Nous Research, an open-source AI collective, has announced the launch of its decentralized AI training network, "Psyche." This innovative platform aims to revolutionize the development of advanced AI models by aggregating global computing resources.
The Psyche network leverages a distributed infrastructure, allowing individuals and organizations to contribute their idle GPU power to train large language models collaboratively. This approach democratizes AI development, moving away from the reliance on expensive, centralized cloud computing.
The test network for Psyche is already live and is currently pre-training a 40-billion-parameter large language model. This initial phase demonstrates the feasibility and potential of decentralized AI training for future open scientific research.
Nous Research has also developed "DisTrO" (Distributed Training Over-the-Internet), a key technology underpinning Psyche. DisTrO significantly reduces the data transfer required during distributed training, making it a practical solution for leveraging geographically dispersed computing resources.
This launch follows a recent $65 million funding round for Nous Research, with $50 million led by Paradigm. This investment will fuel the development of the Psyche network and accelerate Nous Research's mission to create open and accessible AI technologies. The initiative marks a significant step towards a more decentralized and collaborative future for artificial intelligence.