Tag
Russian startup Mostik has developed a method for AI models to communicate in latent space, enabling a small model to leverage reasoning from a frontier model without text, achieving 80% accuracy at 20x faster performance, and they are partnering with inference providers to promote open-weight adoption.