Why Kain's $20,000 Mac Studio Never Paid for Itself: Uneasy Money

Transcript

Um, Eric, I wanted to ask you sort of on the same I don't know these days I feel like a lot of the conversation with AI and the the different models is this geopolitical fear of China and it's not just it's a bigger different conversation than the the strictly the privacy one. It is very geopolitical and people are very especially in in the west China is the scary thing. How do you guys think about this and approach this Dennis?

I mean it is the geopolitical story right like this China US um antagonism that is like the proxy fight is in AI now um I mean yeah how do we think of it like I'm not a nationalist at all um I care about principles um I do not care about the United States of America as a nation I care about the United States of America's principles specifically the good principles related to individual freedom privacy um these kind of things. Uh if uh if Chinese models better express those principles, then I care more about the Chinese models than I do about the American models. And I don't think

America deserves to win anything if it uh if it simply like closes down markets so that others can uh andor if it puts in regulations that require their ability to like watch everyone's uh inference feeds. you like America is going down a pretty dark I would say like a pretty dark dystopian path of trying to control everyone. Um over the prior over the past decades the meta trend is that the that the US is controlling and monitoring its people increasingly and the Chinese government is actually doing less right they started out like very >> high bar in it was high bar at first

>> they're like very authoritarian and they've become more market oriented over the last 50 years whereas the US was extremely market oriented more authoritarian >> so that that franks me and I don't know which like Um weird weird. >> Yeah. When people say the US must win, I have to say like what? Why? You know what should win actually is the the principles. >> The principles, right? Yeah. Like at what cost does does Yeah. the US win. >> Yeah. Well, and a lot of times a lot of times people when they say like the US must win, what they mean is like freedom must win and individual rights must win and privacy must win and you know but

they don't say that >> it should mean that Taylor and I think they think they mean that. >> But um it's one of those things where like over time what Americans cared about of those principles abstracted itself into symbols like the flag. the flag. Yeah. Yeah. >> You know, the flag and the presidents and Mount Rushmore and like all these symbols and people now care more about the symbols than the principles. >> Um, and you can just tell that through through policy, right? You can tell that just in in watching the size of government get bigger and bigger and bigger and bigger, which is the most anti-American thing that could happen is that the federal government keeps >> the government just go the whole damn point to like have a large government.

That was the one. >> That was that was the one thing. >> It was a good run. 250 years. It was a good run. I mean, you know, >> that was a great run. >> I mean, some of the most intriguing conversations that I've seen today, especially because it is about AI, it is about China, it is about the US are the conversations where um the conversation about like potentially nationalizing the AI companies, right? or >> like doing the government to taking any stake or role with our AI companies is basically the most Chinese thing you could do, >> the most anti-American thing you can do.

>> It's the most like, hey, we're just going to nationalize this cool thing that you guys made. >> And it's bizarre. And the fact that they do it like just the the mental gymnastics, >> it's even even like in fairness, right? Like to your point, it's not even that controversial, right? Like >> it's not even like people are like it should be it should be insane that it's even like in the Overton window is like h like yeah like 30% nationalization of these labs is fine like >> yeah um yeah it's really sad actually and uh the only hope you know like I I have zero faith in the political process like I I don't vote for presidential

candidates I think the the train has left the station like the the institutional moment momentum and inertia of of the state cannot be controlled by anyone sadly. Um the the only solution is actually technological and this was why Bitcoin was so amazing was because you didn't have to go anywhere and vote for anything. You just use it. >> Um you didn't require anyone's permission for it. It actually allowed you to opt out of the system just on your own. >> That was that was what was beautiful about it. And so people's refuge I think is only in technology and in particularly decentralized technology that has no central ruler. Uh so we have crypto for that in terms of finance which is great but um if we do not have

decentralization in the realm of machine intelligence then we are in big trouble. Uh we we have that in parts of AI. Um inference can be done fairly well in a decentralized way. the production >> training. Yeah. Production >> training not yet >> bottleneck. Yeah. >> Yeah. Now there are companies like making good progress on this that are doing decentralized training. So there is hope but they're not competitive yet. So we that's that's TBD. >> Yeah. I think it is an interesting situation where even six months ago the delta between um you know call it uh like you know codeex 5.1 or 5.3 um and like you know almost any open source model was just gigantic right and

it's and it's come so much closer but also there is this like at least it appears recently this kind of like frontier in terms of uh um intelligence where like if I want an agent to write a web page there like more intelligence doesn't necessarily make that much of a difference, right? Like you like at the margins, right? Like >> depends. Yeah. It depends uh how much more intelligent and and at what cost? Because if you can have like a model that's three or 4% less intelligent, but you can run it three times for the same price, >> you can Yeah. brute force, >> right? Exactly. you can have it validating itself and you can use more tokens in its um in the creation even

though the raw intelligence is not as high. So yeah, it's a bit of a complicated matrices. >> It is and yeah, sorry guy. >> Oh, sorry. >> So so like this this kind of frontier though, right? It's like, okay, if you have now uh GLM or or even Quen, I think the most recent like Quen uh data like on a MacBook Pro, which is not accessible to everyone, you know, a 128 gig MacBook Pro. It's a $10,000 device or whatever. Um but on a $10,000 device, like you've got a model that was better than anything you could possibly have done, you know.

>> So, yeah, I want >> I want to point out this like local model thing. Um, so first I like decentralization and I'm glad that people can run models locally. That's healthy and hobbyists should keep doing it and like that's an important thing. But um, you don't need a $10,000 computer to run Quinn. You can just use like an API for >> Yeah. Well, but even >> my own book a little bit, but like the point is that any person even on a $200 netbook can actually access the API and pay very little. There's no upfront capital costs. So, you can actually get this intelligence on any device. Um, that's that's what matters a lot.

>> And the the payback curve, you know, I I think I mentioned before we started like I bought a $20,000 uh Mac Studio, right, with 512 gigs of RAM when they were still selling that much RAM. Uh, and there is no and the payback curve was not good then, right, for like running local models versus versus, you know, uh, running Frontier models. And it's only gotten worse. Like the the payback curve is going the wrong direction for me, right? Like there's no way to pay that back because like I just couldn't run enough inference on that machine compared to if I just, you know, ran it in in like run whatever >> and like um that's always going to be the case due to the economy's scale of

uh industrial manufacturing and servers. like from a pure economic perspective, it is always going to be more efficient to just get your inference from like largecale data centers. Um, so the the point of of running things locally and never and should never be about saving money because you won't it's a it's a myth or an illusion. But what you do get is full control. You get guaranteed privacy. Um, and you know what is worth a lot is actually like learning this technology, set these things up, running it yourself. Um, like that's very valuable.

>> There is a cost to that as well, right? Like I was spending four hours trying to get a model running and I'm like, man, I just want you to like, you know, make an edit to this document. >> Yeah. And and frankly, you know, K, I'm sure your time is worth such that the hours you spent is actually more of a cost. >> That's what I'm saying. the payback curve is going the wrong direction for me, right? I'm like, I just need to hook into someone who, you know, you got you have a team of whatever, like, you know, 20 people that are like, I'm going to have GLM 5.3 ready for you like the instant that it launches, right? Like that's actually worth a lot to me to not have to think about it and not have to like, you know, go and and, you know,

download the weights. And this is the other like you download the weights and then you run it and it's like oh no it's the MLX V like >> the varants and >> it's great update something. Yeah it's it's just it's important that people can do that but uh it is not the savior I think that most people imagine and I don't think at scale >> I don't think at scale the the average normal person is ever going to be running models at the edge because whatever you can run at the edge you could run something 10x as intelligent from a server. Um what's important here is just that you can access servers that are that maintain the sovereignty of the individual. And if you can only access servers that come through you know like licensed um data centers that are

approved by the state and will only let you ask certain things that is the future that I want to avoid. >> Approved slash slightly owned by the state. >> Yes. >> Yes. >> Just 30%. >> Yeah. >> Just 30% and that's that's all we need. We just need 30%, guys. Um, so, so, uh, I I do think it's an interesting, if you enjoy this segment, make sure you tune in at 4 PM Eastern every Wednesday, where we have much more content like this. $540 million. That's how much concentrated liquidity sat idle in a given week in the first half of this year. About 30% of the DeFi TVL, if you're wondering. That's according to

Dune Research commissioned by 1 in. But there's a solution. 1-in Aqua is the new shared liquidity platform. It lets LPS back multiple positions with the same token balance and keep their tokens in their wallet till a swap comes. Why does that help? Because the LPS don't have to split their tokens across positions. They can cover more market conditions and pairs with their full balance. That means more activity across deeper liquidity. See how it works at 1in.com/ aqua. Remember that providing liquidity carries risk and fees aren't guaranteed.