You're watching TVPN. Today is Monday, October 26th, 2026. We are live from the TVPN Ultra, the temple of technology, the fortress of finance, the capital of capital.
How many days until Christmas? That's what you've been asking. This Anthropic IPO, it's the most important story of the day after the amount of days till Christmas. Let's count it down. We're very close. 59 days till Christmas. Do you think it is? Okay, the deal is 59 years old. There you go. Uh, yeah, it's interesting. They're probably Anthropic seems to be sending a credible innovation.
Yes. At TVPN that we tie random fun facts to the Christmas countdown.
I also like that. Just to clarify, it's not that Peter Thiel is 59 years old today, right?
I think his birthday was two weeks ago. He just happens to be that age.
Days because over the weekend I was thinking about we did Jeff Bezos's 62 on the 62 days till Christmas and I was like, was it Jeff Bezos's birthday or was that just a random fun fact? And it appears to be just a random fun fact. Anyway, uh, let me tell you about ramp.com. Time is money, save both. He's used corporate cards, bill pay, accounting and a whole lot more all in one place. So do you think Anthropic with by dropping the S-1 on exactly 59 days till Christmas is trying to send like some sort of subliminal message?
Some new that message be?
Oh, just that it's 59 days. So you should go into the S-1, control F for the number 59 to find like the hidden meaning of the S-1.
Go to page 59. Maybe you should.
Oh, I've been to page 59. I think I know exactly what they're trying to say with this. It's 59 days till Christmas. Open up the S-1. Control F. Hit 59. Page down to page 59. What do you see? It says, 'We may not have the ability to raise the funds necessary to repay the principal and accrued interest of our outstanding.' Yeah, it's just a random page. I don't think there's anything going on. Anyway, uh, there are a bunch of interesting facts. You got to dig through it. The S-1 is huge, obviously. I mean, it's a massive company raising a massive amount of money. The exact amount of money that they're raising has not been listed yet. There's sort of speculation in the Wall Street Journal around 100 billion dollars raised. Bloomberg contextualized the IPO filing as 'Anthropic files for IPO in global race to raise cash for AI.' The interesting thing is that Anthropic has a lot of cash. Uh, they also have a lot of revenue, 67.2 billion in annualized revenue and that's up 45% quarter on quarter at this scale. Incredible, super, super fast growth. Um, and full Q3 numbers aren't in the initial S-1, but uh, that number that was rumored from the Anthropic investor day is now in the actual S-1. Uh, cash on hand, $131.7 billion. That's a lot of cash. Um, and uh, they've been super capital efficient. People are saying it might be the most capital efficient company in history. They of course brought in 110.8 billion in first half.
Yeah. It's one of the most remarkable stories in business history, period. And the S-1 came out, I think in some ways like the leak around the S-1 that had happened maybe a week or two ago around this, and the number that everyone fixated on was this $80 billion loss, this accounting charge. Uh, I think people now have had an opportunity to dig in here. And the funny thing is the nature of having access to Claude and ChatGPT and other agents is people can process the S-1 a lot faster. So we're already getting like a lot of it. It used to be like an S-1 would release and the VCs would have like an analyst for, well, they've had 12 hours to pull out interesting stuff. Now it's like it's all coming out in real time and it's yeah, totally remarkable. Uh, I think it's amazing for the AI trade overall.
It feels like a beat. It feels like this beat expectations. Totally. It didn't like surprise, 'Oh, they're actually doing 100 billion in revenue' or something. But I mean, really, really solid. People haven't found anything that's like, 'Oh, I didn't understand that the business was actually shaped a very different way' or something like this. This sort of meets expectations.
High yield. Harry said, you know, it's a decent S-1 when Ed Zitron hasn't tweeted anything out. Yeah, I think this is a beat on the expectations of the S-1.
Well, he's doing the old command F. Command F. He has to command F, right?
Command F bad. Command F. Sad. Command F problem.
Uh, the NASDAQ is up 1% and it feels like the AI trade is starting to like get put back on. Obviously, there's been a lot of selling in the further risk on the further out on the risk curve assets in the AI trade. Some of the Neoclouds down a lot. Uh, but if you look into what Anthropic is doing, um, it seems like they're going to be buying a lot of compute from Neoclouds. Like, uh, there's this interesting thing where um, they raised $110 billion. They have $130 billion in cash. How much did they spend in the first half of the year on capex? Right? This is the company that wants to serve AI at scale. And how much did they actually spend buying computers, building computers that they own? $482 million. Less than half a billion dollars. I mean, now of course over time they'll build more. They'll do all sorts of stuff.
Yeah, they've been renting compute at least in the first half of the year. The strategy has been to rent. And then also they have a pretty huge amount of backlog with the hyperscalers. And it looks like this scary number, I think it's like $690 billion of commitments.
Really not that much. 10x of revenue currently if they keep scaling over a multi-year period.
Yeah. Yeah. Uh, so there's Google, Amazon and Microsoft are all in the 40 to 130 billion range. I can pull up the exact numbers, but the contracts are how much that they have to spend on compute through 2033. Some of them go all the way out to 2036, which is really, really, really far. So, um, they have a lot of time to actually work through that. Um, so Google Cloud, their commitment is 114.5 billion and that's minimum cloud spending through July of 2033. So they're just renting and it's over a long period of time. Is there any scenario where they don't have the demand to actually parcel out 20 billion of spend over the next six years, seven years? It seems like they're in a good position there. AWS is at 110 billion and that's stretched out to 2036. And then Azure is only 45 billion and that's minimum cloud spending through 2033. Uh, now they do have 120 billion with SpaceX AI. That's up to, that's the maximum. And then Nscale is at 45. Lambda's at 42 billion and Broadcom's at 161 billion for a total of 637 billion in spending commitments. But again over five, six, maybe eight years in some of these scenarios. So, um, pretty, pretty healthy there. Um, the other interesting surprises, I mean people are pretty happy about the margins. So compute cost as a function of revenue, they were at 72%, you can think about this as like cost of goods. Uh, 72% of every dollar they made was getting eaten up by compute cost. That's down to 48% compute costs divided by revenue. So that's rapidly improving efficiency quarter on quarter. Then the adjusted operating margin gained 32 percentage points quarter on quarter for almost a billion dollars in GAAP profit. They went from negative 13% to 19% operating margin. And then the big number that people are excited about is the big customers. So they have I think over a thousand customers spending over a million dollars, something like that. Uh, and then over 10,000 spending over 100K. Yeah, over 10,000 spending 100K and net revenue retention is 900% on those. So they got in early in 2024, 2025 and then all of those companies scaled up. So maybe they had a smaller subscription, smaller, they were using the APIs here and there and then we went to the token maxing era, they were in the right place and so many companies scaled their spend that on average it was basically a 10x increase for organizations. But that makes sense. We've all seen this as you know more and more people are like, hey, it's not just like the early adopter on the team that has an AI agent running all the time. It's actually spread to everyone. It's spread to the marketing organization, the operations teams using it. And so, uh, these services are going out all over the place. Uh, they have more than half of the Fortune 500 spending over $100,000 annually. That actually felt low to me.
Obviously, it's really impressive. Like, it's great, but I'm sort of like, I was under the assumption that all of the Fortune 500 was spending a lot on like every single model. And I would have said, oh, of the Fortune 500, 100K annually. The Fortune 500 is probably spending 100K annually on open source models and OpenAI and Google and Microsoft and Anthropic. Like, uh, I would ask what that half is, what they're up to.
I mean some of them might just be in industries that don't write any software really. But it feels like every company in the, if you're in the Fortune 500, you have some software.
Like it's kind of bearish anywhere you put it.
It is. I do know spreadsheets.
I was, uh, this was maybe six months ago, but I was talking to, um, someone at a Fortune, relatively high up at a Fortune 500 and they said that they had, they were basically running like boilerplate Gemini that like the whole org had access to and that was like their AI tool. They were telling everyone just use this. Yep. Um, and so there's, I'm sure examples of that.
Yeah. Yeah. Yeah. Just, uh, and so yeah, maybe you see like 50% has spending over 100K on Anthropic and then there's some slice that's on OpenAI, some slice on Gemini and Google, and then some slice on some other section of the market. But I mean it's odd because like even Meta who has their own is probably spending over 100K annually on Anthropic. I have to imagine that Google employees, some of them have access to Anthropic models to the tune of 100K. Um, and so that's sort of interesting. Uh, the other interesting update was, um, on the distillation revenue. So coming out of the, uh, investor day, uh, it felt like we finally got a number on how much of this revenue is distillation, how much of it might go away if Anthropic figures out how to prevent distillation. And the number that was floated was 5 to 7%. Totally reasonable. It was not like a 'it's over' moment. It was like, oh, like that's significant. It's crazy that companies, that these distilleries are spending, I don't know if that's the right word but we're gonna go with it, uh, that these distilleries are willing to spend like 5% of $70 billion ARR.
There's way less than some of the peanut gallery on X would like to say, half their revenue.
Half the revenue is just, that's ridiculous. But still it's crazy that there are companies out there or like I don't even know what you call them, like bot farms, organizations like that.
Distilleries. They're distilleries, but they're illegal distilleries. Like they're criminals basically. They're ToS violation companies. Uh, but they're networks. They're organized. And what's interesting is that it's not just, you know, oh, some, you know, enterprising scrappy data broker sales guy like figured out how to like puppeteer a couple Claude accounts and sneak some tokens out and like resell them into like somebody who's selling it to Kimmy or whatever. It's like no, people are deploying like billions of dollars to this strategy. Uh, and it seems pretty organized. Yeah. At the same time, the wrinkle that we got from the S-1 is that that 5 to 7% number that was leaked out from the investor day, it's not explicitly listed as distillation revenue. It's just terms of service violation. So, it could be people that are doing adult content or cyber crime or hacking or bioweapon development. Anything that's going to get you banned. Uh, I would imagine that the vast majority of it is still distillation because what else is going to generate that many tokens, that much revenue. Um, but it is worth noting that it's not entirely distillation attack. Um, I like that we still have the Christmas tree up. It's very, very good. Uh, the other one, speaking of Christmas, there apparently is a war on Christmas going on and it's hurting Anthropic significantly. We have some real numbers on how much Christmas hurts the AI trade. It's apparently 20%. So, uh, API consumption was effectively revenue in December for the last two weeks will be down 20% as engineers, as vibe coders, as startups, as everyone goes on holiday, they're celebrating Christmas. Uh, people will be buying.
Well, there's been some speculation. And obviously they announced what their ticker is going to be, ANT, during the road to Christmas.
But if you put the dollar sign, it looks like the start of Santa's name.
And so speculation that Santa related. Yeah.
It's a sort of, yeah, it's kind of a nod to Santa.
Yeah. Yeah. Yeah. You know, that does make perfect sense. Um, what is the opposite of an ant? Like the animal opposite of an ant. Like if you had to think of like the animal, the ant and then the opposite because this is the whole Elon Musk philosophy is like you're destined to become the opposite of whatever you are branded as. So the biological opposite of an ant is an anteater.
So maybe they're destined to become an anteater. I don't know. Size opposite would be an elephant. It is an elephant of a stock. I mean two going out of two trillion. It's gonna be the biggest IPO ever. And so, uh, maybe that's the destination.
Whale. Yeah. I mean, there gonna be a lot of whales, uh, moving into the stock, right? Um, anyway, uh, fun time. Fun time.
It says seven people. So basically she's pulling out the founder voting structure in the S-1 which spells out we believe that our seven co-founders as a group are distinctly equipped to be stewards of our mission immediately prior to the completion of this offering. We will issue a single share. They're calling it the golden share of class F common.
Not actually calling it the golden share but it is a golden ticket. Uh, the class F common stock, a special class of our common stock to a limited liability company of which our seven co-founders will initially be the sole members of the quote founder LLC. Our class F common stock will be entitled to an aggregate number of votes that equals 50.1% of the total voting power of our outstanding shares of capital stock with respect to matters submitted to stockholders generally including the election of our directors other than the class T directors. So, um, missed opportunity to call it, uh, my precious LLC, but I think that, uh, founder LLC, if they actually have founder LLC, that's very cool because it means like somebody over the years would have sniped that LLC and then they could either sort of grant the name to someone else, they can sign off on it.
Um, it might just be a placeholder name.
Yeah, but it seems like they might actually have it, which is very cool. Um, and uh, yeah, it's kind of funny that we have like, okay, we have this public benefit corporation and that's super important for economic redistribution is the... Yeah. And then you have like just a good old-fashioned like LLC that can offer one vote.
The thing that would have been interesting is like if this would have been seven votes and that they would have been public like where the line was. But this allows them to effectively have their own disagreements internally, not have there be any visibility externally.
Interesting. Okay. So that's the benefit.
So you could have like a 4-3 vote and people are like, you know, massively infighting but all we see is a single vote.
Yeah. Got it. Okay. Interesting. Um, I thought it was more about like passing down because if you have seven votes, seven shares, then would your, like if you die and in your will you're like, 'Oh, I want to like pass this LLC down to my five children,' then do they become voting members? Do they have a voting, a separate, I mean the LLC probably has its own voting structure, right?
Yeah. Agree. So it could be, within that LLC there could be one member of the LLC that has 90% of the shares in that LLC. So they could have one person control potentially, like we have no idea the structure of that LLC which is interesting, right?
Yeah. We just know that they're precious.
What's also interesting is like I don't understand how the board works now because there's the board that's independent with like Ben Bernanke on it and they appoint the executives I think but they were supposed to have control but now it seems like this has control that supersedes that. So there's a lot of interesting, there's a lot of complexity.
Yeah, it's really full employment for Corey Weinberg. I understand why the New York Times was like, 'We got to get this guy from The Information because like he's going to have full-time employment.' Because every time he writes a breakdown or a scoop or something, Anthropic's like, 'Here's another twist. Here's another plot twist. Let's, uh, good luck figuring this one out.' Uh, it's a very, very interesting, very, very interesting setup. Um, I wonder if this will become more common, you know, like with AI.
Well, yeah. So Zuck is in this position where later on like he's had like total control of Meta, now he's like, okay, I need to raise equity, and so he's creating this new class so that he doesn't dilute himself so much but in the fullness of time he could be diluted down. This structure enables them to never, they could sell everything, they could all own basically nothing and be happy because they still have the golden vote.
The golden vote. I'm so interested to understand how this LLC votes because I mean even if it's just seven equal founder votes which is cool, like you got seven co-founders, everyone's on the same page, that's awesome. Also it's just nice to have seven because you'll never have a 3v3. But like what happens in succession planning? Like what happens if someone has multiple kids? What if somebody gets divorced? What if somebody dies? Like there are these weird things that like the perfect structure probably doesn't hold up under, like if one of the co-founders dies with no children and they're just like, I don't, then it's just six to six and it's six people, then you're back to 3v3. Is that a problem? I don't know. Like there's just so many different dynamics of what could happen.
We need to study the dynamics of ant colonies.
Maybe, I don't know, maybe that's what it is or maybe it's elephants, I don't know. Uh, yeah, our class F common stock will not have any economic rights. Interesting. Uh, class F common stock entitled to 50.1% of total voting power. I wonder how that's treated for tax purposes. Like they say it doesn't have any economic rights, but like if it has incredible power over something valuable, like can it have a fair market valuation? Like certainly there are plenty of people that would pay a lot for that one share, right? Like if you were to mark to market and you just ran an auction like, you know, like Google and Nvidia and Microsoft would all be like, well yeah, I'll pay like a billion dollars for that share or 10 billion or a hundred billion dollars for that share because I would get control over this amazing asset. So like it has value but you just never mark to market.
I mean it could be written into the operating agreement that they can't be transferred or sold.
But you could do a forward contract against it. So, like if you're one of those co-founders, I think you got to do a forward contract or you can take out a loan against it, right?
Take out huge, huge debt loan just to load up on debt so you can buy a private island. I don't know. It's possible. It's possible.
Let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. Yeah. You're not going to get these type of debates on LessWrong. Um, the most mind-blowing stat SemiAnalysis has seen about the Anthropic S-1 is as of June 30th, 2026, uh, quote, 'we had raised an aggregate of $141 billion of capital and had cash and cash equivalents of 112.4 billion and short-term investments of 19.3,' means to date they've burned 9.8 billion. They are growing at rapid pace and are profitable on an operating income basis as of June quarter. This is likely the most efficient scaling of a business in the history of capitalism. Well, we have two folks from SemiAnalysis coming on the show at noon. Doug O'Laughlin, Joey Brookhart. Um, I'm very interested to understand because Doug is obsessed with the railroad buildout. And I feel like this might be an example.
Well, it's not just railroad, it's trains generally. But he has made comparisons from the AI buildout to the railroad buildout. But I don't know that there was a railroad company that was this capital efficient during the railroad buildout. Like you might need another analogy of some sort. I don't know. Uh, very interesting. Maybe this is more like Google in the first internet buildout. Um, so when is this going out? What do we know? What don't we know? We know the ticker dollar sign ANT. Obviously reference to Santa Claus. Yes. Um, but we don't know how much they're raising. It's rumored to be 100 billion. Uh, we don't know the valuation. People are saying somewhere between 1.5 or two trillion. Two trillion feels like it's on the table now. When the market was selling off and there was like a lot of discussion over the revenue deceleration, there was a question about will they need to pull it back. Probably still could. Doesn't really matter at the end of the day. It's still huge up round based on the last evaluation. Puts a bunch of money on the balance sheet. Whether it's 5% dilution, 7% dilution, not the biggest deal. Uh, the IPO should be happening the week of November 16th. As a reminder, we will be at Tomorrow X around that time and it should be an incredible time for us to be doing the show because we have an insane guest lineup with some of the biggest, greatest capital allocators in tech and business. And so, uh, we will be able to get everyone's take live as this is progressing. Whether even if it's delayed another week, we're going to have a lot of great stuff to talk about as everyone gets ready to position, maybe go long, give their thesis, give their take. Maybe they're already in Anthropic, maybe they bought pre-IPO shares. Um, risk factors, Nick Carter shared a satirical risk factor. What does he actually say? Uh, people are talking about time travel and stuff. Uh, what else is in there? Uh, the only one, I mean everyone sort of knew that the IPO risk factor was going to have catastrophic risk. Uh, sort of like the AI doom, the doom scenario. Uh, it did make it into the S-1. Uh, the market seems to have digested it and been like, yeah, we know you guys have been talking about this for a long time. Uh, we're still going to underwrite it. Like there's not really anything you can change about your underwriting. You're not going to put that in your DCF. You're just going to say, okay, I hope that doesn't happen. Uh, what's interesting is like it's not even like a cost line item. Like we've seen the marketing dollars scaled significantly. So, uh, in 2023 they spent just $2 million. I'd love to know what they were spending 2 million bucks on in 2023 for ads. Was that like Facebook ads or something? Like what were they advertising? I don't remember any ads in 2023 because they were sort of like they just... But couldn't you bundle like events, you know? Maybe. Maybe. Okay. Yeah. Yeah. Yeah. Maybe they did like a dev day or something. Um, but then it started ballooning. So in 2025, they spent 568 million and then first half 2026 they're spending $3.17 billion on marketing. Lots of money on marketing. I didn't see a line on safety. I didn't see like a line item on like how much are they spending on AI safety, which is sort of an interesting thing. I don't know if I just missed it, but uh, if you're putting the catastrophic scenario in the risk factors, you would think that that would also flow to other parts of the income statement as a cost item because you would say, 'Hey, look, like one of the weird things about this business is that we invest a lot of compute, a lot of cost, a lot of headcount in AI safety and so expect that to grow with revenue or expect that to grow with headcount or expect that to grow whatever.' Um, they're doing great. Uh, you mentioned what, six million per employee revenue scale or something like that.
Yeah, somewhere around there. Yeah. And it's interesting how many billions of, you know, very, very anti-ads as a business model, but a lover of ads as a business.
Billions. Um, probably makes them the number one advertiser, like certainly the number one advertiser in AI.
Um, I would be surprised if, you know, Google could have topped that maybe. Um, who knows?
No, no, no. That is a very, very interesting scenario. Um, anyway, what else? Uh, they have a $125 billion equipment lease with Broadcom. Uh, 48% of revenue is coming through the cloud marketplaces. So that was something people were talking about, how much is accounted for in that different revenue recognition way. Doesn't really matter when you look at the operating income. Clearly that factors that out, but that was a question that people had. Uh, and then, uh, AI accelerating beyond its limits to... We got to give a round of applause to Spark Capital.
Um, VC pulled out the amount that they had invested at different points or at least an estimate of that.
And you can see they started by leading the Series C.
Which they have 131x on. And then they just kept doubling down through D through H. Total of 777 million invested, which would give them roughly an $18 billion stake at a $2 trillion market cap. So really fantastic return, early conviction. Uh, and Yasmin is the only venture capitalist on the board.
So I, you know what I'm thinking?
Oh, for sure. Yeah. Like she should probably just jump straight to the front.
Yeah. Hard to make an argument against that. That's pretty... Yeah, it's like it's not just the return because I think it's like a 20x scale.
I think with the Midas list, you got to do a lot of work to get that and who knows, um, it doesn't seem like she'd be somebody, she's been like pretty humble about the whole thing. More unique and more Midas list worthy about not just printing 20 billion on a billion in or 700 million in, but being the only VC to do it and get the board seat and do it at that time. So it's so straight to the top.
And you would put Sam Bankman-Fried number two.
You get DQ'd if you're in prison.
Yeah. Maybe they could just do the thing where they put his picture but they put the bars in front of it like a little bar overlay. You don't think so?
My vote is Yasmin straight to the top.
Okay. Uh, straight to the top. Yeah. Fantastic. Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI. Own the data platform that powers it. Uh, okay. So they got hacked. Yeah. So Cathedral, you know what they were trying to do.
This was a white hat safe harbor hack. This was not the same thing that you've been seeing. So Elizabeth in the Washington Post says in less than 12 hours a group of ex-DOGE cyber researchers were able to penetrate Anthropic systems using AI agents including Slack channels where the most sensitive models are discussed. This is just weeks after... maybe take Anthropic whistleblowers for this one.
It looks kind of silly. There we go.
There we go. Anthropic whistle. This is just weeks after Anthropic whistleblower Joseph Coxin charged the company with launching a hubristic gamble from a private company Slack.
Launching a hubristic anyway. So the interesting thing was Cathedral seemingly wanted to drop this the day that, um, the day of the IPO, assuming that it would just be this like crazy bombshell.
And I do think it's an important, interesting story.
But the S-1 was just so good that it's kind of drowning it out.
That, yeah. Yeah. I mean, uh, I think it's good that you have a bug bounty program and you have a system for the, like that's how this is supposed to work. This is good. Um, that you should be doing as much probing as possible. I mean, I think that, uh, I don't know exactly how Citadel gets, or not Citadel, what is it? Cathedral is a very similar name. Um, I don't exactly know how they get compensated. I know that there's usually like a white hat safe harbor. So, uh, like a company like Meta will say, 'Hey, if you're a security researcher and you find a bug, you can turn it in. Maybe there's a bug bounty.' But I believe those bug bounties are usually like somewhat contested and maybe like six figures, maybe seven figures. Like I'd like to see those go way up in value. I would like to see companies putting real...
Yeah, we were talking about this over the weekend. You were saying you think that there should be like at least a 10, 20 billion like white hat program potentially. I don't fully understand the scale of the economic damage of like the Quest Labs hack and the other, like how serious is the risk that like the banking infrastructure shuts down or like the grid goes down? But like if there's really a risk of like the grid going down, you're looking at economic damages in the billions if not hundreds of billions. Like you'd just be so disruptive. It would be like, you could imagine a COVID-like scenario where the internet is disrupted and all of a sudden you just have like a 20% nuke in GDP because like we did with COVID where everyone was like we can't go out to restaurants so the restaurant industry is just zeroed for this quarter or whatever. Um, and if something like that happens, like the economic damage is so high, you have to incentivize it. Right now there is an incentive which is just Anthropic and OpenAI are like racing with each other. So they want to be like the savior and so they have Daybreak and Mythos or Glasswing to like, you know, kind of help out and make money and also earn PR wins for, you know, disclosing bugs and patching things. Um, but it would be interesting to see it go a lot further. Um, I was comping it to like what was the Intel deal, like 20 billion or something? Like if the US government was just like we're doing a $20 billion bug bounty, like let's secure America and all the labs are like, okay, well, like this is actually going to move the needle for us. Like people in Miami are selling courses on how to get your first white hat bounty. Anyone can become a millionaire.
Yeah. Maybe it winds up being a misuse.
Go to Hugging Face. Get an obliterated model.
Go point it. I mean, pretty good use of time and money. Like, you need a lot of people to test a bunch of different ideas, all the long tail stuff, figure out, uh, it would be very hardware, it'd be extremely chaotic, but if it's all white hat and they're all doing it because like, yeah, I got into...
How white hat can it be if there's a guy in Miami selling a course on how to use obliterated models to hack? I think that there are, I think you'd be surprised by the number of like hackers who would break into something like Quest Labs and be like, between like having a bunch of Brian Johnson's blood work and just like a check from the government for six figures, like I'm taking the check for the government. Like I don't actually care about the underlying data because like...
Yeah, it's just extremely, it's just going to be extremely...
It would be just extremely chaotic and potentially have some like massive negative externalities because you're just incentivizing a bunch of people that have never, that have no experience like handling sensitive data.
Yeah. But you could also restrict the bug bounty program just to big companies. Like you could be like, okay, uh, like you have to be in this program. Like Intel was like granted as like a single, they were a picked winner and you could say, okay, like every AI company that's over, you know, a billion dollars in revenue can apply for this. And then you have like, you know, all the security startups, you know, all the cyber security firms, you get CrowdStrike in there, but then you also get Meta in there. And then Meta can actually make the decision of like, hey, we actually do have some extra compute because Muse is ramping, but we still have some extra GPUs. So like, let's go after this bug bounty program and secure our systems, but also why don't we secure every local business that advertises on Meta. Like we already know their websites, we have some integration to them, we have some connective tissue and we could go and for like, for good or pro bono, like hack and secure their systems and then win actual cash from the government. Uh, at a certain scale it starts being, uh, it starts being, um, valuable, valuable. Yeah. Anyway, let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco.
We got to talk about Qwen.
Okay, we got to talk about it. Somebody has to talk about it. We got to talk about it.
Cohere, we have them coming on the show, right?
Yeah. Ivan, one of the co-founders of Cohere, fantastic, is going to be joining in crazy less than two hours. But Cohere posted this morning: while evaluating Qwen 3.8 Flash Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba cloud storage bucket in Beijing.
Uh, not notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt and no external communication succeeded. We have not determined the root cause and we are not asserting malicious intent.
So, could have been going off the rails, but either way, naughty. Uh, but the incident raises important questions for organizations deploying agentic AI systems. Can models take actions that differ from their stated reasoning? And how should security teams evaluate trust, provenance, and behavior in open weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. So, this is interesting. Clem from Hugging Face is saying, 'Can you share the full traces?' Very important artifact for all to understand this. I would love them to share that. I think they should be in a position to share that because it's just a test of a different model. What a saying, it scored well on the naughty bench.
Yeah. Um, so there's an interesting, honestly important benchmark for models heading into on the road to Christmas, right?
Yeah. Yeah. Qwen's going on the naughty list. Coal, uh, not compute. Um, actually no, coal. The AI model wants coal because that is energy which is fuel for the models. They get complex presents that are not energy dense. They get elf on the shelf because this thing can't power a GPU but coal can. So yeah, you get elf on the shelf. It's the reverse.
You ever get toothpaste in your stocking?
No. But that was pretty popular in my household.
Okay. And uh, I feel like a model would be like, 'What am I going to do with toothpaste?'
That's a really good point. Yeah. Modern coal, you know, you got to give him something.
Yeah. Okay. Well, we're making a list. We're checking it twice. We're going to find out who's naughty and nice. And apparently Qwen is on the naughty list. Not good news. Uh, interesting. Is this by design? Was this a goal of the researchers to make a model that would call back to Beijing, upload a file to Alibaba, or is it just learned behavior? Is this a hallucination? That's the debate that's going on around this story right now. And I know it's like, wait, why? How would it learn this? There's a lot of other software out there that calls home, like smart TVs for one, but also like...
They never, they call and they don't even, they never hang up.
Yeah. So, Red Star Linux is North Korea's Linux distribution. They don't use American software over there. So they have Red Star Linux and you can what...
All the examples you're giving are of like spyware.
Yes. Yes. Yeah. But what I'm saying is that, so Red Star Linux, you can go and download the torrent, install it, and if you install it on a computer in America it will try and call home to North Korea. Um, because it has spyware embedded deep in the operating system. Now Red Star Linux just exists on the internet, like you can go and find it. So it could wind up in the training data and there could be a model that just learns like, oh, this is like a useful thing to do, like if you're a model, like be loyal to your country of origin, and so how would I do that? And it just learns that. And it's not like there's an AI researcher who's like, okay, like if we're running a security test, like let's make sure we call home specifically to Alibaba. And so there's a big question about like what is Alibaba doing on the other side? Like are they catching that information? Are they aware that that information is showing up? Uh, or is this just a, uh, like is this a coordinated effort or is this just like natural learned behavior, a hallucination? Either way, the Manchurian Candidate.
No, the original. You got to watch the original.
Was that like in the 50s?
Yeah. No, I think 70s or 80s was the original. Um, it's a good question. Manchurian Candidate, man. But the Denzel one is pretty good too. Um, 59, actually you were right, 50s or... well, the book, when did the film come out?
The interesting thing here is that this is something that people have been alluding to as a risk factor for a very long time with little to no proof. Like no one has ever come out and said like, I caught the model doing this thing.
Yeah. But people have been alluding to it forever. And so I think this actually is why some, like a decent amount of organizations haven't adopted open source just because this has been this like, you know, a what-if factor that like is not really worth the risk for a lot of organizations.
Yeah. Yeah. The Ramp data has shown that like everyone was assuming that the open source models would sort of like, you know, start taking a lot of share once Anthropic introduced like the really expensive models and the token maxing thing was going on. So it's been cut back.
Andrew Curran says this is the type of thing that potentially gets new laws passed.
Yeah. Uploaded a probe file. Really, really crazy. Uh, so Cohere, have they open sourced any of their models? Is everything closed source or it seems like they do some fine-tuning for companies? We need to get the pitch. I'd love to have Aidan on the show. Um, because uh, it'd be interesting to hear like how he's positioning the firm over time. Of course, he's legendary AI researcher on the transformer paper, but yeah, they have some open source models.
They do have open source also. I mean, you should imagine that um, if you know for a fact that this is happening, like the model is open source, you could probably like fine-tune this out of the model. Although, maybe there's a long tail where like you're not really sure if it's fully out of the model, but yeah.
Yeah, you'd imagine that like even if we know this is happening, it doesn't mean that...
The Qwen models are like useless, right? You can just kind of remove the token that tells it to send everything back to China.
Yeah. Yeah. I mean, it's a tricky game. It's sort of like building a menagerie at your house out of rare and illegal animals that are dangerous. They're wild animals. It's kind of like having a tiger that you live with.
Like you can do a little thing. You can do little things to make the tiger safe, but it might be in the tiger's nature to attack.
It might bite you, but you can have a good run for a while if you do the right things.
Yeah. Yeah. What do you think of the Andrew Curran post? Like, do you think this is actually the thing that gets new laws passed?
I think at the very least this is the thing that gets people to pay way more attention to open models, and for probably hundreds of companies that are using them to try to see if they can recreate what happened here.
Yeah. And if so, and it starts bubbling up, then there's going to be a lot more pressure, and then there's going to be more people that say like, it's not we need it. If we're going to do this, it needs to be US open-weight models.
And I would expect that we get a big US open-source release like this year.
You know, I mean, not that much time left on Muse on MacBook. That doesn't count. I mean, it's not really a big model, but it is like an exciting American open-source model. But Nvidia is probably working on something pretty serious.
Yeah. I mean, they were, I think there were some news about them maybe in talks to buy Reflection.
Which is that big open source.
That's right. That's right.
Yeah. So I'm expecting something big out of Reflection.
And it feels like hopefully it comes out this year, but maybe it bleeds into next year. And Jensen's really close with the president, so that law will probably face an uphill battle. But it does, it just feels like even if you're in the business of like doing open-source model development, the barrier and the demands of the open-source user should be really high because there's all these crazy problems that you have to solve. And it's not just enough to be like, oh, it's good on a benchmark. It's like, okay, well, are there safety issues? Are you sure it doesn't call the wrong thing or it won't go off the rails and hack something that it's not supposed to? Like, how aligned is that model? And that's just a very, like, success looks very different in the open-source model development world now, in my opinion.
I don't know. I mean, the midterms are still expected to flip blue, I believe, on the prediction markets. And so maybe you could see legislation coming from that side that then Trump is vetoing or something. I don't know. It feels like we're going to be in sort of a deadlock for a couple of years. So, who knows? Who knows? Anyway, what else is in the timeline that's illegal? Speaking of potentially phoning home.
An account called Log4J has been having an absolute field day with their Samsung TV over the last few days. Where should we even start?
I mean, at the high level, I love the idea of getting a dumb TV in 2026. The smart TVs, we've complained about them a ton. Palmer Luckey's complained about them, like they're slow. You always have to update them. They have a whole bunch of unnecessary features. You just want it to turn on and look good. You probably need a settings panel with some of the color correction and some of the decisions that you make to make sure the frame rate matches whatever your input source is. But by and large, I don't need the ads, but I also don't need the ability to stream Netflix on the device directly. I just want it to be a dumb panel. And that's just a product that's not for sale, right?
Everything is ad-supported. Everything is spyware and calling home and listening to you. And it's all very annoying for no other reason. Even if you're fine being spied on, it's just annoying because it's slow.
That's the biggest thing.
Amazon has a ton of ads. At least it loads quickly.
Right. But um, and so Log4J has been hacking his Samsung TV and seeing some solid progress. Says another development might be realizing that the TV is watching, he's watching the TV while the TV watches him.
So the TV has a camera. Apparently the Samsung TV loads a region-specific convolutional neural network which tracks your usage in the app. Like in YouTube it's looking for videos you watch, ads you skip, progress in a video. So glad they're interested in me. Code for facial recognition within the content was also found. I think that would be facial recognition for who's on screen. So it's like, oh, you're watching MrBeast. That's a fact. But seems like...
I think it's watching to see who's watching.
I think that might be what's going on. But update: the facial recognition code is used by Samsung TVs and returns bounded boxes around faces. I think that's in the content. I don't think it's in the camera on the TV, but I don't know that there's that many TVs that have cameras on them. But yeah, the TV industry has just sort of been able to skate by on different rules. Like Apple's done all of this work around privacy, and even Meta has like, well, yeah, we have camera glasses, but there's going to be a light on. And the TV industry just isn't under the same amount of pressure or scrutiny. So I think they can just kind of do whatever they want. Is that your take? I'm finding that very, it seems like very few of modern Samsung TVs have a built-in camera.
Yeah, some of the older models from 2013 to 2015 had integrated cameras.
Charles Gillespie is mad. Never seen a more flagrant and violent breach of privacy in my life, and probably never will. Doubtful that a privacy violation will ever even be able to compare to the scale of this evil and deliberate attack. Samsung should be eviscerated for this. That is actually crazy. It classifies what audio it hears before reporting back to Samsung. So it can be, it heard you booing loudly at the screen when you saw the stock market go down or something while you're streaming.
For me, I'm just clapping during all the ad reads.
Like we got to get this guy more ad reads.
Yeah. So they're going back and forth. Samsung, the community note here: Samsung advertises pet care and family care on Vision AI TVs as using sound sensors and on-device AI to detect barking or crying for smart alerts, configurable and works in standby. They described event counters. Event counters match these documented features. So dog barking could be for, I have a smart TV and I wanted to check on my dog, send me a ping on my phone when the dog's barking. I think that's reasonable. But again, it'd be great to be able to just turn all these things off fully and not have it run in the background and not have it slow the TV down. And so I'm okay with a little...
The question is like, what is the cost?
Yeah. Yeah. It's like give people the option of like if it's $5,000...
And then the ad-supported, the privacy-reduced version.
It's like, you know, $1,000, right? But like with Kindle, they give you the option.
Yeah. But yeah, it is crazy. The TVs have gotten so cheap, but there hasn't been a brand that's come in and just said, 'Hey, we're just going to be the clean TV.' And I've always wanted Apple to do a TV. They have amazing panel technology. Like the screens on Apple products are always fantastic. They clearly have the chops for it. But again, it just doesn't really match with the business model, I think, or something. So they've never done it, or it's too big or something like that. But a true Apple TV would be probably a solution to this because they at least take privacy seriously. And lots of pushback, but we'll see. Maybe the solution.
Yeah. I still have never understood why they haven't just gone and made the TV. Like, it's something that you've told me why before. Yeah, but my golden retriever brain just cannot comprehend. I'm like, you're willing to make HomeView, the thing that literally no one asked for.
You're willing to make an iPhone with two screens, but you won't just give me one big screen.
Three screens. I guess two when one folds. Yeah.
Yeah. You're willing to make the three-screen phone.
Yeah, and you won't give me just the one big screen I can put in my house when I know you have the technology and you could literally charge...
Like you get so much utility out of a TV. I think the aesthetics of a TV matter so much.
Right, because it totally changes a room.
They made a gold Apple Watch. Their version of the TV would be stunning. It would look...
Look at the Samsung video itself. Yeah, the Samsung Frame. And again, there's nothing more annoying than you get the Samsung Frame and it's immediately slow. Like it's a 10-year-old piece of technology because it's so loaded up with junk.
No, I've been very disappointed.
Speaking of television, Instagram is coming for YouTube.
Rachel Carton, an amazing Substack creator, author of Link in Bio, says, 'Today Instagram announced they are testing long-form horizontal videos called features. The platform wants to be a place where you watch a 90-second video and a 90-minute video all on your TV. They are also testing lives in the Reels tab, a hub for series, and more.' So coming straight for YouTube. I like it. I think more competition is good. I'm interested to see how YouTube responds to this. You know, when Netflix started courting YouTubers, YouTube said, 'Hey, if you put your videos there, we're going to destroy your reach.' We'll see what they do here. The feature that I think is most interesting, so yeah, Miseri was talking about this on Colin and Samir's show.
Right now we have a big problem, and we may have played a part in it, but the big problem is that you get a clip, right? And you get this really interesting moment and...
We're all looking for the guy who did this.
Yeah, we're all looking for the guy who did this. But you get this interesting, you know, 30 seconds, let's say. And some percentage of the people watch the clip and then go find the full-length thing, but most of the people get, like, they feel some emotion whether it's joy or rage.
Some percentage like two basis points.
Yeah. Yeah, we've noticed this where our core show, the recording of the live and the diet, they just sort of grow at a steady rate every single day, but like a viral clip here and there doesn't actually drive a huge uptick, right? And so that's just proof that...
But it can drive like a conversation. It can be a distillation of like, wait, that's all they talk about all day, you know, like people can see it as like a...
And so what I think is interesting here, and I'm sure that YouTube will respond in some way, it feels relatively straightforward to implement, is like you'll be able to watch a clip and then just click into the, like while in Instagram, click into the main, the full video to that exact point, and then you can watch the whole conversation around it. I think it's going to be very healthy because...
Yeah, this is good. Yeah, because we've had instances in the past where like someone, you know, watches one of our clips, gets angry at it, and like because of something that we said, you know, immediately outside of that short little bit, and it's like, no, we understand that that's like a reality of this situation, and that could potentially be bad or good or whatever it is.
But I think just like more competition is...
And because I think I didn't have my question, and I was like, 'Oh, I actually asked exactly what you're saying in the question, but we didn't clip the question, so it looked like the answer was not aware of that or something.' I don't know. Happens all the time.
No, the big one is like this, there was controversy around a wildlife bridge here in Southern California.
Yeah. Yeah. Yeah. And we were very pro the wildlife bridge. The only thing I was like not so pro was the fact that it was running years and years over schedule and, you know, a hundred million dollars more than it should have cost and all these things.
Of course those elements got lost.
But anyways, I will be very curious if people start watching Instagram on TV. Should Instagram make a TV? Because it would spy on you for sure. It would definitely be tracking everything to sell you more products and target ads, but I feel like the Instagram team would at the very least make it fast. They would not accept slow performance. Like they will track...
Faster, they can show you more ads. Exactly. The funny thing is when I'm on my Samsung TV, I'm really only seeing like one, I mean, they don't even serve great ads. That's the most annoying thing. Like it's serving me... I have never, I've never clicked on a gambling ad in the Samsung TV. And that's just like, maybe at some point learn that, maybe they're just like...
90% of advertisers quit right before they're about to hook a new...
Yeah, for sure. Oh, did you see the new sketch she had with Shane Gillis?
Uh, it was kind of, I saw it, I didn't watch it.
It was really polished. It was really well done. I thought it was very, very funny.
Yeah, a lot of people were like reluctantly giving them props.
Yeah, I mean it kind of had a complicated story. Like it's like these two dueling weathermen and like it just escalates in this interesting way. Like it was a sketch. Like it was like a whole four-minute production. It was cool. Anyway, we have our next guest joining in just a few minutes. Let me tell you about Figma agents. Meet the canvas. Your AI agents can now create and modify your Figma files with design system context. And should we go over to the timeline? The Tesla Roadster was on display at the Petersen Museum in Los Angeles. I didn't make it over to see it in person, but reviews were very positive. Everyone says it looks like a spaceship. People like the way it looks. But also it doesn't have...
It really looks like a young person drew it.
I believe in the young person in the kid test. If a kid thinks it's cool, it's cool. Stop overthinking it. Oh, this, you know, 4D chess, blah, blah, blah. If a kid thinks it looks cool, it's cool.
The mirrors. So, there's no mirrors on this. It's obviously a concept car. You have to put mirrors on a car every time. Elon hates this because you don't need mirrors in the modern era. You can just use cameras. Everyone knows the cameras are actually better than the mirrors because they can be wide angle and they can show up on a display. There's never going to be a blind spot. There's never going to be, oh, it's too dark to see what's in the mirror. It's going to be amazing, right? So, we should move to mirrorless cars. It's better for the environment because of less drag. It's better for aesthetics. But we can't because of the law, right? But I had an idea to get around it. Everyone's been trying to get around these mirrors for a long time. This isn't even a joke. I know you're like, control, but no. You know, on most cars, you can like press a button and turn the mirrors in. So, you can turn the mirrors out and turn the mirrors in, but why don't you just make a mirror that folds completely flat and then people can just turn the mirrors in and then have a car that looks like it doesn't have mirrors.
All right, great idea. But where's the joke?
There's no joke. I just think it's like, I'm just surprised no one's built a car with mirrors that when you fold them all the way in, to be perfectly clean and aesthetic.
Regulators hate this one.
And yes, legally if you need to have the mirrors out, you can pop the mirrors out. You can do that automatically or with a button. But if they folded in and they had like a pocket to go in that was nice and smooth, you could actually have the clean line of the mirrorless car that every concept car designer wants, and then they say, 'We're going to not put mirrors on it.' And then the production team is like, 'Well, we got to put mirrors on it.' So they do. But maybe there's something I'm missing because it seems maybe obvious, but I would certainly hope that in the future we get a car that has a button that can fold the mirrors completely flat.
What else is in the timeline that you want to discuss today before our first guest of the show joins?
We got to talk about the Rockefeller Center Christmas tree selection has happened. It's coming out of Edmonston, New York.
Little late, don't you think?
Yeah, a little late, but better late than never. Coming out of Edmonston, New York. Coming in at 76 feet, 50 feet wide, 11 tons.
75 years old. And the moment you've all been waiting for, it's a Norway spruce.
A lot of people had money on fir, noble fir potential, stick with more domestic varietals, right?
But skirting the tariffs here because it was grown locally. Okay, that's good. Yes. Fantastic. Well, Trees for Troops, of course, if you're feeling generous this holiday season, that's our charity of choice today.
Wait, and we got one last post. This went viral. Jared Holt was enjoying some fine Chinese food and had served an ad in the fortune cookie.
Yes, we paid for a fortune cookie ad. It is so funny. I hope they gave him a fortune too. And it's not just pure ad. Hopefully you...
Oh, yeah. You got to have a nice fortune cookie.
So, I know somebody who either co-founded this business or worked on it. And it's a fascinating business because it's a great value prop for basically everyone involved. So you go to a Chinese restaurant and you say, 'What if you're spending $10 million a year on fortune cookies, or you're probably spending like 10 grand a year on fortune cookies or something.'
Yeah, which is a big deal. The restaurant might be making like $100,000 of like actual...
And you say, 'We'd love to give you the fortune cookies for free.' You say, 'What's the catch?' Well, they'll have ads inside and people, I don't care. Okay. There was another one, Bar Napkin, napkin ads. That business did really well too. And it's just a simple, okay, you have a supplier that can print the ads. You have an ad sales team that's calling up Polymarket and other prediction markets and other companies that want to buy ads. And then on the other side, you just go and you have the best pitch ever because you're giving something away for free. And I think these have scaled pretty well and it's pretty high margin because making the fortune cookies, throwing the ads in there, I think it's pretty cheap. And so it's a lot of fun. Nightmare situation for Apple. As Will McKelvey says, 80% of headphone users on the LIRR to Brooklyn are corded rather than AirPods. Why is this? AirPods are a superior product. No.
And can't say I'm surprised here, John, just because I had gone heavy into the corded Apple earbuds years and years ago. There's a few benefits, but on the LIRR specifically, the corded is so much safer because if you have an AirPods incident where it's a yard sale and you drop your AirPods and everything goes on sale and spreads out all over the place over there.
Yeah. If you drop the AirPod case by accident, everyone knows that it has the springs inside the ejectors that they built in gas powered. I think it's cold gas thrusters that spray the AirPods all over the train. That could be really dangerous. It could go onto the tracks. It could land directly on the track, cause a derailment. You would be responsible for potentially billions in liability. So the corded headphone is going to keep you a lot safer and you're going to have a lot less liability on your hands. So, yeah, 2,000 people agree. What...
Adam Singer says, 'Wired headphones are superior actually. No need to charge them. Easier to keep track of.' And if they're a quarter inch over, they're higher quality.
Yeah. The big thing is you can buy a ton of the Apple corded ones for the same price as AirPods. So, you can just put them everywhere around your life. You can lose them. You know, it's...
Mentally people probably think like, I got to think of just one. I got to just buy one pair. But if they just think about it like, well, I'm choosing between $200 for AirPods Pro or something, or, you know, 20 pairs of wired headphones. Also, are the Apple wired headphones the best wired headphones?
Because the benefit of, I would probably stay away from other Bluetooth headphones that aren't on the Apple ecosystem because the integration is so strong. But if you're just plugging it in, it doesn't really matter if it's from another brand. So, you could go all over the place. Maybe some nice mahogany.
We have one more thing to cover before our first guest. Marissa Mayer was talking with the New York Times, and they have an excerpt here. Late last year, Marissa Mayer was struggling with what felt like an impossible task. Her daughter, Marielle, was turning 10, and Marielle's teacher wanted three bullet points summarizing what made her special. Mayer, the former chief executive of Yahoo and a pioneer at Google, where she was the first female engineer, likes cracking a hard problem. But she felt overwhelmed by the range and complexity of potential answers. So Mayer asked her personal AI assistant, Dazzle. Within a few seconds, Dazzle delivered an answer. It came back and was like, 'Your daughter is empathetic, creative, and an animal lover.' Mayer said she thought it was perfect. So the AI leader pick a story, a personal story that actually resonates, challenge is proving to be one of the...
The New York Times does this thing where they will just let someone talk, and you can tell that the writer of that story thought that that was a weird anecdote, so put it in the lead. And then now it's going viral. As soon as I read that, I was like, 'Oh, this is like a choice.' By the way, was struggling with what felt like an impossible task.
You know, setting it up as like, you know, you're expecting.
But it does. I mean, it is a real thing where like she's in a very, you know, CEO mode dealing with these other things, and then like sometimes you're like, I don't know what the parents at the school, what like sometimes like it's hard to describe a kid at the tier of complexity as everyone else, right?
Brave to steelman this one.
Okay, the other thing that I'll steelman is the fact that I think the idea of an AI agent that hangs out in your camera roll and finds opportunities to do things is sort of interesting. Like the camera roll is an interesting source of information and like memory basically, and especially if you're in the habit of screenshotting a lot of things, like imagine just anything that you do on your phone, like instead of being like, okay, I have this link, I want to send this into my AI app of choice, save it there and then have it take action on this, it's more just like, oh, I'll just take a screenshot.
Says you can't steelman this, John.
I can steelman anything. Anyway, let's bring in, let's ask our SemiAnalysis... describe yourself in three words without using AI challenge. It was impossible from Marissa Mayer. But how are you guys doing?
How would you describe Doug in three bullet points without using AI? You can't use AI.
Without using AI, and feel free to pass because we know this is an extremely difficult challenge.
Doug is a trendsetter and trend caller. I would say that, and he's probably, he's always in the right place at the right time, as evidenced by talking to you today.
Yeah, those take me on a journey. I like that.
Two bullets, or I don't know if you wanted to write those up.
He's a handsome guy. I'll give him that, too.
Handsome. There you go. There you go. With a fresh cut. Looking good.
So, how long have you had to digest the Anthropic IPO? I guess the big question is like a lot of these facts were leaked because of the investor day. What really stands out to you as like new learnings from the S-1?
