00:00$250 million allows us to compete with the large AI model providers, just providing smaller, much smaller and more efficient ones.
00:08You need to have such a large, let's say, amount of money just to try to get into the market and distribute properly.
00:17And do you see yourself competing then with the large language models that your technology compresses for your customers?
00:22Not exactly at this time. We are not building super large models, something that we could do, but we are providing cheaper.
00:28I mean, half the price models and then models for the devices, which is an untappered market so far.
00:33And what is the business model then? Who are you shipping to? Who are your customers? How do you monetize this innovation?
00:38Okay, that's easy. I mean, we just get some open source model. We compress that or we can just tune it up as well.
00:44We compress and then we resell on a fee. Or we can deliver through the cloud providers, for example, API, and then we charge by the token.
00:53And how deep is the technological moat in terms of your competitors? What you're able to do versus your competitors?
00:59I mean, this technology comes from a completely different part of the science. We came from the quantum industry, and this is what is called quantum inspire.
01:08Algorithms that came from a different part. So it's like technologies, like if you are just still making just holes with a show well, or if you use a machine just to do that. This is the example.
01:20How much demand are you seeing for these compressed large language models?
01:25For the compressed larger, I mean, the market is going to explode. Look what's happening in the Netherlands.
01:30They are restricting the consumption of energy because of the AI. So this is happening right now, according to another news outlet like Financial Times.
01:41The point is, okay, what are doing the small companies that want just to use a cheaper LLM? They have to use cheaper AI model builders, and that's us.
01:54On the other side, if you want a model just to be on your iPhone, on your, let's say, Raspberry Pi, or even in your drone, for whatever reasons you think that you can put in a drone, an LLM, okay, that means a super small, a nano LLM, which consumes very few energy.
02:12That's us.
02:12And that's edge computing?
02:14That's edge computing.
02:16Nano large language.
02:17Yes. And this is a market that doubles the size of what we are seeing today in the cloud.
02:22When does that market come and mature?
02:25It's coming now. It's coming now. If you just have a look around, okay, you will see some names from device manufacturers, large American one, some one starting with Hewlett and ending with Packard, okay, that is precisely on that.
02:40But we have more. This is the reason I'm just going to America also.
02:43Yeah, yeah. And you're off to the U.S. and maybe raising additional funding. Talk to us about, just break down the numbers in terms of the potential efficiencies on energy.
02:51We know energy is a major constraint in terms of these large language models. That's a key problem that you address.
02:57And then there's the efficiency and then there's the cost. Just give us broad numbers in terms of what it means on cost reduction and energy and efficiency.
03:03Okay. Compression, 20 times. I mean, just the final model, 5%. Energy savings, half. From 50 to 80% of savings. Cost, exactly the same as energy.
03:14Okay. What does this mean? This is a threat, is it not, for NVIDIA and Jensen Huang?
03:18It's not. Well, Jensen Huang is a very good, let's say, GPUs. But if you try to put Jensen Huang on a drone, do that.
03:26If you try to put Jensen Huang on a drone. On a drone. You mean his chips? His chips on a drone.
03:32Yes, I mean, probably it's going to consume all the battery just at the beginning.
03:35Okay. So this is what we are just, let's say, fighting now.
03:38Okay. But NVIDIA's CEO is touring the world and basically saying you need to build more data centers and more infrastructure.
03:44Yes.
03:44But you're compressing the models, so that means less data centers, less energy, less chip use.
03:49I don't think that it's going to happen that way. I think that we are going to still build data centers and more data centers.
03:55But have a look from the point of view of that data center. You have the same energy, but then you can serve double the outcome.
04:03Because we are just, for a fixed amount of energy, we are just serving double the customers.
04:08Okay. How much competition are you facing from some of the big tech giants, like IBM, like Google?
04:13No. Google, I mean, we arrived to agreements with AWS, for example, and we are just synergic with them.
04:20Because they are serving the super large models. Okay. But they fail. They fail. They are not serving right now the smaller models.
04:29And this is what we are just providing. In the future, maybe, but it looks like those guys just are focusing on the artificial general intelligence and so much.
04:38And you talked about the fact that you are quantum inspired, your technology, quantum inspired.
04:43How do you see quantum computing, large parts of it still unproven, quantum computing and AI, that story evolving in the next two to three years?
04:51This is nearly a sci-fi in the sense that quantum computers are not powerful enough to do that.
04:57But yes, there are some scientific projects just to put a small LLM, our smallest LLM, inside a quantum computer from a French manufacturer and see what happens.
05:08This is going to work very badly, but that is going to be the first in a row.
05:13So at the end, if you put a super small LLM model inside a quantum computer, which is, I mean, a very dumb model.
05:21I mean, this looks like a million monkeys just with the typewriters. Maybe, I don't know. I mean, this is sci-fi so far.
05:29You know.
05:30I mean, this is a major level.
05:31In fact, is something I might have a few years ago.
05:32I mean, this is a Huurtz you is not a scientist.
05:33This is what I mean by the bottom music is definitely seems a large number.
05:34Although over there are leaders, the whole world is a huge thing.
05:37This is high level, all right.
05:37So, I mean, it's a smoothverse that are
05:53people in a way to actually you know new people.
Be the first to comment