So we start with Anthropic's monster week.
We may be at the stage where we throw the humans under the bus, not the AI anymore, which I think at some level is pretty terrifying.
We move to OpenAI killing Sora.
I think shooting it in the head is even more significant.
A big part of the whole strategic direction of the company was flawed.
Agreed. You're seeing the economists, the accountants have wandered into the room and they said, 'We have a scarce resource here. Let's optimize it. Let's devote this compute to the people who can pay the most for it.' And then we finish on the man with the biggest balls in tech, Masayoshi. You haven't lived till you've seen an 85% decline in an index.
This is one where it's just backass backwards. I don't believe there's right or wrong in money. There's just money.
I just don't think raising it 5 or 8 billion when you're at 80 million or 100 million of suspect ARR is the most exciting accomplishment in the world.
Let me be direct. Get over it. You should conform your company around your customers and your model, not your VCs. Being mean to a billionaire is actually a feature.
Ready to go?
Boys, welcome back. It is this week in Anthropic, otherwise known as the SAS OGs, which has been renamed. I want to start with, you guessed it, Anthropic. Unbelievable 28-day month of February where they did 6 billion in revenue, which was more than Databricks has done in their entire lifetime. You know what I think was the most interesting news out of Anthropic this week? It was actually the accidental leak of Claude Mythos. Essentially 3,000 unpublished assets leaked. It's a 10 trillion parameter model apparently that is this next-level step-changing capabilities that they're not releasing because of how powerful it is. This is by far the most interesting to me. Jason, how did you think about this news?
Well, look, obviously it's embarrassing, right, to Anthropic to leak it. I actually just think we're going to see more and more of this accelerate. The faster we vibe code, the faster we ship, the more corners we cut in general on application-level security, it happens. I mean, so many folks are accidentally uploading code to insecure GitHubs, to databases, to Supabases that are by default open. So this is accelerating our data, which is just open on the internet. And you could say, 'God, this shouldn't happen at the Anthropic level,' and I'm sure someone will get scolded, right? I'm sure it will. But overall, this is accelerating, and it's going to accelerate even more as we let our AI agents make decisions. Our agents are going to decide where to put code. They're going to decide what level of security to use. And this is going to become happenstance. And you know, it's funny. I mean, people were like, 'Oh, how could Anthropic have a new security agent and have this happen at the same time?' I think it makes perfect sense. The Anthropic AI security agents, which I've basically used in Replit, are very, very good. And it also makes sense as we rush, we're going to leak source code, data, PII. Right? It was, I don't know whether it's happened, it was reported today, all of Meru's data leaked. It's being held hostage, all of it, every single interview, every single piece of PII, every single piece of humans. And so, you know, we used to mock these. I think it's going to start happening daily and weekly in the agentic era. And it doesn't excuse it, but it's a reality. Agents are goal-seeking, and agents are going to make, not only are they going to make the same mistakes as humans, they're going to work a thousand times faster. So even if they make the mistakes 10% as often, Rory, help me with the math. If they do a thousand times more productive, they're still going to make a hundred times more mistakes. So we're going to see it everywhere.
Again, just for perspective, because there's two things going on here. Anthropic, some data leaked from Anthropic about their new model, Mythos, which of itself is meant to be amazingly powerful in dealing with cybersecurity. And there was a whole consequence that we'll talk about in a second in terms of how that impacted cybersecurity stocks. But as Jason pointed out, the level of irony here is acute because it was an inadvertent leak. So you had the situation where a model that's meant to be amazing for cybersecurity actually leaks via a cybersecurity leak. So that's kind of, so we're toggling between the two on the cybersecurity leak. It was noteworthy, Anthropic, quote-unquote, blamed human error, right? We may be at the stage where we throw the humans under the bus, not the AI anymore, which I think at some level is pretty terrifying. But, and you know exactly what happened. You see this, I mean, not down in the weeds, you often see this where, you know, you're about to do a big announcement, you have your content management system, you stage all the assets, you know, be it their Fed press release or in the UK it happened on the budget, if you remember Harry, the budget, you have the press release ready to hit play the minute the budget is ended and someone inadvertently forgets and puts it on the public side in advance. It's the same thing here. So it probably was a human error with a whole bunch of content ready for, I don't know, pick a date, the March, the May 15th announcement of Mythos. They forget to secure it correctly and out it goes. So that's the first thing, right? So that's kind of the embarrassing part of it. And then the interesting part of it, and you really do have to do this not without sniggering, despite the fact that it all leaked, you also have to separately talk about the fact there's some big claims on Mythos, right, and on Anthropic we're making via, again, via this leaked memo. Reminder, no one else has seen the actual model, at least not publicly available. Obviously some people have seen it, but not publicly available. And even I was trying to get copies of the leaked memo. There's just a few screenshots at this stage. It's hard to track it down at least quickly anyway. But the statement is it's way more powerful. Second thing is it's going to be way more expensive for them to serve, and therefore it's going to be way more expensive for customers to buy. And then the third thing is a particular focus on cybersecurity. It's meant to be, quote-unquote, extremely good at detecting cyber issues. And the result of that was a four or five percent decline in the average cybersecurity stock last Friday when this leak happened.
Yeah, just two, maybe just two other things on the leak just for this tradeoff. You know, I'm dating myself, but when I was at Adobe and we were acquired, we were an early customer of GitHub. And so we were putting source code in the cloud, and that was banned in Adobe at the time. It was banned because the source code was their crown jewel. It was pretty easy to make a crappy PDF reader or a crappy image generator. But to do what Photoshop or Adobe Acrobat did, all the exceptions, all the tens of thousands of corner cases was the crown jewel, right, of the company. And so everything, we got the first exemption to be able to use source code in the cloud, pros and cons. But when they used this on-prem source code management tool, it took a month to do a release. A month, a month. Okay, now we're doing 60 releases a day, right? Or even Anthropic, fastest-growing enterprise company of all time, is still doing massive releases every month or two and dropping features every day, right? So we went to something that took 30 days at a tech leader to something that takes hours. There's tradeoffs there, and I'll take them, but we're going to see it explode in terms of like the stuff that was published today.
Going back a few threads on show number 50 to Rory. One of the things that I thought was pretty cool in Chyros was two things. Always-on background assistant that works constantly. Our AI is working with us 24/7, and agents that can sleep, wake, and self-resume. The autonomous agents, which I've been talking about how this is going to consume orders of magnitude more tokens and change our life. I'm excited to see more coming. And you know, OpenClaw was just this brief thing that woke us up to what Anthropic appears to be all in on, right? Truly autonomous agents running 24/7, hopefully safely, hopefully not leaking all of our source code, but it's coming soon, right? Not these, you know, this whole idea that we've been doing. When we started this podcast, you went on to ChatGPT or Claude. No one had heard of Claude when we started this. I was a quirky guy using Claude. And you talk to it and go back the next day. The next release, it's going to be on all the time. All the time. Debating Harry's latest investment. Was it big enough? Is he too concentrated in the fund? Where should he go? What was Rory thinking on that deal? Right.
Why was Rory abusing Harry by email again for the second time in a day?
Laugh, but this is the future. I'm excited to see it coming sooner when our agents are 24/7. Like they're literally around us, and we give up all of our personal freedoms and autonomy as part of it.
I hear you on the embarrassment of it being leaked and, you know, the human error element. But while Anthropic has Mythos, which is supposedly as powerful as it is, you're juxtaposing that with OpenAI fucking around with killing Sora, kind of ads not really working and people being unhappy with it. And it's seeming like this massive chasm of the progression of force that is Dario and Anthropic continuing faster and harder than ever with a fault-altering, confused and dazed OpenAI wandering around the product desert trying to find some water.
You're just being mean. I mean, again, as I said last, and I'm sorry to repeat,
Yeah, but again, narrative is overdone on both sides, right? So I think some parts of it are true. Obviously, you're true in a bunch of different things. The decision to shoot Sora in the head, right? Almost certainly a good decision. Look, it's obviously embarrassing to say something is going to be amazing less than four or five months ago and then shoot it in the head. But if it's a mistake, give him credit for at least saying it's a mistake. Move on, right? And yeah, that relationship with Disney again, I think I'm going to give, it wasn't me. I was sneering at it on real time when it happened. I think someone else in this podcast said it's really significant. Just saying, right? It's clock.
I do. I think it's massively significant.
I think shooting it in the head is even more significant.
I think it's saying that a big part of the whole strategic direction of the company was flawed.
The whole that we are going all in on consumer, that from what I read, Sora made single-digit millions of revenue, right? And was consuming a million a week, which actually sounds way too low, right? It must have consumed billions and made single-digit millions. It makes no sense as a product either in the short term or the long term. But if you want to own the whole consumer experience with AI, they decided we have to own image and video, and Anthropic never even attempted to do it right. So it's a massive retreat. It doesn't mean it's the right, it's probably the right decision to your point. In fact, almost certainly is. But man, that's a, our strategy was wrong. Like this is a huge own goal. Our strategy was wrong.
Agreed. But I, and I agree with that. But I still think that, as I say, I still think Harry's over, kind of overeagering it a little bit because look, you made a comment about ads that I think is caught effectively implying that the ad strategy hasn't worked. That's a bit of a bigger leap. I mean, Sora hasn't worked. They've killed it. I think I'm with Jason. I think that's smart because I think one of the things you're seeing right now is in a world of scarce compute and astonishingly, despite all the investments that we've seen in terms of actual available compute for people to sell AI on, we're in a scarcity mode. You don't devote compute to things that are highly compute-intensive and low revenue-intensive. I mean, and Sora was almost the definition of that. Video generation is extraordinarily compute-intensive relatively speaking, and the revenue is almost minuscule. Conversely, you know, codegen, while it is compute-intensive, is orders of magnitude less compute-intensive, and there's real dollars attached to it. So you're literally, what's happening right now, I actually think at a higher level, it's actually very healthy. You're seeing the economists, the accountants have wandered into the room, and they said, we have a scarce resource here. Let's optimize it. Let's devote this compute to the people who can pay the most for it. So that's the Sora comment. On the ads comment, Harry, it's early days for, you know, GPT ads, but again, I cite that quote that Brian Kim that I thought was really good. Of course, they're going to run damn ads because there's no other way to build a mass consumer business, and they have no choice, right? Because, you know, their consumer conversion rates run roughly 5%. That gets them to a, I think, a roughly 10 to 15 billion dollar consumer business, right? And, you know, out of their 500 million uniques or whatever it is. So one of two things has to happen in the consumer business. Again, I'm going to leave the enterprise business out on the consumer business. Either A, they take that conversion rate to a number we've never seen before from a typical consumer business. I think that's unlikely. I don't think most consumers are going to pay 20 bucks a month for this. Or option B is you make an ad business work. They got no choice to make it work. And by working, I don't mean 100 million dollars. People are kind of ragging on the 100 million. It's in the noise. It's scale. Big picture here. Facebook and Google each do 200 billion plus or minus a year in digital ads. If these guys aren't doing 20 billion within a couple years, they're not even in the game. And to get to the market cap of, I mean, remember, Facebook has a 1.7, whatever it is, trillion market cap doing 200 billion. Alphabet, Google has a 3 trillion market cap doing 260 billion plus thing. If they're going to grow into the market cap on the consumer side, 20 billion is not enough. They have to do 50 billion, 70 billion of ads. So unlike Sora, this is not going to be a try the ads and then fold. This is a, there's only two existential bets for this company. One of them is ads to make the consumer business work. And then the other is, oh my God, we should have done coding all along. Let's get a competitive coding and enterprise model out there and compete with Anthropic on that side. So those are the only two things they're doing, and they're the only two things they should be doing. It's straightforward. I mean, I actually see this as good news. Like, at least they've, like, we've gone from the let's wander around the woods feeling cool building shit to there's only two things to do. Let's get them done. It's net net. It's a positive. Better late than never.
Man, did they have the Wall Street Journal this week? They had a story of why Dario left OpenAI. Did you?
The amount of tension at OpenAI, the fact that Greg Brockman recruited them and no one would work for him. He and Dario and his sister would not work for Greg Brockman, would not talk to him. They would not allow him to be part of the LLM or GPT groups. Then Sam had to constantly tell each of them that they were in charge, right? Told Dario he was the boss. Then told Ilya and Greg they could fire him at any time if they wanted to fire Sam, right? Then begging Dario to come back. Then Dario saying he would stay only if he directly reported to the board and nobody else. I mean, the level, and then firing Sam and then bringing him back and then Sora and we're not doing coding. It's just, I mean, I'm exhausted. I maybe I'm wrong. I have to think at least someone like me would feel much more comfortable in Anthropic where it appears there's a much more consistent process in leadership, right? Same founders, same things, same goals. It just, I have to think a company organized like that's just going to out-execute someone with that level of drama. I almost can't take it.
You're going to kill me for this, Rory. It's the best thing for OpenAI not to buy Sierra, incorporate that as its customer support product, and have Brett Taylor come in as the day-to-day CEO, and Sam can be fundraiser. Sam can be master of
I'm not in the boardroom. So, you know, I hear, look, at the end of the day, at the end of the day, I think you're right, Harry, and I would favor that as a board member, but I'm not going to say that publicly because I don't want Sam to break my balls. I am too unimportant for Sam to even give a shit about. Right? So I don't worry about that at all. So let me say this delicately. That amount of board-level and senior team-level turnover over an extended period of time is probably the highest warning signal that you could have as a board member about how your CEO is doing, right? And if it was anything other than a founder, let's put it this way, if it was anything other than a founder company and this level of drama was going on, you'd probably be having, you'd probably be sitting down with the CEO and asking how's it going at least and what are you thinking of doing about this? I don't think you turn on people just when things go to shit. But you probably want to cut down the drama from here, build a team, and try and call a shot and play it for more than 20, you know, for more than six months at a time. When you've worked at or observed startups where the CEO is spending so much of their time load balancing talent that can't work together versus when you've worked at one or with one where the talent's rowing in the same direction. To say that it's night and day would be an understatement, right? It's like the backside of Pluto and the front side of Mercury. And it's just, and I think Sam, we can criticize him actually. When I read everything I've seen and then when I read the Wall Street Journal, it's like, my God, this guy has spent so much time load balancing the drama of these extremely brilliant personalities that just, oh my God, that can consume most of your time as CEO, most of your time load balancing.
You're exactly right. It is the drama of, you know, we're not dealing with a bunch of people just trying to crank out some B2B software and make a paycheck. We're dealing with people who are angsting about whether this is going to change the world, who have fears about the technology, who have, yeah, desires to be seen as credited for the technology despite their fears about it. This is a, I mean, as is often the case, extraordinarily talented people come on an extraordinarily high bandwidth, a demand on attention and care and feeding. It's been a real slog, I'd say.
Okay. The man with the most balls in investing, Masayoshi Son. SoftBank gets a $40 billion bridge loan to buy OpenAI stock. How deep can Masayoshi go? He'll go as deep as they let him. I mean, that's the one thing we know. If they give him another 20, he'll borrow that too, right? I mean, look, this is a high, I checked the SoftBank. You've got SoftBank Holdings. Have to be careful. There's the telco group, which is reasonably levered at the Japan level, and then SoftBank Group is around, it's around 2x levered, right? One and a half to 2x levered in terms of equity, right? What that means is a 30, 40% decline, you know, wipes them out, right? It's a very aggressive stance, right? It would be like me taking our $800 million venture, $900 million venture fund, borrowing 1.8 billion and investing it all. And if it works, I make you, I really juice my return. But if it goes wrong by 30%, I'm done, right? And it's just, it's super aggressive. I mean, I suppose his lesson is Masayoshi survived 2002 when I remind everyone the NASDAQ went down 85%. You haven't lived till you've seen an 85% decline in an index, right? And obviously if that happened or anything like it, you'd just be way underwater, right? So it's a fairly high amount of leverage for an investment fund to say the least.
Yeah, I mean, for sure it's dramatic. Having said that, you know, real estate investment funds get the maximum leverage they can by design, right? That is how they work. I would imagine if venture had access to more debt, we'd all load up on it. If we all could do the growth rounds in your hottest company, maybe we would, and we could get all the carry from it. And the worst thing is we leave the keys to fund seven on the table. We might load up too. I'm not sure, but certainly real estate funds load up as much as they can.
But just pushing back again because real estate funds load up because the cash flows are predictable, right? At the end of the day, I mean, look in
Well, and they can, but they can because the cash flows are predictable. They can load up.
