The Decentralized AI War with Talus Network

Transcript

welcome back to the rollup podcast I'm Andy and today we're talking about crypto as a force for the democratization and decentralization of artificial intelligence AI has been a massive narrative in crypto and people are really wondering why it's important a little while we didn't really understand why we need crypto and AI why does AI need to be on chain after this episode we really dug deep into AI smart agents and why crypto and blockchain are very important uh for the future of AI we kind of believe that AI has a Divergent path and we want to see it be used for good hope you guys enjoy and uh let's Dive Right In welcome back to the rollup for an episode today about Ai and modularity AI agents Building Systems

for the future of onchain interactions and maybe they're not going to be humans Rob uh we're in modular march on the tail end of majo March uh welcome welcome we have Mike from Talis and Ben as well who we met in Denver and welcome guys thank you Andy thank you Robbie um Ben nice seeing you again here pleasure pleasure having you guys on uh yeah I think all of us are still humans here at least for the moment um so I think that's like that's kind of like where my head is at is like sooner or later we're gonna reach a point where AI agents are kind of like acting for themselves um and even before we started

you know before we hit record we were talking about why AI is so powerful that it makes it very important not to be centralized so so let's kind of start on this like decentralized AI front like why is it important that AI is decentralized yeah so thing about AI is that it's too powerful it's an exponential exponential technology and at some point like some big firm is going to take all that power and we're currently seeing that with like open AI even like um anthropic like they they're all running all these like resources and models and applications and they're reaping all the benefits of it right they're blocking the gauge for people to like customize to adapt to use to

leverage in whatever tool or setting they want to use AI they're having issues with data with compute power and in the end if we want to see AI transcend what we how we interact with systems and how we uh work with other people because at the end of the day we'll have agents interacting with agents and us not doing most of the work we definitely need some type of ownership of the people uh right like there's no way that we're going to have and see all these like huge entities own what we do what we say how we do because all these other influences and models will have a lot of uh way they'll just like change the way that we think and how we interact with systems and there definitely needs to be a way that we should control it rather than them

deciding what we do and how we interact with all these other applications yeah and before we get too deep guys like I know we can get deep into like decentralization and kind of like AI onchain but like can we just start with first principles like what is crypto and AI what is the use cases why have I seen so much like crypto times AI is is a scam like like why what is the intersection at like the fundamental level that you guys are seeing like for somebody who's new to this idea of bringing AI into crypto like what does that even mean yeah so Talis believes that the two most transformative Technologies in the world today are

crypto and artificial intelligence and artificial intelligence has incredible potential to help Humanity in a plethora of ways um and really improve everyone's lives uh but it's at risk of becoming centralized and controlled by just a few very powerful actors like much of web 2 has you know web 2 in aggregate is a is a fantastic technology that's worked out very well for Humanity but it's come with very serious consequences much of which are caused by um the fact that it's so centralized and that you know the just Network effects and um you know things Trend towards just a few very large companies

controlling most of web 2 in our web 2 experiences AI is very susceptible to this if you think about the inputs of what it takes to to make AI right compute data and then whatever model you're running just just uh breaking this down into the most basic building blocks um compute who has all the computing power uh who uh which organizations right now can subsidize really high Opex at a very early stage right it's very tough for startups to do that you're you're you know we're talking about Facebook and in Google here and and apple these are the companies with the with huge um uh resources to to fund compute uh who has

all of the data again because of the way that web 2 has progressed uh Google Apple Facebook the large web 2 companies are the ones who already have all of the data about us and they have access to a lot of proprietary data that's not on the open web um that's not accessible to open models um and then the last thing is you know where are models trending right now um you know models can go one of two ways we have open models that train on the open web or we could have somebody who looks at this a little bit more like iOS and like chat and says we want a proprietary closed Source model we don't want anyone to be able to look at this we want to monetize this as much as possible and we want to acrew as much proprietary data as possible to to train this model so

that we have the advantage um so right now in web 2 uh again I think AI has incredible potential it's one of the most important Technologies ever created but the way that it's taking shape in web 2 there is some risk there right right there is some risk there that that this could become this you know Draconian technology that um that influences our lives in in more of a negative way so at Talis we're thinking about those inputs and thinking how can we make these accessible to everyone and that's exactly what Talis is we're the Nexus where all of the inputs required for AI can come together and create Pro people can create products with these inputs that um are more decentralized and can benefit Humanity in in ways that

we we aren't able to see in web two like with smart agents gotcha okay so the core idea here is that there's this bucket of compute of power which there's a limited amount of and you have to bid or purchase this compute or this power to function these models to operate these different models and right now what you're saying is that this source of power is either expensive or only available to big companies or companies with a lot of money or one in the same or both and that that access is could lead to a more centralized output

of AI Control and so where blockchain comes in is opening that access amongst other things and even potentially streamlining the processes of like renting Ai and payments and things like maybe break it down even further so like what so that's like it is that the right kind of a mental model to train AI you need a ton of computing power you need a ton of data and you know you need to select what kind of model you run an open source or a close Source model um open source meaning you know just like the software anyone can look at it close Source meaning that this is proprietary data this is closed to everybody else we want to keep this a trade secret with compute you know it's very unlikely that you have a ton of uh gpus lying around

at home if you did congratulations because you're probably you could probably sell those for a lot of money um you know it's very unlikely you you uh create a lot of data every day it's it's literally Impossible on web to this is one of the problems in in web to and another reason we need blockchain technology it's impossible to keep track of the data that you create and to take ownership of your data this is not you can't own things in in web to um and then you know W with models like like I said we're starting to see the trends that a lot of people are prioritizing those close Source models and um I I think it's a much better world where we have open source models that train on open web data I I want to add to what Ben is saying but it's compute it's data

it's storage is models or algorithms that like come together to build all these agents but I think one of the biggest Promises of decentralized AI is transparency and verifiability so what does transparency mean is that you know that the actual model is doing what it should be doing and verifiability means that the output that you you're getting or the infs comes from the right model um and like with a blockchain it's very easy like it's Sovereign technology like you could see everything is like completely transparent like you know you could see what in what input input goes in what what's the output and the actual model should not behave as a blackbox and then with like cryptographic algorithm like

CK proofs now we're seeing like ckl opml um now like crypto cryptoeconomic ml or teml which people are starting to explore there's like different verifiability uh or different computational verifiability that could say Okay so this model this output actually came from the model and we could know that okay open AI is not lying to me when you just like submit like okay do this like task and they just like tell you how to do it instead of doing it because they want to save on comput like all these things that all these centralized companies or centralized AI providers uh remove control from the user from and that's all like the biggest promise of of decentralization as well yeah I I want to double click on something that that

Ben kind of illuminated here which are the three parts the the compute the model and the data I think data is pretty pretty self-explanatory like either we give our data or these big web 2 companies take our data unpronounced to us the models are like the weights right it's going to prioritize some things if it knows that you're a male and someone else is a female then it's going to prioritize some algorithm to show you what it thinks you want to see and then and then the compute and here I want to take a step back because my understanding of AI is that it's basically a big matrix multiplication it's taking the inputs and weights and putting them in a matrix and multiplying them together and so your data is going to impact the weights but all all that's

happening here when we're running these AI models is that it's trying to predict in the case of uh large language model it's trying to predict the next word that a human would say or it's trying to predict the next word that you want to hear so so when it comes to compute all these different AI firms and even users uh are trying to use the computing power for the Matrix that I want to compute it's it's it's trying to like use that power for the prediction that I want to that I want to try to predict the Matrix I want to multiply and so there's an access problem and there's a there's a cost problem to uh using the computing power for for uh the that particular matrix multiplication problem and so um

I just think it's helpful to remember that that these running these AI models is just a is just multiplying matrices together and so when it comes to who has the computing power there's this idea of decentralized compute where it's kind of like this Open Marketplace and anyone can use and use this compute for whatever matrices I want to multiply and then there's like this Walled Garden and and these Walled Garden types of eco systems are cutting off certain matrices cutting off certain models so that only those with access or uh resources to purchase this computing power are able to run the models that they want to run so Robbie you mentioned one word that I want to highlight which is like Open Marketplace um internally we've been

highlighting that a lot because it's not it's not like you need to have people to be able to access whatever resources they need for this Matrix multiplication to like provide uh data to models to like fine-tune them or improve them uh to like get that um storage capability where like actually the models are going to be stored because if the models are like let's say uh 10 of gigabytes that you need to store someplace like it's it's going to get more expensive throughout time and all that access and open aess and verifiability definitely needs to come into place in one single place where everybody could come where Andy could come and say okay I want to use that specific model from that provider with

this GPU uh specific GPU cluster okay you could do that um or if you want to choose like a separate data set that Robbie provided to uh fine CH your model based on Robbie's data you could do it and on top of that something that's like super interesting is how does ownership and monetization look like right like how do you make sure that Robbie's actually paying for what he's using or that he's getting paid for what he provides to the ecosystem and let's say what what if like Robbie and Andy at the same time want to own uh the same model like you can't do that in like centralized ecosystems in a trustless manner yeah and that's the decentralized compute aspect now you guys have hit on

verifiable compute as well as something that's separate from the decentralized or as a is a benefic is a benefit of a decentralized model what does that mean what is verifiable compute well just generally the idea in blockchains is that um it should be very easy to verify that somebody did something right so either blockchains is replicated State machines either means everybody's redoing the same work right like OG e you know just all these nodes that are recomputing all these you know for more more advanced blockchain technology takes that and says okay well we can we have much more creative ways that uh we can tell if somebody's telling the truth right we can we can

verify somebody's telling the truth without having to redo all this work adding this behind uh um you know uh Ai and and AI models would be groundbreaking right that would be in talking about these two past week can go down right like the more more negative more Draconian look for AI in web 2 versus how this could really be beneficial for everybody maximum beneficial for everybody and we mitigate a lot of the risks um you know verifiable compute is something that blockchain technology enables that's right up there with taking the good on um as as Robbie said uh this is a little complicated but yes like a AI training

llms for instance is just a lot of floating Point math like a lot of just replicating all these Matrix operations um you know a lot of the the uh you know that that's that's required for like a lot of the um how the hardware operates uh that's what gpus are optimized for um and you one thing that Robbie mentioned was the the parameters behind changing models so um one thing that uh you could do basically with governance um you know with with uh the the decentralized nature of voting with tokens um is controlling the weights on chain so that even if the model doesn't live on chain

um you can control the weights of the parameters on how that model is trained there's nothing like that in web 2 where um users of a product could actually have a say in how this AI is trained um and can can direct it in in whatever Direction they want so this is something hopefully Talis is going to be able to enable yeah so okay so I think that we've got the and and I'm just taking this a very just like stepbystep approach here guys bear with me and I think our audience is probably in the same boat um we've got decentralized compute we've got verifiable we've got decentralized compute for AI as a means of solving this kind of centralization problem of expensive gpus and and and

kind of gated access to compute then we've got verifiable compute which is like the properties of blockchains what what we know of them as today for creating kind of censorship resistance and uh permissionless systems um and so we have these two kind of core pillars here the the the the thing that that I'm kind of trying to get over or get over in my mind is like there's a finite amount of compute in the world of gpus of like hardware and there are companies and people all trying to access this for for AI models um like how exactly does Average Joe Andy get access to AI models

or AI compute through one of through Talis or through a decentralized AI crypto company which I wouldn't be able to get access to otherwise like like how am I going to go and access that compute and what am I going to go and do with that with that like create a model or kind of like what are the possibilities I'm sure they're kind of endless but yeah I'm curious like how do I go and use this so the way that we look at it is building a platform where everybody could come and like provide their compute provide their models provide their data their data so Andy could come and say okay I I want to use this specific compute cluster that Robbie provided that Ben provided and then I want to like put together that

compute with what the model that Ben provided and let say um if you want to even like fine-tune that model you could use the data that I provided so it's kind of like this like putting together the most like basic building block for AI resources um where like anybody could come and anybody could provide an incredibly neutral manner whatever compute and whatever verifiably verifiably Spectrum there is so it could be like let's say CK op U from like te or it could be like completely centralized right it really depends on the use case uh with whatever model is in in the market or has been provided or committed to the network um and then on top of that you have like a solid like

SDK framework that enables you to put together all these different resources to build the applications and even that was very low level um it just keeping it at a high level like what what does that mean for your users or your listeners uh this might be helpful to get into what a smart agent is but basically what Talis enables this new AI use case that you can't do in web to for a lot of the reasons that I just mentioned um and I I'll throw one in there too uh that's liquidity right AI can't open a bank account so there's no there's no such thing as like a banking API that you can just plug in an existing AI model today and it can AI can execute transactions for for you the future that we see like the next great unlock with with AI um is

actionable AI so right now we a very performative generative AI you go on chubbt or Claud 3 or something and you can you can ask it questions and it'll it'll give you answers right or it'll create new data for you like write me a call whatever um the AI today is still not actionable right so you can't you can ask uh uh chat GPT um you know I I want to leave uh New York to LA what's the best flight to take on on if I want to leave this date and keep it under you know this amount of money it can give you an answer but it can't book the flight for you um and the biggest reason for that is because AI can't open a bank account right AI is going to have to use usdc and not USD um so that's the

in addition to all of those other inputs that we that we uh bring together for AI uh liquidity is the next thing that we want to add in there that's that's Chief among those so I say all this to say that in the future what you're going to be able to do with smart agents on Talis is um execute user intents so right now we're starting with defi use cases taking this idea of intense and defi like um I have uh this amount of usdc giving as much eth as possible you know automating that process to make it very user friendly um and then expanding that's a whole plethora of use cases in GTO but um anyway going back to your original question we've talked a lot about the backend and we keep talking about inputs and data and um you know floating Point arithmetic and all this

stuff uh one of our our chief concerns here at Talis is making sure that we have are very ux friendly right so the average person is going to be able to interact with our platform they're not going to have to worry about all the blockchain and and llm and you know model stuff on on the back end they're just going to be able to execute with our smart agents so cool so cool like that that's like the ux unlock is like I can just natural language type English into a search engine and then it'll go execute whatever action I ask it to it'll do research first and then it'll go start to perform those actions and I think the example of when we start putting money behind these actions that's when it becomes so so important to

verify the verifiability becomes ever so important for those actions um right because I mean it's just obvious if I'm GNA if I'm going to put money behind something I'm going to want to verify that it it's doing the thing that I ask it do so just Ty tying this together it's like okay once we start to insert monetary value into these AI agents like we don't want these agents to go Rogue at any point but let alone if they have if they have money and it's a significant sum of of like savings or checking account right like they're going to have a lot of a lot more power than they do right now and it's very important to make sure that we have those domiciled and we have those like verifiably

Crossing these checkpoints at which point we can say no like you didn't you weren't supposed to do that don't do that or yes keep going I think like right now well everything started blockchain blockchain interaction started as like just being humans like you send the transaction you you're the only one who interacts with the blockchain now we're starting to see a bit more of like AI AED uh blockchain interactions where like okay you let's say asking llm what you should you should do in the blog Sher like how should you interact or like just explain me what this project does or even like audit this smart contract to see if it's worth me interacting with it I think the next stage and is like specifically what

you guys are talking about where like everything is going to be completely AI oriented right like you're not only going to just like ask the llm to do things or the agent to do things but you you won't even like interact with blockchains right like they'll just manage and do everything for you like you just like deposit like let's say 100,000 usdc and or even like one usdc for whoever doesn't have like uh that much money or like much time to invest and the the idea is the agent would like manage and generate wealth and generate value for whoever owns or like provides access to that agent um and it's gonna it's been like slowly but surely and

you've seen this in like other Industries like where starts like only human then human plus AI in a guarded way um until we make sure that these agents actually really work in the way that we want them to work and then after to the feature is okay agents like interact with each other like in an open ecosystem talk to each other like figure out what are the best strategies and strategies and make sure that they all happen on behalf of the users without then them having to do anything I I'd like to revisit like a very early question for in this conversation which was um the decentralized AI question and we speak a lot about cloud computing and how blockchain is now this like new era of cloud and specifically it's like this

verifiable Cloud so could you guys talk a little bit about why smart agents are not possible on the web 2 version of the cloud but they become possible on the web 3 version of of compute so the main definition of a smart agent is a program or piece of soft software that executes actions on a certain direction with like predefined goals like that's a vanilla definition of like smart agents it can act it can communicate it could do all these things in autonomous ways smart agents are sometimes like like you could say that they're possible in web 2 like they are like you could see like Bots that go and like trade and

that is a web to Smart agent right I think what what web3 does and Ben you could add into this as well is it makes it verifiable it makes it trustless and I think the one of the biggest value ads and Ben touched a point around this earlier is that you have the open infrastructure from for them to interact and communicate right like you don't know if like Robbie's agent will be able to send a payment to Andy because he cannot connect to his bank account or even like send a payment through stripe like building all these agents right now and if you know anything about like just how like build agents and and traditional way like you have to code and connect directly to each single like API in their different

like formats and like scrape the web and see how the different like uh front ends look like so you could like provide the data and then like scrape it to like understand what's the whole like context of the website and the applications and that is too cumbersome to scale like I think agents need a whole like simplistic open infrastructure to see okay so this smart contract looks like this the function is like exactly the same as with interacting to any other protocol let's just like put that together and in a seamless way like execute on chain um and I think like web 2 is too brittle or too siloed um and blagi wrote a post about this like okay how do how to like decentralize web 3 uh databases are all like or all like

blockchains have like different like data models within one single set and like putting all this together is like okay so the smart agents will interact with each other we like transfer data uh would communicate um in an open ecosystem and that's what I think that's one of like the biggest Promises of we three like Mike said uh it's tech there there are some uh very low performing smart agents in in web 2 nothing like what we're going after but yes you're correct in that cloud cloud basically scaled the modern internet like as we know it today um so for something like AI these these resources that I talked about Cloud democratizes a lot of them in a web 2 setting it gives people access to scalable Computing resources

you don't you know you don't need like localized gpus to to train a model um and it it completely change data storage right like you could you you could score store data and interact with it some uh somewhere that's not um localized to you uh the biggest uh thing lacking still in web 2 and in terms of performance right like we we've really uh uh I I I I feel like gone over um the benefit of verifiable compute and privacy and giving users more control uh for egalitarian reasons right this is just a better outcome in in in creating AI for for for the average user um all of that applies to every use case in crypto

right transparency we making a more open system uh things are just better that way when they're even more democratized but in terms of performance what we're still lacking in web 2 is intera right so if if you have data on Google Cloud if you have dat and then you know I have data on AWS um not only are they not uh is there not a way for that data to communicate but uh these big web two companies are are centralized you know they have an incentive uh to uh not to interoperate right um this problem is gets orders of magnitude worse with uh anything related to money right if we want AI

to handle money it just has to be crypto it has to be a currency that lives uh on an interoperable common data substrate because there is not this is a huge failure of our banking system there's absolutely nothing like nothing close to like a banking API or something that Banks could communicate with each other very quickly right you know this is going back to like what the use case for Bitcoin but a payments take weeks to settle right this is not something AI is not going to be able to be performant with the tools that web 2 has today for something like smart agents it needs blockchain technology it needs a completely new paradigm for how smart agents can

interact gotcha so we have so Mike was talking about uh about 10 minutes ago about how this kind of works for for the users there's an Open Marketplace for um you've got uh people that can upload models people that can provide compute um um and then what what you're saying now is like okay blockchain's going to also bring in this like this different a incentive mechanism which we should touch on and then B like also this this ability to interoperate between different users effectively having these open market places where people can uh provide models and and rent gpus all on crypto rails so so if I'm a if I'm a person who has compute or you know I

have a warehouse of computers or I'm just really damn good at creating models and and we've created a bunch of proprietary models what what is my incentive to come and and bring this to the Open Marketplace in an open source manner either for the compute or for my my models like if it's a money thing what's the incentives for these decentralized AI marketplaces to actually work like like why would I just not go to Google and be like you know hey guys like I have this like how much is this worth to you guys so this can be like be identified it with a pretty simple example let's say Andy you're the artist the guy who like really knows how to draw and have all the data regarding

like these cool images and all the IP and you want to create like let's say a collection of art AI generated art but you don't really know how to do it in a way that like the actual benefits of it will go directly back to you and then there's let's say the other party who's a scientist a guy guy who's like really good in creating models and really understands the whole like math of it um I think they could in a Marketplace like this they could come together you could have the artist and the scientist come together generate a model or like fine tune a model with data and the actual algorithms themselves that would have a specific use case let's say in this case is like generating images for like an nft collection that could be sold um and then they come together they generate

this algorithm or this fine tune model they generate the collection but really who does it belong to like is was it for the guy who like actually built the model or was it for the guy who had the data and could had the creativity to make things actually uh be PCH purchasable and that people would really want to own and so it's like a two- side incentive like I could provide my abilities somebody else can provide their abilities they could come together and then you could have like an third like an output that is owned by both parties at the same time and let's say every single transaction from that nft collection goes back to the both owners of the initial model that created the the nft so it's like it's an incentive where like anybody could come and

provide their capabilities provide what they're good at and then somebody else on the other side where it's like the demand side could pay for it and it seem lose weight and the idea is that not only could they just I think there's like a lot of use cases that could come and send out for this in terms of like ownership and monetization but it's like the idea of like putting everything together um for it to make sense to people to build these applications Ben you might you might want to add something but I just have a quick thing here which is that the The Entity or the agent that like supplies that model and then the one who supplies the art and the one who goes and buys it those could all be Bots or or AI agents themselves so I think it's it's a really like

peculiar thing where these AI agents like maybe one's really good at like making these models and these weights and then another one's really good at like designing the use case for that model model and like you can just start to see how these AI agents one are potentially going to start owning their own assets with like those that model is an asset if it can be sold and repurposed I just think that's a really funky future that we're getting into because then these things are just going to start acting on their own their own poition and I think it'll be really it'll be really weird I like the cool cool thing about it it's like in in this space you could the output of one model could be the input of another and this just like generates like a

positive feedback loop where like you just outp put some something regarding like content or like an asset and then the other model consumes it and then you just like okay as you say like having self- sustaining growing ecosystem digital assets where they compute their D it's data it's content it's text it's images um and like agents are like speculating on each of them and I think it's what like like hyper financialization of markets is what like I think um Nick from all calls it like where every single type of digital asset would be are represented and like speculated upon off because um there is an entity that could like recognize and keep track of all of them uh and it's

it's a pretty like crazy feature and and I don't know it I don't think a lot of us can even like cope to what it will really look like because of the exponentiality of it um but I'm excited gotcha so let's come back to reality boys I know that we all love Ai and a little circle jerk but if I'm a GPU provider if I have Hardware in a you know Warehouse why on God's green earth am I going to just go give it to some AI Marketplace in crypto when I could just go give it to Google like what's the incentive uh so the more generally the incentive in what's cool about crypto is that you can incentivize any Behavior you want right and it's very crypto

markets are very reflexive right so if a lot of people agree that this is valuable it'll make it even more valuable um if you had a GPU at home like let's say you had a like a graphics card for gaming or something that's not necessarily optimized for or doing floating Point math and training AI but let's say you wanted to contribute some compute um the incentive if like for you there's no way to do that to to send it to um Google headquarters in Mountain View or something like that and and you know contribute to Google um crypto is all about contributing to a communal project um so you know there's there are a lot

of cool um compute uh platforms um that are working on this but basically you can whatever gpus you have at home or whatever Hardware you have at home um you can uh Aid in uh doing this small amount of of help training this model um and you can get paid and say the the the native token of that Network you know they're they're basically compensating you crypto so it's the little guy like the little guy can contribute to the network whereas they couldn't in the past exactly I just wanted somebody to say tokens I mean yeah I think that that's it it's because like in Bitcoin proof of work like yeah you had to have

a bunch of gpus in a garage somewhere to contribute some compute power to the network ortion we don't need that anymore now we have cryptoeconomic security so all I need to do is stake some eth and now I'm contributing compute yeah yeah that's that's uh that's basically um the the the right idea um crypto lets anybody uh participate in doing whatever communal task whatever um common common goal uh the the the network wants to do and you know so for for for providing compute um to to train AI in a more decentralized way there are a lot of really cool projects working on that that are that

are doing a great job so do you guys see like one two three decentralized AI marketplaces winning and like being like the main ones that everybody uses like the Lio and like the rocket pool and The Ether fi or the blur and the Open Sea and the and Etc or is do you guys see a a a a more like specialized like more like modular uh kind of uh situation where we have you know different marketplaces for like more specific uh use cases and more specific AI trainings um and then further like are you guys aiming to be very general in terms of

your reach for Talis or like is there a specific Niche or area where you're really trying to like specialize in that that's a great question so uh for the market and whether or not you know what kind of Market this is going to be I think it's way too early to say um I will say that you know in in our opinion what Talis is is working on what we're building um I don't think anybody is building uh anything specifically like like what we're doing we're the common Nexus of all of these inputs compute data liquidity and models to come together and operate within the same economic zone that's the first thing second of all the reason why uh we're building our own chain is because we didn't we we couldn't see this as an application play um I don't think

there's a single blockchain that exists today um L1 L2 anything uh that could support uh this specific use case of smart agents um even the very very high performant blockchains uh you know we're making huge uh optimizations to our chain specifically geared towards smart agents so um you know in that sense I do think that AI uh compared to other use cases in crypto probably Trends toward more modular just because there are so many different inputs coming in and if you want to build an AI application there are so many external dependencies that you know you you it's it might be better to have those optimized in their own vertical in more of a modular sense um and then the you know the the the

last thing what do we want to optimize for uh we will have uh as we're planning right now general purpose defi capabilities specific to our our special purpose though which is smart agents so you know we need to have um you know the the the the ability to exchange on on our chain um you know we we need to have um certain defi infrastructure in order to to to provide liquidity best to Smart agents um but this smart agent use case is really what we're focused on right now creating autonomous agents that can execute uh user intents starting with Defi and expanding to everything else awesome sweet I guess the the last thing

from us is going to be like um where are you guys at in your product development what do you guys have on the road map uh how are you thinking about kind of navigating the bull market um and yeah where you guys at yeah so it's been a lot of work on our side um currently hopefully launching some interesting Partnerships over the next coming weeks uh so stay tuned um we've also been like internally working like our def which should hopefully be done within a month uh so enable like people to start like interacting with the chain and like seeing what this actually uh looks like in in reality and yeah I think that's like most like we'll hopefully like be

releasing our light Vaper soon as well so um a lot of like a lot of people have been asking what is Salis what is Salis um so I think we're ready to come out and say okay this is Salis comfortable with us and um happy to like look and research and ideate around all these use because they're they're fairly new and we're starting to just like see the tip of the iceberg of what's really going to be the crypto and Ice Space Mike and I are both in lock step in that we set really really aggressive timelines and we're always pinging each other you know talking about how how quickly we can we can get to our our uh our main product but um you know we're we're really trying to tackle something that hasn't been solved yet in web two or in web three and there's a lot of external

dependencies that we have so um the one thing that I can promise is uh we're going to work very hard and everybody at Talis is incredibly focused on product and creating the best user experience possible so um you know if somebody's going to do it hopefully hopefully it's us awesome it's been a pleasure guys thank you for coming on Rob any closing thoughts man I just renamed the podcast Studio to killer use case AI smart agents uh because I think I think that's where we're going I think like AI is potentially the use case that blockchain has been missing out um for all this time so really really cool to see this uh this idea start to get start to get like massaged into the into the real world

and start to come to fruition um best of luck on uh the entire development journey and uh ship and drip boys thank you guys appreciate it