These Startups Are Building Advanced AI Models Without Data Centers
Briefly

New startups Flower AI and Vana have developed Collective-1, a novel language model that disrupts traditional AI building by spreading training across global GPUs using diverse dataset sources, including private messages. With 7 billion parameters, it's smaller than leading models yet illustrates a shift towards decentralized AI. Nic Lane, a Cambridge scientist, expresses confidence in this scalable approach, hinting at future models up to 100 billion parameters. This emerging trend may challenge existing power structures in the AI industry reliant on centralized data and computing resources.
Flower AI and Vana's innovative Collective-1 model exemplifies a disruptive approach in AI development by utilizing distributed, global computing resources to train language models.
Collective-1, though smaller than leading AI models, showcases an emerging trend toward decentralized AI architecture, allowing for rapid scaling and diverse data usage.
Read at WIRED
[
|
]