Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
Confidential AI: Trustless Tech, Trusted Outcomes
Recorded: May 7, 2025 Duration: 0:33:58Space Recording
Short Summary
CrunchDAO and FADA Network forge a strategic partnership to enhance decentralized AI capabilities, launching innovative infrastructure for verifiable AI applications. This collaboration marks a significant growth opportunity in the blockchain and AI sectors, addressing the critical need for data privacy and security in decentralized environments.
Full Transcription
Thank you. Second try is charm.
Sorry, sorry everyone.
I hope you guys are capable of rejoining.
The space just crashed, literally.
Big apologies for that.
So our team are reposting the link
in the different community,
both on Alpha side and Crunch side.
We're going to give a few minutes for everyone to rejoin,
but anyway, there is a recording.
So we'll be able to, people that are not able to rejoin right now,
we'll be able to catch up a bit later.
So I propose to kickstart things.
Hello, everyone.
Marvin, how are you?
Hi, Gene.
Hi, Phil.
How are you?
Good, good. Thanks. Hi, Phil. Hi, Phil. How are you? Good, good.
Thanks.
Hi, Phil.
Hello, guys.
Thank you so much, everyone, for joining the space.
Thank you, Marvin, for taking the time to deep dive today with Phil and I.
So on the agenda today, we want to talk and dive deeper into the partnership and the collaboration that is happening between CrunchDAO and Fada Network.
It's a very interesting use case.
I really wanted to find the time to leave somewhere on the internet a trace for people that want to hear more about it. So I spoke a lot between token 2049 and recent space and et cetera.
So today I'm going to try to shut up as much as possible because we have, we're lucky enough
in crunch to have very smart people in a team, starting with Philip, a CTO here in the call, and Peter, and Abdenou,
and Enzo, and everyone from the Crunch team. So for once, I'm going to leave the mic as
much as possible to you, Phil, which is Crunch CTO, that's going to represent Crunch and talk
about this partnership from the Crunch point of view and from final
network we have marvin marvin welcome hi everyone um yeah uh thanks uh for uh gene and uh crunch
team you know for uh co-hosting this event uh super excited to hear like to discuss what's going on after Denver's talk for the AI space.
Yeah.
Absolutely.
And so we started speaking in Denver, like you just mentioned, and some stuff happened
between that time and now.
I think it would be great for people that are listening.
Maybe some people are from Crunch and don't know about FALA.
And then for the people from FALA that don't know about Crunch,
maybe Marvin, you want to give a quick
and talk about FALA Network.
Oh, yeah.
So, yeah, TLDR, like FALA is working on to provide, you know, a verifiability infrastructure to make sure we can use AI in a safe zone and with, you know, the original idea of Fala start in 2019,
by then we want to leverage a key technology
that called trusted execution environment
to scale these capabilities of layer one
and layer two like Eastern, Paul Dot, Solana.
And after that, we find the real product market feed for this technology is living in decentralized
application trustless and AI agents and AI infras.
So yeah, we keep shipping solutions based on these use cases.
Yeah, that's basically our story.
It makes a ton of sense.
And I think we kind of started at the same moment at crunch 2019, 2020. that feel can you enter a crunch
for people coming up
from the side of
Twitter?
I did crunch for you building a
collective intelligence protocol
basically that turns
models into recurring revenue
so basically behind that is
a protocol that is a
double-sided marketplace.
On the one side, you have industry coming currently to us, but later to the protocol
with their data that they want to get insights out of.
And they build them into kind of challenges, games, or competitions.
And then on the supply side, we call, we call them crunches, which are
machine learning engineers. And they have models that generate alpha based on that data.
And they compete against each other to who can generate the better alpha. And the person who generates the most alpha
or the most insight that the institutions
generate money with gets the reward.
It's very meritocratic and it basically,
I will dive into more aspects of this in later questions.
So I basically probably leave it here
and just see if there's any follow up questions here.
Yes, thank you, Pierre.
And we can hear you very well,
but maybe you have another alternative for you, Mike,
that will give you a bit more power,
but no worry if you don't, we can hear you pretty well.
Is it the quality or is it the volume?
The volume is a bit lower compared to Marvin
and it could be nice okay i will check
then then marvin but no worry otherwise we can we can hear you super well uh thank you so much
guys for this quick anthroles and so um the first question is for you philip um you just described
crunch down but often when we talk about crunch we like to say that it's a palantir with zero
employees can you explain what does that mean and what is the vision behind this statement and
and and how it relates to what we're trying to do yeah sure let's first first, I don't know if everybody knows what Palantir is, but Palantir is known for turning messy, siloed data into strategic insights. And it does that usually for really large enterprises and governments. And it's very centralized and expert heavy. So how do you take that and build it into a company that has zero employees?
So basically, I will basically talk into what I already said before in our model.
We do not have anybody that solves actually the machine learning problem on our payroll. What we do is we designed this protocol that is the in-between middleware
between business problems and then basically getting these strategic insights
out of the cyber data.
And there's three core components here.
There is on the one side, and this is important also
for later, a term that we created,
which is called the coordinator.
The coordinator basically is not part of our organization.
They come from the business fields,
could be anybody really, but they usually have good understanding of the data and the business problem.
And they can become a player as a coordinator in our protocol.
And this coordinator, what they do is they take the messy data
and just stage it on their note, basically.
And they turn this data into a prediction game that everybody else, which I will come to
next, will work on. And the next role that we have is the crunchers. So these are the machine
learning engineers across the globe, anywhere really, that can now come to these competitions
that are hosted by all of these coordinators,
and Crunch Lab being the OG or coordinator, basically, back in the day,
but soon we will have many coordinators.
And they compete and apply the prediction models against the problem given. And they are then assessed on how good their predictions are,
meaning that how much Alpha Generous.
Lastly, in between there there there's the crunch foundation
which at least for the foreseeable future handles the model running and the model
nodes basically in this in this protocol so the differences here is like like pointed out in the
in the statement there is nobody on nobody on the payroll of this protocol.
But the other part that is really unique here is like we're doing the same thing, but it's really from first principle, very open.
It's there that anybody can participate in this protocol.
Everybody can get to the talent and the insight.
So it's not that you get a company in a boardroom with big bucks, basically.
But if you hire a volunteer, there's secret deals and you need to make this happen in
the boardroom.
Here, you can really just talk into a protocol and start talking to thousands and thousands
of machine learning engineers, PhDs, and people that are experts in the fields
to really help you get more alpha out of your data.
And yes, thank you so much, Philippe.
That was a very good definition of crunch.
I love the way you put things.
And the elephant in the room now is,
obviously, we have all these people that are
collaborating that are building machine learning product for this large institution
but that question is for marvin this elephant in the room is how does um what you do with bala
is solving this elephant in the room which is well how confidential computing is enabling this vision at crunch um and why do
you think it's the right time for uh crunch and fella to work together yeah that's a that's a
great question so uh first of all i think uh the uh let me explain confidential compute a little bit, there's several technologies that can make sure people
having confidential compute or, you know, web standard compute in some way.
For example, you know, as we all know that, you know, JGAF, HT, MTC, all great are critical
graphics solutions on that too. The thing is, like if you compare all of these,
you know, include Proof-of-Work, Proof-Stake together,
all of these solutions, technology,
they can help us to have different security level
within certain security cost. So for example, the extremely case
would be Bitcoin. Bitcoin is very, very secure, very confidential, and the security level is
extremely high, right? But as a trade- the uh comparatively a normal computer the uh the
security cost is 10 to the nine power so um uh but uh the confidential compute uh technology
we are using for example like trust execution environment they can have, for example, two times, within two times of the compute trade-off,
we can still have a pretty good security level.
So that's the first thing,
different range of the solutions, the technology.
But in general, I would say in AI space,
computational compute is the missing trust layer for decentralized AI.
Like because blockchain can enable transparent transactions, but they can't handle, let's say,
private data or heavy computations, which both are the foundation of how modern AI models are built.
more than AI models are built.
AI need more than, you know,
on-chain smart contract, they need models.
And so it means that very, very few
on-chain infrastructure can support that.
So it means that we need a new trust-based execution layer
where the model code can run security
and data states in the privacy.
So yeah, our job is to make sure we can provide a trusted execution layer that ensures that
the AI models and AI agents can be trained, provide inference and data happening, you know, handling happened in the verifiable environment.
So I believe that's the missing layer and the foundation of how decentralized AI could be built.
That was a great answer. Thank you, Marvin. It's really that
Thank you, Marvin.
It's really that enabler of a connection where we have, in the case of Crunch, on this kind of decentralized intelligent network, we reached the points where we have, we need privacy for both hands, with the hands that is sharing the data and is trying to build a better product, and
the hands that is providing models in order to build that meta product.
So by enabling this confidential compute is effectively enabling these two parties to
talk to each other in a trustless manner, right?
They don't have to give away the data.
And on the other hand, you don't have to give away your model or your IP. So that's really, I think, a big change
in crunch. But before diving into actually the specific of that partnership and it impacts,
I'd like to ask you both. What do you think are currently the biggest barriers
in building trust in decentralized AI?
And how can this partnership be the first step
in trying to break through this biggest barrier?
I can go here, but it's going to be very similar to Marvin's answer before.
I think that the barrier that we have, like if I don't look at crunch as a protocol as specifically, but overall it is like if I work on an AI model and I want to give it to people, how can I do it in a decentralized fashion
without giving my secret sauce away?
So this is model privacy.
Then on the other end is like, I want to use a model.
I have data.
I want to get the insights out of it,
but I don't want to expose my data.
And then the last thing is like, am I actually calling the model that I want to be calling?
So, can I verify that the right process is actually being triggered on the data that I'm sending?
These are all four problems to basically enable this marketplace of decentralized AI.
And yeah, all of these are being addressed by TEs.
Yeah, totally agree.
I think for me, I will ask some basic question like, who owned AI, right? Models, who actually owned AI model, who are the users, is have total
controller of the agent that they are using, who can access the AI? And is, do we really
understand how AI works while we're using it? and is the AI model as well is trained
is it aligned with us right with human being or does it present some single company's
interest as a maxi goal or does this present like some you know common sense for the human society?
So based on those very basic questions,
I would say like, if I'm breaking down the barriers on this,
I will say, you know, like a trusted data privacy,
most AI still relies on centralized,
you know, single reliance infrastructure.
So that's a problem A, because it means that who control the data, who control some AI
capability.
Like Grok is very good at searching posts on Twitter.
But if, let's say, the AI I'm using, or some open source AI want to have the same capability,
it will be extremely hard
because Elon Musk won't give away the data from X.
And trust in executions,
like while I'm using some AI agents to execute jobs for me,
for example, for some commercial, for CRM,
or while I'm doing sales, I'm doing interviews,
I need guarantees that model runs exactly as intended.
It shouldn't have any backdoor uploading the data
or it shouldn't present the interest party
that against the mind. And of course uh ai incentivize you know
like i always uh uh joking about like after two years uh most of us uh or i have to say all of
us will lose our jobs because of it yeah so what happened by that it means like AI will replace a majority of our today's work.
And we don't want to be know uh monopoly level yeah it's
essentially important so that's why i think a lot of trials on web 3 space today like
you know incentivize the base on you know training uh you know intelligence for working
or incentivize based on computation or incentivize based on data contribution,
essentially important because that's pre state
before AI totally replace our work.
So yeah, I would say those are the barriers I see today.
And now I would like to move forward and thank you for this dive into the barrier.
I want to talk about this partnership and what are we currently building together.
So Philippe from Crunchler's side, how does Phyla Trust that execution infrastructure
is really helping Crunch in bringing our mission further.
Especially, I'd like to hear some specifics
about how this tech is being implemented
and what it actually enables concretely
inside our community.
For sure.
I mean, this was one, like when I first joined Crunch Labs
and one of the biggest asks that the
crunches have is like where does my code live is it private and we wanted to give to the people
that are running the models a guarantee that basically they have the keys to the environment where the model is running.
And they can verify that nobody else can basically get in there.
So that's on the supply side, how that looks in concretely.
And by the way, we're just implementing it.
Currently, we have a POC that's basically
working with the new protocol, which isn't fully in mainnet yet.
But the way that works is like back in the day or right now,
the code would be uploaded and run in a secure environment, but not mathematically or guaranteed secure environment, basically.
Just with company standards and compliance uh guaranteed um the
new the new model is that basically um you launch this uh this uh confidential virtual machine
and you load your model into it um you you're the only person who has the keys uh to this thing you
you then sign a transaction to join a crunch on our platform basically.
And then the data from the coordinators will start flowing to your model.
And you are guaranteed that only you will ever see this because the data basically goes straight
from your machine into a confidential virtual environment, and it never touches anything else, basically.
So this is much better, obviously, because you can basically say you don't have to trust anybody here.
You trust the hardware and the technology.
So that's on the model privacy side.
And then on the data side,
we always had to implement privacy measures.
For every competition we had,
we had to basically take certain data items
and reformulate them in a different way.
Like one of the games
that we're just going to be launching relatively soon that's based on real-time market data um uh basically uh changes
the real-time market data into into uh a dove that's flying and falcons that are attacking it
so you cannot really while you can still help us predict certain outcomes, and that's very interesting for the institutions,
you cannot figure out where this data came from
because there was a meticulous process put in place to hide it, basically.
But that is extra effort for the coordinator side.
Ideally, you would like to say say i can just stream this data to
to a source i don't have to think about um i i can get guarantees that this cannot be made public
or this cannot be leaked and then you avoid all of that step plus you have even more guarantees
that nobody can uh reverse engineer it, et cetera.
And also for this, that's already in place, basically,
in the new protocol.
And those are the two sides that we are putting into heavy use here and that are solving real problems for us,
especially for our current target segment,
which is TradFi markets, where there's low latency
and you need to quickly respond.
TEs are actually a really good technical solution
for this type of problem.
And thank you, Phil, for mentioning the user base
and the real use case,
which is something that Crunch is really, really proud.
Marvin, I wanted to know from your side,
what actually drew you to build this partnership
with CrunchDAO?
What was it acting for you to spend some time with us
in order to build this custom solution with us?
to build this custom solution with us.
I think Marvin, you are not a speaker anymore.
You have to request again to become a speaker
so I can accept you again.
Twitter is very capricious today.
Sorry guys.
I think I had this problem once, too.
I needed to leave and rejoin the space to get it to work again.
Marvin is laughing.
All right, Marvin, try to figure it out.
Maybe I can move on to the next question to you, Philippe,
and I'll get back to that specific question because I really want to hear your answer out here.
Philippe, on the adoption of confidential AI,
you just mentioned, again, crunched out
what you've been working with hedge funds
and banks and other traditional
institution can do you want to dive into the real challenge we face there uh in term of not only
trust but also institutional expectation yeah yeah um and by way, there's a message that there's a hard stop for the final team. I can continue.
I don't know.
I'm back, guys.
It's really weird.
Sorry.
Can you hear me?
Yeah, we can hear you.
You hear me.
Yeah, do you want to answer that question?
And eventually, we can close on that.
So fix your YouTube team saying you have a hard stop here.
Yeah, sorry guys.
Yeah, I mean, I think for a crunch style,
from my understanding is that we both working on the mission
that AI should not be controlled by single corporations.
So you guys are being a future like anyone can
contribute and benefit from the predictive models, especially
for machine learnings. I think like the very interesting
foundation of this is a I think the contributions from a lot of PhD and strong, smart people, they are contributing their strategy and models.
And these informations won't be valuable if all of the information are transparent.
So key privacy is not just for personal rights.
Beyond personal rights, it's also the foundation of intelligence marketplace.
So it means that, for example, if I know exactly a smarter way for the,
you know, for the, for, for, for under trading.
And if I contribute it transparently, publicly,
then the information is known by everybody.
And it means that this information is not a big value contribute anymore. So I think a fair intelligence exchange or
intelligence marketplace definition of it should
be the intention I'm providing is private.
So of course, I think a nature
combine of our best part will be
Fala is very good at providing
privacy preserving compute for sensitive data and AI models.
So I think that's the work we could add value on.
And second thing is a certificate or let's say proof of contribution.
So sometimes when people contribute some work,
how we can prove the work is there
and the correctness of that work
and the value of that work, right?
It's not a one-time job.
It's more like a long-term state of verification work.
And what Fala is also good at is to prove the correctness without revealing the internals.
So based on that perspective, I would say, you know, the TE or artificial compute technology
is a very good fit for CrunchDoll's future admission.
for CrunchDoll's future admission.
Yeah, absolutely.
Thank you, Marv.
That's a great closing
because in the end,
a lot of people don't realize
that in collective intelligence,
often the best answer
is an ensemble of answers.
And this is existing
in machine learning for a while.
And now it's becoming more obvious with LLMs,
where you won't eventually trust only one LLM,
because it could have some hallucinations and et cetera.
So when you have this ensemble of people
and you can track the participation,
have multiple parties involved in producing an outcome,
that's real collective intelligence
where you can have the ensemble of all the solution
is greater and better than any individual vector
of information.
And that's what we're seeing. And on top of that is the fact that this solution also exists
in competitive space.
Like we often speak about open source
and having completely open model,
but we also often forget that most of the software
that is running our world is closed source, right?
And there is a reason for that because different companies have been competing
in order to produce these things and they are producing values
for both their employees and their investors and their own community,
let's say, with this IPs that they have been building and producing.
And thanks to what Fala is doing we can we're not capable of producing that
to the most smaller unit which is any individual and any human being will be capable of contributing
to a greater purpose guys thank you so much that was great to have you Phil Marvin have a great day
for everyone that were in space thank you so you so much for joining. This is recorded.
We are going to delete the previous version that
is 2 minutes and 51 seconds, and that just completely crashed.
Yeah.
You guys, I hope you guys enjoy the space.
And we'll get back to you when this tech is in production
in the coming weeks, and that we will be able to really prove this tech is in production in the coming weeks,
and that we will be able to really prove that Crunch is moving to the next level
thanks to this kind of tech. Have a great day.
Thank you, Jing. Thank you, Phil and Crunch.
You guys are doing amazing work on decentralized AI. And yeah, looking forward to have
a great
co-intelligence
workflow
based on current stuff.
Thank you for hosting today.
Thanks, Marvin.
Thank you, guys. Thanks, everybody, for listening.
Cheers. Have a good day. Bye.
