Near Protocol has unveiled an ambitious plan to develop the world’s largest open-source artificial intelligence model, set to contain a record-breaking 1.4 trillion parameters. This model will surpass Meta’s Llama model by more than three times in size, marking a major advancement in decentralized AI technology. Near announced the project at the Redacted conference in Bangkok, Thailand, revealing it will be developed through a competitive, crowdsourced initiative on the new Near AI Research Hub. Starting November 10, participants can join the training of a smaller 500-million-parameter model as part of this pioneering project.
The project will evolve over seven phases, with progressively more complex models, each involving only the top-performing contributors from previous stages. To encourage collaboration while maintaining privacy, Near Protocol will use encrypted Trusted Execution Environments (TEEs) and a rewards system to protect contributor data and continuously update the model.
Near Protocol plans to fund the project through token sales, providing token holders with returns based on the model’s usage. Co-founder Illia Polosukhin explained that this business model not only provides funding but also allows a reinvestment loop for ongoing improvements. “We have a business model, a way to monetize it, raise funds, and create a loop,” Polosukhin explained, emphasizing that the reinvestments will support continuous training and scaling.
Near Protocol’s unique expertise positions it for this ambitious undertaking. Polosukhin contributed to the transformative research paper that led to AI models like ChatGPT, while co-founder Alex Skidanov previously worked at OpenAI. Despite their experience with centralized AI, both founders are committed to building decentralized AI due to concerns over the potential dominance of large tech companies in the AI space.
Developing a decentralized model of this scale poses significant challenges, especially in terms of computational resources. Skidanov explained that the model would require “tens of thousands of GPUs in one place,” but distributed compute networks might provide an alternative if they can overcome existing technological limitations. Recent research from DeepMind indicates that distributed training could be feasible for this project in the near future.
Privacy advocate and keynote speaker Edward Snowden highlighted the importance of decentralizing AI, cautioning that centralized AI could lead to mass surveillance and control. “If AI is controlled by one company, we’ll effectively be doing whatever that company says,” Snowden explained, stressing that decentralized AI aligns with the fundamental principles of Web3. “If the entire economy is run by one company’s AI, then decentralization becomes irrelevant. Web3’s future depends on building AI that follows these principles.”
Near Protocol’s initiative reflects a growing trend in blockchain to make AI more accessible, private, and decentralized. With the power of crowdsourcing, privacy-focused protocols, and decentralized incentives, Near’s AI project aims to put control of these powerful technologies into the hands of communities rather than corporations.
Source: Cointelegraph
Photo Credits: Cointelegraph
