Researchers have trained a new kind of large language model (LLM) using GPUs dotted across the world and fed private as well as public data—a move that suggests that the dominant way of building artificial intelligence could be disrupted.
Flower AI and Vana, two startups pursuing unconventional approaches to building AI, worked together to create the new model, called Collective-1.
Flower created techniques that allow training to be spread across hundreds of computers connected over the internet. The company’s technology is already used by some firms to train AI models without needing to pool compute resources or data. Vana
→ Continue reading at WIRED
