NEAR Protocol has unveiled new developments in artificial intelligence, including the launch of an AI agent capable of executing cross-chain swaps. The team announced the creation of the “world’s largest” open-source AI model and plans for the decentralization of AI computing.

NEAR AI Alpha is live at https://t.co/WQeAf2cxWs.

This is our first step towards powerful, privacy-preserving AI to ensure the future of technology is truly owned by everyone.

NEAR AI Assistant (alpha): https://t.co/pg0s0FqNzG

NEAR AI Research Hub: https://t.co/dOB3ayPcGx
— NEAR AI (@near_ai) November 10, 2024

“NEAR is the blockchain for artificial intelligence. We aim for everyone to control their data, assets, and choices, creating an alternative to big corporations,” said representatives of NEAR Protocol.

Cross-Chain Swaps with AI Agent

The new AI agent supports cross-chain swaps, enabling the exchange of assets between various blockchains. Currently, Bitcoin, Ethereum, Arbitrum, and Base are supported through the NEAR Intents function. Future integrations are planned with Solana, TON, Dogecoin, BNB Chain, Bitcoin Cash, Ripple, and TRON.

This AI agent can also handle swaps with fiat currencies through market makers like Peanut Trade and utilize liquidity from centralized platforms, including Binance.

“The AI agent can ask another agent, ‘Do you know how to open an account with Revolut?’ and then agree on a service price,” explained Alex Shevchenko, founder of Aurora Labs.

NEAR Protocol’s Largest AI Model

Ilya Polosukhin, co-founder of NEAR Protocol, presented an ambitious plan to create the world’s largest open-source AI model. At the Redacted conference in Bangkok, he stated that the model would have 1.4 trillion parameters, 3.5 times more than Meta’s current flagship model, Llama 3.1 405B.

“The goal has changed. Forget the billion. NEAR is creating the next Trillion (with a capital T) AI agents for Web3,” stated the NEAR Protocol team.

The AI model will be trained on data from 15 trillion tokens, using 16,000 Nvidia H100 GPUs. The creation and maintenance of the neural network will involve “crowdsourced research and development from thousands of participants in the new NEAR AI Research center.” The project will be funded by the sale of NEAR tokens, and the cost of this massive initiative is estimated at $160 million.

Decentralizing AI

Polosukhin emphasized that today, training large AI models requires tens of thousands of GPUs located in one place for fast connection. However, with the development of technologies like DeepMind, there is the potential for decentralized distribution of computations. This, he believes, is an important step toward preventing total control of AI by a single company.

“If AI is controlled by one company, we are effectively doing whatever that company tells us,” explained Polosukhin.

Edward Snowden, former NSA and CIA employee, supported Polosukhin’s opinion, noting that centralized AI turns the world into “a massive surveillance state.”

It’s worth recalling that in July, Polosukhin stated that AI systems do not pose a threat to humanity unless programmed to do so.