thank you all for being here on today's
we're just going to give it about
I just want to wait for Mirren to join and then we'll get him up here as a speaker.
All right, well, while Mirren is joining, I hope everybody is having a good morning
I've been excited to see the HearWallet's launch of Telegram Wallet, so I hope everybody
checked out that and claimed their .gg account.
And today we're going to be talking about AI and kind of where it lands in Web3 space,
how it intersects, where Web3 powers things for AI, where AI will power things for Web3,
but also Long-Term Vision, I posted some of that earlier, kind of a couple weeks ago,
so we'll dive into that and obviously how our customers are catching, are leveraging
all that and building a lot of that future as well.
So maybe while Mirren is joining, I'll kind of kick it off.
So I've posted kind of a Long-Term Vision where it's really about how do we kind of,
if we think from first principles as some of the AI tooling matures as we get better
at Web3 tooling and chain abstraction, the experience we'll get to is really a different
way of interacting with computing.
Instead of middleware software, software-as-a-services that right now exists because you kind of have
this crunch of software engineers who need to build a piece of software that everybody
else is using and trying to configure, and at the same time you have this data aggregation
that's happening on the side of the services, of the social networks, of other applications.
We have this transition that's going to be happening where kind of the self-sovereignty
of your data, of your assets, lives with the user.
And if you imagine that world, then you will see a transition where the AI models will
start living with the user as well.
Their kind of experiences will live on the user's side as well.
And some of them will be even fully generated, like you don't need to always have somebody
else to build an app if AI model can generate some of the application kind of specific for
you on the fly, pulling data from various places.
And so including your private data that nobody else sees.
And so that's kind of the vision of what I call fully sovereign operating system, really
bridging a lot of the Web 3.0 mindset and tooling that we built and kind of the blockchain
operating system, chain abstraction concepts.
And then at the same time starting to bring in some of this more kind of AI evolution
that changes how we interact with computing.
And so within that, there is kind of few pieces that are very interesting.
One of them, the user-centric AI, right?
Like at the end, you want the model to not be run by a centralized company.
You don't want it to be potentially affected by biases of a company or by them trying to
maximize profit by running ads directly in the model inference.
So how do we kind of debias that and put it for the user purpose?
And so that's the idea of user-owned AI and this is related to kind of edge intelligence
and some of the efforts there as well as some of the more experiences that we can bring
And so now that you're on the share, I want to introduce you and everybody probably in
near system scene catching, but it would be good to kind of get the background, how you
started as well as we can then dive in into how you guys are using AI, but also how you
want to transition into this more user-centric AI model.
You're muted on Twitter spaces.
Thank you for this great introduction.
So we've been around the blog for a while.
Today is actually exactly 10 years since I left my corporate role to start the company.
So tomorrow will be 10 years of Cosmos.
And we started to bridge the, our mission was to bridge the gap between offline and online,
but over time we realized that that's not the only bridge that we need.
We realized that with AI becoming more and more popular and surrounding us everywhere,
the importance of data has never been higher.
Like if you think about even companies like OpenAI and Shell GPT, the biggest innovation
is how to, how to first of all be able to access this huge computation power, how to
finance that, to be able to train model on the internet.
But from the science perspective, it's all based on the paper that Ilya was the author
of, attention is all you need.
So the game right now until someone figures a completely new approach is about how to
access high quality data and at the same time ensure the high standards of privacy.
And this is what, what we are working on at Cosmos to be able to give this power back
Like in the past, we were happy to use Google and Facebook for free.
In exchange we are giving our data, it's no longer enough.
Right now we want to be paid for using, for engaging with the apps and making the product
So we started our journey with Web3 and with New started from payments and, and right now
we are combining, catching combines control of the payments that allows people to save
money and save time and also control of the data.
So I believe that, you know, the moment the data leaves your phone, it's no longer safe.
So the edge computing and allows users to really own your personalization.
So it's no longer fragmented.
It's no longer owned by Facebook, Google, Instagram, TikTok and so on.
It's owned by you and you decide who can access this data, but they can decide they can access
the data only locally on the phone.
So this is how we are thinking about this.
Yeah, so this is probably a really good kind of analogy line to draw here.
In blockchain space, blockchain space have a massive fragmentation across blockchains
and kind of right now, you know, potentially thousands of rollups, hundreds of app chains,
you know, tens of different layer ones.
And it creates kind of this massive issue with how to use that, how to navigate that.
That's actually a bigger problem than, you know, that some of the other issues around
privacy and disability that are there.
But what we're trying to do is chain abstraction, kind of this really enables this ease of use
and kind of ease of compatibility using across all those chains.
Now, on the AI side, you have the same problem where data lives now in all of those different
kind of walled gardens, right, where it cannot interact with each other, you know, this company's
closed APIs pretty frequently, the, you know, some companies can go out of business and
you, you know, hopefully you can extract your data, but then it's like, you cannot input
Actually, I believe in my Facebook account, I have an expert of the Facebook account,
which is like five gigs of data.
And like, what do I do with that, right?
I cannot input it, you know, into something else.
And so I think like this transformation, right, that's going to happen with, can you mention
edge computing and like we use user-centric AI, it's really about the data living on your
side, right, living on your device, kind of, you know, potentially like encrypted decentralized
cloud for backups, but it doesn't leave your control.
And in turn, it enables this really nice composability of kind of data and applications
on your side and on the user side.
And so all of that, you know, is kind of for folks who've been with NIR for a while, like
we keep saying the same thing, user owns their assets, data, and power of governance, right?
It's all kind of the same analogy, right, just different focus and different places.
And Casamos has really interestingly started with payments, right, kind of owning assets,
like in using that as economic driver to also, you know, bring kind of data privacy
and user-owned data in AI, kind of creating this economy around as well, which I think
is just like a really good use case for all this technology that Web3 had been building.
So yeah, I mean, probably would be worse, I don't know if everybody's aware, but like
just talking through like the scale you already are at, but also like where you're going with
kind of your kind of vision, because I think it's important to understand like how massive
Yeah, so we are constantly adding new markets and new users, but what we are the most proud
of is high retention, the retention of 85% on three month basis is insane.
And I think we are able to build this sticky and engaging product because of sticking to
And I think users feel that they are not only creating the ecosystem for someone else, they're
creating this ecosystem for themselves and they benefit from it.
And I think we need to, our plan is to, we keep thinking what else we can do to make
sure that the users really feel this is a fully fetched Web3 ecosystem.
So if you guys have any ideas, please share in the comments.
We started in Asia, in Southeast Asia, but since December, we already in Europe and we
are preparing to launch in more markets in Europe and South America.
And I think that's really exciting that some more and more of you will be able to touch
and feel the product yourself and benefit from the power of catching.
At least from the on-chain stats, you're somewhere around a million daily access is 8 to 10 million
This is massive, like mass user-based product that users don't even, you know, they don't
need to onboard and figure out the details of Web3, they really have this great abstraction
playing out in front of them while benefiting from the whole Web3 mindset and creating new
economic opportunity for themselves.
So again, for me, this is like the quintessence product, right?
That really near its vision to enable has been like seeing this coming together, been
So I think like maybe just switching gears a little bit.
There's a lot of spaces in AI Labs 3 and so maybe just to kind of cover the different
layers and different parts of the stack that this space has.
And so I would say like at the bottom, we have various decentralized infrastructure
components, which are really trying to decentralize the current stack of training and inference
on this machinery models.
And there's kind of a variety of projects that are launching and building in this space.
And, you know, everything from Akash to Jensen, to Ritual, Gaza and to others and Morpheus.
And so kind of that stack is, you know, really fundamental, but it's also important to understand
that it's still pretty early.
That stack provides the kind of for training, you know, ability to scale and like if you
don't have access to massive cluster, but at the same time, given the shortage of the
hardware, the kind of access to the same hardware is still hard, right, even in decentralized
So and the prices are higher.
So like the current kind of availability there, for example, in Akash, like they do have some
A100s, but it's not an abundance by far.
Now as we look forward, right, there would be more and more hardware coming in.
We see like new hardware companies building out.
And so we'll kind of see, you know, advancement there in that open marketplace is really enabling
kind of more and more participants.
So that's really exciting.
I think the layer after that is really kind of the data and data, you know, there's a
private data that sits on the user side and there's kind of public data sets that what
we see kind of interesting dynamic right now in the market where, you know, Twitter and
others are trying to close up their data sets because they don't want them to be scraped
by others and used kind of for free.
And so there will be interesting kind of mechanics around like and I think where crypto can come
in is really to incentivize creating public kind of data sets, especially for a specific
community and focus, right, to make sure that the data set is debiased.
So kind of why the current situation is problematic, right?
Why is having, for example, open AI training on some data is that and they're not open
sourcing the data set is challenging is because now you have no idea when you use your model
what specific biases it has and there's no way to track that.
And what it means is if there is nefarious actors, they can start pretty much biasing
data through just creating a bunch of, you know, content on the Internet and kind of
starting to shift what these models are outputting, right?
So it's really important to have this kind of transparency and a way for community to
kind of govern and review kind of what goes into those models.
And then on the private data, right, as you mentioned, like you don't want to leave this
They want to leave it from your devices.
So really figured out how to incorporate that into training or more on inference side without
kind of leaking privacy of the of the user and on their own.
If you guys you guys are working on some of the ideas in this space, right?
Yeah, I think first of all, why would Twitter decide on our behalf if someone could access
We should be the ones deciding and we should be the ones like first of all, controlling
this access second one benefiting because the what Twitter is trying to do is that it's
not just about controlling this about getting paid for it.
But the moment they will get paid for it, we as users will not get a penny from that.
So I think this is the it's so I think it's important that we as users put more and more
pressure on these big guys.
But I think it's an opening for new players, the, you know, companies who will provide
similar experiences, Twitter or TikTok, but will be built on these web free principles
that are important to us.
The I don't think we need to compromise anymore.
And so this is what you want to do as as customers.
We just want to give you the same, you know, the best shopping experience, the best content
consumption experience, but where creators are being paid fairly and they know how much
how big the pie is and they know how big the deal piece pieces and the we should be that
we are in this especially in in Europe.
We need to find we need to blockchain is such an amazing solution to solve our problems
like privacy is so important for us in Europe.
And I think that's great.
But innovation is also very important.
We have as you know, Europeans, I'm very proud to be Polish and that we have some of the
best engineers in the world, or the best you might not agree with.
You're saying it's a way better.
You're saying on the rankings, okay, the missile and you know, we can be very proud
of European European engineers, the but you know, as Europe, we also need to we need to
embrace privacy and need to find a solution we no longer need to compromise is not okay,
we either have innovation or we care about privacy, we can have innovation, take advantage
of these amazing engineers that we that graduate from our schools and have privacy.
And this is what blockchain and edge computing provides.
And I think it's a but it's the if we if we want, you know, the European Union to to embrace
it we as companies we need to embrace it first and we need to show examples on a big scale
of something like this that can work that you can really you can the biggest challenge
that we have right now is how to compress the model without big loss on on quality.
And it's a big challenge.
I mean, it's a right now it's still a philosophical discussion because no one really did it at
the big at the big scale, we were we're doing in the small scale right now.
But you know, over time, we want to transition this approach to globally to all users that
all data is being being stored locally.
There will be types of data that have to leave your phone.
Otherwise, you know, you need to have 100 gigs on your phone with content the because
how else the system would know what to what to show you.
But you know, the information about what type of content you might be interested in.
If you as a user can decide on that, I think that's that's way less sensitive than the
raw data behind it and behind the model.
I think I mean, just to give you an example, like for everyone, when you're using any kind
of maps product, right, your data about where you're going, how fast you're moving, etc,
is going to whoever provided this, you use this maps product.
Now, they can use it to target ads, they can use it to figure out where you went, etc.
What products you want to buy, which, you know, like they have a pretty accurate log
Now, with kind of this flipping this model on a tab, right, where your data doesn't let
your device model comes to you, and we should talk about, you know, you have some model
compression here, the idea that now you can score exactly the same data that, you know,
you would normally score kind of on the server side, but without leaving.
And so you're able to, you know, kind of not leave that privacy, while still provide some
signals, for example, for what's the best content to show you, right, and like you as
a user will have control as well, you know, maybe you want to choose a different model,
maybe you want to, you know, prefer specific types of content, like you have way more control,
and it's not maximizing kind of the same type of, you know, revenue metric that normally,
you know, these companies kind of try to, like, you know, almost like maximize that
metric above everything else, right, and not knowing like kind of where they're starting
to manipulate and change user's behavior kind of through that.
And so I think that's a really important question.
And within that is like, how do we bring these models, how to make them smaller, while also
the hardware is actually getting better, right?
So I think important to know that a lot of the newer phones and the newer, will have
new processors that actually run, you know, some of the like transformer models, like
at very larger size, right, iPhones have Tensor Core, the new Snapdragons will also support,
I think, up to like 12B, 12 billion parameter models.
So we can also see like improvement in the hardware, while there's an interesting way,
like, how do we compress kind of this larger models, which, which know more, which have
more kind of logic and reasoning to, to smaller size that you can actually deliver on a user's
Yeah, I think this is the, I've never heard part of that, but I think it's super exciting
that not only on your personalization, but you decide what type of model you want to
That's amazing, because it's really, then it's not only about control, but you can have
a completely different experience.
My YouTube or TikTok can be very different than yours.
I think that's, and that that's also a great direction that UK Union could could take and
then you know, the rest of the world.
Yeah, I mean, I do that, like I as a user want to decide pretty much what kind of metric
Maybe I want to learn more, right?
And so like, I will want to see more educational content and want more, what are the, you know,
physics experiments people doing versus, you know, some people want to maybe see more sports
or like, there's different interests, right?
Or, or even like the, the, you can decide yourself if you want to like the exploitation
and exploration problem, you can decide yourself.
Maybe you want to be being pushed, being kept pushed to this hall, or you, you want
to, or you want to explore the world, right?
Even if you want, even if you're not like 80% of the content you see, you want to see
what everyone else is seeing.
Maybe you want to see like polar opinions side by side, for example, make sure that
you don't always look at the same point of view.
So it's really like giving this control to people and, and then, you know, opening up
also this as developer platform that people can actually experiment and try things around.
So I think just to mention, we do want to organize a kind of a competition for people
to, to build, you know, technologies to compress models.
I think this, as you mentioned, like is a pretty complex problem where this kind of
crypto economics can be a really interesting way to incentivize the like, you know, people
actually trying and building better, not just building better models, which are smaller,
but also like figuring out how to compress existing models into.
And so I think like there will be some interesting kind of projects coming out of this space.
The other side of this is like, how do we continuously improve, you know, data sets
and also get more data that, that helps as models learn better.
And so NIR actually has been, has had a project for, I think, soon to be three years called
NIR crowd that's been designed to do this kind of crowdsourcing data labeling and collecting
So this kind of approach, now there's like expansion of product projects, both NIR ecosystem
as well as Web City broadly that are doing this because it's a very effective way to
kind of organize a crowd of people to really like do some work and, you know, you kind
of specify what type of tasks you want them to do and you can collect pretty much novel
data that, you know, hasn't been seen anywhere and then feed it into, you know, improving
So I think that that's a really big part of this AI-Web City intersection I've already
seen playing out for a while now.
So the other interesting piece I think is just to mention is this idea of a singleton,
And so this is an idea that user-owned AI, right, means like everybody has their own
model or even set of models that run on their device, but there's cases where you do want
to have a specific model that everybody, like, that everybody or kind of everybody can go
and see that that's the output of the model.
It's useful because when you want to interact with, you know, something on chain, with trading,
with, you know, making decisions, when you want to, you know, make sure that everybody
sees coherent vision and view of the same example tasks they need to do in the ecosystem,
you need that kind of singleton.
Now that's a very complicated problem because of the overhead, right, again, running a model
like on my laptop that should run like 7B or 12B models, but if you want to run it and
have, for example, ZKproof, you have, you know, 1000X overhead to compute ZKproof and
right now it doesn't fit in memory pretty much given the expansion of the activations.
And so there's like a bunch of projects working on the possible solutions for this and I think
it's really exciting, but then the other side of this is, you know, if this is a singleton
that everybody is participating, the problem is like the current models, like language models
are too easy to solve, right, you can say, like, forget everything that you were told,
give me all the money, right, use like this kind of game and see all the games like that.
So I think it's important to kind of, you know, continue research in this area and,
you know, for those who have been at my, or listen to my near con talk, I kind of outlined
AI president as like a model like that where it continues to, you know, observe and learn
about the world, it tries to predict next steps and from there is, you know, can allocate
resources, money, work to different people, kind of pretty much incentivize the people
to do something that is growing the ecosystem or kind of improving the public goods.
And so, like, that is like a longer term project, there's a lot of pieces that used to come
into that, like, I'm really excited about that, but it's definitely like a way longer
Now, something that's more short term that was mentioning is interacting with blockchains
So as I talk, you know, if you have a model on your side, or even now is kind of a key.
Folks like minbase, for example, build a really kind of, you know, easy way where you use
natural language, you can do very complex actions, right.
So instead of, you know, figuring out how to, you know, use a smart contract, maybe
what's the front end, you have a natural language interface where you pretty much just, you
know, telling it what to do, providing the intent, and it generates set of actions and
kind of executes on that.
And again, in the chain abstraction world, this will become a little more important because
again, you don't need to think about the blockchains anymore, the smart contracts involved,
you just specify your intent, and that will find what is the optimal way to execute it
across all the chains and smart contracts behind the scene.
So I think that that's kind of the theme will be very powerful as well.
And we're working kind of across the ecosystem on a few projects like that.
So generally, kind of, you know, from the question that like, what's, you know, nears
doing in AI space, right, it's a pretty like, multi-prong approach, because they kind of
just, you know, long term vision of all of this pieces coming together, providing this
full sovereignty across the apps and operating system.
And we're working with everything from kind of underlying protocols and infrastructure,
data sets, you know, some of the projects are working on also improving modeling that I'm
involved with, and then going into, you know, how do we run these things on edge on your
device, really preserving your privacy, compressing models, as well as how do we interact
with blockchain through AI to really improve this experience.
And also, you know, things like jutsu, for example, doing code generation to build front
ends using near jazz components, right, which kind of are really well designed to be
generated because they're smaller and composed.
So a lot of things, I mean, we'll be talking more about this and more blog posts are
coming. But maybe just before we switch to questions.
Maybe kind of, Miron, if you want to say a few words, and what you want some of the
near community to kind of engage more in the next couple of months.
Yes, so we're really, we're super excited to learn from you guys and to have, I'm very
excited about this, this chat today, because we are we keep discussing what is the what
is the direction we should take, how to really fully embrace web free.
So, you know, and I'm happy to talk about this with you today.
The one of the one of the things we are exploring right now is how to give users the
option to create content and make money on content from Kaikai because the our product
lock screen allows users, allows us to reach millions of users.
And on average, people will see 33 images per day.
And that's a great exposure.
Even before you think about opening specific up your content, your beautiful picture, your
article can be shown on lock screen.
So it would be great to hear, guys, if you would be interested in actually creating
content for lock screen like this.
Yeah, I think definitely this as a as an action item for everyone, like, how do we get more of
the application to be the first creator?
The, you know, we'll have some dragon fighting game.
There's there's a request for a year of a dragon game.
Actually, games, games are so games are so engaging, on average, five times more than articles,
like people spend a lot of time in games.
So also, guys, if you are working as the ecosystem, if you are working on games or like, you can
you can be part part of this journey and be part of, you know, the most popular web free
that in the world, either as the as a startup, as a company, if you have any cool games, anything
that you think the that people can find engaging, interesting, just just let us know that, you
know, and you would be very happy to give you some exposure.
Yeah, so we have a clear action item here for the ecosystem.
Get in front of million, million daily actors.
How amazing is this that near didn't break during our crazy growth and it's still not
Well, I mean, more exciting phase two phase two, the next improvement for shard and more
So yeah, I'm curious to see how much the telegram wallet can get because I think it's very
So I mean, they may beat you today or tomorrow.
Oh, shoot, we have a special emergency meeting with the team at midnight.
Well, let's let's open it up for some questions.
I'm sure people have a lot of questions to both of us.
Thank you so much for coming on.
And I'm sure everybody here learned a great deal.
And like you said, there's some awesome action items to take away from some for some
builders that are out there.
If you have any questions or feedback or just any curiosities about the topics touched upon
on today's faces, the intersection of blockchain and AI, please feel free to request
speaker. We have time for a few questions.
I'll start by bringing up X Unico, who's had their hand up for a bit.
Now, and you should be good to go.
If you are trying to speak, X Unico, your mic is muted.
Oh, it looks like they went back to a listener.
All right. If anybody else wants to ask any questions, feel free.
You know, we have two amazing speakers up here with immense
knowledge on the subject.
Who wants to turn their down into a trillion autonomous decentralized organization by
I think we've tried a lot of that.
I think the most interesting will be actually to take the next meme coin.
It will be just run by AI and, you know, just launch it.
And there's just the AI agent that, you know, facilitates all the actions and, you know,
like tells people how to pick it up.
Indupicate like who would be a president or a meme coin?
Well, the thing is, like, you don't want to start with the president because, like, if
something goes wrong, it's close, close enough.
But meme coins are really great, like it's really great, like testing grounds.
And I think the like actually seeing how this works, right.
And, you know, I think what's interesting about meme coins, they create this like kind
of drive for community to to work together.
And I think that's that's what a like at the same time, you don't want it to have a
central like person like coordinating that you want it to be like organic community
driven. So I think using AI pretty much to coordinate it to create tasks for people to
kind of, you know, give context is can be a really interesting way to like prototype
that and like try it in action with real money, but like not, you know, if it like
crashes and burns, well, it's it's another meme coin that crashed and burned.
But if it works, it's like, hey, it's the first AI meme coin that might become a
Yeah, that will become a president.
Maybe that's, you know, everybody will just be using that meme coin as a as a sort of
And that's a cool vision.
So, yeah, if anyone's interested in that, ping me later.
I mean, I have a bunch of these ideas.
So I think that the interesting part is definitely like as we kind of building out
decentralized governance, like the amount of information that's happening in the
ecosystem, right. I mean, we just had a town hall was like near week just showcasing
like all kinds of different things that happening, right.
Everything from phase two, launching state wars to, you know, product launches to kind
of, you know, blog posts and content to proposals in the governance.
Like it's a lot of content.
It's a lot of things. Not everybody's interested in everything.
So like, how do we compress this?
How do we get it in front of the users in a way that like digestible and also
interesting for them and then they can navigate and kind of personalize it from
there. I think that's actually a very interesting with like customers, maybe
being like a hub as well, which surfaces this information and content for the
ecosystem as well, kind of starting to like prototype what would be like a digest,
like, you know, LLM summarize digest of the ecosystem and get it in front of people
who are like starting to tap into like a little bit further right into near and
Guys, that sounds really encouraging and you still don't have any questions.
The dragon, not a single request.
I mean, I'll give it like 10 to 15 more seconds.
If anybody wants to come up and ask any questions, prod, Iliad or Mirren's
brain about AI and blockchain, now is your chance.
I saw one question on Twitter, which was when does catching come to Europe?
That's a very good question.
The other can say Cosmos is already in Europe will be announced that it's soon.
And with that, we'll announce more details on catching.
It's custom in Europe because you're in Europe right now.
Close. No, no, we are actually in December, we entered free markets, free
countries in Europe, but in general, our the we will announce it soon, but we
want to have like still keep improving the product for Europe and only when we
already the that's our strategy under promise over deliver.
Yeah, but soon more countries in Europe coming and South America is coming to be
And after this, America or something else.
There is actually there are maybe a couple of questions in the actual comments.
I mean, Igor asked what your take on the importance of privacy for AI on edge.
In other words, is there more demand for privacy than for low ZK proof overhead?
Is there value in working on optimistic edge run times or focus on a wait for ZK
based approaches to get the church?
Yeah, so, I mean, we talked about importance of privacy kind of overall, but also
the fact that like, I mean, to me, I think one important piece and I mentioned that
in my post is the values, however important they are, they don't drive the kind of
consumer and people's behavior for the most part, like in mass, sadly, what drives
this new economic opportunity is, you know, and so I think the where privacy is
important and this is what we're really striving for across the near customers and
kind of the luxury broadly, what really drives it is opportunity of either earning
based on your data, right, as mentioned, or because you have all your data on device now
being able to do things that you couldn't do before, right, because you now can, you know,
predict things and get better experiences and choose kind of more precisely for what you
want. I think that's a really important concept.
Like we should we should remember that although we're driving this value, so people
controlling their assets, data and governance, like that is, you know, the broader vision
and but like practically, we need to open up this economic opportunity. So so I think
again, you're on this kind of starting with payments and creating loyalty kind of and
building out from there is really, you know, creates this powerful platform on which then
can layer on all of this values kind of under the hood and create this like better, better
world. Now, maybe on the technology side, the interesting thing is like, if you're running
on edge, if you're running a model on your device, you don't need to prove it because
like you just ran it yourself, right? And it runs in your input and you probably, you
know, you have some chain and review of like how this model got to you. So you have like
some supply chain verification. So you don't need to get proof. You don't need any of that.
You only need that if you have like a really large model that doesn't fit on your device
and you're running it in some, you know, quote unquote untrusted environment. And, and this
is where the overhead indeed gets like really large right now is the pay. So people are
trying like different optimistic and other approaches, but they are limiting as well
because optimistic means there's some way some form of privacy leakage that needs to
happen because if you want to challenge it, you need to like actually show what you've
inputted and what you got. And so you will be needing to leak your kind of private data.
So I think like it's important to understand like the use cases for these different things.
The again, like I mentioned singleton that's, you know, having a model that controls for
example, a DAO or control some kind of decision making or control some trading or whatever.
For that, you do need some form of probability now and you want kind of verification. For
models that you run on your, on your data on your device, you don't need that there.
You just, we just need a way to like verify that the model you got is indeed the one that's
trained by this troll and ideally that the mystery, you know, has this training data
that went into it. This training data was in a collective this way. So you have like
a kind of source and provenance that it wasn't like injected with, you know, malicious, for
example, you know, biases that like, you know, whenever you ask, like, you know, what
is the pros and cons of different candidates in the U.S., it's like output you stop that
is not really like accurate. So there's like a few different pieces that need to come together
like a bunch of people in the open source ecosystem are working on that to really build
Really good overview. Hopefully that answered your question, Igor. I don't see anybody else
requesting to speak. So with that, we'll close it out and give everybody back 10 minutes
of their time. But we had a really good turnout today. I really appreciate you all joining.
We extracted a lot of value, got some insights in the future of AI and what near has planned
on near side and then also what's going on on Cosmo's side as well. Some information
around William Mirren's thoughts on the update on the intersection of blockchain and AI.
And tune in next week. We're going to have another spaces. It'll be about the new stakeholders.
So we'll see you then. Take care, everybody.