Can you sing a little? I can. Yeah.
We've been all on a time searching for love And there are no where is common fun And as soon as we get it
Somehow we seem to think that I work, no. You have the same thing you do to get that love. All the things that you have to do to keep it. You gotta stay on your tail, you'll take all of it. You gotta give a little take or not.
You got all these little diamonds burning, keep the fire burning, keep the fire burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds burning, you got all these little diamonds#
You gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you gotta do it, you#
♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪ ♪#
Kill the music. Kill the music. Kill the fire. Kill the music. All right, cool, cool. Great.
Okay, all right. Good to be here. Thank you so much volume and welcome everyone. We are getting started and this is the volume weekly, Twitter, space.
We are so excited to be here today is June 28th is a Wednesday and we hope it's sunshine and awesome where you guys are at today. We just want to thank of course our volume community for joining us and of course we want to thank the Paloma community for joining us. It is amazing and don't mind it is
and I know everybody is pleased rocking and rolling on the beach and enjoying time off and enjoying what is essentially going to be an amazing week ahead and a midweek. I want to say also that today
we're doing something different. We are rocking out a new topic today so that we're bringing something fresh. Also, if you are reading the blog post or you are listening to our videos on our site and I think we're still posting videos to YouTube, this is definitely a cool video we're going to share so that folks can
was an enjoyed on their own time. So let us rock and roll and get moving with Somersault and talk about who we are. And this is we are volume. And this what is volume, who liked to tell folks before we start volume is software company.
company working on in the blockchain and our goal is to make private key security and private key management massively, massively scalable and easy. We want you to get more use of your private keys across multiple blockchains. No matter whether it trains maybe in a matter of way, you're liquidity to be.
We want to be able to make private keys do things that were previously impossible and you never thought you thought impossible, but you never thought we're possible. All right, and I am excited this morning. We have a special guest with us. We first have a field theory was with us field theory. You're here.
I'm here. How are you? I'm rocket and roll and groove it and cruise in you know how we be do and we max and relax and block change action. Um, lux and TXs. Um, and we have also door is door here is the door. Hi, good morning. Hi, door. It's on this is here. All right.
So, well, I really know Dole was supposed to be at the StarCware Twitter space, which has about 6,000 people in it right now. And Dole was the keynote speaker there. He decided to come to our Twitter space because he said it was too many people over at StarCware. So we want to say thank you Doerr for joining us. How you do it today?
I'm doing great. I see that one of the listeners may be joined because it seems I'm going to talk about different Saturdays. Today, I'll politics only. I have followers from the Twitter community that know that I had some strong opinions about politics.
I want to get into that in the next one.
But today, thank you so much for visiting. And so we hope we can do a few of these. We're really excited. And of course, you know, for those who are in the Pelloma community and the Vaughan community, everything here we talk about is how to use
And how you know volume and our community can continue to add value in the blockchain space. So today's topic is machine learning and AI and blockchains. Is it real or is it even a Raj? And so now to get going what we're going to do is really talk about
You know, one-on-one about AI. So, you know, if you're in the polymer community or a polymer, uh, I'd have a validator member or if you're in the blockchain community, um, really it's like, hey, how do I, you know, what is it this AI thing? And, um, what
what we really like to say is AI 101 is the first way to understand what is AI. So I'm going to ask Doar and Field Theory to give us their definitions of artificial intelligence. Hit it, my friends.
Who's going first? Okay, I'm going to start with field theory. Field theory. Field theory, why don't you go first? Yeah, I'll give a short one and I think Dora can give a much more detailed one. So my interpretation is Michel O'Brien.
artificial intelligence, if you will, Alexa will call them machine learning. These are essentially models trained on huge data sets and it consists of a ginormous data
It consists of a ginormous private set and you also consist of a server. So, you know, farmer users perspective, all they want to, you know, ask chat GBT or they have something they want to predict.
They convert their questions, requesting to some data and then they send to the server and the server process their data with their trained parameters and it returns back some data which shows the answers, prediction, desire, etc. That's it.
Okay, so for the five year old in me, you got a box. It has, it takes some stuff into the box and then when I look into the box, it is a great story. Like I can see the future with the boxes. That what you're saying. - Yep. - All right.
door hit us. Yeah, I would say that machine learning AI I would use these terms interchangeably in this kind of discussion. It's kind of a special type of computer program. So you know that most computer programs are built in a certain way.
you know, some programming languages, first stuff like that, it gets loops and if, then, and all these kind of stuff. Machine learning are very special computer programs that are designed at the fundamental level very differently. It's like we're planting your brain inside the computer, built out of neurons.
stuff like this, like the way that we think that the real human brains are built. And then we kind of train this brain the same way that we teach kids stuff, right? When you grow up a kid, when you get a kid and he grows up, he turns stuff. So he sees things and he doesn't ditches that and this is that and all that. So it kind of happens in the same
way you can plant a brain inside a computer program and then you teach these brain things like you would teach a kid, a child that's going up. And this way you get computer programs that can do very sophisticated things, things that couldn't have been done
using computer programs. So there's no way you can do a computer program with it written in regular programming languages with if done and all that and just tell it to complete your sentence in a meaningful way like the
ask it a question, it would answer. But change of it can do that because it's designed in a fundamentally very different way. So this new type of computer programs is, it seems like it can allow us to do many things that we weren't able to do before that.
But when you say it, you mentioned ChatGPT and that's interesting because now everybody, when they hear AI, ChatGPT is sort of the key phrase that accompanies it and open the eyes to key company.
Why would you say that chat GPT artificial intelligence is so successful? What would you say makes that successful? I'd say two things. First of all, they did take the AI.
the inventory of computers back in the previous century, Alan Turing said that another is a test if a computer can make you think that he is human then
It's kind of a test for intelligence, right? So right now you can have a discussion with which LGBT and it seems like it's human. So like from the computational level that did achieve something that I haven't seen before that, that's one thing. And the other thing, they're solving a problem that is useful for everyone. It's like,
Whenever you do a computer science stuff, most people aren't interested with it. They go to the grocery store and they talk with people. They don't care about these kind of things. But anyone can understand this thing that you guys, you're inside the computer that asked if it's a real problem.
and you can ask it things and handle discussions and it's a really good thing to do with your mind. It makes your computer alive. So even if someone doesn't know anything about computer science programs and stuff like that, here's a computer program that talks with you like it's a person. So it's kind of mind-blowing and it makes the
the way of explaining them. Yeah, I think that's awesome. Okay, so then, let's talk about, we talk about artificial intelligence like teaching, learning models, it's about learning, you talk about kids and how kids learn, it's sort of
of imitating how kids learn and creating this type of approach of improving continuous improvement and continuous learning. When we think of blockchains and AI, we know what blockchains are, right? So everybody here is sort of in the blockchain space. We think of blockchains
as these, you know, a pandually data stores where, you know, each new addition to the data store is verified by, you know, a number of other players in the network, people who are connected, you know, look at, you know, these transactions and say, great, you know, what, you know,
This works and we agree on it and we use it for you know ever since we talked about Satoshi peer-to-peer You know e cache or electronic cache which was his first usage When we think of those two concepts of you know this computer learning plus
So, I think that's, you know, peer-to-peer electronic hash. The question is, what does it mean? And so, you know, what does it mean for artificial intelligence on the blockchain? And I think maybe, you know, I'll start with field theory again. Field theory, when you think of, you know, artificial intelligence,
intelligence and what you understand it and then the blockchain as you understand it, peer to peer, electronic cash, what does it mean for you? Like what ideas come to mind if any and what is stimulates you? Where does your brain go when you think of this thing?
I think that's been very lost in talking permissions, maybe go to Winthole from then. He doesn't want to have consensus on this AI discussion. All right, so we'll talk about it. So I guess the, you know,
What's your view? Why don't you jump in here and get your thoughts on it? First thing I know, I'd go one step back and say something about blockchain. Most of us think about the blockchain as a distributed ledger, right? It's kind of the least that we've ever seen.
all agree on what's on the list and the list is only updated according to certain rules and it can contain transactions or stuff like that. But more generally, blockchains can allow you to do much more than only storing data on an organized manner and in a variable manner. Blockchains is kind of computer
that fits on the sky, it's owned by no one and it's maintained by all the nodes that use it. So this kind of computer allows you to do many other things than just storing information. It allows you to even run computer programs on this computer on the sky.
And this gives all the users trust, like all the users that play within this computer can trust each other because the computer cannot be manipulated by any single entity. Any single no, we cannot do nothing even large. Of course, of course, no, it cannot do anything. So it's a personal blockchain.
the computer on the sky and it acts through smart contracts. So the programs that run on this computer in the sky with the blockchain, the programs are smart contracts. That's how we call them. And as we say, AI is only one type of a computer program. So the question is,
Can you run an AI program on the blockchain? So I think that's kind of the question. Where is volume volume? I think field theory is back and rejoin. Will you add him as a speaker? I think he's waiting to be given the speaker
role and I don't have that superpower. All right, so that's interesting. Computers in the sky, wow, dude, you know, they said, you know, psychedelics of the future, you know, I was like, wow, this is trippy stuff. Okay, so computers in the sky running programs that are owned by no one. And then
have to do these sort of, so can we say that you're saying, hey, really what you're thinking about is using this learning or this way that computers learn in computers in the sky that no one owns. It sounds very sky-knits, very
terminator like, is that a way to think about this star? Am I two stuck in the movies? No, you're just okay because that's the way things happen right now. I've seen that in people, right? So when people put in there, I don't know, if they're or you have to see inside or whatever, whenever there's supply and demand,
there's a smart contract that sees that the conditions are met and it performs the transactions, right? So this smart contract that handles the liquidity pool is essentially just a computer program that runs on chain and checks if certain conditions are met and does something that it was designed to.
So this smart contract is still being in computer science. It's a very simple kind of program. So thinking you could have, like for first example, this kind of smart contract that can handle transactions, not only using this simple rules of if
then and looks and stuff like that, but also using more sophisticated computer science tools such as AI. So that's one example. And another example is just forget about transactions, forget about liquid pools. Let's think about other stuff that people would like to do with trust online.
like a joint computer, let's say we want to play poker online. Okay, that's a very simple example. We can explain it to anyone. If you and I want to play poker online, how can we trust each other? There has to be a deck that deals cards to both of us and this deck has
to be kept somewhere. Now, if the server that runs the game is under your control, you control the deck. You can do it yourself whatever cards you want. Now, many people do play poker online, but they play it on a server that belongs to some company. I know the AAA.
whatever right and then they can do whatever they want right they can just manipulate the cars they can just make it more interesting and stuff it won't be fair right it won't be truly random so think about it that all the nodes in the network can hold this computer together so that is is held
on this computer and it's fine. So, I don't know, just something of this deal. But, yeah, there we go. They want to pass each other. They want, yep, on a server that they don't write. So, the blockchain becomes a server and the programs that Iran's are the smart ones, right?
Right, we want like people want people talk about having AI programs inside this mark contract So it seems like it seems to be very something that's good. Yeah, really change the game. Yeah, you're serious area back John back here. Okay, so we
We got some ideas on playing poker, trustlessly, on the distributed network as thinking about running computer programs in the sky, and learning models in the sky, what any thoughts from you on when you think of blockchains in AI?
Yeah, I think there's in the current blockchain space, there is a lot of very active marketplace for potentially aware supply and demand of
learning models can match. And I think the lack of that also is partially a reason for, you know, because you know, puts your shoes in a machine learning model. You spend a lot of work training models, but it's kind of hard to
to find users. That's basically a market place for very generic models. I think they're useful. That's a powerful idea, having a marketplace for models.
Again, models can be pretty big. We're talking terabytes or how big can models get door. You've seen models of how big in size. >> Just to keep you updated, it can be up to a terabyte and even more than that. >> Since blockchain don't really handle a terabyte, data are really well right now.
How do we, you know, and we want to have a marketplace where people can, you know, you know, sort of trade models and really start trading these things that are so big? I guess the question is how would we want to make that happen? How what would be a pack by which, you know, models and
and marketplaces for these models can come together when these models are so big. Do you have any ideas about that? Yeah, first of all, I want to say that I agree. Definitely agree with Friir theory. So we can say that there is not, we don't see many
actual use cases like actual demand for this kind of thing. Lots of people are talking about the possibility of doing AI on chain but it doesn't seem to be so many people that are currently interested in the things that it could provide so that's one thing that people should think about. Okay, let's say you have a chain, what would you do with that?
and this question still needs good answers, but like he said, even if we had like use cases, good use cases for doing that, just like he said, there's a problem for, like if you want to run an AI model on chain, that's problematic because just like he said, these models
are huge. So what happens in practice, like I'm not only in AI, like in many other things, when you want to run TV programs on chain, and remember, just like we said before, an AI program is just a computer program that is built differently, okay? And this computer on the sky, this blockchain is a computer, but
the site that runs smart contracts which are programs. So, you want to run a program and it doesn't matter if it's in a i program or some other program but it's a very complicated one. It's a program that runs on a lot of data and on a lot of data, you can't actually run it on chain. So what people do, it's called sometimes there are two or
the role of stuff like that, they do the computation of chain which kind of makes you lose the trust, but then they use a very sophisticated computer science tool that it's called the proof system or most multiple call it like ZK or mobile
So it's a computer science tool that it's a proof system and these proof systems allows to generate the proof that says that the computation that was ran off chain is accurate and honest and everything was good with either as no manipulation in the computation, the computation like
went correctly and everything. And then you only run on chain proof. So this proof system is going to compute the programs. One program generates a proof and the other program verifies it. So you run the computation and the generated proof of chain.
But then on chain you can verify that it was done correctly. So if you would want to actually run AI on chain the way that people are trying to do it is by running the model off chain, but then executing on chain it approves that it was executed correctly. Got it. So you run
the model of chain with your big data centers because it's really huge, but then you prove, you generate a proof that the model run, that you've been able to prove that is now available on chain because that proof is very small, it can fit on a blockchain and it can be processed by a smart contract in terms of what happened. Did I read you correctly in that process?
And then when we marry that with with the jibis understanding that you know, you know models need to you know in order for models and marketplaces of models to come together people need to trust us Lee and and and dispassionately prove that
that the models are authentic or can run, this might be a way in which AI finds a pathway into the blockchain and is useful. So the question is, what is this AI proof stuff? Is it easy to do or is it hard to do and why?
That's a great question. I described it very clearly. I'm a great question asker. I just asked it to be a great question. I described it very clearly. It's so simple to just take the program, your writing of chain and then you create the proof and everything.
The problem is that these proof systems are kind of complicated. So in most cases it's not so easy to do that because to generate the proof most often takes much more time than just executing the regular computation that you want. And in certain cases
cases it becomes much more time and one of these cases is AI systems. So the thing is as I said before AI programs are not regular programs okay they're built differently just you know just one technical term they're based over
floats over floating point numbers as opposed to other programs that are running binary or in some other way. So this fact, the fact that AI programs are built fundamentally different makes them hard to
be squeezed into the proof system world because we got like you know 40 years long theory about proof systems and it's like for several decades that people are developing proof systems that are better and better but all of these proof systems are always we're always built for regular programs right right now
Now when people are trying to take this proof system that are built for regular programs and apply them to AI models, it complicate things significantly. If some of the listeners are aware of what's happening, they know about the quantization problem. You kind of have to take these floats and squeeze them into
the binary or finite field word of proof systems and it really makes the generation of the proof extremely long and becomes inefficient to do that. So that's one problem a lot of people are trying to work their way around it and come up with better solutions for it.
Wow, got it, got it, got it. So it's still a big problem and a big blocker to again unleashing new usefulness of AI on shame. All right. So I think the question will be, you know, what comes out of that and, you know, what those solutions look like. It would be very
interesting to hear that and to see that. So I think we need to come back again and talk some more about some of these problems. But the exciting part about it is that as we wrap up on time is that we at volume are looking at this problem and again thinking of ways in which your private keys can participate in the
AI ecosystem. And if your private keys can participate in the AI ecosystem, then that means your private keys continue to do more than just what you thought was possible. And I think is very, very exciting to start touching on some of these ideas and door. I want to say thank you very much for joining us today. It's been awesome to
have you. A lot of deep stuff talking. I know we just barely scratch the service in 30 minutes, but I hope you'll come back again and we can talk some more. Sure, thanks for having me. Yeah. Feel free. Markets, markets, markets, you know, markets from models. Exciting stuff. And I think we should
continue to explore and share some thoughts with the Paloma and Valiant community as we continue. What do you think? Absolutely. I think that's a very interesting and exciting area. Oh my goodness. There's so much dripping alpha in this. I mean, I just can't even get out of my chair. There's so much alpha in this beautiful Twitter space today.
I feel sorry for all those 6000 people over at StarCware. I mean, we love StarCware. But you know what? Yeah, a lot of Alka here that you just won't get anywhere else. So I want to say thank you everybody for coming and we will see you next week. We're kind of excited. We're going to talk about Prime Trust and decentralized
which is even another exciting topic and think about decentralized custody with AI on blockchains. Whoa! It gets to be pretty hot. So thanks everyone. We will see you next week. Enjoy your week ahead. Thank you, Dohr. Thank you, Field Theory. Cue the music. Thank you.
I hit the side of the song. That's it. You got it. Yeah, I think you hit that. Oh, I thought some of you girls like to know what I'm saying with this.