Name: Assisterr AI
Uploaded: 2025-02-01T19:07:27.203Z
Duration: 2 min 33 s
Description: A decentralized AI platform enabling users to create, own, and monetize Small Language Models (SLMs) without technical expertise.

assisterr-ai

A decentralized AI platform enabling users to create, own, and monetize Small Language Models (SLMs) without technical expertise.

Assisterr AI

Hi everyone. My name is Nick. I'm co-founder and COO at Assisterr and we're building a human-driven data economy with small language models approach. 

Um, we are a Cambridge-based startup. We started back in 2022 with thinking like, okay, what can we do by utilizing AI for the web3 space? I had a background in the AI space. I had a background as a head of ecosystem growth in a web3 protocol since 2017. 

We found that there is a lot of inefficiency in how protocols and developer-facing organizations run the processes of supporting developers, building relationships with them, and supporting tech documentation. So we started with this AI agent of tomate and deral support, which right now is a standalone business. 

We already have a lot of integration with Solana, Optimis, and many other guys in the space. We simply save money with the AI automation by automating the documentation part, by automating support, and other things. But as soon as we started working on this, we started getting requests from big protocols like Solana, like, "Hey guys, that's amazing, but we still have too many requests that cannot be supported automatically by AI." 

We started thinking like, okay, we can add a ticket system. We can process this manually. The dev team can process these requests from your community manually if our agents cannot provide support. But then the amount of tickets gets bigger, especially if the protocol grows in price, and there is a question like how to.

Incentivize community to provide support peer-to-peer support to other people. That's where we start thinking like how to scale what we building, how to make it more interesting, and how to put community in the loop. 

We found that there is big problem and we experien this by building this use case for devel. There is this we call it last mile challenge with data required for AI model accuracy. You see these graphs is basically if we tak in a base model they can be quite different depends on amount of parameters, depends on the budget spent by the company. 

So the accuracy would be quite different. But as soon as we add into llms rack process, they became almost equally in terms of accuracy. If that's true, the question is like why should I spend hundreds of millions or billions on training large language models or keep using this like centralized uh llms if I can use open model, open source model, or any model but then add additional knowledge that would be up to date and would be adding this additional accuracy and context for domain main specific purposes. 

These big tag guys which exist in the space they actually end up with the same conclusions because right now you probably heard about Reddit SEC flaw with quora deals with big T guys. They what they doing they need community generated content, user generated content because it provides this last mile accuracy. But they don't want to pay you, they don't want to pay us for.

Using our content our data our our data sets and we believe that it should be fixed because current data market for AI it's kind of broken because there is M resourced data provenance mechanisms that can allow us to check okay was my data used at some AI model was my data used by some specific application. 

And big tech companies they don't have any incentive to implement this mechanism because it means that they need to share profits with you while if there is no such mechanism no need to to prove anything and no any uh instruments to fairly evaluate and reward data contributors. 

So to address these challenges we develop our own chain data Provence protocol that enableing incentive driven network of small language model. What the pin here it's basically we empowering people to contribute data to each small language model. 

This model can be stored on decentralized infrastructure structure centralized on your personal Network and then we also track in who when and how uh change or affect a data set which is behind the model. 

And in that way each of us or any user can re get Fair rewards especially if the model data set which this specific person was contribute to was actually quered. 

So there is a mechanism how to even check and fair will reward let's say there is 100% of data sets behind the model and only 10% of this data is actively used by user basically quaring by users in this way we can Define which data and who was be highest.

Ambition of this data and reward these people most while other people as a contributor would be rewarded less. We recently, like two and a half months ago, we won autonomous agent hakon at East Denver with this solution. 

Uh, let's go a little bit deeper into the role and how everything works. So we have four roles at this moment of time. So you can join ecosystem as a small language model Creator. 

Uh, basically you will define which domain you want to target, which small language model you want set up, and you would be organizing people around this model. So they would be basically contributing data or acting as validators. 

Also, we have data contributors. It's people or experts in specific domain zones who have knowledge, experience, expertise. It can be statistic information, user research, like whatever else. 

Um, and they contribute data to the models which they have expertise in. Then we have a role of data validators, which is highly important because otherwise we will end up with spam models where people would be just like running for this like bounty or reward system or tokens. 

Or, and they would just spam and lower down the quality and accuracy of the model. So the role of validator is to validate the quality of the data based on different parameters. 

And then we have a mechanism if something happening and we will discover that the data contributor and data validator they validated wrong data, low quality data. We have a mechanism how to share AI Agent Store with colleagues.

Penal tize uh these people and they also need there is a taken mechanism to secure the network and other things. 

And also end users they're going to be using applications. Why it's important for them because uh in the future or even now we have three options. We have uh Google search or any search engine. We have llms which are averagely good and averagely bad for any use case. 

And we're going to have small language approach and I I'm betting and our team believes that in the second half of 2024 uh small language approach going to be super popular and hyping because we discovering that large language models with this averagely results for everything that's not something that business and real users want to use and want want to have. 

Um in terms of the user flow basically users right now can register at the Assisterr choose the role set up and trace small language models based on they data sets information that they have. 

Then they can choose different service providers. Right now we are partnering with i.net in terms of decentralized storage zero gravity protocol as a service providers. You can choose basic models that you want use as a basic infrastructure and also add some crypto payroll. 

We're going to add monad support salana support and also evm chain soon. Uh then you can customize uix right now we support different conversational chat bots. Right now we support web widgets that can be inbuild it in any uh web service that your company or.

Your team is using actively and then you're listing it on small language model Marketplace. Uh, we're going to launch it next week. Uh, where anyone can uh find your model. Uh, can choose the model that which they want to contribute in or join as a validator. That's pretty it right now. 

We are running several use cases. So devel use case which I mentioned it's already live. It's already generating revenue. We have 20 live integration with different protocols while we also running proof of concept with um crypto analytics teams with m&a deal flow teams in London and couple other use cases. 

Uh, from token utility perspective so we have tokens that can be used uh as a vesting and also on a staken. Staken will allev you to governance and affect different decision in terms of each small language model but also on a protocol on a infrastructure level as well. 

Also, we have data ownership register which basically keep track records about who when and how submit some specific information to this small language model. And it's allow us not to put the whole model on chain because we don't think that is reasonable at this point of time. 

Time because it doesn't give any any any pluses or efficiency to the uh speed of interaction speed of training and all of this. And also we expect to launch uh next quarter kind of small language model Launchpad which allo which would be allowing people to tokenize um small language models that they want to build. So if you have

Some knowledge uh or or domain specific background and people know that you can build and feed this model with your data. And you can organize community of 10 100 people maybe thousand people around you. Uh they can also stake additional our tokens and incentivize you to build this application to build this use case faster and support your community in that way. And of course you're going to be sharing profits from this small language model with the community who support it. 

Um we see huge market opportunity. So Bloomberg estimates that ji powered application is going to be bigger market than App Store. Uh as it is right now and there are some companies who are running some at least targeting the space in different ways like Sahara mellas F. So on the difference that we targeting we try to tokenize not only the protocol level but also each application each model level where community can contribute and and get and get some incentives for contributing the data that wasn't discoverable. 

Uh before and and time before and right now anyone would be able to get exposure to this. So imagine you want to launch uh your product on b2c product on a China market. So you would get some small language models that train on b2c data uh by b2c experts with specific exposure on China market. And there would be such use cases just like endless amount of use cases that could be done in this model by using our infrastructure by using our approach.

There are several companies as I mentioned who targeted the space. The difference between us and them is we have working infrastructure. We have already revenue and we keep running this. 

So we, we, we, we want to get more exposure with more use cases. So that's my call to action to the audience that, uh, is here. 

So first, we are hiring, uh, AI researcher to work on decentralized Vector DB. If it's something that is interesting to you, I would love to chat after this call. 

We are also welcome to join the ecosystem as a small language model creator, data contributor, and data validator. That's it. 

Note, that there may be slight differences in text and video. Watch video for the original content.

Assisterr AI Video