Hey, good morning everybody.
Terrific. You guys hear me. Okay?
Yeah, we can hear you fine.
Great. Listen, I guess this is Tom Trobridge, one of the co-founders of Fluence and just
psyched to have everybody on this. We want to get into the topic here to AI need blockchain,
which we're excited to talk about something that we think about a lot in Fluence and we're
excited to have a bunch of prominent guests who have I'm sure some real opinions on the matter.
So excited to kick it off and you know I think first maybe just talk quick mention the other
people of Fluence. Evgeny, do you want to want to say hello for a second and then Bernard if you're on?
Listen, while we while we wait for Evgeny why don't we then just walk around the panel here
and if you guys wouldn't mind giving just a quick intro on yourselves and
you know what you're working on then we'll get into the get into some of the questions.
Sure. Hi, I'll kick off. So my name is Richard Simpson, representing Bionic DAO. We are a DAO
that is formed to look at and invest in the convergence of what we think of future technologies,
one of which is AI, another is DeSci, another is smart infrastructure and the last one being
extended reality and we think that web 3 and blockchain is integral to the future of all these
technologies. Hey guys, I'm Andrew Lubon. I was a seven-year professional athlete, got into crypto
recently with Pulsar.ai, which is NFT for discoverability. Can you hear me? Can you guys
hear me all right? Yep, gotcha. I'm the founder of MASMA. We're a web 3 super app. We are basically
bringing the convergence of every application into one ecosystem so that we can help on board
mass adoption through account traction, cross chain technology and simplistic user experience.
So yeah, happy to kind of be in this conversation to kind of talk about
my opinions on the AI and blockchain. Thank you.
Good morning. I'll go next. My name is Travis. I'm with Vaynar. We're an L1
that we'll be releasing here soon and we focus kind of on the entertainment industry
or entertainment customers, but really blockchain for all where we have a lot of focus on green
initiatives and carbon with the chain, as well as fixed fees, low costs, and really making
blockchain more accessible and usable for the everyday person who may not be as familiar with
web 3 in general, kind of making it relatable. Cool. I'll go next. My name's William. I'll
continue the flow, I guess. So Matt, head of marketing at SADA and we are building a foundational
data layer for all of web 3. The best way to think about is a bit of a base layer
that sits underneath current infrastructure and allows any blockchain, any protocol to query both
cross chain and off chain data via their own customized data feeds as well. So I think,
you know, very excited to join this conversation and AI obviously needs a data input.
So very excited to see, you know, share some of our opinions on the matter as well.
Hey, everybody, this is Zania. I'm currently I have a pleasure of working with the Fluence team
and I'm jumping on here to help guide this discussion because it's also a topic that I'm
personally very passionate about. My main background is in web 3 marketing and communications and it's
so fascinating right now for me personally to see all this amazing projects they're building
and the question that is the topic for today's discussion does AI need blockchain?
It's pretty like it's resurfacing, like it's on the rise. That is the point of discussion.
So it's very exciting to be talking about this with y'all today.
Hey, guys, I'm Andrew Lubon. I was a professional athlete for seven years, got in crypto in 2017.
Most recently with Pulsar.AI, we're solving discoverability across the NFT ecosystem
specifically. Check us out at www.pulsar.ai. We're launching our PFP drop and our token
right before the Bitcoin happening. So super exciting time for us. And thanks for having us
on this space. It's right where we live, right at the intersection of web 3 and AI. So lots to share.
Thanks again. Hi, I'm not sure if my mic is working. Can you hear me okay?
Yeah, we can hear you. I think this is a little bit lag time probably.
Yeah, fantastic. Yeah, my name is William Laurent. I'm a journalist covering
AI, blockchain, web 3. I help clients build content factories and lately those content
factories are in web 3. I built two web 3 news desks. I've taught AI classes here at a university
in Tokyo. So this is a subject near and dear to my heart is like the intersection of three things
that I'm really into, which is quantum computing, and then the other being AI and web 3. So looking
forward to the discussion. Cool. I'll go next. I'm Pat. Nice to meet you guys. I'm from NounsDAO
representing Nouns Esports, the first web 3 funded esports team fully through DAO. We primarily focus
on social impact, kind of like a social experiment of memes and also doing good in the world. But
you guys have probably seen the noggles somewhere around Twitter at this point. My background is
actually in investments. And so like consumer AI, a little bit of genomic research and pharmaceuticals
are my investment criteria. So excited to kind of hear what you guys are building on and chat with you.
Amazing things everybody. And I think there were a bunch of glitches when I first started this space,
but now Tom and Evgeny are here on stage with us. And I'm going to pass the mic to Tom to kick
start this discussion. Thank you. Appreciate it. And this is great to be here. I've sort of stopped,
started initially was back and forth as I kind of figured out the system here. But you know, there's
a lot of different, there's a bunch of different perspectives here. And I guess, you know, fluence,
we were thinking about decentralized compute, and areas that is sort of important for AI. And I know
kind of some thought I want to bring on down the road, but to kick this off, you know, maybe it
might make sense to start with kind of an academic perspective, and kind of take it from there. And
then we can kind of transition into more, more, more kind of recent practitioner elements. But if
that makes sense, you want to start with some academic perspective on kind of AI and blockchain
where there might be a convergence. And then we can get some perspectives from anyone else who wants
to talk about where there might either be synergies. And I sort of have a view down the road just that
they're actually, you're actually going to be required, which is, you know, I can elaborate
on that more later. But but first, maybe an academic perspective, if there is one.
Yeah, I guess I'll take this one. I think that when you think of AI, and you think of
decentralized compute, you have a lot of different areas that you can look at, you can obviously
starting data. I might just jump in. But I'm definitely not an academic in AI. But I think,
you know, we're all blockchain academics in one way. So maybe we can approach the conversation
from that side, and you know, the topic being AI, needing blockchain. And I think I understand
where you want us to kind of take the conversation, Tom, just by what you were leaning with there.
And I think, absolutely, the blockchain needs AI to flip it on its head, you know, where we're
moving so fast. There's just so much, so much data that creates an output, or, you know,
people creating things, but there's only so much that the technology with the human input can
control. So when you have that AI, you get that level of assistance. That's how I always really
like to look at AI. And I know there's so many other parts of the conversation that we could
talk about. But for me, when I think about AI, I'm really a big fan of AI assistance,
and building those. There's some really good companies out there doing amazing
AI assistance stuff, as well on the blockchain. And I think when you start building a protocol,
keeping teams lean is very important, as well. And I think AI comes into play there for early
stage startups being able to output more, with less of a financial commitment to building their teams,
as well. And then, you know, for us, at Stata, dealing with data, AI becomes very, very exciting
when protocol builders come to us, and they see some of the data that they can get cross-chain
off chain on chain. And then what they can do with that with an AI assistant, how you can aggregate
data, how you can customize data with AI, is going to be a huge untapped world out there as well.
So I'm very excited to see how the blockchain uses AI. So I guess from my point of view, it's
yeah, blockchain needs AI. It's a great place to start the conversation.
Yeah, well, that's interesting, because you're talking about how AI can further leverage
blockchain and can further leverage all of the, you know, almost limitless amounts of data that
are generated by different blockchains. And so I certainly can see that being something,
and that's something, you know, I had been thinking less of, we've been, I've been thinking
of it more from the AI kind of LLM type of perspective. But that's obviously just one use
case of AI. But Pat, you were, I think you were talking, Pat, and you may have not finished your
thought. Did you want to want to finish where you were started there?
Yeah, I think unfortunately, with X today, there's probably a little bit of lag time between the
speakers, because I've noticed it happen a couple of different times. I think on the side of like
academic side, encryption, something that's obviously important to look at, I think sharding
is also something really important to look at. When we look at the centralized compute, there's
a lot of like, there's a lot of really strong possibilities. But at the same time, we also have
to look at MAB protection, we have to look at the ID systems and social scores. As we I think
we've all seen from at least like the LLM side that we're getting better and better at doing
catfishing and phishing people's accounts and creating kind of like digital twins of humans on
X. And what AI does is AI is really, really good at pushing out content, it's really good at kind
of speeding up the way that we process everything. But then we have blockchain on the other end,
where blockchain is really proof of transparency, proof of work, we have a lot of different ways to
actually use blockchain to verify that humans are real. I'm hopefully that we get into attestation
today. But I think with the idea of attestation, when you think of like, even someone in the real
world right now who is using Tinder, the chances of getting catfished are significantly higher
today than they were two years ago, imagine where that's going to be in five years. And that's
really where blockchain is going to provide a lot of help in the future for us.
So you see AI is being relevant from authentication perspective.
Hold on, William, you want to bounce off of that? No, yeah, 100%. I mean, we're all going to need
digital signatures very soon within the next couple of years to log on to zoom or wherever,
just because of all the deep fakes that are happening with AI right now. So, you know,
I'm very interested in the intersection of like where AI hits blockchain from like the ethics
perspective of AI and being able to get a little bit further, you know, with ethical AI, a part of
that obviously is transparency, right? The transparency that being on the blockchain gives
us and someone mentioned decentralized science, I think, for me right now, that is one of the best
use cases that I see for, you know, AI when we're talking about it running on the blockchain,
because everything from, you know, funding to clinical trials to peer review
can finally be transparent. Super excited about that.
Yeah, interesting. Julian, you want to add something?
Yeah, strange enough. I cannot hear William. That's very strange.
Yeah, just soundless. I was just going to play off the authentication thing. But where I see AI
kind of converging with blockchain is that blockchain is just an accounting like level
three system, right? And so currently, the system we have today, we have to execute contracts,
we have to execute, you know, all these like kind of like web two things that are like the
infrastructure layer, right? AI can kind of move in there and replace those infrastructure layers.
And if you want to imagine a future where, for example, right, I want to create a deal with you
today, right, Tom? And so this goes back into the trustless, right? Web three is about being
trustless, about being transparent. My imagine me going into a deal with you as a like, you know,
right now, let's, hey, Tom, let's, you know, buy by product, sell service. The moment we speak that
the deal is done, right? Our words go on chain. The contrast gets spun up by some AI or some
automated system that kind of put that all together for us. And then, you know, we just kind of go
about our day. So I do feel that AI will play a part in terms of the automation side of the
blockchain and AI. And then basically, we have the blockchains that indicate the specific, you know,
contracts or automation that the AI is doing on behalf of us. And then, you know, one day in
2020, 2030, we're just going to be a whole new society, right? Whereas the moment we just talk,
it just, you know, business is kind of done. That is, yeah, gives you a lot of, you know,
having negotiated lots of contracts, and I think there's going to be a little bit before we get to
that stage. But I certainly think for a lot of simple things, I would make life a lot easier.
That is for sure. Evgeny, did you want to jump in? I know Andrew is waiting also.
Yeah, hey, guys, I hope you now can hear me. Well, I had some problems with connection. But just a
quick intro. Hello, I'm Evgeny, co founder of Luance, and we're building the cloudless decentralized
compute platform, which we believe is like, which is also like powered by blockchain, and we believe
can like really help AI. And I want to actually, you know, add a little bit perspective on like,
how blockchain can help AI versus the other way. So I think actually, like current modern AI systems,
they really lack transparency and verifiability, it's like really, really dangerous, and it gets
more and more dangerous. And like blockchain technologies and web tree is the best way to
address this with like adding like really, really transparency on what is happening on which data
the model was trained, you know, how this data was produced from like from the other data, like from
sources. And, you know, give you like, this AI powered by blockchain can give you proofs that
the right model was executed against user query, without no manipulation, no third party, you know,
access, and so on and so forth. So yeah, so it's, it's kind of like, the bigger AI gets in the world,
the much like the more danger it creates. And you know, the bigger the potential value that the whole
web tree space can bring to to AI. Yeah, and I listen, I got a lot of stuff to bounce off on what
to bounce off on what I said, but I want to get to, to Andrew, because I know, Andrew, you were,
what may want to reference that or something else before.
Yeah, no, thanks. I there's more technical people than I in this room. So I'd be happy to be
corrected. But one thing that's super exciting to me being in the space is it really feels like a
moment where AI is this massive creative force with kind of no parameters or no limits. And in
ironic or amazing moment for the rest of web three blockchains and tokens, it feels like blockchains
and specifically the tokens incentivizing those blockchains has a moment to really reign in this
huge creative force in a way that is practical or useful and all sorts of different use cases. So
I am very bullish and hopeful that that is a scenario that plays out where everyone gets
access to massive creative centralized AI, which I think is probably inevitable. And what we can
use as an industry is use these blockchains and tokens to parameterize or to put rails on this
amazing creative force. Well, so that that's interesting, because you just used a word that
is anathema to us influence, which is centralized. And so one of the things that we and this is to
open up to everybody here is the benefits of centralization versus decentralization. And one
of our thesis is when you have something big and centralized, it becomes a tool for influence and
for manipulation and for kind of governments, etc. And so one of our views is you need to have
when I talk about AI here, let's be clear, we're talking about LLMs, but this also could be true
for legal stuff, whatever needs to be decentralized and running decentralized, but transparently,
so that it's free from that type of influence. So that that's kind of an interesting thing when you
when you mentioned centralized and ways to control it. The question then is, who's controlling that?
And that starts to lead to down lots of interesting paths. But but I know Bionic,
you've had your hand up for a while.
Hi, yeah, yeah, some interesting points. My point was from something earlier, just talking about
some of the use cases and the easy wins or the intersection between AI and blockchain,
some things that interest in, particularly to us is voting in DAOs. So using AI agents to vote in
one of the problems that we're seeing in blockchain communities in DAOs is, it's like
voter apathy, almost. And you could, I think it would be great to see a situation where
you could have AI agents voting, you could split this improvement proposals into different levels,
and perhaps level five is low and is deemed less important. And level one is very important.
So level four and five, you could almost have agents voting on your behalf, and that would
increase the engagement and the way people are voting in these DAOs. Another one is something
that we're seeing in some of the blockchain wallets, certainly Rabi. And I think MetaMask does
it as well, is that an AI is simulating a transaction for you, which is helping you or
allowing you to sort of understand if you're being scammed before you're scammed, which I
think is a great use case. It allows you to sort of, so if you're exchanging token A for token B,
you're not open to, or at least it sort of says, oh, what I think is one Bitcoin changing to
ex-USDT. Actually, it sort of says, you're only getting four USDT and the pool that you're
interacting with is a scam. So I think it's unsophisticated, as it were, AI agents just doing
low level work, but work that is necessary and relatively easily to do on a blockchain is a
great use case and an easy win before we get into the whole decentralized issues and LLMs and
chain endangerment stuff, which is very technical. I just thought it'd be good to just talk about
some of those easy wins. I love the second one. I mean, using it as effectively what you're talking
about is fraud prevention, which to me makes a ton of sense. And you could have multiple different
models out there. They could just lag suspicious stuff. To me, that makes a lot of sense. And
frankly, you're already seeing that employed by banks, right? But you just don't see that
employed by individuals to help individuals out. And I guess we're talking to some sense about the
democratization, AI leads to the democratization of this type of almost personalized fraud
prevention in a way, which is which is terrific. The one thing I would say a little bit though,
on the voting, we were launching a DAO very soon, we actually just incorporated association
Switzerland, you know, participation is a huge issue. And, you know, in the traditional equity
world, you've got shareholder services, which basically do all the proxy voting for everybody.
And so voting there happens by a company, if you want a director, you have to convince one
of two companies. And if they like it, they recommend you and everybody votes for you.
Yes, it's effectively delegation. And so you end up with the problem with that is you end up with
just a couple of groups that make those decisions, and you've got to get ISI or whatever they are
on your side. And so they then are calling the shots. And so it's a little bit, you know, this
would then you then if in the DAO side, you'd kind of need those agents to be supportive of whatever
proposal was, if they have a lot of votes. So anyway, it doesn't, I'm not sure it sort of solves
one problem, I create another, I'm not sure, but, but, but, but, but interesting. But, but sorry,
Vanara, you were you had some you wanted to say, I think.
Yeah, I was gonna say, like, I don't, you know, AI doesn't need blockchain, and blockchain doesn't
need AI. But when you put those two together, it creates like, a whole new force and whole new
dynamic and whole new things that can be done, like, you know, kind of like you were talking about
with the centralization versus decentralization, you, you know, where you can have more accurate
and trustworthy data that comes from blockchain, as opposed to like, you know, kind of feeding
these systems with, I don't know, I don't know the word you propaganda, I guess you could say, or,
or, you know, yeah, with however you kind of like want to influence in that way. And then at the
same time, blockchain has so many use cases that AI can do. You know, one of the big ones that I
kind of think of is, you know, the ability to, you know, your smart contract, your system,
whatever you put online, just being able to like, throw an AI, like, hey, find my vulnerabilities,
or find how to hack this, find, you know, where the weak spots are, and just running, you know,
millions and millions and millions of tests to, you know, kind of find those weak spots that,
you know, even a traditional human audit, you know, could miss. And we see those things get
missed all the time. And you know, that's how some of these hacks and everything else kind of happen.
Yeah, that's all fair. Pat, you had a comment.
Yeah, I actually wanted to just go back to kind of what Bionic was saying about voter atrophy.
At NounsDAO, we deal with an eight figure treasury for voting. So every single person in the world
can actually write a candidate proposal, or if they're a holder of a noun or two nouns, I guess,
now after our last arbitrage attack of $20 million. Voter atrophy is something that's extremely
important to think about from the DOW perspective. And when you bring in, like, when you bring in AI,
I think one thing that you can do is really think about knowledge graphs is that should one vote
stay one vote in the long run when you are fully decentralized? If you are voting on bad acting,
and you are like, kind of like not doing an audit, which is unfortunately what happened with Nouns
the past two years is we never really took the time to really think about should a vote stay one
point or should a vote and be able to be bought in through like $30,000, $100,000? Or should that
vote go up and down based on if you've been voting with reason and after an audit of that
project being completed? I think one thing that is really interesting to look at is looking at
incentive programs. Let's say I voted on a proposal and it passed for $400,000. But that
person was not a good actor, and they actually didn't execute what they were saying to do,
should my vote lose maybe, you know, point zero two out of that. And then that's really where it
kind of gets interesting. And the intersection of blockchain and AI is that blockchain allows us to
prove a lot of things like kind of going to your historical record, whereas AI is really going to
help with processing that data when we're looking at specifically topics like DOW management.
Interesting. Yeah, I we're going to have to spend some time ourselves thinking about that.
Yeah, just a quick comment. I wanted to share a fun fact that influence we basically prepared for
AI agents to participate in DOW in the way that basically this Swiss Association that Tom mentioned,
like we wrote in articles in like a in physical documents and legal documents,
that, you know, members of associations can be like people and like companies, and it could be
AI agent represented by physical person. And that's kind of limitation that we have currently
due to, you know, just laws, we would, you know, just do it agent without any representation,
just like how we had to do it. But it's kind of fun that we are preparing to this world already.
Yeah, for sure. We should have AI agents will be full members at some point, I'm sure. Julian,
Yeah, I wanted to touch on some more use cases that I've been seeing in space, one project in
particular, Parax, or now called parallel finance, where they're using like the alarm systems and
taking the user's intent. So for example, I want to buy, you know, this token, this token, this token,
but you know, some of them might be on different chains. Using their, I guess, their, like,
whatever their model system is, they would actually bundle all those transactions up,
and then actually use account abstraction, and then solve, like, basically route all those,
you know, whatever the user is actually trying to do, and then deliver that in one solution.
And then obviously, the AI, I guess, would be, you know, sorting around to kind of put those
those intents together. Another interesting use case. And these are things that we were doing
at Mads, but we built a proprietary, like, we can buy any single token in DeFi on any,
for many liquidity pool, any collection in one click. So this would be interesting concept
where it's like, hey, I want to like, you just heard about a shit coin, right? And you want to
buy it like this instantly. So you see, you know, it kind of maybe speak to something, I'm not sure
if parallel finance will have such a thing. But you could just say I want to buy a blank or,
you know, this, you know, smog token, and then it would basically, you know, just like chat GPT,
set up all those transactions, and then you just client click Yes. Another cool part actually is
with like, you know, applications like DSO, or any other, you know, social fly application,
is that platforms like Twitter are lagging indicators with information, right? It takes
a little bit of time for us to kind of, you know, the algorithms that kind of push us the content
depending on who we follow, etc. But with on chain conversations, or on chain, you know,
platforms like, you know, DSO, or now, I guess, Farcaster, or whatever, you could have
the more proactive measure for finding maybe the latest, you know, tokens, right? AI could have a
predictive indicator, which then emangulates all conversations, which have a hashtag of something
trending, looking into like the token holders, the token, you know, bag holders or diamond hands,
and really actually predict, or give you better insights of like, hey, take a look at this
specific, you know, market or this, etc. This is kind of trending. That would be a very useful tool
for a lot of people, because a lot of people hear, you got to hop in spaces, you got to all this
manual shit, right? Hop in spaces, see what's trending, you hear it late. So I think, you know,
AI on those specific use cases are going to be quite amazing. I've been playing off my first
initial comment of like, we're going to move into a world where everything is like, transparent,
like, you know, what those like that new on the AI pin, I forgot the name of it, or you attach it
onto your thing, and then you can like, they're saying you can make payments, right? That's also
part of that AI intent model is like, hey, you know, hey, Tom, you know, send me some money,
and then the pain would reiterate your voice, activate the AI wallet, and then execute those
intents to, you know, send you a transaction without even having to even pick up, you know,
move our hands. And also, I might want to just leave the spaces for a second and rejoin, because
a lot of people here are like, I cannot hear, so I would like to hear them. If that's okay, Tom?
Well, we'll wait here for you. All right, thank you.
Go for it. And William, you wanted to say something?
Yeah, absolutely. Thank you. So I have a big data background. And when I look at, you know,
the big picture here, when we're talking about AI, these AI models have huge training sets, right?
And it's just cost prohibitive. I'm not going to drag any blockchains through the mud. But,
you know, a lot of the blockchains that are near and dear to our heart, it's just caused too much
money, right? So you can't really have at scale, you know, data on many, many blockchains for AI.
And I'm not sure what that, you know, if we're going to go to a hybrid model before we can
figure out how to make things cost effective. But, you know, the other option then is to go
to the cloud, right, to go to Amazon or to go to Google. And that's, you know, at the end of the
day, those are not truly decentralized solutions. So I'd be interested, Tom, to hear what your
solution for that is, like what that cost-effective AI solution on the blockchain
looks like when we take into account that we have, you know, terabytes and terabytes of data.
Yeah, yeah. You want to go?
Yeah, sure. So basically, like, as I, you know, mentioned before, there are different
aspects of kind of AI, running AI, basically, like there is there's an aspect of preparing
your model, there's a there's a process of training your model, and then there's a process
of running your model on top of, you know, user queries. And then like the model itself is
basically a black box, like, right, it's a huge thing, huge kind of piece of data that you can
apply to the input and get some output. And this you can run, like, potentially, like,
basically, like on any computer, right, that is performant enough to execute. It's like,
like, in terms of modern computers, like, you can do it on basically any server that has GPU access.
And, and the problem here, like, from my perspective is from
verifiability perspective, like, how do you know that when you do when you send in your query,
your your input, your sentence to this website, this website actually runs the model that was
not manipulated or not, like, somehow, you know, influenced by governments or censorship. And then
this model, this exact model gives you the result. So like, what do you want to get is you
want to get some sort of like proofs, some like, ideally, cryptographic proofs, like in blockchains,
that you, you ran the right model, and this model, you know, really processed your request.
And, and for this, you just need infrastructure that gives you out of the box. Like, for example,
like fluency is a kind of infrastructure, just like built into protocol, whatever you run on
fluence, whether it's AI, or not AI, or just your, you know, simple program, it always like
you always get the proof of computation, you always know that the right computation was executed
against your data. So, and that's basically like the, from my perspective, the main missing piece
for the verifiability of AI. And the one other thing I'd add on to that is there's two other
bits important, but William, we're not talking about on chain here. And so what again, he's
talking about is, is off chain compute, it's doing this, but it's verified and validate on chain,
because to your point, consensus overhead is just too high to run these models on chain. So we're in
sort of violent agreement with you on that. And then the other point there is besides the query
of the model, you still can run models on decentralized networks of computers of CPUs or
GPUs. And if you do that, then you eliminate the possibility of any company owning it and being
subject to influence and manipulation as well. So another way to get to the same answer of getting
ahead about, you know, making sure the question the query was answered correctly. So that's the
other bit. And the first bit is making sure improving the model was trained on a certain
set of data. And that is also used for blockchain, because you've seen New York Times to chat GPT
and open AI claiming their data was used to train that model. Well, I think going forward,
big LLMs are probably going to need to prove the data sets we're trained on. And that's another
thing that, that we think is useful. So you can see we have this very sort of narrow LLM focus
view of AI, what I love about the spaces is we're talking about all these other AI uses, which are,
you know, going to impact all the world in lots of different ways, and even blockchain in lots of
ways. So, you know, I'm learning, learning a lot about it from this and good, good transition to
Bionic, who is right in the middle of that. Yeah, thanks. It's really interesting to hear about that
as well, to try and take the conversation to the point where, how close is Fluence Network to some
of the intersection of Web3 and AI, when we talk about the kind of, you know, take it down level,
the render networks, the IO nets, the get grass and things that seem like much more
infrastructure deep in place. And it'd be interesting to hear how that compares to
Fluence Network. And what you guys just think about that whole deep in scene, which seems to be
blowing up. Yeah, well, we're smack in the middle of deep in so we but we are CPU focused and the
technology we have is applicable to GPUs, but models are trained on GPUs. And so, you know,
IO.net is really a GPU focused training deep in network for models, which is terrific and great.
It's a bit more complimentary to us. You also could run models theoretically in a decentralized
base using IO.net, I believe as well. So we're sort of complimentary to them in terms of using
different types of hardware. We also have a full software stack on top of us, which adds verifiability,
which is what Evgeny was talking about in terms of cryptographic proofs. I'm not sure
that the other networks have that level of kind of stack and verifiability on top of them.
But we are we're kind of an analogous to them with the CPU focus currently.
Evgeny, do you want to add anything there?
No, I think, yeah, I think it's pretty clear now.
Yeah, sorry. So just for my kind of understanding, the IO.net and kind of the bare bones sort of
LLMM training compute, and you guys are more sort of the level of the stack above that the
application layer almost. I would say slightly differently. You're right about them in terms of
they can rent out GPUs and train. And so the idea is they get price training models as cheaper,
a lot cheaper using them, because they have a network and prices could be down as great.
Our network is CPU focused, which is as we all know, most of what the web runs on. And so
AI is not training models isn't isn't what we do. But once a model is trained, some models can
run on CPUs, they can run on the fluence network and then be verified and validated as well.
You also could train and validate the data sets that went into the models using fluence as well.
So we can kind of go before and after but not the actual training of it.
Yeah, and I will add on training that basically like training is a very specific computational
task, where you have like a huge amounts of data that you need to process. And like the best like
it's kind of defines the architecture or the topology of your system or your network, like
ideally, the best in the best scenario, you have all these GPUs, like highly performance GPUs,
sitting next to each other connected with like really, really huge links. So you can process
maximum amount of data per second between them. So you can exchange the data between them,
like super fast. And that's why Nvidia is like is so big with their new things that I forgot the
name of this new kind of container of GPUs that they created. And that's why everyone is so excited
about that. And basically, it's a challenge for any decentralized GPU network is to be
performant and like as performant and fast as these things. And that's basically the challenge
to like really onboarding web to AI people into into web tree decentralized GPU networks.
And like compared to fluence, what we do is a little bit different. So like,
it's more about distributed networks of different data centers and, you know,
and adding security to running computations. And that's why we are kind of targeting it for more
like a model inference use case or like data processing use case, but not really for training
because the training on fluence probably would not be as efficient as it is on centralized systems
currently. Question for kind of the panel here. You guys have views on whether AI is going to be
these AI tools we're talking about, or they're going to be thousands of different AI tools
created independently, run independently, or we're going to have a couple large organizations that
basically propagate, develop, and market and distribute and sell effectively their AI tools.
Like, do you think this is a very fragmented industry, or is it one that's going to become,
you know, much more kind of a benefit of scale with a couple of businesses running that curious
reviews? Julian? Well, I would ask you the question of, you know, how many applications
on product content have you used today? There's thousands of many, many projects that come up.
You know, I think the distribution is going to be key here, right? I mean, we already see with
ChatGPT, they're launching their app stores, right? They're going to be accessing a lot of these
applications from ChatGPT because the branding now, like people just think AI is ChatGPT.
Maybe in the future, it might branch out a little bit more, it's more segmented into like,
you know, micro projects. But I think, you know, humans at heart will still, you know, trust brands
and that's going to take a lot of, you know, a lot of money, a lot of effort. So I do think that
they're, you know, probably going to be, you know, the main players, you know, it could be Microsoft,
Apple, you know, we're seeing that, you know, VR has been like around for a long ass time.
The Apple headphones come out one day and now the market's set to miss shifts, right? So going back
to your point is, I believe that there's going to be obviously one central player. It's all comes
down to simplicity, ease of use, you know, and beautiful experiences. And that is what is going
to continue with the AI side is that people, you know, consumer are going to build, I don't know,
maybe it's probably going to be Apple, right? Or Microsoft, I don't know. But they'll probably
build a very beautiful, you know, experience or ChatGPT, so that anybody comes in can kind of just
personalize your experience to that kind of like a Zapier, you know, but, but yeah, I don't, I don't,
I don't see in the near future that we're all going to be using different type of solutions,
just because like, you know, AI is now just sped up the acceleration of startups.
Thousand years ago, I mean, 20 years ago, it would cost like a million dollars to make websites,
so there's only three players in the space, Microsoft, Apple, you know, Adobe. Now today,
there's like 10,000 startups a day, right? So I just, it's gonna be very difficult to
differentiate yourself in that marketplace. Well, I've got, I definitely have a comment
to that, but I want to give the mic to William before I address it. Go ahead.
Well, just sticking with the ethos of decentralization, I'm hoping that the market
is very segmented and fragmented. You know, look, we're producing as much data in like,
every year is, is, you know, like 10 times as much as the previous year. So that data needs to
make it into training sets that data needs to get somewhere, you know, I see a lot of room
still for niche players to have at that data. Also, you mentioned this lawsuit, the New York Times,
like you're going to see this crazy litigation. So I've spent time looking at what's happening
in AI in music, right? And, you know, the AI models training on on different types of music,
that's just one industry. And, you know, I just I don't see big players being able to get their
arms around or being agile enough to, you know, get through some of these legal, you know, legal
questions around intellectual property, get their arms around the data sets that are very dynamic
to keep coming. So I think at least for the foreseeable future, we're going to see hopefully
decentralized marketplace for like AI, almost like off the like, like taking the algo off the
shelf, you know what I mean, like algo shopping, right, where you can go and be like, Okay, I want
this, I want that. I don't see I see room for small players for a while. Listen, I hope you're
right. I mean, and here's, you know, something that echoes a little bit of your point, William,
but Julian, curious, what do you think of this as a perspective that you've got, you know,
going back to the sort of narrow LLM focus, which is what I just spent most time thinking about,
which is you've got these things, you know, open AI, massively funded, massive developers,
huge head start, huge success, huge use, but controversy in terms of manipulation in terms
of lack of transparency, etc. And there's an open source initiatives out there multiple
trying to do that and trying to trying to become a viable alternative. And that may seem like
insurmountable task, but I go back to the, you know, Linux Windows to fight. And you know,
Windows with his huge head start unlimited resources, and yet Linux with the, you know,
over over a decade with the open source community has become effectively the standard. And so I'm
hopeful, William, to your point, that an open source community for something as important as
you know, it becomes almost a source of truth, becomes not dominated by, by one or two companies.
And you know, that may or may not happen. But that I like to look at that as an analogy there. But,
but obviously, Julian, quite aware of the human nature, we want to go to one thing and use one
thing and people are far more willing to sacrifice privacy and, and, you know, then they probably
should, but that's just the reality we have to deal with. Yeah, I mean, I go back to a lot of like,
you know, like a lot of biopsychologies and stuff like that. I mean, the question there lies,
especially with what William's talking about is, who is the customer, right? Is he a developer or
is he a mainstream user? If it's a developer, yes, I mean, I'm gonna want more tuned systems,
I might want to go to marketplace. And for example, I was playing around with an application just like
with, you know, just as a joke, like, you know, AI girlfriends, right? But when you use the chat
GBT API is that you don't, it's always very restricted versus if you were to actually then,
you know, copy one of these popular models, you have full access. But like, you know,
would my mom use that? Would my brother use that? No, like my girlfriend currently right now she
uses chat GBT, but they don't understand technology, right? They just see they just heard that
everybody's using it. Schools are using it. Everybody's using it in universities. So, you know,
there's going to be a subsection of just like is a regular bell curve, right? You're going to have
the innovators such as your guy's self. You're going to the early adopters, which is basically
a lot of the people who are currently using it now. And in the mass market, it's impossible to
market the mass mark mass, the early adopters, which is as well what's coming up for the next
space, they see 100 ads a day, you guys might see the differentiation between like, Oh, wow,
this is this is really cool. This is like they don't give a shit, right? They to onboard early
adopters, or early majority, you have to have a product in which spreads through word of mouth,
or product led roles, or some sub sub section through community. And it has to be easy enough
for them to share to their friend, which is why child GP was so great, like you would like check
out this thing that does this, this, this, you taught everyone here knows use chat GBT, you've
taught your friends and family to use it. That's how early the majority gets made. But if you cannot
show your your friends and family how to use these specific decentralized marketplaces,
they will never use it. And in the back of your mind, you're probably like, Yeah, you mean this
kind of guy, I wouldn't, I wouldn't, I wouldn't drop this onto my mom or my brother, right? Like,
that's not who they are. So but if the thing is, if you solve that, if you solve that issue,
so if you're decentralized marketplaces, and you have an AI, and I basically maybe might improve
the user experience of like, hey, you're coming in, the onboarding experience is amazing, what do
you think is the solution for that user, maybe in a very easy way, right? So it goes back to that's
how I would go around that. I get I was gonna I was gonna joke and say yes, you have AI build a
good user interface for you. Exactly. Yeah. So I mean, that's that's how I would see it. But you'll
also have to just like I said, right? I mean, and this just boils down to anything is we need to
do a good product. And if it's a damn good product, it can be shared with your friends and family.
Then that's a great market. But then on the other side of the coin, we're building for developers,
we don't have to do that, right? That's a whole different conversation.
No, no doubt. I guess, you know, one other way to take this conversation, which sort of
the answer and a little bit are ethics related to blockchain and AI. And this goes into
everything from transparency to, to privacy, etc. I'm just kind of curious if you guys,
there's any perspectives you have on the panel related to, you know, ethics, whether it's any
of those, those sort of sub topics of it. And have you seen any of that come up in any of your
businesses or any of the, the kind of AI related areas? I mean, we've seen there's some ethics,
you know, people talk about income distribution related in blockchain, potentially, there's also
data, trends, data sharing, is that fair? Is that open or not? Is that good or not? Kind of curious
and it was used there. I mean, I'll try to chip in. In terms of ethics of, I mean, could it be
talking about, you know, taking information and intellectual property from humans and then
using it to replace jobs for those specific humans? Yeah, that's a good one.
So for example, right, like where AI is going to be very powerful is there's a hundred dentists,
right? There's going to be people who are going to try to figure out how to implement an AI system
to read the processes of like, you know, maybe the top performing dentist and take that information,
because obviously all dentists or all, you know, everything that's in a fragmented reality,
there's a different, there's a, you know, there's going to be, there's going to be a difference in
the processes and efficiencies and etc. So, you know, using AI, you can come into some, you know,
some specific job, figure out what's the best way of specifically doing it, kind of stealing it,
and then offering it to everybody else. And then kind of maybe reducing systems in those places,
which could be, I mean, I guess maybe unethical, because now you're looking to optimize a lot of
people's jobs, and that could, and that's, you know, all tools are all about being built to
reduce efficiencies, and efficiencies do mean less jobs for humans. So, I mean, my big question
is, by 2050, is what's going to happen to, especially with robotics as well, what is going
to happen with jobs? Because there's, you know, that question of, will AI help supplement our
jobs in terms of making us more productive? Or will AI as well not only do that, but reduce
inefficiencies, which is where humans can be, whereas like Sam Altman just made a post,
he's, he bets that there's going to be a unicorn company built with one person, because we're all
talking about AI agents. So, does that now trump the initial messages, as he said, he's like, hey,
AI is not going to replace jobs, it's going to basically help supplement you, yet you make a
comment in which you basically replace all the inefficiencies of a whole unicorn company, but
only one person. So now, so now what's the what, what is it going to be right in 2050? We've only
experienced a significant AI explosion in four years. And as you know, with Moore's law, now,
now Moore's law is not even a thing anymore. We move faster than Moore's law. So, does that mean
by 2050? You know what I mean? So I would like to see your, hear your guys thoughts on like, where,
where does what does that world look like? And, you know, yeah. Well, one thought, but William,
you go go first. Yeah, no, these are great points, Julian. One thing I want to mention is, you know,
when we talk about AI ethics, we always get back to implicit bias, right, like all these algos,
to left or to right or to woke. But there's also implicit bias in the data, right, depending on
where we set the outliers, right, depending on what we lasso, so on and so forth. But like,
on that point, what, what concerns me is, we, you know, the, the ethics of what data are,
is being ingested. If we get to a point where we have privacy as a service, and we'll be able to,
you know, have dials where we control our own data and be able to sell that data, maybe to an AI
algo, then things become a little bit more cheery. But, you know, think of think of some of what's
going on with music right now, where you have, say, Michael Jackson, Elvis Presley, you know, these
famous, famous stars that have left behind legacies. And, you know, the model is now training,
training on their music. And it's going to be able to spit out, like, in a couple of years,
you'll probably be able to get a Beatles album maybe better than Sergeant Peppers, right? Like,
is that ethical? Is that fair to fans? Is that fair to the estate of John Lennon? Like, these
questions, like, are going to need to be asked, and then they're going to need to be litigated.
And it's going to just be the wild, wild west, be a good time to be a lawyer and AI, I think.
Yeah, I wanted to add on this, you know, that AI being trained on something that people produce,
that they like really invested resources in, in doing it, and then AI just takes it and, and,
and uses it. And basically, after that, earns a lot of money to its creators. I think, actually,
that's the the angle where the whole blockchain space and blockchain, that kind of web tree
vision can can help because we're kind of trying to enable here the world where, you know, we are
connecting this value, like the content creation with the content monetization, even if it happens,
you know, it's done by very different people, like, blockchain can allow you to basically reward
those who created those pieces of content that were then being used in the model, and then,
like, brought the model, this revenue, so like a fraction of this revenue can go
to those who created content. And this, like, this can be done if it's all on a blockchain,
right, because it's a very transparent and clear thing that basically allows to connect these dots.
And that actually may, you know, even accelerate innovation of AI space, because now people say,
okay, like, I'm, you know, I'm motivated to feed my data to AI, because I will be able to to
monetize it in the future. And actually, it's, it's a little bit like, it's reminds this semantic
web idea, if you remember that, by Timbers Lee from 20 years ago, when he was trying to push
the idea that every piece of content should be kind of licensed, so it has the author, so then
we can see how these pieces of content connected, and we can, you know, go through them, we can
reuse them, we can, you know, follow the license, and so on and so forth. I think it's all applicable
now to AI and blockchain can can help a lot with that. Super interesting. Andrew, you want to say
something? Yeah, I mean, again, I hope the more technical and maybe more seasoned folks in the
room could help me. But am I unrealistically bullish that everyone's been criticizing crypto
web3 for so long, like their use case, we're just sending tokens back and forth. Is AI really that
huge unlock that we've been after? And now we have the infrastructure and the parameters and
the way to rate and check and distribute wealth in the way that we wanted to in a web3 world.
Oh, that's a big, well, hopefully, hopefully, I don't have an easy answer for that one. That
would be great. I don't think in that case, just show my bags, just go back to that. I mean,
there will be links to it. But I don't think it's quite that simple. Unfortunately, Julia.
Yeah, I want to hop off of Gany's point in terms of where, you know, obviously,
blockchain brings a transparency aspect. And I 100% agree, especially with the lawyer thing,
and where this is going to be like, this is going to be this is actually going to be a whole new
like market segment, right? That was previously non existence. Same thing as like, you know,
social media marketing, right? Like before Instagram, these are just going to be new
markets like in a merge. I like to use this example a lot with like understanding like,
you know, web one, web two, web three, or even web zero, right? So like, you know, web zero was
like 5000 years ago, right? A group of Sumerians, there's no such thing as time,
there's no such thing as money, there's no such thing as ownership, right? They basically wrote
numbers on a rock and then created their first base 60 system, which is what we know today as
the 60 minutes on a clock, and then they invented time and then time invented money, and then money
and time then thus invented ownership. But back then, there was no ways of calculating intellectual
properties or, you know, etc. So like, for example, this is why kingdoms were inside of castles,
because the moment I left the kingdom, like, I could be murdered, or I can't say anything,
I mean, it's the Wild West in terms of caveman, civilization. Now we move into web two, a little
bit more, which is like, you know, not web two, but like, accounting, accounting level one was that
system, right? Accounting level two, which is what we use today was started in the 1500s by
Luca Bocoli. He invented the banking system, so like debit and ledger systems. And we have the
ability to create contracts. So we have contracts, we have double ledger systems, right? We have
things to kind of keep us a little more. It's a tool to evolve humanity, right? So that tool,
then I just write contracts, I'm a now my, you know, I'm a record producer, I can, okay, this
music is my record before it'd be impossible to kind of say that that was yours. Now we have
contracts with accountants, we have in databases, which still centralized, but allows us to say,
hey, this is my proof of, like, I own this, but although it's impossible to control it, which is
like, you know, like LimeWire, Spotify, tobacco, right? But now we're moving into a whole new
evolution, which is our like third tool in that segment of humanity, which is blockchain.
Blockchain is now like omnipresent, everything is now accounted for, like, even like, you know,
I don't know, right? The breath you take is accounted for, whereas previously before it was
not. Now we're going to move into a world of like, where I was saying, like the conversations that we
have will be all on chain, every, every sound or everything will be on chance accounted for,
which that opens up a whole new market that we don't even have to have any clue about. And that
goes back into like, for example, the now, you know, the moment I could make like a fart sound,
and that could be intellectual property, I don't know, right, but I can actually then come after
that legally. So I think that's a great way to see the from zero to 100 to where we are right now,
we're moving to a world now where it's like, literally, everything is of a value, right,
like we're talking about token tokenization, anything can be of value. And now we've been
talking about intellectual property, words, anything that we can think about that can be
fit into a blockchain is now going to be opening up whole new markets. And I think if you think
about this way, right, because we're able to like, you know, tokenize everything, and how big like
this new world is going to move into all markets are going to 10 x, like, we're thinking like $38
trillion for the US that is big, we're going to look into a 400, like, you know, trillion markets,
just because now there's a lot more things we can account for, we can come after we can create
businesses for so that's another opinion, I think that's on top of beginnings. Well, definitely
interesting. I think, um, you know, we also means lots of lots of more government things
that governments can tax as well. So it's gonna lead to other other issues, no doubt. But one,
I think, you know, we plan this for an hour. So I think it's about time to, to wrap up. I want to
thank the host. And one thing I somewhat of a potentially depressing but, but opportunity for
everybody here is a quote I heard from someone in Silicon Valley, I can't remember the attribution
somebody may know, but said in the future, everyone will either tell AI what to do, or we'll be working
for AI. So you're either going to be telling it what to do, or it's gonna be telling you what to do.
So guess all of us here want to be on know what side of that we want to be on. So we've got we've
got a head start here. So with that, I want to thank everybody. And if you like this, we're going
to try to do more, more talks every every Friday at this time going forward. So thanks for
participating in our in our first one and appreciate all the all the panelists sharing their views.
Appreciate you for hosting. Yeah, I appreciate that.
Put this guy in this panel was a it was a great chat.
Thanks for having me. Great panel. Great. Thank you guys. Appreciate it. Bye.