The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
00:46:54 · Harry Stebbings, Daniel Dines
[00:46:54] S1: be clear then, so we will actually still use the core provider, which is OpenAI and Anthropic. It'll just be deprecated older models. [00:47:01] S0: I will still use Anthropic and OpenAI AI with their cost efficient model, but I will have a verifiable backup on open source all the time. I should be... As a responsible enterprise, I should be able to switch models. I cannot be locked in. [00:47:19] S1: Can I ask you, we invested in? I'm just checking my portfolio against your brain. I believe strongly in open models, and I think that every company, not every company, but mid to large scale company will have their own model, own their own intelligence, and feed their own data into it. And, you know, I think it goes to the statement of kind of owning your own intelligence, not renting it. That's why we invested in fireworks, and I believe in the open model ecosystem. [00:47:41] S0: You know, I'm a big fan of fireworks, and we are using them quite a bit. [00:47:46] S1: Do you like
00:47:46 · Harry Stebbings, Daniel Dines
[00:47:46] S1: Do you like [00:47:46] S0: them? Yes. We like them a lot. And I'm I'm a big believer that an enterprise should distribute their bets, and one of the bets should be on open source and very importantly, should be on this map of work. Because think about, if I wanna train a model, my own model with who am I, I need to have this who am I very well documented. I need to create this manual because I'm training one model today. But in next two months, there is another better model, base model coming into the picture. How can I do transfer learning from my old model into the new model if I don't have the data and the exact menu? I cannot. Well, there are, you know, terrible losses when I do this. The real investment for an enterprise is to creating this map of work that documents how they actually work. And with this one, they can train their own models. It's in fireworks or other provider, doesn't matter. But this is their IP, and this is their core data. Makes sense? [00:48:48] S1: And so you do, sorry, just so I understand, so you do believe that companies, and a lot of them, will have their own models with their own data?
00:48:48 · Harry Stebbings, Daniel Dines
[00:48:48] S1: And so you do, sorry, just so I understand, so you do believe that companies, and a lot of them, will have their own models with their own data? [00:48:55] S0: I do believe that they will at least have their own models as a backup to frontier models. To me, where I'm not clear, if I can provide the same cost efficiency with my own model versus a cost efficient model from Anthropic and OpenAI. Because I think these guys are in a position to truly optimize large infrastructure. So part of their business model will be to deliver, you know, more intelligence per dollar than even I can squeeze from my own models. [00:49:28] S1: If you have highly specific data that is exact to the request that you have, which is your data, I think you'll get more token efficiency with your own model than you would an optimised model. [00:49:41] S0: Only if you are training your models, that might be true. And only if you can deliver... If Fireworks can deliver at a large scale and in a very optimized way. [00:49:52] S1: Would you invest in Fireworks at $15,000,000,000? [00:49:55] S0: Probably, yes. If this hypothesis of open model is true, which I believe is true, I think they are under... Undervalued. I think they will have to get very soon in this big game of securing compute. Because if they don't secure compute, I don't understand how can they give me the inference at the scale that I want. What do you think? Because you invested in them.
00:49:55 · Daniel Dines, Harry Stebbings
[00:49:55] S0: Probably, yes. If this hypothesis of open model is true, which I believe is true, I think they are under... Undervalued. I think they will have to get very soon in this big game of securing compute. Because if they don't secure compute, I don't understand how can they give me the inference at the scale that I want. What do you think? Because you invested in them. [00:50:18] S1: I did. I think you're absolutely right that they need to move into the compute layer, and I think Lin is doing that, I'm sure, very soon. [00:50:23] S0: So they will have to raise, you know, Tensor will be the last guy. [00:50:27] S1: Yeah. And she did that at Facebook. I think that's a unique thing to this team. They did compute securing. [00:50:34] S0: Yeah. It's a great team. We really like them. [00:50:36] S1: Great fucking team. [00:50:37] S0: And we worked with them before the big hype. [00:50:39] S1: I also think the data providers are massively underpriced and underappreciated. Macore and Surge in particular. Everyone's like, oh, they're commodities. You're just buying data. Data is the most important thing to feed model quality.
00:50:39 · Harry Stebbings, Daniel Dines
[00:50:39] S1: I also think the data providers are massively underpriced and underappreciated. Macore and Surge in particular. Everyone's like, oh, they're commodities. You're just buying data. Data is the most important thing to feed model quality. [00:50:54] S0: But what's the difference? Think one thing is storage, and one thing is understanding of the data. Because if I have a storage, I can have a tape, and I can put data on the tape. Would we invest in a tape company? I don't think so. You need to invest in the intelligence that understands the data and feed the model and, you know, extract the right data at the right time, feed feed really the model with the data that is needed with the context. Because if you don't have if you have just data, but you don't have a way to create a really good context to give the model when they ask something, it's useless. I I think you would say that they have more data than anyone else across more categories than anyone else. So when the model requests highly specific data, because of the breadth [00:51:38] S1: of their library, they're able to provide it in a way that others aren't. [00:51:41] S0: If it's their own data and is valuable for models, I'm sure the models will buy will buy that data in an instant. [00:51:48] S1: Can I ask you, what have you changed your mind on most in the last twelve months?
00:51:52 · Daniel Dines
[00:51:52] S0: I didn't understand this necessity to have a manual in order to work. That was maybe the biggest breakthrough in my understanding. That every time I'm running a query towards AI, AI should have at their disposition the entire way, you know, my company work or this particular process work. When I realized this, I understood also that this thing that it's the the biggest differential between memory and true learning. This is how we started the discussion. And I'm not sure I really make a point, but it's a huge difference between just laying down, having a scratch pad, or being transformed by an experience. And that's the thing that I realized the most. And I think I realized also what's kind of human for us Because I experienced a lot with AI writing, not code, but writing a book. So I've been through different styles. I I understood a lot how to prompt them. AI doesn't have a style. And you you realize why they don't have a style, because they are an averager of anything. In order to have a style, you need to have kind of a body. You need to have individuality, because we are the choice that we made and the choice that we don't make in a sense. So you need to be transformed because otherwise, I will just ask AI, read this book and write in the spirit of this author. And it's not really working because you need to be transformed by the experience. So to me, this is gonna be the biggest breakthrough in the AI technology when I can have models at the size of mythos being transformed on a job, being, you know, putting in a laptop. And it might be possible, who knows, you know, the pace of technology. Maybe twenty years from now, I can have a 10,000,000,000,000 model that is my own model and is getting transformed along with me. But we need to see. I think there might be a few series of innovations to get there. Because I wanna give you also an interesting data point. AI is solving very interesting math problems that humans didn't solve before right now. But AI still, it's not capable of creating frameworks. I don't know. Relativity is a framework. Okay? And I was thinking, why so? I think one of the main reason relates still with this not being transformed when you are on the job. When I'm writing a book, I am being transformed by the act of writing this book.