Sahara AI decentralizes artificial intelligence with verifiable data ownership

0:01I'm Ashan Addison from the Cryptocoin Show and today on blockchain interviews we have Sean Ren, co-founder and CEO of Sahara AI. Here to talk about the intersection of AI and blockchain, decentralized AI and why it's important and much much more. Sean, welcome to the show and thanks for taking the time. Thanks a for having me. Glad to be here. Yeah, AI is growing so quickly. I need

0:27experts like yourself to keep us up to date on uh where it's heading especially with the intersection of blockchain and what value that can provide to have both of those together. I would love to kick off our conversation by first of all hearing a little bit about yourself and what your team have done at Sahara AI so far and then we can dive into all the

0:46latest topics. Yeah, totally. Um my uh I'm Sean Yen. I'm the co-founder and CEO of Sahara AI. My personal background is kind of an interesting mix of uh research product and technology transfer. So I've been working in AI for the past like 15 years starting even just like a senior in undergrads that I really got into this machine learning concept which later

1:10everyone called it AI or deep learning. Um so I uh basically started um uh working as in my PhD uh career the sort of a problem of like what we call knowledge extractions or information extraction which is like creating structured uh representation of human knowledge of the world from the massive unstructured data sits on the internet right and that is like a really um

1:38intriggering problem to help me understand oh uh once we actually collect and construct all this like representation of the knowledge. What if we can help AI models learn about that? Right? So that basically pushed me further into the later of my career where I spent time at Stanford as a postar and later become an associate professor at University of Southern

2:01California working with over 20 PhD students on various uh fronts of AI including how to make AI cheaper to be customized for various business usage and how to make AI safer to be deployed while preserving user data security and protecting user privacy. while being able to properly make use of this personal data to improve the model capability and it's exactly during my

2:30tenure at USC especially uh in late 2022 I uh have a lot of discussions with my co-founder Tyler who is a former investor of finance lab and taught me a lot of concept from blockchain and crypto where I start realizing hey uh AI is not only about just productivity, right? Not just about how to improve the capability of the models and the systems to uh be better than human. It's also

3:00about how to facilitate us to build up a more sustainable and healthier ecosystem where everyone that have interacted with AI either in the way of contributing their own data to improving AI or interacting with AI by providing feedbacks or like providing good model recipes for a better AI architectures or like providing GPU compute or training AI.

3:28All of them is part of ecosystem and supposed to get their own sort of uh fair compensation for what they have done in the process of improving AI and this is not how today's web two uh AI landscapes looks like. So that was really um u concerning but also very intriguing problem to me that we start uh thinking about how to use blockchain to build a platform where on top of that

3:57platform if you create AI data sets or models or agents or any kind of applications that are powered by AI in that process we can sort of have a open ledger type of a copyright system to automat automatically to record and trace everyone's contribution when the AI is being created. If that could happen, this is a very powerful provenence structure that we can use it

4:26to enable some new business models among these participants. For example, I can split the model uh into multiple shares that are owned by different parties. The model developers own some of them, the compute providers own some of them, the uh data contributors to the model own some of them. And this is um recorded on trend uh immutable, transparent and

4:52auditable by anyone. And you can just use smart contract to autonomously attribute every single dollar made by the model or the application powered by the model to all these upstream contributors. uh and this is what we call a sort of a new revenue sharing mechanisms powered by such a provenence structure and we really want to sort of realize this vision and that's where we start the

5:16whole Sahara AI platform in order to build towards this open and collaborative AI ecosystem. Wow, that's really cool, Sean. and being in AI before it is even called AI you know in machine learning and deep learning uh you obviously have uh a strong expertise in how AI has been developed and it's probably been you know most people they've only followed

5:40in the last two three years or so when when chatpt really broke you know millions of into millions of computers in the first days of release and now it's open to the world and this is something that you know it's like being in Bitcoin in 2010 like you were there you've seen it develop uh but on the AI side and um I I'm glad that you are incorporating the the principles of of

6:05blockchain in in a sense that that fair compensation model that you're talking about because if not it's just version 2.0 0 of the technocracy of the search engines that are using all everyone's input to almost in a way using it against them to sell them more or to build more power uh in a in a centralized structure and um there I think AI is even better at collecting

6:31and analyzing and using data than the search engines were um so that would make this problem even even bigger of a problem. Um, so, uh, I would love to hear more on how exactly that works, uh, in building a fair compensation model and like decentralized AI. So, if I was a user that went to ChatBT and I've used that before and then I see Sahara AI,

6:59does that look different from the front end? If I go there and I'm like doing inferencing, uh, what is the main difference from a user perspective? Yeah. Um, great question. So um Sahara AI is having uh multiple different target users right um for a user of AI um the the the first principle is very simple it's just like frictionless right it's just like using any other AI apps

7:28you didn't notice anything too different uh but for developers uh and builders who own the AIS things works different because they now start seeing Oh, I have this like account and building that I sees this revenues come in for every single different access I co-owned with others and I got maybe one 10% split on that model that I contribute and then

7:56another 20% split on the data sets that I built and so on so forth right and this is like a continuous revenue streams as the other models that you contribute to are being deployed and used by the others Right. So just to give a little bit more context on how this is actually uh operationalized on our products. Right. So if at the very high level uh we can see Sahara AI

8:23ecosystem has two layers. Uh on the bottom layer is what we call the blockchain infrastructure layer. Uh we created Sahara blockchain that basically serve as the prominence infrastructure as I mentioned like a copyright system for the plat for the entire platform. uh that really monitor every single important events and actions happens in the process of developing AI agents and

8:46models. Uh it's a purpose uh layer one blockchain for AI. It's EVM compatible. Uh it's specifically we are building a lot of AI native pre-ompile functions and access management pro protocols that specialize for the purpose I was mentioning. And on top of this blockchain layer we have the application layer. This is the actually our mode compared to many other uh crypto AI

9:13infra projects uh that no one else in the entire industry has. We create two natively building an enterprisegraded applications. One is called Sahara data service platform. It's basically a decentralized data collection enabling uh platform leveraging crypto incentives. Everyone from um are able to select some of the projects that they have expertise. For example, listening

9:37to a audio, trying to write down the text of it, looking at a video, write a very like detailed and thorough descriptions of what's going on in the video, even taking pictures of something that in your real life or like just like recording part of your uh screens when you are operating on certain softwares that being very useful for training agents uh in in the in the AI industry,

10:02right? many types of data, many types of task, many kinds of projects sitting on them. We have a network of over 500,000s uh g online gig workers or data service pro providers are actually using this platform. We have a daily active users around 200,000s right now sitting on the platform. This is only doing a wide listed phase not even like a general

10:23access as access yet. And then the other um key application we have is called Sahar AI developer platform. uh basically the one-stop shop for the end tom process of AI developments provide all of the developer tools and services for the AI constructions or the AI life cycle anyone uh whether you are tech techie or non- techy person you can create and bring AI models to their own

10:48use cases and also further monetize them um through the process I mentioned right you can publish them on a marketplace where other people's can browse discover and then purchase the license to use it and you can further even tokenize it into a sort of a AI access token where the DeFi ecosystem can come in and invest on those tokens and you have all the upsides if

11:16you you are actually a very valuable and sort of continuously growing good AI access then you have more liquidity coming to your access that you can use them to build the next iterations of the models right this is really like a onch chain uh proprietary version of hugging face uh with the provenence supported by our blockchain platform. Yeah. So I would say this is kind of like an

11:40overview of what we are offering today. That's really cool and yeah I would love to learn more about the tokenization part of it how how tokens can fit in there. blockchain can provide that transparency, trust, security, ensuring that the developers are getting what they rightfully need and that ownership can reside uh with them or the credits for building. Um where is that where is

12:05this at with the with the blockchain and the crypto side of it? Is that already integrated or um are the the starting models and can people go right now and use the models with or without blockchain? Yeah. um our data service platform launched uh almost like a three months ago. We've been go to market with the with it uh on enterprise clients even started like a year ago working with uh

12:31very big logos like Microsoft, Amazon um Snapchat, Capture AI uh many schools research labs creating over five million uh uh revenues within just six months of time of total market with enterprise client and then we now gradually open this up to all of the uh retail users. uh starting with a wireless of 200,000 uh having a very high active user ratio

12:59probably in the next uh couple months we'll make it it generally accessible to anyone there on the AI developer platform side we start beta testing about like a month ago uh we have uh uh dozens of initial beta testers giving very good feedbacks on the tools and um we are to we are to going to launch part of the developer platform months. For example, the model uh uploading, data

13:26set uploading, registration on trend park uh probably done uh down the month, maybe early next month. Um and then we'll open up the marketplace uh for everyone to come to discover and browse and publish data sets. And then we're going to gradually open up everything from a developer platform and everything we open up to Publix will be already uh on trend right for example if

13:53you upload a data set or model the last steps of the uploading process is to register on trend and then you will actually have a ownership NFT token that you have to claim and it could be a co-ownership token as well that you co-owned with you have your NFT which is associated with a bunch of other that jointly owned a data set or model. Um and then down the road this will be then

14:22really integrate with our licensing tokens that shows how usage comes into the access how revenue flow in the access and then you follow the ownership tokens to split the revenues to attribute the revenue back to all of the contributors. That's like roughly how it's going to operate down the road. It's very cool. I'm looking forward to the full uh public access wi with the

14:47blockchain crypto integrations. I'll be jumping on. It sounds like, you know, the the white list and the the daily active users as you mentioned, it's already overs subscribed. There's a huge demand for this. Um and with those partnerships that you mentioned, you know, I saw there's Microsoft, Amazon, you mentioned 5 million in revenue as well. That's incredible. Is that through

15:07the dev environments uh on those platforms or is that with the organization? Um how did you guys manage and how do you manage those huge partnerships? Yeah. Yeah. I mean for uh just the $5 million revenue is just for the data service platform alone. We've been using these tools to actually work on some of the very challenging AI data projects from these enterprise clients right on

15:32different modality and uh use cases and that's been very successful because uh number one we have very competitive price because we have a huge network of um online gig workers that willing to uh uh take sort of like attractive pay to them because they come from different regions of the And then number two, because of the huge network, we cover a wide range of

15:59different expertise that rarely um exists in the existing web two uh data service companies like scale AI. Even we managed to really like compete with like scale or multiple of these enterprise client projects to deliver both high quality and very fast-paced results. uh but on top of this the developer platform we have also done successful go to market with enterprise clients like

16:24for example uh monison is a public company in India doing like a car manufacturing they want to use agent view from this developer platform for various use cases such as steady design like uh supply chain optimization dashboard monitoring legal document reviews and so on. So we start building uh in a sequence various agents for their use cases using our develop

16:50platform while recording the sort of the provenence of them. Um uh and and and the main reasons is still like number one because we have a very strong AI native development team here in the projects that even just creating web two AI projects. We are very competitive on the landscapes but really the mode for us today is to adding this prominence and adding this liquidity on the access

17:17token that really bring a huge difference to the web to developers because there's no way today for web developers to monetize their models and data sets. It doesn't happens on hugging face or github because these are all open source platforms right we have to build a proprietary privacy protect platform for them plus introducing liquidity come from the web three retail

17:41side uh so you get best of both worlds definitely obviously that's important to make it worthwhile for for the developers and I saw on you know on X openai CEO was talking about the fact that you know people are really as same here you're the the users are using the platform a lot once they have access it's like daily users like everyone's the daily user um and for them it was

18:09like they're almost using it too much you know he made a statement like can you guys tone it back like first we're we're losing money and second like the you know once you add in the graphic capabilities and audio video people are like doing it all day and like they're actually losing money. So does the blockchain economics and the supply maybe having like decentralized compute

18:30does that change things and give an advantage to decentralized AI like Sahara and that you know the comp there's no company itself that's going to be losing money that the supply comes from the the users itself. Um yeah I I think the I think to answer your question maybe uh better way to understand this is to first see what happens on web two right I think on web

18:56two AI landscape the problem is all of the revenues and monetization opportunities are controlled by the app builders like simply put like this tech giants right like a meta google open aai they are the application builder they touch on the applica the the application users which directly give them revenue And then once they have this like huge revenue, they can they can have a

19:19dominant sort of a say on how much I'm going to pay for the GPU compute, how much I'm going to pay to hire these like AI talents. They control the entire landscapes. So there's not much the rest of the ecosystem can say because they don't have the revenue directly versus the sort of the paradigm that Sahara AI is trying to create is a very open and transparent

19:46monetization architecture, right? If you own something or you co-own something, the revenue going to flow to you autonomously through the smart contract. there's no like centralized accounting, centralized like um bookkeeping going on that like stop you earning money and then that's a huge trend and I think number two the huge difference is the liquidity today if you are a great AI

20:12developers you want to monetize yourself the best way is to get hired by open AI get the $1 million package with the equity and that's probably the best upside you could have. But then what if you can just like tokenize your models and data sets and get people invest into those tokens, bring the liquidity onto those tokens. I think that's like the two huge opportunities doesn't exist for

20:39the web two people. Yeah. No, I'm really excited to see that liquidity side and the tokenization done in a proper way. Um and so if for example, I'm not a developer. I know there's a lot of viewers that aren't developers as well. they're always looking for uh great AI platforms, but also revenue opportunities are nice. So for non-developers um the there's different ways that they

21:04could potentially contribute and earn as well, right? So what are the what are the different ways that uh regular business users can uh earn while using or contributing in the platform? Um today we are seeing two types of retail user. One type of retail research are more like this online gate workers, right? They spend time online. They they do work. They create good data sets.

21:29They label data sets. We have 5,000 uh 500 thousands of these users very actively working on our data service platform. There's more to come once we open up the data service platform for anyone to open up data project. So there will be more project available for them. And on the other side, if you are a D5 retail user that you been investing into, say MC coins or

21:53all sort of like opportunity um space, then now you have an entirely new brand new space where you see a bunch of data sets and models being tokenized. And the most different things from Mingcoin is like you know the fundamental of these access, right? You know how much API calls been on these models. So you know that it's popular, it's actually making

22:15more revenue than the other models and based on that you can speculate where the coins is going to go and that's very different from the Minkcoin market. Definitely. Yeah, it's good to have some underlying fundamental value in this and and with all the active users in using AI every day, I'm sure it's going to grow significantly once it opens to the

22:37public uh as well. I'm excited to see how that liquidity aspect works. Is there any uh obviously that's a big part of the road map. Is there any timeline on when people can expect to see that? Yeah, you can um already apply for wireless access to our data source platforms and the developer platforms today and we are opening up the wireless every uh two three weeks. So uh stay

23:03tuned on there. You can find the uh links on our websites which uh as to gonna attach to the podcast links and then follow our Twitter. Everything will be announced there. Down the road we're going to uh launch our AI marketplace. Everyone can come to here to upload your useful data sets, get them uh verified by the community and then earn incentives for publishing good data sets

23:28and you can use the incentives uh the Sahara points to also purchase good data sets for your own use cases and models as well. So uh stay tuned on uh our series of launches related to the AI marketplace and the developer platform. Sounds great, Sean. I'm looking forward to it. I really appreciate your insights into decentralized AI and building a platform for fair compensation for the

23:53developers, users, everyone involved uh and bringing those principles of blockchain transparency, smart contract ownership. If you build it, you should be rewarded uh no longer serving to the technocracy uh at least in part. So, I'm looking forward to those uh that the blockchain crypto liquidity uh add-ons in the coming months. I will be following along. I'll leave the in the

24:17show notes those social links and the platform links uh and anything else we need u so people can access that and would love to have you back on when we get closer to that uh crypto liquidity launch and wishing you and the team all the best in the meantime. Yeah, appreciate it Tom. Great chatting with you as