NetMind.AI: Who says academics can’t do business? Tsinghua team creates the first decentralized AI full-stack model

Introduction: d/acc movement and decentralized AI

In the article "My Technological Optimism" published in November last year, Ethereum founder Vitalik proposed the philosophy of d/acc (defensive, decentralized, democratic, differential acceleration), emphasizing the selective advancement of technological development, avoiding excessive concentration and monopoly, to ensure the overall benefit of technological progress to human society.

Especially in the field of AI (artificial intelligence), Vitalik pointed out that AI has the potential to surpass human intelligence and become a new dominant species, so its development needs to be treated more cautiously. However, there are many problems with the current AI governance structure. OpenAI is tied to Microsoft, Anthropic is tied to AWS and Google, and a few technology giants monopolize GPU computing power and user data.

"We need a truly open AI ecosystem." This concept has driven many Web2 AI entrepreneurs to turn to Web3 to seek decentralized and democratic solutions.

The Founding of NetMind.AI

Kai Zou, founder and CEO of NetMind.AI, is an innovator committed to integrating AI with Web3. He has accumulated rich experience in the field of Web2 and firmly believes that in the future when AI models grow in scale and are widely used, the distribution and decentralization of AI is an inevitable trend. He believes that the framework of Web3 is an important and necessary solution that can provide a broader space and more possibilities for the development of AI.

Kai Zou graduated from the Basic Science Class of Mathematics and Physics of Tsinghua University in 2010 and received a master's degree in mathematics and statistics from Georgetown University in 2013. Inspired by DeepMind and OpenAI, he founded the artificial intelligence company ProtagoLabs in 2016 and served as CEO. This company is more like an AI research studio. The team is committed to AI research, and research papers have been published at top industry conferences such as AAAI and IEEE. Among them, the paper "EDA: Easy Data Augmentation" co-authored by Kai Zou and OpenAI researcher Jason Wei has been cited more than 2,000 times.

In addition to his academic achievements, Kai Zou has also invested in multiple AI startups, including Haiper.ai, Auto Edge, Qdot, and Orbit. In 2021, he decided to start a product-oriented company. When training and fine-tuning language models, he realized that GPU graphics cards were very expensive and difficult for ordinary AI enthusiasts and small research institutions to afford. So he began to study asynchronous training architectures and tried to combine small GPU cards to train neural networks. In this way, he created NetMind Power to encourage more people to contribute idle machines to create a large distributed computing network.

Kai Zou firmly believes that only through decentralization and open collaboration can AI technology truly benefit all mankind.

NetMind Power NetMind Chain

NetMind Power is the core product of NetMind.AI, a decentralized AI full-stack development platform based on NetMind Chain.

Operation Mechanism of NetMind Chain

NetMind Chain adopts the Proof of Authority (PoA) consensus mechanism, with 21 Master Nodes responsible for verifying transactions and generating blocks. These Master Nodes are elected by the community and must be participants with a good reputation and continuous contributions to the community. Compared with mechanisms such as Proof of Work (PoW) and Proof of Stake (PoS), PoA significantly reduces energy consumption and maintenance costs while ensuring decentralization and security.

In addition to the masternode, there are several types of participants in the NetMind Chain ecosystem:

Computing Power Providers: Individuals or institutions that contribute idle GPU computing power resources receive NMT rewards according to effective workload.

Computing Power Users: Developers and researchers who use the computing power resources of the NetMind Power platform for AI model training, fine-tuning, and reasoning need to pay NMT tokens as service fees.

Stakers: Support the operation of the masternode by staking NMT tokens and receive staking rewards.

Liquidity Providers: Provide liquidity for NMT on decentralized exchanges and earn transaction fees.

In order to efficiently aggregate globally dispersed GPU resources, the NetMind Power platform uses innovative decentralized clustering technology. When a user initiates a model training, fine-tuning, or inference task on the platform, the task will be automatically assigned to the GPU resource with the best performance and lowest latency for execution. NetMind.AI achieves unified scheduling and management of large-scale heterogeneous computing resources through peer-to-peer networks, containerization, Kubernetes and other technologies.

In addition, NetMind.AI has been developing asynchronous training technology to solve the network bottleneck problem in a decentralized environment. Traditional distributed training requires real-time communication between GPUs through a high-speed network, which is difficult to achieve in a decentralized scenario. NetMind.AI's asynchronous training divides the training data and AI models into smaller parts, assigns them to different GPUs for parallel processing, and finally aggregates the results. While ensuring the training effect, it can significantly reduce the requirements for network bandwidth.

Finally, in order to ensure privacy and security, NetMind Power is using model encryption and data isolation technology. The platform encrypts AI model code, training data, and model weights at each node and during transmission. In addition, through data partitioning and model segmentation, it is ensured that no single node can access the entire data set or model, thereby greatly improving security and privacy protection.

NMT Token Economics Model

NetMind Token (NMT) is the native utility token of NetMind Chain, designed for a variety of platform functions, including paying for training and inference services, rewarding users who contribute computing resources, and participating in platform governance.

The total supply of NMT is 147,571,163. The platform controls the issuance and destruction of tokens through smart contracts to regulate market supply and demand. For example, the platform will adjust the allocation ratio of computing fees based on market demand and NMT prices to ensure the stability of the token value.

NetMind.AI launched a new token economic model on April 16, 2024, and a total of 113 million NMT tokens will be issued in the next 10 years, of which more than 50% of the newly issued tokens will be used to allocate to the platform's computing power providers.

Token Distribution (Total 10 Years):

・62.5M NMT (55.31% of new issuance and 42.35% of new total supply) are designated as Hash Power Rewards, awarded to contributors who provide hash power to the network.

・16.5M NMT (14.60% of new issuance and 11.18% of new total supply) are allocated as Staking Rewards, awarded to users who participate in staking and provide liquidity.

・16.5M NMT (14.60% of new issuance and 11.18% of new total supply) are designated as Ecosystem Growth Fund, used to expand and enrich the NetMind.AI ecosystem.

・7.5 million NMT (6.64% of new issuance and 5.08% of new total supply) are reserved for strategic investors and advisors.

・10 million NMT (8.85% of new issuance and 6.78% of new total supply) are allocated as team funds to reward the NetMind.AI technical team.

As more and more users choose to hold and stake NMT for a long time to earn returns, the actual amount of NMT in circulation will decrease significantly. Currently, there are more than 20 million NMT staked, accounting for 57% of the circulating supply.

Through a carefully designed token economics model, NetMind.AI closely combines economic incentives with system contributions to build a virtuous cycle ecosystem: the more people use the NetMind Power platform, the greater the demand for NMT, the more stable the price, which in turn will attract more computing power providers and developers to join. NetMind.AI is exploring a new and sustainable path for the development of decentralized artificial intelligence.

Strong computing power supply and demand

As Nvidia's revenue and stock price continue to hit record highs, the concept of "computing power is the new oil in the AI era" has taken root in people's hearts. Decentralized computing power leasing has become a highly competitive track, with the emergence of many similar projects, such as io.net, Aethir, Akash, Gensyn, Golem, and so on.

To judge the product market fit (PMF) of these computing power rental platforms, it is necessary to consider the two dimensions of computing power supply and demand:

Computing power supply: Can the platform aggregate enough GPU computing power resources?

Computing power demand: Is there enough demand for the computing power aggregated by the platform?

Continued growth of GPU computing power

According to the official dashboard data, the GPU computing power of the NetMind Power platform has been growing significantly since February. As of now, a total of 2,076 GPUs have been connected to the platform, and 4,804 GPUs are waiting in line to join.

The NetMind Power platform provides GPU computing power of various specifications and models, ranging from low-end and mid-range GPUs (such as RTX 4070 Ti, V100, RTX 4060 Ti, RTX 3070, RTX 3060) to high-end GPUs (such as 3090 Ti, 4090, A100 and H100).

It is worth noting that the utilization rate of high-end GPUs on the NetMind Power platform is very high, especially the utilization rate of A100 and 4090 is close to full capacity (98% and 99% respectively). This shows that high-performance GPUs are in great demand on the platform, mainly used to handle complex and high-computation tasks.

Diverse partner network

Partners use more than 1,000 GPUs on the NetMind Power platform, accounting for more than 50% of the platform's total computing power. This shows that NetMind.AI's success not only relies on its technological advantages, but also benefits from its broad and deep partner network.

In the commercial field, NetMind.AI has established strategic partnerships with many innovative AI companies, such as ProtagoLabs, Haiper.ai , Auto Edge, Qdot and Orbit. These partners use the computing power resources and AI ecosystem of the NetMind Power platform to accelerate the development and iteration of their own products.

The NetMind.AI team has a senior AI background. Founder Kai Zou has published many papers at top academic conferences and has also collaborated with OpenAI researchers. This background enables NetMind.AI to connect with professional AI researchers and institutions. Currently, NetMind.AI has established academic partnerships with many world-renowned universities such as Tsinghua University, Cambridge University, University of Wisconsin-Madison, Shanghai Jiaotong University, Huazhong University of Science and Technology, and Fudan University to jointly carry out cutting-edge AI research projects. These projects involve hot areas such as graph neural networks, neural architecture search, natural language processing, and data privacy protection, and make full use of the computing power support and AI ecosystem provided by the NetMind Power platform.

In April this year, the NetMind Power platform began to support legal currency payments, further lowering the user threshold. Users can directly use credit cards or bank transfers to purchase computing power and services without having to go through the cryptocurrency exchange process. This move not only facilitates traditional corporate customers, but also attracts more individual developers and AI enthusiasts to join the platform.

Model as a Service (MaaS)

NetMind.AI not only provides computing power leasing, but is also committed to providing full-stack AI development services, including open source model libraries, high-quality data sets, and easy-to-use training and inference platforms. This model as a service (MaaS) model makes AI development easier and more efficient. Developers do not need to train models from scratch, but can directly fine-tune and deploy pre-trained models on the NetMind Power platform, greatly reducing development costs and thresholds.

It is worth mentioning that NetMind.AI's data sets are also used as evaluation benchmarks by Google Gemini models. In the math benchmark test of AGI Odyssey, Google DeepMind's latest AI model Gemini 1.5 Pro achieved a breakthrough score of 55.8%. The data set for this math benchmark test was originally used in the Global Artificial Intelligence Championship (GAIC) math competition hosted by AGI Odyssey and supported by NetMind.AI. It was designed by math experts and professors and covers high school, college and Olympic levels. Currently, this data set has been open sourced on GitHub: https://github.com/protagolabs/odyssey-math/tree/main .

In addition, NetMind.AI also supports users to upload their own models and provide services to other users, thus forming a virtuous ecological cycle. Thanks to such a complete AI ecosystem, the computing power gathered by NetMind.AI can be fully utilized without worrying about excessive computing power.

To sum up, NetMind.AI has established a strong moat in both supply and demand of computing power. On the one hand, thanks to innovative technical solutions and token economic incentives, NetMind.AI has aggregated a large amount of idle computing power resources around the world; on the other hand, NetMind.AI's unique AI ecosystem and top partner network have found the real demand side for these computing power resources.

Complete AI Ecosystem

In addition to the decentralized AI full-stack development platform NetMind Power, NetMind.AI also provides a series of decentralized AI products and services, committed to building a complete AI ecosystem so that everyone can use AI conveniently and affordably. As founder Kai Zou said: "We are working on all levels. I know it sounds ambitious, but we are working on all levels. All levels are basically ready and open to the public."

NetMind Chat Gaia

NetMind Chat is an AI chatbot platform where users can experience and choose the large language model that best suits their needs. The platform provides a simple chat interface for users to compare different language models and fine-tune these open source models through the NetMind Power platform.

It is worth mentioning that in addition to well-known open source language models, NetMind Chat also provides its own Gaia chatbot. Gaia is based on the generative pre-trained Transformer family and is fine-tuned using supervised learning and reinforcement learning techniques on the dataset collected by NetMind.AI. In addition, Gaia adopts a retrieval enhancement approach to access extensive knowledge on the Internet, rather than just parameterized knowledge within the model.

Digital Human Avagi

Avagi is a user-customized digital human model that allows users to design realistic appearance, movements, and expressions. Users can animate digital humans using voice, text, or gestures for a variety of purposes such as entertainment, education, marketing, or social media. The platform introduces digital entities embedded with AGI (artificial general intelligence) to provide an immersive conversational experience through intelligently generated video clips. In addition, NetMind.AI is developing a 3D version that will be launched soon.

AI Life Healthcare Solution NetMind Life

NetMind Life is committed to using AI to provide personalized health solutions to help users achieve longer and healthier lives. Through comprehensive AI-driven assessments, it provides users with tailored lifestyle recommendations, health tests, nutritional supplements, and treatment plans.

AI Agent Development Tool and Platform NetMind XYZ

NetMind XYZ is an AI agent development tool and platform that provides users with all the modules to easily build AI agents. NetMind.AI will also build a multi-agent system on the platform to allow AI agents to communicate with each other and collaborate on projects.

The platform was inspired by founder Kai Zou's thinking about AI Agent and Multi-Agent systems: "I believe that in the future, humans and AI agents will live together in a society. AI agents may have similar needs to humans. They may need to go to school to learn, go to companies to make money, and communicate with each other to evolve into better intelligent agents."

Currently, NetMind.AI is building a marketing company based entirely on AI, which will become the first multi-agent system on the platform.

AIGC Tool Portfolio NetMind Studio

Finally, NetMind.AI also plans to provide a series of AI-generated content (AIGC) tools, covering a variety of content forms such as text, images, audio, video, etc., to help creators improve productivity and creativity.

NetMind.AI has established a complete and robust AI ecosystem by creating competitive products at all levels, including computing power, models, data, tools, and applications. This full-stack layout can not only better meet users' diverse AI needs, but also generate synergies between different products and services, forming a stronger competitive advantage.

Conclusion: Endless imagination space for decentralized AI

According to CoinGecko data, the current market value of AI coins is only 35.4 billion US dollars. Compared with other fields, this number is still relatively small. For example, the market value of DeFi has reached 105.6 billion US dollars, and the market value of L1 public chain is as high as 2.16 trillion US dollars. Looking at the Web2 field, the top 47 AI companies have a total market value of 9.945 trillion US dollars. These data show that the decentralized AI track has huge development space and growth potential. From all aspects, we are on the eve of AI Summer.

Compared with other head decentralized AI projects, NetMind.AI is expected to have greater room for growth. This is mainly due to two aspects: one is the deep technical accumulation and professional background of the NetMind.AI team in the field of AI; the other is the all-round AI ecosystem created by NetMind.AI, which covers all aspects from computing power leasing to model training, reasoning and deployment. This dual moat of technology and ecology makes NetMind.AI stand out in the decentralized AI track.

With the continuous improvement of the NetMind.AI ecosystem and the continuous expansion of the user scale, its native token NMT will also gain more demand support and value capture.

欢迎加入律动 BlockBeats 官方社群：

Telegram 订阅群： https://t.me/theblockbeats

Telegram 交流群： https://t.me/BlockBeats_App

Twitter 官方账号： https://twitter.com/BlockBeatsAsia