Cirus Foundation lets users earn by contributing data to AI

0:01I'm ashon Addison from the cryptocoin show and today on blockchain interviews with Michael lucku co-founder of Cirrus back to talk about wallets AI blockchain the merging of AI and blockchain and how can we create synergies between these two industries Revenue generating opportunities and keep up with the insane demand that AI is asking for uh this year and over the next few years so

0:29thank you so much for taking the time Michael thanks Ashton good to be back here likewise it's been a few times that cus has come on and the platform continues to grow I'd love to hear a little bit about the latest on what you and your team are building at Cirrus and then we can dive into all those topics definitely yeah it's been a it's been a wild ride uh to say the least um

0:51we started Sirus with the concept that data is monetary value it's the currency of the internet and at that time 2021 um it was a new topic people were a little bit afraid of it um you know where is this data going why you know how is it being used and we have to do a lot of convincing to say that this data has actually been used in lar large parts to building these platforms and

1:16advertising companies and so forth uh fast forward to today I think people are realizing that this data has actually given rise to some super intelligence which is AI and over that past eight years this has been happening behind the scenes been scraping public data from every Source uh on the internet and now we have this beautiful thing called AI but people were really the makeup in

1:43creating it and and that's what's interesting now I think the topic has become easier because people are making more relevance around what data actually gave power to so with Cirrus um we've become a little bit more relevant than previously because now we can kind of say well look at the products that are being built around AI based on your data how do we now transform you to be part

2:07of that process and what can we do next so that's kind of the long and short where we are but it's it's been a whirlwind to say the least I love that and I think a lot of people understood that data is important but taking data and turning it into information or things that can be used that's where AI really comes in handy and I think that's like that was the missing key that maybe

2:28in 2021 people didn't iiz the full potential of how data can be transformed to be used so efficiently um and you guys have been starting as a a blockchain based project but focusing on data now that's emerging into okay AI is the one that's using this data for many many different purposes I want to start off this part with a question on why do you think AI needs blockchain or where

2:58is that Synergy between the two industries that's a that's a great point I think the the concept of where AI right now is is there's a few things that are hard to find let's just say just in the traditional AI side we know that there is going to be a shortage of compute if and when that happens uh there's definitely a supply chain issue we know that it's comput is in high demand even

3:24though they're making the models a little bit smarter and a little bit more easier to distill um and what we do know is that layered with that there's two elements there's compute and there's data data is being harder to find so they're trying to unlock new types of data especially from offline uh autonomous driving you know physical robots understanding layouts for uh

3:45factories and understanding what that data looks like because a lot of the public information from the internet has been already scraped so how do they unlock the next uh version of data where can they find that so those two components are pretty critical the the the great part is in the world of blockchain and decentralization those two Commodities uh live and they're

4:05abundant um so we do have compute around the world whether it's the laptops we are we're on right now or gpus like an RTX 490 or whatever gamer is using at that point in time and the interesting part is just like a peer-to-peer system those resources can come to use into this Global uh platform where you and I can share those resources so if you're a

4:28model builder you're a Creator in some form of of AI you can ask the network or ask me to say can you provide GPU compute or rendering services like some of the platforms you see now and we can share that resource so instead of you going to AWS or gcp um there's a there's a abundant resources as long as we can tie that in so blockchain is a mechanism

4:50to broadcast to find those resources to record where it's being allocated and to share the value uh layer between the two which is the cryptoeconomic layer so there there's a lot there that we can really tap into that the world hasn't really seen yet MH I think a lot of people understand that there is a high demand and it's probably going to be uh there's an even higher demand as the AI

5:14gets bigger and requires more processing to process all of the information in the world do you think that with blockchain mobile phones laptops graphics cards on your gaming computer in the corner of the room is that that computing power enough or a a decent chunk that will allow people to get involved with Computing and Ai and sort of getting a little piece of that pie that that AWS

5:42and gcp have yeah that's a big topic because what we've traditionally seen these big models these large Lang models had to be built on very Advanced Computers like h100 you seen Elon Musk build a a factory of these 100,000 h100s those are hard to come by those are not easy gpus to purchase for the average person uh that's specifically used for model

6:05training so model training is heavily uh is heavy on compute resources and you need low latency so you need Super clusters to all come together in unison to basically power these train these models um what we've seen though that these large language models are becoming more efficient and they're becoming smaller in in in parameter size so from 400 billion down to like 1 billion 2

6:27billion so a lot of those smaller models can actually on device on edge essentially being able to perform some of those compute processes right from your phone or from your laptop so we're finding that the energy required to actually not just train but to actually perform inference is uh becoming a much more efficient so we're now starting to see new compute Hardware come out that's

6:49a Le less costly um I think Nvidia just launched one um I forgot the exact name and they're being able to be more consumer friendly so you can actually store a large language model directly on your device that's interesting because not only I can use it but I can share that with you so you can access it when I'm not you know in use so that the concept around compute and GPU resources

7:11are becoming more ubiquitous I would say um And in regards to not just training but inference for like structuring data and so forth um we don't necessarily have to work in unison as long as we're providing pieces of that data back to the original uh repo uh we can do it asynchronous that's kind of the word um so there there long short it is becoming much more efficient and the use

7:34usage of our compute can be uh a part of that system as well that's really cool and I want to dive into that part more about the data side uh structuring so say for example we do set up a blockchain network and we want to contribute to Growing this AI Monster uh what exactly are we doing with our machines interacting with using the data to

8:01interact yeah so we think that I mean there the two big parts for the AI it's the training um it's also how you provide and serve that model it's inference and how you provide reinforced um human learning feedback so the idea is like once it's been done how do we provide better feedback to the system how does a human now provide some level of intervention and we think in

8:24those three parts we can serve different areas specifically um like I touched on the training side is difficult uh it's not something that's as efficient we've we've seen a couple examples of that distributed training in our system I think uh like Prime intellect and noose that have done it a very small scale there's a long way to go there um but we think that inference

8:45especially we can use some of our our power to like ask the GPU or ask the the AI hey I have this piece of data I need to be structured and annotated and labeled in a certain way instead of me having the tools like the ETL traditionally data you would have to have uh extract Transformer load tools instead of me having all that I can basically place the request with AI they

9:06could format it specifically how I want it and then send it back to the main uh to the main repo so we're now instead of the user having to have all these different specific tools and being part and commit these these resources as long as I can call on that AI to help you know pass that information through it's much more achievable it's what we call

9:24uh structuring so that's one part uh we can become you know subsidiary I think of sending data making that data more real time um another components called rag the idea is you can make this data much more efficient much more real time sending that reinforced learning uh and once that model is built people can look at that and annotate and label that data and

9:49labeling very very simply put is just human intervention human eyes looking at that piece of information saying yes this is accurate no it's not it can be an image it can be piece of you know written text it doesn't matter as long as that is being done and that's a very high demand sector um right now so yeah there's a lot of ways we can play into it it's really interesting and with so

10:14there's two parts there there's being able to tap in and have your bandwidth and you know your your processing power in more of a passive kind of way where like you don't really have to do anything you can connect your device and you don't have to be working 8 hours a day to be able to contribute and then there's the labeling and structuring where you actually have to do more

10:37active work but the rewards could be greater in that yeah exactly that uh when you're not using your computer we call this sort of like system standby right it's lat and compute it's really interesting there's actually um this has been done before in traditional science uh there was open- Source uh work by Berkeley University for example uh project called

10:59b o i n c and it was just people as volunteers plugging their computers at home their desktops at home and providing um standby or Laten compute to these research facilities and all the research facilities would do would you would download a Docker image to your uh to your desktop and then you would connect to their Network and they would ask you for specific casts like your

11:22computer would fold protein uh it would look at um telegraphic imagery and then provideed a labeling to that data set and send the back and the best you know um it would take a it would actually take an aggregate of different types of data sets and see what was the most accurate so folding protein and these things have been traditionally done it's it's really interesting because this is

11:42what's coming back into our space I'm sure we'll talk about uh this category called decentralized science so we were able to do that passively now and actually earn a crypto reward uh from that um but yeah exactly you said we can supply that information the labeling side is a little bit more human intervention so it should technically um equate to more value um traditionally

12:04labelers that are built uh doing this overseas they make it anywhere from a dollar to $6 per hour on providing that information um scale AI is a great comparison they make $760 million in Revenue a year just from labeling services for AI so it's a great B comparison um but we do definitely see there's a multiple areas people can be part of the data system at this point

12:30that's really um interesting about this the science part I I do recall that way back in the day I think it was so novel that people realized their computers could contribute to something passively that they're just doing it but there was no reward um essentially except besides the goodness of your heart that you're you're helping science U but I have seen

12:50this notion of decentralized science or Desi in the last year through the crypto industry I don't think it's really taken shape maybe AI could play a big role in that um but I've heard the term thrown around but there is no really big names that are actually helping to do this and I feel like the missing key is that crypto Rewards or blockchain integration

13:13where people can contribute either passively or be actually getting something for it because nowadays people are so busy um at least if you're doing it passively you can earn uh while you're sleeping or if you're doing it actively um it's something that you need to be rewarded for nowadays totally um I think down it's it's a definitely an emerging sector in our

13:34field um decentral sign I think there's a company called bio bio protocol that's just launched what they're typically doing right now in daa is just setting up a dow so anybody can contribute to that Dow if they use those funds for major research and there's a win inside of that research they would pay back the recipients uh of the Dow a reward and You' be a contributor and you would get

13:55some type of profit um I think that's step one but I think step two to would like we said is providing the latent compute and the compute resources that the science companies may actually need and uh pass back some of that that income stream or that Revenue to to the user so we can supply data that is Noel we can perform work like fold and proteins uh and we can also even Supply

14:18funding to these DS as well so the user can be part of that entire ecosystem uh for decentralized science I think that's fantastic I think it's it's more practical use of how um you know the world of decentralization is coming to use for a better good uh so I find it I find it super interesting definitely and the the notion of dows taking over companies and and being decentralized

14:43autonomous organizations in themselves has been around for a few years but I feel like there's so many moving parts that it's it's hard for people to know where to start or where to focus on and I feel like smaller specific tasks like this or small initiatives are are a great way to uh start figuring out how exactly this will work in a decentralized manner and and how the

15:07governance works on it um how do you see people actually you know without having to understand all the technical jargon and barriers to entry can people go through just their browser or Chrome extension to start getting involved in a data da or something that's a DA but contributing to to dii or AI yeah you know and that's interesting point you made I think typically

15:36speaking Dallas have not taken off they haven't really been truly adopted I think a large part of that was this idea that there was governance there was a voting mechanism there was a snapshot you know and hopefully you provided some change to this protocol um most of those votes would go back to management the management would say okay yeah we can do

15:57this or not and like nothing really happened or whatever um where we think a dow stands to really take off is that it's truly automated so if you think about a data Dow very simply put uh let's just think of it as a a Google folder it's a folder where everybody's it say 100 gigabytes and you're submitting data to this folder and if I submitted a one gigabyte I own

16:211% of this folder I own 1% of this this um Dow essentially and every time that Dow is being interrogated whether it's uh someone using it for research whether it's somebody building a new model whether it's um supplying some form of sh chain to another system where basically needs more relevant data uh we can see that there's a there's a record and a receipt so that Dow all that Dow

16:46is really doing that drive that Google Drive is making that return on that data more sustainable so it's not onetime use it's a renewable resource so every time that folder is being accessed there's an access of of payment um we're basically saying that the recipients who own a percentage of that can get paid um and so it becomes the the interesting part

17:08about data and these these these uh assets is that is actually not just one time use it's renewable so we can share that down into multiple different areas and say hey do you want to access you know the point so I think this is more sustainable ways that Dows can be actually integrated and and we can adopt it in a much better way um but typically speaking people wouldn't have to do that

17:29much as long as you're connected to the system like with Cirrus we want to take that friction off of the user we want to actually plug into these Dows plug into these systems so anytime information is being done the user can be hands off they can just earn they know exactly what uh project they're supporting uh what network they're they're part of and

17:47it's more handsfree yeah yeah I'd love to dive into that uh the sort of the original vision of of the Cirrus wallet and then how AI is fitting into that because from what I understand in our previous conversations you know there's thousands and thousands of people that have just using the the serious Chrome extension uh so you really just download the extension quick and you have a

18:11wallet and then you're earning from the data that's there but you're not giving up your personal information per se and it's all passively done but it wasn't towards you know it was before you know open AI was even around and it wasn't for llm machines so how is this evolving into contributing into AI more yeah definitely and like I said like the the the first part was how do

18:38we provide insights how do we gain that piece of data make sure that we can re responsible and and you utilize and monetize on a traditional data market and people get paid so the first key component with Cirrus was insights right so we understand browsing traffic history of people like building that data set essentially um the second third part is what we're moving into this year

19:02is having the user more integrally involved so when you're sharing information you have to be online you have to be surfing the web that might be three hours four hours 5 hours a day but what can you do after those uh hours so entire entirety the 24 hours you're earning something so the second component I think we touched on was structuring so that's simply having your

19:23computer connected to the system will'll send you a piece of data a micro piece of data you can Str structure it annotate label it just from the use of your computer because it's set up with AI call it passes that information through so now we have one giant data set that we can present uh that might be for a third-party um uh project uh for example right now we have a automotive

19:48um logistics company that's basically taking a lot of traffic data and imagery from the car and they need that data structure because it's really raw and that needs that needs two power is a structure so we can basically say okay well here's our user base it's a decentralized network that is providing latent compute on average right now that would be you know anywhere from 8 to 10

20:11cents per hour so in the equation of our of our system we have couple hundred, worth of compute at any given point in time um that's step number two that's what we're really focus on testing right now uh step number three would be looking at the labeling um components and all that would happen is that a user would see in their wallet there's a notification here's an image or here's

20:32something on a dashboard to go and check you're almost like a QA service to what the data to structuring the data and you're saying okay this makes sense here's an image here's how I've labeled it and that's another form of Revenue the user can be involved in any three of those points or just one uh we want to be more comprehensive so we can at least

20:51offer all of these different tools with Sirus so that you don't have to go anywhere else um and the the great part is the wallet would be able to direct directly receive and earn that income directly to that wallet so you don't have to change that or move somewhere else yeah yeah I know that makes sense to have the wallet in there and and from what I see you can have different coins

21:13different chains um is there a uh in terms of how that would go in contributing and then receiving do those funds actually come from for example that automotive company or you know where's the money being generated from for the revenue for the power that you're giving to the AI yeah well now the Partnerships that we're looking at right now I think this year is for like

21:38for Sirus we really need to focus on meaningful Partnerships that we can become a part of that supply chain of that data system you know whether it's structuring pre-processing labeling um most of these projects right now that we're we're speaking to already have their own former token that's how they're monetizing that's how they're they're they're giving back users so uh

22:01we can actually embed that new token into our wallet so that the reward correctly comes from that existing project whether it's a a dow like Vanna or whatever project we may work with um they can see that signature token from that project um when we step outside of the box of crypto then we'll look at direct Revenue um with projects that are like traditional companies are saying

22:24hey we have this piece of you know data that need needs to be labeled this is direct Fiat involvement you know and we can look at that we haven't explored that directly yet but we know that there's a lot of overflow and demand for labeling um a close company that I know is called sapen they're getting requests pretty much by the day um from traditional companies so that's a that's

22:46I think sapen would be a really good interesting um project for us to communicate with because we can at least support some level of of labeling outside of you know the norm that they get yeah yeah no that's that's a great to hear because I think this is so much bigger than crypto you know crypto's continuing to grow dramatically but uh every you know open AI had few million

23:11users in in the first day it was a fastest growing startup from day one and I think people are still figuring out how ex where that information is just coming from magically when they ask a question um and if they can tap into those Revenue Jer nating opportunities and if they're not in crypto yet um Fiat rewards are great but being able to tie that into a wallet where they can see

23:33Fiat rewards and then other stable coins and other crypto and hey I can contribute to this new company and they already have a token and I'm and earning that and I'm getting in at the beginning maybe it's the next open AI or something like that um so I think that the the reward potential is there um and it's very exciting and I actually haven't heard of this before at all where where

23:56it's like my crypto wallet you know traditionally people are have been earning rewards through staking their assets and they're earning a few percent um so like yeah I earn from my crypto wallet but I've never heard of a crypto wallet that is contributing to Ai and earning from that so it's like my my crypto wallet's helping AI grow is like that sounds like the next level and

24:22logical step for most crypto wallets yeah I appreciate that Ashen it's you know this is this is this is going to be really interesting because I think there's going to be a world in the next X amount of years that there might actually be a Frontier Model or a big large Lin model that is actually built from the participation of people and that hasn't been done yet that's that's

24:45the amazing part there's you know four or five top llms to be an llm provider right now the stakes are massive you have to have billion dollars to probably enter entry into computes and to finding the new the new data unlock so I think it's going to be such a fun next few years to see how does the community get involved um you know a quick example is a a project I studied

25:09it's called poolside and they know that they don't have the compute to compare to xai or claw and so forth they've got $500 million which seems like a lot but they know they don't have enough so they're finding new data unlocks like from code bases of like repos on GitHub to say maybe that code actually gives them a key insight to get to AGI quicker

25:30and so they're basically trying to find scrape code bases specifically not data or text or imagery just really focusing on on on code and I think that's quite interesting so what if you know the users can say hey I've got these private repos in this code base or whatever and we want to con conscribe that to that Frontier Model in response I can get form of income from that large language

25:54model taking off I think that's really interesting and I think at least there's an that's alternative to traditional ways the frontier models are being built um we just need to get the word out there I think that decentralized world has been hidden a little bit and uh the more we talk about it do things like this we can at least get that that awareness out there definitely I love

26:13what you and the team are doing I hope we can get some more decentralized llm models or at least more options uh that that people can choose from and as they continue to grow there will probably be more small language models or uh things where if people want to specifically focus on their life's work on in one specific industry uh there's going to be

26:35even more information that that you know these llm companies are looking to tap into new information to grow there's so much information that hasn't yet got into the uh llms yet so I think there's definitely room there and now with the Cirrus wallet I know it's live right now but these different Ai iterations and and new Revenue gener generating opport

26:58unities are coming in so is there a a timeline for that or you know how can people follow along start earning and then as these new Revenue opportunities come in get involved with them Yeah we actually spent most of the last year taking the Deep dive in AI um a lot behind the scenes before we actually produced a product we wanted to try it for ourselves find the way we can

27:24actually even build a network that can do these functions and we did a lot of that heavy lift last last year we didn't come out with the product underneath the hood we just wanted to make sure we knew what was right how this model was built like all those things it was it was a a huge learning experience uh going forward into this year we want to come out with

27:42some of these functions we've already previously built and we want to tie it into the product um it's been a longer slower process I know a lot of people are kind of frustrated sometimes of our length of time to get a product out there um but we really want to make it beautiful we want to make sure that it's an easy seamless experience and it ties in um with everything we're doing so

28:02from a timeline perspective uh the first thing is we're going to be testing uh traditional structuring models with an existing project to see what kind of feedback we can get how the system looks from a distributed level like where does everything kind of cohesively put together within the next few months and then we want to build out like a backend

28:22dashboard so people can quick on quick fire open their wallet see an expanded dashboard and and see all all the data points that are being committed into the system all the compute that's being supplied how much um we are enabling as a network uh to touch on that right now we're getting 2.4 2.4 million impressions per day from the users and we think these are valuable points we

28:45you know unfortunately our users don't even know how much they're actually sharing but we want to bring that to the Forefront uh that's our that's our real core Focus once we can do that then we can take the next steps and say hey there's other opportunities of how you can get involved and you can literally toggle on toggle off which ones you want the wallet will signal new events you

29:06know it won't just be a Red Orb maybe maybe a green or blue but you can see what your wallet is actually doing at that given point in time and then the last part in the middle of to later part of the year we'll we'll at least test and look at structuring I'm sorry uh labeling and see where we can draw on from that so uh long and short we've been building a lot uh I think in the

29:26next few months we'll be able to test and roll some of these prod at least get some functions out there and uh yeah I wish I could say more um but you know it's been it's been a slow progress but we we definitely are heading in the right direction yeah yeah I think you've said a lot um I learned a lot about there's a lot to know about Ai and and different ways that it gathers

29:49information uh transforms information and we can how we can earn from that uh and the potential opportunities and and how blockchain does play a role I'm excited to see those next steps for the cus wallet and and the revenue generation opportunities passive and active uh I'll be following along I can leave a link in the show notes to to the Cirus wallet extension and and the

30:13Sirrus website and the X and the other socials if people want to learn more as well uh wishing you and your team all the best Michael on getting uh these advancements out and let's contribute to Ai and put it in the right direction so thank you so much for taking the time thanks s this is awesome