Crypto's Answer for the Global Web Resource Race | Andrej, Grass

Transcript

data that's been present on the internet for five minutes is exponentially more valuable than data that's been on the internet for 10 years or longer right when it comes to the most valuable data that an llm can actually access it's data that's happening immediately if you ask an llm about an event that happened last night you want the ability to go and retrieve that data in real time and that's what grass actually provides and that's where it excels because it's fast and it can access anything on the internet the polygon Community grants program was launched with one ion tokens all four polygon Builders season 1 is live right now and features 35 million in madic to support the next generation of polygon products builders on polygon these grants are for you join the aggregated future today by applying at

polygon.net on zerox Research is a recommendation to buy or sell Securities or tokens this podcast is for informational purposes only and any views expressed by anyone on the show are solely our opinions not finan advice bacho Ryan and our guests May hold positions in the company's funds or projects discussed what's up guys welcome back to another episode of xerx research today I'm joined by Andre from grass um Andre thanks for coming on yeah thanks for having me guys before we get started do you want to tell us a little bit about grass and what you guys are building over there definitely um so grass is a network that anyone in the world with an internet connection can join by installing a node on their Internet connected device

uh and what that note is doing is actually selling some of their unused bandwidth for the purpose of web scraping now on the other side of this network are AI labs and large companies that actually very heavily rely on datadriven insights and need access to the public web um so they tap into a network like this and are able to access any public website in the world um and this is actually a practice that has been going on for 20 or 30 years uh where companies are tapping into residential internet connections in order to scrape the public web uh but to dat they haven't actually been compensating users for doing this so so that's a big reason for why we're you know why we're working on what we're working on perfect I um I kind of want

understand why you guys decided on building grass so you mentioned there that this has been happening for 20 or 30 years but people just weren't getting uh they weren't seeing the benefits of it is that are there any other problems that you guys wanted to solve absolutely I mean like I mentioned we started working on this because we just hated the fact that so many people were essentially being robbed by these middleware Solutions providers who are then selling people's resources to very large companies um and this industry has grown to a point where it's NOW essential for any company that wants to compete in the modern world because they do need access to public web data um and what we found extremely

unfair was that the average person who's actually helping Supply this network is not taking part in the upside of the growth of that industry um so we started working on that just because we found it incredibly unfair and we figured well if this is already happening to all of us when we download a free screen saver or free VPN uh we might as well take back a little bit of control and have the ability to opt into networks like these of course as we started working on this Chad GPT came out and the land Cape of demand for data shifted very dramatically by and the demand actually grew in orders of magnitude and what we started realizing was given the internet and the practices

of web 2 sort of tend to move in the path of least resistance uh there's no reason why these types of broken incentive structures would just continue to compound so what we started seeing was not just users being cut out of the equation by having the resources like their CPU and their bandwidth being abused by large companies we started seeing you know media and Publishers and all kinds of other actors and like stakeholders really in this industry uh being robbed in different ways as well so you know since the day we started working on grass to today um the scope of what we're trying to build and the problem that we're actually trying to solve has grown pretty proportionally with the growth of AI so I'd say it's

100 or a thousand times bigger than what we originally said set out to do that makes a lot of sense I um I understand the need for Quality data especially for AI models but I kind of want to understand like how you guys see that fit in because I think I mean obviously a lot of people think different parts of the AI um I guess resources are more important so something computer is more important something data is more important but I kind of want to understand how you guys are approaching that what's your thought process uh definitely I mean at the end of the day um an AI model is a representation of some amount of knowledge and that knowledge came from the data that it was either trained on or that is being provided to

it in the inference process uh and if we just look at the AI industry and the way it operates today the companies that are dominating the development of AI are actually the only two companies in the world that have access to an internet scale web crawl and companies that are able to crawl the ENT ire internet and actually leverage all of the data on the internet in real time effectively manage to build multi-trillion dollar businesses with that as the foundation so for us it makes a lot of sense not just to focus on data but to build the first ever user owned internet scale web crawl and and that's something

that we actually see us you know see grass being capable of doing as it grows so makes sense I um how does it I guess like as a high overview I download a grass node um and grass utilizes my unused bandwidth but what how does that work um what does the architecture look like I feel like there's a decent bit of misinterpretation of how this exactly the process works out so would love to hear like an overview of it definitely uh so let's say you have a need for a certain data set uh grass recently opens sourced a ton of Reddit data so I could probably use that as an example um you've got a validator that sends a set

of instructions uh which would look something like hey scrape every single subreddit on Reddit from the year 2024 now this is sent us a ton of web requests which are then processed by routers now the routers on the network what their job is to actually receive those web requests and send them to nodes and all the nodes on the network that's what people are installing on their devices now when you have 2 million nodes on a network like grass which is you know the size of the network today what you're able to do is distribute 2 million web requests through all these nodes in parallel and hit different pages of Reddit and download all of that data at once and what you end up having is actually two key advantages by

scraping this way the first one is very obvious it's parallelization right you can scrape all these websites in parallel uh and the other one is a bit more nuanced but it's probably the most important one and it's the reason why companies have been opting to abuse people's resources and not just set up IP addresses and data centers and the reason being uh every IP address in the world is effectively kyc you can see whether it's a residential IP address or whether it's a data center IP address um so from the perspective of a web server that doesn't want to be scraped for whatever reason let's say they're you know gatekeeping some information or or trying to hoard their user data if they see data centers they'll usually just

block the data centers and if they see residential uh traffic they can't really differentiate that from real user traffic which they're actually trying to attract um so in terms of this data that's being scraped and like I you mentioned there were some misinterpretations there are actually a number of users who um they they think that grass deals with private data or like user data um it's quite the opposite uh grass can't see any user data uh all the node is doing is literally using some of your bandwidth to scrape up publicly available website much like Google does right when Google goes and crawls all these websites and shows you previews of them that's kind of what grass is doing really is just reading public websites yeah that's why I wanted to bring it up because I saw a

few of these comments over the past few weeks and I I knew that that wasn't necessarily the case I um and you guys have decided to build out a sovereign data rollup what were there other Alternatives that you considered how exactly does the rollup fit into the the equation here definitely so one of the major problems that we want to solve with grass is actually the issue of data poisoning um if a web server has a ton of valuable information being hosted on it um they might have some level of incentive to actually go and poison a data set that's being scraped from that server so if it sees let's say an i

let's see it let's say it sees an IP address that's coming from you know a data center they can go and change the data on their website and effectively Honeypot a scraper and mess with AI data sets that way um another way that people are poisoning data is actually retroactively um they're looking at data sets that have already been scraped and they're introducing some level of bias the number one reason for this is probably going to be advertising and it's already starting to happen um but some of the more dangerous ones are more political right where a political party might want to introduce some amount of bias to a data set now what you need to end up creating is some lineage between the data set and the source of the information now just given the volume of

traffic that passes through grass like I mentioned two million nodes earlier in order in order to achieve an a a true internet scale web crawl like grass will need a 100 million nodes U if you can and if you think about the number of web requests that have to go through all these nodes on a constant basis it's about 70 web requests per session so 70 times 100 million nodes on the network um every moment there's not a single blockchain in the world that can actually handle that level of throughput in order to act publish all the lineage that's necessary on chain um so what we're effectively doing with this piece of infrastructure is rolling up all the web requests like hashes of the web

requests um so that someone can go and explore and see the lineage on this rollup and what gets posted to salana is really just summary statistics so how much bandwidth got used and uh which nodes need to be compensated and which routers need to be compensated as well uh you mentioned data poisoning attacks as a reason to like it's partially why the grass Network exists it's a great reason to use the grass Network could you help us understand kind of like the fre like how common are these data poisoning attacks um how often so does someone like building a foundation model run into these data problems yeah it's a great question uh right now it's AI is

still very nent and it's hard to say you don't have enough of a sample size to say the frequency of how much this is happening but if you just look at the natural progression of data access you can liken any llm to a search engine they effectively serve the same purpose you enter a prompt and you and it returns some amount of information then you can go and explore that information further uh if if you look at the way seos are being monetized today and have been for a very long time now it's been through data poisoning we just call it SEO now if you think about how llms are going to be monetized in the future it would be really silly to think that they'll do anything

different and as this constant pressure for a lot of these large companies to monetize further increases there's no reason why they wouldn't do that now this scary thing is when you're poisoning data in an llm as opposed to a search engine U an llm has a lot more influence than a search engine um in many cases it can pass the teering test so if you think about something very convincing telling you something that has been given some amount of human bias by some third party that wants you to think a certain way whether from a consumer's perspective or from a voter's perspective or anything like that uh it becomes a very scary world to live in and just the fact that that incentive

exists means it's very likely going to happen and that's a reason like one of several reasons why grass needs to exist if we want AI to be fair yeah I'm thinking that that's really interesting I'm thinking about like who are your like most obvious like users like obviously if you want to train a foundation model you might want to use grass but like what if you want to train a multimodal model or what if you wanted to do like some sort of rag method over data like are all those potential customers for you guys uh all of them are yeah u i mean grass is currently scraping some very large multimodal data sets some of them will

be open sourced um so modality isn't really a restriction any data that's present on the Internet is that is public and not behind some login wall or something like that is actually accessible by grass um and in terms of where we see the network being monetized long term I wouldn't actually say it's necessarily in training foundational models in terms of foundational models especially for text Data a lot of this data will become pretty commoditized and that's why we were more than happy or that's why grass was more than happy to open source over 600 million rows of Reddit data um that's data that companies out there paying ridiculous amounts of money to access grass is fine

with open sourcing that because where we actually see the real monetization happening is in live context retrieval um data as we know like the its value decays very very quickly data that's been present on the internet for five minutes is exponentially more valuable than data that's been on the internet for 10 years or longer right uh so when it comes to the most valuable data that an llm can actually access it's data that's happening immediately if you ask an llm about an event that happened last night you want the ability to go and retrieve that data in real time and that's what grass actually provides and that's where it excels because it's fast and it can access anything on the internet um

so that's really interesting so if if there's a large opportunity in live context retrieval um I'm wondering like what are either the cost or product advantages of something like grass versus a traditional way of doing that web scraping like is there a difference there well you end up consolidating um I I guess a couple of middleware providers uh where where I was like you know if you wanted to build an LM that does this I mean Google can do it right now because they own a knowledge graph of the whole internet so they're already doing this in a sense um the issue with the way

they're doing it is the fact that it's closed and none of the people who've actually contributed to the internet are getting any level of ownership in this or any level of compensation in this or or say in how it's being developed even so I guess there's a huge ethical and like moral reason for why grass should be building this and why this should exist on the grass Network U but in terms of cost and um and ability anyone other than Google that wants to build something like like this is going to have to pay for a third party knowledge craft is going to have to pay for some Ser API of some sort and is going to have to pay for hosting if

grass is capable of doing all of this and effectively doing what Bing and Google are doing um then you end up having exactly the same cost advantages that those two huge companies have hey guys permissionless the event for crypto natives is heading west to Salt Lake City and we're bringing the biggest names in crypto together for an incred ible can't miss event join me on the rest of block works on October 9th to hear from more than 200 Industry titles including Nick white from Celestia shiram from Igan and Keon from Mona the conversations at permission list will be covering the hottest FES in crypto including modularity reaking the Bitcoin and Solon ecosystems Ai and crypto rollups l2s institutional adoption and

of course the November election that will be right around the court we couldn't be more excited about this event and it's happening in one of the most beautiful places in the US ticket prices will be increasing towards the end of June so get your ticket today make sure to use their code so I get some credit and bragging rights from the rest of the team uh use ZX10 for 10% off on your ticket one more thing we're also hosting the permissionless re hacka font in collaboration with blebox the theme this year is a multi-chain future to build a better cross chain experience so I don't have to have a heart attack every time I bridge to a new chain all Hacker found participants will receive a free ticket to attend permissions so so if you're a developer come join us in Salt Lake City team up with fellow devs

and build the multi-chain future polygon Labs is developing the next generation of open- source ZK Tech to aggregate crypto liquidity and user bases empowering developers to grow in a unified web of interoperable chains with the a layer the a layer is a machine for Network effect don't worry about boost wrapping liquidity just build anywhere in the a layer to tap the liquidity and users everywhere to support the aggregated future and boost scrap quality project the polygon Community grants program was launched with 1 billion tokens All For polygon Builders season one of the community grants program is live right now and features 35 million in madic to support the next generation of polygon projects for Builders on polygon or projects interested in migrating these grants are

for you join the aggregated future and feel empowered to grow in an infinite web of unified interoperable chains apply today at polygon.net context reval is and how it would work in practice because I feel like that would be something that's interesting for the listeners sweet yeah uh the the way it works really is if someone asks a question or sends a prompt to an llm um and that llm hasn't been trained on data that is recent enough to answer that prompt or or or doesn't have enough context or whatever it might be um but the llm is capable of

doing if it's hooked up to grass is to go and access that information directly from the internet add it to the prompt and from the users's perspective like they're not really seeing that happen this is all happening in about 500 milliseconds but you send a prompt to the llm all this context so like let's say 10 or 100 news websites get added to this prompt with a huge context window and then the LM can go and then it has the information that it needs to actually give you um some response with that level of insight that that's required um so in practice it's actually fairly simple um where the difficulty comes in is building the underlying knowledge graph of the entire internet uh because you do you do need to know

where to search U and in order to do this like not only does grass need to crawl the entire web it needs to build a knowledge graph of the entire web and this is only possible with a certain level of nodes and that's probably 50 times more nodes than grass has in the network today so you guys did upvote web for um Reddit post the data set and one thing you mentioned was that it's becoming commoditized or it will be commoditized which makes a lot of sense given that I assume there are other parties also doing this including you guys um do you have any plans to scrape other websites or is there anything that is like that you guys have already started on how does basically what I'm trying to understand is reddit is obviously very Val valuable Twitter I

assume also and Publishing websites how do you decide on what to make public is this and who is the customer for the data set specifically maybe apart from llms or I guess including llms yeah there have been a number of research groups playing around with upot web that are not building llms um a number of them just given the topical thing today is the US election a number of them have just been doing sentiment analysis on the election and trying to inform prediction algorithms um there have been another a bunch of other random tools that people are building like movie recommendation engines for instance uh uh using you know people's thoughts on certain films and certain subreddits uh

basically any information across the 70,000 subreddits that are found in that data set can be leveraged to build all kinds of really cool and exciting things so it's been pretty fun for us to just watch people reaching out to us and saying hey look look what I built with this data set um and so yeah it's not necessarily just llms it's just that the llms are probably the only use case that are going to gobble up the entire data set without without picking and choosing little pieces of it um and as far as other I guess domains and other types of data that is currently being scraped and crawled by grass the vast majority of it is going to be open sourced um and this ranges anywhere from Price data for

retail and e-commerce websites to long form articles on all kinds of different publishing domains uh so and and I mentioned earlier multimodal data as well um there are tons of videos and like audio clips that are found all over the Internet and at the moment these are things that are actually very difficult to find with existing Solutions such as common crawl um the way we're informing how or the way grass is currently being informed in terms of what it's scraping and what it's crawling is mainly through conversations with some of the more prominent open- Source developers in the AI community and them saying hey look this doesn't exist um this is sort of being kept gatee kept by a small number of

companies and this is the type of thing that would bring us one step closer to being able to compete with some of the best Clos Source models and that's how things are being prioritized currently longer term the idea is to just crawl the entire web uh so there won't be any amount of differentiation it'll just be we want the whole web as frequently as possible and everything being cash and everything being stored and anyone being able to participate in that I'm curious about regulations um I think of like the increasingly um active Regulators in the EU I think about the right to be

forgotten um there's the controversy over open AI using YouTube um using that data set um have how easy is it to be compliant when you're crawling the web and how does grass think about like compliance in this environment definitely I mean compliance is actually surprisingly straightforward when it comes to crawling the web um there are you know as as I mentioned there are two huge search engines that have been doing this for a very long time and have laid out effectively the pathway to being compliant uh what we find actually has a bigger issue is the fact that incentives

aren't properly aligned so sure it's people are allowed to go and crawl basically anything on the public web because it's on the public web otherwise search engines wouldn't be able to work where we see the issue being is the people who are publishing some of that content aren't getting to take part in any of the upside of posting that content to the to the internet in the first place um and that's one of you know one of the future things that we're looking at at in terms of building a fair Knowledge Graph is making sure that anyone that's not just contributing to scraping the web but also supplying the information that's going to the web can also take

part in something like those I know you guys have a an rhf or a data labeling program could could you touch on that yeah uh so grass has quite a bit of distribution especially for for a crypto project um I actually think in terms of real users at this point in time it's one of the largest crypto projects in the world um and given that distribution and given the fact that it's actually a pretty passive Network to participate in um and a lot of these users in our community are actually quite active on a day-to-day basis we realized they want a more active way to participate other

than just downloading a note and referring their friends so we figured well data actually benefits quite a bit from Human feedback we may as well add a section to their dashboards uh where they can go and label or tag data sets or or responses from prompts um I guess one thing I'll mention on that topic though is that we've always seen human reinforcement or human feedback for reinforcement learning um in data be being something that's going to be saturated very quickly as an industry um almost every company like that that has lws of distribution is starting to like dip into this just because there's been

a sudden increase in demand for it um where we see it going longer term is probably a similar direction as online surveys have where you know in the in the late 90s and early 2000s there was a ton of uh valuable or I guess there's a ton of opportunity to dve a lot of value from people completing these surveys and that became extremely sad saturated as the number of companies doing it increased um so for the grass Network it's something that we see as beneficial uh but long term probably not going to be the focus of what we're doing gotcha I'm curious I know that different types of devices can join the network right like like from like a technical perspective is one device

better for scraping or crawling than another device I'm thinking Mobile versus desktop yeah uh mobile devices tend to be better yeah and the the main reason for that or for there's two reasons the first one is the mobile device fingerprint generally is a lot more likely to be a human and the second one is the IP address so ipv4 which is the IP addressing protocol that the internet kind of created on uh was exhausted over 10 years ago we ran out of IP addresses now there are a lot more devices in the world than there are IPS so when IP addresses are being assigned to mobile

devices in a certain area in a certain city a lot of ips or a lot of phones are going to share the same ipv4 address which corresponds to a cell tower that they're all connecting to now if someone wants to block an ipv4 address that corresponds to that cell tower they're going to be blocking all the US users in that geography and that becomes quite difficult um and there are also a lot of other random things too um so like bot bot detection um when you're on a computer it's really easy to detect a bot by tracking the mouse movements um but on a phone uh people tapping on the phone is actually very similar to way

the way a bot operates it's very hard to differentiate all you have to do is add like random time spacing I guess between each of those tabs whereas um when you're uh when you're moving a mouse it's uh like people can detect that with over 99% accuracy right now so that that's really interesting so and and what are the the and the core reasons why you would want to block Bots like I would think that a lot of websites would want to welcome them um just like as distribution and for hits but like why do why do these websites want to block Bots to begin with yeah it's a pretty interesting phenomenon where it started out in e-commerce because the e-commerce

companies were the first ones to start scraping each other at scale the reason they had to do that was because they wanted to make sure that on price aggregators so whether it's like you know Google's price aggregation or some uh travel website or whatever it might be they wanted to make sure they have the lowest price so they'll scrape their competitor's entire inventory and make sure that in each geography their prices are lower than their competitors and and as I mentioned earlier they're using you know your internet bandwidth to do that because you installed some you know screen saver that they own anyway um this has been happening in e-commerce for a very long time uh and very recently it started happening in the

rest of the internet reason being as to your point five years ago most websites that are hosting any language content they want to be scraped because well most of the time they're being scraped or crawled or indexed by an engine so it increases you know their visibility or they just like the vanity metric of having more views on their website so they don't mind it um something really interesting happened in the last three to four years where uh language data suddenly became incredibly valuable so they said okay how have people prevented this previously and they looked to e-commerce and started adopting the same sorts of tactics um and and and the funny thing is it's

it's almost one for one with the way they operate like a lot of these platforms are now scraping each other because they can scrape another platform's content run it through an llm and then create their own content in case they missed a story or something like that um so we're starting to see a lot of the patterns from you know Industries where you have uh like a much shorter path like gratification where we have like more instant gratification such as as e-commerce um they're starting to repeat themselves in like the rest of the internet but at a much larger scale so yeah most of the time anytime a website finds out that the value on their website or the data on their website has some value and

itself they're going to try to gatekeep it I was just GNA ask about um if you have an I guess not internal but maybe like a way to score nodes so like would a grass node that I run if I'm based out in Europe um be worth less than a grass node that Ryan runs if he's based out in the US does it matter if or not so geography is pretty important um the biggest reason for its importance is actually latency um so if there are a lot of web servers being hosted in the US and many are then a node in Europe is going to

just be physically further away and it'll take longer to Route traffic through that node in Europe so for you know a scraping or crawling job that is focusing on websites being hosted in the US US nodes are going to be a lot more valuable uh now beyond that websites are also going to display different information based on your location so there are cases where you want to scrape the same website but from a lot of different locations in that case they'd probably have equal value so it's very dependent on what's being done in general I'm wondering about um just to kind of like go more high level here um there's a lot going on in crypto AI this is a non not grass

specific question um but there's a lot going on in crypto AI um there's a lot of narratives um it's hard to really um to to to get a good grasp on it I'd say because there are a lot of moving parts from where you sit in the indust what what are you most excited about at that intersection we obviously know like you know one thing you're obviously excited about is grasp like what are some other sub sectors that you find interesting yeah um some of the things I find pretty cool are actually um some of the training benchmarks that are being evaluated in decentralized ways um as you know there's so many uh I guess

models that are being overfitted to certain benchmarks like a new Benchmark comes out and then all of a sudden like the next day you have 200 models with 99% accuracy and then you try doing anything else with them and they're kind of useless um I've heard of a few crypto projects that are trying to address this problem I think that's pretty interesting actually getting humans to go and interact with models and score them um and having some incentive structure for that uh that actually kind of closes the feedback loop for open source llms that actually didn't exist for them previously but did exist for closed llms where they can just hoard the data that is provided by users interacting with them um another one I find pretty interesting is actually decentralized

inference um it's one of those things where if you have a model that's small enough or even a large model and the sharding mechanics are properly figured out um anyone can go and monetize their Laten GPU I think that's pretty interesting one thing about crypto AI though that I I I see a lot of people talking about it as the stack so like you need this entire decentralized stack from like start to finish in many cases I actually find that a little bit over the top um so you don't always need to decentralize everything and in many cases especially in AI um by Distributing a system you're actually making it less

performant and one pattern I've noticed has been if you're decentralizing something U and you're Target audience is you know developers and but but the reason you're decentralizing is just for the sake of decentralizing there's a good chance it'll be very difficult to find pmf um so where I think the focus of decentralization in AI should really like be or where it should live is uh on user focused products because at the end of the day AI has been built on top of the internet and the internet wouldn't exist without the billions of users contrib buting it to it from day-to-day on a day-to-day basis um so from our perspective like that should be the core

focus of decentralizing things in AI if they're going to be decentralized that makes sense I um I know you just kind of said we don't need to decentralize all parts of the AI stack but this is something that I was wondering beforehand as well so I'm still going to ask it I um we have a lot of AI projects decentralized AI projects I want to understand how like because you are I would say grass is in its own sector where it's we have a lot of diff other types of AI projects SE projects versus grass is doing something quite unique um and taking a different approach I want to understand how is there like BD through that do you interact with other decentralized AI

projects and say hey like we can provide data how does that work or is it just open source is there like or is there I guess like more Partnerships not in the way of partnership but more collaboration wise is what I'm trying to say absolutely um so at the end of the day whether you're a decentralized project or not if you need web data grass is able to provide it um so if there's a decentralized compute protocol out there which for whatever reason you know let's say they're decentralizing inference and they want to add live context retrieval to their offering so like you know if people are hosting

models on their residential devices and they want to add the ability for that model to go and query things on the web um they could plug into grass the same way that a non- decentralized version would be able to um so the way we see it is like we don't really see the need to differentiate between them in terms of who grass is partnering with um as long as they're you know looking to accelerate the development of AI we're always more than happy to work with them that makes sense I um I want to bring it back the grass and away from this broad AI discussion you mentioned decentralized knowledge graphs I want to understand if that's something you're specifically working towards or if this is something

that is just happening as you guys build out data and crawl the internet yeah um I would say both really um so the what we find probably the coolest feature of grass is that it's your truest definition of of um Network effect where you know at 10,000 nodes grass wasn't really good for anything at a bit over two million nodes grass is able to scrape the entirety of you know Reddit let's say in 2024 and under a week and that's what got up you know that that's what got uh uploaded to hugging face in open source was a bit

less than a week of scraping um at 20 to 25 million nodes grass will actually be capable of scraping enough data to train Chad GPT 3.5 from scratch on a weekly basis and as I mentioned earlier at 100 million nodes grass will actually have the infrastructure necessary to build an internet scale web crawl and when you've got the ability to crawl the entire internet it would be really silly not to build a knowledge with that because that's that's how you really leverage that information is by creating a semantic tree that connects

all these different websites like you're creating a map of the whole internet that anyone or any model can go and use do you see adding more businesses to grass in the future um this decentralized knowledge graph is is for example one that I'm more so thinking of this in terms of okay we have data sets these essentialized Knowledge Graph um are there plans to do additional stuff because like you said if you have 100 million notes and you're scraping the internet yeah I mean we we touched quite a bit on realtime data um that's something that grass is capable of right now um where you can plug any model into grass and without grass it would start

getting rate limited and blocked for browsing the web and an automated way um and grass enables it uh as far as other business models I mean as I mentioned two companies in the world can crawl the whole internet and those companies are now multi- trillion dollar businesses and if you look at them as companies they're what each of them you can probably make the argument that there are at least a hundred businesses inside of those companies um so uh I think it's inevitable that grass continues to verticalized into other fields uh but but in order to be able to do that uh it needs this crawl as a base

layer I was going to ask this before but Ryan kind of interrupted my questioning I um why did you guys decide to go on Solana I mean the throughput reasons probably but anything else throughput reasons were the biggest one uh but aside from that um we've been talking to salana Foundation for two years now they've been incredibly supportive and the ecosystem that lives on Solana has been amazing for distribution and for bootstrapping the network um I don't think there's a single L1 ecosystem out there that can actually compete when it comes to things like that so uh yeah it it just made a lot of sense for us to do

it that that's interesting um when when you when you hear this um when you hear conversations about like L1 ecosystem distribution like it's people Point very obviously to the evm ecosystem like could you maybe dive into that um a little deeper yeah well when it comes to some of the most notable and prominent projects on Salata what they're trying to do is fix real world problems and use crypto in order to enable like solving those problems and most of the users of those platforms end up very naturally becoming users of grass as well because of the

very similar philosophy of we're not decentralizing things for the sake of decentralizing things it's a much more practical and not philosophical approach um I you know in terms of the evm ecosystems uh what we're seeing is actually quite a bit of fragmentation across a lot of these communities in the evm space you just don't see the same types of projects that are happening on salana and and the cool thing about salana is the fact that it's integrated everything exists in one place um we can focus on building grass and not have to worry about having to rewrite some consensus mechanism as a lot of projects in the evm space have had to do in order to make them work properly yeah that's

that's what I've noticed with the cosmos as well a lot of people have to just rewrite common BFD um just to get it to perform to a to the level that they need um what's next for you guys like over the next let's say year what's what's coming up what should people look forward to yeah so one of the more exciting and most asked for features within the community is a broader mobile roll out um as you guys probably know we've been live on the Saga uh for about seven or eight months now and the goal is in the next couple of months to actually roll out to all Android phones we're expecting to see a ton of new

users joining the network from that beyond that um in the next four to six months we do plan on releasing data labeling as we mentioned earlier as well as doing a more Enterprise level roll out of live context retrieval so that's something I you know I can kind of touch on right now but um I think showing is a lot better than saying so just keep your eyes peeled for that perfect Ryan do you have anything you want to ask no I love it showing is better yeah all right um Andre thank you for coming on yeah thanks for having me guys are you currently building on

polygon or interested in migrating season one of the polygon Community grants program is live right now and features 35 million in madic to support the next generation of polygon projects builders on polygon these grants are for you join the aggregated future today by applying at polygon.net [Music]