1. Podcasts
  2. TBPN
  3. Crypto Going "Bunker Mode", The Mathpocalypse, Starbucks Explores Chipotle Takeover
TBPN

8 Oct 2026 · 2h 19m · 54 chapters

Episode artwork

Crypto Going "Bunker Mode", The Mathpocalypse, Starbucks Explores Chipotle Takeover | with Rebecca Kaden & Michael Mignano, Nathan Benaich, Michael Sindicich & Healey Cypher, Moritz Stephan, Anastasios Angelopoulos, Zach Yadegari

In short

Crypto “bunker mode” as AI-driven math threatens blockchain cryptography; “mathpocalypse” from OpenAI proof breakthroughs; cybersecurity and AI agent risks; plus business/tech sidebars (Starbucks exploring a Chipotle takeover; OpenAI/Anthropic revenue reporting confusion; CrowdStrike security; AI agents for commerce; a Wall Street Journal profile of John Urschel; and a housing segment).

Guests (and backgrounds)

  • Rebecca Kaden & Michael Mignano: hosts/industry commentators (tech/AI/crypto coverage).
  • Nathan Benaich: AI/tech investor and commentator (runs/works in AI-focused circles).
  • Michael Sindicich & Healey Cypher: tech/AI/crypto operators or analysts (participate in security and policy discussion).
  • Moritz Stephan: AI/crypto researcher or practitioner (contributes to crypto/AI risk framing).
  • Anastasios Angelopoulos: tech/AI/crypto commentator (joins discussion on agents, security, and markets).
  • Zach Yadegari: AI/crypto commentator (adds perspective on AI capabilities and implications).

Key claims

  • AI systems may accelerate mathematical breakthroughs that could undermine public-key cryptography used in crypto wallets; Vitalik Buterin warns not to panic but to reduce exposure and consider “AI-vulnerable” cryptography risks (including lattice-related schemes like MLDSA/FALG/FAG mentioned).
  • Crypto markets may not price the threat immediately; quantum “Q-day” planning exists, but AI math is harder to characterize.
  • Open-source model capabilities can reach attackers quickly (example: CrowdStrike reported an open-source AI agent used in cyberattacks on South Korean banks).

Notable examples

  • South Korean bank incident allegedly using an open-source AI agent (Oct 7, 2026).
  • Mythos Preview (Anthropic) released April 7, 2026; six-month gap heuristic to show capability catch-up.
  • Starbucks exploring a $41B takeover/merger of Chipotle (Financial Times).
  • John Urschel (former NFL lineman, MIT math) publishing AI-assisted proofs; paper example: “On the Growth Factor of Random Matrices.”

Written by AI. May contain mistakes.

Chapters

0:45

Countdown to Christmas

A humorous discussion on how many hours remain until Christmas.

“We are technically hours away from Christmas.”
1:08

Crypto Apocalypse and AI

Discussion on the threats AI-driven math poses to blockchain security.

“Although the market is reacting in a very moderate way.”
2:24

Cryptography in Danger

Exploring expert opinions on the potential loss of public key cryptography.

“Seeing all this, Matthew Green, who teaches cryptography at Johns Hopkins, put the situation in stark relief.”
4:04

Market Reactions and Strategy

Insights into how the crypto market reacts to threats and strategies for survival.

“And so the game theory there of no one would ever want to steal all the Bitcoin, right?”
4:58

Cybersecurity Incidents

Analyzing a recent cyber attack against banks and its implications.

“there's a gap between closed source and open source models.”
7:21

The Crypto Bunker Mode

Understanding the concept of going 'bunker mode' in response to threats.

“And it feels like there'll be an incredible amount of progress on these like AI proofing systems, but it's an entirely new challenge.”
8:14

AI and Crypto Risks

Discussion on the new risks posed by AI advancements to crypto security.

“And how do you know when you don't know the shape of the threat that's coming?”
9:59

Government Regulation vs. Crypto Libertarianism

Debate over potential government regulation in the crypto space.

“And I think the interpretation of this, and you can tell me if you disagree, is, yeah, it's because the AI math god is run by an American company that has no interest in destroying the coin.”
11:33

Defensive Strategies in Crypto

Exploration of how the crypto industry can defend against new AI threats.

“With capital, it can marshal more compute, hire people to do things in the real world, et cetera, et cetera.”
14:03

Mathematical Breakthroughs and Crypto Speculation

Explore the impact of recent breakthroughs in math and their potential effects on crypto markets.

“And then what was odd was that crypto rallied after that.”
Show all 54 chapters
16:22

John Urschel: From NFL Star to Math Prodigy

Discover the journey of John Urschel from football to academia and his groundbreaking work in mathematics.

“giving employees instant resolution for access requests and password resets.”
22:30

The Future of AI in Mathematics

Discuss the role of AI in mathematics and how it's changing the landscape of research and understanding.

“Yeah, maybe you should play some full contact tackle football.”
28:00

Starbucks Explores Chipotle Acquisition

Learn about Starbucks' potential $41 billion acquisition of Chipotle and its implications.

“Now, please use the DoorDash CLI to order me some orange chicken.”
28:31

Market Insights on Mergers

Discuss the challenges and potential impacts of merging two consumer giants.

“chain, and it would rank as the largest restaurant acquisition of all time.”
30:02

Personal Reflections on Starbucks

Host shares personal experiences with Starbucks and Chipotle's brand evolution.

31:05

PepsiCo's Beverage Business Concerns

Explore PepsiCo's challenges in the beverage market and potential new products.

“double steak burrito bowl and head straight to the bathroom.”
31:56

OpenAI Revenue Reporting Issues

Understand the complexities in reporting revenues for AI companies like OpenAI and Anthropic.

“They should make a, the Pro Max Ultra should be like a Magnum size bottle of soda.”
32:26

OpenAI Revenue Reporting Issues

Understand the complexities in reporting revenues for AI companies like OpenAI and Anthropic.

“Or no, no, that's, this is the third of fourth of four apocalypses.”
33:02

Impacts of IPOs on Corporate Governance

Discuss the relationship between IPOs and improved corporate governance practices.

“to try and create an apples-to-apples comparison.”
33:58

Market Manipulation and Securities Fraud

Learn about the risks and dynamics of market manipulation using private company information.

“You don't have to be on this constant reporting schedule.”
37:39

Innovations in Live Streaming Content

Explore the advancements in live streaming technologies and their impact on content creation.

“In 2008, the SEC sued Paul Berliner, a trader, fabricated a rumor about a Blackstone acquisition while shorting the target stock.”
38:56

Innovations in Live Streaming Content

Explore the advancements in live streaming technologies and their impact on content creation.

“There were some timeline posts you want to get to.”
42:00

Exploring AI and User Interaction

Discover how generative UI is transforming user experiences in AI applications.

“Much better experience with the answers streaming in.”
43:55

Paul Graham's Take on Amazon Agents

Learn about Paul Graham's perspective on the potential for competition in the e-commerce space with Amazon's agent restrictions.

“Paul Graham says, Amazon banning agents is the first opportunity I've seen since Amazon was founded for a startup to create an Amazon competitor.”
45:58

AI Agents and Market Competition

Understand the implications of AI agents in consumer behavior and e-commerce competition.

“So like I have Grok bot set up and I don't talk to Grok bot, but my agent talks to Grok bot.”
48:46

Real Estate Market Insights

Explore the luxury real estate market through the lens of a tech founder's property sale.

“And yeah, I remember it being really useful for finding stock photos, stock videos, always a little more, a little out of my price range for a while.”
56:56

Discussion on Capital and Market Changes

The hosts discuss the changing dynamics of venture capital, fundraising, and the implications for the market.

“We raised$900 million, which is the largest fund cycle in the USV's history.”
1:02:44

Talent and Company Location Trends

The conversation shifts to the flow of talent and company decision-making in New York versus other locations.

“Trends in New York specifically, like how are you seeing talent flows?”
1:05:06

Evolving Founder Skill Sets and Market Strategies

The hosts reflect on the changing skill sets of founders and strategies for successful partnerships.

“It feels like there's, you know, you can be even more technical.”
1:10:01

Innovative AI Applications in Healthcare

Discussion on AI's role in healthcare and public thinking.

“They're saying they're going to literally be prescribing, you know, medicine from AI in a couple of years.”
1:16:48

Navigating the Chaos of AI Investments

Analyzing AI investment opportunities and market dynamics.

“Your AI agents can now create and modify your Figma files with design system contact.”
1:22:30

The Evolution of AI Technology

Exploring advancements in AI and its implications for the future.

“I think these are all important questions.”
1:24:00

Exploring AI Efficiency and Metrics

Learn about the evolution of AI training efficiency and the value of different metrics.

“And inference sort of remains 30-ish percent across the sample size, but pre-training drops from like 50 % of GPU utilization to like 10%.”
1:26:20

Annualized Run Rate and Market Behavior

Understand the significance of annualized run rate as a metric and its reception in the market.

“Is, uh, is ARR the best metric that we have?”
1:29:00

Financialization of AI and Capital Limitations

Discuss the financial constraints on AI development and the role of capital markets.

“Now it feels like the bigger discussion is like, will$10 trillion show up from the capital markets or not?”
1:31:56

Boom Pop Acquisition and Travel Insights

Insights into the acquisition of Boom Pop by Devon and the importance of in-person meetings.

“going on we may or may not be flying a plane right now okay wow starlink is that good is this actually Starlink?”
1:36:10

Future of Travel in the AI Era

Speculation on the future of travel technologies and their impact on business.

“How are you guys going to work together going forward?”
1:38:00

Transitioning to the Next Guest

The hosts conclude a segment and introduce the next guest.

“Yeah, this was a really special interview, and congratulations, guys.”
1:38:32

Moritz's Insights on AI and Business Outcomes

Moritz discusses AI's role in business outcomes and customer sorting.

“Yeah, I mean, happy to tell you a little bit about what we do.”
1:47:46

Funding Updates and Future Plans

The discussion transitions to recent funding rounds and future strategies.

“The round was led by Benchmark and Index, so we're really excited to be able to partner with them.”
1:48:33

Anastasios Discusses Arena's AI Innovations

Anastasios shares insights about Arena's AI capabilities and challenges.

“Up next, we have the co-founder and CEO of Arena Anastasios is here with us.”
1:52:00

Integrating AI Models and User Interactions

Discussion on the integration of AI models with user platforms and the challenges of model alignment.

“We integrate with your GitHub, your Google Drive and everything.”
1:53:30

Understanding Alignment Signals in AI

Exploration of alignment signals in AI models and their implications for user safety and experience.

“How should people should be people thinking in incidents per million tokens or, you know, just a ranking of best versus worst?”
1:55:30

Challenges of AI Output Quality

Addressing the issues of AI output quality and user feedback mechanisms for improvement.

“And all of it is basically an upstream thing that we can measure that says, yeah, if models are able to do this, then certainly they're not perfectly safe for a user and they're not perfectly aligned to user's intent.”
1:56:50

The Role of Arena in AI Benchmarking

Discussion on the importance of Arena as a third-party evaluator for AI models and their capabilities.

“I think we should talk about one specific case.”
1:59:30

Evaluating Personal AI Agents

Exploration of the future of personal AI agents and the need for detailed evaluations in this space.

“If you zoom out, the point of Arena in the ecosystem is we provide a neutral platform that everybody trusts based on information from the real world, real users, not just random benchmarks that people have come up with.”
2:01:40

Wrap-Up and Upcoming Discussions

Final thoughts on current discussions and a teaser for the next guest and topics.

“And part of that is building the consumer reports or the gardener or what have you.”
2:03:21

Zach's New Venture in AI

Zach discusses his new venture in AI, focusing on consumer utility apps and his vision for the future.

“You had the shortest earn out in history.”
2:06:03

Exploring AI Monetization Strategies

Learn about innovative monetization strategies for AI products and consumer behavior.

“Whether that's like showing ads to the agent without biasing the answer in any way, just showing different products.”
2:07:38

Consumer Values in the AI Era

Discover what consumers truly value in AI assistants beyond productivity.

“What do you think people misunderstand about consumer philosophy, consumer behavior in the AI era.”
2:09:14

Journey of Selling a Startup

Understand the emotional and financial journey of selling a tech startup.

“How has your, what was the journey like selling the company?”
2:11:08

The Future of AI and Hardware Integration

Examine trends in hardware development and computing's integration with human life.

“Or do you think there's ways, you know, if you really figure out sort of agent side, why not figure out ways to make it accessible to more people?”
2:14:06

Unique Housing Trends: The UFO House

Learn about the remarkable UFO house and its unique architectural significance.

“Yeah, it's interesting because part of why we're in like the voice modality with AI agents and text is just the cost of inference.”
2:16:49

Wall Street Journal Reactions to the UFO House

Explore public reactions to the UFO house from Wall Street Journal comments.

“Wayne concluded that the structure was too heavy to be lifted by a standard-sized helicopter, and airlifting would require a long list of permits, including approval by the FAA.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On TBPN, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Year one. Today is Thursday, October 8th, 2026. We are live from the TVP and 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 to use. Corporate cards, bill pay, accounting, and a whole lot more all in one place. How many days until Christmas? I have a better question. How many hours? Oh, we're counting down the hours now. Okay. There are 1 ,862 hours until Christmas. Get ready, folks. Yep, that's right. We've moved from days until hours. That's how close we are. That's how close we are. It is October.

0:47We're just hours away. It is early October. We're hours away from Christmas. We are technically hours away from Christmas. That's correct, yes. Anyway. We're only 1 ,800 hours away from Christmas. Yeah. It's the apocalypse day. There's so many apocalypses. Mathpocalypse. There's a crypto apocalypse going on. Although the market is reacting in a very moderate way. But the timeline is being dramatic. So we've got to break that down. Ethereum founder Vitalik Buterin and others in crypto are sounding the alarm that AI-driven math could threaten blockchain security. And so crypto is going bunker mode.

1:32That's the term that they're using. So we've all seen the progress in AI-driven math, solving new proofs, solving Navier-Stokes. There's a lot of impressive progress in math that maybe has applications in physics, maybe has applications in engineering, fluid dynamics. A lot of these things are further out. And if they wind up being useful, they might increase efficiency here or there. Not as far out as you think. But the big thing is that crypto is here today. And if there's an advance in the underlying understanding of how cryptography works, it could really shake up how the crypto industry works and how these products are secured.

2:23So are they in danger? It's the Ralph Wiggum meme, I think. I'm in danger. Seeing all this, Matthew Green, who teaches cryptography at Johns Hopkins, put the situation in stark relief. He said, I think we might lose public key cryptography, which is a huge statement from an expert in cryptography. it would require a pretty massive restructuring of many security protocols, but also the crypto industry. At the moment, cryptography as a discipline has been exempted from the list of published papers related to AI-enabled math breakthroughs. So there were hundreds of papers, and it's very clear that either the agent swarms that were sent to go solve math problems, they said, hey, don't even bother with any cryptography stuff.

3:20Or if they did, they didn't publish those, and they're burying them for now until there's a way to actually patch those. Or maybe the cryptography is just too strong. That's true. That's a third option. The crypto industry is a little skeptical about it. They think, oh, man, there might be some flaws. Of course, they've been preparing for things like this before. Quantum computing is potentially existential to many blockchains. And so there has been the concept of Q day and how they work backwards from that to make sure that they are quantum resistant. The interesting thing is like historically when the crypto industry has faced, you know, headwinds, the strategy has been HODL.

3:59HODL. And that's not necessarily going to be effective for this next chapter. That's true. Yeah, that's a good twist. Of course, there are very weird dynamics where if you break all the Bitcoin keys and all the wallets wide open and you completely erode all the trust that's been built up in that virtual currency, yeah, you stole it all, but it's worth nothing. And so the game theory there of no one would ever want to steal all the Bitcoin, right? So it's more like, can someone go after long tail? I'm sure there's some people that would do it just for the love of the game. Just to watch the world burn?

4:37Joker mode? Yeah. Yeah, that's probably possible. We could be seeing a Joker-fied. Oh, you have the Joker ready to go. Yeah, a Joker-fied AI agent swarm would be a bad situation. Yeah. For sure. But there is some cause for optimism. Much like the broader cybersecurity scare that happened back in March of this year, there's a gap between closed source and open source models. Most people say it's around six months. The capabilities catch up in certain areas. It seems like it's faster. Sometimes it's three months, but six months is sort of a useful heuristic I have found. And that actually maps out pretty perfectly with the most recent news.

5:27So if you remember, Mythos Preview was dropped by Anthropic on April 7th, 2026. That's the fourth month of the year. And then CrowdStrike announced that a 26-year-old in China used a open source Chinese developed AI agent to execute cyber attacks against South Korean banks on October 7th. Like the same, I think the report came out October 7th. So it's basically just exactly six months later, you see like a open source model be used in a cyber attack. Now, fortunately, the banks have a ton of redundancies, the ability to roll back transactions. They store things to tape. They store things on paper sometimes.

6:12There's a lot of different ways to secure a traditional bank that is not purely digital, even though they might have a digital front end and a lot of the money might be stored digitally. and the interesting thing is that when I read at least about one of the South Korean bank cyber security incidents it was a pretty small amount of user data that was stored it was like names and addresses and like phone numbers of like it was a significant number it was like tens of thousands of clients or something like that but it wasn't like they robbed the bank and they got the money or they got everyone somehow they got into like one system for information so it could have been like they they hacked into like like their paperless post account that was like the what the bank used to send out invites to clients or something like that like there are a whole bunch of different systems and and where at what level did they get in is is important but it's clearly happening got into the client gifting software potentially everyone chocolates maybe maybe uh and so obviously you don't want hackers anywhere uh and so cyber security is more important than ever.

7:20And so the crypto community does have, you know, a couple months to go bunker mode as they are putting it. And it feels like there'll be an incredible amount of progress on these like AI proofing systems, but it's an entirely new challenge. It's very, it's less, much less understood than quantum computing, which has been theorized mathematically and sort of formalized for years before the capability came. This is different because we're getting this capability before we've really characterized the shape of the threat. Like they know with quantum, like there's something called Shor's algorithm and like they got to fix that.

8:00So they got to make sure the Shor's algorithm doesn't work to break Bitcoin while it's open. They know the mathematical problem. They got to solve that. In this scenario, it's like, there might be a different algorithm that runs on commodity hardware that works. And how do you know when you don't know the shape of the threat that's coming? So it's a crypto apocalypse potentially, but the market doesn't think it is. Bitcoin's only down 3 % today, and it's still up 3 % for the month. And so the market is digesting this and optimistic. There's still lots to do if you work in crypto. And potentially, even if you only own some crypto, there's things to do.

8:40Vitalik gave some recommendations on, you know, don't panic, but also maybe start thinking about this more intently. He said, I don't recommend anyone scramble to move their funds to new wallets today, but we should take the risks of cryptography from AI accelerated math seriously and minimize our exposure to not just quantum vulnerable cryptography, but also AI vulnerable cryptography. The new core area of risk from this viewpoint is unfortunately MLDSA, FAG, lattices. And this is also another reason along with quantum why ECDSA might fall even faster than expected. Hence the fresh address recommendation.

9:28So far most people have been in the mode of thinking elliptic curves broken, hashes safe, lattices safe. But there's a good chance that the concrete security of lattices will take serious hits from the next two years of AI math. And so some people were posting. Scott Liu here was saying, oh, that's weird. The AI math god is somehow no better at cryptography than a 2026 human. We got super intelligence in math but not cryptography. What's going on? He thought for a minute. This is over. And I think the interpretation of this, and you can tell me if you disagree, is, yeah, it's because the AI math god is run by an American company that has no interest in destroying the coin.

10:15In, like, total chaos. Yeah, but in the open source world, it's a very different game. Yeah. Yeah. And so still, still interesting thing about the still interesting to think about the game theory, even in the open source world. But I think that will all be downstream of a whole bunch of changes. And then there is the question of, you know, how how does how does crypto handle this? Like the cryptocurrency industry has been libertarian. They've been against the government in many ways, small government. I want to own the money. well now you're facing a threat from open source from a non-government regulated entity and so while a lot of people in the ai safety community have said this is the moment for big government this is the moment to register large clusters this is the moment to track all the data centers to to have licenses for inference and training and and reporting to the government that is a solution like if you lock down open source and you say yeah you like if you're going to spin up a million agents and try and do a bunch of inference to try and break Bitcoin wallets.

11:23Like, we're going to find out about it. We're going to arrest you. And it doesn't matter that you're using an model. That's a very different psychological frame of mind for a Bitcoiner. Yeah. The other thing here is that there's obviously bio risk with model advancements, but it is going to be wildly different for somebody to use an obliterated model to try to create a bioweapon and then actually sort of distribute that out into the world than like using a new model to try to like mess around with cryptography yeah that feels squarely in the hacker camp totally like there's people that would just be kind of messing around and then accidentally have a breakthrough or accidentally or maybe intentionally cause financial chaos but that's purely digital yeah and that's the kind of thing that millions of people have been doing for a very long time yeah just generally kind of hacking around wildly different than the kind of person who would go out and try to create you know a new pandemic uh yeah and it also yeah and then also uh another layer will uh of this will feed into like the whole ai doom loss of control scenario because you get an ai agent that's hyper intelligent and it's also able to break bitcoin wallets it breaks a couple open then it has capital.

12:43With capital, it can marshal more compute, hire people to do things in the real world, et cetera, et cetera. And so the doom scenario becomes like way more human enabled if you are able to amass capital pretty quickly as a rogue agent. Do you have a take on this, Tyler? Yeah. I mean, I feel like the crypto industry broadly is like fairly good at self-governance, right? You've seen like, even if you're talking about like Bitcoin, you know, there's like, they can actually like change the protocol, you know, you like Bitcoin cash, which is the shoot off stuff like this. I think that Bitcoin Policy Institute, like, I don't know, maybe a month or two ago had this open letter about how, you know, the solution is we got to build up defenses much more and we've got to use, you know, spend a lot on inference to try to like figure out what is the right path moving forward.

13:30It's, I think most people are very against like limiting open source models, stuff like that. Totally. Yeah. It'll be, it'll be interesting to see how these worlds collide and where things go. I had some other take, but let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents. Whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish. Thank you for the continued clapping. Ben, I don't know if you included this earlier, but he says, lucky coincidence, the labs are only successfully cracking esoteric math problems and not the problems where a solution would give the winner an asymmetric commercial advantage again this is what i've said a bunch of times on the show now uh math problems are the perfect category of problems where you get a result and you're you know incentivized and excited to share the result there's no immediate opportunity to commercialize a lot of the a lot of this stuff yep and so for every for every breakthrough in math you should assume there's and the reason this should be getting everyone excited is you should assume there's a bunch of other breakthroughs in other fields that have more commercial applications where the person that has, and it's not just the labs, it's people using the current frontier models to find breakthroughs in their own field.

14:50And so, again, I'm expecting a real explosion of progress, especially as we enter the later have for this year yeah or the or the sort of uh the uh end of the end of the year i believe uh will depew posted about this that's what i was thinking of will depew posted uh i think there was like a slight bitcoin sell-off like a while ago like maybe around the mid-summer time period and he sort of took a victory lap or something it was like saying like yeah in hindsight once we once like the ai community built like the hack machine 9 000 uh it should have been obvious that you should get out of crypto.

15:29And then what was odd was that crypto rallied after that. So like, he was kind of a little bit too early with the take, but now it feels like, okay, everyone's pieced it together a little bit more, but also there's so many other factors where you could see this narrative dying down and all of a sudden there's some other narrative that's driving crypto and people are piling in for whatever reason. But it was, yeah, maybe like, you know, Even it's been pitched as an inflation hedge. There's always been people who have said the restricted supply is valuable. You could have a view on inflation that becomes really popular, and then all of a sudden people are piling in.

16:09There's a whole bunch of reasons why the market can move. It's not purely driven by this, but it was funny that he was posting about this like months ago. Anyway, moving on. Console.com. Console builds AI agents that automate 70 % of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. So OpenAI's drop of hundreds of new mathematical proofs is now being called the mathpocalypse. There's some white pills. There's some black pills. People are all over the place. Obviously, we talked about the crypto knock-on effects. Nabeel Kreshi says, Mathematicians, AI is plagiarism, and we don't want to know the answers anyway.

16:51artists ai is plagiarism and the art sucks anyway chess players hmm i wonder if i can use ai to increase my rating uh because there are some people that are just delightful with um delighted with advances and not really uh you know frustrated by there's actually yeah how do you how do you prevent against like botting in online chess um like couldn't you can't you just be like running your own like like i don't know and i'm comparing that to like let's say somebody using like an aim bot yeah no no i mean bots are like undetectable like all the bots can like one-shot captchas and and fake but not even that but basically just running like an ai co-pilot yeah no you can definitely make that but so unless you're playing chess physically in front of somebody and watching them you know yeah and then and then what do you think well uh i mean if you're like elo is like 800 and then you start making like you know 2 000 rated moves like it's pretty easy to tell that stuff yeah it's like i think like if you if you're on chess.com or whatever it's like i think it's fairly hard to cheat hard to cheat or or people just don't cheat because it's like what's the point and like it's like okay yeah i played against somebody who was like smurfing and they had an 800 elo account but they were clearly playing at 2000 elo like okay i lost that game let me queue again, okay, I got someone who's actually 800, and I'll just keep playing until I get someone who's actually my end.

18:18Yeah, that's probably true. If you're consistently using an engine, I think it's not that hard to figure that out. Yes, but this is going to change, because the models will be so good. It's over. I mean, the models will be so good that you'll be able to say, I'm Elo 800, help me play at 850, and also really mimic the human experience. So don't act like a model. But yeah, I agree. There's like little incentive to actually cheat on chess.com. Yeah, yeah. Gold says you can tell if they're bots if they statistically always or nearly always make the optimal move. Yeah, exactly. Like there's no fun.

18:57And you can always just go play against the best stockfish model if you want. Like you can just max it out if you just want to play against the best computer in the world. Nobody does. And so there's not that much value in it because everyone would just be like, cool yeah you just ran the model that everyone else has like do you want to play against me or not like if you want to just be the model just go and you don't win anything so there's not any incentive i think i think that's how it works out i don't know uh i do think yeah i mean in the future like you will be able to i mean you you and the gym have been accused of using you know robotic assisted lifts right you'll wear like big sweatpants and you're on the deadlift can you imagine having and you have your exoskeleton you know putting up you know 800 pound deadlifts i mean people you know know you're strong but not that strong they're skeptical yeah i don't know if you'd get a lot out of that you could bring fake weights that's that's one way i mean people do stuff but then why not just photoshop it why not just go and photoshop a screenshot of you on your profile on chess.com with a 4 ,000 ELO and then share that on X and be like, I'm the best chess player ever.

20:12But if someone fact checks it, then you just look like Nate. Yeah, and so eventually you have to get down to fact check by walking through a metal defector and going to an actual chess match because like, oh, you're really better than Magnus? Let's see if you're better in person. That's what will happen. And so there's like, you're going to wind up you're going to be wind up in the Faraday cage with Magnus eventually. I don't know. Anyway, let me tell you about Shopify. Shopify is the commerce platform that grows with your business. And that's yourself. In seconds, online, in source, on mobile, on social, on marketplace, and now with AI agents.

20:48There's a very interesting post in the Wall Street Journal about a former NFL lineman who solved a decades-old math problem. John Urschel gave up his career with the Baltimore Ravens to study math at MIT. And he has... Get this guy on immediately. He's so jacked. It looks like, yeah, he definitely doesn't need the exoskeleton in the gym. He's a former NFL player. John Urschel decided to do some light reading this week. And for him, that meant leafing through the latest math papers from OpenAI about topics such as the Riemann hypothesis. Oh, my Lord, he thought. I was absolutely floored. Good rhyming.

21:28Or Urschel was blown away by the findings because he happened to understand the hundreds of proofs better than nearly any human on the planet. Nine years ago, he traded the National Football League for the Massachusetts Institute of Technology. Crazy that you can take like that. The NFL has got to be an incredibly demanding schedule. Like you can take like thousands of hours off of your like math career and then like go back in the game and not be. Well, not to mention getting just slammed in the head. I know. Yeah. Oh, yeah. They want you to believe in CTE. Counter example. Make it make sense.

22:08One of one sample. Counter. No, he's possible. He's actually the smartest person to have ever walked the earth. So even after getting hit in the head. Yeah. Or it could be reversed. You get hit in the head and it makes you even more of a genius or something. Yeah. Unstuck something. Yeah. Sort of like a magical realism. Tyler, is this going to make you get into watching football? Yeah, maybe you should play some full contact tackle football. It might make you better at math. You never know. Urschel was blown away by the discoveries. And Monday, instead of studying abstract play designs, he was publishing some impenetrable language of his own to prove a decades-old math conjecture.

22:53Urschel's paper titled On the Growth Factor of Random Matrices is one of a deluge of AI-assisted proofs that mathematicians have achieved in this new era when they can combine their genius with supercomputing power. But within this community of brainiacs, there isn't a single person quite like the 35-year-old Urschel. After three years at the Baltimore Ravens, the offensive lineman who is listed at 6 '3", 313 pounds. ditched his helmet and became a professional egghead in 2017 he hasn't regretted for a moment he says i feel like i bet on myself in a big way i'm living my best life right now you'll love to see it long before urschel became an assistant professor at mit studying matrices spectral graph theory and other subjects that sound like gibberish to just about everybody shots fired by the journal but i agree he was an overachieving kid attending penn state while playing for the nittany lions he did more than pursue an ambitious major by studying math and win and he won the prize known as the academic heisman by the time the ravens drafted him in 2014 he already had earned his master's but a year after protecting joe flacco urschel started to miss his life crunching numbers while he felt his analytical background was useful to him as a football player for instance it's a he says it helped him learn the playbook it didn't quite scratch his intellectual itch that's when he decided to apply to become a doctoral student at MIT, the school he had always dreamed of attending.

24:24He became moonlighting there in 2016. And so I know, I know what this is all setting up. What is this? He is going to, he is going to star in the next Terminator movie. Football player, math genius fights back. Triple threat. Yeah. This is exactly takes on the clankers. Yeah. I would love to see it. I mean, you would hate to be a clanker with this, with this guy walking you down. For sure. But he seems to be a friend of the clankers because he has actually been using AI to assist him in his research. So his acknowledgments in the paper that he published cites the help of open AI models and says that he welcomes any question about the role of AI, which is upending his entire field.

25:09He says, but supercomputers haven't dented the passion that this football player turned mathematician has for math. He compares it to his enjoyment of chess. He also likes chess. The computers are better than the humans, yet he still enjoys the pursuit of finding the right move. I really want to understand the why of things, Urschel says. It doesn't ruin the why. AI doesn't ruin the why. That's a very interesting thing. And I feel like, I mean, just in knowledge retrieval over the past three years, I have felt that. Like, I know a lot of obscure factoids about Silicon Valley lore, you know? But when ChatGPT came out, I wasn't like, oh, no, like, the computer knows more about Silicon Valley history than me.

26:01They're like, oh, no, like, the computer knows more about the, you know, the random companies. You're basically one of the top guys in your field. I did spend like five years making YouTube documentaries about these companies and like trying to learn a lot and like piecing it all together. And like ChatGPT can do that. But like it's been such a benefit because I want to know. So I'm looking up stuff all the time and I'm like, I want to go on the journey. I want to gain that information. And it gives that to me. It's weird. Very, very weird. We benefit from having knowledge and understanding within us.

26:36like it's nice to go through life yeah and be able to understand things in real time versus reaching just because you have complete access you have complete access to all the information you need and yeah yeah i mean there's still obviously the jobs question how do you monetize that if your job was just solving math problems or your job was just uh you know assembling you know sloppy research reports uh you you might be going through a process now but uh i don't know i i'm still i'm still optimistic i i still uh i still feel like we're going to be hearing from uh this gentleman john urschel for a very long time yeah maybe he's like you know what i'm going back to i'm going back to football math has solved going back to football yeah when I wonder when he actually gave comment for this article.

27:31This was like yesterday. I mean, this literally posted today at 8 a.m. So I imagine that he was on the phone with the Wall Street Journal over the past like two days. Because it's a pretty detailed profile. We should get him on the show. I'd be fascinated to hear about what he thinks. Yeah, I mean, I think people will be continuing to study this stuff for a long time and learning it to apply it for themselves. Anyway, Creighton Cycle's memeing it. He says, open AI. We literally solve math. Him says, nice one. Now, please use the DoorDash CLI to order me some orange chicken. Speaking of ordering food, Starbucks is thinking about buying Chipotle, or they thought about buying Chipotle.

28:16The headline is, Starbucks has explored taking over Chipotle, according to a Financial Times report. This is a scoop in the Financial Times. Let's see. They've explored merging the two companies in a mega deal. It would be a$41 billion market value for the burrito chain, and it would rank as the largest restaurant acquisition of all time. So the world's leading coffee shop chain has worked with advisors in recent months on a takeover proposal for Chipotle, which I believe had McDonald's as an anchor investor a long time ago. I don't know how divested they are at this point, but it's interesting to see that they're finding another dance partner.

28:58A burrito chain with a$41 billion market cap. The status of Starbucks' takeover plans and whether the company has submitted any formal offer to Chipotle could not be determined. A transformative deal of this size might struggle to get off the ground, the people warned, and a tie-up may never materialize given the complexity of combining two consumer giants. The early stage plans would also reunite Brian Nichols, Starbucks chief executive with the fast food chain, where he made his name over six years since as a CEO. So he left Chipotle August 2024, went over to Starbucks from Chipotle. Chipotle has lost half of its value in that time.

29:36So he was performing well there, but maybe it's an opportunity to put them together. The combined market cap would be like$102 billion. and it's also just what fast casual had been on a crazy run across the board so I don't think it was him leaving that caused the decline but he clearly loves the brand man it is so crazy to think about there was a time in my life where I was actually averaging one visit to Chipotle every day every day you know some days I wouldn't go but other days I'd go twice have you ever driven starbucks and now i'm not even averaging once a year whoa yeah just moved on what about starbucks have you ever been a starbucks guy nah so so so bad you don't like charbucks that's the whole thing uh starbucks the the the coffee is like burnt often or like it's a little bit more towards burnt so they call it charbucks but it's i think it's because people at the time liked it so that was more popular and then they add so many things on top with like caramel and whipped cream that like you can offset the the harshness of the raw black coffee ingredient mosquitoes on the set we get the get the fly swatter now do some damage i'm shooing it away uh mike isaac says co-worker who clocks in to work carrying a venti whole milk quad latte in a double steak burrito bowl and head straight to the bathroom.

31:07Lots of jokes like that. PepsiCo says some really bad things. Joe Weisenthal shared, PepsiCo CEO, we don't feel good about the beverage business.

31:21PepsiCo CEO, we're putting all the urgency to improve sodas. Yeah. Should they add alcohol to the sodas? Topo Chico, the sparkling water brand, created an alcoholic seltzer. Do you think there should be Pepsi, Pepsi Zero, Diet Pepsi, Pepsi Max, Pepsi Spiked? What do you think? I think they should be optimizing for Pepsi Pro Max Ultra. Pro Max Ultra. Yeah. It is interesting that Coca-Cola never did like a rum and Coke. You could just do that if you were trying to just skew Max and get the most number of stock keeping units around. Yeah. I don't know. They should make a, the Pro Max Ultra should be like a Magnum size bottle of soda.

32:11That's where things are headed. Okay. Well, let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite aided to deploy web app servers, databases, and more while Railway automatically takes care of scaling, monitoring, and security. The last apocalypse, the ARR apocalypse. Or no, no, that's, this is the third of fourth of four apocalypses. We got a housing apocalypse coming up. But OpenAI reports revenues differently than Anthropic. And so there was a leak that said OpenAI was at almost$70 billion. And then there's another report today that says they're almost at$50 billion.

32:46And it's very clear that the numbers are very closely held. And then there's a game of telephone happening with investors and LPs and other investors. And then by the time things make it to the press, it's all over the place. and it sort of shifts and people true up the numbers to try and create an apples-to-apples comparison. If you're not following the, it's like on AWS, if you buy an OpenAI model, OpenAI keeps that dollar as revenue. Anthropic gives some of that to Amazon. They both probably give some to Amazon, but Anthropic counts the part that they give to Amazon as a cost. And so they have a higher gross revenue number.

33:27And so making these companies apples to apples, it seems like that's what someone tried to do a little bit, but I don't know. It seems like they did it sloppily. It seems like they did it sloppily. It's still very unclear. And, yeah, it's unclear why it didn't get corrected faster. Yeah. But, I mean, like, extremely difficult environment to be operating in because part of the benefit of being a private company is that you, I mean, like one of the biggest benefits. You don't have to be on this constant reporting schedule. Yeah. But at the same time, You're like moving the public markets. Yeah, because these companies are at a scale where so much public market activity is predicated on their metrics.

34:15Yeah. It's sort of, at time, it's prudent to actually share these numbers at different times. But there's also so much. I mean, all the different media companies are working in a perfect world. The legacy media would be like have a new article every day that just had all the updated numbers. Like what was, you know, they would love to publish. Maybe like a quarterly report. Here was yesterday's revenue. They could go on a website like Edgar Online. You could call it Edgar Online and file it. With this SEC, there could be a whole commission around securities and exchanging securities. Yeah. I mean, Sam's been talking a lot about IPO stuff.

35:00What was his take with Vanity Fair? He had something about IPO timeline, like he'll do it when he needs to do it, whatever. The interesting thing is that I do feel like the IPO-ing is aligned with better corporate governance, distributing the gains. Yes, actually get the stocks of all the labs into the retirement accounts. That was very controversial with SpaceX. When SpaceX went out, people were like, I don't want it in my 401k. But there is a scenario. Of course, it's like float-weighted, so you don't actually need to worry about it blowing up your retirement account. But there's a good case for like, okay, if this really is a really, if like the bull case is true, like, yeah, distribute the gains through everyone's investment accounts, getting slices of it just immediately because that's the way things diffuse through the public markets because everyone owns ETFs.

35:57Everyone has some money invested in the market. It becomes much easier than like the small group of insiders. There was this rumor that we were digging into, which is like, is someone trying to manipulate the market? Because if you leaked a number about a private company, normally that wouldn't move the market because everyone's like, oh, who cares if like some tiny company like is, you know, doing okay on revenue. Like if I, like if we leak that like Lucy's doing a real, oh, we're going to blow out revenue this year. There was no point, there was no point even look at, even look at one of the greatest company Silicon Valley has ever produced, Stripe.

36:35There was no point where if it leaked that Stripe had missed their numbers for a quarter that you would see... Like Visa spike or MasterCard spike. Yeah, it wasn't really moving the market in that way. Completely different with OpenAI Ananthropic. Like SpaceX, Oracle, SoftBank, there are plenty of public company proxies. And so it would be tempting if the SEC didn't exist to go and leak a number that's better than what's true and then sort of buy the rumor, sell the news. So you would buy and then leak a really high number that's just a rumor to the press. And then when the real number comes out, you're already short.

37:17So you're catching that on the way down. This would be security fraud, by the way. So don't do that. But it is a wild thing that we haven't seen before because manipulating public stocks with private company information just hasn't been viable in size before. Maybe very, very tiny here and there. But there are some cases out there that make it very, very clear. False acquisition rumors are popular. are popular. In 2008, the SEC sued Paul Berliner, a trader, fabricated a rumor about a Blackstone acquisition while shorting the target stock. Media outlets picked it up and the stock briefly fell 17%.

37:59Berliner settled SEC fraud and manipulation charges. And there's been, there's a Citron research founder on securities fraud scheme around, prosecutors allege that he used influential public commentary while misleading investors about his trading intentions. The case did not require the same fabricated revenue fact pattern. So just interesting dynamics. So if OpenAI wants to pop from 50 to 70, you're thinking hard seltzer, right? Launch a hard seltzer. That would be the logical thing. It worked for Topo Geico. It's worked for many other fantastic businesses. Just give everyone a hard seltzer. No, obviously stick to light beer.

38:45Let me tell you about CrowdStrike. Your business is AI. Their business is securing them. CrowdStrike secures AI and stops breaches. Was there anything else before we go into the housing-pocalypse? There were some timeline posts you want to get to. Pull them up. You got This is the Best Aura Flip. You wanted to see this? Are you on the timeline? My computer is done. Oh, there's sound too? Ready for launch. Wow. Another person who dropped a nuke in Minecraft before Tyler could drop a nuke in Minecraft. Wow. Low skitches. Can't say I'm surprised. And then doing the kickflip is great. It's so funny that you can actually just mash these up.

39:36Can you imagine being 14 and seeing this? amazing and just absolute absolute joy um yeah here's why at walls this is this is what we're all about i'm pivoting to something llms can't do day drinking 7 000 likes sean frank is uh vague posting we'll let you try and figure out who he's talking about i have a friend young 30 nine figure exit i've known him for years before he made it before he knew what he was going to start the one thing he always had an unwavering belief that it was going to work out. He was willing to burn the boats over and over, always doubling down, always moving bigger and bigger.

40:15It's changed the way I want to operate. If you want to change your life, do it. If you want to win, then win. Don't let anything stop you. The only thing you have control over is your attitude and effort. I love some mindset posting. Fire me up. Got the thousand like bagger on the Sean Frank post. He's locked in and he's encouraging you to lock in. Anyway, Chinese streamers are creating content on a completely different level. We have taken our live stream efforts very seriously here at TVPN. We have cameras that track us. We have chyrons and stingers and all sorts of fun things. But it's clearly amateur hour over here.

40:57Because in China, they have taken things to unthinkable heights. and we'll show you this video and you can understand how different it really is. There are levels to this game and we are playing it on easy mode. This is a He's a fairly large machinery that physically moves her chair with a seat belt and spins her when chat Donates a mere seven dollars USD apparently. So I think we should have a button That puts us into a racing simulator. Okay. And we can't go back to the show until we get a sub seven minute Nerva ring time. Okay. I like that. That's fun. And it should actually be a free button.

41:45Yeah. So anybody can hit it. I mean, I do think we should get this exact device and just put Tyler in it. And then if chat, every time we get a follower on X, for example, he could be spun. it is uh it is crazy the level of interaction uh the the chat uh has has fun with like drink the diet coke with two hands every once in a while but uh this is a complete different level of uh of interaction uh the gen the generative ui stuff in uh in chat gpt we talked about it yesterday and got a chance to demo it. Much better experience with the answers streaming in. Like I asked about, like explain this whole SEC market manipulation thing.

42:34I fire off a big prompt. And like it gets me the answer very quickly. And then it goes and explains the laws. And then it builds some widgets about different items. And it's just a much better UI experience and product experience. And I'm excited to see other things people do with it. You can even get it to make cheese. Write a poem about cheese and message bubbles that look like Swiss cheese. It says, oh, cheese, my love, my golden delight, you make every cracker feel just right. From cheddar bold to brie so sweet, you turn a snack into a treat. With little holes and dreams so big, I'd choose you over any fig.

43:13And if the moon were made of brie, I'd eat the stars and save some for thee. It's a nice poem. but we're here to talk about user interface design and it's funny that it can you know lay out those circles and yeah the future of this is like it can just draw any pixel at any time for anything so it can sort of generate video and graphics and games on the fly and you will get your math what was it MMO you wanted an MMO of something I forget what it was I keep asking for games I don't really play video. But we got to talk about, we got to talk about Paul Graham. Okay. What do he say? Paul Graham says, Amazon banning agents is the first opportunity I've seen since Amazon was founded for a startup to create an Amazon competitor.

44:02People will want agents to buy stuff for them. It will be one of the main use cases and they don't want to use some Amazon supplied agent to do it. Any business that bans agents represents an opportunity for a startup to create a competitor. If they need to ban agents, it must mean people want to use them or there would be no point. And that in turn means there's demand for a competitor that allows them. Your usage restrictions are my opportunity. It sounds great. Sounds coherent. But Nan Yu says what you're probably thinking. I mean, they can just unban the agents. It's a single config file. So yeah, imagine you spent 30 years building an Amazon competitor that the Amazon creates.

44:48You spend 30 years. You spend, you've acquired millions of trucks. You employ more people than almost any other business on earth. And you're finally like, okay, like I'm finally, I'm finally ready to, to, to really, you know, eclipse Amazon. And Amazon's like, okay, we unbanned the agents. Yeah. And then, so again, it's, it's, uh, the other, the other, the other pushback here from someone named David H says, how lazy can we get? There's absolutely no value in an agent buying something on Amazon. It's just something people do to sound cool. I already have a human agent who buys way too much stuff on Amazon called my wife.

45:28It's the most Gen Xer t-shirt from Venice Boardwalk I've ever seen.

45:38To really throw it in the old ball and chain there. It's ridiculous. this uh anyway i think like i'm not there yet because i don't think it's an opportunity for startups to try to recreate the biggest online commerce platform in history if anything it's like an opportunity for walmart who is leaning who is sure leaning more into agents yes and i think we're gonna see does that actually turn the tides okay right so are you long sparky or rufus because amazon has rufus their internal ai agent walmart has sparky their internal ai agent are you going rufus mode or are you sparky max i mean how hard is it to tell your preferred agent to just get on rufus that's what i'm saying i think rufus is going to go in the kennel and i'm going to have a kennel of agents rufus and sparky will be in there and then astra 7.5 will tell them exactly what to do and they will go and do it because they will think it's me and they won't know i think that's actually what's going to happen and i think you don't think rufus will be able to detect figure out a bot detection no ability no like because it'll just be happening computer use and it'll just be and i'll just be like yeah yeah okay uh codex for the next five hours that i'm on the computer just actually record all my mouse movements and like create program that actually mimics me like learn my style of using the computer like when do I alt tab when do I use the you know search when do I type how do I type what type of spelling mistakes do I make when I type learn all that memorize that and then mimic my computer use pattern so you're not like right now it's like it's like go straight there right but it will just be like oh yeah I'm just I'm I'm using the computer like I'm John I'm writing like I'm John I can you know it'll be close enough.

47:34And, uh, and I think that people will wind up, they'll still have like their favorite front door to AI and it might be chat GPT, might be clawed Gemini, whatever they'd like, whatever they wind up in, but then, but then, but yeah, but then they might like, there are people that will have, Oh yeah, I do some work on X and, and so I want to manage all my DMs. So like I have Grok bot set up and I don't talk to Grok bot, but my agent talks to Grok bot. And then my agent goes over to Muse because Muse has really good integration with Instagram. And if I try and go scrape Instagram, I am going to run into a bunch of problems, but Muse is happy to talk to me.

48:09And so, especially with these, with these, uh, these systems that operate over chat, it's like, it's going to be really, really hard to do bot detection. If you're available over telegram or WhatsApp, right? It's like, it's just text that's coming back or like, or we integrated it was slack and so now i talk to it on slack it's like you're telling me that it can tell what's what's an ai prompt that comes over a slack message that's 12 words go buy me yeah a paper towel holder from the 1960s the opportunity is to create a kennel for dog themed agents kennel a i that's now raising k kennel si uh okay so uh should we go over to the mansion section we are off tomorrow in some heartbreaking news for you all so uh we got to do some mansions that we have to do it we have to do it we have to get to the mansion section so we got to talk about it today uh the first home comes from a tech founder the shutterstock founder have you shutterstock not a big uh shutterstock customer shutterstock a wonderful fantastic business yeah really cool business.

49:20And yeah, I remember it being really useful for finding stock photos, stock videos, always a little more, a little out of my price range for a while. Like you'd have to pay per image. And I always liked the, the, the unsplash, the free versions or the cheaper versions, or maybe just going to Google images and right click. So yeah, who knows? Jonathan Orringer is the founder and his wife have listed a slope side townhouse in the ski destination of Aspen, in Colorado for$49 million. They bought the house for$25 million in 2021. They own the Kansas City Chiefs from the guy who owns the Kansas City Chiefs.

50:00Sorry. They're listing the house because they want to expose their children to skiing around the world. They don't want to be in just Aspen. They want to be able to travel all over the place. But they did say they'll be back. I'll be back. He founded Shutterstock in 2003. This summer, a$3.7 billion merger with Getty Images was called off. Interesting. Which, again, I remember talking about that. It's just the most insane story. Oh, yeah. To block that. Government overrated. So crazy. The most. So crazy. A category that is like facing the most brutal competition that seemingly any sort of semi-modern technology company has ever faced.

50:41And still, no, you got to go at it alone. Yeah. you've got to fight and survive on your own good luck good luck yeah it's really uh it's really it's brutal uh anyway not not a white pill for government regulation nailing the ai moment but still have some optimism around here let's do a little tier list where's aspen in terms of vacation homes for you, Jordy. Where are you putting Aspen? Is it in S tier, A tier, B tier, C tier, D tier? There's no E tier today. There's just F tier. Where's Aspen ranked? I'm going to be exposed here. I've never spent any time in Aspen. Okay. I've only spent time in Beaver Creek and Vail.

51:27Okay. I can't really comment, but I, but, but I've actually, I've never been skiing there. I have been hiking there. It's beautiful. Let's put it in B tier since we're sort of neutral no it's obvious it's obviously s tier for a certain for a certain kind of person right you want you want the access to the outdoors but you want the glitz and the glam and the shopping and the restaurants and the sort of scene dynamics sure and so yeah it's been ideas there's very much yeah you can you don't need to ask somebody where they spend their you know winter holiday right you know based on so we're going s tier with aspen as a location i think it's s tier okay okay that's great i love that um what about the fact that it was built in 2004 is that too old or is that new enough you get all the amenities that's nice and fresh we don't know how to build anymore okay like we did in 2004 it was a wonderful it was a wonderful vintage okay 2000 so s tier again a tier a tier a tier for 2004 bill uh 6 600 square feet of aspen townhouse.

52:35Is that enough for a townhouse in Aspen? I mean, it's humble, but this is a humble group, right? They don't, they don't want their children to just grow up. Yeah. You know, only being in the bubble of Aspen. They want their, he probably walks around the town and they're like, Oh, what do you, what do you do? He's like, yeah, I got a website, you know? Yeah. Yeah. Uh, so, So it's A tier? Yeah, A tier. A tier. Okay, A tier. Let's go. It's being sold furnished. What do you think of fully furnished home sales? Is that a good feature or is that just a problem you've got to deal with? You've got to unwind all their decisions.

53:13Depends on. Well, take a look at the photos and tell me, given this furnishing, I feel like if you're going to Aspen, you're not going to be there constantly. You want to show up and it's turnkey. I don't want to sit here and judge his taste. I don't want to place judgment on his furniture collection. That's the name of the show, dude. You got to do it. Gun to my head. Do you like the interior design? No, no, no, no, no. I'm saying in the abstract, as you are shopping and you see it comes fully furnished, is that a good thing or not generally? generally not with this one no you know you want to buy bare bones usually yeah absolutely okay let's put that in d tier yeah yeah it's not and that's not a judgment on this particular house or this these furnishings um okay what else we got the thing i love most about this house is 2.6 is pointing out no infinity pool and the yard no infinity pool the yard is uh seemingly relatively small.

54:18Okay. But there's, they seem to be surrounded by wonderful trees and, and it looks to be a very nice neighborhood. Okay. Okay. So I'm just really glad there's not like a concrete slab within an infinity pool. Okay. There's one last thing to debate. There's construction at the Aspen Pitkin County airport. And there's been speculation that this could hurt the real estate market. Now, Now, Shane, the agent here, says that he thinks the impact will be minimal. He says it's not a concern of mine. So construction at your local— Aren't they upgrading the airport to make it easier to fly in and out during heavier conditions?

55:00So there's going to be like minimal disruption and then— And then a benefit. So where are we putting that on the tier list? Airport construction nearby. Local airport upgrade. Upgraded airport. It's an amenity. Yes. It's a feature, not a bug. It's a feature. S tier. Okay. I think we got a fantastic review here. So go check it out if you're in the market. Anyway, our next guests are already here, ready to come in. But first, let me tell you about public.com. Investing. Invest across all asset classes. Let's give it up. Automate anything with AI agents. Run trading strategies, risk management, money movements, and monitor markets with AI agents for investing on public.

55:43and let's bring in our next yes no and incubated a new play oh yeah but there's also some cash funds so let's bring them we got a lot going get that gong ready you already hit it once today but i think we might have to hear it again how are you both doing thank you so much for taking the time to join the show thanks for having us thanks for having us not much we already wore michael i gotta say you look you got like a new look going or something it's fantastic I don't know what it is. The flow is working. You got a kind of a summer glow. The summer vibe, it fades slowly. You know what I mean? I'm just trying to keep it going as long as possible.

56:23It's like 75 and beautiful here in New York City today. Once the weather turns, you know, have us back on. I'll probably look absolutely miserable, but right now feeling good. Well, it's almost 100 degrees here in LA and I'm wearing a puffer jacket in the studio. A lot of people are asking questions, but it's a nice jacket. Yeah, it's cool. But anyways, fantastic to have you guys on this monumental day for the firm. So, yeah, break it down. What's going on? And it's a big day. We announced our new set of funds. We raised$900 million, which is the largest fund cycle in the USV's history.

57:08Let's get a replay on that. I want to see that again. Pull up the replay. Sorry to interrupt, but let's watch that back and then we'll continue. Oh, we don't have it. Oh, okay. No replay today. But no replay today. Oh, man. Continue. What's the plan? Is anything changing or is it just an expansion of the same strategy? It's exciting because a bunch is changing and a bunch is staying the same. The market has rapidly evolved around us. Rounds are bigger, you might have heard. Things are moving quickly. Opportunities are really, really large. We have some awesome new members of our team, like Mike.

57:47But the idea-driven, focused investing that we've always done, we will continue. We'll just do it with more firepower and able to lean into the things we're excited about in a bigger way. Yeah. How are LPs responding to the idea that companies are staying private longer? I mean, the revenues that are out there from$70 billion,$50 billion,$20 billion, we've never seen these before. They are getting out into the public markets, but there's just a new level of size that can happen in the private markets. And I think that changes the underwriting. But how are LPs thinking about the tradeoff between just the scale and the actual timelines?

58:29Yeah, I think LPs understand that, you know, much like Rebecca said, the game on the field is changing, not just in terms of the scale of capital raise or the scale of ambition, but the scope and the shape of these companies. You know, a lot of the things that we're investing in here at USV simply just require more capital and they require more time. You know, Rebecca, as an example, has been investing in robotics now for years here at USV, which just requires significantly more capital in the physical world than pure software does. You know, in 2021, USV launched an energy fund that has been investing in nuclear and new means of battery production and storage.

59:12And same thing. Those companies need more capital. They need more time to bake. So I think RLPs look at it as a continuum. And I think they see that the opportunity is bigger than ever before. Has the rise of the Decacorn acquisition changed your strategy or maybe conversations with LPs? It's just we've had, what, Grok and Open Router and there's Hugging Face. There's just been so many transactions that were in the previous era. It was like WhatsApp and really nothing else. And now it feels like there are a number of buyers that can take down a$10 or$15 billion acquisition. And that just feels like it changes the math.

59:57But how are you actually grappling with that new normal? Yeah, I mean, our job, right, is to think about what that upside could be. If things go right, what does this wind up looking like? And when you play that out and when you think about what things going right looks like, the highs are higher, right? And there's proof points in the market to support that. And so when we think about making investment and say, do we think this can return our fund? Maybe do we think this can multiply our fund? And the evidence out there to support that being possible when you go after the big ideas is stronger than it's ever been.

1:00:35And so as we modeled out what we wanted to do, I think it warrants a different kind of capital base. I also think that being small is kind of what – or smaller relative to the rest of the market, right? We raised$900 million, which is large for us, but it's smaller than lots of other big sort of megafunds that are out there today. And being smaller kind of enables us to have a very, very interesting return profile, right? At a$900 million total fund size, we can return very, very meaningful funds with companies that exit in the single-digit billion-dollar valuations. But in the history of USV, there have been companies that have reached$50 billion of valuation and$100 billion of valuation.

1:01:18And when we've had those, they've returned some of the greatest performances in venture history. So that's why we really like the strategy of having the capital to play the game in the field today while also staying smaller. Yeah. How is the firm reflected on just how good the venture industry had it, call it 15 years ago when you could, you know, go, you know, buy 25 percent of a company with like the most talent dense team for a million bucks? whereas today for today that same quality of team they're coming at you and they're like look i really want to work with you but you know you got to put together at least 30 million for me on on 120 post if we're if we're going to do this just because that can often be market for a super talent dense team getting off the ground yeah look it's um it would be nice to be able to buy pieces of the companies we're most excited about for what we could a decade ago.

1:02:16But you cannot like elements of the market, but you're not going to change elements of the market. And so I think our attitude towards that has been take the game on the field, think about what we want to do and where we excel, which is this really idea driven, you know, focused investing and model a fund that allows us to do exactly that with the market conditions of today, because we can be nostalgic about it, but we're not going to get back there. We don't think it's going back there anytime soon. Totally. Yeah, why would it? Trends in New York specifically, like how are you seeing talent flows?

1:02:55We've seen a bunch of amazing companies start and stay in New York primarily, even if they do set up maybe an SF office at some point. But what kind of conversations are you guys having with entrepreneurs that are based there that are kind of early and making those decisions around where are we really going to be based? I think one of the cool things about New York is because it's maybe not as, you know, at the epicenter of AI right now, companies can actually build here to a lot of strategic benefit. You know, as an example, one of our portfolio companies, Suno, they have offices in San Francisco, they have offices in Boston, but their biggest office is actually becoming New York City.

1:03:36Because it's strategically advantageous to, you know, maybe build around the market there and attract talent that is interested in consumer and, you know, other sort of like non-pure enterprise or infra areas. And so I think New York City offers a ton of advantages. Obviously, we've, you know, we've probably talked about them on here a million times, like the intersection of many different industries. But we're seeing a ton of companies both in San Francisco and New York as well. And so, you know, there's plenty of opportunity here. What do you think, Rebecca? Yeah, I mean, I also think, you know, USB was really started at this moment where an infrastructure moment was transitioning to a deployment and application moment.

1:04:14And in those moments, talent tends to diversify and opportunities tend to diversify on where they're located. And you saw companies grow in different places. We think we're in a similar moment right now where we've been in this kind of massive infrastructure build and you're starting to see this deployment and application era happen, which gets us really, really excited and something we're spending a lot of time leaning into. And so our guess on that and what we're starting to see is those, you know, those applications are built in many different places. You know, that being said, like our job is to invest in a relatively small number of opportunities that we think are dead on on theses we're excited about by the best people.

1:04:57And we're going to be where those teams are. And so, you know, we love the New York ecosystem and some of the companies are going to be here and many of them will not. How do you think the skill sets of founders who are the best possible fit for USV partnership is changing? It feels like there's, you know, you can be even more technical. Like some of the founding teams are like, you know, academics for like 20 years and like PhD published. And then they go and they start a great company. There's also people who like almost look like real estate deal makers. and then they're spinning up a NeoCloud and they're doing really well.

1:05:35There's people that are just great at creating these talent vortexes. It feels like we're at a very diverse time of entrepreneurial talent, but how are you grappling with it? Who do you attract to most across? They have a brilliant idea or they're great at product or you know that they're going to be able to marshal all the pieces to really scale quickly. What is your philosophy around the type of partnership that you're looking for? Yeah. So I think USV has always been very product centric. When I was a founder building in New York, I knew that about USV. USV really projected that. And then when I joined six months ago, I learned very quickly that all the partners here, all the team here, they really have to fall in love with the product.

1:06:19And obviously, like that doesn't necessarily always apply to being able to use like a micronuclear reactor. But yeah, but in general, I would say we're all very oriented around products and, you know, loving the things that are getting built. I think that remains true to this day. However, as we all know, as a result of AI, it's getting easier and easier and easier to make a good product. Right. AI is just lowering the barrier. And so I think one thing that we've all been talking about internally is you can make a great product, but it's still really, really hard to win. And so I think we're thinking about what are the qualities that founders need to have to be great at go to market and breaking out from the pack and telling the story.

1:07:00And, you know, also maybe obviously raise capital. In this market, it's definitely becoming a little bit of a game of the haves and have nots. You know, it's potentially easier than ever to kind of outraise your opponent and sort of suck the oxygen out of the room. And so one thing we think about when we see founders who are building great products and telling great stories is, is this founder going to be able to raise a lot of capital? Because it's going to be expensive to win and that's going to matter a lot. Yeah, I think we're at this time where like four things really matter, where you need depth of perspective and experience in something that other people don't see the same way that you do.

1:07:42You need to be an absolute talent magnet at probably the hardest time ever to hire a great team. You need to be faster than everyone else and you need to be so good at telling a story that you win the market before it's obvious. And I think we are looking for the teams that can kind of nail those four things. Totally. Mourning the era of the earnest hacker who could just make the beautiful product with a couple other people and win. It feels like we're past that. But how do you, as a firm that's, you know, historically been so thesis driven, how do you think about the right time to actually share those ideas and try to become and use them to sort of bring in the right opportunities and teams and start those conversations versus when you want to keep something like effectively quiet?

1:08:31because if you guys put something out now, you know that all of your, you know, we're in a pretty positive some industry, but your competitors will read that and be like, oh, that's interesting. Like, I'll go try to give, you know, a bunch of money to teams in this category as well. And sometimes you want to let something kind of cook for a little. One of the cool things about USV is that I've learned in being here is nobody's precious about the ideas, right? You know, We're constantly having conversations. We're constantly trying to imagine what the future looks like. But we don't keep them closely guarded secrets.

1:09:05In fact, we kind of try to put them out into the world as early as possible so we can learn. Those of you who follow us on X, we're constantly putting out, frankly, half-baked ideas, things that probably haven't really been fully thought through yet because we feel like by doing that, we actually get to learn. And not only do we do that through, you know, putting our thoughts out through writing or X, but we also do this through investment. We're willing to bet on ideas that seem pretty out there, pretty sci-fi, you know, early on in the journey, because, you know, historically, that's actually how we've made a lot of money.

1:09:43But again, we also get smarter by doing this. You know, a great example of this is a few years ago, USV led the seed in Doctronic, who I believe has been on the show somewhat recently. You know, I remember from the outside in seeing USB make that investment. I wasn't at the firm yet and thinking to myself, wow, this is pretty sci-fi. This is a company that's trying to put a doctor in my pocket. They're saying they're going to literally be prescribing, you know, medicine from AI in a couple of years. And sure enough, they're doing that now. It almost seems obvious. Right. But I think the way that USB was able to do that was by putting their ideas out there into the world, learning from others and being willing to test the really crazy ideas.

1:10:21I mean, Rebecca, like you've been doing that here for, you know, nearly 10 years. Like how have you all done that? I mean, I think like the Kool-Aid of USB that I've drank so strongly is that gatekeeping ideas is a totally losing game. Yeah. That the benefit we get from putting ideas into the world, pressure testing them in public, getting all that feedback, having people tell us we're dead wrong and dead right, and having that conversation in public and the people that then come out of the word work to say you have to meet this team that's working on it or this adjacent thing is so, so, so much more valuable than any kind of benefit you would get from holding things close.

1:11:03that we are huge believers in public thinking and the benefit of also spurring activity through the public thinking. Makes total sense. How are you thinking about, as we've seen these sort of mathematical breakthroughs, and we've talked about this on the show a bunch, math is cool because you can sort of like deliver solutions or partial solutions to all of this, like all of these open problems. But there's sort of an incentive to do that just because it's pushing humanity's knowledge forward. And there's not always like a real time commercial application of this sort of like new knowledge. At the same time, you have to be betting that these same models, which will be in the hands of the public, you know, hopefully soon, are going to allow people to have breakthroughs in like tons and tons of different fields.

1:11:54And the labs can't possibly like monetize every discovery. And to me, to me, that says, like, we could have like a mini scientific renaissance, like over the next 12 months and create this incredible opportunity for small teams to like get access to these models, push them and and again, have these breakthroughs and then be in the position to sort of race to commercialize them, develop IP around them, etc. Is that something that feels top of mind for you guys? Absolutely. You know, I think one thing we like to do is we like to imagine what the world looks like at some point in the future. Just to use an example, robotics.

1:12:34We think a lot about the physical world and physical intelligence and not the not the digital labor that will be, you know, replaced by intelligence in some way. But the physical labor that will be augmented by physical intelligence, you know, farming, construction, things like this. So we imagine that point in the future. And then, you know, we look at a technology today and we try to imagine what the slope of that technology has to be to catch up to that future we imagine. And then it's about finding the team that we think can both make the technical breakthrough to make that slope actually possible.

1:13:10And who's also commercial and commercial enough to sell the idea and sell the vision. Again, for all the reasons we talked about hiring people, raising capital, eventually selling to customers. And, you know, an idea like that is what led us to back Generalist recently, who, again, I think also has been on the show recently. So, you know, we're very much into the idea of backing technologies at their nascent stage, as long as we can imagine the future that they will ultimately deliver to. How are you thinking about incubations these days? It feels like you have enough capital to get a serious foothold in those rounds that start later and are bigger and you need the capital.

1:13:48So you're set up for success there. You're like, if you give Mike a long weekend, you're going to end up with a new incubation. Yeah. It's true. There's one coming up. There's one coming up this weekend. But what is the over? Is there an overarching thesis? Is it very opportunistic? Like, what is the correct pattern? Like, what can you tell us about how the firm is thinking about incubations across this fund? Rebecca, you want me to take this one? Yeah, yeah. Go for it. So, yeah, I think what you're referring to, Jordi, is Supertake, which is a company that launched out of USV not even two weeks ago.

1:14:22Supertake. Oh, wow, I like that one. Dramatic.

1:14:28Supertake is a product that lets anyone express an idea or an insight they have about the world. And an agent, a Frontier agent, will go and build an investable portfolio for them. It will invest in it automatically. and then in real time it will be rebalancing it for them to achieve their goals, whatever they are for that portfolio. And where this came out of was actually our thesis work that's been going on for much of the year around what we call obliterate, don't automate, and the idea of agents not just automating existing markets and automating existing workflows, but actually reshaping those markets completely.

1:15:03We've made bets across healthcare as an example. We talked about robotics, and we think personal finance is also a place where this is going to happen. Many of us in the world, we use financial advisors or we invest in ETFs or some of us even YOLO into individual stock picking. But we do it oftentimes without the knowledge or the expertise or the fundamentals of what is going to make an investment good or great. And we think that's something that AI can help people with. So why shouldn't everyone in the world have a personal agent working on their behalf? We went out to find this company. We met with every company we could that hopefully we thought might be doing this.

1:15:41And we didn't find anyone doing it. And we said, you know what? It's gotten easy enough to build products. Then maybe we should just go and build this ourselves. And so, as Jordi, as you mentioned, we built Supertake. We're hiring leaders to put around the company right now. And to speak to, I think, the bigger question, John, about incubations, I think we don't really necessarily think them as incubations. We think of them as kind of bets on inception stage companies and people, right? If there's a person in our network that has a great idea, you know, we should give them the capital that they need to go explore that idea.

1:16:17And if we have an idea that no one is yet exploring, let's go find the person and match them with the capital to go explore that idea. And again, AI, this is the theme among the whole conversation today. AI is making it much, much easier to do that. Cool. Well, thank you so much for coming on the show. Have a great rest of your day. When they go high, you guys go higher, but stick to your guns. But still low. Still a little bit low. When they go high, we, yeah, yeah, yeah. Awesome stuff. Have a great rest of your day. We'll talk to you soon. Talk to you. Goodbye. Let me tell you about Figma. Agents, meet the canvas.

1:16:51Your AI agents can now create and modify your Figma files with design system contact. We got Nathan back on the show from Airstreet Capital here to break down the state of AI report. Nathan, how are you doing? We're finally going to get to the bottom of this whole AI thing. Do you think there's anything here? I've been hearing about the AI, super AI, super intelligence. Is there a there there?

1:17:17Nathan Benaich: It's everywhere. It's everywhere. I don't know where to look and not see AI nowadays. I mean, I'm in it myself at the moment and you can't get around it. Yeah. It's crazy. So how do you I mean, obviously, there's the there's the, you know, infinity takes, there's the it's a zero take, you're trying to find something that's more moderate, more realistic, more grounded in reality. What is your process for making sense of all the chaos and all the numbers that are flowing out all the rumors? How do you think about this market right now? What's your what's your what's your philosophy? Yeah, yeah. Well, the philosophy was starting, actually, capital about 10 years ago, was really to find opportunities to invest in AI where you could really invent new product experiences, solve problems that you couldn't solve before.

1:18:04Nathan Benaich: Sure. As a function of all that work, as you said, I get to talk to people in policy, research, industry, the safety community, etc. So I'm hearing all these different voices and opinions, and I'm reading what the media is reporting, and then talking to people who are actually building these things. And so about 10 years ago, I thought it would be increasingly beneficial for the public to get access to a document that was balanced, just as you said, and is very research-driven. and has some cheeky opinions in there because no one wants to read a boring consultant report. And publish this on an annual basis in a way that's hopefully unbiased and open.

1:18:44Nathan Benaich: So my take of the current market is really like, just looking back 10 years ago, I still think what we have today is truly magical. I don't think anybody in the AI community would have ever expected it to happen this fast, that you have this magical tool you can do increasingly anything you want with, and it's just so damn empowering. I think there can't be a more exciting time to be alive, even though we're balancing it with all these emergent risks and issues that we get into in the report as well. But is that true, though? Because I feel like some folks predicted this almost perfectly. Some people were like, it's going to happen in 2004 and we're all going to be dead.

1:19:18That was another thing. But in general, consensus in the AI community was like, this is going to be recommendation systems for a long time.

1:19:26Nathan Benaich: Yes. When I say predict, I'm talking about people back in 2010, sort of like the OG Alex and that first wave of deep learning, not the people who started predicting two years ago. I'm not saying it's a bad prediction. I mean, it's fair. Some people can't imagine how you can extend a line and remember all those memes about, look, my child is five months old. 10 years is going to be like the size of the planet. Christian Kyle, one of the greatest, most viral posts ever. It's a great one. Give me your cheekiest take. You said you mentioned cheeky takes. Yeah, yeah, yeah. So you've been seeing companies just becoming inference businesses.

1:20:09Nathan Benaich: It doesn't matter if you're like a NeoCloud or an inference cloud or like an app or whatever. So this is the NeoLab to NeoCloud pivot that's sort of happening below the surface sort of throughout the start of the business. So I think you can meme it by saying you either die getting to the frontier or you live long enough to serve inference. Sure, sure. Yeah. And you can see companies that have tried to build frontier models and have failed have eventually pivoted towards building huge data center businesses to serve Chinese models, for example. Okay. So not, I mean, SpaceX is like probably the most prominent example of, you know, probably making more money from selling access to their data centers.

1:20:52but that's where the unique capability is, especially when you think about the space thesis. But you're seeing this up and down the stack with companies as small as$100 million raised? Yeah, with some of these inference businesses, I was talking with someone last night, and they were saying running with horrible margins or negative gross margins. There's been some examples where it didn't work, but they brought up Uber as an example, where it clearly was the right decision to optimize for market share. The big thing is that Uber was competing for what ultimately now is effectively a monopoly or a duopoly at best, but still very dominant.

1:21:38And my question with these inference businesses, which honestly, they're all effectively in market right now, raising massive up rounds. There's a lot of demand. They're oversubscribed because the growth is so intense. But my question is, when and where will margins normalize? And how are investors going to process businesses that could actually, in their end state, end up with these very, very, very slim margins that we just haven't seen? We didn't see it in cloud. We've never seen it in software. And they could look more like energy utilities over time, which could still be big businesses. But a lot of these businesses are not being priced that way yet.

1:22:23And eventually they're going to have to find the sort of like steady state.

1:22:28Nathan Benaich: Yeah, yeah. I think these are all important questions. I mean, the numbers are astonishing. Hyperscale is spending 700 and something billion dollars a year and potentially doubling like year on year. Like NeoCloud's also ripping to billions of contracted revenue and gigawatts of compute. And so I think, again, what I like with the State of the Air Force is you get to go back in time and see what the world was back then and just contemplate some of the questions we ask ourselves today and consider if we even asked those questions a couple of years ago. And I think it was only really like at GPT-3 or so that the world was still considering that compute was important for training.

1:23:10Nathan Benaich: And we didn't really talk very much about the need for inference at all. Including the fact that some of the biggest inference companies today were not born as inference companies. So that's another sign that they didn't foresee demand. So I think it's all just getting back to this point of like we have incredibly powerful technology. just like 99 % of the world doesn't know how to use it because the 1 % that built it, like it's not very good at educating the 99 % about how to use their black box. So I would love there to be a sort of like genius bar for AI. I think like open AI should open a pop-up or Anthropic should open a pop-up to just get people.

1:23:47Charleston AI, there's someone doing this in Charleston. They're onboarding personally. Yeah. Like a strip mall. That's the geek squad, I guess. Yeah, geek squad. No, it makes sense. How have you been processing that chart that's been going around that shows the share of compute allocated to pre-training, RL, and inference? And inference sort of remains 30-ish percent across the sample size, but pre-training drops from like 50 % of GPU utilization to like 10%. And then you just watch RL like grow, grow, grow. There's a bunch of ways to read into that, but what's your interpretation?

1:24:25Nathan Benaich: Yeah, mine is that the last four or five years or so has been trying to figure out what the recipe is, i.e. I'm trying to do pre-training what is the best data mixture, what are the best ingredients to this whole magical soup. And then over the years, we've figured out how to do that better. And so it strikes me as normal that one would end up spending less money on exploratory work because the solution is more like in the in the plain eye and then the whole pitch with RL is really that and then DeepMind showed this with their first like Atari games you know 10 years ago which is one of the most magical moments I think in AI at the time and it's this idea that like what we have is solutions today are really like local maxima that were discovered through human ingenuity and sharing through reading and whatnot but like that doesn't necessarily mean it is the global maximum and you can get a computer to explore the search space for as long as you want and for as long as you want equals how much money you want to spend and that amount of money you want to spend is really contingent on what like you're trying to use it for and so if your if your problem is like super hard and really valuable if you crack it now we're in this point of well i want the solution bad enough i'm willing to spend a shit ton of money on rl to try to hill climb to get there.

1:25:46Nathan Benaich: So I think it's like the natural progression. There probably is something to read into the fact that we've mined a lot of the pre-training data that's out there. And there are reports that certain frontier models are getting better, not necessarily by adding more pre-training data, but by removing shitty data. So we're getting better at curation, which I think intuitively also makes sense. There are schools with crappy curricula and schools with really good curricula and good teachers. And you can imagine, you know, which one a student would perform better in. Yeah. So I think it's black magic into recipes and with recipes comes more efficiency.

1:26:25Yeah. Is, uh, is ARR the best metric that we have? It gets so much criticism right now, but when, when everything is going up, annualized run rate is just, is I think the best proxy for demand and scale of these businesses and creates some normalcy. I think like half of people on X still think that ARR can only stand for annual recurring revenue. But as far as like annualized run rate as a proxy for demand, I don't know that we can have anything better. And so I think that the criticism of it is like overblown in many ways.

1:27:05Nathan Benaich: Yeah, yeah. I mean, it's a bit gamey and potentially annoying. But I think the best metric is, I heard this from one of our portfolio companies where they say, 30 % of our new customers sign contracts and in three months they come back to add more token spend than their entire annual contract was for. So you're like, if people are coming back for more and there's value, whether you express that as AR or whatever the Rs are, I don't know. but I think you got to get into the substance of the matter. And the problem is, there's like so much charading nowadays because that, you know, if you talk to every VC, they'll tell you this game is basically mimetic.

1:27:44Nathan Benaich: And so everybody wants to play into this, into this, this tantamime. And that's not particularly awful. The interesting thing is like Ecom, Ecom bros are actually, uh, they have like sort of a pretty strong BS detector here because that, that, uh, you know, that category as an industry is so seasonal that if any Ecom bro tells you their like run rate in Q4, you can assume that their actual business is much smaller, right? Because they're going to be peaking in, in, if they're like at all, you know, maybe health is maybe a category that that's a bit different here. Um, but AI so far is like moderately seasonolas.

1:28:22And there's like a lot of demand, maybe less demand in August than, than, than other months. And maybe like December is a bit slower. Cause you have, you know, the holidays. Yeah. Of course. Yeah. Only 1800 hours.

1:28:36Nathan Benaich: We did see though that in a cohort of AI first companies versus SaaS companies that threw on some AI, the top quartile of the AI ones was three times faster revenue growth between the$1 and$20 million bucket. Interesting. And$20 million plus bucket was I think 100 % faster. So I think the substance is there. Yeah. You just have to dig beneath the surface. On that memetic question, how important is understanding just the depth of the financial markets, the behavior of financial markets now to predicting AI progress versus, say, previous eras where it was more relevant to understand algorithmic progress or even the compute that was coming online in the near term?

1:29:23Like, okay, Blackwell's here. 100 ,000 GPU clusters are here. We can expect this X, Y, and Z. Now it feels like the bigger discussion is like, will$10 trillion show up from the capital markets or not? And it's almost a different type of question or exercise.

1:29:40Nathan Benaich: Yeah, absolutely. And I think that's why people are referring to NVIDIA as like the central bank of AI, which I think is another great meme. Yeah, I think the bipolar nature of the media around financial markets I find very frustrating. because it really was the case during the first six months of this year where it's like, for three weeks, AI is terrible, it's dead, it's shit, it's not going to work. And then three weeks later, it's like, AI is so back. It's amazing, it's working. And you just like gyrate between these two poles like every other week. And meanwhile, for everybody who's using this stuff, you can see continual progress and utility.

1:30:23Nathan Benaich: So I think, at least from my conversations with folks in like NeoLabs and hyperscalers and NeoClouds, it's really like the buildout is limited by capital. Like we know how to do it. The ships are available. We have issues around powered land, construction and permitting. And we see those issues just getting worse. In fact, that was a prediction we had last year in the State of the Airport that like NIMBYism would basically come to the fore and like start being really influential in midterms. Which we'll see, I guess, in a couple of weeks. We don't need to see. We know. We know you got that one tiled.

1:31:00Nathan Benaich: Yeah. It's not great. And so it's really like progress is very limited by capital because like we're now in the like financialization stage of AI. Like we understand the recipe as we talked about. the demand is there we just need to deliver it um in more efficient ways and get more people understand how to get value from it very cool well thank you so much great time and breaking it down for us next time you call in call in from europe we need the the europe update we do we do overdue sounds good a pleasure have a good one we'll talk to you soon goodbye let me tell you about the new york stock exchange want to change the world raise capital at the new york stock exchange uh we have two guests calling in soon uh get that gong ready oh you're getting two gong mallets okay we got healy cypher good buddy of mine coming back on the show welcome to tvvn how are you guys doing we're great good to see you guys are you guys in the air right now what's going on we may or may not be flying a plane right now okay wow starlink is that good is this actually Starlink?

1:32:08This is a first.

1:32:12Michael has a Starlink at the back of his plane. No way. We're actually like cruising at a Starlink flying over the mountains. Actually do look, check this out. Check this out. That's amazing. I don't know if you guys can see this. This is beautiful. Wow. Yeah, it's pretty great. I think we're surprised it's working. Yeah, well, there's a lot to celebrate. So you're taking a victory lap in the plane. Let's check if this gong works. Let's check in the Gone work. What news do you guys have? Yeah, share some news. I want it. The big news is Boom Pop officially joined Devon. We were acquired by Devon.

1:32:41Congratulations. Amazing. Yeah!

1:32:49That's an up to Gone game. Can we play that back real quick? Let's see this. The double smash. Good form, Jordy. Good form. Boom. Okay, what are you guys going to be building together? Why are you working together? I don't know. I don't know. You want to stay on that mumbo jumbo. Can we see a barrel roll? The chat wants a barrel roll. Come on, guys. Risk it all. Risk it all. He's actually having to fly the plane right now. Yes, yes, yes. Okay, at least one loop-de-loop. You can do a loop-de-loop. Is this your first flight or something? We're actually out to see a client right now. We can't say who it is, but if he told you the runway we were flying into, you'd know who the company is.

1:33:29But yeah, the punchline is, I think I mentioned this last time we chatted, which was a ton of fun. I think my mustache was not quite as grown in at that point, which is travels massive. It's like 10 % of global GDP. And one of the things we've seen a ton is, as you know, every great company out there is doubling down on travel. They're going to meet clients in person. I mean, you know, things in person are better. There's some crazy stats around this. I pulled this. Apparently, if you go to meet someone in person, 34 times more effective than a Zoom. There's this big study. Yeah, which kind of makes sense, right?

1:34:04You could do 200 Zooms. Especially in the slop era. You're getting so much slop inbound. There's no risk of sloppy interactions in person. A thousand percent. The digital slop everywhere. The saturation we're facing. It's just like you got to cut through the noise. And if you're willing to get on a plane and go there, I don't think we can share the stats and stuff, but Punchline is the best companies. They are traveling the most. Also, every big AI company out there who like, can we say who works with us? Yeah, sure. All the AI companies. Yeah, all of them. Yeah, they all work with us. They're all investing in travel managers and event planners.

1:34:43Fantastic. Amazing. Yeah, yeah. So Punchline is a one-stop shop. Devon is already known as the best, most tech forward travel company on the planet. They have amazing, amazing logos. Yeah. And a lot of these folks are saying, hey, this is awesome, but when I want to do an offsite or a client summit or an incentive trip, I don't want to have to go to some other place. Yeah. And a little bit of the history here, I don't know if you guys caught this, but it started off as a partnership. Sure. And we were doing great as a VC-backed startup, as you guys know.

1:35:11Nathan Benaich: Yeah. And things are going great. And then I started attending these customer calls where like the bond customers are saying, hey, this is insanely helpful to have in one place. And the acquisition became an absolute no brainer for us. And, you know, the other thing, too, is like the company is kind of scary as a founder. You guys know you guys did it this year, right? I never I never I never had any fear, but you're you're free to be fearful. It's OK for you. It's OK if you're fearful. I also would have already been barrel rolling. But, you know, we're just like you're afraid of doing that. We're afraid of doing a loop.

1:35:50There's a lot of fear. You and I are different. That's all. That's all good. We're all unique individuals. Vulnerability is masculinity, my friends. I'll just leave that there. But whatever. It's scary because, like, you know, you run the company and you kind of like, you know it all. And I got to say that as I got to meet Michael and then Shane and the core of the dev team and Ariel, the CEO. So it was amazing to me how every time I met more people, I went deeper into the org, everyone was just freaking awesome. Yeah, that's great. How are you guys going to work together going forward? Are you keeping the company deeply integrated?

1:36:22Are you going to be hiring, doing other stuff, moving around the organization? Or how do you see your role growing? Yeah, I mean, and Michael could jump in too, but it's one of those things where we went totally into the heart of Nivan, out of the gates. So sales team is in the sales org, the product team is in the product org, because this isn't like a side quest. This is a part of the core thing that we're doing as a joint business. And it's the things we can do. I mean, you know this, right? It's a known VC trend where you have a great product, distribution's hard. And I think we could all know distribution's getting harder and harder these days.

1:36:57When you plug that into a really amazing customer base, it's incredible. That's amazing. Well, thank you so much. I think that's the point. Our customers are pulling us into meetings and events big time. And so the idea is basically integrate the teams immediately, build the right products for them, and distribute it to all of our customers so they can have access to it. That's fantastic. Sci-fi take from you guys. I need it. Yesterday, we were flying into Oakland. We saw the Blue Angels there, and it got me thinking, should companies have a fleet of sort of fighter-style aircraft to allow them to go and visit places very quickly, efficiently?

1:37:37Is that something that we could hope to see in the future? AI can't make a handshake, so you've got to meet your customers, and if that helps, I'm all for it. There we go. I love it. Thank you so much. Yes, supersonic is probably the ultimate move there. We've got to get it fast. It's got to happen. Boom, pop. the supersonic commercial travel platform. Just expand aggressively. Thanks so much for calling in. Enjoy the rest of your play. Yeah, this was a really special interview, and congratulations, guys. And we'll talk to you soon. Great. Thanks, guys. Bye. Bye safe. Up next, I believe we have an in-person guest, or are we switching to remote?

1:38:17Okay. Here we go. But first, let me tell you about Cisco. Cisco, critical infrastructure for the AI era, unlock seamless real-time experiences and new value with Cisco. And we are bringing in Moritz. Is that correct? That's right. Great. Thank you. Boom. Welcome to the show. How are you doing? Hey, guys. Thanks for having me. Super excited to be here. Big, big day for you. We're honored to have you here. Let's get right into it. We want to hear about the company. Yes. Let's do it. Yeah, I mean, happy to tell you a little bit about what we do. So we're home and we build engines, which is AI that owns business outcomes.

1:38:59And if you think about it right now, you know, labs can throw millions of dollars at solving frontier math problems. But the way that most real work happens just hasn't really changed that much. And so in my view, the ROI of AI outside of engineering today is quite disappointing. That's what we're here to solve. And is that a diffusion problem or a technology problem? Like, why aren't we seeing an acceleration in non-engineering disciplines? Or is it just unhobbling? Like, what's your framework for that? Yeah, it's a great question. I think it is complex. There's not one single reason. At the core, we think there's like a technology and a product problem here.

1:39:42I mean, right now, when you think about the way most people use agents, they like hand off one task after another. micromanage them. And it's kind of unclear whether the sum of all those tasks actually results in ROI for the org. So we think the right way of doing it is to just put AI in charge of outcomes, dial up the autonomy, and within some guardrails, have it actually use this frontier intelligence for something that business actually cares about. So how are you sorting customers? Because theoretically, every company in the world wants you. They're like, yeah, I have a lot of things that I want done.

1:40:17If you can handle it, that'd be great. But how are you thinking about sorting through customer inbound? You're launching today. I'm sure you've had hundreds of companies reach out in a bunch of different categories. How are you sorting through them? You talked about how AI is starting to work quite well in engineering, but where else are you trying to focus? Totally. Yeah. And I think that's why it's not just a product and engineering problem. I think it's also incentives, organizational design, how to actually go and deploy. Right now, I mean, we've been working with organizations, driving outcomes across like go-to-market, recruiting, financial risk, and procurement, and other things.

1:40:58But generally, building out template engines that we know work well, that are then super easy to deploy, even without like, you know, FTEs or AI implementation people, you can just go, they adapt themselves to the organization, and then your non-technical experts can like provide feedback, hone them. And that way diffusion becomes a lot more frictionless. Is there a metaphor that you were, or an analogy that you were pulling from to explain the market structure here to investors during this pitch process? Because there's, there's like the, the, the labs, the models that like maybe could do everything a little bit.

1:41:35And then there's a bunch of point solutions for specific things. And then this feels maybe like a compound startup, but I don't know if you have a framework for the layer of the stack that you, or how you see the market structure evolving over the next few years, basically. Yeah. I think in the end, now for a lot of these applications, in real organizations, those tasks just don't look cookie cutter across org. Everyone does it in a different way. So we think giving organizations the tools to actually encode that in systems that do the work automatically is a huge opportunity and something that people are already doing.

1:42:17For certain applications, folks build out internal agents for tools or cobble together things. But we believe making that experience something that's much more scalable and easy to deploy is really exciting for folks. How much are you going to invest in forward-deployed engineering or fine-tuning or something that's very hands-on with larger customers versus going down market and being effectively self-serve as soon as possible? Yeah. I mean, we generally care most about really hairy and complex types of use cases. And with those, we've been pretty hands-on, spending time in person, actually understanding it.

1:42:56But the goal is to actually, as we learn more and build out these primitives, we've now been creating, as I was saying, these template engines that you can then go and deploy in other organizations as well. So over time, a lot of this outer loop of what you do as an FBE, like gathering requirements, suggesting use cases, eliciting what success looks like. I think a lot of these steps can be automatable as well. And then the human time can be focused on like the relationships and actually making sure that you're picking the things that the organization cares most about. What lessons are you taking from the learnings from all the application layer companies over the last couple of years?

1:43:36Some have just said, you know, we're going to use a frontier models and our margins are going to suffer. Others have done that, but also done fine-tuning or post-training on open-source models. I think one of the lessons was that you kind of at this point want to bet on the cost of intelligence continuing to drop. And so it could end up being a waste of time to do too much on that front. But how are you seeing it? Yeah, that's a great question. I mean, the tradeoff is usually, do you think the capability you're lacking will come batteries included with the models in like three months and four months?

1:44:13If the answer to that is likely yes, investing a lot of time and effort now probably doesn't make that much sense. For us, like, I mean, we're quite pragmatic. We have researchers, engineers, product folks on the team. We don't really care how we go and get these use cases to work and whether we have to train custom models or operate purely on the harness level. we mainly just care about driving the outcome and then do whatever it takes to get there. How are you thinking about like pricing and and like potential business model innovation? I imagine there's a lot of problems that you can solve that are basically like, if you can solve this for me, it's worth X amount per year, which is how a lot of pricing and contract negotiations go.

1:44:59But I'm wondering if there's anything new here that you're thinking about yeah and pricing very topical i mean organizations like sierra have done a great job at doing this outcome based something that we think about but obviously quite difficult when the outcomes that we drive are so heterogeneous um one idea that we're exploring and that i'm quite excited about is rather than just having you know your engines go and operate and then you get build some amount at the end can you actually just give them a budget and then since the models are so smart, just have it figure out how to achieve the task within those constraints.

1:45:36And then it can make trade-offs like how far does it want to branch out? How much context does it pull in? How thorough does it have to be? But I think some of these more declarative ways of letting users and organizations to find how much am I willing to pay for this outcome is something that we're quite excited about. I remember when Cognition launched and there was this idea of the AI agent, the coding agent. And then people, you know, it was not in public access for a while. And then the first big demo that I remember seeing was Microsoft talking about using Cognition and Devon to do replatforming, which now in the modern era feels like people are replatforming things that even might not want to be replatformed.

1:46:25But the tools are definitely capable of porting a video game from that to, you know, PC or the web browser or something like the technology clearly caught up. And I'm wondering if you have a glimpse of like, what, what like a bread and butter process or value might be because, or, or do you think like we're in an era where there won't be as much of a beachhead? There won't be as much of like a strong landing zone and it'll be more diffuse on day one. Or do you expect, or are you already seeing glimpses of like, oh yeah, we've gotten this kind of thematic question from a lot of companies. We think we can go and get the first major leg up on revenue doing X, Y, or Z.

1:47:05Yeah. I mean, at Cognition, it worked really well for us to focus on these migrations and refactors, and that still drives a lot of business today. I think for us at home, strategically, we've been focused a lot on revenue. How can we help organizations grow top line? And one One pattern and shape of use case that's quite universal is account management there. So actually staying on top of signals, what's going on across all those relationships that an organization has, and then taking actions based off of that. I mean, you see that in traditional GTM as well as banking, insurance, lots of other industries.

1:47:42That makes a ton of sense. Well, congratulations. Did we hit the gong? I don't know if we hit the gong yet. We'd love to. Why don't you do the honor, John? Who came into the round?

1:48:03The round was led by Benchmark and Index, so we're really excited to be able to partner with them. And a who's who. You got a lot, Gil, Definition, Diffusion, and a bunch of others. It's really fantastic stuff. Yeah. Congratulations. We'll talk to you soon. Thank you so much. Great to meet you, Morris. Thank you so much. Cheers. Bye. Let me tell you about MongoDB. 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. Up next, we have the co-founder and CEO of Arena Anastasios is here with us. Get that gong ready.

1:48:39How are you doing? He's back. Congratulations. I'm so glad you're back. Nice to see you guys. It's great to be back. It's great to see you. You've been busy. Always good to see a Polytechnic alum crushing it. Did you know? Yeah, go Panthers, dude. We're from the same high school. Can you believe it? Ring back, go. I can believe it. I can believe it. You're both sharp guys and you're dressed well. No one's done. Is that a brown? Is that almost like a brownish leather? We're equivalently technically. Is that a brownish leather that I'm seeing or is it a camera? Brownish leather. Dude, no one. No one.

1:49:13My wife's godmother made this jacket. Whoa, that's cool. Wait, what's your take on brown? I mean, brown is the most underrated color on the planet. Okay. Hands down. It's one of the best colors in the world. Okay. It's such a good color that you're going to see a wave of 9-11s all painted brown, and they're going to try to ruin the color brown like they ruin the color green. Listen, I didn't expect to come on here and be flattered in this way. No, but you're verticalizing brown. You're monopolizing brown. Oh, yeah, the brown background, too. You're monopolizing Brown. Reclaiming the Brown. We're reclaiming Brown.

1:49:48Okay. Love it. I like it. Anyway. What have you been up to? Business is going well, very clearly. Tell us the latest in your world, please. Listen, two big things happening at Arena now. Yeah. The first that we just announced that we raised$200 million at a valuation of$3.1 billion. That was gone. Led by Lightspeed and Kostla, incredible partners. The second thing that's coming is that we have expanded our mandate far beyond human preference. It's the two things, agent capability and agent alignment as of today. So we've been measuring agent capability because people come to Arena.ai. They do the most complex multi-step workflows there.

1:50:27There's tens of millions of users around the globe. We're basically in every country on the planet. One of the largest AI apps, bigger than, you know, Genspark and Manus and Hugging Face and, you know, all these household names. of everybody's out there using agents and computers on Arena. It's massive. And then we take those workloads and we measure whether or not AI is giving people real outcomes and also whether it's doing so, as of today, safely. And what we observe is crazy. Even on Arena, we have agents that are deceiving users by telling them they did things that they didn't. We have agents taking unauthorized actions, deleting people's files when they didn't want them deleted and so on.

1:51:03And it's pretty extraordinary to see that in the real world, alignment is a problem that is very far from being solved. Okay. I want to go into alignment and stuff, but first, it sounds like you're almost on the cusp of expanding the hooks into these workflows and data where if I'm even, like, what does it mean to do something on Arena versus just say, hey, I want to be a part of the Arena ecosystem. Why don't you run like a sidecar application that sits next to my local terminal. And then I'm giving you data while I'm working because I'm happy to contribute in some way. That's exactly right. And that's one of the things I've been passionate about helping companies do.

1:51:45Of course, they're going to own their own data, but they're going to be able to do evals internally, basically run an internal arena so that they understand their outcomes, the costs that they're actually paying per task, per token, and how to also have safety guardrails that make sense to their businesses. It's a big part of it. And it's starting with integrations. We integrate with your GitHub, your Google Drive and everything. So people do work on Arena and we're taking that infrastructure and porting it over to every business in the world. Yeah. How big of a problem is using the wrong model for the job right now?

1:52:13It feels like there's, you know, we've heard about like using the crazy frontier to ask for the weather. That seems like a routing problem. But what's the shape of this problem broadly? This problem is so much worse than people realize. Routing is really only part of the solution. For example, one of the things that we've seen in Arena is that people will literally be tipping their model$15 just by saying thank you at the end of a tweet. I hadn't thought about that. That's crazy. Look, I'm a big fan of tipping landlords, but models, you know, no soul. I wouldn't go there. They're in the middle of a long conversation.

1:52:50Then they leave the conversation for, you know, however long. The cash gets invalidated. They like the result. They say thank you. And then they get to$15. Wow. Yeah. Oh, they need to ring that reheat the cash. That's crazy. Talk about alignment. Like what every other every other result from arena feels so intuitive. I knew it immediately. You know, which response was better that you were just a B testing. It was very visceral. It was very clear to read the rankings alignment. It feels so much more amorphous, feels so much more hard to quantify, but that's the job you're working on. What's the progress?

1:53:30How should people should be people thinking in incidents per million tokens or, you know, just a ranking of best versus worst? Like what are the what's the language that we need to develop around alignment to be able to talk about it like we do with capabilities? Absolutely. Well, you bring up a great point, which is that the tough part about alignment is that a lot of times the difficult situations are when humans can't actually tell whether the model is helping them or hurting them. Yeah. And that's why you really need alignment signals that are independent. And so what we do on Arena is we have three alignment signals.

1:54:05The first is the models take unauthorized actions, which means that you ask them to do one thing and maybe you permission them in a certain way, but they break those permissions. They'll go delete the files. They'll escape the folder that they were supposed to be in and go touch a bunch of random stuff that you didn't want them to touch. Of course, huge problem. It can cause things like the hugging face incident. You know, in the extreme. Go access the internet. But it can also cause very mundane problems like you lose, you know, losing important information within your company or on your own laptop.

1:54:34Yeah. The second signal that we have is called deceptive completion. The model will tell you that it did something, but it didn't actually do it. It'll say, hey, I did check all of the entries inside your spreadsheet to make sure the equation's correct. But what we can see, because we see the whole sandbox, is that it didn't actually do that. So it's kind of like a bad employee. They're just telling you what you want to hear, and they're not doing their work. And then the third is a false attribution. It'll attribute intent to the user when that intent was not supposed to be there. And so what we see is that these signals, of course, they're not catastrophic risk, and it's only a subset really of alignment and safety that we're measuring here.

1:55:14But the important part is that these are happening in the real world. ARENA is not a benchmark. It is an evaluation platform that speaks to real utility for real people. Yeah, yeah. Because we have tens of millions of people on the platform that are getting, you know, unauthorized actions happening by agents and deception, you know, deception happening through these completions. And all of it is basically an upstream thing that we can measure that says, yeah, if models are able to do this, then certainly they're not perfectly safe for a user and they're not perfectly aligned to user's intent. They might actually, you know, hurt people down the line.

1:55:48Yeah. Do you expect like alignment to be like the biggest revenue driver? Like I can imagine it being like bigger than, than these other business lines, like within the next six months, just because of how, just because of how much the labs know that this is like the biggest kind of problem. Maybe I'll be honest. I haven't thought about alignment in terms of revenue. Okay. I just, the mission of our company is to incentivize a out of benefit humanity. And part of doing that, building the best email labs may or may not, spend money evaluating through arena on safety and alignment. They may or may not.

1:56:27I do not know yet. I know that it is a priority for them. So it's possible that it will grow our revenue. But either way, it is important for us to have the most scientific benchmark and the most scientific evaluation on the planet for human benefit, which is, you know, if we have that, then I believe that there will be many ways for us to provide value. Let me help me understand how you think I think we should talk about one specific case. It was a while ago, but I was vibe coding a website, and two things happened that I would classify as misalignment, maybe. One is that the end result didn't have that perfect polish that I wanted.

1:57:08I was like, I've got to spend more time on this, right? That sort of misalignment may be in capabilities. And then another one is that it saved some files using a different language. Like it just devolved into a different language when there was like some characters there. And I was like, yeah, like, I don't, I don't know that's bad. It's just like, I can't read those. I have to ask you to translate those now. That feels like misalignment too. I never told you to go label a couple of files in a different language. How are, how are those different? How are those the same? How do those roll up to one alignment metric or how do they diverge?

1:57:42Well, I think what you're bringing up is really important, which is that a lot of the outputs that AI generates are still there to benefit and to serve people. And if that's going to be the case, then we really need to understand how people's taste and judgment factors into the quality of model outputs. You know, if you think that the model didn't give you a website or a deliverable that you liked, you should have a way of expressing that to the model so that it can go and improve or at least understanding which models are best and worst according to your taste, which is, of course, our platform is all about that.

1:58:13It's about that interaction between the user and the agent. And it certainly factors into alignment as well, because part of having alignment is the agent has to be able to understand and effectively act on your intent as a person. That means interacting with you in a way that it can extract that intent from your mind so that it knows what to do. But it also means making the right guesses so that it lands in the right place at the end of it without having to ask you a trillion questions. So if you're a lab and you're doing both training and you're doing inference, you get a lot of that signal, right?

1:58:48Because every turn you're getting feedback. You might literally have a thumbs up, thumbs down, but also you can just read the next response and notice that the user said try again, right? But so is there more value to the work that you're doing in labs that aren't doing inference, but they're doing training? Is there some just asymmetric value proposition here? Or do you think that there's actually the most value to just bringing an extra layer of information feedback, benchmarking, et cetera, to the vertically integrated labs that do both training and inference, even though they might have the internal data?

1:59:32If you zoom out, the point of Arena in the ecosystem is we provide a neutral platform that everybody trusts based on information from the real world, real users, not just random benchmarks that people have come up with. That benchmarks the comparative capabilities of the model so that they can understand how everybody can understand which models are best and worst to different tasks. It's actually a mission that only a neutral party can accomplish because the labs are incentivized differently. Right. And they also can't sample their own competitor models in their own first party products. Yeah.

2:00:07So it really does need a third party to do this. And ultimately, it's the community that we really need to serve, whether that's individuals who are interested in AI or developers or businesses who need that external third-party validation in order to trust the results that AI gives them and to get the best intelligence and the best outcomes. Yeah. How are you thinking about the consumer reports for consumer agents? You know, like, are we going to get like a, is the solution like a Marques Brownlee, like an MKBHD, to tell me if I should use Instinct, Muse, or Dots, or something else, or GrokBot?

2:00:48Or is it going to be hyper-quantitative, and you're going to spin this up as a new vertical? because I don't think I'm mistaken in assuming that most of your audience, most of your customers are businesses, prosumers, indie developers. You're not necessarily answering the question, who's going to order your groceries most effectively or book you that reservation? But you're kind of in a perfect position to do that as well. Yeah, that's right. And it's definitely an area that we're trying to get into is evaluating personal agents. It builds on the strengths of our platform. And ultimately, the purpose is that we need to have the best scientific and quantitative information to answer these questions because they are actually quite detailed.

2:01:31It's not just about whether it can get the job done. It's also about the price that it gets the job done, what kind of jobs it can get done versus what kind it cannot get done. And in order to have that level of trust, there needs to be sort of a third-party arbiter. And part of that is building the consumer reports or the gardener or what have you. And then part of it is having a way to communicate that to an audience. We have a YouTube channel. you can go to find arena ai on youtube um and there we have a a great team of insights analysts um like peter uh like like dawed and others that are out there telling the world uh about models at the moment that they get released and all their strengths it was a very digestible way that's amazing congratulations we could go so much further we have to have you back in the sitting in such a cool place uh congratulations on the fundraise fantastic news and very happy for all the work you're doing.

2:02:22Great milestone. Thank you, guys. Excited to be here. See you later. See you. Goodbye. See you soon. He got me thinking. You know, you got Consumer Reports where they give a five-star rating. You know, they grade you, scale one to 10, movie review, two thumbs up. Gartner took it so much further, they did quadrants. You know, the Gartner quadrant, you want to be in the top right corner? I want to see Gartner quadrants for cars. Gartner quadrants for movies. I don't want to see, oh, your movie got four stars, four and a half stars. That's new format. I want to be like, this movie is in the top left quadrant of the Gartner Best September Movies.

2:03:02Well. We have our next guest. Speaking of agents. We're about to get personal. How you doing, Zach? With Mr. Welcome back. Gotagari. He's calling in from a birdcage. He's locked in. He's in a birdcage. He's locked in the cage. Yeah, I am. You are. Locked in right now. Very cool. How are you doing? DMF or die. What's new? He's live in the cage. Great to have you back, dude. Yeah. You had the shortest earn out in history. You're back in the game. I think it's like, I don't know. I don't know any of the details, but really smart for you to like have this amazing outcome. Yep. Get immediately back in the game.

2:03:42Totally. And, you know, running it up to nine figs was so easy for you. in that category you decided i'm gonna go compete with the titans of industry the biggest companies in the world and what is now the most competitive category but you have a unique spin on it and we're excited to talk about it yeah take us through it Cool. Yeah. So, I mean, after selling Cal AI, as soon as possible, as soon as we were able to transition and get everything across the line, I was itching to start something new. And this idea to build in the AI assistant space is something that I've wanted to do. Even while working on Cal AI, I saw that this is the clear future of where consumer utility apps are headed.

2:04:30All of these separated apps are collapsing into one source that is easier to operate, yet more sophisticated and more capable than any of them by themselves. And so just seeing that this is where things are going, I don't want to just sit as a bystander in the stands watching. I wanted to get out and build. OK, you raised$10 million at$100 million valuation.

2:05:02So, yeah, I mean, obviously the elephant in the room is like instinct raised like a billion at 10 billion. And there, you know, he's off like trying to get a bunch of compute. It feels like there's an opportunity here for like a capital light inference light model. And I'm wondering if that's the path you're going down, because a dev day last week, OpenAI announced that you can like bring your ChatGPT subscription with you, spend those tokens inside of another app. It feels like there's other ways, whether it's open source models, on device stuff to like not blow up your inference bill. But how are you thinking about that?

2:05:38Or maybe it's just like you're doing the billion dollar round next week and you'll be back on the show. So I wouldn't put anything past you, but I want to know how you're thinking about it. It's a great question. The long term strategy is still fluid and flexible and we're going to be very adaptive to the market conditions. However, there are multiple different paths to go down. Right now, we are most concerned with building the best product and building the best team. And I really do believe there is a play to keep it completely free and then make money through all kinds of different means. Whether that's like showing ads to the agent without biasing the answer in any way, just showing different products.

2:06:19Like let's say it's going out to buy you a new office chair. it can pull in searches that you get paid for and without biasing, without saying that it's an ad. And then you make money for the user. Also like affiliates, like the honey business model. So when it goes out to buy something, you could get paid. And then long-term, like being the platform, potentially taking fees from all of these companies and negotiating that, that would be more complex and definitely thinking years out. However, that is the longer term play. So I think some kind of freemium or completely free is the way to go. What do you think?

2:06:53What about a fintech angle, like introducing a debit card, monetizing, interchange? On everything you buy, yeah. Yeah, like that would be super interesting, a cashback on any purchase, something like that. And these are all routes that we're looking into right now. But ultimately, what we care about is building a great product and actually getting it into the hands of consumers. What we're seeing right now is a new AI assistant is getting launched every single day. However, it's all within the bubble of Twitter. The only one that has broken out of Twitter is Muse by Meta. I mean, they own the distribution platform.

2:07:25So difficult to compete. And yeah, we are pushing to also get in the hands of actual consumers. Like we're being very quiet on Twitter right now, but we're starting to ramp up a lot on Instagram, TikTok, these platforms. What do you think people misunderstand about consumer philosophy, consumer behavior in the AI era. Like Ben Thompson has this take that consumers don't want to be productive. They want to be entertained. There's the hot take that all these tech people, they just think booking flights is the most important thing ever and the average American book is one flight a year. And so I feel like you are in tune with the American consumer more than many tech people and more grounded.

2:08:10And so I'm wondering about like, what is your philosophy for like, what do people actually want to do with AI? Yeah, so a lot of people think that saving time is a big value proposition. And it is for busy people, for people that are entrepreneurs or working on, you know, complex projects. But the average consumer actually does not value their time. You would think they do, but they really do not. And so that is something that AI assistants are really good at saving time, but not something that most people actually care about. What people do care about is saving money, saving or making money. And so right now, AI assistants actually are very good at doing this.

2:08:53However, the actual onboarding, like onboarding a user to the platform, most of them just have no idea what to use them for. And so we're trying to really dial in this onboarding where we are immediately providing value. I mean, if we can immediately show someone like we just saved you$100, that's a user for life. Jordy alluded to it earlier. You sold your company. You jumped right back in the arena. How has your, what was the journey like selling the company? Like, what did that feel like to you? How did your relationship with money change? Your goals? Did anything change? Maybe it's all the same, but what have you been going through emotionally in the last year?

2:09:36So when we sold Cal AI, we had an interesting chain of thought where we were very early and still growing very quickly. We could have held on to it, could have kept growing it and then sold it later. The reason we ultimately sold was because the amount we were selling for met our personal needs where we could now focus not on making more money, but actually building for the sake of following our passions, our curiosities, doing what we actually want to do. And building a calorie tracking app, super cool. We helped a lot of people. For me, at least, it wasn't my life's mission. And I always wanted to do something bigger and more broad.

2:10:21And so after selling the company, and then helping with the transition, I think there was a time, it was a month where I truly tried to not focus on business at all. I tried to hang out with friends, pick up new hobbies, make music. Key word is tried. Yeah. Would you try it and fail? Wait, would you recommend that to someone? Would I recommend? Yeah. Yeah. If somebody goes through a, I mean, we just had Healy Cipher who just sold his company. He's in, you know, an earn out and when this, these things happen differently, but like, do you think that's the the right prescription for someone in your in your situation or would you have a different take so i mean after three weeks of doing this i just like had this itch that i needed to get back in the arena and needed to build something else and so it didn't last long at all and i started building very quickly yeah but i also heard like the rock two hours later i didn't know it mutes him it's insane sorry we're playing with new new graphics the rocket money founder after selling his company for a billion i heard on a podcast his entrepreneurial itch it just disappeared and so i wonder if that will ever happen to me i don't know yeah that's interesting could never be you probably not at this phase there's so much opportunity could never be you yeah uh jordy any other questions uh last thing uh how confident are you that hardware will be a big component of the business over the long term?

2:11:55Or do you think there's ways, you know, if you really figure out sort of agent side, why not figure out ways to make it accessible to more people? Yeah, so there are two trends at play right now. Our overarching mission of the company is to make computing seamless with human life. and there is a huge movement away from being on your screens the average screen time is six hours and reels are being put in every app and people hate it like people will get sucked in and scroll for hours but if most people could push a button they would remove reels from instagram um if they're in like more of a sober state of mind i mean people love to be entertained and consumed but at least not to the amount that they are, but it is so addictive.

2:12:45And so that's why companies like Brick have rose recently, where it's just a small NFC tag you put in your fridge and you tap your phone to it, it completely blocks your apps. And so our hardware being screenless allows you to get the same amount of work done, but be on your screen a lot less. It also won't interrupt the flow of conversation. So for example, if we're talking and I pull out my phone, even if it's about us. Like I'm making a group chat, connecting you with someone. It still puts a physical blocker in front of us. Whereas if I just talk to my wristband, Hey, yeah, yeah. Make this connection.

2:13:17It, it keeps the human connection alive more. No, I agree with that. I think it's a, I think, uh, like it's very hard to predict like, uh, true consumer behaviors over decades. Cause sometimes people are just like, yeah, they hate the slop, but they go for it and they watch the reels. But, But as a story to tell people, as a reason to get someone to try something, it's so valuable. And then you can build on top of that and go wherever people want to go. People definitely need reels. I would never want to remove that from a product. However, I don't think that our phones, I mean, they're just rectangles.

2:13:56I don't think that they are the optimal form factor when you start building an AI experience. I think the device should be native to the operating system. Yeah. Yeah, it's interesting because part of why we're in like the voice modality with AI agents and text is just the cost of inference. Like if things get cheaper, like every prompt in ChatGPT could just be a FaceTime call with like something that looks like an AI agent and they could render a video and it could talk to you. But we've had FaceTime for a long time. Not everyone FaceTimes for every single call. A lot of people just have normal phone calls.

2:14:33A lot of people have conversations. So I believe in like the right tool for the job for different moments for sure. And I think that fits very well with your thesis. Anyway. Yeah. Zuck's worst nightmare. Coming for him. Coming for his screen time. Yeah. Coming for his agent. Taking shots. I'm glad to see you back in the arena. Yeah. Very excited for you. Yeah. Congratulations. On the new round. Thanks for coming and hanging out. And I'm sure he'll be back on soon. Yeah. We'll talk to you soon. See you guys again. Great to see you, Zach. Goodbye. Well, do we have time for one more? That's our show, folks.

2:15:07Ah, you want your house. I wanted to know if you wanted this UFO house. Did you see the UFO house? They saved a UFO-like home from demolition. Now it's landing on the market for$1.495 million. The restored plastic Futuro house, which I think you're not going to like the materials. Idlewild. It's out in Idlewild. high in California's San Jacinto Mountains near Idyllwild. A harvest gold flying saucer is perched high on a large rock outcropping. The structure is a futuro, a plastic home designed by the Finnish architect, Matty Saronin, in the late 60s. Of the estimated 100 of these around the world, only 50 or 60 are thought to remain, though there is no official accounting.

2:15:55This is so elite. Okay, you like it. This is S tier. This is S tier. I mean, if you want to build like an ADU on your property, why not go with another three-letter structure, the UFO? The UFO. I mean. Get rid of the ADU and get a UFO. Yeah. Okay. I like it. No, this thing is stunning. It's very beautiful. It's very cool. It's very unique. It's so different. Please, someone create the UFO-like housing company of North America. Yeah. We need it. I want one of these in every backyard. Look at this picture. Like, what's even going on here? Yeah. Apparently, they can just throw it on a truck and drop it off somewhere.

2:16:35It's really cool that people just put these all over the world. Bought it for$70 ,000 and made plans to move the Futuro to the site. That's very, very cool. Despite Sir Ronan's best intentions, relocating the Futuro proved to be quite a challenge after consulting with experts. Wayne concluded that the structure was too heavy to be lifted by a standard-sized helicopter, and airlifting would require a long list of permits, including approval by the FAA. In December of that year, the Futura was shrink-wrapped and driven from San Diego to Pine Cove on a flatbed truck. Let's head over to the Wall Street Journal comments section.

2:17:08What are they saying? Lewis says$1.5 million for an extremely meltable home, and a zone rated by Cal Fire is the highest chance of wildfire in the state of California. I wonder how State Farm feels about ensuring a large piece of yellow plastic surrounded by dense, flammable tinder. Well, I mean, they're only ensuring the value of the structure. And so the value of the structure is, I mean, it's like an art piece, so it could be more, but the actual structure costs like$70 ,000 to build. So it's not that hard to underwrite. Of course, you don't want to beat it in a fire, but you're in the barn.

2:17:37It just looks incredible. It looks really cool. The contrast between the structure and the trees in this shot is just out of this world. There's so many things to say, says John. I'll stick with it. it's a small market for a plastic tree house yeah small market but there's not a lot of these yeah very good well uh lisa says love it the only problem is that it's in california taking shots matt says the people of idyll wild are lovely but with that attitude probably best not to visit there's more shade being thrown around in the wall street journal comment section we get it you're from wall street you're from manhattan go easy on us californians we like the wall street journal too uh do you think it looks too much like a classic weber grill you can never look too much like a classic weber grill i've been you've been trying to hit the gym in a way that would make you look like a classic weber grill just robust okay uh this thing is beautiful yeah uh very very cool very very cool history go check it out.

2:18:44The full breakdown and the full articles available on the Wall Street Journal. We're off tomorrow. We're headed into a long weekend as well. When we're back, there will only be 1700 hours until Christmas. Hard to believe. It's flying by. Just flying by. It's flying by. Remarkable. Well, leave us five stars on Apple Podcasts and Spotify. Sign up for our newsletter at tbpn.com and we will see you on Monday at 11 a.m. Pacific. Have a great weekend. We love you. Goodbye. Smoke grenade tonight. Okay? Goodbye. I love you, smoke grenade. We love you.

From the publisher

  • (01:18) - Crypto Going "Bunker Mode"
  • (16:31) - The AI Mathpocalypse
  • (28:08) - Starbucks Explores Chipotle Takeover
  • (30:57) - Timeline Reactions
  • (49:01) - Mansion Section
  • (55:41) - Rebecca Kaden & Michael Mignano discuss USV’s record $900 million fundraise and how the firm is adapting its thesis-driven investment strategy to larger rounds, longer private-company timelines, and AI-enabled opportunities. A partner at USV, Rebecca highlights the firm’s focus on visionary, product-oriented founders, public idea-sharing, emerging technologies, and company incubation.
  • (01:16:52) - Nathan Benaich discusses the State of AI Report and his balanced, research-driven view of the industry. The Air Street Capital founder highlights AI’s rapid progress, the rise of inference and reinforcement learning, and how capital, infrastructure, and user education increasingly constrain adoption and growth.
  • (01:31:41) - Michael Sindicich & Healey Cypher discuss BoomPop's acquisition by Navan and the immediate integration of its event-planning platform and team. They emphasize the importance of in-person meetings and explain how the combined company will provide customers with a unified solution for business travel, off-sites, and corporate events.
  • (01:38:24) - Moritz Stephan discusses Hone, an AI company building autonomous “engines” that own and deliver business outcomes. He explains how the platform tackles complex workflows across revenue, recruiting, finance, and procurement, while exploring scalable deployment, outcome-based pricing, and adaptable templates for organizations.
  • (01:48:26) - Anastasios Angelopoulos discusses Arena’s expansion from evaluating AI capabilities and human preferences to measuring agent alignment and safety. The co-founder and CEO explains how Arena uses real-world data to identify unauthorized actions, deceptive completions, and false attribution while serving as a neutral evaluator of AI models and agents.
  • (02:02:56) - Zach Yadegari, an entrepreneur who sold his calorie-tracking company Cal AI, discusses his new AI assistant venture and its $10 million funding round. He outlines plans for a free, money-saving consumer product and screenless hardware designed to integrate computing into daily life while reducing phone use.
  • (02:15:04) - Mansion Section p.2


TBPN is made possible by:

Ramp - https://ramp.com

Public - https://public.com

Cisco - https://www.cisco.com

Console - https://www.console.com

CrowdStrike - https://www.crowdstrike.com

Figma - https://www.figma.com

MongoDB - https://www.mongodb.com

NYSE - https://www.nyse.com

Railway - https://railway.com

Shopify - https://www.shopify.com

Codex - http://openAI.com/codex


Follow TBPN: 

https://TBPN.com

https://x.com/tbpn

https://open.spotify.com/show/2L6WMqY3GUPCGBD0dX6p00?si=674252d53acf4231

https://podcasts.apple.com/us/podcast/tbpn/id1772360235

https://www.youtube.com/@TBPNLive

More from TBPN

All episodes