AI x Crypto The 30k Foot View | Niraj Pant - Ritual

is this on can they hear me is this on yeah hey guys is that coming through I feel like it's not they can hear okay hey guys thanks for coming to this talk um I'm going to be presenting on AI crypto and uh really talking about the the kind of fundamental underpinnings of this space and um you know really giving a kind of a 30,000 foot overview of of why I find this space so interesting and why I'm working on on this company ritual which which I'll also go into uh later on in the in the talk so the motivation here is AI is eating the world um from education to Media defense art culture and the usage of these products have gone up pretty significantly in the last in the last year and the you know you could think of it as the intelligence quotient for AI has gone up pretty dramatically in the last year so what are the problems of AI today one is that the infrastructure is highly centralized model storage governance ownership compute is all run by a couple of large corporations and a couple of large Cloud providers the AI infrastructure is also highly permissioned there's high costs to using it it's permissioned there's no privacy or computational integrity and a lot of geographical limitations then finally it's becoming increasingly regulated incumbents and governments are trying to jam a bunch of um rules and regulations into the works and make it difficult for new players to come into the market and uh innovate so what is the future impact of this AI is going to be inserted into every technology product in the world over the next 10 years and with that with this centralization backdrop that I just described censorship and manipulation will take place for sure and llms and and other Foundation models are creating these highly engaging new forms of of um you you could call it propaganda or or social control in some ways it's already being used in the 2024 us election media and different sort of governmental bodies and so the people that control the capital distribution and access to models really control a lot of social influence with this technology so the open and transp arent governance of AI models is key and very very important a good analogy to describe what's happening today with AI is very similar to what happened with cryptography in the 9s in the 9s there was almost uh a rule placed that the export of cryptography would be treated as a munition you know as effectively as a weapon and obviously we you know now know that's sort of a free speech thing and I think we're seeing very similar parallels here with AI and the government trying to say you know we should control it or people that we pick should control models instead of open and transparent governance so this is where AI crypto comes to the rescue crypto has given us these great tools for self- sovereignty privacy and computational integrity that bridge these trust assumptions around different modes of compute we get great coordination and incentive engineering and we also get open infrastructure that catalyzes and and sort of uses and kind of promotes this credible neutrality crypto I takes this high level centralization and turns it into a into a decentralized one we've decentralized money and art and IP and storage and and why not AI it's a it's a massive Innovation and and it should be decentralized and deserves to be so this is where ritual comes in ritual is really aiming to be the convergence point of crypto and AI the task that we're trying to solve is we want to make AI highly accessible for a crypto native audience we want to make it very easy for crypto developers to utilize AI in their applications we view Ai and crypto having these these sort of bilaterally beneficial relationship that AI can help crypto and crypto can help Ai and I'll go into that a little bit more um in the next couple slides so we want to create a whole new class of AI enabled daps and use this cryptographic financial and sort of coordination print Primitives from crypto in the AI supply chain so ritual is being built out in two phases the first phase is called infernet infernet is a decentralized Oracle Network that is optimized for AI it's now live and we built an application called friend rug which I will uh show what that is in a in a second and you can actually set up a node today the idea with this is we have an offchain compute Network that any blockchain can tap into and call inference fine-tuning quantization models in their application directly the second phase is called ritual chain it's a sovereign chain that's evm compatible it has a a custom VM that is purpose-built for AI native operations that will enable a whole new class of crypto daps to use it this is just a quick kind of overview of the the kind of setup we have today where we can plug into different different environments different layer ones and layer twos and it's sort of all powered by this this AI setup that we have with these AI enabled um um pre-compiled so what's on ritual today ritual in rituals infernet already supports taking AI models and putting them into applications with the additional option to add in Integrity proofs uh of various forms ZK or or optimistic so any evm compatible chain can use the ritual SDK to coordinate this work ritual chain the next iteration of this is a set of stateful pre- compiles that make AI operations natively accessible to any developer with just a few lines of code what this means is AI models are a first class citizen in in this network and really the the AI model is the ad of this network and you can iterate and do operations on on top of that we have a proof Marketplace that allows users to express their preferences for different guarantees around privacy say for a defi use case you may really want strong integrity and you'll want to use a ZK solution maybe if you're building a um you know a social application maybe you don't care about integrity and you want to go with something weaker so we want the option for users to be able to pick whatever type of Integrity that they would like through this proof Market maret place and then we have a modular stack that allows us to plug into any decentralized model storage layer and use interrup solutions to plug into into various different chains so this is you know not just an ethereum thing although it's sort of starting on it and you can develop natively on the ritual chain or on your own chain we're not picky about where the apps are built we're really just excited to have this technology used everywhere so what are the pillars that we're trying to go for here one is is censorship resistance and decentralization can everyone in the world access these models can developers build products where they aren't worried about losing access to models privacy and integrity do you have trust that the inputs models and outputs are staying private and you know can you make sure that the output generated by a model wasn't tampered with during the whole process and this is really where the cryptography of this this whole network comes in then finally the the blockchain sort of incentives part is you can really now incentivize the entire ml back and offer these these sort of token based rewards for contributing to a model or doing governance on a model or on the labeling side of things or on the compute side of things so there's really a lot of ways to program new incentives for new applications at various levels of the stack and uh we think end users should be should be rewarded for contributing to these networks it's something you don't get in in web 2 AI today so a question a lot of people ask is what are the applications of of AI crypto and and we think there's a ton of really interesting ones and I'll go through a couple that that we're thinking about pretty actively one is defi with models in the loop so going and instead of using human governance and having to go convince a couple of people off chain to vote Yes and having them to vote on every single decision that the Dow does you can now instead do automatic parameterization via you know via defi model that's pretty powerful you can do it for for Lending you can do it for yield generation next is around gaming and entertainment so there's a ton of interesting live projects that are using onchain incentives to build Creator economies around a NPCs and owning AI models that you know play onchain games and win rewards and things of that nature so it's really this highly sort of generative thing that you can build with these different types of AI models content in media this one is pretty straightforward being able to autonomously generate new types of Works whether it's video audio text and it allows really anyone to become a creative and to compare that with the benefits we get of of things like nfts being able to create subscriptions and ownership it's a really really powerful primitive to be able to combine together then finally is governance so automatically generating proposals natural language interfaces and and a whole bunch more so these are just a few but we think there's a lot of really exciting things going on in this space so what is crypto an abled AI crypto enabled AI gives you a couple of interesting properties you get these cryptographic and game theoretic Primitives for AI plug andplay of different types of proofs and giving various sort of guarantees data provenance so creating incentives around rewarding contributors for data for fine-tuning and training and and Rag and incentives enabling users to contribute and be rewarded for it in addition to you know allowing other users with capital to crowdfund models which we think is a really big exciting thing and then finally is governance model preferences collectively curating prompts and a whole bunch more so this is a quick case study um that that you guys should all check out what we did is we created an onchain smart agent this is a few months ago that buys and sells friend Tech keys so instead of it being a human behind the friend Tech bot or behind the friend Tech account um it is a bot it's an llm that's running on top of infnet and what you're trying to do is you're trying to convince the llm to buy or sell your shares so it's kind of a fun almost like a word puzzle that you can go and and try and it's fully open source it's built on infnet it also uses Ezekiel proofs to you know with a classifier to basically able to prove the Integrity of the the action and uh we released a website called frog.com that you can check out more about the process and sort of the build out of this so this is one of the first things we've built on on infernet and many many more things will be come coming out over the next few weeks both by us and partners but also through the community in our Discord so we have a bunch of open research directions and we're really excited to collaborate with people open model governance and ownership proofs of compute privacy improvements open and transparent Foundation models incentives and gain Theory optimal compute routing data retrieval and provenance autonomous agents and and a whole bunch of others I could have you know three slides on on stuff that you know we're excited about and want to want to collaborate on so how can I participate so an easy way is to run an infinet node if you go to our website it's fairly easy you don't need a GPU as many tasks that you you don't need one for you can build on our Network and we're we have a Discord we just announced a few days ago so you can go there ask questions the developers on our team will go and and answer them for you and and help you if you if you need help with building uh daps that are bu being built on ritual several of them will be announced over the next couple of weeks and you can actually start using AI on chain which is it hasn't really been the case so far and more ways that are coming out soon so that is it thank you guys and um more information if you want to contact us or or learn more about what we're [Applause] doing