Latest / The Paul Barron Crypto Show / Decentralized A.I. At Warp Speed🤖Gonka INTERVIEW David & Daniil Liberman

Transcript

Decentralized A.I. At Warp Speed🤖Gonka INTERVIEW David & Daniil Liberman · The Paul Barron Crypto Show

• 0:00
  Speaker 1: Let's get into AI compute at warp speed. Today. We'll
• 0:03
  Speaker 1: be breaking down a little bit of where decentralized AI
• 0:06
  Speaker 1: compute power is going to be coming from, how it's
• 0:08
  Speaker 1: going to be integrated into the blockchain, all of that
• 0:10
  Speaker 1: good stuff. You guys are going to love it. It
• 0:12
  Speaker 1: is going to be great. Let's just jump right into
• 0:14
  Speaker 1: it today. I want to bring in two guests, and
• 0:16
  Speaker 1: of course that is Daniel and David Lieberman coming in
• 0:19
  Speaker 1: from of course, you guys know it as Ghanka AI.
• 0:22
  Speaker 1: How are you guys?
• 0:24
  Speaker 2: It's been great, Paul, let's jump in.
• 0:28
  Speaker 1: I was up on your website and a little bit
• 0:31
  Speaker 1: about what you guys are doing. First of all, a
• 0:33
  Speaker 1: lot of partnerships, key correlations to what's going on in
• 0:38
  Speaker 1: the AI space. Obviously your statement here, AI compute is
• 0:42
  Speaker 1: the new currency. This is something we've been following for
• 0:44
  Speaker 1: a while, along with a lot of different projects that
• 0:47
  Speaker 1: look at decentralized compute as one of the major things
• 0:51
  Speaker 1: that could lead the AI revolution. David, let's start with you.
• 0:56
  Speaker 1: Can you give me kind of an insight as to
• 0:58
  Speaker 1: what ganka is really trying to do here.
• 1:02
  Speaker 3: Gunkai is decentralized community of harder providers who jointly inspire
• 1:10
  Speaker 3: to create one of the biggest computation networks around the world.
• 1:15
  Speaker 3: You know, we all hear about these massive hyperscalers projects
• 1:21
  Speaker 3: like Stargate or Colossos from XCI, when just one organization
• 1:28
  Speaker 3: built a million GPU data center to supercharge the future.
• 1:34
  Speaker 2: Of AI, and we believe that.
• 1:37
  Speaker 3: It is really important to create an alternative when smaller
• 1:42
  Speaker 3: players around the world can join forces and be competitive
• 1:47
  Speaker 3: with those big organizations.
• 1:52
  Speaker 1: We've seen a few companies that have been fairly successful
• 1:55
  Speaker 1: in doing this, some you know company like Render if
• 1:59
  Speaker 1: you look at their compute and what they're doing around
• 2:02
  Speaker 1: graphics and kind of the innovation on the creator side. Obviously,
• 2:06
  Speaker 1: the general use case for AI tool sets is going
• 2:09
  Speaker 1: to be continuing to ramp up, Daniel. If we can
• 2:13
  Speaker 1: kind of follow along what David was talking about the
• 2:16
  Speaker 1: decentralized ideas smaller players creating a bigger PI for AI compute.
• 2:21
  Speaker 1: What is the threshold for someone or some platform to
• 2:25
  Speaker 1: be able to utilize saganka.
• 2:28
  Speaker 4: So in case of Gonka, we from the very beginning
• 2:32
  Speaker 4: aim to build a platform for the largest available models,
• 2:38
  Speaker 4: like as long as we're using the open source models,
• 2:41
  Speaker 4: anything which is open source can be ran on top
• 2:44
  Speaker 4: of our network hardware and therefore like the entry level
• 2:49
  Speaker 4: of hard risk quite high. This we we're talking about
• 2:52
  Speaker 4: high end and vidage abuse. And in the future some
• 2:55
  Speaker 4: other producers like A six and maybe AMD and more
• 3:01
  Speaker 4: chips will be produced and available on the market, but
• 3:04
  Speaker 4: those going to be professional chips like with mainly with
• 3:07
  Speaker 4: higher amount of shared memory and the bendwates of the
• 3:12
  Speaker 4: members high as well. So something which can run the
• 3:16
  Speaker 4: latest version of deep sik or Lama or or the
• 3:20
  Speaker 4: very open source from open Ei and Google and in
• 3:24
  Speaker 4: the future of some even larger models interesting.
• 3:27
  Speaker 1: So this is in a point of growth. I was
• 3:30
  Speaker 1: looking at the Binance story here Ganka's ai network see
• 3:34
  Speaker 1: significant surge in computing power. This course was just here
• 3:38
  Speaker 1: recently published. One thing that was interesting to me is
• 3:41
  Speaker 1: the capacity here. First of all, how are you guys
• 3:45
  Speaker 1: able to see this kind of capacity growth?
• 3:49
  Speaker 4: Is this?
• 3:49
  Speaker 1: How long have you guys been doing this? What has
• 3:51
  Speaker 1: been the path to get to this level for you guys?
• 3:55
  Speaker 3: The network launched just recently in August late August last here,
• 4:00
  Speaker 3: so we are actually amazed of the size of the
• 4:04
  Speaker 3: community which grew around the protocol and you write like
• 4:09
  Speaker 3: ten thousand, six hundred gps a lot. It's a big
• 4:14
  Speaker 3: data center already right now, just in four months. So
• 4:19
  Speaker 3: this you write that, this project, this protocol is built
• 4:23
  Speaker 3: on the experience which Crept the industry already managed to
• 4:27
  Speaker 3: demonstrate through all these years.
• 4:30
  Speaker 2: A few people look.
• 4:31
  Speaker 3: At bitcoin as data center, but its actually the largest
• 4:35
  Speaker 3: data center in the world Bitcoin in terms of the
• 4:40
  Speaker 3: compute power comprised from data centers together, relating twenty three.
• 4:47
  Speaker 2: Gigawatts of power.
• 4:50
  Speaker 3: It's more than Google clouds, Amazon Clouds, Microsoft Cloud combined.
• 4:57
  Speaker 3: You see, it's so amazing of what people can build
• 5:01
  Speaker 3: together is the community. And then you write there other
• 5:05
  Speaker 3: projects like file coin built one of the biggest storage
• 5:08
  Speaker 3: capacity in the world, and then Render build this beautiful
• 5:11
  Speaker 3: tool for artists to use this compute power around the
• 5:16
  Speaker 3: world to make their movies or cartoons. So for AI,
• 5:22
  Speaker 3: it was the obvious step that we need to make
• 5:25
  Speaker 3: the same and we need to use this principle which
• 5:30
  Speaker 3: called proof of work. When you actually every participant they
• 5:34
  Speaker 3: prove that they have the capacity and through that they participate.
• 5:39
  Speaker 3: And this model previously showed this enormous potential to build
• 5:46
  Speaker 3: the large infrastructure and we believed that the same can
• 5:50
  Speaker 3: be repeated with AI and when we launch it.
• 5:55
  Speaker 4: Okay, so when we lunch, it was just a belief
• 5:57
  Speaker 4: and we launched it with only sixty four GPUs and
• 6:01
  Speaker 4: six participants, and we just believed in this and launched
• 6:04
  Speaker 4: this and then we never we never expected to have
• 6:08
  Speaker 4: such a growth. Yeah, growth was like four times for
• 6:12
  Speaker 4: the first three months in comparable in the four Now
• 6:15
  Speaker 4: we're like five months from the start having ten thousand GPUs.
• 6:20
  Speaker 4: Wren't expected to have like maybe in four or five
• 6:22
  Speaker 4: months from now in the best case scenario. So we're
• 6:25
  Speaker 4: like twice faster than our best dreams.
• 6:31
  Speaker 1: I'm looking here that you guys are currently supporting five
• 6:34
  Speaker 1: different inference models. This is since the launch is three
• 6:38
  Speaker 1: months ago. The usage of these inferts now one hundred
• 6:41
  Speaker 1: million tokens per epoch. So is this how is this
• 6:45
  Speaker 1: going to scale? Is this going to scale at that
• 6:47
  Speaker 1: kind of growth in this shorter time or is this
• 6:50
  Speaker 1: kind of well, we're going to see this amount of
• 6:52
  Speaker 1: growth because it's just getting starting to see the network
• 6:55
  Speaker 1: explain what this is going to look like in the
• 6:57
  Speaker 1: eighteen months.
• 6:59
  Speaker 3: So if this protocol will be able to repeat trajectory
• 7:05
  Speaker 3: which we see in this other projects in the past,
• 7:08
  Speaker 3: which we've seen with bitcoin infrastructure for example, it should
• 7:12
  Speaker 3: reach gigawatts of power through the next years, and from
• 7:17
  Speaker 3: this perspective, it's aimed to become the alternative to these
• 7:21
  Speaker 3: big players. That that's why if the protocol works, it
• 7:27
  Speaker 3: will bring millions of GPS in then even more advanced
• 7:31
  Speaker 3: equipment which in the industry called a six but a
• 7:35
  Speaker 3: six will which will be targeted for EI compute, and
• 7:39
  Speaker 3: this from this moment, the girls can be even faster.
• 7:43
  Speaker 3: That's what we are seen with this previous protocols, So
• 7:49
  Speaker 3: we tend to believe that it is just the beginning.
• 7:53
  Speaker 3: It's definitely hard to say. As Daniel mentioned that it
• 7:57
  Speaker 3: was a really great surprise for us that community managed
• 8:01
  Speaker 3: to build.
• 8:01
  Speaker 2: So much in this short period of time.
• 8:04
  Speaker 3: But looking at the protocols of the past, we actually
• 8:09
  Speaker 3: do believe that it is possible to build even something
• 8:13
  Speaker 3: even greater, both from compute perspective, but then also from
• 8:17
  Speaker 3: the usage perspective that it can be billions and dozens
• 8:22
  Speaker 3: of billions of EA wards stockings generated daily and in
• 8:29
  Speaker 3: this case it will be able to sort through developers community.
• 8:33
  Speaker 3: It will be able to serve million people in kind
• 8:36
  Speaker 3: of GIPT like experience.
• 8:39
  Speaker 1: Yeah, well, I think you have to concern yourself as
• 8:42
  Speaker 1: we're recording, these centralized systems continue to show outages. Decentralized
• 8:48
  Speaker 1: system a little different game, which is kind of unique
• 8:51
  Speaker 1: in itself when you look at the ability to grow
• 8:56
  Speaker 1: and obviously we've seen networks like this in the past,
• 8:59
  Speaker 1: some have grown very fast, have kind of grown and installed.
• 9:03
  Speaker 1: First of all, what are you guys going to have
• 9:05
  Speaker 1: to do to kind of make this version of ghanka,
• 9:07
  Speaker 1: the AI or maybe the Bitcoin, I should say, of AI?
• 9:11
  Speaker 1: What would be the key steps here, Daniel, that you
• 9:13
  Speaker 1: have to do to kind of make this happen.
• 9:17
  Speaker 4: As you rightly mentioned, the other part of the equation
• 9:21
  Speaker 4: is usage, like how many millions of AI talking's being
• 9:26
  Speaker 4: processed by the network, And the usage comes not from
• 9:30
  Speaker 4: huge come from the developers. So our network, the network
• 9:34
  Speaker 4: which we build gone ki, is a protocol which connects
• 9:40
  Speaker 4: the hardware provider with the developers through the API and
• 9:44
  Speaker 4: access to the inference of AI models. So in order
• 9:48
  Speaker 4: for this not to stop and not to stall, this
• 9:52
  Speaker 4: development must continue. The development of both is like the
• 9:57
  Speaker 4: technology on the level of the technology on the side
• 9:59
  Speaker 4: of the hardware would be an introduction of new models,
• 10:04
  Speaker 4: introductions of new drivers, introduction of new features, introduction of
• 10:09
  Speaker 4: privacy a concern SEB technologist and features. This is on
• 10:13
  Speaker 4: the side of the supply of the hardware provider providing,
• 10:18
  Speaker 4: and then on the other side, developers will be using
• 10:20
  Speaker 4: this building tools, if more more successful applications as that
• 10:26
  Speaker 4: it is like something like chupt or like the image
• 10:29
  Speaker 4: generation or voice application at agenttic applications.
• 10:34
  Speaker 5: All of these applications would need to be built by
• 10:36
  Speaker 5: the developers, and we hope that there will be one
• 10:39
  Speaker 5: hundreds and thousands of developers who will be building applications
• 10:43
  Speaker 5: on top of our network using our API in order
• 10:47
  Speaker 5: to create the demand for this network. And if the
• 10:50
  Speaker 5: demand grows, there is a demand and supply mechanics here.
• 10:55
  Speaker 5: And this is sort of a marketplace, an open marketplace
• 10:58
  Speaker 5: where anyone who have the highest efficient hardware to be
• 11:03
  Speaker 5: provided for the network and anyone who want to use
• 11:05
  Speaker 5: it are willing to pay with Gonko coins with the
• 11:10
  Speaker 5: coins of this network for the API inference, creating this
• 11:13
  Speaker 5: economy of utility talking around the network.
• 11:19
  Speaker 1: All right, so you've got the utility token itself obviously,
• 11:22
  Speaker 1: as kind of the infrastructure for the transaction and the
• 11:25
  Speaker 1: use case of these AI models. You look at the
• 11:29
  Speaker 1: decentralized model, which I think is now starting to be
• 11:32
  Speaker 1: even more accepted because people are figuring out how crypto works,
• 11:36
  Speaker 1: especially as you see some of the centralized models failing,
• 11:39
  Speaker 1: which we've seen multiple times occur. When you look at
• 11:43
  Speaker 1: that along with kind of just the transition of decentralized infrastructure.
• 11:48
  Speaker 1: Now with most of these scenarios with AI, it's all
• 11:51
  Speaker 1: been around centralized infrastructure. What makes you guys feel like, hey,
• 11:57
  Speaker 1: that might work, you know, to a certain extent, But
• 12:00
  Speaker 1: the decentralized model is really the model going forward. Is
• 12:04
  Speaker 1: it primarily power usage and access to the hardware or
• 12:08
  Speaker 1: is it something more On a philosophical level.
• 12:12
  Speaker 2: There are two things.
• 12:14
  Speaker 3: First, AI is so powerful tool that it entered our
• 12:21
  Speaker 3: lives and lives of billions of people really fast. And yes,
• 12:26
  Speaker 3: we all like like what it brings us.
• 12:29
  Speaker 2: But at the same time, people.
• 12:31
  Speaker 3: Start to be concerned about the safety, about what actually
• 12:37
  Speaker 3: it can give someone if they control AA right. And
• 12:42
  Speaker 3: that's what we've seen, not only three D, not only
• 12:47
  Speaker 3: among professionals. You know, previously, like ten years ago, we
• 12:50
  Speaker 3: were part of these discussion groups among scientists who discussed
• 12:54
  Speaker 3: all these issues. But now these issues start to be
• 12:58
  Speaker 3: kind of part of the can verization of regular people
• 13:02
  Speaker 3: who haven't developed AI, but they see how it can
• 13:05
  Speaker 3: affect their lives, and that's created a movement, a reaction.
• 13:10
  Speaker 3: That's what we see with Gonka, that that's developing so
• 13:14
  Speaker 3: fast because people react to this really fast. They understand
• 13:18
  Speaker 3: that it took too much power to several organizations and
• 13:23
  Speaker 3: that we need to rebalance this while we can.
• 13:27
  Speaker 4: Yes and sense is a necessity rather than like, So
• 13:31
  Speaker 4: it's a feature itself, the desimptlization, So it's decentralizing the
• 13:36
  Speaker 4: power and control. But at the same time, if you're
• 13:39
  Speaker 4: talking about it from the technological point of view, to
• 13:42
  Speaker 4: run a model, you don't need a huge cluster. The
• 13:45
  Speaker 4: model itself fit in a specific smaller cluster of let's say,
• 13:51
  Speaker 4: eight GPUs connected with each other with special technologies, which
• 13:56
  Speaker 4: allows gpu cluster to become like acting as one. So
• 14:01
  Speaker 4: for one instance of the model, you need only eight GPUs.
• 14:04
  Speaker 4: Let's say, when you have eight thousands of them, eight
• 14:07
  Speaker 4: thousands of the clusters. Each clusters, each cluster acts independently
• 14:12
  Speaker 4: serve developers and users consumers independently, and in this sense,
• 14:17
  Speaker 4: it's quite rational to think that there is no need
• 14:21
  Speaker 4: to build those clusters altogether in order to serve users
• 14:25
  Speaker 4: with access to these models. They can be like except
• 14:29
  Speaker 4: that they need to be existing in one point as
• 14:33
  Speaker 4: a cluster, like a large cluster of a GPUs, but
• 14:36
  Speaker 4: not of thousands of GPUs.
• 14:38
  Speaker 1: Okay, I get that. Are you concerned? I should say,
• 14:41
  Speaker 1: are people concerned? Because this is one of the things
• 14:44
  Speaker 1: that right now everybody's kind of bringing up in a
• 14:46
  Speaker 1: lot of these conversations. Are just listening to some interviews
• 14:48
  Speaker 1: from WEF, and this was being discussed in terms of
• 14:52
  Speaker 1: the security mechanisms because obviously with centralized networks and many
• 14:57
  Speaker 1: of these you know these clusters that are cook where
• 15:01
  Speaker 1: we're seeing compute power basically being isolated. When you get
• 15:04
  Speaker 1: into decentralized compute power, is there any security risks that
• 15:09
  Speaker 1: could ride into this? Can you explain how you would
• 15:12
  Speaker 1: battle that?
• 15:14
  Speaker 3: I think that that's what we see in general, not
• 15:16
  Speaker 3: only with the AI, but with other spheres of our lives.
• 15:21
  Speaker 2: There are.
• 15:23
  Speaker 3: Use cases when we need to use centralized solutions. There
• 15:26
  Speaker 3: are use cases where actually centralized solutions are quite risky.
• 15:30
  Speaker 3: In an opposite, we need to create decentralized networks. You
• 15:34
  Speaker 3: write that in terms of outages or genuine in terms
• 15:38
  Speaker 3: of control which some can get if they if they
• 15:42
  Speaker 3: control the entry points to the AA infrastructure. So I
• 15:46
  Speaker 3: think that just different type of securities. For sure, in
• 15:51
  Speaker 3: some cases you would want to have just one actor
• 15:55
  Speaker 3: like counterparty. You deal with one counterparty and it's clear
• 15:59
  Speaker 3: for you what can go wrong. But at the same time,
• 16:04
  Speaker 3: in many cases, especially when it's public infrastructure, and think
• 16:09
  Speaker 3: about AI, not just about as a chat but about
• 16:13
  Speaker 3: public infrastructure. We consume information and we share the information.
• 16:16
  Speaker 3: We rely on this as the truth. And from this perspective,
• 16:21
  Speaker 3: public infrastructure never worked as centralized, and public infrastructure works
• 16:27
  Speaker 3: much better when it's decentralized. When we all can be
• 16:31
  Speaker 3: sure in trust algorithms that the data which we get
• 16:36
  Speaker 3: is exactly the data which those models were supposed to produce.
• 16:40
  Speaker 3: And that's what would power the power of blockchains that
• 16:44
  Speaker 3: you actually can trust without trusting one individual. You can
• 16:47
  Speaker 3: trust that what you get from the network is exactly
• 16:52
  Speaker 3: the result which was predicted by your request. So and
• 16:57
  Speaker 3: that's what is important with realize AI network like Gonka,
• 17:03
  Speaker 3: that every inference is verified so it's very date, so
• 17:08
  Speaker 3: you can be sure that it's exactly what the model
• 17:11
  Speaker 3: was supposed to produce.
• 17:13
  Speaker 2: And this is the security.
• 17:15
  Speaker 4: And from the point of personal data, there is no
• 17:19
  Speaker 4: centralized server which stores all these data. There's no servers
• 17:22
  Speaker 4: which actually store this data, and there is no one
• 17:25
  Speaker 4: entry point to hack and get all these data from
• 17:28
  Speaker 4: one place when we build it the centralized network. No
• 17:32
  Speaker 4: one even can predict where your specific request will be calculated,
• 17:38
  Speaker 4: like there are servers worldwide, data centers worldwide currently where
• 17:42
  Speaker 4: everywhere in Europe, in England, in the United States, in
• 17:46
  Speaker 4: Singapore and Dubai, like everywhere the data centers which now
• 17:50
  Speaker 4: support Gonca's network gonk network, and no one can predict
• 17:54
  Speaker 4: which particular GPU cluster will be calculating your individual request.
• 18:00
  Speaker 4: And in this says no one can trace this and
• 18:02
  Speaker 4: no one can get the data.
• 18:04
  Speaker 1: Interesting, well, that creates privacy. I mean, I act see
• 18:07
  Speaker 1: nation states even looking at this as a tool, especially
• 18:10
  Speaker 1: as you start to apply a lot of what we're
• 18:12
  Speaker 1: going to see in terms of AI development on advancing
• 18:16
  Speaker 1: the superpowers of the world going forward. That's an intriguing component.
• 18:21
  Speaker 1: Let me let me ask you this when you look
• 18:23
  Speaker 1: at the token allocation per for the GPU owners, it's
• 18:26
  Speaker 1: as I saw here on the note, eighty percent of
• 18:29
  Speaker 1: tokens allocated to host GPU owners. How do you guys
• 18:33
  Speaker 1: balance this to basically ensure kind of the long term
• 18:37
  Speaker 1: value for investors that are investing in that token.
• 18:40
  Speaker 3: Yeah, so, first of all, I think that we were
• 18:43
  Speaker 3: on the stage where it's not to say about investments,
• 18:46
  Speaker 3: it's about participating in the in the community, in participating
• 18:50
  Speaker 3: in building these granges infrastructure, and what we lowerarn from
• 18:56
  Speaker 3: the previous attempts. You know, like most of the attempts
• 19:00
  Speaker 3: before with around decentralized AI, they were built on the
• 19:04
  Speaker 3: principal proof of steak when most of the rewards are
• 19:08
  Speaker 3: received by stakers, by capital electors, and it's a nice
• 19:14
  Speaker 3: model for finance, it's a nice model for his investment tool,
• 19:19
  Speaker 3: but it limited the amount of hardware, amount of GPUs
• 19:25
  Speaker 3: which were connected to the networks, because the main thing
• 19:28
  Speaker 3: is how to build this million GPUs million A six infrastructure.
• 19:33
  Speaker 3: So everything which we need to put into designs so
• 19:36
  Speaker 3: that it will happen.
• 19:39
  Speaker 2: Need to be need to be there.
• 19:40
  Speaker 3: And the fact that most of the rewards, all the
• 19:44
  Speaker 3: daily rewards are allocated to hardware owners actually create this
• 19:50
  Speaker 3: long term approach because hardware owners, they are much more
• 19:54
  Speaker 3: long term than the capital allocators, sure, because they need
• 19:59
  Speaker 3: to you see the years of usage, how they amortize
• 20:03
  Speaker 3: these equipments for many months, and that's what creates stability.
• 20:10
  Speaker 2: Also in the protocol itself.
• 20:13
  Speaker 4: Is ability, but also intrinsic value because there is a
• 20:17
  Speaker 4: hardware behind it and there is an actual cost of
• 20:21
  Speaker 4: this hardware behind every coin. So every coin which is
• 20:25
  Speaker 4: mined today have the cost of production, and this cost
• 20:29
  Speaker 4: of production comprised of the total amount of the GPUs
• 20:34
  Speaker 4: of the hardware inside of the network. So the more
• 20:37
  Speaker 4: hardware in the network, the higher the cost of every cycle,
• 20:42
  Speaker 4: the higher the value of the coint be even produced.
• 20:46
  Speaker 3: Imagine that ten thousand GPUs is actually one hundred and
• 20:50
  Speaker 3: fifty million dollars a year works of equipment it imagine
• 20:54
  Speaker 3: this number. So we when we look at this number,
• 20:58
  Speaker 3: and we got two hundred million dollars worts of equipment
• 21:02
  Speaker 3: just in three months. And when we look at this,
• 21:06
  Speaker 3: you know, like projects got to one hundred million dollars,
• 21:10
  Speaker 3: one hundred million dollars of the volume. Maybe usually like startups,
• 21:14
  Speaker 3: usually by years, yeah, or something like that. In this
• 21:19
  Speaker 3: case that this community we're able to bring in to
• 21:24
  Speaker 3: online that amount of equipment just in this short period
• 21:28
  Speaker 3: of time. And as Daniel mentioned, this create the value
• 21:32
  Speaker 3: like you know, we we don't. This network never had
• 21:36
  Speaker 3: air drops when someone got these coins for free or
• 21:40
  Speaker 3: something like that. It's only like all of them earned
• 21:42
  Speaker 3: and all of them built for utility, and that's what
• 21:47
  Speaker 3: makes them in general, this project to be quite long term,
• 21:52
  Speaker 3: and it's embedded in all like small decisions like for example,
• 21:56
  Speaker 3: when your mind you don't get all the on the
• 21:59
  Speaker 3: coin in imediately, they invested over six months.
• 22:03
  Speaker 2: Okay, so if you short orm.
• 22:05
  Speaker 3: You will never join you will you will never be
• 22:08
  Speaker 3: part of this network. You need to be a long
• 22:11
  Speaker 3: storm contribution to this network.
• 22:14
  Speaker 1: A couple of last questions here I wanted to get
• 22:16
  Speaker 1: into before we let you guys go first is incorporation
• 22:20
  Speaker 1: of AI agents. This is something that we're seeing more
• 22:23
  Speaker 1: and more use case becoming real.
• 22:26
  Speaker 4: Uh.
• 22:26
  Speaker 1: Now there's varial you know, or various types of AI
• 22:29
  Speaker 1: agent layers that are being built into many of these
• 22:32
  Speaker 1: large language models. Today, Anthropic just launched their copilot or
• 22:37
  Speaker 1: their coworker I think it was that was a new
• 22:39
  Speaker 1: agent that's going out. Then you have, of course what
• 22:43
  Speaker 1: Perplexity is doing with their agents that are active now,
• 22:46
  Speaker 1: where you can deploy multiples and they can stack and
• 22:49
  Speaker 1: start working with each other. Explain to me, first of all, are
• 22:53
  Speaker 1: you guys going to integrate this into what you guys
• 22:56
  Speaker 1: are doing at ghanka.
• 22:58
  Speaker 3: It's kind of integrated by by default, but but we
• 23:01
  Speaker 3: go in with that's even further that imagine currently you
• 23:06
  Speaker 3: only allow agents to maybe browse your internet or filter
• 23:10
  Speaker 3: your emails, right, but it's coming the moment when you
• 23:14
  Speaker 3: will give the access to your accounts and so to
• 23:18
  Speaker 3: the right to transact. In this moment, this verification which
• 23:25
  Speaker 3: I mentioned start to be crucial. You want to trust
• 23:29
  Speaker 3: it that it will do exactly what it should do
• 23:33
  Speaker 3: with your finance for example. That's what we all get
• 23:37
  Speaker 3: used to already with the defy with with smart contracts
• 23:41
  Speaker 3: that we are for sure know that the smart contract
• 23:43
  Speaker 3: will execute exactly what was programmed, so that you don't
• 23:47
  Speaker 3: afraid that your money.
• 23:49
  Speaker 2: Will be lost.
• 23:51
  Speaker 3: With agents, you need similar quality. You need immutability, which
• 23:56
  Speaker 3: called yes, that's that no one can affect no one
• 23:59
  Speaker 3: can affect your transactions which are performed by this agent.
• 24:04
  Speaker 3: They exactly perform what it was supposed to do based
• 24:08
  Speaker 3: on the language model which run the agents.
• 24:12
  Speaker 4: So we believe technical and then a little bit to
• 24:15
  Speaker 4: the technicality is what we provide is the bottom layer.
• 24:20
  Speaker 4: Like all of these companies open EI, Anthroopic, Gemini XAI,
• 24:24
  Speaker 4: they do provide the API layer access for the developers.
• 24:29
  Speaker 4: For example, perplexing themselves, they use the API layer of
• 24:34
  Speaker 4: other AI providers like Anthropic or or openI or or Gemini,
• 24:40
  Speaker 4: like all of them. So basically Perplexity come up as
• 24:42
  Speaker 4: an idea of what if it will just basically orchestrate
• 24:46
  Speaker 4: in between all of the existing models and use the
• 24:48
  Speaker 4: best room all of them and then build our own
• 24:52
  Speaker 4: layer on top of this. So we are what Gonk
• 24:56
  Speaker 4: is is just one of the bottom base layers of AI,
• 25:01
  Speaker 4: of like API access for developers to use. This whether
• 25:04
  Speaker 4: it's for the IGIC systems or for any other specific
• 25:07
  Speaker 4: use cases. This is sort of like an infrastructure layer
• 25:12
  Speaker 4: which everyone can use.
• 25:14
  Speaker 1: Yeah, okay, well that makes more sense, I think, especially
• 25:17
  Speaker 1: as we see use case going into it. And I'm
• 25:19
  Speaker 1: glad you guys explained how the token value, especially from
• 25:23
  Speaker 1: a utility aspect, really starts to accelerate based on the
• 25:25
  Speaker 1: participation side. Last question for you guys is when you
• 25:29
  Speaker 1: look at the just the compute platforms that we've seen
• 25:32
  Speaker 1: you in the crypto side of things, whether you look
• 25:35
  Speaker 1: at bittens or render a handful of file coin, et cetera,
• 25:39
  Speaker 1: that have had different models of use case, do you
• 25:43
  Speaker 1: think there will be use cases like those being built
• 25:47
  Speaker 1: on top of Ganka or will you guys see more
• 25:50
  Speaker 1: competitive elements that might rise out of the development of
• 25:55
  Speaker 1: what you guys are doing to compete with projects like that.
• 25:58
  Speaker 3: I think that each of the projects, they have their
• 26:01
  Speaker 3: unique structure, the unique protocol which drives the direction of
• 26:06
  Speaker 3: the projects. Like for example, bititans are which you mentioned,
• 26:10
  Speaker 3: it built around proof of stack. It created some advantages,
• 26:15
  Speaker 3: but it also lead to some disadvantages when the amount
• 26:19
  Speaker 3: of GPUs in the network is quite limited. Even though
• 26:23
  Speaker 3: the project exists for four years, it was the most
• 26:25
  Speaker 3: prominent projects in decentralized AI field, but amount of GPUs
• 26:31
  Speaker 3: high end data center great GPUs in one network already
• 26:36
  Speaker 3: like a couple of times larger on what bitans are
• 26:39
  Speaker 3: able to achieve. I still think that different projects will
• 26:43
  Speaker 3: coexist in different projects will have their.
• 26:46
  Speaker 2: Own markets and.
• 26:50
  Speaker 3: Kind of the same with render, the same with different
• 26:53
  Speaker 3: projects which right now targeted for example AI training like
• 26:58
  Speaker 3: Prime Intellect or or Genzeen. We we see this as
• 27:03
  Speaker 3: a system. What Gonka is in to build is this
• 27:07
  Speaker 3: massive infrastructure. And to build the massive infrastructure, we believe
• 27:13
  Speaker 3: the productocols set in the right way to get to
• 27:17
  Speaker 3: these millions of GPUs, millions of h hundred GPUs.
• 27:21
  Speaker 2: That that's what you need in protocol.
• 27:22
  Speaker 3: You need proof of work, and that's what is missing
• 27:25
  Speaker 3: in these other protocols. So everyone will have their own
• 27:30
  Speaker 3: kind of pass and join me. We hope that actually
• 27:34
  Speaker 3: a lot of projects which right now built around decentralized
• 27:37
  Speaker 3: training will use Gonka as infrastructure, and so that it's
• 27:41
  Speaker 3: actually that we were more into collaboration with all these
• 27:45
  Speaker 3: different ecosystems rather than to competition, especially seeing in that
• 27:53
  Speaker 3: in terms of amount of GPUs in the network, we
• 27:55
  Speaker 3: don't have any competition, and that's where where the focus
• 28:01
  Speaker 3: of the protocol will be for foreseeingle future.
• 28:04
  Speaker 1: Very good, Hey Daniel and David, thank you so much
• 28:07
  Speaker 1: for coming in today. Of course, we've learned a lot
• 28:09
  Speaker 1: about where maybe decentralized comput is going to be going,
• 28:13
  Speaker 1: we'll be following what you guys are doing at GANKA,
• 28:15
  Speaker 1: so thanks for stopping in.
• 28:17
  Speaker 2: Thank you.
• 28:24
  Speaker 1: If you like this video, like and subscribe, drop a
• 28:27
  Speaker 1: comment down below, and also join our free private member group.
• 28:31
  Speaker 1: The link is in the description