Near Protocol has announced an ambitious plan to build the world's largest open-source artificial intelligence model on the opening day of the Redacted conference in Bangkok, Thailand. The model with 1.4 trillion parameters will be 3.5 times larger than Meta's current open-source Llama model.

The model will be developed through a competitive research and development process involving thousands of contributors on the new Near AI Research platform, with members able to participate in training a smaller model with 500 million parameters starting today, November 10.

The ambitious AI model of Near Protocol

The project will evolve in scale and complexity over seven models, with only the best contributors allowed to participate in increasingly complex and larger models. The models will be commercialized and privacy-protected through the use of encrypted Trusted Execution Environments to reward contributors and encourage continuous updates as technology advances.

The high costs of training and computation will be funded through token sales, according to Near Protocol co-founder Illia Polosukhin, at the Redacted conference in Bangkok.

'The cost is about $160 million, which is clearly a large number, but in the crypto space, this amount can be raised,' he shared. Polosukhin added:

'Then, token holders will be reimbursed from all inferences that occur when this model is used. So, we have a business model, a way to make money, a funding mechanism, and a closed loop. Therefore, people can reinvest into the next model.'

Near is one of the few crypto projects capable of executing such an ambitious project. Polosukhin is one of the authors of the pioneering transformer research paper that led to the creation of ChatGPT, and co-founder Alex Skidanov worked at OpenAI during the preparation phase for the launch of this important model at the end of 2022.

Skidanov, currently leading Near AI, acknowledged that this is a significant task with many challenges to overcome.

Decentralized AI addresses privacy issues

To train such a large model, the project will require 'tens of thousands of GPUs at one location,' which is not ideal. But to use a decentralized computing network, 'you will need a new technology that does not currently exist because all existing distributed training techniques require very fast connections.' However, he added that new research from DeepMind shows that this is possible.

Polosukhin stated that he has not yet discussed with existing projects such as the Artificial Superintelligence Alliance but is open to considering collaboration possibilities. Regardless, he asserts that decentralized AI technology must prevail for the benefit of all of us. Guest speaker at the conference, Edward Snowden, emphasized that centralized AI could turn the world into a massive surveillance state.

'This is probably the most important technology today and will be even more important in the future. The fact is, if AI is controlled by a single company, we will have to comply with all the demands of that company,' he explained. Snowden added:

'If the entire AI and economy are run by one company, there will be no decentralization. This is the only way that Web3 still makes philosophical sense, if we have AI that adheres to similar principles.'

Source: https://tapchibitcoin.io/near-len-ke-hoach-xay-dung-mo-hinh-ai-ma-nguon-mo-14-nghin-ty.html