OpenLedger lets you earn by contributing to AI language models

0:01I'm Ashen Addison from the cryptocoin show and today on blockchain interviews with have Rah core contributor of open Ledger here to talk about AI agents payable AI models the intersection of AI and blockchain and much more ROM thank you for taking the time to come on the show hey Ashton thanks for having me here I'm very glad um glad to see how this one goes yeah definitely I am super

0:27excited to learn more about AI agents and how the payable models work I've seen a lot about AI agents uh recently throughout Twitter sphere and across web 3 and I really feel like I need to learn more because I need AI to take over my workload so that I can relax more uh and I feel like that's a part of what this technology that's being built is meant

0:48for so I'm excited to dive into your insights into it I would love to start out by hearing a little bit about what your team has built at open Ledger and how that relates to that and then we can dive into all the details no absolutely uh maybe I can start with some introduction about us if that helps um open Ledger is a data block shet for AI uh we've been in the space building

1:12over a year right now um if you take a look at current market today right most of the models that we see which are being used for the current chatbots like CH gbt or other platforms out there they use a large model right they use an llm which is proprietary of their own or a foundation model but going forward when AI becomes massive and it's going to be

1:40used in every other sector uh presumably you know helping Ashton um do his chores as he as he tries to enjoy life right um you would want the AI to be um very domain specific right the model should have capabilities to execute certain task which are particular to that particular domain um so a large llm would not be surprised even it would have context it would be good enough but

2:09it would not really do the right job um in in the most precise manner right so that is where you know we you we bring in specialized models these specialized models will work along with larg nage models these are not a replacement to the existing large language models that are out there it works along with a larger language model where a larger language model is used to get context of

2:36uh what is a task it is and what's the input from a user and um a smaller or a specialized language model is much more precise and actually executes the task right and it can execute it at much lesser uh error uh lesser hallucinations and uh and precise outputs that's the overall idea and what we building is that right we building this platform where you can build like specialized

3:03models on top of us and the first step to that is to collect like specialized data so we have a data net on which people can start a topic around um anything that they believe is going to be useful to build a model um and then um that is then finetuned that specialized data is then fine tuned to build models on top of us which are hosted in our platforms which then can

3:25be consumed by apps and agents which is the agentic cay of offers where they would use these models which are very precise for example an agent if we wish to um you know want to be used for like trading like a trading agent it would need a model which has some very good insights about trading as a knowledge and then probably have some insights about like web3 Alpha right has some

3:50news about what's happening in hisry what has happened over the last 10 years and then probably an an additional rack to it to understand the current market so that you would need a model like that for you to actually probably invest a million dollars in that agent so that it can actually make you $10 million right because it needs to have all of that

4:08context to make very uh important decisions autonomously it's not a bot where you use it just to execute your task it's an actual agent which has its own brain to go ahead and do stuff for you to where you can actually make money so it needs precise model to give it instructions to do the right thing and that's what we will right we we build models which can help these much more

4:34purposeful agents compared to what agents are today which are more um just mem materials and it shouts out tweets um if you want to bring much more serious and purposeful agents you would need models which are going to be useful for that U so yes that's the platform we will we have the data layer the model layer and the agent layer and all of this brings together to build a whole AI

4:58ecosystem and um lot of people are building um various parts of this and we're building an entire ecosystem around that it's very cool ROM and I feel like I need to know more about the the slms the small models because chat gbt and the and the llms are are great but they can't do everything that we need them to do without our permission and asking them every time and following up and

5:26really we're still doing the work they're just helping out but I I I from what I understand the the slms will be able to do something more precisely especially if you're get in you know a specific Niche or an industry whether it's web3 and crypto or any of the uh specialized Industries there's so much information to know people go to school for 10 years uh to to study and learn

5:51from the best how can the llms know every single thing in the world and execute it perfectly at low cost is that right yeah so you I think it's it's a right example think of llms to be sort of like your generalized knowledge and like smaller or specialized models to be um you know very specific um models or like a like an intelligence which has a certain

6:16particular data fine tuned to it so it knows that very precisely it's an expert in that particular domain right so and that's how we've been using um you know products today right if you take a look at it um an OS is like an LM and you know an OS would provide you with the gentle task that is required and then you would have apps which do very specific tasks for you uh you would have

6:42and like if you take a look at search you will have like Google search to do like your basic search that is required and then you would have like very specific searches like search engines for restaurants search engines for your job to do very specific needs right so that's that's very similar how this Market also is evolved um as you rightly put llms are very

7:03gentle purpose and then there are specialized smaller models which do very specific work so how does that uh differentiate in building a small language model at open Ledger getting the the data and the information the input into the model um it from what I understand with uh open AI it sort of scrapes the internet and also uses the user contribut contributed inputs to

7:29learn from that and then get smarter but with getting expert information for example in in a small language model do you have to have the experts give all that information and and input it into the model or how does the data differentiate from the llm yeah so if you take a look at like large language models right as you rightly put they're trained on largely

7:52internet data and um you know they scrape the internet data using various tools that are there or use like common call to feed that as a knowledge to the current models that you see today and all of these models are um as you know has a very generic knowledge of the market and you know they further kind of make it better by getting human feedback to that so that it is it is more precise

8:21to what the actual uh user needs but in a smaller language model or a specialized language model you would not need that you I would not need like the entire intern data to go ahead and train a model because you're going to go ahead and use an existing llm fine tune that with very specific data so it's already has context it already has that inter data with it um what we do on top of

8:44that is to get the specialized data and the specialized data doesn't mean that it has to come from experts it could come from um proprietary firms like it could come from a business which has a particular proprietary data which they want to contribute and they want to monetize like think of this to be like internet or the early era of YouTube right so where when we when internet was

9:06open and you know people were just using it for entertainment like how um llms are used today where you just use it for like information seeking it it it's not very useful it's not really monetizable but once businesses start to contribute to this right that's where Commerce happens like if you take a look at in YouTube and pre monetization of YouTube

9:28it was it was this garbage content and like no one really cared about it like once it got monetized um YouTubers became a thing and then businesses started contributing to YouTube it's exclusive content right which they they knew that if they contribute they could monetize it and it became much more Purpose Driven it it was becoming much more useful for every other um you know

9:52needs of us and I think that's how models today will be once businesses come forward and contribute like proprietary data use propriety data you would get models that are very specific to a particular domain which which can be useful for example uh we have businesses that own IPS large amount of ips who are building an IP data n on top of us we have deepens coming forward and

10:16building a deepen data where they're contributing the deepen data which which are useful for like Vision models to be built right and we have um the ethereum community going forward and contributing solidity data to build a solidity model so this is where we see like it could be like a subject matter expert an individual or a business coming forward

10:36and contributing data to build models which are very purpose duen for a particular use case yeah there's so many large large businesses that have information that they probably wouldn't want available to everybody on something like open AI if I could ask about the the internal financials of IBM or Microsoft U obviously there should be a a specialized language model just for them

11:03where their proprietary information is available and they can be analyzed and the AI could help them be more efficient and profitable but without releasing that information into a generalized uh AI for everybody to use I feel like that's a great use case for these specialized models so what I would I would think is that that are do pass to this when businesses contribute they

11:26could contribute for the greater good uh so that you know they could contribute data which uh which is open which they can instead of their internal financials uh there is so much data that a business has which uh which they can monetize and other users can use that data to as an intelligence forsel right on on other hand if you know that's that's where we

11:48come in where we want businesses to contribute data which which is which can be accessed by others and which can be monetized on the other hand if a business wants to have a model forsel where they don't want to use use an existing model because they fear that you know they might the model get might get drained on the data that provide and that's the most that's the common fear

12:08with most of the Enterprise today right which are not using these closed systems like uh open AI is because they worried if you know the model just learns out of their uh business secrets and it just goes out to other other businesses which are in a very similar demography as them um I think for that there is a separate way of business itself you know they

12:29could go ahead and take an open source and then just start fine tuning that with their internal data and use it as their own local model right and that's how the businesses today are act you know thinking about this and we've spoken to multiple people on this where you know they would they would want to go ahead and take a lava model and then fine tune that with their own data and

12:51then start using it for their own purposes and that's those those are the two parts to this right and we are tackling the earlier one where we would want them open up and contribute data which they think can be useful for others and they would like to monetize that mhm so what are the the bottl next and and the the limits of growth and adoption for this right now why is it

13:13not already bigger um you mean like specialized models yeah just getting this into the hands of those Enterprises and then getting that information out to the public yeah I think uh AI has been um more speculative at least in the web3 industry uh in in the the traditional industry I think it has been more experiments people are just experimenting with it um

13:40Enterprises or small and medium businesses have just started to adopt it right only when when you see businesses starting to use it and making it part of their day-to-day lives I think you would have a need for Solutions like us to be rapidly growing and that is what is happening over the last 3 to 6 months right we've been seeing businesses constantly use um models and AI systems

14:05like for example in our overall workflow and productivity we use um AI models very often right right from writing content and like creating images and stuff like that I think as it goes rapidly as it becomes a common phenomenon where people use models um in basically AI within their within their productive systems I that is where you would see a rapid need

14:31for something like this um and I that's what is happening I think 2025 2025 is going to be that where you would see like widespread adoption of AI uh you would see AI is being used by the end consumer very rapidly where you know we would start using it and we might not even realize that we actually got got an addicted to it right so that's I think that's this year and if if that happens

14:55there would be need for something like that that we are building and we've already seen that happen we you're working with a couple of interesting projects um so I think this year is going to be much more much more than that it's going to be 10x of that it's very exciting uh to see where it's heading I I've seen a lot of growth and in the interest of AI agents and the

15:16agentic world uh recently so I I feel like this is now the time to be paying attention to this uh and and ROM where does the blockchain fit into the AI agents and the payable models and and how does it drive value to that part yeah so blockchain is is always there where you want to implement um code as a law right um where you want to inbuild a system which is completely

15:44trustless where anyone from any part of the world can actually use a system and and know that it's going to work fine right so I know that no one is like having a back door in this um I think that trust system needs to be built for a platform like us because you need to figure out a way where a business or a subject matter expert who has a proprietary data which

16:09they think is a gold mine um you know they could come forward and contribute that to us and then they could be someone who they don't know at all could build a model of of that and then um hoping that an agent or an an app will actually use that model and give back Revenue to the model creator and the data provider in order for a system like this to function you need a trustless

16:34system like a blockchain to connect all of these guys together and uh bring that um you know proof and verifiability on chain right for them to see and see what's happening and I think that's where we use um blockchain um you know it's it's more as to bring that kind of a tusler system and that's I think that's that's one of the most important and the central point of what is open

17:00um the proof of attribution that we do right we create something called a proof of attribution which is an onchain attribution on top of our own L2 where uh we basically showcase how a data provider's data influences a models output and how much influence was it and how much of that is attributed back to that data provider we show the entire Providence on shap right from the data

17:29colle ction to the model training to the output and how much it was attributed we show all of that on chain so that you could use that for rewards you could use that for Price Discovery you could you could figure out how much your your data was useful and what is priced for and you could also explain the model better right um you know data provider knows

17:49where it where the data was used and how much uh he's going to get paid for and then a model provider knows how his model is actually being used and and how much of data was was he that he collected what that he made use of was actually useful for him and then an app developer or agent developer knows which model is he using and what kind of data is he actually accessing right so that

18:14he knows whether it is bias or he knows the data source so that he can he knows the decision that the app or the agent makes is is solid it can be trusted so I think that is where we use blockchain to bring that um in a trustless ecosystem where anyone can come and work together and be sure that they get attributed back yeah I think that there's real

18:36value in that sort of reminds me of uh either you know like nft creators or artists where you're creating something you're adding that onto the blockchain through that trustless manner you will get paid for the value that you drive or the value that you've created with the data um or a Wikipedia and how it should be where if you contributed the information if other people want to use

18:58that and Iz it in AI models you should actually be getting rewarded through the blockchain um and ensure that that happens through a trustless manner exactly and uh I think that that is I think is always one of the biggest plus points of blockchain right where uh you could bring in people across the globe to come forward and contribute to something which is a common Mission

19:22among a lot of people right um You can't see that happening with a centralized system I don't think a centralized system is naturally built to be like that and that why blockchain is is like really connects people together right I've been telling this to a lot of people that decentralized gpus is a great POC of how um you know crypto can add value to AI because it just brought

19:45people across the globe uh to contribute gpus and that made the price of and affordability of gpus much much more lower than what it is in the centralized platform and that was just possible because there was a trustless system that can connect of these guys together and that is what we're trying to do with data as simple as that we just collecting data and building better

20:05models which can be easily available at at a lower cost to people and it helps people bootstrap and build like Innovations on AI and that that is simply what we doing and blockchain is a great driver for that to happen very cool Rah and at the beginning you gave an example about uh giving your AI agent a million dollars and having him trade for you I wanted dive a little bit more

20:29into the the payable models for AI agents and what are the other potentials for that is it true to say that uh the AI model would have its own crypto wallet and it could be on a certain blockchain or it could be using stable coins or other currencies and it could be able to manage transactions for you and and how would that work I think of payable models is

20:51basically like if you contribute to that model it's going to pay you back right so that's a simple thesis behind that and what can we contri cont to a specialized model the biggest changer in that uh the moving point is data so if you could contribute data to a model um you get paid back for the data that you contribute we make sure that attribution

21:11happens and every time this model is being used we make sure that uh you know if your data was part of the process of the outcome of the model you get attributed for that and based on the attribution you get rewarded um as I told you you get you can disc price on how valuable your data you could Better Price your data you could provide much better data to have much better revenue

21:34and stuff and so so it's basically a model paying bag for the data providers and that's why it's a pain mod and all data is not valued the same so if you have more is is it through a price Discovery and the demand of those specific AI models in how valuable the information you input into uh a specialized model is um so yes so I think price not as you

22:02told um I think data is priced very differently it's very relative uh a data in a particular country would be much more valuable compared to another country just because of certain uh needs that we have uh a data in a particular domain must could be much more valuable than a data in a particular domain by by another group of people so it it all comes down to how valuable your data was

22:27for that particular model right and and that is known when the model is being used and the model is being paid for uh if the model was used extensively and your data was one of the most useful data out there and um you get attributed more you get paid more uh you just know that the data that you contributed was valuable how people today like if you index to Google and if you had the right

22:53information Google is going to pick you up and then you're going to just going to get more users right very similarly when you index your data to a model um if that model is being used extensively and the model um gets paid you're going to get paid more and that's how the price disy happens and how far along is is open Ledger with the ability for users to

23:17come in and and add data into specialized models or create new models and also have those models available for people to use yeah very soon so we're going test net in about uh probably like few days more and it starts with a common data pool that we are collecting which is basically internet data um think of that to be the first step where we we creating a data intelligence layer which

23:41uses your bandwidth like an end users bandwidth to collect intern data so that's the first step and the second step to that is to actually let people contribute data on top of us like a data net right which is which I explain to you the ones that are building on top of us are building on close beta so we would open up they would start coming up with with their own test net versions of

24:00their U data Nets and there could be it could it could open up it would be permissionless so anyone could start their own data net and start contributing data to that and then that would evolve on to opening up the model where people could start hosting models and start fine tuning a model with the particular data they connected or provide that data itself as a rag to

24:20that model where it is much more easier and quicker that anyone can start using the model with that particular data source as a rag and uh you would see that happening in the next two three months and you could start testing the L as well very cool so if somebody who didn't consider themselves an expert but they wanted to contribute or learn more the

24:42first step is actually just contributing your internet power and you don't really have to have specialized information to contribute to the network yes the first step is that uh as simple as that uh we would want this to be as wide as possible as much as Community to participate so the first aspect is to just provide your bandwidth so that we could use that bandwidth to consume

25:04internet data um and it's not like it's not like we're going to go ahead and scrape uh the internet again right we going much more deeper where we are verticalized on various topics that has been collected in defin web so that we can have much more larger and domain specific internet pool and then you know when data have contributed if you don't have enough information there you could

25:25access this data pool as sort of like your uh data layer right where you can if you don't have any information that you need you could probably index on this you could search on this and figure out if you have that information so that is why we building this and um you know we would want the community to contribute for that and you could with just an internet connection you can

25:43start doing that and then the second level is a bit more um you know expert uh contribution where you either be a subject matter expert to contribute to some um data let's say it's about um trading you know you would have Traders contribute for that you would have a trading company which has a lot of information about onchain trading let's say to go ahead and contribute for that

26:06uh so that's a much more higher and much more um you know intense process compared to the one that we're doing right now MH very cool and with that test net launch and the iterations that you're talking about after that what is the best way to follow along with the launch and get started and learn more about how how to contribute or how to use specialized llm models so you would

26:33see that more on our website and our Twitter so uh the test net would be running for the next two three months and then we would aim for M net uh we would love to get community's feedback on what we building um you know we believe in Collective output and we are one of the core contributors of open leg and we would want Community to contribute as well um so we would use

26:56testet to get feedback and probably the next few months as I told you we go main it and where it can be accessible live and anyone can start using it awesome I'm looking forward to it I'm going to test it out and see how these models specifically work I'm very interested in learning more about uh slms and specific AI agents that you can earn money from or you can use data from

27:20others that might not be available on an llm uh because there's so much data that still needs to go inside of AI uh we're very early I'm excited to see where where open letter is going with this so thank you so much for the time Rah all the best with everything your team's working on and let's definitely follow up in the near future awesome thank you

27:40it was lovely talking to you Ashton and I hope to see hope to see you soon