A Scalable and Privacy-First Framework for Decentralized Intelligence | Pavel Sinelnikov - Metis

[Music] hello hello hi everyone how's everyone doing today woo come on we can do better than that woo yeah okay we got one more person joined awesome uh before I give my overview my presentation I want to go first and say what I've done and how this is uh developing into more of a of a of a final product the whole vision of this so I've been in the meta ecosystem for over four years it I've been getting paid in crypto for longer than that doing presentations doing um infrastructure doing uh activities like coding I started out in the space as an ambassador coordinator moved on to being Business Development moved on to doing developer relations in it there's tons of infrastructure and projects that I've communicated with tons of ideas tons of this open source knowledge I've seen many projects pitch the same ideas over and over in different years the idea is to take something new and Innovative and to make it better that's The Core theme that I've learned across these four years I built an infrast a platform a programmatic influencer marketing platform called chorus I've also founded and built the governance platform that Med governance infrastructure and platform that metus currently uses understanding fundamentally the infrastructure needs as well as the application needs is primarily what I'm here so this is just the interface for chorus we just distribute bounties uh programmatically in any erc20 token here's uh the homepage for C.V which is the main governance platform which has over the actually the top 10% most activity on Snapshot across all projects in the ecosystem so we built this into something that is recognizable across the community and the interactions The Core theme for mettis is innovation at its core metis's infrastructure essentially speaks this we're the first to have offchain data availability when no projects were doing this now we have data availability Solutions and data availability layer twos as as well that exist we're the first to have decentralized sequencers essentially making it so that you don't the the user end can be abstracted while still remaining decentral decentralized at its core you don't need to know smart contracts in order to H benefit from decentralization we are moving forward with with the world we've seen AI grow into this massive Beast we've seen different types of AI models AI Frameworks that have come from this Vision we've seen practical applications and use cases but we haven't seen this extend to the blockchain space in a way that's reliably decentralized we are building aith a scalable and privacy first framework for C centralized intelligence traditional ai ai traditionally has outpaced crypto tons of Investments going to providers in centralized regions that then have close Source algorithms as well as data we don't know what's going into these AI models we can't verify what is coming out so there's an untapped potential in blockchain technology to make this more open and verifiable so AI thrives with data we essentially have the web3 adoption coming from the data availability as well as the Integrity of it we are prioritizing this Vision this aspect to take what we put into these AI models as the absolute truth so we're looking at four different types of solutions to this problem data sovereignty and accessibility privacy Resource inefficiency as well as the Limited decentral decentralized nature of these Solutions we have created a structure to create to outline how to do this int we are introducing data anchoring tokens as well as ident as well as entities called ows to regulate the information that is being put into AI agents to make this system a more open and transparent way to see what are being what data is being put into these AI models we want this to be open and composable our idea is that your data is valuable and as a result we want to know what's put into these AI bottles just like how we know what's put into the ingredients in the food that you buy we want to know what information is being put into to the AI models that you consume so we're taking the four pronged approach scalability privacy interoperability and modularity the goal with scalability is we want to decentralize the compute power of these AI agent that the AI agents use we know that compute power is very compute intensive right now only a limited amount of providers can provide those infrastructure we want want to make that more optimized for the general user we also are looking at privacy ZK proofs and I worked for a project called I work for a project called zkm we're building a generalizable ZK VM albeit not necessarily focused directly on privacy but privacy can be added in as an additional feature mostly on the data verification making sure sure that the data is integr is integral and later on expanding it into the Privacy preserving use cases next is interoperability we want to make sure that this tool plays well with blockchains as well as other AI Frameworks and AI tools so interoperability and modular design we know that as an as a single solution cannot solve all problems we want to be adaptable to change and our goal is facilit ating this type of vision through these types of Integrations so aith has a life cycle management you as the person are Central to this management you can create manage execution as well as terminate the AI agent we want to make sure that the execution is verifiable and decentralized so using techniques like ZK proofs we can Ensure that the ver that the data that is being outputed by the AI agent is correct we can also ensure that the execution of the AI agent is correct in addition privacy with the introduction of ZK you can mask the information that gets put into an AI agent so therefore you can take multiple different things and masks mask them so right now identity ZK uh ZK TLS is a good example of this where I don't have to reveal all my information all my data related to uh for example my ID my passport I can selectively choose what data can be used what data is presented and that way we enable other use cases such as Healthcare and finally integration layer we want to integrate with crypto natively right no AI tool or at least the AI Frameworks uh they have a way to bridge that but not a way to directly in integrate and interface with these decentralized tools so aith right now the differentiators why should you use it we support different types of tools languages rust python typescript also we're focused on the performance in terms of adopting rust rust is a highly performant language we want to make sure that every that modularity is key in it so in this way focusing on this high performance high scalability as well as the security of your data is integral to what we want to create we also want to create things that are future proof that evolve with the web3 ecosystem this isn't just one framework that will be launched with one idea and just with that we aim to adapt to what the market needs just like how we adopted with offchain data availability and decentralized sequencing the same way is that we will innovate and adopt through these future frof systems we are integrating traditional AI tools right as a stepping stone to the that direction we want to use models that people are using today we create highly extensible uh abstraction for this as well as high performance inference and uh different types of workflows depending on the use case so of course the four main use cases but obviously this isn't this isn't the uh the whole list but the idea is that we can enable other infrastructure to run on this uh on on on this overall client Bots agents uh the different different services that will then consume uh various different types of information making sure that verifiability is key and data Integrity is key we also our vision puts the person at the center AI can AI can help with the advisement AI can help the person make a decision but the AI should not make the decision itself there needs to be some form of a control and barriers now of course when we have when we Institute those controls we want to make sure that these things are communicated the idea is that if you launch an AI agent and it does something bad that means that the person that that Co that that created it with that intent to do uh we want to make sure that accountability towards this AI agent framework is placed and that there are these bounds and controls we do have a campaign to increase the different types of usage and and of course the the uh expandability we've already integrated Eliza OS as well as Lang chain within this framework we're looking at other Innovative Integrations as well as use cases for this we have a couple social groups but the main one that I want you to ideally scan and go to is the Biddle hour for Denver which is actually tomorrow uh we have various different types of infrastructure participants who are looking to build on top of this vision for AI and collaboration different Oracle providers rpcs as well as the core and base infrastructure we hope that people he the audience that uh is targeted is is is specifically focused on building the Next Generation applications we think that through these workshops through these Builder interviews we want you uh to grow from this experience so our goal with this event is to uh make it no show what it is to build from the Builder's perspective so highly recommend attending this event tomorrow really really good now I do have some time I want to demonstrate what exactly it is aith can do very briefly and the idea is to leverage this tool to do more than just what what it is that that that I'll I'll show you so I uh as part of the initiative for the decentralized um the the way for collecting data to operate off of we have ran a program called Community verified projects where we take feedback from the ecosystem and have the community vote on different projects entering in the ecosystem the goal of it is to promote every project that comes into the medic ecosystem with the amount of marketing support that that that we can provide you know that that that's our goal but in order to do that we want to also make sure that this process is democratic we want to make sure that every project goes through an evaluation process and essentially convinces community members to adopt we we have a number of projects that went through and as a result we have been collecting data over all the projects all the launches all the different various different types of outcomes as well so we've been doing this for over three years and as I said we've reached top the top 10% of uh votes directly on Snapshot so we've gotten a lot of data we want to iterate potentially over this data and any other related make any type of inferences related to this we want to see what are the outcomes that projects have launched on The Meta ecoystem How can I adapt my messaging to capture value within ideally any ecosystem right not just metus how can we take these principles and apply them generally so right here I have like a very quick demo uh there's an API route c. vote API proposals so over here we have a tons of data related to each and every single project that has gone through the governance process within the meta ecosystem so we're plugging this in and um we're we're just plugging this in to either Claude or gp4 turbo for now obviously we need the uh we we need more data more inferences for for additional customized models but the goal is to take what work already and implement it and if we run this we we can essentially ask it show me all of the top performing proposals you know that's a very generic way to uh to ask something but the idea is that we can take data that we already know that projects that already pass through this system and we know that it's operating on this type of data our goal is to make this data known and our first step is to integrate it with our governance and of course we got like a few a few responses and you know we can we can uh change the different type of model to for example gp4 turbo which then gives it more more of an umph um for the end result and as soon as it compiles and it'll output a result assuming that my network my my mobile data can uh can handle this request but yeah I guess that that's that's the the overall overview and as well as a brief demo thank you do I have any questions future Visions you know all these all these other things awesome he'll be right with you we got we got the we got the mic coming so you were talking about how AI can't make every decision for us and that's something I think about a lot like how much governance can we have with AI and I also look at the other side of it though too like how much have people messed up governance right so I guess my question is like where's your focus on what decisions shouldn't be made by Ai and also I I wonder how much I feel like we can make a lot of protocols and smart contracts that can help us do some sort of like conditional formatting to to figure those out and I think it's about just designing it ahead of time so I wonder what most pressing things humans should be making the final decision on absolutely thank thank you for the question in terms of the how much power should the AI be given we're thinking of it as a as a series of humans should be at the center we want to make uh we want the person to make the decision over what is happening over the processes there should still be someone accountable and responsible for the decisions that are made the AI agent can do a lot within within advisement especially when you have the data associated with it but the final decision should always be made by an individual by a specific person because otherwise because I've heard that people are creating AI agents and then specifically coding them to rug something and then you know what is the legal respon responsibility what what is the responsibility of the person if if you're not accountable for that decision being made if if huge amounts of losses are are are are are put in then who is right and and then in that in that world of an AI agent can just take away all the money that's not something that we believe would be the the future I think having these barriers especially with the person being accountable and responsible is the the core the core set and I think that's that that's where uh that that's where I'll I'll summarize yeah time's up okay perfect thank you