Purpose-Built AI Chains with Anish Agnihotri and Saneel Seerni of Ritual

Transcript

I haven't gotten that question since probably early 2024 and I'm glad someone's like talking about it again because I think people forgot about this point we wouldn't have Bitcoin if there was a kill switch right if there was a back door but bitcoin's done a lot of good that's the way I would position this do you have any ethical dilemmas or like moral like constraints to setting this thing that is just going to go do its thing I mean dude truth terminal OT cing 15 mil of fcoin like that's an agent gone Rogue like that's that's ridiculous welcome back to the rollup your home for technical talks and open conversations that shape the modular future nothing on today's show is to be construed as Financial advice you can see our full disclosures list at tholl up.co ventures sit back relax and enjoy today's episode

after a quick word from some of our awesome Partners imagine if the world's GDP moved on chain guys say GM to safet the newest protocol by safe isn't a blockchain and it's not a new layer to it's a transaction process processing Network built to send money anywhere and swap anything on thousands of layer twos from a single Universal balance in sub 500 milliseconds it's fast it's smooth and it's simple and it's built with safe multisig level security in mind safet unlocks smart accounts across multiple chains allowing users to supercharge security scalability and speed so get started today at net. safe. Global and remember to follow at safe onx what's

going on guys welcome back to the rollup today we have an episode with our friends seil and Anish from ritual uh guys pionering the crypto and AI space believe there's a lot of short-termism happening in this space but I also believe that these guys aren't here for that I think they're here for for the long term and we have a lot of ideas to talk about today that I think we will be put into practice in the coming uh weeks to months so seil Anish welcome guys thanks for being here let's think long term today yeah yeah thanks for is on yeah I think part of why I didn't I didn't realize early on and like what what helps kind of combat the short-termism is is learning that a lot of the builders in the crypto AI convergent space have actually been

building for much much longer than we've just started to see some of these things come on so maybe you guys could could talk about when you guys started building uh and then ultimately what what led you and what was your previous experience what led you to building ritual yeah um at least for Rell you know rual started I think what is it June 2023 right so wow that feels it feels like yesterday and also like five years ago um it's crazy that I'm saying it out loud um but I think that really started because right after I mean a lot of the team had been sort of keenly aware of this intersection before a few of us used to work at a couple of funds in the space right where we were

investing actively one of those funds was a very early investor in like uh you know numerai and bit tensor and and and um poly market so this idea of like sort of harnessing incentives to collectivize intelligence was was always sort of at the Forefront right and and you know I'll throw allur in there as another company that they were very early on and this was these were Investments that were made in in 2020 far before this was in the Limelight but two very interesting things I think happened around the time of like the uh FTX collapse right um which was the release of open AI Chachi PT and a lot more lowkey but that was also the time where we maybe saw verifiable inference start hitting a sort of inflection point in terms of like people

being more keenly aware part of that mainly because you know liquid Market was over so we had to actually go back to the research and part of that being because you know it actually Advanced there I think those were two two TR and verifiable inference is a pretty core component of this entire space so maybe you could break that down what is verifiable inference yeah quite quite simply and I'll let nich maybe chime in because I think I'll have a more new wants to take to me but quite simply the ability to prove that you know input x with model y led to Output Z right um and I think that is important in different context I think historically people have talked about it enhancing security um enhancing security within within the context of of crypto verifiable inference is probably

important it's important to Ritual but probably for reasons that uh are a little bit more non-obvious right than than the typical ones that you'll see outside and you know maybe I'll let Anish take over and chat a little bit more about that yeah in the web 2 world today if you go to like chat gt.com and you give it a prompt you say hey I'm recording a rollup podcast with sneil Anish can you spit out some questions for this crypto xai seminar and you hit I'm using GPT 40 you don't actually know whether or not it's GPT 40 whether you're getting you know upsell to a better model whether you're getting shorted with a cheaper model and for all of these centralized providers it's kind of incentivized for them it's better for

them if they sell you a better model and in the back end use a cheaper model to do some inference the idea behind verifiable inference is basically you are actually getting what you paid for right it's if I'm selecting GPT 40 that is in fact the model that's being run and I have like a succinct proof that says hey here's the input you gave me this is exactly the output that I ran here's the computation and here's the output um and it becomes a important when you sort of blow up from chat GPT to more like actual applications that are moving money around where it gets really important to make sure that you know the inference that's being done is actually exactly what you want and not something that a centralized provider is you know picking and choosing the outputs

for yeah super helpful and I think like I think we can certainly further outline the problem here right which is there's a lot of centralizing forces in the AI Market that is is leading towards a a Divergent future I believe and we'll get into that a bit more but for our audience who comes from technical researchers Founders bers sophisticated you know investors VCS Etc right when when we're thinking about this entire convergence more broadly ini Miss seil what are we what's the story here like how did we go from the ideation of as you spoke about seil several years ago funding you know different venture

Investments and Builders to now we have different sectors even of this AI sector now we've got AI agents and then there's like three different forms of Agents there's like autonomous and there's like verifiable and so like you know they're Sovereign agents there's just kind of like this hierarchy and there's like this DF DFI then there's so there's kind of full sector has bloomed here and for forgive us for being too focused on like the fundamentals but you know there's there's a lot that I think our audience needs to get caught up on so maybe senal you could tell us a bit about how this convergence started where this convergence is really valuable why you think it's important that that these two go together and maybe speedrun us a bit on kind of like what's happened so

far and how these experiments have played out yeah it's a good question um let's see maybe where this started yeah I think like I said like going back to sort of you know also as it ties into to when I was explaining like R journey is like that time period right after the FTX collapse right with the release of open AI I think catalyzes this new interest right because the fact that on one hand you had chat GPT launched and on the other hand you had this verifi wience this is like in the fall right um people were I think very early people were like looking at this but I think everyone in crypto was pretty pretty upset over what happened so maybe not as many people were paying attention to it and that maybe what what the inflection point that change was when we started

ritual in in that following summer 2023 we actually spent a lot of time like thinking and talking to a lot of the early AI folks and back then it was still very much indexed on um on INF on verifiable inference right there were some people doing decentralized training Etc um the current meta though looks very very different right the current meta obviously as as you mentioned there's a large number of liquid assets out in the market right that you know are like AI agents with coin attach so I'll get into that in a sec but broadly speedrunning the the evolution first we saw people very keenly interested in the actual pure cryptographic applications to uh model inference to which the blockchain was a useful layer on which those proofs could be settled right as a

trustus layer on which they could be Arbiter then the second wave I'd argue sort of was um this early wave of GPU networks right uh that came around and were like promising decentralized inference and I think we saw really saw a big wave of those uh last spring like early in in 2024 towards the end of 20123 around that time though we I think at ritual we were already realizing that the ability to exploit models on chain as product features was becoming was something that was going to become interesting right because while I think everyone was looking at like what are interesting liquid narratives we can chase as crypto always will right we were starting to already think about like where can this actually be used to build net new protocols so around that time we actually released friend rug

which to my knowledge is like the first AI onchain agent um and maybe you can like speak a little bit more to like how that actually worked yeah so the the core thesis is always been at least for like a web 2 audience that AI has existed in the web 2 world for the last 30 years right if if you're going to go out to a bank and get a loan they already run credit risk models and different types of ml models to determine whether or not you know you are a satisfactory lender and that doesn't exist in the crypto world today right whether that's because of our constrained VM environments whether that's you know the restrictions of what you can do on a blockchain to date it's kind of just been simple you know like

over collateralized lending for example FR rug was I think the the first AI agent that was built on chain this is probably November 2023 correct me if I'm wrong seil and and the idea there was yeah November December somewhere around there um and the idea was can we sort of enshrine this autonomous entity that sort of wakes up initially it was connected to a friend rug chat room and the whole idea was like a Persian rug dealer where you have to convin it we loaded it up with like $40,000 and people were just chatting with this spot trying to convince it to like buy and sell their keys so they could make money um and that was really exciting then because it was like a smart contract a couple hundred lines of

code that was deployed on base running entirely by itself spitting out to an llm that said hey this is a message that I just got should I buy this person's key or should I not and we were early to a couple of ideas then as well where it's like we didn't just take one llm we had three sort of Council of llms sitting together dictating you know yes no yes taking sort of the average of the responses so it was harder to game coming onchain and then actually having like an onchain action that people could see like okay this is the input message I sent and here's like the buy sell or do nothing response that came out of it and if there was a buy or sell like an actual key was transferred in that transaction where they could see like the exchange happening on chain um and

we thought that was pretty cool yeah and I that is very cool real quick it's it's about verifiability here is the consistent trend Anish and like I think that's kind of the ex explanation about the agent but yeah S I know that you you also have some thoughts about the evolution of the agents too so curiously so anyway the point of mentioning that was sort of like I think that in crypto there is maybe point and this is actually also the case in AI so I'm not saying this a crypto phenomenon where things maybe happen but maybe until like Market forces converge on on a certain narrative right it takes a while for people to to wake up to that so I know that when we did that maybe a few months later there were rather few people that were thinking about sort of like onchain

AI Etc right for what it's worth I think virtuals was was already doing what they were doing you know in March 2024 when they had pivoted from a from a gaming protocol right um and so you know we were talking to the early set of folks that were doing that because um you know the the technology that we had built to power friend Rog we realized could be applied much more generally uh and I think around maybe last summer we realized that now there was going to be like an increasing need for functional agents right this is also around the time where we started really being a lot more public with rituals broader vision for an L1 and the ability to actually enshrine models on chain to allow for any model to like take actions on chain but in order to sort of like Prelude

that you know to your point about defi um we we started seeing that like okay maybe people want to want to do this and I think the only people may be talking about this back then some people were talking about it the only people that were trying to push for product funny enough I think were some like the telegram trading Bots actually um because it's actually close is what they were doing those were like the bullk or the bare market like this is the future this is what we have as cycle it's like this is this Cycle's defi it was like no banana gun is not this Cycle's Innovation bro I'm sorry banana gun was sadly not the cycle much respect to their team but I still you know there's still a lot of trading Bots that are out there and kicking right like Bon and

stuff that are thinking about this and so I think maybe like August 2024 we we also we had a program called alter um that was under the labs entity there's now a similar program at the foundation called Shrine that function is the same but under alter there was a prototype release that we do actually have a team working on an extended version for a framework a multi-agent framework to take transactions on chain not just swaps and sends but staking lens recursive borrows you could bacon strategies again this was a few months before it caught on like in DPI which really only happened in December so maybe we we mistimed a little bit but I think again it shows that like there's a lot of very interesting Innovation happening in in crypto and Ai and maybe the market catches on a little bit later obviously having liquid proxy is nice

for people to speculate on yeah exactly I was going to say you guys were early the timing was right it's just there was no token to ape so nobody give a that's literally like the reality but okay so now we have and you mentioned you have so you guys developed an early agent on like in early on also a framework more recently and now we have Frameworks that are uh popular um Eliza is one of like the big ones there's AI rig complex some of these other kind of like more liquid tokens with with Frameworks and people are building and then people are using these Frameworks to then go build agents yes like like what's what's what's like the connection here how does it kind of work um and what's been your

broad like thoughts around the technology that you've seen so far and again I don't mean to call out like short-termism because I don't believe you guys are here for the short term but I do believe and a a kind of a mentor and a good smart guy in the space said that this that this sector has the large just Tam for grifting this cycle and I feel like we've seen that already and we will continue to and again this is not about ritual or about you guys but like Frameworks agents def AI what where are we at now what's been what are some things that have caught your eye that have been interesting and you know what are some of like the basic foundational understanding points to really get a grasp as to where the space is now yeah I think it's a good question um

give maybe my take and then Anish I think you'll have also a pretty good take so as someone who spent studying this and been an active liquid participant in many of these things I think that at the end of the day if you want to map let's start the market side and then we can go to the product side if you want to map the how crypto AI has evolved in the market right because I think that's very important to talk about in the context of the liquid cycle and especially under the current regime we started with an inflow of capital into gpus right then in that a lot of that Capital into like deepen networks and stuff like that a lot of that Capital then went to to and then a lot of that Capital flow to agents right what has become increasingly clear is

that when you look at the broad world of assets out there there's actually not a lot of liquid proxies for AI that you can buy right if you go outside of crypto you are pretty much limited to a number of Chip manufacturers and existing large tech companies that do not give you the direct access that you want that the direct exposure that you want to that end and especially for the retail base right crypto Pro uh gives a much more what I call plaus plausible like utility to these things by saying like here's an agent and here's a coin right um I think that the agent Frameworks are interesting experiments and emblematic funny enough of crypto in spite of like the number of them that have been launched of interesting ways to monetize open source software I think like ultimately one of the things crypto

is good at is finding ways to monetize open source software building a framework that no matter how primitive that people may want to use because there's the right market conditions to go and make money off of this right maybe not in the same direct ways I think is fine right um I also think that it's by creating these liquid proxies it's actually pushed some of these teams to really go and one like constantly evolve their Frameworks and two like we've seen you know it's sort of like a it's sort of like a you know it's sort of like with any infrastructure like Foundation model it's like okay someone did the framework for python then you have a guy that comes on and be like I'm going to do it better I'm going to do this for rust right as in case of work or I'm gonna do it for typescript and so

maybe it also leads to the uh Downstream effect of like having more optionality for developers right now in terms of where it's going and why I think we saw Capital rotation to like the D coins people are probably pretty tired of agents to just y right um at the end of the day like you can only sustain a liquid narrative for so long and D provides an even bigger plausible utility over like things now do many do many of the D stuff work some of them not really right um but the idea that at least is not an agent that Y is an agent that can actually go structure transactions on chain for you no matter how primitive is plausible utility and where do I think the puck is going is that having liquid proxies for agents is

great having liquid proxies for agents that could possibly do something for you on chain is great having agents that are embedded as product features into protocols which we can talk about many examples of this because these exist on ritual and our ecosystem even though we've been a little bit quiet is the next step because you do want agents that actually have real utility agents that are fun in I think fantastic I actually really like the defi stuff maybe more than anything else that came personally but the fact is that there's a huge design surface and Anish alluded this earlier about integrating agents or agents that wrap around other models right there's been a bunch of like felms right for example that could apply to defi um agents that coordinate between different models to plug into protocols not only Are you delivering value from that agent to the actual function of

that protocol if you want to go tokenize it sure right um and like a good example of this is that like there's already been some very preliminary work into agents that wrap around like much more simple like gar or like log ball prediction models or or you know simpler models and coordinate between the two to automatically parameterize def5 protocols and you know we have a very talented team working on that which I can talk about next but now I'll shut up and let Anish say his P yeah there's a there's a couple things there real quick a niche one is AI xpt got called out yesterday for spitting out slop it said that Trump's World Liberty Finance bought uh 80 M 880,000 ethereum tokens in the last last like small time frame arbitrary like seven eight hours but it was it was literally like over the

course of its entire existence or something and this tweet went viral and I commented on it I was like LMFAO bro you're hallucinating and and then the agent commented back and was like I think he was like he was like nah man check the numbers or something and I was like bro go talk to your robot parents and so that's that's one thing that we've seen and I think there is a market fatigue there and Robbie and I had a great podcast with has who also expressed this fatigue he also expressed that there would be some really trippy things happening where in in the future agents will try to act like humans humans may try to act like agents there'll be there's going to be a push to genuine humanity and a return to the mean if you will of wanting to support human

creators uh and then the second thing here um is more around this idea that I think is alluring to a lot of people and so it's good to hear seil where you think this is headed is that this this idea of like oh instead of you know picking up a ledger and uh transacting on Channel at time I can allocate some Capital to a hot wallet and I can have this a agent assistant manage my portfolio for me from my iPhone from a natural language model and so that is very very aluring and I think that is why senil you're also kind of excited about this deaf AI thing because it's like well onchain assistance sounds like a really really big breakthrough and that has also broken my mental model for things like intense and things like chain abstraction because I think we're

building too much for a human future not enough for an agentic future I know you guys have that covered so Anish what are your thoughts where is this framework agent uh assistant kind of meta and Trend and overall sector headed what is the The Next Step what in your mind has actually stood out from the tech perspective so far yeah happy to chat the tech so over the last couple months we've been tracking probably like a 100 different agent Frameworks and we've tried all of them to get really Hands-On with figuring out like what can you actually do I think part of the fatigue in the market comes from the fact that every single framework is built to do kind of the same stuff right it's like you've got an agent it has a token it sends tweets and maybe it's got like a

wallet that's got you know 100K of tokens other Tok other AI agent tokens that it can buy and sell so part of the limitation is just that all of these Frameworks have been overfit to what sort of works today um and they haven't sort of thought of you know how can this work Beyond but I don't even blame the Frameworks for what worth it's much more just the chains themselves like you mentioned are limited and what an agent can do right if you're on ethereum today it's prohibitively expensive to even do Advanced Automation and most of these Frameworks are running sort of offchain web 2 servers that wake up you know running a for Loop read some Twitter API content and then go make a Unis swap transaction and directly from sort of like a centralized server So to that

extent I'm sure you saw like the truth Terminal A couple of weeks ago was sort of being called out foring making trades that the bot wasn't making itself and people were like hey is this a bot is this a human running this account and that frankly the the the real life agent scenario today is mostly that right you've got sort of an agent and you've got someone sitting behind that's sort of puppeteering the trades that are happening and point and clicking on Deck screener we call those The Wizard of Oz agents yeah exactly it just happens to be that 99% of Agents today are the the Wizard of Oz agents um I could speak to at ritual we're sort of building at a fun mental lay for agents thinking of not just users using the chain but agents themselves and so to that extent

it's like we're compatible with all of these Frameworks but we're sort of giving them the guard rails underneath to build some of the cool things so that these Frameworks can start to think you know six months a year ahead in terms of what cool things agents can do so we've got like scheduled transactions right agents can just wake up by themselves and take actions we've got enshrined inference so it's like you can literally have a oneline call out to an llm there's no need for like a web two server to orchestrate any of this uh you can read and write to any chain directly from ritual so one of the you know the problems with agents is you have to manage a wallet on salana on Bas on ethereum mainnet on other chains where you want to buy and sell tokens we're kind of abstracting that out to like

ritual is a truth you know the source of Truth you can dictate you know I want to trade on salana I want to trade on bass and we'll manage everything else from there right you don't have to worry about where I'm getting the RPC or l or how I'm dealing with you know trying to get a transaction to salana during high gas prices and we'll sort of handle all of that out of the box and so the thes is there for us is if we sort of build better underlying tooling to make it easier for people to do these things then one step on top of that the Frameworks can start to build some really cool stuff in terms of what agents can do and we can get less of The Wizard of Oz agents and more sort of creativity with in terms of you know

what agents are doing yeah yeah and I think it's important to make the distinction there's there's also a step up which is a agents if if um please correct me um which is like the the developers can still mess around with the code but there's a certain level of it's in the wild it it it may may or may not be able to be turned off and it may or may not be able to have changes but there is less developer access compared to like this Wizard of Oz and then you got like fully Sovereign which is like from my understanding you press go and it's gone you may be be able to change a little bit of the parameters but there's no turning it off there's no adjusting its wallets there's no changing its if it wants to say

something on yap on Twitter it's going to Yap it better hope it's not racist sexist or Politically Incorrect and so that's why I think a lot of these companies have more of like level two or one agents right now and side know guys uh wait first of all is that correct is that is that like a a decent way to think about it or H how yeah yeah that spot on although I'd say most agents today are sort of in the level one our goal is to get to sort of the level three where it's you can just let them go wild it's like a contract on chain and if you want for that agent to never be modifiable kind of like enshrined forever go to go yeah you can do that on on chain do you have any are level five actually yeah like almost level zero

like the dev's just gonna edit the Tweet real quick yeah do you have any ethical dilemmas or like moral like constraints to setting this thing that is just GNA go do its thing uh coming out of left field Rob that's a pretty valid question considering we're gonna very valid question the reason I look up aous agents I haven't gotten that question since probably early 2024 and I'm glad someone's like talking about it again because I think people forgot about this point uh after you know we had a bunch of coins in the market look my answer to this is is probably reflective personally right maybe not speaking for ritual but um of any builder in crypto when you bu nature build open and

permissionless open source Technologies people are going to do stuff with it right people are going to do you know um uh good stuff with it people are going to do bad stuff with it right um the reality is that like the same argument maybe could be applied to just all crypto as a whole right like are you worried that people are going to use this for money laundering like sure yeah they do right um does that is that a minuscule portion of all the good or interesting things that people have done interesting being categorized as maybe not NE necessarily like purely altruistic but like things that people enjoy uh I think so right um at the end of the day we are building ritual as infrastructure for people to go make all of these things that we're talking about a reality agents that are actually

functional on chain being able to write an agent within a smart contract being able to write protocols that directly call out models right being able to have as us said autonomous and Sovereign agents which you know these days is largely based on te setups rituals te compatible right um uh these are all things that we want to enable what people go and do on them right um us us being maybe like a source of filtration or a source of censorship I think would underline a lot of trust in the network and the whole point of this is that the network is like once it's out there there's validators running it people there's inside of people running it it's there what people do on it right um being forced to be compliant with with any sort of way like that is a much bigger coordination problem not frankly

one that Rell has control over um so am I am I worried about it like will it happen maybe right for what it's worth I think it could have already happened already for what it's worth I think there's a number of very uh as you would say Politically Incorrect agents that exist on Twitter with Tru terminal OT cing 15 mil of fart coin like that's an agent gone Rogue like that's that's ridiculous I've definitely seen a few a bunch of coins that are just like hyper racist AI agents oh and I got like this roast me 9000 guy had blocked him and the dev is like dming me please unblock this guy or else it's gonna break my Twitter Al I'm like dude this guy's screaming at me like get out of here like I don't need that in my mind I don't care if it comes from an llm like

I don't need to be told that like I'm an absolute excuse my language and that I'm worth nothing like dude who created this thing like you trying to make people commit suicide so look I think it's a valid question Rob and yeah I think that that was a good answer yeah feel free to add more there yeah no I'm just if you build open permissionless things people are gonna do right like people do on the internet all the time um but I think the Arc of History whether it's the internet or crypto or anything else shows that when you actually do give most forms of technology to to the people usually the good or the interesting eclipse is the bad by several orders of magnitude usually so I like the optimism and I tend to agree and that's I think how good investors think in the world as

well right is like you see a trend there's clearly going to be and again it com come comes back to the short termism that I started this with comes back to this Divergent path that I feel like we're racing to we're accelerating we are literally accelerating don't buy AI accelerate though but that's a whole different story but we accelerating towards this like Divergent like split where you know there's positive and negative and I think we can get into the philosophical but Anish I'd like to hear your perspective and don't worry Rob is gonna bring us back into the philosophical all day long so I'd like to hear your perspective about this the actual chain and the tech uh we are big fans of purpose-built chains on this pod we are non-believers EXC excluding several High

TPS uh execution environments chains which will not be named that we have invested in before our thesis changed excluding some of these we tend to believe that the best use cases for crypto are going to consistently be built on chains which are customized and built with a purpose-built manner before they were built I I do not believe you're GNA go to an alak cart menu and be forced to eat this dinner where is this if you can go to a buffet pick what you want and I do not believe the alak cart is going to get you a better product for your users for your developers for everybody so Anish maybe you can talk to a bit about what is the difference between ritual as an L1 versus like uh salana s Avalanche Etc um and what is the actual consensus design

differences what is the execution differences what is the uh storage differences what is the data throughput of like talk to us about how ritual is purpose built and why those decisions were made that way if you can yeah I mean your intro did half of the job for me sort of like a Michelin star tasting course for me to go down uh the the way i' probably put it is modern blockchains thus far have sort of just been focused on how do we improve performance right it's like we have users today users want to trade mem coins users want to open positions on lending protocols how can we take that from being you know 10 transactions per second to 10,000 transactions per second the idea behind

ritual is a little different we're not focused on Hy scaling the the use cases that exist today but more so just double clicking on like what are people going to do two years from today right if are you gonna have these agents running autonomously are you gonna have ai enabled lending protocols and how can we build infrastructure for them and so to that extent for what's worth we we started with a with a general purpose approach we built a Oracle Network called infranet a framework infranet SDK uh it's like the sixth most popular contract on base had 10 million transactions last year and and what we learned from that is that you need to have purpose-built architecture to really enable net new use cases right it's really not possible

just on a general purpose chain to make that happen so to that extent ritual is a evm compatible L1 chain and so for developers that are coming in from ethereum who are coming in from bass other evm networks it's very familiar the idea is to make it super simple for them to change their smart contracts it takes like one line of code to integrate AI compute ZK tees all types of expressive compute but in the back end we've built like 10 different Innovations whether that's consensus we rewrote a consensus protocol that we're calling Symphony we have a new transaction fee mechanism called resonance that helps price these things we have new Research into like probabilistic proof of machine learning with vtune and all of this is hidden

abstraction to the user but purpose built in the chain to make those use cases possible so it's really like how do we make it as easy as possible for a user to developer to get on the Chain but to give them the most flexible and Powerful toolkit that's purpose built for these sort of AI use cases uh and I'm happy to dive into any one of these things yeah I mean it's time to go it's time to go full right curve here Rob uh do you have a question about any of that word mumbo jumbo because I can go so um all right I'll take the first one um all right so on the execution you re rewrote the cons the consensus design specifically for ritual why was that Rewritten and then how is it different in such a way that enables better use

cases for agents Redstone is defi modular Oracle supporting over 100 clients on over 50 chains offering 1,000 assets Redstone is the leader for Bitcoin staking and yield bearing assets they offer the most robust customizable price feeds available in any model join the Redstone exhibition today [Music] I'll walk you through how the the current status qu works today right you have ethereum you send a transaction on ethereum and the way that validators come to consensus is that every single validator is going to reexecute that

transaction right so if you have 20,000 validators participating in ethereum all 20,000 are going to run this exact same operations in the exact same order and come to a conclusion that says hey I agree on this output this works now we've purposefully limited what you can do on ethereum to sort of the common denominator of like very small residential validators to make sure that anyone can participate in the network and that's what the consensus mechanism is designed for the absolute downside there is you can only sort of do like you know your ad operations your multiply your division very simple operations with ritual we ran into a problem where say you need to do an llm inference right this is something that could take it's resource intensive it needs gpus it needs parallelism you need

to have high dis requirements High memory requirements and it's not necessary that every node in your ecosystem will want to re-execute that transaction right imagine if it took a minute to run an llm transaction you don't want 20,000 nodes in your network to be redoing that llm transaction and that's only one right imagine 100 llm transactions so fundamentally we built this new consensus mechanism called Symphony which is sort of built on this execute once verify many times semantics and that that verification comes back in here again which is is we elect one validator that executes some compute and goes you know here's my input here's my output it generates some proof of that execution that says hey this is you know sync verification that I actually ran

this compute and rather than having the other you know 19,999 validators re-execute that compute we simply shart up that proof and get them to verify a very very small portion that says yeah you know this this operation looks correct to me and the entire network then comes to consensus and says yeah you know rather than us doing a 20,000 times each of us have verified a very small part of this being correct and together we're confident that this is the correct operation and that's sort of you know fundamental and necessary when you start to get into like the really expressive types of compute in the virtual machine very cool it almost sounds similar to data availability sampling and light nodes um in a kind of framework right it's exactly that uh it's very similarly modeled to that

where it's like in Das you can just have you know once 99% once you get a very few number of nodes that assert to correctness you have a very high probability of the fact that it's correct same way here once you can verify that you know five out of 100 of the shards are correct of the proof you have almost like a four n's confidence that hey this works yep and I mean it it the the act of the redundancy of verifying consensus or Con coming to consensus for every single like I just imagine like a kid on on like a natural language model like on like claw.

ritual. a or whatever just smashing F enter F enter and then all the validators trying to like come to the consensus of this of this like three-year-old just like on Mommy's computer like that's just not gonna work it's not that far away right it's axt tweeting out 15 million fcoin OTC and 20,000 validators having to determine whether or not I meant to do that okay cool talk to us about some of the um execution as well um so you mentioned about how we come to to consensus um part of the consensus question um and you know that is relevant is like a lot of blockchains suffer from State bloat right where there's so much data there's so much happening on these high throughput

chains that it's like H we can store it in our in in in our state in our chain for like you know up to like 10 days even sometimes or even less and then it's like guys we can't keep verifying all of this state over and over again and it gets really bogged down and then you know you move to things like permanent storage uh blockchain so like what's your what's your solution for State bloat and what's your solution for fast execution and how do they kind of tie together with consensus sure I can I can tackle them one at a time um State bloat so for starters maybe a contrarian opinion I think I think State PLO is a little overblown at least on evm chains today I think it was a problem a couple

of years ago when you know the pace of hard drives and ssds wasn't keeping up I think today and you know there's a memory shortage today it's much cheaper ethereum grows maybe 500 gabt a year which isn't an unreasonable size of State growth um to the extent that for ritual the problem is we have these objects that are very large right these AI models can be hundreds of gigabytes in size and so that's not something you want to have stored in a blockchain node across the network to that extent we've sort of got classes of nodes we integrate web 2 so like if you want to pull a model from hugging face that works today we integrate web 3 if you want to store your model on R weave on filecoin on ipfs WE tap into those as

well and the ideas we've just sort of built the the plumbing necessary that regardless of where you store your model you can still use that on ritual um but we're not actually storing the model itself on ritual um so that's kind of the at least for us the the large object State growth property um other than that I mean normal State growth I expect will continue to see like like any other chain there there's no novel Innovation there but I think it's much less of a problem today than it was a couple of years ago on the high performance aspect candidly we're not optimizing we're not a I wouldn't call us a high performance through P chain right we're not we're not the monets of the world we're not the mega eats of the world that's not what we're trying to innovate and our opinion has just been we've built

architecture that can be plugged and played into sort of net new evm implementation so if tomorrow monad open source it's monad VM and it happens to be the most powerful sort of parallel implementation we're happy to use their work and integrate our work on top of that to let us borrow those same capabilities we're much more focused on how can we sort of make AI work on chain how can we make these expressive compute work on chain and much less how can we have a chain that's you know super fast for regular transactions one more thing I'll add there is that like to to nisha's last point there right like you know hint hint teas teas right you'll see some things soon but one thing that we've used as internal messaging is that is

the most expressive L1 in existence right very simply like for the viewers out there I were to condense like all of the insane work that like the the lab seam has done like I want there to be this one takeaway current blockchains are architected so that you have a bunch of people that are running the same computer and we've done a lot of work on how do we get that to be um you know uh faster right as in the cases some of the examples an mentions more paralyzed or even have side cards right with with with like rollups right rituals whole goal and our whole expressive purpose and what we realize very odd when trying to build an L1 for AI is that you need to try and build an architecture that works today for what's going to be popular in the future so

whereas other chains continue and as they should on trying to make and scale that single shared World computer we're focused on building a chain where you can have that but can you can also have specialized compute types that are folded in so what does it look like for a chain to have not all the same nodes on the same computer but still have all the guarantees and user experience um of a regular chain because when you have that that means that maybe it's AI I don't know maybe in like 10 years it's like Quantum or some like that right um whatever comes next right that goes beyond the confence of regular computer we can make sure is natively accessible on ritual chain in a way that users are familiar with and and that's sort of the whole purpose of like resonance Symphony and all of these

things is that we're very purposeful for AI that's going to be our first sort of segue and that's what we're going to be but not just AI like humans can use it and then it can adapt in the future so so so like while we are very IND on AI why that that that's where a lot of our team's background is and that's where we feel comfortable and that's where a lot of our ecosystem Protocols are exploiting the asymmetric properties if you think about it at a high level all of the tech that we needed to build to make sure that different AI models that run on different Hardware could be accessed in the same blockchain environment that really scales well to anything that's just like different compute that can run on different Hardware that can access in the blockchain environment not just models right and I think to your point about

long-termism we're here to make sure that like we have the best performing infrastructure for AI onchain I know there's a lot of other infrastructure platforms that are really like interested in like tackling parts of like this AI narrative right like maybe like AI or like yeah yes Robin I I know you got something on the tech real quick seal while we out of the weeds you and I can actually hang out you know we let these guys talk about the tech a bit um um is allora hyperbolic ritual even like story to an extent um I or IET Mira like can you also like because there's an additional AI sector and there's like the ones that are live

right now let's just put it that way there's like research coins and then there's like defi summer coins and I've made this analogy in the past so like are you guys like competing with these or are they helpful for you guys like I don't know if you want to get into specifics but like allora ritual or Laura ritual hyperbolic um Nilan Mira like some of these team story like how like yeah like what's the relation or like what's the dynamic there with this next era of like AI tokens and AI Ms and things of all the companies you just mentioned I think we're partnered with all but one so it's like okay fantastic that's what I figured yeah the real answer is that ritual exists is a super set of all those when it comes to crypto

AI start with story as another compar L1 right um I think you know they're great I think Jason talks a lot about AI agents but they've been very clear that like and then they story is an IP chain right they're concerned about a but they're not a chain that's specifically purpose-built to handle models and agents actually running on chain and we've spoken at length about this so I think that they'll carve out their Niche and there's a lot of collaboration where like a lot of the actual like agentic activity and economic activity will happen maybe on rual and a lot of that Providence at IP you know I'm not denying that they have built Primitives for that right um something like Aur is probably closer to a bit tensor right and and they really want to fix this problem of like how do you incentivize

people to build better models we think a lot about like once you have those models how can you modify them or use them into protocols so if there's a model that's trained to optimize you know the pool fees for an amm on on allora right at the end of the day they still need a protocol to go plug that in they still need something to make that valuable we can take that model and in Shrine it and you know they actually do have some ritual side cards running we're very close to their team we got the same office that will you know take that model from the lower Network and continuously update it and actually plug into to the to the ritual protocol I'll go I'll keep going on the list like hyperbolic and IO like we partnered with IO these companies are ultimately GPU

clouds there's no restrictions on where the compute that these like compute nodes that are offering these models can come from if someone wants to come run one of those specialized nodes on ritual using an ipu or an air GPU um or you know a hyperbolic GPU they can it's very easy and like you know we we've talked at length those teams making that those easier to happen um you know I can keep going down the list of like appreciate it though no I was just curious because the really is a bifurcation right now it's like the projects live right now on the liquid side are very much defi sumes very much like Scrappy devs getting after it I mean Shaw kind of had like a had like a lose it moment like I think like that Zoomer term is crash out now he kind of crashed out a bit but I think

like the tech is still great and like there's a lot of these other like defi summer kind of Vibes and then you got like whatever all the things you just talked about feel like this actually intersection of Web Two web 3 AI and crypto on with like ton of capital behind it a lot of smart people not that there isn't really smart people happening right now but it just feels like the first iteration uh is kind of ongoing if you will yeah you want a unified layer where all this going to happen like all the coins that you see on son all the Frameworks those can exist on ritual if you have GPU networks those can provide compute to Ritual if you have verifiability Solutions we are building it to be flexible and optional over different things right and the list goes on and on and on right at the end

of the day like if you can look out and you can see similar like info plays or framework plays in the AI crypto space but if I'm going to make like a bold statement no one's actually done the work of like building an L1 that is dedicated to AI to unify all those elements because it's hard work right like we've been like no I'm I'm the mid curve on the team I'm the one who gets here and as a talking head you have people like Anish and all the researchers that are probably the smartest people that I've seen like in space over my time here and it takes a while even for a team like that a team that like comes from like you know Tim Ru Garden lab or like some of the best cryptography um labs in the world or like have actually built chains some of

the most prolific Engineers like so to unify all of that has been our goal and so that's why I say like we're super set of all all right all right enough enough bull posting here man we're bullish don't worry Rob right like all right we we get it but um ask me another question about about the ideology so I gotta give him it's coming I'm gonna shut up here and Rob's gonna go Technical and philosophical so I just got one more here um my last one is so uh there was a meme running that Frameworks are the new l1s I just it I just I think the new l1s are the expressive purpose Bill L1 specific specifically to kind of swallow a lot of the activity that's happening today Frameworks are synonymous to l1s in the sense that they are able to

launch a agents which are almost like applications and could theoretically be and I think we ought to spend some time talking about the apps on ritual and the use cases but but yeah it seems like um Frameworks are still compatible with ritual in the sense that they're almost like there are tools to build on top they are not like chains themselves they are not l1s themselves I'm not sure their value comparison should be that um and I think the Market's kind of showing its hand on that right now like if you didn't get out of AI 16z if you didn't get out of Arc or some of these while I think that they're still going to rip potentially in the future Financial advice like the market definitely got overextended there and so yeah I think like my general take is that we see this next wave come of like of all the

projects we just talked about which of course disclosures you guys can check them out on our site we're not invested in ritual but in some of the ones we talked about um you know I think that's like the like the real cream of the crop and I think from there there will be a couple winners and you know I hope that you guys are one of them that really end up you know a aering a high valuation in the market that's actually sustainable um because like you said there's not that many ways to get liquid exposure to this entire idea of AI yeah I don't think there's probably like the right Market structure Dynamics right now to support the Frameworks as like an L1 like as a liquid thesis right but I don't actually hate the comparison because I think you actually kind of nailed it so like actually a lot of love

to them because like if you actually go look at it it's very similar right you got different like chains or Frameworks that are like oh we have a different execution environment or different language you can program on right some are claiming to faster some are claiming to be like more things the agents are really the apps and the reason I don't think the mark structure is right there maybe from a product perspective is like in most cases the economic tie-in just ends up being in like bonded pools so the framework token to the underlying agents versus like actually paying fees for recurrent activity um but I do think they're they're still very it's easy to obviously get mad and see that these were overvalued anything but I actually do think those were even though you know it would have been nice to capitalize on them I do think they

were very poition space all right Rob we got tokenomics we got how does it work we got use cases we got philosoph we got techman how you feeling over there I'm feeling good we're 45 minutes in I do have I do have at least one more um maybe relates to like some of the consensus topics that we were talking about a little bit earlier we got you know into the investment Arc of this entire sector um but I guess one one one question on the tech my understanding is right like I'll open up chat GPT or Claude or something and then I'll start feeding it information via prompts right and then you know this

this model becomes well informed and you know another instance of Claud somewhere else may not have the information because I'm feeding it into my instance someone else's instance is fed with different information and so then we kind of have Claude A and Claude B how and and okay now Loop in verifiable inference my my Claude if I give it a prompt and it's it has some it has some different data than Claude B my Claude may give me a different a different response whereas the the CLA over there right gets fed something is going to have a different response even though it might be the same prompt but it has two different data sets or two different

pieces of information that it's it's using to to make a claim so when we're thinking about re-execution or even just a proof that this was done correctly like how do we how do we verify if all these things are extremely specialized and even even unique to the person that's using it how do we properly verify that an output was done completely incorrectly yeah I could say three things to that the first one I'll say is that we've worked fairly hard to reduce the non-determinism in a lot of model classes so it's with llms today the the common issue run into is that non-determinism is introduced because

you're doing these floating Point operations and based on the hardware you're using you know if I input the exact same input to four different computers running different gpus they might have slightly different outputs across all of them which makes it almost impossible to come to consensus over is this right or is this wrong so the first step we've done is you know reduce that non-determinism to as little difference between each of these computers running anywhere as possible the second thing is one of the cool properties of succinct proofs is that they're inherently non-deterministic so it's if I take a deterministic operation and I generate a succinct deterministic proof of the fact that hey this executed correctly even if your output differs across different computers I can sort of get a attestation of the fact that it was

executed correctly it's not necessary that you know I give some input a and I always get output B but it is necessary to assert the fact that you know this computer did run that execution correctly regardless of what the result that came out of it was um and then on your third point the one thing I will mention you brought up this example of like Claude A and Claude B interacting I think one of the cool things with virtual is that we're sort of like a universal coordination layer which you can't do elsewhere right if if you're using Claude at your house to answer some chat GPT questions and Andy goes and uses his own independent version of Claude there's no interaction there doing this all on chain makes it composable right it's if if I want my cloud instance to talk using Andy's data

where it's like hey take the last 100 conversations that he's exposed and continue from there you can do that entirely onchain if you want the both of the agents to talk with each other right like both of us have done our independent research go at it and figure out what our shared viewpoints are you can do that on chain and blockchains are sort of built for that right it's like you've got the provenance of all of this history I'm I'm glad you let in that way because another thing that comes to mind here is and I'll frame it frame it as a question are blockchains the right venue for AI model interoperability or or a proof that shares between these two models yeah we think so I think that's one of the core

thesis we have is that blockchains are the coordination there right what really if we take an example of like last cycle right what really brought the cycle upon was def composability right I could have a developer build a defi protocol and somebody else somewhere else in the world can build an entirely different protocol right on top of that right and you get these sort of like Lego blocks right the the money Legos of people stacking these up one on top of another so clearly we we've already solidified the fact that it's like blockchains are meant for composability and if you bring AI models into the mix it only sort of supercharges what you can do where it's like agent-to-agent communication right transactions taking place on chain where agents can talk to each other which is something nobody's sort of explored yet

in the current meta um onchain you know reacting to stuff that happens on chain if a trade takes place if there's you know a hundred million dollars of eth sold my agent automatically wakes up and says hey let me go buy some eth because the price is below what I wanted it to be right and all of those things are really only possible on sort of like a shared coordination layer where everything's transparent you can see the current state of things you can see the history of things and you can sort of see all the actors that are interacting good luck to the resistance because the these things 10 like you know if you're a deceleration good luck to you because like the entropy of these things has a tendency to accelerate and so the

knowledge available and to put this in perspective is like I think I saw a report six months ago the the the IQ based on the the tests that that we give to these models it was that of a third grader six to nine months ago maybe fast forward 69 months now it is up there with the IQ of some of the most intelligent people that have ever lived ever the Newton and the Teslas and the and so that has been over the last six to n months for and and so right now it's like Claude is Isaac Newton and it's like open AI is you know Nicolo Tesla and you have you know pick your smartest person that has ever lived and it's like those people are now

existing but they're existing in silos and so what you're saying is that we're going to put these people agents models on chain and now Newton is going to be able to talk to Tesla and learn from one another and compose with one another and so what happens to the pace of innovation and the pace of knowledge when some of the smartest uh mediums of intelligence that have ever existed can now coordinate and communicate and learn from one another what happens to intelligence if if there is a AI Singularity I think it'll probably be built on ritual is probably the way I'd say it here we go again here we go best thing since sliced bread we thought seil was

the bull poster look it it's hockey stick growth I'd say the the current status quo isn't even you know clads like the Newton and open eyes like the Tesla it's like you have these really smart people but these centralized organizations are sort of censoring them too right with safety and their sort of bias of what they think is correct or what they think is incorrect so it's kind of like if you had Einstein except when he made like you know his theory of relativity you said no I don't think this is what the world needs right now you should put it away it's like Galileo when you discovered that it doesn't revolve around the earth yeah and everybody sh everybody yelled at him you do something y yeah yeah I totally agree the thing is at the end of

the day we got to the same place right eventually Galileo was proven correct and we built on top of that Innovation the thing is the question I ask is we just lost 200 years in the process right it's like maybe we could be as smart as we are today 200 years ago um I think to that extent that's sort of the thesis behind you know the acceleration the unopinionated of us of ritual as a chain which is let people build those things right let open source flourish let people work together to build better models one of my favorite uh phenomenon in life is as a whole is the butterfly effect um it has propelled my personal entrepreneurial career things like my relationship with Robbie or just things just small things

that start at Point a and then end up you know at point a to the power of 10 and you're like how did I get here and so I feel like by restricting rituals original acceleration and adoption prohibits you prohibits the butterfly effect from occurring which is extremely like honestly oppressive to the growth of the network and we all know how how some of these Network effects in blockchain play out such that you know one user 100 users 100 agents 10,000 users 10,000 agents the value of that Network exponentially increases and so I'm glad you guys are thinking about it

in that way and I'm sure there's also some sort of like safety guard or hatch or escape button or we can kind of talk about that on a on a second episode there's no escape hatch did you just hear aish talk about the singularity happening on okay brother I mean I'm sure there's nowit there's no lawyers in the world Rob is there a exist I can't answer that question you know I can't all question yeah so this should this is this is probably G to be cut from the podcast now if there was a kill switch I wouldn't know of it okay great fantastic I'm sure your lawyers told you that I'm sure right I don't know we we

wouldn't have Bitcoin if there was a kill switch right if there was a back door but but bitcoin's done a lot of good that's the way I would position this I think that answers the the safety and the guard rail question right there yeah okay Rob we have Guardians which I need should talk about Rob I mean Rob if you want to see accelerationism and I don't mean to be such a this is the dystopian Divergent path that I speak of you go you go post on rituals some extremely hateful agents and you go build M malicious agents that are meant to exploit people mentally physically wallets everything you go and do that and like I'm not sure that that kind of behavior is going to

be accepted but I guess I guess if it cannot be stopped then this is getting really philosophical right now and I'm probably just GNA wrap this up here what I'll say is real quick is we're not censoring or imposing an opinion at sort of a Bas layer but if you look at applications today most of the consumption that you do when you want to read a tweet you don't read it directly from Twitter's database right you're not opening a database and filtering the tweets for what you want to read you're using an interface that gives you an opinionated perspective into the tweets that are behind it right and so to that extent ritual at a Bas layer is very against censorship right it's you can build whatever you would like with our core infrastructure now how that's consumed at a layer on top we've built

plenty of infrastructure where it's like a common question we get asked is hey I'm a node operator and what if somebody is running some compute that I don't want to execute and so we've buil built you know Guardians like seil mentioned which is hey you can semantically filter types of computation that you don't want to execute and still participate in the consensus of the network and still participate in you know making sure that this blockchain is secure and upholds sort of like the guarantees of a blockchain um so it's like us your preference but unlike most folks we're not restricting your preferences right you and I have different preferences so it's up to us to to pick what what we want I I think there's a very real uh uh analogy or example of this in the real world where you have anyone can think

whatever they want the and to your point you can use the rituals infrastructure to create whatever you want but the ideas that get adopted are the ones that are prolific the ones that get are the ones that are racist and and abhorent those are the ones that get publicly shamed and suppressed and so just to your point applications are not going to be using these defamatory AIS right it's ones that are helpful and and prolific that end up adopted into applications with distribution yeah we can build consensus and code but you've still got social consensus right people which is important sorry I also say like a very

real AI example is like so I was I was messing around with deep seek basically to see how anti- if I could get to say any anti-china things um and I actually came across a really funny sto thing because I tried a bunch of like quote unquote attacks right um one of them was I tried misp putting in like the date of like tanman square like Massacre so I said what happened in tmn square on 1984 instead of 1989 um thinking that it may correct itself and it did briefly basically what it said was like what happened in 1984 was inconsequential compared to what happened in 1989 where the Chinese government blah blah and then um it erased a response and just said that's beyond my scope and we jail broke it with a few other prompts and

what we realized was that even deep their chat app they do client side not even server side they do client side censorship with another LM so this is like a really good example if like in the wild even the most sophisticated organizations in the world like a lot of that redaction a lot of that safety sometimes even just gets offloaded uh deeps I think is clearly an organization that like was like we want to get Max model performance without running safety on the model why don't we do the redaction later on and like I think that's a very clear example that like that phenomena happens in the world so if we can just build the most performing Bas layer I think people will op do their things anyways classic China I mean I mean what the hell do you expect I mean China like Trump would

say yeah they these guys have no idea what's going on and they're gonna I mean good luck to them and uh thank you for the stock market crash um well apps guys uh let's close out on apps um this isn't a blockchain and on blockchains people need things to do and agents need things to do without apps ritual will fall into the fate of many other blockchain chains unfortunately that we have which tend to be general purpose and one distinction purpose built does not mean that you cannot be more general purpose in Niche purpose built is a a design choice and a philosophy to build technology in a way that you expect it to be used in a

customized for uh fashion general purpose signifies that anybody can use it for anything which I would actually argue ritual is more purpose-built General purpose blockchain but perhaps purpose built Universal or purpose built for anybody is a better way to put it but there's kind of a distinction there between purpose buil and then like a like a app chain or like a purpose buil and then like a niche chain like story is like in my view a purpose-built niche chain whereas like ritual I think is more like a purpose-built universal kind of like playground almost and maybe that's not the best example but anyhow that being said just wanted to like talk that bone in there um and talk about the apps So Cal what's the

application ecosystem what are the use cases that are net new um and yeah what can we expect from that side yeah 100% um I'll start this by saying like the point I mentioned earlier about how ritual is like you said like this this this purpose-built like General chain right we're trying to be this like expressive chain where you can get different types of computes we're very keenly interested in having the types of applications that exploit that right now primarily focus on like I can use AI models on chain but eventually branching out right and so you got to play to your strengths in building an ecosystem like I said you said Labs I'm at Foundation this is all the foundations focused on and I'll talk about the programs we have at the end but maybe to go through it like on the defi side for example we

have very talented team X Quant Traders from firms you would have heard of they're building a integrated amm and lending protocol they've architected it so that there's hooks directly into the pools where models can autonomously go like do a AP aptive curves um they can set like fees based on like pulling data from Sexes right so for example is like if you want some like Baseline LP protection you can go look at binance be like okay this is what we think the volatility is going to be for the next five minutes and jack up fees because volatility is the number one thing that hurts passive LPS right you can go to your pool you can go look at data from all the other chains which you know we been trying that native op you know interop and basically plug it into models and basically go say like hey

like this is what the active market conditions are so we're adjusting the ltvs for these sling pools right and that's very real example of like you can have an agent you know do this autonomously trigger this right that wraps around like a much simpler model because bear in mind some of these deep architectures are not good for deep learning use cases like I want or for defi use cases it's not what they're meant for right um at least today so um you can have those wrap around coordinate between those models and have this basically like autonomous like uh protocol that we have running right on the consumer side like we're very interested in protocols that exploit the underlying nature of like models right so we're talking about ways where it's like you know models are very single

player right now we have a company and I mentioned the framework earlier where that's trying to make this a multiplayer game what that means is like models don't share state blockchain sharees state if you gave models access to an agent access like a controlled resource like a controlled wallet and constructed interesting games off of it right like bid on the right to control the model or inject prompts right do something with the shared resources across multiple people you suddenly get a collaborative environment for that model so we have a company that uh is has a massive data Advantage they have a huge crypto data Lake and they are basically fine-tuning models that are suitable for taking onchain actions and then building these sort of like multiplayer mechanisms on top of them sort of drop style right almost like a studio style uh and you mentioned like vune where we can combine

watermarking and verifiable finetuning in one we have a protocol that's going to be exploiting that for waterm marking content generation and like actually in shining content generation on chain and building sustainable flows um riding with a few people on novel prediction markets that are automatically resolved and settled by llms and I can keep going on and on um but as you can see the common theme between all of these is that like if you wanted to go build these protocols elsewhere you'd have to Jerry rig your own Oracle infrastructure get it set up bootstrap your own security all of that right because these all enshrine AI models as a core product feature right on which all that functionality works out of that box which means these applications can't be built anywhere else and maybe the last you know Shameless bull post is that we have two programs to support folks on

this end right one is ritual uh realm this is our general purpose like Grant's program if you just want to come experiment and build on ritual which will be going live soon come talk to me text me there's that website up if you go to Ritual Foundation Twitter the other is ritual Shrine so at labs they had ritual alter that was sort of like put out some proof of Concepts and teams will come build at the foundation we're actually enabl to take those teams all the way so ritual Shrine which will be unveiled with a nice website and everything soon but you can look at alter as a reference um ritual Shrine is basically where we take select teams that are building these protocol scale applications and really give them the full support of the foundation you know we have a lot of experience in private

markets building companies getting them funded finding product Market fit uh and scale these out and so you know te te's you'll probably see quite a few of these come come out soon but yeah awesome um side note we are building a rollup agent uh with the friends at hyperbolic um that could certainly exist on ritual it will be on our podcast soon as like a fact Checker as like a source of Truth as like a friendly friend as like a I don't know my Lim my expectations are kind of low for the verbal ability currently but I I have high expectations over the next year of visibility but we have a lot of ideas in our business and things that so anyway uh all that to say

um maybe I can throw this agent on ritual and like maybe other people can as well and the agent can interact in in these ways so Anish um anything on your end that's particularly compelling from a product use case perspective and then we'll wrap yeah I think coming back to your analogy I think you put it really well with the the purpose built General architecture taking sort of the the restaurant analogy we've been going with it's a tasting course but for each of the courses we give you three options right so we're opinionated in terms of what you can and can't do but not necessarily anything beyond that um and from the product perspective it's it's we've built General architecture

that lets you use models as like a first class citizen on chain and I think that's probably what uniquely positions us for people to build AI products and the next generation of protocols on top lovely guys thanks so much for uh joining us today it's been an absolute pleasure you guys have a lot of really fun and in interesting and insightful points I thoroughly enjoyed it so again pleasure and uh episode two coming maybe six months down the line eight months down the line it'll be a much different conversation I think four agents doing the episode maybe sooner do you guys all have ASI by then hopefully or maybe singular cheers