Sakana AI, Founded by Ex-Google Researchers, Unveils Open-Source Models in Tokyo - EconoTimes
Sakana AI introduces an evolutionary approach to developing generative models, treating model merging like natural selection. By iteratively breeding and selecting the most effective designs from hundreds of generations, this method offers a significantly cheaper and faster alternative to traditional AI training processes. The company launched three Japanese-language models, two of which are open-sourced. These include a large language model, an image generation tool, and an image-to-text system. Being open-source, these models allow the broader community to inspect, adapt, and build upon the underlying technologies, fostering transparency and collaborative innovation. This initiative is highly relevant to open data because it democratizes access to advanced AI capabilities. By making the core models publicly available, Sakana AI enables researchers and developers to study the efficacy of evolution-inspired architecture. This promotes knowledge sharing and accelerates the development of more efficient AI solutions within the open-source ecosystem.
Source: econotimes.comPublished on 2024-03-22