Kimi K3 Reactions, The Odyssey Hits $264M in Debut, Netflix Embraces AI | Diet TBPN

0:01John Coogan:It's Monday and there's a lot of news, the big stuff. Kimmy K3 dropped, but first we should recap the weekend. How was your weekend? The elephant in the room and I'm wearing a suit. Yes, the elephant in the room. Did somebody say that? The elephant in the room. Jordi's suited up. He's feeling good. He's refreshed. Yeah, refreshed. I was in the elephant for the weekend. So the old country. Yes, yes, exactly. Really fun time. Glad to be back. Very, very refreshed. Very fun. I was at a Caruso property this weekend, and I got to say, being there, being at the Miramar, Rosewood, I can never vote for him for public office ever again.

0:43Why? Because you want him to develop more properties like that?

0:45John Coogan:Yeah, I want him laser focused on resorts and malls. It doesn't really hold up because the whole point was that there was some possibility that he could make greater Los Angeles something like his incredible progress. You've got to get a Rosewood in every city in America. Or take private of the city of Los Angeles, turn it all into a Caruso problem. Potentially. But it was a good weekend. Glad to be back. The timeline was burning up over the weekend over the AI Cold War that's breaking out between the U.S. Frontier Labs and Chinese open source. There's been a whole bunch of new Chinese open source models that have dropped with some promising benchmarks.

1:27John Coogan:The latest one that everyone is up in arms about is Moonshots Kimi K3. It was unveiled last Thursday. Now, they aren't actually. I kept opening X and thinking, looking, scrolling a few times, thinking, oh, I'm happy for you or sorry that happened. I put my phone away. I feel like I miss a lot. Yeah, I mean, it's gated behind an app right now. The waits are not actually open just yet. That should happen Monday if things go to plan. But, of course, there's a lot going back and forth with the administration. People are weighing every possible consequence of this. It's the classic, it's over, we're back.

2:14John Coogan:Is this good for semiconductors? Is this bad for semiconductors? Is this Jevons Paradox? Is it the counter example of Jevons Paradox? It's all very interesting debates, and everyone's coming out with their own opinion. Dean Ball was going back and forth with two people, David Sachs and also Emil Michael. We're duking it out on the timeline, fighting it out back and forth. But, you know, decent takes on both sides, lots to chew through. The headlines for Kimmy are impressive, but not wildly surprising based on the current trend lines and model progress. But it does seem to invalidate the thesis that open source would start falling behind, which was something that was start.

2:55John Coogan:There was that chart from the government, I believe, that showed closed model progress was accelerating on this particular exponential and open models were falling behind. This seems like a proper catch-up, something that was predicted from many lab leaders that the six-month gap, three-month gap, nine-month gap, it will remain just that gap for the foreseeable future. But there are, of course, a bunch of other nuances here. In the most above-board scenario, sort of here's what happened. Like the moonshot team, it's just a bunch of highly talented AI researchers who marshaled enough compute to release a great model.

3:32John Coogan:Open sourcing Kimi is a great way to grab attention, attract talent, drive demand for hosted API services that are more turnkey eventually. There's a very reasonable business model behind open sourcing. Red Hat Linux, of course, is an open source company owned by IBM now. I think the acquisition price was in the$30 billion range. And at any point, you could just say, I'm just doing Red Hat on my own. But a lot of companies pay for consulting, hosted, and all sorts of different services around the open source piece of software. That's sort of like the bull case, like everything's going well. The more skeptical folks on the timeline are pointing to ideas such as maybe the compute was smuggled into China against chip controls.

4:17John Coogan:Maybe the data was exfiltrated from frontier lab APIs, routed through wrapper companies. Maybe this is a distillation attack. The open source strategy is a deliberate attempt to destroy American companies. It's geopolitical warfare. It's the AI Cold War. And there might even be spies inside companies stealing IP directly. And so that's sort of like the bad version. The reality is probably a mix of both, but we'll know more over time. And we'll hear more statements from the labs and from all sorts of different participants to understand. People are going back and forth on, you know, if you ask Kimmy its name, will it just say it's Claude or GPT?

4:53John Coogan:And there were some examples of people doing that. Seemed like it was Photoshopped. That also seems like the easiest thing to fix, possibly. It's just put it in the prompt. Put it in the fine-tuning. Find and replace. Seems pretty easy. Yeah, but they were showing it in the reasoning tracing. Yeah, but I don't know. All these things are hard to assess over time. It takes time to actually work through what's real, what's fake, what's a hallucination. They all make mistakes from time to time. But there are bigger questions about AI safety in a world of powerful open source AI. I haven't seen a ton of takes from this crowd, but you have to imagine that they aren't happy about it.

5:31John Coogan:Maybe the Stop AI protests will head over to Beijing next and beat down the doors of Moonshot directly. But it is an odd thing because even if you're not worried about the fast takeoff, you know, gray goo, paperclip monster. From a cybersecurity perspective, things are going to get more complex. There's reasonable solutions to all of this, but it is a more powerful tool in the hackers or the attackers tool chest. Fortunately, Defenders have had roughly six months of frontier exclusivity with GPT Cyber and Mythos for a while. And so there's been a lot of preparation. We've talked to Nikesh at Palo Alto Networks and George at CrowdStrike about this.

6:11John Coogan:They have been aware that this is coming for a while. And so a lot of systems have been patched. But at the same time, you can just imagine now anyone who's a hacker can spin up a very, very intelligent model and start sending a bunch of spam emails or spam text messages that stop seeming so AI. And they're not as clockable because it's just another level of intelligence. and it can do reasoning traces and research. And so the spammer or the phishing attack can actually know what insurance you have and then pose as your insurance agent and actually know a little bit about your insurance agent instead of just sort of sending you a generic spam text.

6:55John Coogan:Things are going to get more customized, more personalizable. Of course, there will be more advanced defense layers, but the war is continuing. There really is no debating that open source is just cool, like open source AI. It just feels punk rock to me. It feels like Napster. Like as a kid, the idea of just like liberating all the information. Yeah, it's culturally aligned to the internet itself. Yeah, it's just a cool thing, even though you do have to grapple with all the different problems that come from it downstream. uh like the there's something just deeply satisfying about the idea of like a solar powered server rack serving you up unstoppable intelligence which is another good name for a neolab you know a country of genius it feels yeah in some ways to me it feels like aligned to america's ideals oh yeah it's like it feels like it feels like like very empowering it's total second amendment yeah it's totally second exactly yeah it's like i should be able to go off grid and have not a country of geniuses in a data center controlled by a corporation, but just like my own genius who's riding with me no matter what in my basement.

8:01John Coogan:And it's like, even if you're not thinking like nefarious, it's like this idea that I like to have a copy of a movie in case the grid goes down. The more like apocalyptic thinking, this is a very general thing that I think a lot of people align to, especially if they're at all crypto native. It's like, why do you want Bitcoin? I mean, well, it's unstoppable money. The government can't take it from you in certain scenarios. Of course, like there's a bunch of ways that they can, of course, like anything else. But there is just something very, very attractive about having just a machine with all of human knowledge on it.

8:36John Coogan:It's like every house used to have the full Encyclopedia Britannica. You can just look up anything. And now you have the LLM that can solve any math problem for you. But, you know, the U.S. government is going back and forth on this issue, how to handle the situation. And I imagine we'll see developments this week. The Kimi K3 weights drop next Monday. And so this is the week where if the government wants to do something one way or another, they have the opportunity to do that. There were some back and forth on, like, would there be a direct ban, a hard ban, a soft ban, maybe just some, like, messaging around, hey, like, please be a patriot.

9:17John Coogan:Don't use this foreign product. You know, there's all sorts of different ways. Like, the original, like, buy Made in America, buy American. That's just marketing, but it does have some effect. It's not enough to make GoPro a$10 billion company, but it is enough to cause a little bit of motion in the economy. One thing that was notable over the weekend, Kimmy on X said that, Kimmy K3 has received far more love than we expected, and our GPUs are feeling it over the past 48 hours. Demand has pushed close to the limits of our current capacity to protect the experience of existing subscribers. We're temporarily pausing new subscriptions and prioritizing compute for current members.

10:00We're adding capacity as fast as we can, and we will reopen new subscription spots and batches. So Kimi and the world more broadly are quite supply constrained and showing that the other dimension of this whole geopolitical battle is around the hardware side.

10:18John Coogan:I mean, in general, I think my takeaway was, and Tyler, I'd be interested to hear how you process this, but I just don't see this as a black pill, even if it's like a deep seek moment and the markets sort of bounce around for a little bit. But it seems like there's still a lot of demand for frontier intelligence, a lot of demand for near frontier intelligence, a lot of demand for really cheap older models that have been baked down to basically intelligence too cheap to meter. And we are going to be in this process of go and find an actual use for something and then bake it down to be really, really cheap so that it can be offered.

10:54Yeah, and in so many moments over the last two years, there's been a company that's come out. Maybe they've raised some money. You maybe forget about them for like six months. And then they pop up, and it turns out even though they're the number 10 company in their category, they're still growing faster than almost any company that existed like six years ago.

11:18John Coogan:Yeah, and a lot of it goes down to deployment too. There's like certain companies, if they're in consumer, like they just have a really good go-to-market motion. Like they're just able to get customers to subscribe or use their product. Or if they're in enterprise, like they just have a great like forward deployed motion to actually do the change necessary as opposed to just like here's an API law firm. Like they could have done that. And yet you have legal AI companies that have done a great job of coming in and actually creating a product that works within the confines of the broader organization.

11:49John Coogan:What are your Kami K3 reactions, Tyler? Yeah, I mean, I think it was definitely like a surprise, an update. Like, this is an extremely good model. The idea that open source is falling behind seems like very much not true. We sort of got a preview with GLM 5.2. Yes. Like, people were already saying, like, oh, GLM 5.2 is sort of catching up. Yes, but even that was closer to, you know, maybe a year, maybe a little less than a year behind. It seems like much closer. Okay. Yeah. But I think I broadly agree, like, okay, there's this great open source model. And now we have, okay, the Jacobian, you know, conjecture.

12:22Like this kind of storied math problem.

12:24John Coogan:Yeah. We'll have to see. And so, like, yeah, there's just going to be a lot more use, like, everywhere. I expect, you know, these kinds of models to be used everywhere. It seems like a lot of hype is around, like, specifically the parameters, the size of the model, the way it was trained. And then also it did very well in the front end arena, which was interesting. Because that's, I mean, it jumped ahead of everything there. and that feels like a particular... It'll be interesting to see because you see more and more taste going into these models. OpenAI is really big into the math stuff and Anthropica was really big on code early.

12:57John Coogan:And there's all these different... You can just tell that, oh, this lab cares about chess now and now the model's good at chess or the model got better at writing because they took the time to spend a month sprinting on that. And it'll be interesting to see where did they sprint? If anywhere, where didn't they sprint? and it's sort of lagging, and then what harness it works best with, how important is the harness relative to everything else. But it will be interesting to continue following. Semi-Analysis said, similar to the panic over DeepSeq R1, some uneducated people think Kimi K3's use of linear attention, KDA, is bad for NVIDIA, HBM, DRAM, and networking because it has relatively lower KV cache requirements.

13:40John Coogan:The opposite is true, and they'll explain. Kimi K3 is actually quite positive for NVIDIA as large model inference is where the NVL72 shines because K3 has more than 2.8 trillion parameters. It requires a large scale-up domain to store its weights. Secondly, although Kimi Delta attention has up to 10x lower networking requirements for KVCache transfers, its large weights require even more network bandwidth to implement an optimization called WideEP, which spreads weights across different GPUs. Is China getting TPUs? Those have to be chip export restricted and probably more difficult to get into the country because there's less resellers.

14:17John Coogan:And in order to set up a TPU-based data center, you probably have to work very closely with Google, whereas NVIDIA GPUs are sort of going to, like, all sorts of small Neo clouds, and there's a lot of places where seven steps down the chain can wind up in a suitcase with a hairdryer removing a label or something like that. Well, it's not even that. Matt, you can imagine if you were a Chinese lab and you wanted to get access to compute, there would be a number of ways to set up U.S. entities that then basically get compute through a variety of different sources here in the U.S. Yeah. And you're just getting it remotely.

14:54John Coogan:Yeah. Nicholas Bustamante at Microsoft, who is working on AI for knowledge workers, says, Guys, the world is so compute constrained, I honestly don't understand how anyone thinks otherwise. Almost nobody is seriously using AI today, and we're already hitting capacity. Everything is a bottleneck. Land for data centers, permits, electricity, grid connections, chips, construction, cooling. Demand is growing far faster than supply, and expanding supply is insanely expensive in this game. The best positioned companies are the ones with massive net income from non-AI products. They can fund hundreds of billions in capex.

15:30John Coogan:He's given a little bit of a bull case for Microsoft. Guillermo over at Vercel says Kimmy. Tested it. What's that? Yeah, that's good to hear. Yeah, Kimmy K3 is top tier at cybersecurity. That will be interesting as basically the weights are released. Do we see an uptick in tax? And it's also sort of a a refutation of like the, it's just distilled because the models that you would be distilling from are pretty severely nerfed on cybersecurity, at least in theory. And so it's like you would need to distill on the non, the non nerfed model, which is probably way harder because those are much more restricted.

16:15John Coogan:And so it does feel like there's a, there's a little bit of refutation of the idea that these are purely distilled. Sol is a leap ahead in cyber capability at a significantly higher cost, but quite remarkable still. Fable refuses everything. We couldn't get it to complete the run at all. What is interesting is that Sol, in comparison, was much more open to helping with defensive cyber hardening. TLDR Frontier open-weight cyber capability is here. This was good. Hi, Harry says, RIP the bear case, July 16th to July 19th. perfectly coinciding with my trip, which is nice. Dean Ball was going back and forth.

16:57John Coogan:Oh, some deleted posts now. His observations were getting taken to task, going back and forth. He says, it's a very good model. I don't think its performance can be explained away by distillation or anything like that in agentic coding sessions. It seems pretty much on par with the best public models of Q1 2026 in my fairly limited use. It also seems very token hungry. Dean says, I am personally surprised the Chinese state continues to allow the open sourcing of models this good, given potential risks. Now, it seems like that is very much like from the top, they are pro open source right now. Dean says, to be clear, I myself might be fine with models presenting this level of marginal risk being open weight, but I'm surprised that China is fine with it.

17:38John Coogan:I suspect the reason is that they are. It's 75 % explained by strategic blindness, lack of AGI pilledness. The CCP is very Yan Lacuni in its view of AI. The other 25 % or so is their lack of compute for customer inference, making China's open weight strategy an unintended byproduct of US export controls. That's interesting. And the normal Chinese strategy of aggressive exports for the companies as opposed to the government, the decision to open source is partially ideological and partially because they are behind. And they know that very few people would pay for sub frontier models from China.

18:13John Coogan:Yeah, you have to imagine that it like the idea of sending, I mean, people are having discussion over like, should you send your corporate data to an American frontier lab? Imagine that same level of, should we upload our entire code base to the moonshot API because we've gone all in on Chinese closed source? So it seems like that's a very, very hard sell. So there is a world where there's like no money to be made in America. I don't know. Maybe there's some American businesses that would just be like, yeah, I'm cool with it. Take my, take everything. Take all my IP. I'm anti-fragile. I can't be, you can't clone me.

18:49John Coogan:So, and I have no IP. So just upload it all. I mean, if you're a local plumbing company. Yeah, maybe. Probably in a good spot. Maybe, maybe. This is a pretty hot take. Dean says open weight models are inherently decelerationist. And Tyler and I were debating this before the show. He said, I'm continually surprised to see the so-called accelerationists so excited about open weight models. I suspect the reason is that they know open weight models are effectively ungovernable. That's correct. And they simply like the overall cloak of ungovernability open weight models create over the whole of AI.

19:26John Coogan:It's not a bad strategy. It reminds me of James Scott's recounting of the Hill people in The Art of Not Being Governed. Still, in the end, open weight models deter further AI CapEx. And this is the debate that we were having was like, how CapEx intensive are these frontier training runs? Because Moonshots raised, I think, a billion dollars recently. And it just feels like even the latest labs, like there's a lot of cost, a lot of researchers, and they do a lot of experiments. But the core run might not, we might not be talking about 50 billion yet. Maybe in the future, like you need to accrue towards that.

20:04John Coogan:And maybe people wouldn't fund it if they're like, oh, you're only going to be able to monetize this for like three months because you're going to get distilled so fast or you're going to get copied so fast. That is tricky. Like it is nicer when you have a monopoly or oligopoly for some period of time to actually earn profits off of what you invested against. Well, if anyone listening to this is over at Kimmy or Moonshot. Come on the show. No, don't come on the show. Just feel free to let us know how you made it. Yeah, maybe. Feel free to leak the secret sauce. On the show. Come on the show. On the show, yeah.

20:38Yeah, if you're so open, why not? The Odyssey is up to$264 million. It's a really good selling. Biggest non-animated opening of the year until Spider-Man in two weeks, according to Lucas Shaw over at Bloomberg.

20:53John Coogan:$375 million budget. Production was$250 million. They recouped that immediately. $125 million in global marketing, apparently. And looks like a big financial win, but very much priced in. The base case here, the film will make a little bit over a billion dollars. The cash flow to Universal and Comcast, who are the studio behind this. You're looking at$250 million in pre-tax lifetime profit. That's 0.2 % of Comcast's market value. So it's not like this crazy thing for the corporate entity, but it is exciting for the film. And it's exciting for Christopher Nolan, who's in the journal today in the business and finance section.

21:40John Coogan:Because Nolan's The Odyssey shows directors are franchises now. They have an interesting statistic in here. 53 % of Odyssey attendees said that the director was their number one reason for attending. not the actor not the fact that they love the book and they were waiting for the ip to get adapted not that it's a you know so they like sword and sandals they were looking or that it was imax they wanted a nolan film and that's i mean everyone knows this intuitively from talking to everyone everyone's like i want to see the next nolan film but uh it's just interesting that that there has been this huge shift towards uh franchises built around particular intellectual property verticals Marvel, Star Wars, Disney stuff, Pixar and whatnot.

22:26John Coogan:But now people are saying, I'm just going to ride with this director and whatever he does. And he can take me to the past. He can take me to the future. Wherever he goes, I'm riding with Nolan. I had a funny run-in with Christopher Nolan. I was walking on the beach and Christopher Nolan was walking in front of me. Yeah! I'm not hitting any sound effects right now, to be clear. And he's walking with his dog. I don't know who he is or what he looks like. Someone else who I'm with says, that's Christopher Nolan. I said, who? And then it came to me. But then this dog ran up behind us. He was walking his dog.

23:13Another dog joins and is, like, really messing with his dog. And so I had to be like, I just had to say, hey, man, it's not my dog. Because it seemed like my dog was off leave. I was being framed for having a poorly trained dog that was harassing his dog. But he was pretty unbothered. Let's head over to this picture from Gautier. He had the best seat in the house.

23:38John Coogan:Is this a real photo? Or do you think this is like take a photo and then put it in AI and say, like, make it even more distorted? Because this is what it looks like to be truly in the front row of an IMAX showing. People were really crazy about the IMAX looked way better. I think some folks on the team saw the film once, twice, three times. Who saw it in IMAX? Okay, we got some IMAX chats back there. I haven't seen it yet. Mike's seen it twice. Give us a review. You liked it? How many thumbs up out of ten? What do you got for me? Tyler, you can chime in too. What do you think? It was fantastic.