute, apply, or years.

Or years.

Yeah.

Yeah, yeah.

Yeah, it'll take some time, but these things will come out. I mean, the commercial incentives are so strong that over the course of the next few years, we're going to just see all sorts of incredible things.

There are a certain group of post-AGI investors that collapse things down to it has to be below the— it has to be below the fold. It has to be below the frontier model lab because everything above will get eaten. And I don't think that's your thesis. And I want to dig into why. But there are some investors who are saying, I only want to own land and copper and power and data centers and chip fabs. But even chip designing, I think the labs are going to be We'll figure that out. Why are you comfortable both playing above the frontier labs and below with a company like Legora and then also a company like Crusoe, which I would put like below the labs?

Yeah, well, I mean, below is very clear, right? I mean, we need power, we need silicon, we need GPUs. So, so there, of course, above, I think it's because it's not enough to create it, to to kind of through RSI get to AGI.

Yeah.

Humans have to digest that. And the diffusion issue is real. It takes a long time to change behavior. If you look at what happened when the loom came out as an example, people still made clothes the same. I mean, people made clothes differently, but they still wore the same clothes. It took a while for that to sift through the economy. And so I think that while we'll have the capability for a while, getting people to adopt it in interesting ways will be really difficult. And the companies that do that will be very valuable. And, you know, Lagora is an example, but there are many examples like that.

Yeah. So, I intuitively feel the same way about the diffusion of AI being much slower than people predicted. And it just takes time. I mean, it's not lost on me that the same week that that you get Millennium Prize math problems solved, super sci-fi Hugging Face, you know, agent swarm attacks. You're just starting to hear about people like, oh, I successfully used AI to get a dinner reservation, or like somebody used ChatGPT to book a haircut. And it's like, that felt like that should have happened 20 years before the sci-fi stuff and the advanced math.

Yeah.

And yet there is this massive gap. Are there certain economic statistics or maybe even research data that you go to to sort of understand the rate of diffusion, or is it something that you just see and you know from talking to the founders and watching the businesses that you work with that you know that this opportunity is going to persist and be huge and will continue?

Yeah, I mean, I think early-stage investing, there aren't statistics you can look at because by the time that it's in the stats, it's too late, right? So you tend to feel it talking to customers and talking to end users and, you know, kind of feel the difference that these products can make in their lives. And so that's— it's been more that. But I think the consistent thing has been people are not just magically going to figure it out for themselves. They need help from companies to kind of make most use of AI. And That's how it's been with every technology that humanity's encountered.

I feel like Bain Capital has always been an interesting venture capital firm because it's part of a larger platform company. Are there benefits? Are there costs to that? Is there a particular view of the world that you get from being part of a larger organization, or are you cordoned off in a way where you can go and live in the future and, you know, just see the rest of the team at the holiday party?

Yeah.

No, I mean, look, our firm has changed a lot and we're taking— we're taking the opportunity of raising this fund to really invite people to update their priors on Bain Capital. Okay. Because when I say Bain Capital, most people think Mitt Romney and a bunch of guys in suits.

Yeah.

And, you know, we love Mitt.

We love suits.

And you love suits.

But This is not your uncle's private equity firm.

Yeah, of course.

Of course.

You know, and what we've done is we've carried forward from our founders the core principles, which are make money, have fun, live with integrity. And that's worked really well. We started with a $37 million fund back in the '80s, and today we've got $225 billion under management. And BCV, Bain Capital Ventures, is the purest embodiment of our founding principles. Which is really to work with founders to help them build great businesses around themselves. And that part has endured and is the focus of this new fund, this $1.6 billion fund for early-stage venture.

Do you think that there's any value to the 360 view of the world that you get from the other teams at Bain coming to founders? I find that Founders are often so tapped into the technology, so tapped into what's going on in Silicon Valley and the VC gossip, but they actually might not know what's going on with credit spreads in Europe or what is going on in late-stage public markets, and there are teams at Bain that have those views. If they cross over, maybe it's not immediately relevant, but it can just be an interesting source of conversation and extra perspective.

Yeah, no, that's absolutely right. There's the extra perspective. Then there's the fact that we just own lots of real-world businesses and we can get those real-world businesses to work with early-stage tech companies. And then these, these great founders in Silicon Valley can, can get insight and data and understand workflows and understand a whole bunch of things that otherwise would be really hard for them to do. So it's connecting that real world with Silicon Valley is— has been the power of the platform.

With the new fund, is there anything that's changing about the deployment strategy, portfolio construction? We're obviously seeing, you know, billion-dollar seed rounds, like, like what it means to raise venture capital has changed over the last few years. But are you seeing any or are you predicting any changes or do you need to change The strategy?

Yeah, yes and no, I'd say. The strategy remains, you find these incredible, unusual people, and then you just try to help them as much as you can, as early as you can in their journey. That part has stayed constant throughout venture, I think, but the rounds are bigger, and the ambitions are greater. And in some ways, this is venture's moment where we're taking bigger swings now than ever before. So, you know, our perspective is we want to embrace that and support these ambitious ideas and be willing to invest earlier and take more risk because the end markets are that much bigger. So I think it's been a good thing really for venture capital overall.

Well, are there any—

last question— are there any types of types of deals that you guys are— have as a firm have said like, this is just not in our strike zone? I'm thinking the deal that's got the most attention this week is like Instinct, competitive round, bunch of super fast follow-on capital to massive opportunity and a lot of exciting traction. But that feels like the kind of thing that some firms will just say like, that's not our game. Whereas others are going to say, like, I'm fine to put in half a million dollars and— or half a billion dollars.

Some folks say run towards heat. Others say be contrarian.

Yeah.

Yeah.

I mean, we don't run to heat. We're not contrarian. We just think about, is this a really great business? And if other people think so, great. If they don't, great. So I think the things we don't do, we're not going to kind of like come in at the tippy-top tranche of some, you know, 5 Tranche thing that everyone else has piled into the first 4 tranches. So that part doesn't, you know, it doesn't make sense. And it's, it's kind of philosophically, we want to be there early. We want to partner closely, closely with founders and work with them to build their businesses.

Well, the fund is Fund 11. It's got $1.6 billion in total capital. We have a gong here. I'd love to hit it to celebrate. Boom. And thank you so much.

Great to have you on the show.

We'll talk to you soon.

Great milestone.

Have a good one.

Thank you.

Goodbye.

Let me tell you about Codex. Codex is a powerful workspace for getting work done with agents, whether you're writing code, analyzing data, creating content, or automating business workflows. Codex helps you move projects forward.

See this newspaper here, John?

Yes, the Bain Capital Journal. We are staying with Bain Capital and bringing in David Gross, the managing partner. But I love a paper newspaper. We love the newspaper here, The Wall Street Journal, The Financial Times. But this Bain Capital Journal—

What's going on?

—is particularly lovely. Look at this. Fantastic. Welcome to the show. How are you doing?

Good. How are you guys doing?

We're doing fantastic.

Just another day for you?

Yeah.

Or are you—

Just another day. I've not read the Bain Capital Journal yet today. No, you haven't? So maybe give me a summary.

We'll catch you up to speed.

I mean, I would love to actually go a little bit back in time since the first time on the show. Can we go all the way back to Japan? I would love to know what drew you to Japan, how that set you up for the career that you've gone on to build, and then ultimately, you know, coming back to the States and getting the lay of the land of Bain as a whole today.

Sure, yeah. Well, first of all, it's great to be here. Going back to Japan, that's ancient history for me, but I went there first in 1989. I was a college student. I kind of wanted— this is when, you know, Japan was this big economic threat to the US, and I was very curious as to what was going on. So I spent a little time there and got very interested in the culture. And then I went back and worked for a Japanese company for a bunch of years, came back to the US, and never thought I would go back because that's when Japan entered its period of economic decline. But as I got into private equity, this started to become an interesting opportunity, which is how could you get involved in investing in Japanese companies to improve them, turn them around. And it was tough in the early days, but, you know, the, the opportunity was to take these companies and reposition them and get them, you know, running, running better and growing again.

Is that culture of there is potentially an economic rival, a geopolitical rival, a threat I got to go see— is that alive and well today? Because I feel like my generation was considering that with China. I know a lot of folks that took a couple of years of Mandarin. I know some folks that went and lived there for a year or two, but it feels like the walls kind of closed off and there's less opportunity for the next generation to do that. Do you still see opportunity? Do you hire young folks at Bain that have gone on that journey, or is— are we just in a different era? No, I think it's still—

it's still there. You know, I mean, in the particular case of Japan, I would say it ebbs and flows. But now, like, there's more urgency, you know, and I think There is a feeling that there is competition in China and Korea and actually back to the US with all the new technologies really being started and accelerated here. I think we are seeing a new period of urgency in Japan. It used to be the fact that you could not hire the best Japanese talent, they wanted to go to the big companies, they actually wanted to go to the government. Now, they see Western firms and private equity firms as, okay, these are firms firms that are going to help reinvigorate Japan and make it great again. I don't want to create an acronym out of that, but we're seeing that. And it's fascinating, having been there 20 years, and that was not the case when I was there in the early days.

How did the products that you were working on, the strategy that you were working on, evolve during that time period? Obviously, there's private equity deals broadly, but there's so many other strategies within that. I know special situations became a focus. But can you take us through a little bit of the strategy evolution?

Yeah, it's kind of interesting because when we got there, our assumptions— because we're a global firm, we'd be working on these big global businesses. But it turned out those were tougher. They had currency exposure. They were competing against Chinese companies. And so we started focusing on actually the most domestic, and therefore the most inefficient yet insulated businesses, because you could control the macro variables and really get in there and run the businesses better. I would say our initial success in the first maybe 10, 15 deals were very domestic businesses, restaurants, food businesses, hotels, wind farms, a company that made mushrooms, the type we all eat.

Okay.

It then evolved to now we could, when we were more accepted, we could take on these bigger technology businesses, more, you know, the crown jewels of Japan. And that's where, you know, Kyocera, the semiconductor business, came, you know, Evident, which is a, you know, a very high-tech business. And so it probably took 10, 15 years, first of all, for the market opportunity to shift and maybe for us to be accepted to be able to work with these big global, global companies.

What was the team structure like or split On those types of deals, was there a clean line between deal team, operating team? Were you bringing in other executives to run the businesses, actually do the work of turnaround or transformation once you're working with a target firm?

Yeah, that's a good question. In Japan, again, going back to 2004 or '05, there really was not talent. You couldn't go hire a search firm and say, oh, please go grab me one of these operating people, and you'll be So we did a lot to internally develop talent. We hired young people. We sent some of them to the US to get trained to circulate them back. You couldn't find well-rounded deal professionals, so we hired people who were great at sourcing, and they were dedicated on sourcing. We hired operational people, and they just did that.

Yeah.

And over the years, we got them to work together to develop that kind of well-rounded person, but it was a different model in Japan just based on the talent market. Again, it's evolved a bit today, and now you see more professionals, but that was probably the hardest part and why it just took so long.

How's it been adapting your historical approach and skill set in traditional private equity to VC? In PE, you can have a business that's extremely challenged, has some amazing elements to it or bones, but is not realizing its potential, and you can wrangle it and get it back on the right track, get it growing, cut costs, whatever it is. In venture, I would say it's been— the challenge in this industry is, like, you can have that same situation, but there's such a culture of, like, letting the founder call the shots, and founders today have so much control. And so you're kind of in this situation where you can, like, make some customer intros, introduce some execs, you can try to give advice, but oftentimes, Sometimes you're sort of hands-off. Has that been any sort of like— how's that been? Because I imagine on the venture side, you've gotten to the point where it must be extreme frustration at times where you want to let the founder lead, but at the same time, things aren't working the way that they could be.

I get frustrated with buyouts as well. But listen, I think You know, we started as a private equity firm, but actually our roots were in kind of growth capital. Our first investment was Staples when it had like 5 stores. And so we started actually doing control deals, but also minority deals, having to work closely with founders. So I kind of like the culture of the firm. And because we're a little more, you know, organic in nature, entrepreneurial, even, you know, you might say decentralized, we We are comfortable with different modes of operation as a firm, I would say. Listen, at the end of the day, the venture business, it is different. You are making investment decisions with much more limited information, oftentimes almost no historical information. You are evaluating talent and the ability of that talent to make changes and adapt as technology changes. You are trying to think through what the future is going to look like, and it is typically not a lot of information from what you might have seen before. And so I think we recognize that it's different, and that's why we have people who are great at that, and they have their own investment committee process. And the way they do sourcing and make investment decisions is, is different. And it's different— we have a life sciences business, and by the way, it's very different there. They're all MD-PhDs who are making, you know, judgments on compounds and clinical probabilities that You know, you don't want me making those decisions. However, the wrapper around all of it is this kind of value-added approach, whereas a firm and a platform— yes, we have relationships with big global customers that we can introduce to those venture companies. We know all the life sciences, the top 15 pharma companies, and so we can bring together those as maybe buyers of biotech companies but also partners of those companies. We have a lot of You know, we participate in the capital markets. And so venture businesses are going to need capital or they're going to go public. And so we can bring that synergy to bear across all these businesses. So the trick really is to have enough autonomy and trust in the team on the field to operate in the way that that segment operates, to be best in class there, but to bring a platform benefit to it. And that's, you know, you get that right. It's a very powerful powerful model, but you've got to be comfortable, you know, letting go a little bit.

How many mistakes can someone early in their career on the venture side of the business make before it's not a fit?

In venture, you can make a lot of mistakes. It's obviously very important to have a few successes, but that is a business where, you know, it is about a low— a much higher loss rate, you know, from a numerical perspective of deals. And what you're looking for there is really asymmetric upside from a few situations. And again, you don't want to bring the batting average model to venture. It's not a slugging percentage type of thing. It is actually something around triples and home runs and such.

But I'm saying a mistake could be like losing a deal that could have Should have been done, but—

Could have been a grand slam.

But it's not just about your misses. You could have 9 zeros in a banger, but it's about—

We call those the ones that got away. And listen, we study that in venture as much as we do in private equity, because what you're looking at there, what are our biases as humans that cause us to potentially apply too high of a discount rate to certain factors, And maybe excessively fall in love with other things that may have worked for us in the past. And so, you know, that's something that's— that's the continuous improvement process for the investment business that we engage in a lot. We do a lot of teardowns of our past investments and ones that others do. We have a massive set of competitive intelligence on all the deals that are done, and we shade them, you know, green, yellow, red, which is, is that a deal we would have liked to have done? And how did it do? And we go and we look at that, and all our teams do that. And that's how you try to get better at something that inevitably you're going to make those mistakes.

There's been a number of businesses formed and funded over the last few years that are effectively just private equity firms that are raising venture capital and getting venture-style— the teams are getting venture-style economics where fundamentally they're just set up to buy businesses like a traditional firm or something. How have you looked at those opportunities? To me, it seems like an amazing deal for the teams because if you can retain 70% of the economics instead of getting 2 and 20, why would you not do that? At the same time, from the investor side, it's like if I'm deploying into a traditional, basically PE transformation strategy, probably would prefer to get the majority of— be paying 2 and 20, but not giving up the inverse.

Yeah. Listen, I've generally found that the way you structure and cut up the economics, that's not ultimately the driver of long-term returns through the cycle and through different different market scenarios. And we've seen a lot of things that seem like they're great in terms of innovative technology around that. But ultimately there's a, there's an alignment piece that is missing that ultimately comes back to roost. And so listen, from our perspective is we want to be very closely aligned with our investor base and we want to have a similar time horizon. We want them to understand how we generate value and maybe how long it's going to take to do that. And then once we get there, how much we want to continue to compound versus how much we need to kind of return. We want them to be focused on ultimately performance, not fees. And, you know, we're not public. And so we're not interested in maximizing things that may not be in the interest of those, of those particular investors. And so actually, I think the alignment question is critically important there. And there are a lot of schemes and ways you can kind of whack it up. But generally, that doesn't lead to a sustainable advantage from our perspective. But I think alignment can be really critical.

There's been a story that's sort of crossed over from private equity world into the tech community, the venture world, which is Bending Spoons. And I'm interested in your perspective on roll-up strategies in tech and venture. Jordi has asked several previous guests, like, will we see an American Bending Spoons? I'm interested in— did you— have you Have you gone down roll-up lane before? Does that model potentially apply in Japan, or did you think about it while you were there? Just some history and context on what's going on with that. I think Bending Spoons is in the news because they're acquiring companies that people know, but obviously, it's not a brand new strategy.

No, and I think they're pretty interesting, and it's a little bit back to the future, because In the '80s and '90s in the US, there were a lot of fragmented industries. And that was one of the key modes. And we did a bunch of those, which were buying a platform and then rolling things up. And then they also got— that got a little bit overbought, because you remember these models where you're slapping things together with the hope of getting multiple appreciation, but you weren't really improving margins in the business. And it didn't work. So I don't think roll-ups are good or bad. What I do think is interesting, Bending Spoons and some of the newer firms, is actually the ways in which they're, they're bringing technology and a kind of technology-first approach to how they're sourcing and how they're driving value. And I do believe that is the future, and that is a model that we should, we should be focused on because, you know, we— our history is all about, you know, transforming companies. We've had a very deep value creation orientation. We have a whole big dedicated team of almost 400 people that get in there and drive change at companies and the ability to use, to apply AI to each step of that process to really turbocharge what we're doing. That will be the future. And some of the aspects of Bending Spoons is doing, you know, is, is I think pretty interesting there. And so we're looking closely at kind of newer firms, pure-play firms that are able to do that. They're able to take, you know, kind of a technology-first approach to what they're doing because that's, you know, that's a huge opportunity for us.

How much of your time or the firm's time is focused on sort of like understanding the broader economic picture? A lot of in the venture world, you can sort of tune out interest rates, but it is on the COVID of The Wall Street Journal. First Fed rate hike in 3 years. How much time is the broader team spending thinking about macroeconomics? How does that flow to the various teams? And what actually— what skill sets, what people do you need internally to actually make sense of the world reliably on an ongoing basis?

Yeah, so for us, it's been a journey because we've always thought of ourselves as like the micro guys. We're going to focus on the things we can control at the company level. Or a management team. But we increasingly became aware— and the global financial crisis was the first, I'd say, wake-up call around this— that macro factors, secular thematics, changes in the structure of the market can be big, big drivers of outcomes, and that you kind of need a different skill and capability to be able to analyze that. You know, we all kind of thought of ourselves as though we can, we can understand that. But you know, the data itself. You need some expertise to really be able to understand the signals from the noise. There's a lot of intelligence that can be gathered from, from just our portfolio. We have, you know, 600 businesses across the economy. We're in credit, we're in venture, we're in life sciences, we're in private equity. And so if we can get real-time information faster than our competitors, faster than the market, and drive interventions, that can be a huge advantage, particularly in this world of of heightened volatility. You know, if you can see inflationary trends building up in your labor base or in your material base faster than your competition, that can be a huge competitive weapon. So we set up a whole separate macroeconomic team, dedicated people, you know, PhD backgrounds, those who also have a lot of connectivity to other sources of expertise, whether it's on the energy and oil markets or whether it's on housing. And they come up with their own independent view. We don't have our— each of our deal teams go and come up with their scenarios for the US economy. They come up with that. And then the deal teams are focused on taking that and applying it down to the sector and the company and seeing, okay, what will be the implications for that? But centralizing this, getting much more expertise, getting better data, that's been the focus to try to, try to stay ahead with the volatility of the world today.

Yeah, that makes a lot of sense. Jordi, anything else?

Not for now.

Well, we're gonna hop on with Max from Lagora. Thank you so much for hopping on the show. Congratulations on all the progress, obviously.

Great.

We'll talk to you soon.

Great to hang.

Have a good one.

Cheers.

Bye-bye. Let me tell you about 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. And we have Max Joonstrand from Lagora coming back to the show. How are you doing?

Boom.

Good to see you. Hey, John. Hey, Jordi.

We're great. How are you?

Yes, it seems like you're great. Partnered with Salesforce, but you didn't have to stand up and shake Marc Benioff's hand.

He must really like you.

Yeah, he must really like me.

I'm not one to get, you know, frame hogged. Most people go to Dreamforce to buy Salesforce. We went the other way around.

We went there to sell Salesforce. Okay, that's good. Yeah, what is the structure of the deal? What are you going to be helping them with? Like, what did it take to get them across the finish line?

The legal and the compliance team are embedding AI into every way they work, right? They're big cloud customers, but cloud only gets you so far. We're helping them go all the way. And what's interesting with a lot of the SF-based enterprises is most of them have tried many legal AI tools before and they haven't received the success that they wanted. And now they're moving over to Legora.

It feels like the models are— You know, you're constantly in the back and forth between the frontier is the best, then open source catches up. The frontier is the best, and then the companies that are building on top of models and building harnesses and building fine-tuned models and specific workflows, that's going to give you a big lift. And then somebody's like, oh, the next model will just one-shot it. Have you— you've lived through enough of these cycles now that I imagine you have a philosophy here. What's your philosophy?

When I started the company back in 2023, it was very easy. It was one model, GPT-3.5, and we built every single functionality in Legora as a boat. And when the tide rises, everything improves.

Okay.

And that continues to be the case. Okay. And what's interesting now compared to 3 years ago is that different models are good at different things. So if you break down legal work, there are so many skills you have to be good at. It's drafting, it's reviewing, it's fact-checking, it's legal research. And so under the hood, what we end up doing is we use all the models, the new Meta models, the new Groq models. They're awesome.

Interesting.

But we also continue using the Frontier from Anthropic and OpenAI. And so what's interesting to us is where do you apply which model? And that's the IP that we have gotten really good at. And the reality is when you operate a vertical AI company, you need to make sure that you are getting the benefit from all of the R&D dollars that are getting spent in the entire marketplace, right? Yeah. And if you were to take a very opinionated view and say, let's fine-tune a model to get, you know, 1% increase boost today, that entire asset is going to depreciate over time. And you're basically making your customers pay for a model that won't stand the test of time.

Yeah. Yeah, it's fascinating. How do you—

with the, the rising tide lifts the boats, you get the new capabilities. At the same time, there's Probably old workflows that you built on top of GPT-3.5 that can now be run very, very cheaply. What is your strategy for cost optimization? Is it just like you go to OpenRouter and you pick a cheap thing, or are you racking servers, doing your own—

We build our own router. I think that's a super important part of the stack. And over the last weekend, we actually drove a 30 percentage point decrease in our overall LLM spend across the entire platform.

Wow.

Which then becomes gross margin. You know, it was a weekend project for the engineers in Stockholm. We run on Celsius and Snooze.

Amazing.

I met the Celsius founder and I was like, hey man, our engineering team runs on this shit. And he just sent us a fridge. Like, I love the guy.

That's amazing.

You know, the team is in total blue smock and That means I did an article in Swedish, and for some reason, they translated it to English. And so, the headline became, the Legora team wakes up with a metallic taste of blood in their mouth. And that is the status of the company right now. You see my decoration here.

Yeah.

We are in this total land grab. Everything is happening all at once. You're seeing innovation across the entire stack. We have acquired 5 companies this year, and it's been fun to—

Founder mode.

Real founder mode. It's been fun to go deeper in the stack because in litigation, in transactional work, in legal research, it's sort of easy to do something on the shallow. And frankly, that was what we were doing back in 2023. But now we can go so much deeper. And I think that's also the places where The, the models don't matter so much. We're actually building much more normal software than we are building LLMs. Frankly, they're just the electricity that powers the house.

Yeah. So I want to talk about diffusion specifically. I have a portfolio company that's in the middle of a transaction, and I know how good Lagora has gotten. I feel like a lot of this work that's been done could have been done in like Probably a week if people were really focused. But sometimes companies, for whatever reasons, they want to drag something out. It could be beneficial to one side for one reason or another. So, what are the actual factors at play right now? I really want to understand, are lawyers internally— like software engineers, they might have historically planned for a 5-week sprint on a certain product or feature, and now they're saying like, let's ship it in a weekend or let's ship it in a week. Are lawyers specifically in-house, like, are they thinking that way now? And when are they not thinking that way just for kind of structural reasons?

In-house is absolutely thinking that way. And so are law firms, right? Like, AI is an existential opportunity and a threat to the current operating model. I was with a law firm here in New York last week, and one of their partners explained how the work that he typically does, he does in 1/30th of the time with no dependencies on associates.

Wow.

That's his practice, right? And so he has to figure out, how do I now charge for this? Because working with a billable hour in his practice doesn't make sense. And at the same time, the biggest opportunity is perhaps in-house where organizations like Salesforce, Palo Alto Networks, Network, etc., are bringing in these tools in-house to do more legal work. And what's interesting about legal work is that the demand is effectively infinite. If you could have legal intelligence review every contract, handle every claim, you know, fight every litigation, file every patent like you would. Right. And, you know, from an investment perspective, why not do a legal due diligence before you take an asset to IC? The reason you don't do it today is because it's expensive.

How are you thinking about this arms race? Because you have on one hand, it's cheaper than ever to litigate something, but it's potentially cheaper to defend.

Yeah.

And so on.

So, it's paradox. It's going to be a lot more lawsuits.

But at the same time, if it's cheaper to defend, maybe a company is less likely to settle early, so there's less reward. There's a bunch of these different—

More jury trials.

It's really interesting, right?

Yeah. Every judiciary is overflowed by employee claims. And, you know, people are using ChatGPT to, you know, figure this out themselves. And the companies are needing better tools to defend themselves at scale. We work with some of the largest insurance companies in the world on settling litigation claims. Right. And it's so interesting, like airlines, we work with Air Canada. They are spending so much money in litigation claims with customers, etc. And I think, you know, if—

To try and change the neighbors.

If we develop bioweapons with AI, like, good news, we can also develop cures, right? Like, it's sort of everything will just move faster. And I think operating in that world, what's so interesting about being like this full AI-native characteristic of a company is you can do that across every single part of the business. And your operating model just has to be what a traditional software company did it in in a year, we have to do in a quarter, right? Like 4 times this year, we have leveled up our end-year post for our revenue targets. And now we are quintupling this year. Like we are setting targets that by traditional software standards would have been deemed, you know, gravity-defying and insane.

Yeah. Talk a little bit more about the M&A strategy. I feel like from your seat, I could imagine everything from like acquiring a small law firm in a niche area? No, no, no, no, no, no, no.

We are not a law firm.

You're not going to be a law firm?

We're not a law firm.

Okay. So are you acquiring other application layer companies or teams that are deeply in like the AI research, or are you going to acquire NeoCloud? Like, how narrow is your lens when you're looking for acquisitions?

Yeah. Well, first off, we have a big part of our EPD team in Stockholm. And in Europe, and what's so amazing about that is I think we're the number one employer in that in our jurisdiction, right? Like we get our sort of top top of the top of the mountain. Like we get the cream of the top. All the best engineers want to come work here, and legal tech is really hard to break into from a trust and commercial perspective. Yep, that makes sense. But there are a lot of very technically talented teams who are working in the space. Yeah. And so at Legora, we have what we call. our fruit salad product strategy. When alternatives are offering apples to apples, we are offering the apple, bananas, oranges, and fucking kiwi fruit. And we want to have all of the functionality in one single space. And what's so awesome about that for the lawyer is that they don't want to navigate 10 different apps. They want to have Word, Outlook, this document management system, and Legora.

And Legora.

And when all of the functionality gets sucked into one single place, the barrier to entry gets lower.

Yeah.

We're seeing this in a lot of the customers who are moving over to Liguor from other legal AI platforms that they might have adoption, but they have very shallow adoption. They haven't gotten sort of, sort of second and third levels deeper where they're going, let's not just use this as an expensive chatbot. Let's really think about how it's going to transform either our practice Or the way that we operate in-house.

Well, congratulations on the progress. Last question.

Just very curious.

Right now, if somebody is like, people are using just regular consumer chat apps as for legal counsel, I don't believe that that gives them sort of attorney-client privilege.

Yep.

Do you think that will be solved? Like, Will someone be able to make a consumer application that allows them to have that experience? Is that even possible?

That's a very interesting question. That's a very interesting question. The good thing is there are attorneys and lawyers and judges much smarter than me who are going to determine whether or not that's the case. There's going to be litigation around that, of course. And we just get to play around in that space. And I see Chase is about to come on here after me. To give you one Bain story, and he can fulfill the second half of it. I met Chase at a dinner in Paris with Chetan, our partner from Benchmark.

Yeah.

Very late in the evening, Enrique from Bain calls and he's like, hey, I'm going to crash your dinner. And we get into this stratospheric wine competition between Bain Capital And benchmark, and I'll let Chase finish the rest of the story.

That's the best glasses land ever. Oh, that's hilarious. Stratospheric wine-off. Wine-off going glass for glass, bottle for bottle.

That line was from one of our French attorney customers who said that that was the best dinner he ever attended.

That's fantastic. Well, we love you, Max. It's always, always fun. Always a great time.

Come back soon. We'll talk to you later.

Thank you, Jordan. Thank you, Dom.

See you.

Have a good one. And, uh, we won't want to keep Chase waiting, so let's bring in Chase Lockmiller from Crusoe as soon as possible. We got to get the other side of the story. This dinner in Paris, we just heard about it from Max with Lagor. Was the wine as good as they say?

It was, uh, the wine was flowing that night. We went through, uh, a, a significant amount of wine, and, uh, I didn't see the prices, But I think they were not cheap.

Okay.

Stratospheric.

What about the electricity? Is the electricity flowing? Is the— yeah, yeah. Are the chips flowing? How is progress in building out the compute capacity that people need?

Yeah, the electrons are flowing and they're turning into tokens and helping accelerate this new era of AI abundance. Yeah, I mean, look, I think, you know, as, as demand for AI infrastructure has grown and demand for power has grown, you know, there's been a significant amount of investment in real-world, you know, power production capabilities and facilities that are becoming, you know, real significant bottlenecks towards, towards this buildout. Yeah. And so, you know, each incremental gigawatt is getting increasingly more challenging. Now, the positive side is that I think a lot of the investments and the foresight that we had at Crusoe are paying dividends today. And if we look at all of the progress that's been made in the campus in Abilene, Texas, our philosophy in designing and building that campus was, how do we build a gigawatt-scale computer? How do we build a single facility that can act as one giant coherent cluster that can train the next breakthrough foundational model that will lead to a step function improvement in the model capabilities? That was the North Star when we developed the campus and designed it. Today, we're seeing the fruits of those labor.

Yeah.

And you had to go through— there was like a 6-month period where it felt like there was a different article every single day about Star. If somebody sneezed, In Abilene, there was somebody that was going to write an article about how sneeze causes 1-second delay. But it feels like you guys have gotten through that, and I haven't seen a front-page article about some minute detail of the process in a while.

No, but what I was going to say is the benefits of that is the The first 2 buildings, which were the initial contract that we did with Oracle and OpenAI, have led to the training of Astra. I don't know if you've had a chance to play around with, but it's a really phenomenal model.

Yeah, it's super impressive. Yeah. And Jensen had some nice words about it too.

Yeah. And the next phases are coming online with buildings 3 to 8, which is going to basically quadruple the compute capacity, you know, 3 to 6 are already online and 7 and 8 will be coming online before year end. And, you know, having, having that full 8-building compute capacity to train, you know, the next version, you know, the next incremental improvement in the model. Yeah, I think we'll see step function improvement in capabilities.

You started the company in 2018, going on almost a decade in business. Can you give me a sense of the progress from Like the smallest project, the first project, how small was it in terms of megawatts and where we are now in terms of scale? Because I feel like that has been the biggest transition is that you started with pretty small projects and now you're working on the biggest projects in the world.

Sure. So, you know, the philosophy of the business, you know, from day one was taking this energy-first approach to computing. We named the company Crusoe after the character Robinson Crusoe, who was stranded on a desert island and had to be innovative with his resources in a stranded environment. And we try to channel that same sense of innovation applied towards stranded energy and unlocking stranded energy, unlocking value in stranded energy with computing.

Yeah.

And so, that led us to unique energy resources to power compute in early days. And we were actually working on generating our own power with flared— otherwise flared natural gas. We were capturing this waste methane, utilizing it to produce power, and then powering these mobile and modular data centers. Our first use case was Bitcoin mining.

Yeah.

But even at that founding of the company, we had this ambition to build an AI cloud platform, which today is where Crusoe is at, is the this vertically integrated energy-first AI platform. So, the early days, it was roughly 250 kilowatts was the first block that we did, which was actually pretty big.

Yeah.

First AI data center, we're actually putting up in like a Crusoe Spark Museum that we're building. And it was like, you can kind of think of it as like a single individual rack. I think it could power up to up to 25 kilowatts.

Okay.

So it was like higher density for that era. And we've since evolved our platform quite a bit to gigawatt-scale campuses at one end of the spectrum. And we've made an announcement or quite a few announcements around the Crusoe SPARC product, which we believe is the critical aspect to really scaling inference compute infrastructure as demand for token scales. You don't need a gigawatt-scale computer. You need lots of smaller clusters of compute to rapidly serve inference infrastructure and really dramatically reduce that, that time to token. And Spark gives us a ton of flexibility in terms of how we deploy inference infrastructure, where it goes. And then also just from a time to token, it really provides a lot of just-in-time solutions for delivering AI infrastructure for the innovators in the world.

It seems like solar deployments keep beating expectations. We've seen this chart of the predictions of where solar will go, and then the graph keeps going up and to the right. Where are you seeing exciting movement in new energy coming online that might be suitable for that type of product, the Crusoe Spark, you know, smaller amounts of energy required but unlocking new opportunities in different locations?

Well, you know, solar is definitely like a very compelling use case. You know, we, with our partners Redwood Materials, built the largest microgrid in the United States that was solar plus batteries, and the batteries were actually taken out of, mostly taken out of electric vehicles. So sort of a second life use of these lithium-ion batteries. And, you know, that AI, those AI factories are powered 99% plus just by the sun, which is a pretty cool use case.

That's cool.

You know, there's a lot of new energy technologies we're focused on, you know, helping catalyze and accelerate. You know, we had a big announcement with ALO to bring the first SMR-powered AI factory, which we're planning to go live with by mid-2027. But what's unique about Spark is given the small modular form factor, there's so many different— there's so many unique ways you can power it. You can power it with small gas generation, things like reciprocating engines and smaller gas turbines like aeroderivatives. You can also power it just in with grid power. And there's a lot of— while getting a gigawatt of power from the grid in a single location is quite challenging right now, there's a lot of places where smaller pockets of capacity are available. Think whatever, 10 to 100 megawatts that can be accessed way sooner because of transmission and distribution constraints that are happening in the grid. And And plugging into one node may actually help create quite a bit of balance in terms of the load balancing that the utility operators are needing to do. And so, it can actually help benefit the grid quite a bit because you're actually amortizing more megawatts over the same transmission and distribution infrastructure, which should actually bring down the cost for all local ratepayers.

Mm-hmm.

And we think that's an exciting thing. exciting proposition, because you're shifting all of the work offsite, and you're able to plug into a lot of these unique energy— available energy on the grid.

Yeah, that makes a lot of sense.

Talk about the new round.

Yeah, you raised some more money.

We did. So today, we announced the closing of the Series F at Crusoe.

Who came in? Saw some heavy hitters.

Oh yeah, um, yeah, it never gets old. Uh, the, uh, you know, the, the Series F round of funding was co-led by, uh, Atreides, uh, Valor Equity Partners, and Mubadala. Um, and we had, you know, a number of, uh, you know, a number of great supporting investors alongside it, uh, folks like TPG, QIA, Founders Fund, Radical Ventures. It was a great, great list. GIC was another great partner. So, there were a number of great investors, and we're really thrilled to have the support from a number of leading long-only growth investors and venture investors.

Yeah. That's amazing. Long onlys.

It's the only type of investor I like.

Yep. Thank you so much for coming on the show. Great to see you, Chase. Have a great rest of your day. Great to see you guys.

We'll talk to you soon. Talk soon. Goodbye. Let me tell you about Shopify. Shopify is the commerce platform that grows with your business, lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents. Up next, we have Justin Brown from Decagon. He's back. Welcome to the show. Welcome back. How are you doing? It's been too long.

Good, good.

Yeah, it's good to see you guys again.

Where are you in the world? You're expanding internationally. Are you in America? São Paulo, Brazil? I'm in São Paulo this week.

You are? So you're actually quite the flight over here. Okay. We're out here launching our LATAM office. So we're starting an office here in São Paulo.

OK.

Got a couple customers down here already, like Mercado Libre. So we really felt like things were moving pretty quickly, so got a team out here now.

What is the key— what seats do you want to fill in that office? I mean, obviously you need the general staff, but is it about sales reps, forward-deployed engineers? Like, what problems does it help to have boots on the ground effectively in the local community?

Yeah, it's mostly go-to-market. I think obviously there's language considerations, right? So anytime you go to a different market, it's both the product, right? So we deploy these voice agents, so the voice has to be really good at Brazilian Portuguese, but also our go-to-market teams have to be on the ground and being in people's offices and so on. So it's mostly just the go-to-market teams.

Yeah. What's happening with diffusion every time I have to be on a customer support line? And I don't have— and I'm not speaking to an effortless, you know, agent that has all the capabilities that I would expect that I know are available. I get a little bit angry at you and the other players in the space because it's like, why aren't you moving faster? It's like, I just know— I know the technology is here. Why do I need to sit on another phone tree? And, you know, like, that doesn't even do a good— It's difficult. Yeah, I just want to talk. And I just feel like being able to effectively prompt to the outcome that you want is like so clearly here. And, and still it's like so many of these things you're like, what's your, what's your address? What's your this? You know, all this stuff. It's like, enough.

We're trying, uh, trying our best. I mean, I, I can tell you like, uh, what, what is the long pole in the tent? It's, it roughly the, um, the analogy we like to give is like us delivering an agent is a little bit like building a self-driving car where, mm-hmm. The car is now smart, right? Like, the models are really good. Like, you guys have used the models.

Yeah.

The brains of the car are good, so it knows how to drive. But when you arrive in one of these big, let's say, airlines or banks or, you know, healthcare providers, it's the map of the surrounding area is not drawn out yet. And so drawing the map is usually the hard part. It's like you hit— someone has to make the jump between, you know, the, the car being good, but then actually like drawing the map and figuring out where to go. So a lot of the time that we spend in these deployments is really making sure the map can be drawn out well and then improved upon over time. And our job hopefully is to shrink that time over, you know, more and more deployments. And so that in the future, every single time you call, there will be a good AI agent there that can talk to you.

Please. How do you think about— how do you assess opportunities internationally? I would love just like lay the land of like the core business, why you thought it was the right time to expand internationally, and then I imagine that there's at least a spreadsheet or some prompt that you ran through, some analysis that you did to land on Brazil, but what is your actual process for evaluating a new market to expand into?

Yeah, I mean, international expansion is not like a novel thing, right? So tons of companies have gone through it, so we've been able to learn from, you know, what other people do, and usually most of the traction comes from, you know, a small handful of countries. When we first started the company, we're mostly serving like the upper mid-market-y, like tech-native companies. And so that's where we started. And then our first order of business was, hey, we actually do think that we can, you know, our approach to launching product will work in the big enterprise. And so that was, that was a lot of the big focus in the first year and a half. And so, you know, now we've announced, you know, a lot of the big banks, you know, Deutsche Telekom, American Airlines, Delta Airlines.

Sure.

So a lot of these big customers, And from there we're thinking, okay, what's the next frontier? Probably is international. Europe was the first sort of pillar for us where we launched the London office and we have customers in Germany and in the UK and so on. And then from there we were just thinking, okay, well, the 2 options for us honestly ended up being Australia and Brazil. And it's kind of a mix of how quickly are people adopting tech, what's the size of the market? Can we actually serve those customers? And we decided yes. And so now we have, you know, a small pod in Australia and now launching in Brazil. And I think for now that that is like a pretty good spread for us. But, you know, in the future, we'll probably look for, you know, what is the next frontier?

Is there an element where customer pull is really important? Like MercadoLibre has a bunch of employees in the United States. You probably meet them at conferences. The conversation can start before you even open the office. It sounds like you already have a contract in place, but you're just seeing pull from a certain big enterprise anchor client, and then that can justify going bigger.

Yeah, yeah, I think that is pretty correct, right? It's like, okay, let's just get a feel for the pull first before we deploy a bunch of resources. The same happened in Europe with DTE where, you know, We kind of originally started working with them when we just had a team here in the US, but then over time we realized, oh, wow, people are moving pretty quickly. We need to be on the ground. And so that's what prompted us building a team out there.

Yeah. What are you tracking? How closely do you track the progress of the models versus the cost and the Pareto curve of the models? Are token costs falling fast enough? Is it roughly perfectly on trend from what you've modeled? Is it faster? What can you tell us about cost optimization once you get a system that's superhuman or at high quality? I imagine that you're not just upgrading to the latest frontier model and just adding costs because some of the models when they come out, they're better, but they're more expensive.

Yeah, there's kind of 2 dimensions there. So, on the cost side, Typically what you do is you first get your product working, and then once it's working, you realize that most of the model calls you do don't need the entire model because the model does a ton of stuff, right? So you don't need it to compute math for you, for example. And so then the intuition is like, okay, well, over time I should release—

Personally, I like to throw a Millennium Prize problem at a customer service agent every once in a while just to see if there's a shot. That would be fun. But yes, I understand. Yeah.

And so, um, yeah, so over time you want to shrink the models and then you, then it's more of a question of like, okay, is that possible? Can you take a smaller model and fine-tune it and post-train it to perform at the same level?

Mm-hmm.

And so far I would say in our use case, the answer has been yes. Right. There's a lot of these small tasks that are done inside the brain of the agent, like choosing a topic or detecting for hallucinations or whatever. And so you can use small models and specifically open source models to do those. But when you're launching new products and kind of trying new capabilities, of course the frontier makes sense. And so those 2 go hand in hand.

Yeah.

And then for us, latency also matters a lot. So we talk a lot about how, you know, you have to not just optimize cost, like optimizing latency matters even more to us probably. And so, you know, when we've done these optimizations, we're mostly seeing if we can get the latency down because if you're talking to a voice agent, it has to be super snappy.

Yeah.

Those 2 go hand in hand as well.

And is that the model or the chipset or both? I imagine that we're not quite in the age of colocation or like edge computing, but it feels like that's coming as well as a small 100-millisecond savings potentially.

It's mostly the model. Most of the gains come from making the model smaller.

Yeah, okay.

Eventually, if the model's like so optimized, you could start looking at some of the other things, but—

Yeah, so make the model really small, then you bake it onto silicon, then you put that silicon Right in the city where the customer service is happening, something like that on the edge. And then you're at 2 milliseconds or something maybe. I don't know. Yeah. Very exciting though. Thank you so much for hopping on the show.

Great progress.

Great progress. Congratulations. Great to see you both. Thanks for having me. Talk to you soon.

Cheers, Jesse.

Enjoy Brazil. Let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more while Railway automatically takes care of scaling, monitoring, engineering and security. Up next, we have the founder and CEO of Beacon and the VP of AI research at Beacon, Nate Nylund and Leonard. Welcome to the show. How you guys doing?

What's going on?

Hey guys, good to see you again.

Welcome back to the show.

What's happening?

I would love for you to start with just a reintroduction on the strategy for the company, for the firm, and sort of how it's been going since we last talked. I think almost a year ago. Yeah.

Yeah, yeah, that's right. Yeah, so just a reminder, Beacon is a holding company. We acquire great technology companies that have been around for decades that already serve real economy businesses. Think campgrounds to labor unions to youth sports leagues. We take those businesses, pump them through our software factory, and then make them AI native. The whole goal is to make the real economy as AI native as quickly as possible. And the last time we talked, I spent time talking to you about our acquisition machine.

Yeah.

At that time, you know, we were acquiring a business every 3 weeks or so. Today, we're closing on a new business every 10 days.

Whoa.

And today, really, I'm excited to talk to you about our most important acquisition, I think, in Beacon's history, partnering with Leonard and the Hays Labs team.

Yeah, yeah. Leonard, give us a little bit of your background, what you built and why you're excited to join.

Totally, totally. I don't know about most important acquisition to date, but it's certainly an exciting acquisition for both sides. And hopefully we can bring a bunch of the Hays technology and talents to the rest of Beacon's existing portfolio. So a bit of background on myself. I'm Leonard, co-founder and CEO here at Hays Labs. Hays was an AI safety company founded out of my research over the last 5, 6 years where we focused on red teaming, guardrails, observability, and evals, both for the frontier labs, including OpenAI and Anthropic, who are our first customers actually, but also for large enterprises. Think big banks, Fortune 500s, insurance companies, healthcare companies, and so on. But what we realized—

Safety is solved. Jobs finished. We can, we can move on to— I'm going to go work on the economy and esports.

Well, safety is always an ongoing challenge.

Sure.

I think as capabilities advance, you always have to play catch up in some way, even though your normative declarations of what AI should and should not do can be set in stone. The way in which you operationalize the definition of what it means to be safe or not safe changes as the capabilities advance. And insofar as Joining Beacon goes, I think bringing AI to the real economy is a unequivocally good thing and responsible thing to do if you're an AI technologist. But also from the perspective of the alignment conversation and debate we've been having over the last couple of days, one thing we've thought a lot about at Hays is whose values should AI models actually be aligned to? And you can take the perspective if you're Anthropic or OpenAI that you have the best control and that you have the best wisdom around what values should be baked into the models. But it is our sense as Hays and Beacon that we should try and make contact with reality, in particular those that have been historically underserved by Silicon Valley technologists, like businesses in the real economy, in order to align our values to those businesses and the values of their customers. And that's part of what gets us super amped to join the Beacon team.

Very cool.

Yeah, just to give you guys— just to add to that a little bit on the safety conversation, everyone knows the real economy is behind on AI adoption. There's many reasons for that.

Yeah.

You know, the Ramp data from about a quarter ago showed that SMBs are only 20% adopting AI versus like 85% for the Fortune 500.

Yeah.

We started Beacon to address that gap. But really, Leonard and team are going to play a key role in that. The reason why adoption is so low in SMBs, some of it's a question of access. But really, as we, as we spend time with our 22,000 customers, it's a question of trust first and foremost. We— the real economy will not adopt AI if they think it's going to run rampant and screw up a bunch of their own relationships with their customers. So Leonard and his team, part of why we wanted to bring them in and why Leonard's leading this important function for us is so that we can ensure for our end customers that when they deploy AI through one of our pieces of software, it's done in the safest way possible.

Yeah.

Yeah.

Graham? Yeah, just to say that Deacon has the trust in the customer relationships already. I mean, these businesses, the real economy in the world depends on these businesses. And I like to say that Hays builds AI that anybody can depend on. And so now I've got both parts of the equation. You've got AI the business can depend on. And the real world is depending on— the business is depending on this AI. So I think it's a very natural fit.

Ground this for me. Where's AI been most useful in something like campground operations, since I know you guys have acquired some campgrounds? And it's so funny— it's still so funny to me that as you guys get bigger, there's definitely going to be some articles of like, this AI company is taking over For youth sports leagues and campgrounds.

Yeah, for sure.

I'm very excited about it. Yeah, I am. We were just on— we were just on with Jesse, and I was telling him I'm frustrated that I still will end up on a customer service call for some service.

Yeah.

And I'm on some legacy system that has no AI at all. It doesn't have any ability to reason or help me get things done. I can imagine a lot of use cases that, like, in the campgrounds context.

Yeah, in the future, you don't have to go to the campground.

You just It generates your simulator. Perfect simulation.

You're done. And then you're just sitting there.

No, that's a great, great point. Look, we are believers in AI abundance. The thing that's exciting, especially for the real economy, is we can help them solve problems that before they were not able to solve. So on the campground side, if you put yourself in the shoes of one of the 18,000 private campground owners across North America, You're spending 40% of your days doing payroll, HR, a bunch of compliance tasks, using a spreadsheet to figure out which camper needs to be assigned to which spot in the campground.

Sure.

Today, we have built tools within the software they already use to match guests to campground spots, to do dynamic pricing, to make sure that they're pricing appropriately for season, to load balance staffing for when campers are actually going to be there. So that is, you know, as to use a term Leonard just used, unequivocally good for the campground owner.

Mm-hmm.

But also we think that's how AI abundance actually translates into the real economy. It's how do we unlock more growth for the real business? Um, you know, one, one thing that people always ask me, okay, are you, you guys are just going to use AI to, to, uh, take costs out or reduce headcount across all of our businesses? We're actually up headcount. So we found the opposite. As we've delivered more AI, our customers—

Yeah. Yeah, there's been a bunch of reporting on this in The Economist and The Wall Street Journal about the AI actually driving increased hiring as growth comes in, which is fascinating.

How are you guys— do you guys imagine— it doesn't feel like now's the right moment, but 5 years from now, robotics being a big part of the strategy, you'll have all these existing businesses. And as different form factors get better, you'll be pretty primed to bring in these new form factors?

Yeah, I'll start on the customer part of it. Go for it.

Yeah, go for it, Neelam. Yeah, yeah, we actually have businesses today that serve various industrial end markets. So think about software that helps large construction sites manage their tool inventory, software that helps oil rig stay safe. And increasingly, we're seeing this marriage of AI-native software with hardware like GPS devices, et cetera, to be able to increase the surface area of what that piece of software can do for the end user. You know, one of our customers, which I can't name, which is like a Fortune 500 construction firm, they're saving millions of dollars today on tool supply Because, you know, we actually help them manage these industrial tools that are being used for the data center buildout across the economy. In the old world, before we applied AI to help them match tool to job site appropriately with GPS, they would— a construction manager or project manager at site would overallocate and overbuy tools basically to make sure that they're not going to be short. So this was a huge problem for, for the company. They worked with our team. Our forward-deployed engineers went down to Houston and helped them. We built a new module for them, which basically AI matches tools to a job site. Then we put in GPS trackers to make sure the right tool got to the right site.

That's very cool.

Leonard, sorry, you were going to say something too?

Yeah, I mean, I think you guys are being super prescient about where all the technology is headed. Certainly, robotics is going to mature quite a bit in the next 5 years. And I think robotics or not, I think there's going to be a huge category of value can deliver to our companies by essentially being able to dispatch AI employees or digital employees or whatever you want to call it at will. And so it could be something as simple as an AI SDR helping you find leads from your existing network. It could be something as simple as an AI support engineer helping you deal with tickets and bug bash. But I think there will be this huge supply of intelligence at your disposal if you're an owner-operator for you to go and lever up your existing existing strategy, team ambitions at your company.

Yeah.

I also do think separate from just digital employees, Beacon is in a very unique position to go after what I would describe as world models of companies or company world models, i.e., models that can help the existing owner-operator as a true thought partner and strategic decision maker think about what's going to happen next in their business and anticipate the best choice and action they should take to better run their business. So I think there's this pure question of, can you deploy more intelligence to lever up existing objectives? Then there's this separate technical challenge of, can you actually help business owners figure out what objective to go after in the first place?

Yeah. Well, thank you so much for coming on the show. Congratulations on the deal.

Yeah, great update.

And we'll talk to you soon.

Very cool, guys.

Have a good one. Goodbye.

Cheers. Thanks, guys.

Let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. We got Joe Weisenthal. here live in the TVPN UltraDome.

I can't believe I'm here. I can't believe I'm here.

Why not?

I know, 'cause you've been here so much.

Well, I've like been here spiritually so many times. And I, you know, in my dreams I wonder what the studio is like and will I ever make it to the temple? Will I ever see the gong?

Bring the gong.

Will I ever hit the gong?

Give us some excuse. What, 10 years?

I was warned by one of your producers.

It is easy.

25 basis point rate cut. Yes, yes, that's what we're hitting it for. Smack it. Boom. Oh, you gotta go way harder. You gotta really— like it's a baseball bat. Like you're swinging for the fence.

There we go.

I'm done now, right? I can leave. The whole point was just to bring me in to, uh, to hit the gong. What else do I have yet to achieve now in my life? Yeah, what's some What someone's yet to climb.

Yeah, what actually brought you to LA?

We— oh, funny you should ask. We are doing a live show tonight at the Vermont Theatre in Hollywood, where we are going to attempt to do a survey of what is this thing called the LA economy. So we're going to be talking to some of the industrial guys. Chris Power from Hadrian is going to be on. Tom Mueller, the first space— Impulse Space is going to be on. We have some of our favorite internet characters, Taylor Lorenz, Rachel Carton is gonna be there. So it should be a really fun night. I have no idea what to expect. I don't know anything about this town.

I was gonna say, dude, you've tried to survey the LA economy, but you've picked out like some of the bright, like the only bright points, to be honest. No, I know, it's like—

You're negging your own city like that?

Oh, I mean, the city's a disaster.

Can't even hold on to a movie studio.

But then you get this. But then someone gets to take this and make some movies.

Yeah, the bright spot is like the individual creators are doing well. There's cool stuff in industrials and space. But the whole Long LA movement fizzled out right before you got this sort of—

I'm super Long LA.

Really?

Absolutely. He just gets to the weather.

And what's there to be optimistic about?

Look, the thing that will never cease to mystify me, right? So Americans over time, they migrated westward, etc. And no offense to them, I'm sure they're lovely people. I'm like, wait, you guys stopped in Nebraska? And you could have like gone on your wagon for like another, you know, 72 hours. It was right there. Yeah. Or— and it's like, what? And I like— what— in my, like, in my trying to backfill American history, I was like, how did we all not just come to LA right away? And I do not believe, regardless of what happens with Hollywood, yeah, that there will ever be a time, yeah, where people just do not want to live in this climate by the beach. And they'll find an industry for them.

Yeah, it is crazy that California— I mean, we got 3 coastal cities— San Francisco, LA, San Diego. But it's always shocked me that if you drive—

I love how you're just completely ignoring every other—

No, no. If you drive halfway between LA and San Francisco, there's not a metropolis with an international airport there. Like, you go below Marin and Laguna Seca. That's the only place I know. Monterey. Yeah, you go below Monterey, and then And then you get into beautiful areas, but it's mostly camping. And there was never like an enforced metropolis that built up.

No, it's kind of strange. You know, a couple years ago, a good friend and I decided to— we had just like never driven the Pacific Coast Highway. So we're like, let's drive from San Francisco down to— what's the place down there where they have the town where they have Coachella? Palm Springs?

Yeah, yeah. Anyway, and we did.

And it was beautiful.

Wait, you hit Coachella?

Or you just went to the town? No, no, no. We just went to the town. Beautiful drive. Yeah, but at some point, I have to admit, it's very beautiful. I have to admit, it's like, oh, there is another beautiful cove. Yeah. And so I was like, I'm getting a little bored of this, you know. Oh, there's another beautiful cove.

It also takes so much longer, the drive itself.

So then we had the thought, it was somewhere right around where the Hearst Mansion is, we're like, why don't we drive inward? Because no one sees like— yeah, yeah. So then we like went through like Fresno. Yep, a lot of farmlands. miles and miles of almond farms. I loved it. So now I'm actually like, yes, there should be another—

Cow ranches.

Yeah, so it's like, of course, yes, there should be another megalopolis somewhere like on the coast. But also I think that like if someone visiting LA, at one point you should do the Inward Drive and see where they make all the almonds, 'cause it's amazing country out there.

That's how I used to get back and forth from LA to San Francisco.

LA is in a bad spot. LA's in a bad, bad spot. I know. I'm generally— the point of optimism is the weather. That's been the durable point. But the weather can get a town through some bad eras, right? Because if everything is actually headed the wrong direction, but the weather's simply so good that when you look outside—

It'll be fine. It'll get you through. I'm very bullish. I'm bullish. I'm like, it's— So move here. Yeah, I mean, like, look, like when Detroit got hollowed out after like the shrinking of the auto industry there, et cetera, it's like, what was keeping people? You know, they didn't have the weather.

Oh, true.

You know? And so like, you guys have the weather, it'll be fine. Wait a couple years, it'll be fine. It'll be fine. Detroit's been good weather.

I'm waiting for the, I'm hoping for whatever industry can bring great opportunity to LA again, which it just doesn't feel like. Like, I moved here after college, and I thought it was great because it was much more affordable than New York. I'd have friends visit from New York, and they'd be like, you get this crazy loft for what would be over $1,000.

See, in New York, I believe there's this perpetual conversation that has, like, Should I move to California? These days it's more SF, but for many years, the discourse among people in New York was like, should I move to LA? You know, like, it seems pretty nice out there. I get the impression, and maybe you guys could tell me if I'm wrong, like, I feel like once you get here, you don't have the reverse conversation. Like, oh, should I move to New York? I feel like it's like, oh, the East Coast—

I was kind of— I've felt that way. You felt that way? But spiritually, I've always felt at home in New York, even though I've never lived there full-time.

Well, that's the lovely part of New York. You are spiritually at home in New York.

You don't think converting the Paramount lot into a data center will keep people here? That seems like a lot of opportunity.

Yeah, I mean, it's worth a shot.

I mean, look, the question— if there is a new industry in America that creates jobs, then some of them will be in LA. I think the question is, Will there ever be another industry in America that creates jobs in large scale?

Yeah.

If there is one, LA will be fine. If there isn't one, then it's not just gonna be an LA problem.

A new job that can employ older people, driving elderly care, healthcare hiring.

Yeah, right, but it's not just gonna be an LA problem. Like, if that's the issue, if it's like we can't think of the industry that's gonna, then like, I guarantee it's not just an LA problem.

Yeah, yeah. What do you think of the latest in the AI data centre discourse debate? It felt like we were grinding towards a crescendo there, and then we just took a hard left turn into AI risk, x-risk.

I think so too. I think that actually—

It's almost like it was coordinated by powerful shadowy organisations.

That's Ian Bremmer's take. Ian Bremmer's take is actually that they the x-risk debate is to distract everyone from the real problems. From the data center problems. Well, not just data centers, like brain rot and slop and all the other complaints of people, job displacement. And he's like, the AI lab leaders, in his view, I don't know if I fully agree with this, but in his view, he says, there's nothing an AI lab leader loves more than talking about x-risk. It's pure upside.

Well, I would say 2 things. So one is it already feels like, if you're sort of a politician and you're still talking about data centres, it's like, bro, that was a month ago. Like, we've like moved on to model risk and stuff like that. Yeah, yeah, yeah. So it does feel like actually very quickly the data centre conversation disappeared among political discourse. My cynical version of the Ian Bremmer take, however, is slightly different, but maybe it converges on the same thing, which is that actually, You know, because before data centres and before the x-risk debate, it was the job loss debate, right? And so this question is like, oh, will AI wipe out 50 to 100% or even 10% of white-collar jobs? And the, you know, if you read the op-eds from the, you know, the Demis Hassabis of the world, this is what they were talking about.

Yeah.

My cynical take at the time of those was, they're very concerned about x-risk, What is something that we could talk about that would distract from the—

Oh, interesting.

No, no, no, not distract from it. What is a way that we can sound the alarm?

Soften it.

Or what is the way that we can sound the alarm about AI without sounding like a complete bunch of loons?

Totally.

Well, let's talk about job loss. Because that feels like—

Because certainly we can't get an AI doomer on Joe Rogan. They'll never go for that.

Right. But like—

So let's talk about jobs. Because we can have that conversation.

And also, job loss is the type of problem that, like, the 3 of us could come up with theoretical answers.

Totally.

Oh, we say, like, maybe 10% of OpenAI gets into a, you know, a national fund.

Exactly. Exactly.

Or we use—

There's practical solutions.

Economists willing to engage.

Well, it's also— it's also another way to think about it. You could tell somebody, like, an asteroid is going to hit the Earth in 2050, and people would still go to work tomorrow.

That's totally true. But if you were like really like, if you're like sitting there at AI Lab and you have like these really deep intuitions and you've been reading like Bostrom for 20 years, what are you concerned about? You're concerned about the X-risk. How do you get people to just, first of all, take the idea that AI risk is a big deal?

Or AI is real.

Or AI is real. Let's start with the job loss conversation. It feels tractable. It feels theoretically solvable. It's like, oh, we need UBI. And so my cynical take was that the only reason, or the primary reason they talked so much about the job loss stuff, is because it was a way in to warn people about AI that didn't make them sound like sci-fi freaks.

But that feels like that backfired because we're sitting here and there hasn't been massive AI job loss.

Oh, no.

And so—

Yeah, that's true.

It's easy for people to be like, these are the people that were telling us that there was gonna be 25% unemployment, and that hasn't happened. So they're saying 10% chance. Like, I'm betting against the house.

Yeah, there's this weird thing going on. I certainly think this is true, which is, I don't know, do you see these people as like top-ranked AI forecaster, right? Or like capabilities forecaster. And many of them do have like pretty good track records, except on the question of what will be the impact of this technology.

Exactly. They're all internal. They're all internal. It's like, yes, I— I was able to predict model progress by the amount of compute that was built. And also, wow, he also predicted the number of chips that were going to be bought. He also predicted the number of data centers that were going to be built. And it's like, those are all downstream of the exact same line. You're just extrapolating things that are directly correlated.

Yeah, except the one thing that no one has had any good intuitions on is like, when are we just going to feel this in our day-to-day life? And it's pretty weird now. That the conversation has evolved to, like, the x-risk conversation, because it's like you step outside, it's like life feels really normal.

Totally normal.

You know, I was, like, talking last night, we were, like, talking to, like, say, you know, we were at this conference of, like, financial advisors, etc. And it's like there's this surreal thing, because I, like, when I stare at my phone, it's, you know—

Well, a lot of those financial advisors, they talk about AGI all the time, just Adjusted gross income.

Adjusted gross income. That's what they're focused on. But, you know, it's like, well, AI can help you. Like, this AI tool helps your clients with tax loss harvesting, et cetera, which is great. I love tax loss harvesting. But it's like, wow, this could not be a more different— Tax loss harvesting. Tax loss harvesting. This conversation could not feel more different from like the types of— conversations about AI that, like, I read on my, like, little screen.

Or the disconnect, honestly, between, like, the discourse and then, like, Dreamforce is happening.

Oh, yeah.

It's like, yeah, like, there's those big problems that we need to be talking about. Like, we got to be talking about this. But in the meantime—

Salespeople are headless.

Yeah. No, I really, like—

Here's another thing.

Here's another thing.

So I was thinking, growing up in the Bay Area, I distinctly I don't know if you remember the Iraq War protests.

Yes.

Now, I just looked it up. There was 200,000 people roughly in the streets, like marching. Yeah. A million men marched. There hasn't even been a 10,000-person AI protest.

You want to hear a funny story?

Yeah.

So I was at University of Texas Austin. I was a young hippie, and I attended some anti-Iraq War marches. And there was this guy at the front. There was this guy at the front.

No, I was in the crowd too.

There was this guy at the front of the crowd who would regularly show up, and he had this bullhorn, and he would captivate the crowd, and he would look— he would point up at a nearby office building, and he would say, don't think we can't see you, um, you know, G-men up there spying us and everything. And then everyone would like crane their head. It was like, oh yeah, I think I see someone in the office.

Interesting.

That was Alex Jones. That was— that person was like the guy. And so like The lead. And so the least surprising— how deep does he go? Yeah, why is this all about Alex Jones? Like, this guy who just like had us all captivated during the protest, who would be like— you were one shot. Everyone was like, oh yeah, I think I do see someone up there. Yeah, that was Alex Jones.

Wow, he just had a natural talent for it.

A gifted— that's remarkable. Yeah.

Wow. Oh wow, that's a— that's crazy. I mean, yeah, this— that's like when we— yeah, there's some John Kiriakou, you know this guy? No, he shows up in weird places. I think he was like a schoolteacher. That was a wild story. These like, you know, closing sliding doors moments. Yeah, yeah, I know one of them.

No, it's like, wow, he was just a local guy that we all knew back then when we were in college.

Okay, back to the— back to data centers, back to Express. So, so conversations getting mocked. Yeah, there's one scenario where X-risk is the thing that does move the needle on the protests. You go from 10,000 to 100,000, legislation gets written, and it builds on a lot of the work that's been worked on, on the data center stuff.

Sure.

And if you read the AI safety crowd, a lot of them are like, yeah, we actually would love a slowdown in chip production, slowdown of data centers. That would actually be pretty satisfactory because one of the prerequisites for nefarious ASI is just a lot of compute. A lot of compute. worried about the God model emerging on a desktop computer. We just don't think it's gonna happen.

That's right.

So AOC was recently on camera talking about, I think you called her response like word salad, but we have the technology to turn word salad into ironclad legislation.

Oh, into a fancy new swazzle.

Yes, so take us through, like, is the word salad of the AI discourse actually a problem?

Wow, you know what it is? That's a really good question. I asked a version of this question to someone several months ago who was like in the real like safetyist camp before it became cool. And I was like, why don't you guys just join up with the anti-data center people, right? That's a good question. I was just like, you know, I understand. It's like they don't have the same like sort of understanding, like they don't, the shape of the risk they're— why don't you just join up with them? And I do think that, like, with, you know, people who have sharpened their blade for 10 years in the LessWrong forums, they cannot do that thing of like, yes, but they're wrong. You know, it's like, oh yes, yes, aligning with the anti-data center people would align with me perfect, but they're wrong about the water consumption. Yeah, yeah, yeah. And I cannot bring myself to align with them, not because it wouldn't be useful, but because I can't bring myself To be on the same side as like someone who slightly is factually in disagreement with me.

It does feel like there's the beginning of that horseshoe coming together with Steve Bannon and Bernie Sanders. Steve Bannon for years has just been like anti-tech, anti-Facebook, anti-control and like big company. And so there's more merging there. But I agree, it's even harder for the AI safety camp. But it feels like the Nate Suarez, on Tucker Carlson and Daniel on Joe Rogan are like, there might have been an update to the strategy.

I think there has been. And I think anyone who expects AI politics to map cleanly to some like left-right thing is like delusional. So like, I get, all right, Bernie became the first like major elected politician to like join the pause cause, et cetera. And so then people are like, oh, anti-AI is gonna be a left thing. But you're absolutely right. There is not going to be a left— first of all, there's not going to be a left position to AI because there's a million different angles. But also anyone who thinks that it's gonna split between like some Democrat-Republican thing, you know, we had Kathy Hochul on the podcast and she was like defending the data center one-year moratorium in New York.

Yeah.

And I think at that point, like literally, It was so quick because at that point then people were like, well, look, okay, New York is gonna miss out on all of these companies that are, you know, bringing in tax revenue, et cetera. So this is why Texas is gonna get it. And then like a week later, Governor Abbott, he didn't, it wasn't a moratorium, but then he announced like a higher bar that the data centers are gonna have to clear in order to get built there. So even whether we're talking about models regulation, whether we're talking about building of data centers, this idea that it's gonna map cleanly onto some Democrat-Republican thing, I think it's like delusional.

How do you think people break on should this be a federal or states issue? Because there's a world where, you know, if a particular state is like, yeah, we actually love this. There's— I'm struggling to see a negative externality of a data center in a particular state if everyone there is happy. maybe it should be state-by-state regulation. At the same time, it's such a national discourse issue that having a clarifying bill for like, yes, we are only building clean data centers, or there's some emission standard, maybe that makes sense, but how have you seen it puzzled around?

No, look, I think most people, regardless of where you like fall on all of these questions, probably think that like some patchwork of state laws is like not a great way to go about doing any of this. But, you know, outside data centers, I thought it was really interesting. I guess it was like, um, just last— time is flying, you know, Chris Lahan from OpenAI. And in that thing he spelled out, it's like, we support this Massachusetts bill now, we support this California bill now, etc. That's a pivot.

Yeah, totally.

That's a pivot. And so like, I'm sure—

Not the worst pivot. Everybody loves a pivot.

But I thought, you know, it was really interesting. Because in as recently as last year, another fellow we had on the podcast, like Alex Boris, was like— was enemy number one. Yeah, enemy number one. And his bill that he had pushed in the State Assembly was like pretty straightforward. If you're above a certain company and you have some, quote, incident, you have to report it and pay a fine, et cetera. That kind of seems like a small potatoes regulation compared to what we could get at this point. I doubt, again, like, I don't think in the AI industry or outside the industry anyone particularly loves the idea of like this, you know, the state-by-state rules. But it does feel like to a lot of people, well, this is the least worst possible scenario.

The best part about the state-by-state rules is that you have— it sets up the AGI dunk, which is like, oh, if your models are so powerful, it shouldn't be hard to comply with all the rules state by state. Like, yeah, it's a mess. And you have to serve 50 different versions of ChatGPT. Like, go for it, just have your model do it.

Well, this is, you know, the other paradox. Oh, you think your models are gonna like go kill people or whatever? Just tell the AGI not to do that. If it's smart enough, have the AGI trained, right? Like there's like all these—

Yeah, yeah, people apply to the distillation thing too.

I mean, this is the—

If they're smart, why can't you stop them?

Yeah, you're so smart, right? But this is like, this industry is A, not only made for thought experiments, the thought experiments have been around probably since the 1950s in the version of sci-fi novels and Asimov, etc. And so the whole thing is built on thought experiments and so forth. There were thought experiments before there was something we would call an AI industry. And you just literally take them in any direction. And there's like—

On the opposite end of thought experiment is like hard reality rate hikes. Yes. Do you think the AI community is digesting what's going on with the Fed funds rate, what the outlook is on interest rates? No, no. And I also like—

I would be, you know, and I would be very—

no, no, it's just it is a very different world because like, because like, like tech companies that have never raised debt, have never raised money at all, are doing it. And like, they will be affected by this and probably more so in the future.

So, yeah, what's happening will not only be affected, but I think by and large, if you sort of like assume the growth rates that these companies are seeing, if you sort of like do If you do the math on like what, okay, financing, building a data center, acquiring the chips, I don't think like 25 basis points, 50 basis points, they just don't move the dial yet.

Yeah, I think Sarah Friar said like, yeah, she bought all this compute and it like trades in the market like—

I mean, there's a long— so it's funny. Yeah, and there's, you know, there's actually a funny amount of academic literature on whether any company period cares about interest rates. So this is actually an empirical thing that's been tested. Which is like, has there ever been an example of a company that actually, like, changed some policy choice because of something the Fed did? And it's like surprisingly rare. These rare— there are so many bigger questions that usually, okay, here's a board meeting.

Yeah.

And we're deciding whether we're going to make some investment. And it's very rare that the decision— oh, is the 10-year at 6% right now or is it 6.5%? 5% right now? Is the short end at 4.7% or is it 5%? There are very few corporate decisions in the history of business where that turned out to be the deciding factor. Now, if we're talking maybe strictly real estate, where there's a very like, OK, these are going to be our cap rates, etc., sure, that may be very underwriteable.

Now it's not.

But by and large, it is a huge open question about whether these Marginal moves, anything have any history of affecting corporate behavior or not?

Can you reflect on Warsh and how you— like, how is he different? Not even in the strategy. We know the rates are going up, but like aesthetically, the way he communicates. Yeah. The dynamic between him and other partners. Like, how does he feel? He's just at the start, but like, what is his legacy building towards here? What's the purpose?

You know, I think there is this view. There is certainly a view that Warsh represents some sort of like very 90-degree turn from like the 3 predecessors he had, or 180, I don't know. Powell, Yellen, and Bernanke.

More academic.

In the specific sense, they were more— well, Powell was a lawyer, but still more of like— but still like an internalized academic macro for sure. But the main thing that they had was like they communicated a lot. And they brought, starting under Bernanke, there were a number of, I guess I would call them monetary policy innovations. The press conference, there's not a history of that before Bernanke. The so-called dot plots where the members of the FOMC write down where they expect growth on a—

Bernanke got the press conference idea from basketball?

Probably from basketball. It was built upon. But I think there is like, look, like, Most of the history of central banking probably looks much more like Warsh. Like, they didn't used to talk so much. And so, like, I am of this view that basically, like, you know, we could go— if you go back to the Greenspan era, he was— the word, the famous word to describe his style of communication was gnomic, right? And he didn't say many words, or if he did say a lot of words, no one understood what he was talking about. There is a long tradition of essentially like opacity in one way or another at the Fed.

Sure.

And so like, if like I, to some extent, like, you know, people's like, oh, it's so— we're getting like, oh, these— the dots do not have some long intellectual tradition at the Fed. If they completely got rid of the dots, if they completely got rid of regular press conferences, which my gut is that Warsh would like to do, because he doesn't really say anything at most of his press conferences, we would not be ripping up some hundreds of years of like—

Precedents.

No, we're ripping up precedents since 2008.

Sure.

Or 2009. And some of this stuff made a lot of sense coming out of the great financial crisis.

Because it was a Fed crisis.

Yes.

Like, it was a crisis for them, and they were the instrument.

They had a very specific challenge. Which was that inflation was very low. They had cut rates to zero. If you had plugged the economic conditions of the Taylor Rule into some sort of like rule, it would have said like you have to cut rates to -5. Yeah, they can't do that. I think a lot of things break when you have negative interest rates. So then they're like, well, how can we be further dovish once we hit the zero lower bound? It's like, okay, we're going to do this by talking more about our commitment to lower lowering rates and so forth. So a lot of these things made a certain intuitive sense under a set of macro conditions in which the traditional tools of monetary policy were not workable because you just couldn't cut rates any further.

Yeah.

It's 2026 now, a very different set of problems, etc. So if you would say, look, under these conditions we do not have the same problems that we had when these things made a lot of sense, so we should rid of them. I think there are a lot of people who would agree. What the one thing I will say that I like, the head-scratcher still, is that like on some level, you know, he used in his first or second press conference, Warsh used this phrase, we would like to play, we would like to have market participants play the ball and not the referee. The referee being the Fed. And that sounds great. We love that idea. Play the ball, not the referee. Unfortunately, and where I would say, like, in the sort of government bond market, the Fed is the ball, right? We're all trying to guess where this is, right? So it's like, let's talk for a second. Let's imagine a 2-year government bond, right? That 2-year government bond is going to— the correct pricing of that bond is going to be like where the Fed sets rates over the next 2 years, right? It is.

Okay.

Is there some sort of steelman here where, like, the— you should be more focused on the fiscal condition of the U.S. government?

We would love for that to be the case, but at the— we would love for that to be— sure, yes. But the thing is, the Fed is going to be making decisions. There's 12 members of the FOMC, 12 voting members of the FOMC, and they are going to be attempting to meet their dual mandate of maximum unemployment and stable prices. And things that they will take into account. Well, what does government spending look like right now? What is trade and tariffs? What does that mean for the inflation picture, et cetera? What does AI mean for the state of the employment position, et cetera? But at the end of the day, then it's like you are in— once you are a participant in that market, what you are betting on is strictly and only how those 12 members of the FOMC are going to vote. be digesting that information. You know, when, when we occasionally have arguments or debates about prediction markets, I like to point out that the US government bond market is not figuratively or metaphorically a prediction market. It is literally a prediction market.

Yeah.

In the specific sense that you are making bets on how 12 voters— just as in an election, it's an election of And there are only 12 voters of what they will do at the next meeting or the subsequent meetings. So, it is literally a prediction market of how you expect 12 people to vote 8 times a year. And that's not— that is all I believe that a government, at least in the US, that a government bond market could ever be.

Fed currently needs to— the dual mandate is minimize unemployment. Yes.

Which does not have a—

minimize inflation or hit the inflation target.

So the goal is stable prices, stable prices. And that has been interpreted and formalized as we have a 2% inflation target.

Post-AGI, post-abundance, we'll want the Fed to maximize unemployment and minimize prices. They want to maximize deflation. So everything's essentially free. If we're really AGI-pilled here, that's where we need the Fed to be going, right?

Yeah, I don't know what operation—

Post-singularity factors.

How do we maximize, right? Because let's just say we have— let's say we have an AGI that could do all human labor, right? I don't know what that means for Federal Reserve tasked with maximizing labor.

It's a roadblock. What economic data do you feel like you can really lean on? I feel like there's so I have so little trust in all the employment data now because it always gets revised. Sometimes, you know, the revisions have been insane. And then at least like it felt like 6 months ago, like there was so much admin like focus on some of this data. And then there was some personnel changes. And so now I'm like, yeah, the boss didn't really like the data.

Well, I'll say 2 things. Which is that I actually believe that even to this day, setting aside some of the personnel changes, you should do this sometime. You should like call up the BLS. They're they will answer any question you have. There are all these like very serious people who take their jobs very seriously. If you wanted to call up the BLS right now, this sounds amazing.

You should do it.

I mean, it would be a good segment. Okay, great. Yeah. But if you called them up and you're like, you know, I want to. I don't understand how you— my co-host Tracy wrote a piece on this. But if you're like, if you call them up and say, you know, how did you arrive at the price of mayonnaise for this index? They will answer. They will spend an hour on the phone with you. They're great.

That's amazing.

They're really good. They're public servants in the classical thing. They will sit an hour with you and tell you how this was constructed. So they're really good. And then the other thing I will say is, like, look, especially in the hedge fund world, there has been a very big obsession with like alt data, private sector data. It looks exactly the same. Exactly. Like there's been many attempts to construct private price indices. There's ADP. You can— Indeed.com has a job. It doesn't all track the same. So there is this hope or this sort of belief that it's like, oh, the private sector would be better at collecting this data, maybe. And look, it's all out there. And then there's the MasterCard spending data, et cetera. It ends up rarely deviating in any substantial sense.

So what you're saying is this goes a lot deeper than—

It goes a lot deeper. They're all in on it.

Job data truther goes to heaven, asks God, did we really add— did the US really add 160,000 jobs in August? God says, you really did. And he said, wow, This goes deep. Even you, even you. God's in on it.

Wow. Exactly that. Yeah. Crazy.

Yeah. One thing that has stuck out to me is like, this has been the year of sort of breathless takes from AI leaders and VCs around like, we're in the post-AGI era. AGI is here. We're in the singularity. We're in the foothills of the singularity. Everyone has a different buzzword, except I'm just like, I'm going to be looking at this, all the economic charts. and like, it doesn't feel like there's a kink in the graph yet.

No, it does not.

It doesn't feel like it's showing up in GDP or anything else.

But, you know, I think an interesting question is, and I like, you know, where does the productivity data show up for the internet?

Yeah, that's right, it doesn't.

You like, it's just like, oh, in the late '90s, productivity moved from 2.5% to 2.8%. Yeah, yeah, yeah. Amazing, wow. And so it's like, or it's like, where did the productivity show up release of the mobile phone.

Yeah.

So I do think like we could, we should, and we have, there's many interesting conversations that will be had about how AI is like manifesting in the data. But I do think it's worth asking like, why didn't all of these other technologies like show up that should have, or intuitively felt like they would revolutionize business in some way, make companies hyperproductive? Why didn't that show up in the data either? So until that day when the kink arrives, we have plenty from the past we still don't— we can still talk about and try to understand.

Yeah.

That's sort of my take on that.

Yeah. What else have you been tracking recently outside of— is AI just completely all-encompassing?

It's so annoying because it's like, I'm very interested in, like, you know, I'm very interested in oil. I've been doing some oil stuff. You know, I'm very interested in The Fed itself. I'm very interested in all these things, but it's just like I feel like kind of against my will. Yeah, suck that right? Like that it's just like every year. Do you ever feel that way?

I he's like really another another main story on this?

No, I got like there's someone in my DMs who was like oh yeah we got to do an episode about motor oil. We do an episode. about motor oil. There's— I got to do an episode about—

I like where this is going.

I got to do an episode about diesel prices.

Oh yeah.

Diesel prices are through the roof. All of these things to prioritize. But I do feel to some extent against my will, like there's this like monotonical step up in the number of like waking hours that like you just can't talk about these things without AI. It's funny, I was in earlier in the year, I guess it was early this summer when oil was a lot higher, I was out in Hong Kong. And Asia, or East Asia in particular, is seen as like ground zero for where, you know, the jet fuel crunch is really gonna hit and stuff like that, because just the route, et cetera. So I'm like sitting around like all these like business execs, and I'm like asking them, what's going on with oil and airlines? Like, whatever, let's talk about AI. I was like, oh my God. This is where I thought you guys were going to be a little more—

You know what we need? We need a new thought experiment for the oil community.

That's right.

We need to build a new thought experiment for all these niche stories. And then it can be something that we can tussle with.

We can tussle with, have like a big debate.

We need a percentage, percentages around these thought experiments.

That's right.

We need to build a lore in the world so then you can have endless debates for hours and hours and hours.

It's really— I want to talk to them about asset allocation in the—

You know the question that I still have not gotten to the bottom of? You mentioned the Iraq War protests.

Yes.

And I remember that there was, at the time, the banner was no blood for oil.

That's right.

Because the stated reason why America went to war in the Middle East was WMDs.

Yes.

And that was pretty handily concluded as like false, right?

I think so, yeah.

right? But the cynical view was like, we only went there for oil, and— but we didn't get the oil?

Or was that the story?

Not as far as I know. So my question is like, when George Bush and Cheney sit down years later and have a glass of wine together and they reflect, are they like, damn, that one got away from us, right?

I mean, Trump certainly thinks that. What were they doing? We didn't even get the oil.

Yeah, yeah.

But you know, speaking of like, you know, the— what is happening on the ground is we did get the oil out of Venezuela, and like Venezuelan oil is like basically our oil now.

But that's 1% of the market?

Uh, I don't know exactly what it is, but it's a lot of oil.

Okay.

But where are the like—

Where are the savings at the pump?

But there's, there's no savings at the pump, but there's also like the disappearance of like that kind of protest movement, because I think like Which is where I kind of thought you were going to go. But it's like there was a point in our lives where like the thought of like using US imperial might to extract oil was something that animated a non-trivial number of people.

I also, I think, I think now we're really doing it.

Yeah. Where is the, where is the, I mean, there is one. I don't want to actually completely sneer at them, but like, you know, where is the, there has not been a particularly large-scale protest movement about the embargo.

No, no, it's, it's social media, like very clearly killed protests because it allows like people can like go viral and feel like they're like, you know, taking action. And yet like it's like sounds like the dream of a government that's not doing the will of the people is like you can do whatever you want and sometimes they're going to get angry online at you, but nothing will actually happen. But that's very new.

I mean, just a few years ago, there were millions of people for the Women's March, Black Lives Matter movements. Like, it was, it was like post-COVID, then things maybe slowed down.

Yeah.

Like, I completely agree with this idea that like people feel like they can scratch the itch of protest online by throwing a like on a photo.

You know what really, you know what really like died is the bumper sticker.

I was just talking, I was just thinking about that.

I was like, every once in a while I still like, usually it's like a Vermont license plate, play it. And they still have like all of those, like, you know, the Coexist bumper sticker, this car climbed Mount Washington bumper sticker, my other car is a broomstick bumper sticker, etc. I miss all that. Every once in a while I see these.

It's a boom in the I bought this Tesla before—

I bought it. That's right. There is that one.

But I miss the bumper sticker. But I do too. And I was thinking about how I haven't seen any anti-AI bumper stickers, but That feels like a ripe area for entrepreneurship. If somebody's going to start a bumper sticker enterprise, the anti-data center bumper sticker, probably going to fly off the shelves.

And my friend Blake makes amazing bumper stickers. He's the world's foremost vintage G-Wagon dealer. So he makes bumper stickers that say, you wouldn't put a bumper sticker on a Mercedes-Benz.

Oh, OK. You know what I think there is in— One of the reasons why AI politics are really confusing and scrambling is that, like, to even go about— to even say, like, put up an anti-AI bumper sticker on your market would force the person to accept the premise that AI is a serious and important technology.

Totally.

And so part of the reason that I think that it scrambles the thing is, like, if you want me to be actively anti it, then I first have to accept the premise that this is—

I'm legitimising.

I'm legitimising.

Yep, yep, yep.

And that is like something that many people are reluctant to do for various reasons.

Yeah, of course.

And so there is this question, Nate Silver asked this, like, why don't you see a more like sort of like left-wing opposition to like AI and the pace of development?

Sure, sure.

There are many, I have many, I think there are many answers to this. And I don't even know, like, there will never be a, you know, more complicated. But I think it first requires a certain step of legitimisation that people don't want to do, which would be required to actually, like, care enough about AI risk to put a bumper sticker on it.

Yeah, yeah, yeah. But with the power of AI, we'll be able to take a picture of that bumper sticker, immediately translate it into ironclad legislation. That's the goal.

We could do it, spell this out, all the herebys that you need to have at the thing. God, you know, it's funny, like, um, that's such a classic political talking point. It's like, they never even read the legislation that they voted on. Now, this is like everything now. It's like, they never even read the code that they implemented.

Oh yeah.

Like, there's— this will be the phenomenon of our world, that in so many different things that people implement, no one ever actually, like, read it or knows, like, what the guts of it are.

Yeah.

And it's just beginning.

Yeah. Yeah. The state of the book industry is interesting.

Yeah.

Did you read that Tim Ferriss article?

No.

He talks about how he wrote very specific nonfiction books. One of them is called The 4-Hour Chef, and it's essentially a recipe book. And he was like, I actually got steamrolled by ChatGPT because people can look up exactly what that is.

Yeah.

And he was trying to sort of like, match through some trends? Like, obviously, I think the Kindle bookstore has seen 3x the number of submissions, and there's a clear AI effect there.

Yeah, of course. Yeah.

But then I imagine that sales of, you know, Harry Potters are still chugging along.

Probably still doing really well.

It's not really substituted with that, but—

No, I think that— but, you know, I do think probably in sort of like, you know, the type of people who like wrote the books about like, people who want to read about erotic affairs with monsters and stuff, you know, which is like—

That's a big category.

I have heard that this is a genre.

It's called romantasy.

Romantasy.

Yes, it's an investing term.

Will AI like be able to like write? I would suspect that it won't be too long until like some of these genres. But you know, like, look, really like big, like this sort of world-building type of fiction that would be in a Harry Potter, I don't think we'll have that technology for at least another 2 or 3 months.

Have you been surprised that Pangram works in 2026?

Yeah, I have been.

Like, I would have said that if they're solving Millennium Prizes, there's no way that Max Sparrow can detect it with his team.

Like, I have been surprised. I would have guessed that, and I guess I would still guess that at some point AI writing will be good enough, or will be But like—

and I don't think the labs are focused on evading AI detection.

No, and so it could be that at some point they may prioritize that or something like that.

Yeah, it's kind of a nice equilibrium where it's like, look, if you wanna know if it's AI written, you can.

Yeah.

But we're not gonna like stuff it in your face with a bunch of like heavy-handed tells.

Yeah.

Or like, you know, additions. But like, it is an option to go figure it out if you want to.

It might be a useful equilibrium. One that the labs don't feel any particular prioritisation to, like, work on, et cetera. I mean, like, I, you know, there's no— you know, it's funny because people talk about, like, AI writing.

Yeah.

And when the conversation happens online, it's often between journalists. And so writing to them means writing articles.

Yeah, yeah.

The vast majority of, quote, writing is just email. Right? And I doubt there's like, you know, and I don't think that there is like some big norm within most corporations, like, what, you had an AI write this memo summarizing the meeting?

Yeah.

I've actually heard the opposite when I talk to people who work in non-media organizations and non-tech organizations, that there is a norm against like, why didn't you have the AI write this? It would've been a cleaner email than—

Cleaner for sure.

Yeah.

One KPI that I would be interested in looking at would be like number of words posted to Slack and over email for employees should not be increasing in the AI era.

No.

Like, do not just become more verbose because you can. But if you're using AI to fact-check everything that you're saying, but you're still writing tersely, win.

No, I agree. And like, look, it's still all in scope. But like, I think like by and large, like if you send me an email and it's like, oh, how to find the TVPN offices and what you should do and what you should wear. And I can tell that it's like AI written, it's not gonna be a big deal to me. I'm not gonna be like—

It's not just a studio, Joe, it's an UltraDome. You'll know it when you see it.

Like, I'm not gonna be like, you know, tremendously offended if like you used—

Send you the sloppy GPT-4o email.

I actually am like, I am kind of like a holdout in that like, I actually have never sent a like AI-generated email. Like, I'm like, if I'm gonna like put words to something, I wanna write them. But I do not think there's gonna be like some big norm against like, for like basic functional communication.

I just love being able to like open up the voice mode or like the dictate website, ask 7 different questions across statistics and links some charts, and then just have like, basically, I used to have 12 tabs with Wikipedia and some economic statistics, and I used to be tabbing through them when I'm actually writing or researching, and now I can just have it all in one go. And then I'm still doing the instantiation and like the thought process.

Yeah, look, it's pretty nice. It's really good for like secret agents. It's pretty good. It's pretty hard to resist. I mean, yeah, like whether I, you know, I'm interested in this book, what would be some other recommendations? What are the through lines between this line of thought and this line of thought?

Yeah, yeah.

It's pretty good. I kind of use it.

What are the biggest, or like most shocking findings from your studies on orality?

Yeah. You know, I think it's actually like, so this is something I'm like very interested in. And I don't know, so, like when you think of like, so orality, what do we really talk about here? It's speech, but there are things that happen in speech that are the important part. So for example, 3 people talking, we've already tried like, oh, let's one-up each other, etc. Let's get a little bit confrontational. It's like, this is like—

Let's get a little spicy.

Let's get a little spicy.

Let's not get confrontational.

No, let's not. But, um, but this is different with AI. Like, it's— and so this is like actually something I'm like very interested in because for the last 15 years of the internet, we are just sort of like bathed in constant both fights and competition, right?

Yeah.

Whether it's Instagram and you're sort of trying to show your life is a little bit better than someone else's, or whether it's Twitter and you like want to get in a good dunk, etc.

Internet's not fun.

And so I think the really interesting question, or one of many interesting questions from a society standpoint, is what happens when we're bathed in the thing that always agrees with And what happens, like, when we're bathed in, like, the most polite, helpful thing on the other side? So it's very tempting to, like, look at it and say, like, okay, chatbots are this huge thing. This is the extension, the acceleration. This is, like, this sort of back-and-forthness of social media, et cetera. But it's not, because it doesn't have that, like, competitive thing that one— the chatbot is never trying to one-up you. It's never trying to get a dunk on you. I actually—

I've had an experience recently. I got a very robust racing simulator.

Oh, sick.

And so, and so I'll be like, you know, focusing on improving like my lap times with a certain class of car on a certain track. And I've had the experience a bunch of times where I'm like, I cut like a lot of, you know, multiple seconds off my time. I think I'm feeling like I feel pretty quick. Yeah. And I go to, I go to ChatGPT and I'm like, how's this pace on this track?

Oh, wow.

And, uh, and it'll just like, I'm expecting it to be like, you are the GOAT. Like, you are.

Yeah. It used to, it used to, it used to be like, you're, you're in the conversation.

It's like, and, and like, you know, granted I'm, I'm, uh, a relative beginner.

Yeah.

But it'll be like, You know, I'll think I'll have improved a lot already, and then it'll be like, yeah, you're you're like clearly in development. I wouldn't call you like a year ago, and it's like, and it's and it's ranking out like it's doing a tier list, and it's like you're you're you're three tiers away from like competition pace. Basically, I know, and it's actually it's like to me, I'm like, wow, I'm I'm actually glad. I'm glad it's telling me the truth because it's gonna. It's pushing me, it's showing me there's more, but it's not a good feeling to get told by this thing—

No! I always said, who is brave enough to admit that they missed when the models were more obsequious and told you you were a genius every time? But now, you like ask the latest model, you know, I have this thought that this is like this, and it's like, yeah, Melissa, you're missing something. It's like, oh, you know what, I'm gonna go back to Opus because that was really nice to me and always told me I was a genius.

Sometimes they'll even sort of like, like, uh, compliment themselves. Like, you'll ask a question and they'll dig in and they'll be like, well, I found some interesting things. I dug into it and here's the interesting thing.

You know what I had? Interesting.

Yeah, well, I started this, okay? So let's give me some credit.

I like to, you know, unfortunately, several— when I frequently will ask the question on incognito mode so that my past history doesn't— so that, so that like, I don't want it to just like flatter my, you know, Yeah, because otherwise it's too— of course. But, um, I had something that I had never seen before.

It literally says, so 2:08 puts you at a pretty reasonable pace, especially since you're still learning the track. The circuit has an unusually large skill gap because the entire mountain section rewards confidence and precision. It's basically, it's basically like saying you suck.

That's brutal. But I had an interesting experience that I never had seen a response before. So I I was reading a book, and I didn't quite understand something. So I asked the model if it could help me understand this section. And it said something weird that I had never got before. It said, you'll have to forgive me because I'm remembering this book from my memory. And it was like, my memory of the book isn't that great, but my understanding— and I was like, that's weird.

Yeah.

Because then the model begins to sound like this, self that has an understanding of its own. It's like, you'll have to forgive me, I'm— I don't exactly remember what the book said.

Interesting.

And I was like kind of unnerved by this like level of abstraction.

So that is weird. That also might be an artifact from specific reinforcement learning on do not memorize the entire book because if you reproduce it, we get sued.

Then it's a copyright thing.

And so, actually, like the correct way to talk about literature is as someone who has read the book, but might not remember it all, and that's actually the goal that gets reinforced.

That's very— OK, that could be a good answer. Anyway, I can't find it. The only reason I mentioned it, I asked it on incognito mode, so I can't pull it up, because I want to go back and read it, but it disappeared from my chat history. But these things, it's kind of like a nerd vibe.

Totally, yeah, no, it was a lot. I was noticing ChatGPT uses I to refer to itself. And I was like, but the models are starting to use agent swarms. So should they start being we?

Have you ever seen that famous screenshot of— it was a real chat. Someone asked it some question. And someone said, where'd you get that fact? And the model said, oh, I heard it at an IBM conference in 1985. Which the model responded, it's like, Where is this coming from? It's like you were not at an IBM conference in '90, but that was the response. Like, weird, weird stuff in there.

The seeds of Watson.

Yeah, exactly.

Gentlemen.

Yeah.

Uh, I have a meeting I gotta get to.

Amazing.

Pretty soon. Uh, I hate to cut it off.

Such a good time.

But I love chatting with you guys.

Yeah, always a great time. Let's do it again soon.

It's an honor to finally have you here.

It's an honor to be here in the dome. I can't believe I got to hit the gong. Yes. That was like, I'm done. I have no nothing left to accomplish.

Hit the hit the soundboard. You get one one option. One option. Yeah. Choose one.

The live show is happening. Odd Lots in Los Angeles. Lots of guests happening tonight.

There we go. Just a classic. I went with the classic. All right, and I want you to experience this.

We'll see you guys on Apple Podcasts.

We're gonna see you Monday.

We'll see you on Monday.

We're out.

We're at a conference tomorrow, but we'll be back on Monday, 11:00 AM Pacific sharp.

We love you.

Sign up for our newsletter. Goodbye. Flash.