So I have somewhat of a hot take. Good. The big labs already are behind. In two very important ways. What are the leading personal agents? Muse and Instinct. Not the big labs. I know they don't launch DOTS. We'll see what happens. But Muse and Instinct are out in front leading in those two. When you look at specific applications, legal, finance, etc., the large language models are not the leading models. They're leading in general purpose, but when you start to get specific, they're not. When you look at it on a cost, perspective i can just go off of our data if i process 500 million tokens from uh on claude about five thousand dollars if we process the same 500 million tokens on our own model and our own hardware it's 97 cheaper so what i think is going to happen but i think the fact that you and i don't agree on this actually is like maybe that's the story of like nobody really knows right is You're going to start to see the large language models say, we got to pick our battles. Where do we go and fight to win? Because just having the general purpose model is actually not going to be as valuable because of the commoditization. And so we should accelerate into specific use cases. Anthropic recently announced that they're going to go really heavy into bio and all the biotech stuff they want to go do. it feels like they think that is a massive opportunity if they can win there that is trillions of dollars of value that they can create OpenAI is saying dots and personal assistant and they've always been more of a consumer type company than an enterprise type company and so i don't think it's just like hey go all in on one specific use case but it does feel like they're now realizing that the general purpose model can serve as like a foundation but the monetization of this and the sustainable moat is going to be in the applications and This goes back to the idea of in crypto, if you go back to 2017 or so, there used to be this thing called the FAT protocol thesis. And the whole idea was like all the value accrues to the protocol layer. And I could be wrong on this, but I just continue to go back to the FAT protocol thesis. The equivalent in the AI is like the FAT model lab thesis, where all the value accrues to the model lab. What we found in crypto was there was definitely some value that accrues to the protocol, obviously Bitcoin, Ethereum, etc. but actually a ton of the value accrued to either the infrastructure underlying it in terms of miners and things like that or the application layer it feels like that's the same thing happening here where as the technology gets built out we have tons of infrastructure that's making money we have tons of applications that are yet to be built but I just find it really hard if like Anthropic and OpenAI just stayed as general purpose model providers I just think that is like a cutthroat business it's like kind of a race to zero who can get the you know highest level intelligence for the cheapest cost and so like what's the moat what is the the the you know thing that defends against everybody else but i don't know maybe you think differently i do um i don't
The equivalent in the AI is like the FAT model lab thesis, where all the value accrues to the model lab.