NEAR AI: Building the most powerful open source AI for users

NEAR's continuous expansion in the AI field has the potential to become the preferred platform for decentralized AI development, paving the way for the next generation of AI-based dApps.

Author: IOSG Ventures

Original Title:IOSG Weekly Brief｜NEAR AI: Building the Strongest Open-Source AI Owned by Users #254

Cover:NEAR

This article is for learning and communication purposes only and does not constitute any investment advice. Please indicate the source when reprinting, and contact the IOSG team to obtain authorization and reprint instructions. The projects mentioned in the article are not recommendations or investment advice.

Near.ai → NEAR Protocol → NEAR AI

In late 2017, a groundbreaking AI paper - NLP & AI Revolution - was published, and 4 out of the 6 authors went on to start AI-related companies (later joining openAI), while the other one was the founder of NEAR AI, Illia. When we saw this paper, we started to reach out to Illia, and I first visited the NEAR office in 2018, which was in an underground shared space in San Francisco, where a group of engineers were gathered for a workshop. We had a long conversation with the founding team from Ukraine, helped Illia obtain his visa to come to China, and assisted them in organizing more than 4 offline developer events within 3 months.

As a crypto OG project with a 7-year history, NEAR's two founders are well-known in the field of artificial intelligence, including Illia Polosukhin, the co-creator of the Transformers architecture, and Alexander Skidanov, a genius in the MemSQL domain. Their initial vision was to train models through global computer science students and apply program synthesis. Therefore, NEAR was initially established as "Near.ai", driven by the belief that they could not only make breakthroughs in program synthesis research, but also launch a revolutionary product powered by this technology. (Disclosure: IOSG participated in NEAR's early investment rounds in 2018)

However, in 2017, machine learning, especially natural language processing (NLP), was far from as mature as it is today, and the real-world application of program synthesis still faced enormous challenges. Therefore, their advisors suggested that they try to apply the program synthesis technology to solve the problem of developers programming verifiable smart contracts on Ethereum. However, at the time, Ethereum had many limitations, especially in the availability and scalability of micro-transactions, and no blockchain or traditional payment system could meet the demand.

As a result, NEAR spent six years building a blockchain that is both highly scalable and user-friendly, one that can adapt to the needs of micro-transactions, provide a familiar onboarding process, and solve a series of problems such as cold start, key management, and account availability. But this also shifted the market's perception of NEAR from artificial intelligence to an unclear Layer 1 blockchain use case. With the emergence of more and more cryptocurrency and AI innovations, including computing markets, reasoning networks, and the emergence of numerous AI agents, as well as the impressive price performance of high-value AI tokens like TAO and WLD, people seem to have forgotten that NEAR was the pioneer in this field.

After seven years of development, with NEAR's infrastructure now mature, NEAR has a scalable, user-friendly blockchain and an ecosystem of application builders. Illia and Alexander can now refocus NEAR's goal on their original vision, which is to build NEAR AI as the center of user-owned artificial intelligence in the next stage of development. At the IOSG OFR event in Dubai in April this year, Illia shared his views on "Why AI Needs to be Open" and expressed his determination to build NEAR into the leading ecosystem for next-generation AI research and applications.

NEAR is driven by the vision of user ownership of AI, and is building an AI ecosystem on the Web 3 orbit. Compared to the popular centralized AI model controlled by large tech companies, NEAR believes that user-owned AI provides developers and users with a truly better choice. NEAR hopes to maximize the benefits of AI for humanity and communities through Web3 and the entire NEAR ecosystem, while minimizing the potential risks of centralized AI.

Considering that the Crypto x AI field is still in its early stages, NEAR's strategy is to plan to enhance the development of the entire Web 3 AI ecosystem from a more non-profit perspective, whether through incubators, accelerators, investments, building its own open-source AI research centers and tools, or leveraging competitions to cultivate community-driven AI development methods, with the strategic goal of realizing the vision that the fruits of AI can be truly shared by the community. The improvement of its token economic benefits is not currently the highest priority for NEAR.

This article will focus on NEAR's re-layout in the Crypto x AI track, helping everyone better understand its positioning in the Web 3 AI ecosystem.

NEAR Foundation Incubator

In response to this theme, the NEAR Foundation announced the launch of the NEAR AI x HZN Incubator in May, and also set up an investment department focused on AI projects and a research and development laboratory.

The NEAR Foundation has selected 11 projects to support, and the participating projects in the incubator program will receive the following during the initial 12 weeks (June to August 2024) and subsequent phases (until May 2025): access to NEAR's AI expert network, dedicated technical and token launch support, computing resources, and financing opportunities.

These projects cover multiple aspects of the AI technology stack, including data, models, computing, and reasoning, aiming to provide users and developers with a secure and decentralized open AI solution, breaking the monopoly of centralized AI service providers.

Highlights of some of the selected incubation projects include:

• Hyperbolic: an open-access AI cloud platform that hosts the latest open-source AI models, including Llama 3.1 405B.
• Mizu: launched its beta data preprocessing platform, attracting 20,000 users in the first week.
• Pond: launched its first graph neural network (GNN)-based model, using on-chain data for wallet prediction and achieving a 20% prediction success rate.

NEAR x Delphi Labs Accelerator

Due to the good results of the early incubation, NEAR further collaborated with Delphi Labs (another highly active research and venture capital firm in the Crypto and AI fields) to jointly announce the first joint AI accelerator program. This is a strategic move aimed at supporting and rapidly expanding those projects with high potential in the intersection of AI and Web3.

The selected teams will benefit from comprehensive support, including technical guidance, funding opportunities, and access to the expert networks of NEAR and Delphi. In addition, each team will receive a $100,000 investment from the NEAR Foundation and up to $250,000 in potential investment from Delphi Labs for further incubation or acceleration after the program is completed. The program will be showcased at the investor-centric Delphi Labs demo day in mid-December. Each selected project will also receive $50,000 in computing credits from the accelerator program's advanced computing partner Aethir, ensuring that each team has the computing power needed to develop and deploy high-resource AI models, allowing them to focus on project building without being constrained by infrastructure limitations.

NEAR AI R&D

In addition to funding and incubating other projects, NEAR itself is conducting a lot of AI research and officially launched NEAR AI on November 8th at its REDACTED event, positioning itself as an AI blockchain and launching three important developments.

NEAR AI Assistant

The NEAR AI Assistant is a chatbot-like assistant that can answer a variety of general questions, such as "What is IOSG Ventures?" and "What is NEAR AI?". The AI assistant provides local memory functionality for each user, associating with each NEAR account and remembering information during the user's conversation with it, and utilizing these memories in interactions with different agents on its platform.

More importantly, it allows non-technical creators and developers to easily perform many crypto-related operations, especially those related to cryptocurrencies. Users can simply prompt the AI assistant to complete operations in a few sentences, such as creating a Memecoin image for the user, generating a website for that Memecoin, and deploying it to the Pump.Fun platform, realizing AI agent Token issuance.

Open-Source Toolkit for Building AI Agents

NEAR has also released an assistant API toolkit to help users build, measure, and deploy AI agents, and integrate them into specific applications.

Currently, over 60 agents have been uploaded to the NEAR Agents Hub, each focusing on different use cases in Web2 and Web3, such as web crawling, token swap agents, YouTube transcription agents, etc. Each agent can connect and call other uploaded AI agents to perform tasks.

In future updates, agents on NearAI may even be able to transact and interact with external services and APIs.

Access Community-Built AI Research and Resources for Cutting-Edge Model Development

NEAR believes that open-source contributions are key to realizing user-owned AI. To achieve this goal, NEAR has launched an AI Research Hub that aggregates all the toolkits and infrastructure needed for researchers to build open-source AI, including datasets, models, inference, decentralized storage and computing, etc. An important feature of this open-source toolkit is the benchmarking tool, which allows developers to compare the different implementations of agents and models with the same functionality, helping users continuously adjust and improve the performance metrics of their agents and models over time.

To ensure the development of state-of-the-art (SOTA) models, the NEAR AI Research Hub also adopts a competition-based approach to continuously optimize models. Currently, NEAR is running an ongoing Model Training Series competition, encouraging the community to gradually train larger-scale AI models (from 500 million parameters to 14 trillion parameters) with perplexity as the benchmark. These models use the FineWeb dataset and are trained on NEAR protocol core nodes using Trusted Execution Environment (TEE) and GPUs, ensuring that the models are both privacy-preserving and have monetization potential.

Clearly, to capture the value created by open-source contributors, appropriate rewards and revenue sharing need to be provided. NEAR is already planning its future roadmap, aiming to establish a reward and copyright allocation system for competitions and model usage.

Conclusion

In NEAR's vision of user-owned AI, NEAR welcomes anyone to contribute, and every contribution will be valued and rewarded. Code contribution is not the only way to help the community. Answering questions, helping others, and improving documentation are also extremely valuable.

As NEAR continues to expand in the AI field, NEAR has the potential to become the preferred platform for decentralized AI development, paving the way for the next generation of AI-based decentralized applications (dApps).

In fact, we have already seen some AI-related dApps being built on NEAR. For example, a protocol called Bitte is working on decentralized agent discovery, with use cases of its agent plugins including Non-Fungible Token minting, cross-chain transactions, and more.

Clearly, the NEAR ecosystem has many innovations beyond just AI. To explore more projects in the ecosystem, you can refer to a tweet shared by Marcus:

Currently, NEAR's AI strategy and its integration with the NEAR Token and blockchain, as well as how to accumulate value for the NEAR Token, have not been particularly clear. However, in the future, Web3 dApps developed based on the NEAR AI platform may be deployed in the NEAR ecosystem. The NEAR Token may be used to reward and incentivize users participating in model training competitions, or as a payment method for using platform proxy services. In the future, when users need to call proxy services on the NEAR platform, they may need to pay NEAR Tokens as a usage fee.

In any case, this AI strategy will greatly promote more users to embrace decentralized Web 3 AI, taking an important step towards users fully controlling their own AI future.

Disclaimer: As a blockchain information platform, the articles published on this site only represent the personal views of the authors and guests, and are not related to the stance of Web3Caff. The information in the articles is for reference only and does not constitute any investment advice or offer, and please abide by the relevant laws and regulations of your country or region.