These Startups Are Building Advanced AI Models Without Data Centers
1 min read
Summary
Startups Flower AI and Vana have partnered to create a new large language model, Collective-1, trained using GPUs located across the world and both private and public data.
This method could disrupt the way AI is built, allowing smaller companies to create advanced models and nations without the infrastructure to build powerful data centres to do so.
Flower AI’s approach enables training to be spread across hundreds of computers, whilst Vana’s technology allows for the acquisition of private data.
There are concerns, however, that this “data democratisation” could make datasets more vulnerable to hacking and other security breaches.
Additionally, this method of AI training could enhance competition, allowing newer companies to catch up to industry leaders.
Collective-1 is a relatively small model with 7 billion parameters, but Flower AI is training a model with 30 billion parameters and another with 100 billion parameters.
