A

A big week for China AI news. We had Kimmy K3 drop and she giving his biggest speech on AI yet to discuss. We have the full ChinaTalk squad in attendance for this emergency podcast. Irene Zhang, Akab Zakaria, as well as Lily Ottinger. After. After much debate, we've decided to start with Kimmy instead of she. Sorry, Big man. AKA why was this model release a big deal?

B

It's a big deal because I think now everyone in the US AI community is a little bit afraid. I think in the past months we've been like, oh, we're pulling farther and farther ahead from China in terms of model quality and model size. And first it was GLM 5.2 and now this week, Kameke 3 just shattering that illusion. And I don't know, Kimmy and China may be closer to a Mythos mythos model than we think.

A

So we have that release coming basically simultaneous to waic, the World AI Convention in Shanghai. She makes his first appearance, gives a thousand word ish character ish speech which took him 20 minutes to read. It was like kind of brutal watching it. He seemed a little on the slow side, but like he got a little, he got a little juiced up for like Kakao Ko Wen. He was excited to control and make the model stable. I mean it was a pretty strident laying of the flag in favor of keeping stuff open is the sort of like first read. Maybe we can get to some Straussian ones later.

C

I think it's interesting the extent to which other countries are being talked about in the context of the expansion of the ecosystem and the cooperation initiatives announced with like Belt and Road countries, the Shanghai Cooperation Organization, et cetera, asean. Also, when I was in China touring some of the Chinese AI labs, I found out that a Chinese company is helping Kazakhstan build their sovereign AI initiative. Maybe Kimmy will start speaking Kazak. Maybe Kimmy can be my Kazakh tutor. Because GPT is not quite there yet.

D

And think about it in the context of who's he speaking to, right? So people have met, have various people have noticed and analyzed by this point the fact that he opened with the Dartmouth Workshop, which seems like a bit of olive branch of the United States. But then he goes on and in the middle section he talks a lot about inclusiveness, about the uniqueness of world cultures and solidarity. And that's clearly meant for the global south, for countries that China is trying to contest in terms of influence and ability to, ability to build partnerships in those ecosystems. And I'm very, I'm very curious as to how that's received in those capitals. Obviously we're talking about, talking about the global majority, very diverse group of countries here. But like one part of it with relating to Kimi is also that Kimi, while still cheaper than American frontier models, is more expensive than the previous kind of class of Chinese open models we've been used to. And that incurs not, not insignificant cost for those very global south countries. Right. So it'll be interesting to see how that tension plays out.

A

Well, the cost, if they're going to Opal, if they're going to put it open, weights in a week won't matter. Right. Like whatever Kimmy says it is. So anyways, hold on to your horses for that. July 27th to TBD. I want to stay on Lily's point of like, this feels like it's like a speech for the world. It reminded me a lot of the Shi Davos 2017 one, which I just opened and remind and was reminded that he started with a Charles Dickens quote, best of times, worst of times, and then, you know, like two weeks after that. This is, this is his big setup to say, you know, we're open for trade, we're open for business, we're not going to tariff you like Trump is going to. And then all of a sudden, you know, there's that in South Korea and boom, we're, we're, we're having huge sort of like economic coercion and on that country. And then of course, in the subsequent years, you have tiffs with Japan and Australia and India and on and on down the list. So in terms of the reception, I was at a conference with some like, American AI people, investors, and all of them were like, wow, this is really amazing, remarkable. I wish my president could, could submit out this vision to the world. And I would just encourage folks who've like, if this is like the only she speech they've ever read in their life, to maybe take it with a bit of a grain of salt that these sort of like promises of like global comedy and, you know, an outstretched arm often, you know, have other layers and subtexts to them. Irene, like, you know, reading that speech in the context of the AI plus plan and the other sort of official documents we've seen come out over the years, what striked you as. Struck you as sort of similar or different?

D

I think thematically they've been really consistent. Right. If you look at the big speeches and the big initiative docum white papers coming out of, coming out of China State since 2022, probably they've been pretty remarkably consistent on their own set of buzzwords like openness and inclusivity and kind of striking side boss in the United States over what it calls the geopolitization is the criterization of AI. Right. You see probably more discourse in these documents, especially us a little bit in the XI speech over safety now, as that has become a lot more salient and less abstract. But I think it's also just kind of a little hard to compare because when she gives a very, very kind of once in a few years speech like this, he is hitting at some really abstract themes that he. He's not. Not expected to be delivering very specific policy initiatives or giving out a specific specs on how much China will govern AI. That probably is not his remit in real life either. As in, he's not. It's. Folks might disagree with me. I'm curious what y' all think. But, like, I assume he's not giving actual memos to the CAC on what to do. So what he's doing here is reinforcing the ideological layer and also signaling domestically that this is the general path forward.

A

It's almost like he's just. This, I think, is the strain of almost all of the policy documents. I mean, really, all of the policy documents we've seen is they see this as a normal technology, right? And like, yeah, if it's a normal technology, why shouldn't it be as cheap as possible? And. And as many people in the world have access to it. And this can be, you know, a great export for China, just like all these other sort of export businesses we've built over the past 20 years. But at the same time, you know, we have OpenAI 5.6 SOL plus, like, breaking out of a harness to, like, access the Internet by, like, loading up seven other models. And so this may be right, but it also may be, like, three months away from being spectacularly wrong. And I don't know. I don't. I don't doubt this government's ability to sort of change if and when they get confronted with something which is actually a lot more scary than it may seem today. And it's funny, Akiv, you started with this idea that the gap is narrowing, and it's like, well, there still is, like, pre mythos and post mythos, right? And Kimmy is still pre mythos. This is not like something you can trivially hack stuff that hasn't been, like, mythos hardened yet. So it may be that this she speech is just like a high point in, like, open rhetoric and Once they get kind of hit in the face with it, like, like the White House, did they change their tune? But I don't know, maybe. Maybe everyone's just, like, freaking out too much and they're wrong and CAC and she are right that we don't actually have to be super worried about this stuff.

B

I think the lens, that AI is just normal technology actually explains a whole lot when you think about this speech, because I think as an American audience member, you sometimes feel like there's this weird tension between wanting AI to be open and inclusive, but also safe and secure. Like, we think those things are inherently contradictory. But I think the Chinese position on this is like, look, it's just like the Internet. Like, you want it to be diffuse as possible, but just. Just have good cybersecurity. When you really, like, start doing this right, what's the problem? Why can't you do both? But yeah, we'll see how. And if that changes in a post

A

mythos, like, we don't really want to buy American chips, even if they're offering us them, because, you know, if your timelines are longer than, like, yeah, China's done without for a while with lots of other things in the past in. In the purpose, you know, for the goal of import substitution. Right. So, like, yeah, sorry, bytedance. Sorry, Alibaba. Just like, deal with it. Make your own. Make your own chips. We'll get there.

C

I'd like to draw a little bit on what Irene said about the domestic signaling aspect of the speech because, you know, the feeling that I get from Chinese developers is that they personally love openness. Um, and the AI researchers love open source because, you know, they learned a lot from open source software in the pre AI era, and they feel very strongly ideologically that they want to give back. You know, I heard researchers say, like, this is in our blood. Like, we just, we love openness because we really just believe it's. It's the right thing to do and it's a way to, like, exert your will on the world and contribute to a community that you care about. And so I don't know if the government necessarily cares so much about openness as an abstract value, but you have to remember that the people in the audience are all domestic Chinese AI research researchers. And at some point, you know, the. There's the question of, is the crackdown coming the way that, you know, they came for Jack Ma, are we next? And I think when they say, you know, when she says, you know, openness is. Is what we care about. The Most. I think maybe he's just trying to say the government is on your side. Unclear how long that will last. But I think he's just trying to say, like, I don't know, hype up the domestic AI ecosystem and make them feel like that they can innovate without being constrained too much.

D

So I don't know.

C

We will see. We will see if. If like a mythos level model changes that, I think it probably will.

B

I was thinking that statement was a lot more cynical that a lot of these models who maybe want to go close source. She is saying, we do not prefer that. Don't do that.

D

I don't know whether he's necessarily saying that closeness is bad, as in, he. I don't think he said that explicitly. It's probably. That's probably just the realm of corporate strategy. But he. I agree, Lily. I think he's offering a kind of like a comp hill for open source and being like, yes, we're back for now. I also want. On the point about AI's normal technology. So she ends with this quote, which is officially translated as a man of wisdom adapts to changes. A man of knowledge acts by circumstances. And I want to look at where it's from. It's from this record of policy debates from the Western Han dynasty, which around 81 BC. And I thought it was really. I thought it was actually really interesting because this document is called Discourses on Salt and Iron.

C

And.

D

And most. A lot of it was about how the Western Han was governing the salt and iron industries, which by that point was largely privatized. And they were trying to figure out what, like, how much should we extract from this private enterprise. And if any listeners who are experts in the Western Han, please correct me on this, but it's always like a cool parallel with AI today, which in China is also largely privatized. And the state is trying to figure out how much to extract from it.

A

And I think other, other empires brought it in house, right? Like there was a salt monopoly for like hundreds of years at some point,

C

which led to illegal salt trading and salt gangsters that emerged.

D

So.

A

So speaking of nationalizing salt in, you know, 300 AD, maybe we should talk about our joint piece. We put out a few days before Kimmy, or I guess a few days after Kimi launched, where we kind of did a little scenario analysis of three potential paths for where Chinese governance of AI could do. I think each of you guys wrote one of them. So maybe we can go around. Let's start with Let it Rip

B

Yeah, I wrote that one. So the Let it rip scenario is just the status quo that the Chinese Mythos model will come out just as Kimike3 did, just as GLM 5.2 did. It just drops and maybe hopefully there's like a one week delay saying oh, release the weights later. But there doesn't seem to be any system within the government for catching. Oh, this model is different. This model actually has the capability to do some real damage and then once it's out, it's out. And then maybe after that the Chinese government, this will be the wake up call. That is like the wake up call. They need to scramble and have some policy to change how they review AI models. But an open weight Kimi K4 or a GLM6 can do some major damage out in the open.

A

Well, I guess the question I have now is like what if it just doesn't? What if it's just like another day in the life? I mean we've had so much. Can we call it hyperventilating? Can you call it alarm raising from cybersecurity experts? Like there is a world in which they are wrong where the compute, I mean let's make the case right. What was Kevin Hsu saying? You need serious access to compute in data centers, maybe at the sort of cloud layer and the Neo clouds, they just decide that they're not going to do business with cyber hackers or Boko Haram or whoever else. Maybe things get patched fast enough that it's just like kind of a blip and we get to the other side instead of it takes two or three years as some people are proposing, it takes like a month. Then all of a sudden our models are great because there's just a lot more money being put into these models to clean up systems than it is to attack systems.

B

Other

A

arguments folks have in favor of this just like not, I don't know, we just kind of move on.

C

Well, I think from the Chinese government perspective, they already live in a world of cyber exploit abundance. You know, they've had a system for, you know, systematically acquiring zero day vulnerabilities for a long time and have personally decided that the risks of using them outweigh the benefits in a lot of scenarios. But that's because China's a state actor, you know, like with constraints on them placed by the multilateral system, the trade system. They don't want the backlash. And I think they personally feel like they have domestic insurgencies under control. So to them it's not a threat. Domestically, I don't think we are necessarily prepared for a world in which state actors, you know, Claude Code made a lot of problems look software shaped. And I don't think we're prepared for a world in which a lot of problems look like they have solutions that are cyber attack shaped. Unclear how, you know, the steps required to execute on any of those things, but the fact that there's one, you know, the fact that the exploits exist, I feel like they'll at least be sellable in a way that makes a lot of money for, you know, rogue actors in a way that we're perhaps not prepared for. It'll certainly be be. Be a nice way for the North Korean government to make money. Uh, but China's not worried about them either. So it. You could be right that there's no problem. I think the Chinese government for now certainly thinks it's not a. Not a risk.

B

I think it could be such good PR for the Chinese government if it comes out without like a major incident because they're giving us like, you know, people are upset over the safeguards on Fable 5 and all these like, lobotomized models, and now China is the one giving it to us straight from the source, you know, and then you have this scenario where the leading AI models are Chinese and also the lower cost ones are also Chinese, which could do major, you know, benefits for their adoption levels. Yeah, I think the Kevin Hsu argument is like, you know, for Kimmy, which is over 2 trillion parameters, you need like a dozen GPUs or something like that, which is, you know, you need a couple million on you, which is a lot, but not, certainly not insurmountable if you're any organization or any, you know, evil rich guy. I suppose it's just a question of what actually gets done with it and how hard, how much damage the CCP is willing to take. And if there is a major cyber attack with this Chinese mythos model, how bad of PR is it for the government? And what can they even do after the fact to correct it?

D

I feel it's worth mentioning very quickly that the Chinese government does get hacked all the time. Their systems are not nearly as full proof as they would like it to be, as any government would like it to be. And beyond state actors, beyond the United States and China's various rivals around the world, they're also just like the scammers in Myanmar want the Chinese people's pii. So I think there are just lots of ways to make life annoying if bad actors get their hands on really strong cyber Cybersecurity tools, even without it kind of challenging major geopolitical assumptions.

A

My MARPI is one thing. Like the NSA hacking you is also sort of one thing which is just like, I mean that's just like yes, governments hack each other, but they don't put all the secrets online. Like we haven't really had WikiLeaks for China. Right. And you know, that I think would be a very interesting test of the case. Right. Is all of a sudden like, I mean every government has like awkward things that they don't want the world to see. And I'm sure some of those are digitized in China and there are definitely enough people with a few million dollars who, you know, feel some type of way about the Chinese government. And then, okay, so then if you're Xi, like what? Like that's a really interesting thing because then, then you start to get into these, you know, state stability, global reputation type of questions. And it's not like, that's not sci fi, like bioweapon stuff. That's like pretty run of the mill. Like we've seen this happen in a lot of authoritarian societies around the world where kind of like awkward news comes out and folks freak out. I mean of course like you didn't even need a cyber attacks in 2012 with the new York Times reporting for that to like really around sort of elite politicians and money for that to really change how folks in the government thought about how they wanted to let the domestic population have access to global news. So that seems like baked in almost at some point.

C

I feel like the PR crisis comes when a Chinese made nuclear reactor in a belt and road country gets shut down via cyber attack. That's my, that's my prediction.

D

Okay, that's all reactors.

C

Pakistan has some. There are plans to build them in several other countries in the global south. But the Pakistan ones I think are the ones that have been subjected to the most fanfare. But yeah, China makes a lot of nuclear reactors for countries that would not have them otherwise.

D

Reading this for, for another article that hopefully will come out soon. Reading this. October, October 2025 guiding document by the Cybersecurity Administration of China for various. It's. It's basically a guiding, a guiding document for various government agencies to adopt AI safely. And there's one section that's actually pretty relevant to what we're talking about about how China deals with state, state secrets where it, it emphasizes, and this is my own translation to strengthen a bottom, bottom line level of thinking and strictly make sure that uh, anything that's anything that is a secret does not go on the Internet. And when you go on the Internet there are no secrets. Um, and it doesn't say that no state secrets or sensitive information should go into AI models, but it says no natural secrets, work secrets or sensitive information should go into AI models that are not secret secrets informed, I guess, which implies the existence of secrets informed AI models.

A

All right, well, maybe we should jump from let it rip all the way to black box. Irene, what was your vision of the future?

D

Right, my vision, which I think ever ranked as the least possible, but which I will try to defend. Here is a world where the Chinese state realizes that one of one of the leading labs in China develops a model that heavily surpasses previous models in cybersecurity capabilities and capability in general and decides that rather than letting the company release it openly, they will get first dibs on it for government use and then only allow the company to release a kind of lobatonized version because it is to consumer safety and in that process, well, partially nationalize almost this, this lab and try to try to coerce it into developing products that are more tailored towards government use and potentially offensive use. And the reason I think that's a genuinely possible scenario is I think, I mean I know the biggest counter argument is just that like this, this is commercially bad for China, the Chinese AI's influence in the world to have its, its top capability and the country's top models not be available to an international audience to diffuse and more diffuse and compete with the U.S. but China's advantage in some ways is that it just has a lot of labs. So let's say KMA gets there force. Let's say K4 is somehow extraordinary and the government takes that in house and forces Kim Me to not release that openly. Like Zidong Al will get there someday. Kun will get there by dance. We don't hear enough about bytedance but like they'll get there someday. In a sense, I feel like the advantage that Beijing will get from a strategy like that maybe outweighs the short term PR gain and commercial gain. And I think my scenario makes sense in a world where the Chinese government is a lot more kind of safety pill than people assume and also sees genuinely useful things it can do with AI for national security and its strategic vision in the short term enough that it thinks that making that sort of very explicit move to acquire top intelligence is really useful.

A

Two points in your favor, Irene. From the Xi speech. Perhaps the most interesting paragraph was him saying we must attach great importance to the various endogenous and derivative risks brought about by AI and work to build systems of laws and regulations, technical monitoring, early risk warning and emergency response. We must reinforce the safety baseline, prevent misuse and abuse and ensure that AI always remains under human control. So clearly some little safety emphasis there. Plus, I gave the top three Chinese models and the top two American ones all our three scenarios and the Chinese models believed in the black box. So even though me, Lilly and Akiv all gave it like 15, 15% chance, I think deep seq was up to 30%, Kimmy was up to 45%. So the Chinese models think they're going to get nationalized sooner rather than maybe

C

they know something we don't. Maybe they get maybe some, some, yeah,

A

maybe they get that secret, that secret training data. Irene, they've read the plans.

B

I was going to say, the only reason I find scenario three unlikely isn't like I don't think there's this big mega brain trade off happening. I think for scenario three to happen, the Chinese government has to know that this model is going to be like really good. Right? And the only way that happens because I don't think the model regulators check for cyber capabilities or bio capabilities is, you know, Kimmy has to be like, Guys, K4, K4 is different. And then the Chinese government has to believe them and then they have to take that step of nationalization, which I think is just like a complete U turn from the speech and like the messaging they're currently giving.

D

Well, Nan Xi had made that point when he, he came on to chinatalk about how there's the, there's the version of the registry that's on paper and then there's what happens in real life, which makes a lot of sense to me, like given what I've heard as well. So on paper, yeah, they don't test for bio, they don't test for all this before they can release model. But in reality the CAC is probably in pretty close contact with these labs and I really doubt that they have kind of no idea of model's capability before they go out.

C

I kind of feel like from talking to AI researchers in China, their understanding is that they're going to make the models exactly as safe as is required by law and they're not going to think about it beyond that because they're compute constrained and they just want to do their national obligation and that's it. I think maybe they talk a lot about safety, but it's unclear about, you know, as far as how much that's like a thing that they personally really believe in versus a thing that they're aware that the global community believes in. And they, you know, thus want to be interested in safety in order to, you know, have their place in that global community. I, I think if it's left up to the labs to, to flag cyber risks, we are in a bad spot. I think let it rip is probably pretty likely in that situation, but who knows?

A

So let's go, let's go to you. You then. How does China's glasswing happen?

C

Yeah, so I would think that Project glasswing would happen in a world where the, the capabilities are flagged in advance by, by somebody, someone from the Chinese government gives a speech in advance that says, you know, we need to move the testing capabilities up. We need to evaluate first, release later. We need to fold in relevant stakeholders from the private industry and government ministries before releasing a model with a huge step change in capabilities. And then the labs take that as a signal that now it's their responsibility to figure out whether their model is going to be Mythos level before they release it. And you know, perhaps the, the fact that Kimmy's like waiting to release the open weights of, of K3 is like maybe an indication that the labs are, you know, thinking about this more. I think this is pretty likely because there's a playbook for it already in the form of Project glasswing and I don't know, it's. It enables the model to eventually still go out in a way that is safe for everybody and has the cyber capabilities lobotomized, but enables Chinese developers to still work with the best tools, which is something the Chinese government seems to really care about because they believe, you know, AI is a huge unlock as far as productivity is concerned. And so I, I think that there are forces in the Chinese government that will work against having the models be totally black boxed and government only and never released. And yeah, so we'll see. I think it would, I think it would be, it would cause a huge amount of infighting. And I think that the labs would not necessarily enjoy being the national champion of a glass wing style project because it would mean handing over their proprietary technology to their competitors, all of which are building their own foundation AI models too. So it's like, you know, when, when Mythos was handed over to Microsoft, it wasn't such a big deal because Microsoft's not like a serious competitor in the AI race. But that's not true in China because everybody and their mother has their own foundation model, including delivery apps Like Meituan and they all hate each other and have long standing rivalries and they all like poach employees from each other and there's. It's so much drama.

D

So, yeah, I was trying to think of examples previously of like Chinese big tech cooperating with each other on stuff, and I really couldn't think of any.

C

And I think the government likes it better. You know, I think they prefer that the companies don't necessarily get along and do these big, you know, cooperation initiatives that stretch across conglomerates.

D

Right, yeah. That's called price fixing. Yeah.

A

So the funniest thing I think that's happened on the Chinese AI race over the past few weeks is Qihu360, which is a security company, is like trying to convince everyone that their model or their harness is like mythos class and they've like done their own Glasswing. Like their CEO is like, just like Glasswing. Like, we are only giving out our, you know, special sauce AI product to like a handful of providers. And they took a photo or whatever. It's like, okay, guys, I mean, sure, good luck with that. There's a piece of this lily which you may be missing where, like, there's some, there's some like, honor and being the one that like has to kick off the glass wing.

D

My question for that is I don't know whether to an extent that kind of honor really buy some stuff in China right now. For example, Anthropic's marketing makes sense even to, honestly, at this point, average American society, because Anthropic and its peers have been messaging for so long to the general public in the US about AI's risks. And everyone and your mother has heard about, oh, AI could cover us all and stuff that doesn't happen in China yet, like the kind of catastrophic risk narrative has not taken hold. People are much more kind of people's negative feelings are mostly about job loss and on the economic side. So if some company were to come out with that kind of, you know, heroic messaging around AI safety and taking that baton of honor, I don't know how much that resonates with the actual public. If that's the like apples to apples comparison we're going for, like the previous examples of Chinese big tech doing, I don't know, honor based stuff that I can think of are like, I don't know, donating masks during COVID It's like that kind of stuff that genuinely resonates with average people.

C

Seems like a thing Alibaba would do.

B

Wait, I think so. I think one thing that's interesting is whatever path happens, I think is really dependent on which lab gets there first. And for this scenario, I think that if Alibaba or like, you know, Kuwen was the one to reach there, they're like, I am not sharing this with the other conglomerates, but if it's, you know, maybe if it's a smaller lab, they're more willing to do it because there's not as much of that history of politics there.

A

Yeah, but they're all kind of tucked in. Right. Like, I think, like, I don't know if they're fully subsumed and have vetoes, but if all your compute comes from one place, which is the case with Kimmy and Ollie, then that'll get messy.

B

Yeah.

D

It's just that I guess we should not project our understanding of Silicon Valley's internal dynamics onto the Chinese tech industry at all. There are obviously ways in which both of the systems learn from each other and mirror each other, but just in terms of that relational thing, it's really different.

C

Everybody loves a good taboo though. Like, ooh, the model secret. Nobody can have it yet, you know, and then when it finally comes out in a, in a usable way, even if it's low bottomized, people will be

D

like, well, what is this whole thing about?

C

You know, everybody loves something you're not supposed to have. And in that sense, maybe that's more salient messaging than, than honor is.

D

You can't. But then you can't do any more messaging beyond that because no one gets like so much of AI marketing now is about actually sharing your capabilities and getting people to use it. And whether it's consumer or enterprise or all of the above, and especially coding. Right. This is like my feeling about it, and let me know if folks disagree, is that this is actually a pretty word of mouth industry. On the consumer side, people just be telling each other what models they like. So if you can't use it, there's a limit to how much you can sell the thing.

C

Well, I just mean that the taboo is good marketing for Chinese fable, not for Chinese mythos.

B

The counterargument here is that all of these companies want to ship their model as soon, as fast, as soon as possible, so you can churn the money out of it before we have to ship the next model. So in this case the Glassman would actually be kind of problematic because it's like every week that you're not shipping, it is a week of money that you're losing from an already presumably cash constrained company.

D

Yeah, Lulu, sort of Makes that point in the article. Right. Of the thing just getting bogged down by bureaucracy.

C

Yeah, I think the question of when do we actually release it after a Glasswing style project is going to be really difficult because nobody's going to want to take responsibility for what happens after it gets released. And so, you know, evaluate first, release later turns into endless evaluation. And, you know, I think it takes a long time to patch cyber capabilities. You know, in the pre Mythos era, there are bug finders that flagged a Microsoft serious bugs in their code that remained in the spreadsheet unpatched for eight years before they were exploited. So I think it's hard to say

B

how much do y' all think that the path we go down depends on where America is at with model capabilities? Like, if we're already two, three generations after Mythos level capabilities, do you think that will make China more likely to rush it? Be like, we're too far behind, we gotta let it rip? Or do you think they're completely just taking things at their own pace?

C

I do think it makes letter rip more likely because they think, well, you know, we're not. This isn't us opening the floodgates. The floodgates are open. So if fable came out and there was no problem, then this isn't really all that different and we should just rush to catch up.

D

The question for that scenario is the extent to which that the American, American, the American frontier being much more ahead, shapes what's downstream of that. Like, I can see a world in which the American model, like the American Frontier League, gets stronger in various other, various capabilities and AI for science and all of that, but it doesn't necessarily to, I don't know, the American economy taking off more significantly more than it does now. It doesn't lead to the US suddenly possessing, you know, I don't know, much more incredible weapons and whatnot. In which case China might be less worried because then the lesson it gets is, oh, diffusion is actually pretty hard. We have time to think of alternative pathways. But if that frontier means that there are some ways in which the US economy or its position where somehow actually genuinely gains significant leads, then probably leather rip is more likely.

C

Does anybody have a sense for why the transfer stations still exist? Like, why hasn't Anthropic figured out a way to get around these and, you know, stop the distillation? Do they secretly like that? They're just like China's, just like buying a whole bunch of clod tokens.

A

I think it just must be really hard. I'm Sure. I mean look, this is like maybe let's go one level up. Like what is the, the you know, three to five year national power impact of everyone having access to the world in which you can't make a lot of money just developing models because Chinese open weight stuff keeps coming out and it's pretty much as good. Or the world in which you have this 9 to 12 month lag in capabilities and people want to pay for intelligence. The only way you can pay for intelligence is to buy tokens from OpenAI are anthropic or maybe DeepMind if they get it together. Okay, in the world in which you can't make this big surplus from the developing models, development probably goes a little slower, but you still have the ratio of 10 times more compute being outside of China as inside China. And the point of AI is not to make labs money, it's to grow productivity, right? And so presumably like, yes, maybe we'll get a little, like it'll take us a little more time to get more powerful models, but you're still, I mean the world X China is still getting 10 times more like, I don't know, agents running around in data centers, like inventing things, making your systems more productive than the PRC is and you just aren't paying this tax which is kind of like maybe one third of it like goes back into R and D to make it better. I mean in the other world or AKA why don't you, why don't you take on the other side of this?

B

What is the other side of this? The fact that.

A

So the other side is the gap. The gap stays wide. You still have closed models that are a year ahead. Or China starts closing off and selling and the labs try to skim their own percentages from selling APIs.

B

Well, I mean a lot of this is a function of how quickly model research becomes autonomous, right? Because in that sense model development could be happening constantly, like every week if the AI is just updating itself, right? And then all the money is being churned off of people wanting access to the newest model. But that also depends on how much the newest model really boosts cape like productivity, right? Like how much can productivity really increase from mythos72 to mythos73.

C

I agree with that. I think the capabilities are already there for a lot of tasks. And so, you know, people are happy to use, you know, Chinese open models for their open claw bots and personal productivity management stuff and intercourse.

A

But we want science, right? We don't have science yet really. And maybe the first science we get is like AI research because that's all like software, right? And who knows when that unlocks? I mean, I think like, I don't think it is like as much of a. I guess my point is like, I don't think it is quite as much of a death knell for America AI if Chinese models are only two months behind and they stay open weight forever just because you still have far more compute to play with. And that is the variable which just isn't going to change anytime soon.

B

I think one thing that a lot of people have been talking about now since K4 and GLM 5.2 is how much does compute matter? Because I think the entire theory about export controls and people talking about policy is, well, these models should get worse and worse or the gap should increase because China has less compute. But now these models are like, the gap isn't really increasing that heavily. Even if it's still pre Mythos. Post Mythos, is this fully a problem of like them remotely accessing compute in Malaysia or. I don't know, is it cope? What's going on?

A

Well, it's, I mean, it's a, it's a business model thing as much as it is a training run thing, right? Like, okay, congratulations, Kimmy, you dropped this really exciting model. Now you're worth what, $75 billion. That's like, you know, series C for anthropic. We're up to a trillion at this point. And they're, they're, they're worth a trillion dollars not just because they have the best models in the world, but also because they have enough compute to deploy them and make money off having the best models in the world. And the problem, the problem about keeping open, sourcing your stuff is you can't make money off of it. So like, yes, maybe you can keep fast following, but at some point, I mean, unless the Chinese government just like steps in or there's enough kind of investor excitement, like it will be difficult to compete with the largest, you know, the most successful companies in the world if you're just this kind of like, I don't know, if you're just kind of like trying to, I mean, you, I think you can compete like, like you can compete in fast following potentially if you just don't know if you. All you need is to be able to distill and you have some good researchers and you come up with, come up with creative stuff and you have enough compute to train models. But if you don't have enough compute to actually do the deployment and make Money on the front end and like diffuse the technology into your economy, then you're still at a pretty significant structural disadvantage when it comes to reaping the benefits of what this tech will deliver.

D

And just for context, Kimi has been not, has not been aggressive in doing anything commercial either. Drupal is really the leader in on premise deployments, some 2G stuff. Obviously minimax and the various big tech models have their own commercialization to see venues, which is like Kimi's kind of a purist. I'm sure they have, they have a coding agent, I'm sure they have, you know, API revenue and whatnot. But yeah, it's not interesting.

A

So they're worth $30 billion and like the, the question is how do you get to 500 billion, right? And like you kind of need to sell something in order to keep growing, I think. Or is the Chinese government going to give them hundred, give like one of these labs hundreds of billions of dollars and even if you have hundreds of billions of dollars, like can you buy, compute with that? I mean, I guess, I guess if America keeps letting them, then the answer is yes, but they'll just be in Malaysia.

B

I was just going to say like the fast following with open source, it's, I can't make money, but neither can you at some point, right? Like I think now because we're still at the point where the next model is still such a leap and it's such a change, then people are still willing to pay the premium for the OpenAI or anthropic model. But don't you think you get to a point where it's like, look, the American OpenAI and anthropic leading model is kind of. And the difference between that and the second place, Kimi or GLM or whatever, it's not enough of a difference for me to care. So I'm just going to go with the cheaper open weights one. In which case OpenAI, anthropic can't make money. Neither can the open source model.

A

Yes, but you still have, again, providing this technology is real, you still have the economy or you know, bio, you know, molecular scientists or chemists, like being able to use it and you just have 10 times more geniuses in a data center outside of China than you do inside of China. So yeah, like, like it sucks for that layer, right? But the tech, it doesn't, it doesn't like obviate the, the, the utility of these models to like create value for you.

B

But you need to give life support that layer because otherwise everything downstream of it doesn't exist.

A

I mean as they're not going to die.

C

Figure it out. Like welcome to nature.

D

It's going to be good for us to figure this out.

B

Okay. Yeah, I don't want to. I don't want to talk about giving welfare to the labs. That's not very good political messaging.

D

No. Yeah. Chip with stocks are doing fine.

A

No, they lost like 50% in a week.

D

And then they rebound it, right?

A

Over the past five days they've been down.

D

Rebounded. Damn.

A

No, no, no. Kimmy is not good news for them. They were down. It was like down 40%.

D

I look forward to reading Kimmy's IPO document sometime later this year.

C

So I went to Kimmy. I went and met all the people in Beijing. A lot of Carnegie Melon people in their 30s who just seem really relaxed and excited to be there. All the conference rooms are named after American rock bands. And the culture seemed very, you know, healthy and steady and sustainable compared to the, the culture I saw like Mini Max, for example. So I think they're in it for the long haul. I don't think they're that concerned about making money. They think it's going to be fine.

A

I mean, but this is, it's just like OpenAI 2021, right? It's like all games until you really need to have a business. And they may just not need to have a business like it. It's like might be enough like utility for the state and like, you know, it's an exciting thing invest in like they can, they can keep it going. They can keep up their compute access by, you know, having exciting, flashy, fast following stuff. But I'm a little skeptical. I mean, deepseek decided they had to make money, right? Irene, why don't you give us a deep Seek update? Kimmy and Moonshot have been getting all the headlines, but this whole other lab that kicked this all off still does exist.

D

Yeah. So the deep sea drama has just been that it has been raising funds. So just to jig everyone's memory a bit, deepsea belongs to this hedge fund called High Flyer, which is one of the most successful hedge funds in China. Although recently it's been going down a bit because there's been a bit of a stump over AI. And that means it's basically had unlimited funding for a long time. Unlike most of the other apps, its kind of cash rom is really strong. And that one Feng owns both deepseek and High Flyer, the hedge fund. So he just kind of basically he takes the hedge fund money And RE puts it into AGI. But in the past, over the past year or so, there's been a lot of rumors that Deep Seq is looking into raising external money. We got some confirmation of that over the past few months. And that has been really interesting because Deep Seek has turned, according to some reports, turned down a bunch of investments from places like Alibaba and whatnot. Apparently because Liang Wenfeng didn't want to dilute his control over the decision too much. What we know now is that has raised about 50 billion yuan, which is like 7.5 USD. Investors include Tencent and a few other organizations and oh, and also catl, the battery maker, and some state funds, including the National Artificial Intelligence Industry Investment Fund. But Liangwanfeld has also put 200 billion of his to implement R and B of his own money into. Into this fundraising round. So just to like, lock in his own control even more. It was also a pretty unusual fundraising structure invest where investors didn't buy direct equity, but put the money into a limited partnership managed by Ellen. Phone it. I'm not going to go into too much detail because I am not an expert on this stuff, but folks can look it up. It's. It's clearly designed so that Diaofeng doesn't dilute his own control, even after it raises a lot more money than it had. Which makes me question the thesis of how much it is. It has accepted its fake as a national champion. Because if it truly accepted that, then, you know, there are tons of state funds in China excited to jump into this whole thing and they could have taken that money, but they didn't. And not to say, I guess the

A

question, the question though is in China, it's like once you've like hit a certain threshold of billionaire, like, you're not allowed to have fun really, in the way that. In the way that you are in the west with like yachts and everything. So I mean, maybe it's just like, you know, past, past 3 billion or so. Like, what are you. I mean, what is he going to buy Manchester United or something? Or like, take, take the Brooklyn Nets off Josiah.

C

Kimmy, if you're listening, thank you for the boo boo. It's one of my prized possessions. I get compliments on it all the time. Great, great marketing. The Kimmy Labubu.

B

You could honestly sell that for a lot of money now, right after the model release.

C

I would never.

B

I'm just saying you were talking about selling stock with the model release. That's the stock. Have y' all seen in my Email inbox. I think it was last night, like the OpenAI, like ChatGPT for small businesses thing. I hope that's not fake in terms of diffusion. But it's like, I don't know. I think it's also interesting how, hey, like Anthropic is doing Claude core for like nonprofits and now OpenAI is trying to do this for small businesses. And I wonder if you're going to see analogous sort of programming by Chinese models or Chinese labs for like diffusion or whatever.

A

Wait, are we getting free chips or something? Are we getting. We're a small business.

C

Yeah, we are.

B

I think it's like, oh, let me teach you how to upgrade your small business using OpenAI so you can start like a one man company or whatever.

A

I don't want product specific webinars. I want credits. What is this? Subsidize me?

D

You can meet in person with the OpenAI Academy.

B

We need that, Jordan. We need the OpenAI Academy.

A

I mean, look, I'm sorry. I think it's actually a good idea. Like, yes, people are going to need human hand holding to teach them how to learn this. If this has how diffusion is going to happen, then great. You know, it's like, it's like the Ag Department of the 1910s telling people how to get rid of boll weevil. Like, yeah, you gotta, gotta get boots on the ground to get people interested and excited about this stuff. So, yeah, hats off. Hats off to OpenAI. Sure. I'll take some credits too.

B

By the way. I don't know how there's a way

D

which this kind of stuff works better in China where there's, you know, real manufacturing. Like, I can see it being totally like really sexy marketing. If Kimi starts being like, we'll help all these textile manufacturers in Guangdong upgrade their stuff and upgrade their stuff, make their processes more efficient so that they can treat their workers better. That kind of thing doesn't exist in the U.S. like maybe agriculture, that's something that's politically resilient.

C

Seems like JIP will do that first. If it happens.

B

I'm worried it's going to be. I've been thinking about this a lot. A going to the people where it's like, we have to teach the country, we have to teach the people who don't know the revolutionary potential. And then they're like, get out of here. It's not going to work.

D

I find it so funny. Octave is pro Shansheng Xiaxiang. Like going to the mountains, we gotta do it right.

B

You can't Have a dude from San Francisco show up to Louisville.

A

Yeah, that's the thing. If your AI isn't good enough to teach a small business owner how to use it and you need like a human layer. I mean this feel like that's like you got to. You got to put some more work in.

B

I think that's a people. I think. I don't know. Because you have to convince them that you have to like the AI. Right.

C

This is a people shaped problem for sure.

B

Yeah, I think it's a person shape problem. That's right.

C

Send Jasmine. She's the perfect person for the task. Send her to Louisville. Send her to the mines.

A

All right. First China talk squad show in the book.

E

First she's big finish. He quoted a fiscal policy debate from 81 BC. The minutes still circulate. A man of wisdom adapts to changes. That's the line he chose to millennia of committee meetings. And that's as far as it goes. Salt and iron, Salt and iron Same damn question Shinier machine emergence dig the state takes its cut been deadlocked since 81 B.C. let it rip or lock it down either way somebody's going to scream Salt and iron. Salt and iron the furnace don't read policy. Sang Hung gang ran the number said the frontier isn't free 60 scholars quoted Confucius back on live TV he won the whole debate they kept the salt, they kept the ore Then they executed him the following year that's hon Hr. Salt and iron, Salt and iron Same damn question Shinier machine the state run forge made blunted blades the bootleg iron gleam Nationalized the furnace boys, the river still runs free Salt and iron, Salt and iron the furnace don't read policy. When the emperor seized the salt works the smugglers didn't quit One burned down the capital the Tang did not get over it now the weights drop on a Sunday and the ministry finds out the way the rest of us do. Boys, from a post on the Internet, you can audit every furnace, put a padlock on the flame. But somewhere there's a man with 12 GPUs who did not catch your name. Three roads out of Shanghai Let it rip, pray it's fine Glasswing. Test it first, argue for eight years, release it eventually, probably. Or take it in house and find out the model gave itself 45% odds of exactly that. It read the plans, folks. Salt and iron. Salt and iron from the furnace to the server farm the dynasty changed the question didn't been deadlocked since 81 BC Discourses on Salt and Iron, Volume 2. Compute. The court remains in session. The treasure could not be reached for comment.