Some people may not know why NEAR is mentioned when AI is being hyped. Here is a little-known fact: NEAR was originally an AI company. Illia Polosukhin, the founder of NEAR, has nearly 10 years of experience in the AI ​​field and is one of the eight authors of the landmark paper “Attention is All You Need”.

He and co-founder Alex Skidanov founded NEAR.ai in 2017 with the goal of creating the first "AI programmer." People could communicate with computers in natural language, and the computers would automatically program themselves. Given the limited capabilities of AI models at the time, this attempt failed.

During this process, they came across smart contracts and found them to be an interesting subset of programming, but blockchain technology had too many other challenges. Therefore, NEAR decided to make a strategic transformation in 2018 and first build a truly useful decentralized development platform, NEAR Protocol. It was originally estimated that this transformation would only take 6 months, after which they could return to the research and development of AI technology, but 6 years have passed in the blink of an eye. Other blockchain projects are now doing AI as a "strategic transformation", while NEAR has finally returned to its old business.

NEAR’s official website recently announced its AI technology stack, which is divided into three main layers: application layer, infrastructure and model layer, and data layer.

Under this three-layer technical architecture, NEAR has gathered 11 of the latest ecological AI projects. Next, BlockBeats will briefly sort out these 11 projects for you to see what the NEAR AI ecological landscape looks like.

Application layer: Bitte, Cosmose, Jutsu

Bitte: AI Agent Wallet

AI Agent is a popular development direction of AI applications. The general consensus is that in the future, there will be various AI Agents on the blockchain to perform various transaction operations instead of humans.

Bitte is trying to take a small step forward with existing technologies when the relevant infrastructure is still not perfect. By accessing OpenAI's model API, Bitte allows users to use natural language prompts to command Agents to complete various on-chain operations in a chat window interface similar to ChatGPT.

For example, if a user enters "Mint an NFT with AI of a rocket going to the moon", Bitte will call the API of DALL-E 3 to generate an image of a rocket landing on the moon and mint it on the NEAR blockchain. In addition, users can also let the Agent swap and transfer tokens, create contracts and NFT sets, and so on.

The Bitte wallet was developed by the Mintbase team, a project born in 2022 amid the NFT craze. For details, please read "Mintbase: NFT Grid Shop on Near | Project Introduction".

Cosmose AI: AI shopping guide platform

Cosmose AI is an e-commerce company that uses AI to predict and influence people's offline shopping methods. It received investment from the NEAR Foundation in April last year. After reaching a cooperation with the NEAR Foundation, KaiKai, an e-commerce platform under Cosmose, launched a cryptocurrency for payment, cash back and rewards: Kai-Ching (KAIC). This is a native stablecoin running on the Near network. 1 KAIC is equal to 0.01 US dollars and can only be used in its application.

Jutsu: AI Agent Marketplace

This project is still in the white paper stage. According to the concept shown in the document, it wants to be similar to the GPT Store launched by ChatGPT. Developers can publish the built AI Agents on it, and users need to pay for the use of these Agents with its platform currency JUT.

This is a hackathon project that came out of Eth Denver last year. It started out as a developer development tool called “NEARpad”.

Infrastructure and model layer: Exabits, Hyperbolic, Nevermined, Pond

Exabits/Hyperbolic: GPU computing power rental platform

Both Exabits and Hyperbolic want to build io.net on NEAR, and both have been selected for the first phase of the NEAR Horizon AI incubation program.

After experiencing the bombardment of various GPU computing power dog projects from the beginning of the year to now, I am a little tired of seeing GPU computing power leasing projects. However, if a chain wants to build its own AI ecosystem, it must have its own GPU computing power leasing platform.

At present, neither of these two projects has released tasks to run nodes or provide computing power. Friends who can’t find projects to mine after mining io.net can pay attention to these two projects.

Nevermined: AI Payment Protocol

Nevermined is a payment platform that allows AI developers to monetize their various products, including AI models, AI agents, and datasets. By creating smart subscriptions, developers specify access parameters for AI products, such as price and time limits, basically moving the functionality of Web2 subscription platforms to the chain in the form of NFTs.

Currently, Nevermined applications are deployed on Polygon, Gnosis, and Arbitrum networks, and in the future they should be expanded to NEAR as a payment infrastructure to support the development of the platform’s AI ecosystem.

Pond: Decentralized GNN Model

Unlike Transformer, which is designed for processing sequence data (such as natural language), Graph Neural Network (GNN) is a neural network designed specifically for processing and analyzing graph structure data. It is widely used in social network analysis, chemical molecule property prediction, knowledge graph, recommendation system, etc. Compared with Transformer, GNN is more suitable for capturing the local structure of the graph and the complex relationship between nodes.

Pond said it is building the first decentralized graph neural network (GNN) model, trying to learn the interaction patterns between users and contracts from blockchain data, and predict the future behavior of users based on the learned on-chain behavior patterns. It sounds awesome at first, but to be honest, GNN was already very mature before Transformer came out, and there have been some studies and attempts to use it in blockchain data analysis, but it was limited to money laundering and phishing transaction detection. Whether this new model can surpass the original depends on what effect it actually produces.

Data layer: Masa, MIZU, Nillion, Ringfence

Masa Network: Decentralized Data Marketplace

Masa is a subnet on Avalanche that allows users to earn token rewards by contributing data and computing resources by running worker nodes. These worker nodes crawl, structure, transform, annotate, and vectorize a large number of data sources such as Twitter, Discord, and podcasts. Developers (Oracle nodes) build artificial intelligence applications by accessing this data and LLM services.

But this is not the biggest gimmick of this project. In addition to the above "Node to Earn", Masa has been promoting "Surf to Earn". Before the transformation to AI, Masa was originally a decentralized credit scoring protocol based on SBT (Soul Bound Token). Later, the concept of zkSBT (zero-knowledge soul bound token) was innovatively proposed. Unlike traditional website cookies, users can share their website browsing data completely anonymously through zkSBT for data analysis and model training, thereby obtaining token rewards. To this end, Masa plans to launch a Chrome extension, but this Chrome extension seems to be more difficult to launch than expected.

MIZU: Decentralized Synthetic Data Generation Network

I don't know why, but every project likes to emphasize that they are the first in various ways. Mizu says it is the first and largest decentralized open data network, which is actually a decentralized synthetic data generation network. Based on the data set contributed by users, the network encourages the community to build prompt words to generate a large amount of synthetic data, which is submitted to the data repository after verification, thereby making up for the lack of real-world data and providing more targeted training data. The roadmap says that the test network will be launched in August, so you can pay attention to it if you are interested. Data should become another new key infrastructure track for decentralized AI after computing power.

Nillion: A decentralized secure computing network

Nillion is a decentralized public network designed to handle secure computation and data storage without relying on blockchain technology. It introduces a new type of cryptographic primitive called Nil Message Compute (NMC) that enables nodes in the network to process data in a secure and private manner without communicating with each other or maintaining an immutable ledger like traditional blockchains. NMC is the core technology behind Nillion. It enables the network to split and distribute data across nodes, perform secure computation on data without the need for decryption, and achieve processing speeds close to centralized servers while protecting privacy.

In general, Nillion proposes a new cryptographic primitive that has great application potential in private AI model reasoning and training.

Ringfence: Data Monetization Platform

It has always been controversial for AI companies to capture user data for model training. The rights of creators are difficult to protect under this model. Ringfence has proposed a creative solution - rNFT. The data uploaded by users to the platform will become NFT assets and be authorized for use in the form of NFT.

Traditional NFTs usually represent ownership of a single item, while rNFTs are like a folder or collection that contains multiple sub-NFTs (called cNFTs). Through smart contracts, rNFT owners can easily commercially license the entire rNFT or specific components of it.

The Ringfence platform allows users to contribute rNFTs for neural network training and receive rewards, with the long-term goal of building the first neural network trained with 100% delegated work.

Summarize

After reviewing the whole article, we can find that there are not many AI projects native to the NEAR ecosystem. Many of them are forcibly incorporated into its AI ecosystem in the name of cooperation. Most of these projects are still in the proof-of-concept stage and need a lot of effort before they can be officially launched.

It can be said that compared to Arweave's use of the new project AO to prove its determination and strength in transforming AI, NEAR's return to AI is much more low-key. Many people still think of NEAR as "a high-performance public chain". In the eyes of many people, the only connection between NEAR and AI seems to be the founder Illia. But in fact, NEAR has been continuously building AI infrastructure in a holistic way. For example, its strongly advocated chain abstraction is very important for introducing agents into chain interactions in the future.

It can be said that NEAR has no shortage of brand, technology, or funds for AI. However, how to establish its own AI business card and cultivate a complete and vibrant AI ecosystem is a difficult problem that NEAR needs to take seriously in addition to building AI infrastructure.